diff --git a/src/Makefile b/src/Makefile index dcb2eecc6..46d21d247 100644 --- a/src/Makefile +++ b/src/Makefile @@ -379,6 +379,16 @@ endif # I wish I could make dependencies out of rc files :( $(objdir)/win32/Srb2win.res : \ win32/afxres.h win32/resource.h +clean: + $(REMOVE) *~ *.flc + $(REMOVE) $(OBJDIR)/*.o +ifndef NOBLUAJIT + $(REMOVE) $(OBJDIR)/buildvm* $(OBJDIR)/minilua* $(OBJDIR)/lj_vm* $(ALL_HDRGEN) + +cleanjit: + $(REMOVE) $(LJVMCORE_O) $(ALL_GEN) + $(REMOVE) $(OBJDIR)/buildvm* $(OBJDIR)/minilua* $(OBJDIR)/lj_vm* $(ALL_HDRGEN) +endif # dependency recipe template # 1: source file suffix diff --git a/src/bluajit/jit/p.lua b/src/bluajit/jit/p.lua new file mode 100644 index 000000000..ef2ee82f5 --- /dev/null +++ b/src/bluajit/jit/p.lua @@ -0,0 +1,309 @@ +---------------------------------------------------------------------------- +-- LuaJIT profiler. +-- +-- Copyright (C) 2005-2023 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- +-- This module is a simple command line interface to the built-in +-- low-overhead profiler of LuaJIT. +-- +-- The lower-level API of the profiler is accessible via the "jit.profile" +-- module or the luaJIT_profile_* C API. +-- +-- Example usage: +-- +-- luajit -jp myapp.lua +-- luajit -jp=s myapp.lua +-- luajit -jp=-s myapp.lua +-- luajit -jp=vl myapp.lua +-- luajit -jp=G,profile.txt myapp.lua +-- +-- The following dump features are available: +-- +-- f Stack dump: function name, otherwise module:line. Default mode. +-- F Stack dump: ditto, but always prepend module. +-- l Stack dump: module:line. +-- stack dump depth (callee < caller). Default: 1. +-- - Inverse stack dump depth (caller > callee). +-- s Split stack dump after first stack level. Implies abs(depth) >= 2. +-- p Show full path for module names. +-- v Show VM states. Can be combined with stack dumps, e.g. vf or fv. +-- z Show zones. Can be combined with stack dumps, e.g. zf or fz. +-- r Show raw sample counts. Default: show percentages. +-- a Annotate excerpts from source code files. +-- A Annotate complete source code files. +-- G Produce raw output suitable for graphical tools (e.g. flame graphs). +-- m Minimum sample percentage to be shown. Default: 3. +-- i Sampling interval in milliseconds. Default: 10. +-- +---------------------------------------------------------------------------- + +-- Cache some library functions and objects. +local jit = require("jit") +local profile = require("jit.profile") +local vmdef = require("jit.vmdef") +local math = math +local pairs, ipairs, tonumber, floor = pairs, ipairs, tonumber, math.floor +local sort, format = table.sort, string.format +local stdout = io.stdout +local zone -- Load jit.zone module on demand. + +-- Output file handle. +local out + +------------------------------------------------------------------------------ + +local prof_ud +local prof_states, prof_split, prof_min, prof_raw, prof_fmt, prof_depth +local prof_ann, prof_count1, prof_count2, prof_samples + +local map_vmmode = { + N = "Compiled", + I = "Interpreted", + C = "C code", + G = "Garbage Collector", + J = "JIT Compiler", +} + +-- Profiler callback. +local function prof_cb(th, samples, vmmode) + prof_samples = prof_samples + samples + local key_stack, key_stack2, key_state + -- Collect keys for sample. + if prof_states then + if prof_states == "v" then + key_state = map_vmmode[vmmode] or vmmode + else + key_state = zone:get() or "(none)" + end + end + if prof_fmt then + key_stack = profile.dumpstack(th, prof_fmt, prof_depth) + key_stack = key_stack:gsub("%[builtin#(%d+)%]", function(x) + return vmdef.ffnames[tonumber(x)] + end) + if prof_split == 2 then + local k1, k2 = key_stack:match("(.-) [<>] (.*)") + if k2 then key_stack, key_stack2 = k1, k2 end + elseif prof_split == 3 then + key_stack2 = profile.dumpstack(th, "l", 1) + end + end + -- Order keys. + local k1, k2 + if prof_split == 1 then + if key_state then + k1 = key_state + if key_stack then k2 = key_stack end + end + elseif key_stack then + k1 = key_stack + if key_stack2 then k2 = key_stack2 elseif key_state then k2 = key_state end + end + -- Coalesce samples in one or two levels. + if k1 then + local t1 = prof_count1 + t1[k1] = (t1[k1] or 0) + samples + if k2 then + local t2 = prof_count2 + local t3 = t2[k1] + if not t3 then t3 = {}; t2[k1] = t3 end + t3[k2] = (t3[k2] or 0) + samples + end + end +end + +------------------------------------------------------------------------------ + +-- Show top N list. +local function prof_top(count1, count2, samples, indent) + local t, n = {}, 0 + for k in pairs(count1) do + n = n + 1 + t[n] = k + end + sort(t, function(a, b) return count1[a] > count1[b] end) + for i=1,n do + local k = t[i] + local v = count1[k] + local pct = floor(v*100/samples + 0.5) + if pct < prof_min then break end + if not prof_raw then + out:write(format("%s%2d%% %s\n", indent, pct, k)) + elseif prof_raw == "r" then + out:write(format("%s%5d %s\n", indent, v, k)) + else + out:write(format("%s %d\n", k, v)) + end + if count2 then + local r = count2[k] + if r then + prof_top(r, nil, v, (prof_split == 3 or prof_split == 1) and " -- " or + (prof_depth < 0 and " -> " or " <- ")) + end + end + end +end + +-- Annotate source code +local function prof_annotate(count1, samples) + local files = {} + local ms = 0 + for k, v in pairs(count1) do + local pct = floor(v*100/samples + 0.5) + ms = math.max(ms, v) + if pct >= prof_min then + local file, line = k:match("^(.*):(%d+)$") + if not file then file = k; line = 0 end + local fl = files[file] + if not fl then fl = {}; files[file] = fl; files[#files+1] = file end + line = tonumber(line) + fl[line] = prof_raw and v or pct + end + end + sort(files) + local fmtv, fmtn = " %3d%% | %s\n", " | %s\n" + if prof_raw then + local n = math.max(5, math.ceil(math.log10(ms))) + fmtv = "%"..n.."d | %s\n" + fmtn = (" "):rep(n).." | %s\n" + end + local ann = prof_ann + for _, file in ipairs(files) do + local f0 = file:byte() + if f0 == 40 or f0 == 91 then + out:write(format("\n====== %s ======\n[Cannot annotate non-file]\n", file)) + break + end + local fp, err = io.open(file) + if not fp then + out:write(format("====== ERROR: %s: %s\n", file, err)) + break + end + out:write(format("\n====== %s ======\n", file)) + local fl = files[file] + local n, show = 1, false + if ann ~= 0 then + for i=1,ann do + if fl[i] then show = true; out:write("@@ 1 @@\n"); break end + end + end + for line in fp:lines() do + if line:byte() == 27 then + out:write("[Cannot annotate bytecode file]\n") + break + end + local v = fl[n] + if ann ~= 0 then + local v2 = fl[n+ann] + if show then + if v2 then show = n+ann elseif v then show = n + elseif show+ann < n then show = false end + elseif v2 then + show = n+ann + out:write(format("@@ %d @@\n", n)) + end + if not show then goto next end + end + if v then + out:write(format(fmtv, v, line)) + else + out:write(format(fmtn, line)) + end + ::next:: + n = n + 1 + end + fp:close() + end +end + +------------------------------------------------------------------------------ + +-- Finish profiling and dump result. +local function prof_finish() + if prof_ud then + profile.stop() + local samples = prof_samples + if samples == 0 then + if prof_raw ~= true then out:write("[No samples collected]\n") end + elseif prof_ann then + prof_annotate(prof_count1, samples) + else + prof_top(prof_count1, prof_count2, samples, "") + end + prof_count1 = nil + prof_count2 = nil + prof_ud = nil + if out ~= stdout then out:close() end + end +end + +-- Start profiling. +local function prof_start(mode) + local interval = "" + mode = mode:gsub("i%d*", function(s) interval = s; return "" end) + prof_min = 3 + mode = mode:gsub("m(%d+)", function(s) prof_min = tonumber(s); return "" end) + prof_depth = 1 + mode = mode:gsub("%-?%d+", function(s) prof_depth = tonumber(s); return "" end) + local m = {} + for c in mode:gmatch(".") do m[c] = c end + prof_states = m.z or m.v + if prof_states == "z" then zone = require("jit.zone") end + local scope = m.l or m.f or m.F or (prof_states and "" or "f") + local flags = (m.p or "") + prof_raw = m.r + if m.s then + prof_split = 2 + if prof_depth == -1 or m["-"] then prof_depth = -2 + elseif prof_depth == 1 then prof_depth = 2 end + elseif mode:find("[fF].*l") then + scope = "l" + prof_split = 3 + else + prof_split = (scope == "" or mode:find("[zv].*[lfF]")) and 1 or 0 + end + prof_ann = m.A and 0 or (m.a and 3) + if prof_ann then + scope = "l" + prof_fmt = "pl" + prof_split = 0 + prof_depth = 1 + elseif m.G and scope ~= "" then + prof_fmt = flags..scope.."Z;" + prof_depth = -100 + prof_raw = true + prof_min = 0 + elseif scope == "" then + prof_fmt = false + else + local sc = prof_split == 3 and m.f or m.F or scope + prof_fmt = flags..sc..(prof_depth >= 0 and "Z < " or "Z > ") + end + prof_count1 = {} + prof_count2 = {} + prof_samples = 0 + profile.start(scope:lower()..interval, prof_cb) + prof_ud = newproxy(true) + getmetatable(prof_ud).__gc = prof_finish +end + +------------------------------------------------------------------------------ + +local function start(mode, outfile) + if not outfile then outfile = os.getenv("LUAJIT_PROFILEFILE") end + if outfile then + out = outfile == "-" and stdout or assert(io.open(outfile, "w")) + else + out = stdout + end + prof_start(mode or "f") +end + +-- Public module functions. +return { + start = start, -- For -j command line option. + stop = prof_finish +} + diff --git a/src/bluajit/lib_base.c b/src/bluajit/lib_base.c new file mode 100644 index 000000000..7ffdc3dd9 --- /dev/null +++ b/src/bluajit/lib_base.c @@ -0,0 +1,709 @@ +/* +** Base and coroutine library. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +** +** Major portions taken verbatim or adapted from the Lua interpreter. +** Copyright (C) 1994-2011 Lua.org, PUC-Rio. See Copyright Notice in lua.h +*/ + +#include + +#define lib_base_c +#define LUA_LIB + +#include "lua.h" +#include "lauxlib.h" +#include "lualib.h" + +#include "lj_obj.h" +#include "lj_gc.h" +#include "lj_err.h" +#include "lj_debug.h" +#include "lj_buf.h" +#include "lj_str.h" +#include "lj_tab.h" +#include "lj_meta.h" +#include "lj_state.h" +#include "lj_frame.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#include "lj_cconv.h" +#endif +#include "lj_bc.h" +#include "lj_ff.h" +#include "lj_dispatch.h" +#include "lj_char.h" +#include "lj_strscan.h" +#include "lj_strfmt.h" +#include "lj_lib.h" + +/* -- Base library: checks ------------------------------------------------ */ + +#define LJLIB_MODULE_base + +LJLIB_ASM(assert) LJLIB_REC(.) +{ + lj_lib_checkany(L, 1); + if (L->top == L->base+1) + lj_err_caller(L, LJ_ERR_ASSERT); + else if (tvisstr(L->base+1) || tvisnumber(L->base+1)) + lj_err_callermsg(L, strdata(lj_lib_checkstr(L, 2))); + else + lj_err_run(L); + return FFH_UNREACHABLE; +} + +/* ORDER LJ_T */ +LJLIB_PUSH("nil") +LJLIB_PUSH("boolean") +LJLIB_PUSH(top-1) /* boolean */ +LJLIB_PUSH("userdata") +LJLIB_PUSH("string") +LJLIB_PUSH("upval") +LJLIB_PUSH("thread") +LJLIB_PUSH("proto") +LJLIB_PUSH("function") +LJLIB_PUSH("trace") +LJLIB_PUSH("cdata") +LJLIB_PUSH("table") +LJLIB_PUSH(top-9) /* userdata */ +LJLIB_PUSH("number") +LJLIB_ASM_(type) LJLIB_REC(.) +/* Recycle the lj_lib_checkany(L, 1) from assert. */ + +/* -- Base library: iterators --------------------------------------------- */ + +/* This solves a circular dependency problem -- change FF_next_N as needed. */ +LJ_STATIC_ASSERT((int)FF_next == FF_next_N); + +LJLIB_ASM(next) LJLIB_REC(.) +{ + lj_lib_checktab(L, 1); + lj_err_msg(L, LJ_ERR_NEXTIDX); + return FFH_UNREACHABLE; +} + +#if LJ_52 || LJ_HASFFI +static int ffh_pairs(lua_State *L, MMS mm) +{ + TValue *o = lj_lib_checkany(L, 1); + cTValue *mo = lj_meta_lookup(L, o, mm); + if ((LJ_52 || tviscdata(o)) && !tvisnil(mo)) { + L->top = o+1; /* Only keep one argument. */ + copyTV(L, L->base-1-LJ_FR2, mo); /* Replace callable. */ + return FFH_TAILCALL; + } else { + if (!tvistab(o)) lj_err_argt(L, 1, LUA_TTABLE); + if (LJ_FR2) { copyTV(L, o-1, o); o--; } + setfuncV(L, o-1, funcV(lj_lib_upvalue(L, 1))); + if (mm == MM_pairs) setnilV(o+1); else setintV(o+1, 0); + return FFH_RES(3); + } +} +#else +#define ffh_pairs(L, mm) (lj_lib_checktab(L, 1), FFH_UNREACHABLE) +#endif + +LJLIB_PUSH(lastcl) +LJLIB_ASM(pairs) LJLIB_REC(xpairs 0) +{ + return ffh_pairs(L, MM_pairs); +} + +LJLIB_NOREGUV LJLIB_ASM(ipairs_aux) LJLIB_REC(.) +{ + lj_lib_checktab(L, 1); + lj_lib_checkint(L, 2); + return FFH_UNREACHABLE; +} + +LJLIB_PUSH(lastcl) +LJLIB_ASM(ipairs) LJLIB_REC(xpairs 1) +{ + return ffh_pairs(L, MM_ipairs); +} + +/* -- Base library: getters and setters ----------------------------------- */ + +LJLIB_ASM_(getmetatable) LJLIB_REC(.) +/* Recycle the lj_lib_checkany(L, 1) from assert. */ + +LJLIB_ASM(setmetatable) LJLIB_REC(.) +{ + GCtab *t = lj_lib_checktab(L, 1); + GCtab *mt = lj_lib_checktabornil(L, 2); + if (!tvisnil(lj_meta_lookup(L, L->base, MM_metatable))) + lj_err_caller(L, LJ_ERR_PROTMT); + setgcref(t->metatable, obj2gco(mt)); + if (mt) { lj_gc_objbarriert(L, t, mt); } + settabV(L, L->base-1-LJ_FR2, t); + return FFH_RES(1); +} + +LJLIB_CF(getfenv) LJLIB_REC(.) +{ + GCfunc *fn; + cTValue *o = L->base; + if (!(o < L->top && tvisfunc(o))) { + int level = lj_lib_optint(L, 1, 1); + o = lj_debug_frame(L, level, &level); + if (o == NULL) + lj_err_arg(L, 1, LJ_ERR_INVLVL); + if (LJ_FR2) o--; + } + fn = &gcval(o)->fn; + settabV(L, L->top++, isluafunc(fn) ? tabref(fn->l.env) : tabref(L->env)); + return 1; +} + +LJLIB_CF(setfenv) +{ + GCfunc *fn; + GCtab *t = lj_lib_checktab(L, 2); + cTValue *o = L->base; + if (!(o < L->top && tvisfunc(o))) { + int level = lj_lib_checkint(L, 1); + if (level == 0) { + /* NOBARRIER: A thread (i.e. L) is never black. */ + setgcref(L->env, obj2gco(t)); + return 0; + } + o = lj_debug_frame(L, level, &level); + if (o == NULL) + lj_err_arg(L, 1, LJ_ERR_INVLVL); + if (LJ_FR2) o--; + } + fn = &gcval(o)->fn; + if (!isluafunc(fn)) + lj_err_caller(L, LJ_ERR_SETFENV); + setgcref(fn->l.env, obj2gco(t)); + lj_gc_objbarrier(L, obj2gco(fn), t); + setfuncV(L, L->top++, fn); + return 1; +} + +LJLIB_ASM(rawget) LJLIB_REC(.) +{ + lj_lib_checktab(L, 1); + lj_lib_checkany(L, 2); + return FFH_UNREACHABLE; +} + +LJLIB_CF(rawset) LJLIB_REC(.) +{ + lj_lib_checktab(L, 1); + lj_lib_checkany(L, 2); + L->top = 1+lj_lib_checkany(L, 3); + lua_rawset(L, 1); + return 1; +} + +LJLIB_CF(rawequal) LJLIB_REC(.) +{ + cTValue *o1 = lj_lib_checkany(L, 1); + cTValue *o2 = lj_lib_checkany(L, 2); + setboolV(L->top-1, lj_obj_equal(o1, o2)); + return 1; +} + +#if LJ_52 +LJLIB_CF(rawlen) LJLIB_REC(.) +{ + cTValue *o = L->base; + int32_t len; + if (L->top > o && tvisstr(o)) + len = (int32_t)strV(o)->len; + else + len = (int32_t)lj_tab_len(lj_lib_checktab(L, 1)); + setintV(L->top-1, len); + return 1; +} +#endif + +LJLIB_CF(unpack) +{ + GCtab *t = lj_lib_checktab(L, 1); + int32_t n, i = lj_lib_optint(L, 2, 1); + int32_t e = (L->base+3-1 < L->top && !tvisnil(L->base+3-1)) ? + lj_lib_checkint(L, 3) : (int32_t)lj_tab_len(t); + uint32_t nu; + if (i > e) return 0; + nu = (uint32_t)e - (uint32_t)i; + n = (int32_t)(nu+1); + if (nu >= LUAI_MAXCSTACK || !lua_checkstack(L, n)) + lj_err_caller(L, LJ_ERR_UNPACK); + do { + cTValue *tv = lj_tab_getint(t, i); + if (tv) { + copyTV(L, L->top++, tv); + } else { + setnilV(L->top++); + } + } while (i++ < e); + return n; +} + +LJLIB_CF(select) LJLIB_REC(.) +{ + int32_t n = (int32_t)(L->top - L->base); + if (n >= 1 && tvisstr(L->base) && *strVdata(L->base) == '#') { + setintV(L->top-1, n-1); + return 1; + } else { + int32_t i = lj_lib_checkint(L, 1); + if (i < 0) i = n + i; else if (i > n) i = n; + if (i < 1) + lj_err_arg(L, 1, LJ_ERR_IDXRNG); + return n - i; + } +} + +/* -- Base library: conversions ------------------------------------------- */ + +LJLIB_ASM(tonumber) LJLIB_REC(.) +{ + int32_t base = lj_lib_optint(L, 2, 10); + if (base == 10) { + TValue *o = lj_lib_checkany(L, 1); + if (lj_strscan_numberobj(o)) { + copyTV(L, L->base-1-LJ_FR2, o); + return FFH_RES(1); + } +#if LJ_HASFFI + if (tviscdata(o)) { + CTState *cts = ctype_cts(L); + CType *ct = lj_ctype_rawref(cts, cdataV(o)->ctypeid); + if (ctype_isenum(ct->info)) ct = ctype_child(cts, ct); + if (ctype_isnum(ct->info) || ctype_iscomplex(ct->info)) { + if (LJ_DUALNUM && ctype_isinteger_or_bool(ct->info) && + ct->size <= 4 && !(ct->size == 4 && (ct->info & CTF_UNSIGNED))) { + int32_t i; + lj_cconv_ct_tv(cts, ctype_get(cts, CTID_INT32), (uint8_t *)&i, o, 0); + setintV(L->base-1-LJ_FR2, i); + return FFH_RES(1); + } + lj_cconv_ct_tv(cts, ctype_get(cts, CTID_DOUBLE), + (uint8_t *)&(L->base-1-LJ_FR2)->n, o, 0); + return FFH_RES(1); + } + } +#endif + } else { + const char *p = strdata(lj_lib_checkstr(L, 1)); + char *ep; + unsigned int neg = 0; + unsigned long ul; + if (base < 2 || base > 36) + lj_err_arg(L, 2, LJ_ERR_BASERNG); + while (lj_char_isspace((unsigned char)(*p))) p++; + if (*p == '-') { p++; neg = 1; } else if (*p == '+') { p++; } + if (lj_char_isalnum((unsigned char)(*p))) { + ul = strtoul(p, &ep, base); + if (p != ep) { + while (lj_char_isspace((unsigned char)(*ep))) ep++; + if (*ep == '\0') { + if (LJ_DUALNUM && LJ_LIKELY(ul < 0x80000000u+neg)) { + if (neg) ul = ~ul+1u; + setintV(L->base-1-LJ_FR2, (int32_t)ul); + } else { + lua_Number n = (lua_Number)ul; + if (neg) n = -n; + setnumV(L->base-1-LJ_FR2, n); + } + return FFH_RES(1); + } + } + } + } + setnilV(L->base-1-LJ_FR2); + return FFH_RES(1); +} + +LJLIB_ASM(tostring) LJLIB_REC(.) +{ + TValue *o = lj_lib_checkany(L, 1); + cTValue *mo; + L->top = o+1; /* Only keep one argument. */ + if (!tvisnil(mo = lj_meta_lookup(L, o, MM_tostring))) { + copyTV(L, L->base-1-LJ_FR2, mo); /* Replace callable. */ + return FFH_TAILCALL; + } + lj_gc_check(L); + setstrV(L, L->base-1-LJ_FR2, lj_strfmt_obj(L, L->base)); + return FFH_RES(1); +} + +/* -- Base library: throw and catch errors -------------------------------- */ + +LJLIB_CF(error) +{ + int32_t level = lj_lib_optint(L, 2, 1); + lua_settop(L, 1); + if (lua_isstring(L, 1) && level > 0) { + luaL_where(L, level); + lua_pushvalue(L, 1); + lua_concat(L, 2); + } + return lua_error(L); +} + +LJLIB_ASM(pcall) LJLIB_REC(.) +{ + lj_lib_checkany(L, 1); + lj_lib_checkfunc(L, 2); /* For xpcall only. */ + return FFH_UNREACHABLE; +} +LJLIB_ASM_(xpcall) LJLIB_REC(.) + +/* -- Base library: load Lua code ----------------------------------------- */ + +#if !LJ_SRB2LIB +static int load_aux(lua_State *L, int status, int envarg) +{ + if (status == LUA_OK) { + /* + ** Set environment table for top-level function. + ** Don't do this for non-native bytecode, which returns a prototype. + */ + if (tvistab(L->base+envarg-1) && tvisfunc(L->top-1)) { + GCfunc *fn = funcV(L->top-1); + GCtab *t = tabV(L->base+envarg-1); + setgcref(fn->c.env, obj2gco(t)); + lj_gc_objbarrier(L, fn, t); + } + return 1; + } else { + setnilV(L->top-2); + return 2; + } +} + +LJLIB_CF(loadfile) +{ + GCstr *fname = lj_lib_optstr(L, 1); + GCstr *mode = lj_lib_optstr(L, 2); + int status; + lua_settop(L, 3); /* Ensure env arg exists. */ + status = luaL_loadfilex(L, fname ? strdata(fname) : NULL, + mode ? strdata(mode) : NULL); + return load_aux(L, status, 3); +} + +static const char *reader_func(lua_State *L, void *ud, size_t *size) +{ + UNUSED(ud); + luaL_checkstack(L, 2, "too many nested functions"); + copyTV(L, L->top++, L->base); + lua_call(L, 0, 1); /* Call user-supplied function. */ + L->top--; + if (tvisnil(L->top)) { + *size = 0; + return NULL; + } else if (tvisstr(L->top) || tvisnumber(L->top)) { + copyTV(L, L->base+4, L->top); /* Anchor string in reserved stack slot. */ + return lua_tolstring(L, 5, size); + } else { + lj_err_caller(L, LJ_ERR_RDRSTR); + return NULL; + } +} + +LJLIB_CF(load) +{ + GCstr *name = lj_lib_optstr(L, 2); + GCstr *mode = lj_lib_optstr(L, 3); + int status; + if (L->base < L->top && + (tvisstr(L->base) || tvisnumber(L->base) || tvisbuf(L->base))) { + const char *s; + MSize len; + if (tvisbuf(L->base)) { + SBufExt *sbx = bufV(L->base); + s = sbx->r; + len = sbufxlen(sbx); + if (!name) name = &G(L)->strempty; /* Buffers are not NUL-terminated. */ + } else { + GCstr *str = lj_lib_checkstr(L, 1); + s = strdata(str); + len = str->len; + } + lua_settop(L, 4); /* Ensure env arg exists. */ + status = luaL_loadbufferx(L, s, len, name ? strdata(name) : s, + mode ? strdata(mode) : NULL); + } else { + lj_lib_checkfunc(L, 1); + lua_settop(L, 5); /* Reserve a slot for the string from the reader. */ + status = lua_loadx(L, reader_func, NULL, name ? strdata(name) : "=(load)", + mode ? strdata(mode) : NULL); + } + return load_aux(L, status, 4); +} + +LJLIB_CF(loadstring) +{ + return lj_cf_load(L); +} + +LJLIB_CF(dofile) +{ + GCstr *fname = lj_lib_optstr(L, 1); + setnilV(L->top); + L->top = L->base+1; + if (luaL_loadfile(L, fname ? strdata(fname) : NULL) != LUA_OK) + lua_error(L); + lua_call(L, 0, LUA_MULTRET); + return (int)(L->top - L->base) - 1; +} +#endif + +/* -- Base library: GC control -------------------------------------------- */ + +LJLIB_CF(gcinfo) +{ + setintV(L->top++, (int32_t)(G(L)->gc.total >> 10)); + return 1; +} + +LJLIB_CF(collectgarbage) +{ + int opt = lj_lib_checkopt(L, 1, LUA_GCCOLLECT, /* ORDER LUA_GC* */ + "\4stop\7restart\7collect\5count\1\377\4step\10setpause\12setstepmul\1\377\11isrunning"); + int32_t data = lj_lib_optint(L, 2, 0); + if (opt == LUA_GCCOUNT) { +#if LJ_INTONLY + setintV(L->top, G(L)->gc.total/1024); +#else + setnumV(L->top, (lua_Number)G(L)->gc.total/1024.0); +#endif + } else { + int res = lua_gc(L, opt, data); + if (opt == LUA_GCSTEP || opt == LUA_GCISRUNNING) + setboolV(L->top, res); + else + setintV(L->top, res); + } + L->top++; + return 1; +} + +/* -- Base library: miscellaneous functions ------------------------------- */ + +LJLIB_PUSH(top-2) /* Upvalue holds weak table. */ +LJLIB_CF(newproxy) +{ + lua_settop(L, 1); + lua_newuserdata(L, 0); + if (lua_toboolean(L, 1) == 0) { /* newproxy(): without metatable. */ + return 1; + } else if (lua_isboolean(L, 1)) { /* newproxy(true): with metatable. */ + lua_newtable(L); + lua_pushvalue(L, -1); + lua_pushboolean(L, 1); + lua_rawset(L, lua_upvalueindex(1)); /* Remember mt in weak table. */ + } else { /* newproxy(proxy): inherit metatable. */ + int validproxy = 0; + if (lua_getmetatable(L, 1)) { + lua_rawget(L, lua_upvalueindex(1)); + validproxy = lua_toboolean(L, -1); + lua_pop(L, 1); + } + if (!validproxy) + lj_err_arg(L, 1, LJ_ERR_NOPROXY); + lua_getmetatable(L, 1); + } + lua_setmetatable(L, 2); + return 1; +} + +LJLIB_PUSH("tostring") +LJLIB_CF(print) +{ + ptrdiff_t i, nargs = L->top - L->base; + cTValue *tv = lj_tab_getstr(tabref(L->env), strV(lj_lib_upvalue(L, 1))); + int shortcut; + if (tv && !tvisnil(tv)) { + copyTV(L, L->top++, tv); + } else { + setstrV(L, L->top++, strV(lj_lib_upvalue(L, 1))); + lua_gettable(L, LUA_GLOBALSINDEX); + tv = L->top-1; + } + shortcut = (tvisfunc(tv) && funcV(tv)->c.ffid == FF_tostring) && + !gcrefu(basemt_it(G(L), LJ_TNUMX)); + for (i = 0; i < nargs; i++) { + cTValue *o = &L->base[i]; + const char *str; + size_t size; + MSize len; + if (shortcut && (str = lj_strfmt_wstrnum(L, o, &len)) != NULL) { + size = len; + } else { + copyTV(L, L->top+1, o); + copyTV(L, L->top, L->top-1); + L->top += 2; + lua_call(L, 1, 1); + str = lua_tolstring(L, -1, &size); + if (!str) + lj_err_caller(L, LJ_ERR_PRTOSTR); + L->top--; + } + if (i) + putchar('\t'); + fwrite(str, 1, size, stdout); + } + putchar('\n'); + return 0; +} + +LJLIB_PUSH(top-3) +LJLIB_SET(_VERSION) + +#include "lj_libdef.h" + +/* -- Coroutine library --------------------------------------------------- */ + +#define LJLIB_MODULE_coroutine + +LJLIB_CF(coroutine_status) +{ + const char *s; + lua_State *co; + if (!(L->top > L->base && tvisthread(L->base))) + lj_err_arg(L, 1, LJ_ERR_NOCORO); + co = threadV(L->base); + if (co == L) s = "running"; + else if (co->status == LUA_YIELD) s = "suspended"; + else if (co->status != LUA_OK) s = "dead"; + else if (co->base > tvref(co->stack)+1+LJ_FR2) s = "normal"; + else if (co->top == co->base) s = "dead"; + else s = "suspended"; + lua_pushstring(L, s); + return 1; +} + +LJLIB_CF(coroutine_running) +{ +#if LJ_52 + int ismain = lua_pushthread(L); + setboolV(L->top++, ismain); + return 2; +#else + if (lua_pushthread(L)) + setnilV(L->top++); + return 1; +#endif +} + +LJLIB_CF(coroutine_isyieldable) +{ + setboolV(L->top++, cframe_canyield(L->cframe)); + return 1; +} + +LJLIB_CF(coroutine_create) +{ + lua_State *L1; + if (!(L->base < L->top && tvisfunc(L->base))) + lj_err_argt(L, 1, LUA_TFUNCTION); + L1 = lua_newthread(L); + setfuncV(L, L1->top++, funcV(L->base)); + return 1; +} + +LJLIB_ASM(coroutine_yield) +{ + lj_err_caller(L, LJ_ERR_CYIELD); + return FFH_UNREACHABLE; +} + +static int ffh_resume(lua_State *L, lua_State *co, int wrap) +{ + if (co->cframe != NULL || co->status > LUA_YIELD || + (co->status == LUA_OK && co->top == co->base)) { + ErrMsg em = co->cframe ? LJ_ERR_CORUN : LJ_ERR_CODEAD; + if (wrap) lj_err_caller(L, em); + setboolV(L->base-1-LJ_FR2, 0); + setstrV(L, L->base-LJ_FR2, lj_err_str(L, em)); + return FFH_RES(2); + } + if (lj_state_cpgrowstack(co, (MSize)(L->top - L->base)) != LUA_OK) { + cTValue *msg = --co->top; + lj_err_callermsg(L, strVdata(msg)); + } + return FFH_RETRY; +} + +LJLIB_ASM(coroutine_resume) +{ + if (!(L->top > L->base && tvisthread(L->base))) + lj_err_arg(L, 1, LJ_ERR_NOCORO); + return ffh_resume(L, threadV(L->base), 0); +} + +LJLIB_NOREG LJLIB_ASM(coroutine_wrap_aux) +{ + return ffh_resume(L, threadV(lj_lib_upvalue(L, 1)), 1); +} + +/* Inline declarations. */ +LJ_ASMF void lj_ff_coroutine_wrap_aux(void); +#if !(LJ_TARGET_MIPS && defined(ljamalg_c)) +LJ_FUNCA_NORET void LJ_FASTCALL lj_ffh_coroutine_wrap_err(lua_State *L, + lua_State *co); +#endif + +/* Error handler, called from assembler VM. */ +void LJ_FASTCALL lj_ffh_coroutine_wrap_err(lua_State *L, lua_State *co) +{ + co->top--; copyTV(L, L->top, co->top); L->top++; + if (tvisstr(L->top-1)) + lj_err_callermsg(L, strVdata(L->top-1)); + else + lj_err_run(L); +} + +/* Forward declaration. */ +static void setpc_wrap_aux(lua_State *L, GCfunc *fn); + +LJLIB_CF(coroutine_wrap) +{ + GCfunc *fn; + lj_cf_coroutine_create(L); + fn = lj_lib_pushcc(L, lj_ffh_coroutine_wrap_aux, FF_coroutine_wrap_aux, 1); + setpc_wrap_aux(L, fn); + return 1; +} + +#include "lj_libdef.h" + +/* Fix the PC of wrap_aux. Really ugly workaround. */ +static void setpc_wrap_aux(lua_State *L, GCfunc *fn) +{ + setmref(fn->c.pc, &L2GG(L)->bcff[lj_lib_init_coroutine[1]+2]); +} + +/* ------------------------------------------------------------------------ */ + +static void newproxy_weaktable(lua_State *L) +{ + /* NOBARRIER: The table is new (marked white). */ + GCtab *t = lj_tab_new(L, 0, 1); + settabV(L, L->top++, t); + setgcref(t->metatable, obj2gco(t)); + setstrV(L, lj_tab_setstr(L, t, lj_str_newlit(L, "__mode")), + lj_str_newlit(L, "kv")); + t->nomm = (uint8_t)(~(1u<env); + settabV(L, lj_tab_setstr(L, env, lj_str_newlit(L, "_G")), env); + lua_pushliteral(L, LUA_VERSION); /* top-3. */ + newproxy_weaktable(L); /* top-2. */ + LJ_LIB_REG(L, "_G", base); + LJ_LIB_REG(L, LUA_COLIBNAME, coroutine); + return 2; +} + diff --git a/src/bluajit/lib_bit.c b/src/bluajit/lib_bit.c new file mode 100644 index 000000000..ada9614d0 --- /dev/null +++ b/src/bluajit/lib_bit.c @@ -0,0 +1,181 @@ +/* +** Bit manipulation library. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#define lib_bit_c +#define LUA_LIB + +#include "lua.h" +#include "lauxlib.h" +#include "lualib.h" + +#include "lj_obj.h" +#include "lj_err.h" +#include "lj_buf.h" +#include "lj_strscan.h" +#include "lj_strfmt.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#include "lj_cdata.h" +#include "lj_cconv.h" +#include "lj_carith.h" +#endif +#include "lj_ff.h" +#include "lj_lib.h" + +/* ------------------------------------------------------------------------ */ + +#define LJLIB_MODULE_bit + +#if LJ_HASFFI +static int bit_result64(lua_State *L, CTypeID id, uint64_t x) +{ + GCcdata *cd = lj_cdata_new_(L, id, 8); + *(uint64_t *)cdataptr(cd) = x; + setcdataV(L, L->base-1-LJ_FR2, cd); + return FFH_RES(1); +} +#else +static int32_t bit_checkbit(lua_State *L, int narg) +{ + TValue *o = L->base + narg-1; + if (!(o < L->top && lj_strscan_numberobj(o))) + lj_err_argt(L, narg, LUA_TNUMBER); + if (LJ_LIKELY(tvisint(o))) { + return intV(o); + } else { + int32_t i = lj_num2bit(numV(o)); + if (LJ_DUALNUM) setintV(o, i); + return i; + } +} +#endif + +LJLIB_ASM(bit_tobit) LJLIB_REC(bit_tobit) +{ +#if LJ_HASFFI + CTypeID id = 0; + setintV(L->base-1-LJ_FR2, (int32_t)lj_carith_check64(L, 1, &id)); + return FFH_RES(1); +#else + lj_lib_checknumber(L, 1); + return FFH_RETRY; +#endif +} + +LJLIB_ASM(bit_bnot) LJLIB_REC(bit_unary IR_BNOT) +{ +#if LJ_HASFFI + CTypeID id = 0; + uint64_t x = lj_carith_check64(L, 1, &id); + return id ? bit_result64(L, id, ~x) : FFH_RETRY; +#else + lj_lib_checknumber(L, 1); + return FFH_RETRY; +#endif +} + +LJLIB_ASM(bit_bswap) LJLIB_REC(bit_unary IR_BSWAP) +{ +#if LJ_HASFFI + CTypeID id = 0; + uint64_t x = lj_carith_check64(L, 1, &id); + return id ? bit_result64(L, id, lj_bswap64(x)) : FFH_RETRY; +#else + lj_lib_checknumber(L, 1); + return FFH_RETRY; +#endif +} + +LJLIB_ASM(bit_lshift) LJLIB_REC(bit_shift IR_BSHL) +{ +#if LJ_HASFFI + CTypeID id = 0, id2 = 0; + uint64_t x = lj_carith_check64(L, 1, &id); + int32_t sh = (int32_t)lj_carith_check64(L, 2, &id2); + if (id) { + x = lj_carith_shift64(x, sh, curr_func(L)->c.ffid - (int)FF_bit_lshift); + return bit_result64(L, id, x); + } + setintV(L->base+1, sh); + return FFH_RETRY; +#else + lj_lib_checknumber(L, 1); + bit_checkbit(L, 2); + return FFH_RETRY; +#endif +} +LJLIB_ASM_(bit_rshift) LJLIB_REC(bit_shift IR_BSHR) +LJLIB_ASM_(bit_arshift) LJLIB_REC(bit_shift IR_BSAR) +LJLIB_ASM_(bit_rol) LJLIB_REC(bit_shift IR_BROL) +LJLIB_ASM_(bit_ror) LJLIB_REC(bit_shift IR_BROR) + +LJLIB_ASM(bit_band) LJLIB_REC(bit_nary IR_BAND) +{ +#if LJ_HASFFI + CTypeID id = 0; + TValue *o = L->base, *top = L->top; + int i = 0; + do { lj_carith_check64(L, ++i, &id); } while (++o < top); + if (id) { + CTState *cts = ctype_cts(L); + CType *ct = ctype_get(cts, id); + int op = curr_func(L)->c.ffid - (int)FF_bit_bor; + uint64_t x, y = op >= 0 ? 0 : ~(uint64_t)0; + o = L->base; + do { + lj_cconv_ct_tv(cts, ct, (uint8_t *)&x, o, 0); + if (op < 0) y &= x; else if (op == 0) y |= x; else y ^= x; + } while (++o < top); + return bit_result64(L, id, y); + } + return FFH_RETRY; +#else + int i = 0; + do { lj_lib_checknumber(L, ++i); } while (L->base+i < L->top); + return FFH_RETRY; +#endif +} +LJLIB_ASM_(bit_bor) LJLIB_REC(bit_nary IR_BOR) +LJLIB_ASM_(bit_bxor) LJLIB_REC(bit_nary IR_BXOR) + +/* ------------------------------------------------------------------------ */ + +LJLIB_CF(bit_tohex) LJLIB_REC(.) +{ +#if LJ_HASFFI + CTypeID id = 0, id2 = 0; + uint64_t b = lj_carith_check64(L, 1, &id); + int32_t n = L->base+1>=L->top ? (id ? 16 : 8) : + (int32_t)lj_carith_check64(L, 2, &id2); +#else + uint32_t b = (uint32_t)bit_checkbit(L, 1); + int32_t n = L->base+1>=L->top ? 8 : bit_checkbit(L, 2); +#endif + SBuf *sb = lj_buf_tmp_(L); + SFormat sf = (STRFMT_UINT|STRFMT_T_HEX); + if (n < 0) { n = (int32_t)(~(uint32_t)n+1u); sf |= STRFMT_F_UPPER; } + if ((uint32_t)n > 254) n = 254; + sf |= ((SFormat)((n+1)&255) << STRFMT_SH_PREC); +#if LJ_HASFFI + if (n < 16) b &= ((uint64_t)1 << 4*n)-1; +#else + if (n < 8) b &= (1u << 4*n)-1; +#endif + sb = lj_strfmt_putfxint(sb, sf, b); + setstrV(L, L->top-1, lj_buf_str(L, sb)); + lj_gc_check(L); + return 1; +} + +/* ------------------------------------------------------------------------ */ + +#include "lj_libdef.h" + +LUALIB_API int luaopen_bit(lua_State *L) +{ + LJ_LIB_REG(L, LUA_BITLIBNAME, bit); + return 1; +} + diff --git a/src/bluajit/lib_table.c b/src/bluajit/lib_table.c new file mode 100644 index 000000000..f4266b2f0 --- /dev/null +++ b/src/bluajit/lib_table.c @@ -0,0 +1,331 @@ +/* +** Table library. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +** +** Major portions taken verbatim or adapted from the Lua interpreter. +** Copyright (C) 1994-2008 Lua.org, PUC-Rio. See Copyright Notice in lua.h +*/ + +#define lib_table_c +#define LUA_LIB + +#include "lua.h" +#include "lauxlib.h" +#include "lualib.h" + +#include "lj_obj.h" +#include "lj_gc.h" +#include "lj_err.h" +#include "lj_buf.h" +#include "lj_tab.h" +#include "lj_ff.h" +#include "lj_lib.h" + +/* ------------------------------------------------------------------------ */ + +#define LJLIB_MODULE_table + +LJLIB_LUA(table_foreachi) /* + function(t, f) + CHECK_tab(t) + CHECK_func(f) + for i=1,#t do + local r = f(i, t[i]) + if r ~= nil then return r end + end + end +*/ + +LJLIB_LUA(table_foreach) /* + function(t, f) + CHECK_tab(t) + CHECK_func(f) + for k, v in PAIRS(t) do + local r = f(k, v) + if r ~= nil then return r end + end + end +*/ + +LJLIB_LUA(table_getn) /* + function(t) + CHECK_tab(t) + return #t + end +*/ + +LJLIB_CF(table_maxn) +{ + GCtab *t = lj_lib_checktab(L, 1); + TValue *array = tvref(t->array); + Node *node; + lua_Number m = 0; + ptrdiff_t i; + for (i = (ptrdiff_t)t->asize - 1; i >= 0; i--) + if (!tvisnil(&array[i])) { + m = (lua_Number)(int32_t)i; + break; + } + node = noderef(t->node); + for (i = (ptrdiff_t)t->hmask; i >= 0; i--) + if (!tvisnil(&node[i].val) && tvisnumber(&node[i].key)) { + lua_Number n = numberVnum(&node[i].key); + if (n > m) m = n; + } +#if LJ_INTONLY + setintV(L->top-1, lj_num2int(m)); +#else + setnumV(L->top-1, m); +#endif + return 1; +} + +LJLIB_CF(table_insert) LJLIB_REC(.) +{ + GCtab *t = lj_lib_checktab(L, 1); + int32_t n, i = (int32_t)lj_tab_len(t) + 1; + int nargs = (int)((char *)L->top - (char *)L->base); + if (nargs != 2*sizeof(TValue)) { + if (nargs != 3*sizeof(TValue)) + lj_err_caller(L, LJ_ERR_TABINS); + /* NOBARRIER: This just moves existing elements around. */ + for (n = lj_lib_checkint(L, 2); i > n; i--) { + /* The set may invalidate the get pointer, so need to do it first! */ + TValue *dst = lj_tab_setint(L, t, i); + cTValue *src = lj_tab_getint(t, i-1); + if (src) { + copyTV(L, dst, src); + } else { + setnilV(dst); + } + } + i = n; + } + { + TValue *dst = lj_tab_setint(L, t, i); + copyTV(L, dst, L->top-1); /* Set new value. */ + lj_gc_barriert(L, t, dst); + } + return 0; +} + +LJLIB_LUA(table_remove) /* + function(t, pos) + CHECK_tab(t) + local len = #t + if pos == nil then + if len ~= 0 then + local old = t[len] + t[len] = nil + return old + end + else + CHECK_int(pos) + if pos >= 1 and pos <= len then + local old = t[pos] + for i=pos+1,len do + t[i-1] = t[i] + end + t[len] = nil + return old + end + end + end +*/ + +LJLIB_LUA(table_move) /* + function(a1, f, e, t, a2) + CHECK_tab(a1) + CHECK_int(f) + CHECK_int(e) + CHECK_int(t) + if a2 == nil then a2 = a1 end + CHECK_tab(a2) + if e >= f then + local d = t - f + if t > e or t <= f or a2 ~= a1 then + for i=f,e do a2[i+d] = a1[i] end + else + for i=e,f,-1 do a2[i+d] = a1[i] end + end + end + return a2 + end +*/ + +LJLIB_CF(table_concat) LJLIB_REC(.) +{ + GCtab *t = lj_lib_checktab(L, 1); + GCstr *sep = lj_lib_optstr(L, 2); + int32_t i = lj_lib_optint(L, 3, 1); + int32_t e = (L->base+3 < L->top && !tvisnil(L->base+3)) ? + lj_lib_checkint(L, 4) : (int32_t)lj_tab_len(t); + SBuf *sb = lj_buf_tmp_(L); + SBuf *sbx = lj_buf_puttab(sb, t, sep, i, e); + if (LJ_UNLIKELY(!sbx)) { /* Error: bad element type. */ + int32_t idx = (int32_t)(intptr_t)sb->w; + cTValue *o = lj_tab_getint(t, idx); + lj_err_callerv(L, LJ_ERR_TABCAT, + lj_obj_itypename[o ? itypemap(o) : ~LJ_TNIL], idx); + } + setstrV(L, L->top-1, lj_buf_str(L, sbx)); + lj_gc_check(L); + return 1; +} + +/* ------------------------------------------------------------------------ */ + +static void set2(lua_State *L, int i, int j) +{ + lua_rawseti(L, 1, i); + lua_rawseti(L, 1, j); +} + +static int sort_comp(lua_State *L, int a, int b) +{ + if (!lua_isnil(L, 2)) { /* function? */ + int res; + lua_pushvalue(L, 2); + lua_pushvalue(L, a-1); /* -1 to compensate function */ + lua_pushvalue(L, b-2); /* -2 to compensate function and `a' */ + lua_call(L, 2, 1); + res = lua_toboolean(L, -1); + lua_pop(L, 1); + return res; + } else { /* a < b? */ + return lua_lessthan(L, a, b); + } +} + +static void auxsort(lua_State *L, int l, int u) +{ + while (l < u) { /* for tail recursion */ + int i, j; + /* sort elements a[l], a[(l+u)/2] and a[u] */ + lua_rawgeti(L, 1, l); + lua_rawgeti(L, 1, u); + if (sort_comp(L, -1, -2)) /* a[u] < a[l]? */ + set2(L, l, u); /* swap a[l] - a[u] */ + else + lua_pop(L, 2); + if (u-l == 1) break; /* only 2 elements */ + i = (l+u)/2; + lua_rawgeti(L, 1, i); + lua_rawgeti(L, 1, l); + if (sort_comp(L, -2, -1)) { /* a[i]= P */ + while (lua_rawgeti(L, 1, ++i), sort_comp(L, -1, -2)) { + if (i>=u) lj_err_caller(L, LJ_ERR_TABSORT); + lua_pop(L, 1); /* remove a[i] */ + } + /* repeat --j until a[j] <= P */ + while (lua_rawgeti(L, 1, --j), sort_comp(L, -3, -1)) { + if (j<=l) lj_err_caller(L, LJ_ERR_TABSORT); + lua_pop(L, 1); /* remove a[j] */ + } + if (jbase+1)) + lj_lib_checkfunc(L, 2); + auxsort(L, 1, n); + return 0; +} + +#if LJ_52 +LJLIB_PUSH("n") +LJLIB_CF(table_pack) +{ + TValue *array, *base = L->base; + MSize i, n = (uint32_t)(L->top - base); + GCtab *t = lj_tab_new(L, n ? n+1 : 0, 1); + /* NOBARRIER: The table is new (marked white). */ + setintV(lj_tab_setstr(L, t, strV(lj_lib_upvalue(L, 1))), (int32_t)n); + for (array = tvref(t->array) + 1, i = 0; i < n; i++) + copyTV(L, &array[i], &base[i]); + settabV(L, base, t); + L->top = base+1; + lj_gc_check(L); + return 1; +} +#endif + +LJLIB_NOREG LJLIB_CF(table_new) LJLIB_REC(.) +{ + int32_t a = lj_lib_checkint(L, 1); + int32_t h = lj_lib_checkint(L, 2); + lua_createtable(L, a, h); + return 1; +} + +LJLIB_NOREG LJLIB_CF(table_clear) LJLIB_REC(.) +{ + lj_tab_clear(lj_lib_checktab(L, 1)); + return 0; +} + +static int luaopen_table_new(lua_State *L) +{ + return lj_lib_postreg(L, lj_cf_table_new, FF_table_new, "new"); +} + +static int luaopen_table_clear(lua_State *L) +{ + return lj_lib_postreg(L, lj_cf_table_clear, FF_table_clear, "clear"); +} + +/* ------------------------------------------------------------------------ */ + +#include "lj_libdef.h" + +LUALIB_API int luaopen_table(lua_State *L) +{ + LJ_LIB_REG(L, LUA_TABLIBNAME, table); +#if LJ_52 + lua_getglobal(L, "unpack"); + lua_setfield(L, -2, "unpack"); +#endif + lj_lib_prereg(L, LUA_TABLIBNAME ".new", luaopen_table_new, tabV(L->top-1)); + lj_lib_prereg(L, LUA_TABLIBNAME ".clear", luaopen_table_clear, tabV(L->top-1)); + return 1; +} + diff --git a/src/bluajit/lj_asm.c b/src/bluajit/lj_asm.c new file mode 100644 index 000000000..c31dd1dec --- /dev/null +++ b/src/bluajit/lj_asm.c @@ -0,0 +1,2631 @@ +/* +** IR assembler (SSA IR -> machine code). +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#define lj_asm_c +#define LUA_CORE + +#include "lj_obj.h" + +#if LJ_HASJIT + +#include "lj_gc.h" +#include "lj_buf.h" +#include "lj_str.h" +#include "lj_tab.h" +#include "lj_frame.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#endif +#include "lj_ir.h" +#include "lj_jit.h" +#include "lj_ircall.h" +#include "lj_iropt.h" +#include "lj_mcode.h" +#include "lj_trace.h" +#include "lj_snap.h" +#include "lj_asm.h" +#include "lj_dispatch.h" +#include "lj_vm.h" +#include "lj_target.h" +#include "lj_prng.h" + +#ifdef LUA_USE_ASSERT +#include +#endif + +/* -- Assembler state and common macros ----------------------------------- */ + +/* Assembler state. */ +typedef struct ASMState { + RegCost cost[RID_MAX]; /* Reference and blended allocation cost for regs. */ + + MCode *mcp; /* Current MCode pointer (grows down). */ + MCode *mclim; /* Lower limit for MCode memory + red zone. */ +#ifdef LUA_USE_ASSERT + MCode *mcp_prev; /* Red zone overflow check. */ +#endif + + IRIns *ir; /* Copy of pointer to IR instructions/constants. */ + jit_State *J; /* JIT compiler state. */ + +#if LJ_TARGET_X86ORX64 + x86ModRM mrm; /* Fused x86 address operand. */ +#endif + + RegSet freeset; /* Set of free registers. */ + RegSet modset; /* Set of registers modified inside the loop. */ + RegSet weakset; /* Set of weakly referenced registers. */ + RegSet phiset; /* Set of PHI registers. */ + + uint32_t flags; /* Copy of JIT compiler flags. */ + int loopinv; /* Loop branch inversion (0:no, 1:yes, 2:yes+CC_P). */ + + int32_t evenspill; /* Next even spill slot. */ + int32_t oddspill; /* Next odd spill slot (or 0). */ + + IRRef curins; /* Reference of current instruction. */ + IRRef stopins; /* Stop assembly before hitting this instruction. */ + IRRef orignins; /* Original T->nins. */ + + IRRef snapref; /* Current snapshot is active after this reference. */ + IRRef snaprename; /* Rename highwater mark for snapshot check. */ + SnapNo snapno; /* Current snapshot number. */ + SnapNo loopsnapno; /* Loop snapshot number. */ + int snapalloc; /* Current snapshot needs allocation. */ + BloomFilter snapfilt1, snapfilt2; /* Filled with snapshot refs. */ + + IRRef fuseref; /* Fusion limit (loopref, 0 or FUSE_DISABLED). */ + IRRef sectref; /* Section base reference (loopref or 0). */ + IRRef loopref; /* Reference of LOOP instruction (or 0). */ + + BCReg topslot; /* Number of slots for stack check (unless 0). */ + int32_t gcsteps; /* Accumulated number of GC steps (per section). */ + + GCtrace *T; /* Trace to assemble. */ + GCtrace *parent; /* Parent trace (or NULL). */ + + MCode *mcbot; /* Bottom of reserved MCode. */ + MCode *mctop; /* Top of generated MCode. */ + MCode *mctoporig; /* Original top of generated MCode. */ + MCode *mcloop; /* Pointer to loop MCode (or NULL). */ + MCode *invmcp; /* Points to invertible loop branch (or NULL). */ + MCode *flagmcp; /* Pending opportunity to merge flag setting ins. */ + MCode *realign; /* Realign loop if not NULL. */ + +#ifdef LUAJIT_RANDOM_RA + /* Randomize register allocation. OK for fuzz testing, not for production. */ + uint64_t prngbits; + PRNGState prngstate; +#endif + +#ifdef RID_NUM_KREF + intptr_t krefk[RID_NUM_KREF]; +#endif + IRRef1 phireg[RID_MAX]; /* PHI register references. */ + uint16_t parentmap[LJ_MAX_JSLOTS]; /* Parent instruction to RegSP map. */ +} ASMState; + +#ifdef LUA_USE_ASSERT +#define lj_assertA(c, ...) lj_assertG_(J2G(as->J), (c), __VA_ARGS__) +#else +#define lj_assertA(c, ...) ((void)as) +#endif + +#define IR(ref) (&as->ir[(ref)]) + +#define ASMREF_TMP1 REF_TRUE /* Temp. register. */ +#define ASMREF_TMP2 REF_FALSE /* Temp. register. */ +#define ASMREF_L REF_NIL /* Stores register for L. */ + +/* Check for variant to invariant references. */ +#define iscrossref(as, ref) ((ref) < as->sectref) + +/* Inhibit memory op fusion from variant to invariant references. */ +#define FUSE_DISABLED (~(IRRef)0) +#define mayfuse(as, ref) ((ref) > as->fuseref) +#define neverfuse(as) (as->fuseref == FUSE_DISABLED) +#define canfuse(as, ir) (!neverfuse(as) && !irt_isphi((ir)->t)) +#define opisfusableload(o) \ + ((o) == IR_ALOAD || (o) == IR_HLOAD || (o) == IR_ULOAD || \ + (o) == IR_FLOAD || (o) == IR_XLOAD || (o) == IR_SLOAD || (o) == IR_VLOAD) + +/* Sparse limit checks using a red zone before the actual limit. */ +#define MCLIM_REDZONE 64 + +static LJ_NORET LJ_NOINLINE void asm_mclimit(ASMState *as) +{ + lj_mcode_limiterr(as->J, (size_t)(as->mctop - as->mcp + 4*MCLIM_REDZONE)); +} + +static LJ_AINLINE void checkmclim(ASMState *as) +{ +#ifdef LUA_USE_ASSERT + if (as->mcp + MCLIM_REDZONE < as->mcp_prev) { + IRIns *ir = IR(as->curins+1); + lj_assertA(0, "red zone overflow: %p IR %04d %02d %04d %04d\n", as->mcp, + as->curins+1-REF_BIAS, ir->o, ir->op1-REF_BIAS, ir->op2-REF_BIAS); + } +#endif + if (LJ_UNLIKELY(as->mcp < as->mclim)) asm_mclimit(as); +#ifdef LUA_USE_ASSERT + as->mcp_prev = as->mcp; +#endif +} + +#ifdef RID_NUM_KREF +#define ra_iskref(ref) ((ref) < RID_NUM_KREF) +#define ra_krefreg(ref) ((Reg)(RID_MIN_KREF + (Reg)(ref))) +#define ra_krefk(as, ref) (as->krefk[(ref)]) + +static LJ_AINLINE void ra_setkref(ASMState *as, Reg r, intptr_t k) +{ + IRRef ref = (IRRef)(r - RID_MIN_KREF); + as->krefk[ref] = k; + as->cost[r] = REGCOST(ref, ref); +} + +#else +#define ra_iskref(ref) 0 +#define ra_krefreg(ref) RID_MIN_GPR +#define ra_krefk(as, ref) 0 +#endif + +/* Arch-specific field offsets. */ +static const uint8_t field_ofs[IRFL__MAX+1] = { +#define FLOFS(name, ofs) (uint8_t)(ofs), +IRFLDEF(FLOFS) +#undef FLOFS + 0 +}; + +#ifdef LUAJIT_RANDOM_RA +/* Return a fixed number of random bits from the local PRNG state. */ +static uint32_t ra_random_bits(ASMState *as, uint32_t nbits) { + uint64_t b = as->prngbits; + uint32_t res = (1u << nbits) - 1u; + if (b <= res) b = lj_prng_u64(&as->prngstate) | (1ull << 63); + res &= (uint32_t)b; + as->prngbits = b >> nbits; + return res; +} + +/* Pick a random register from a register set. */ +static Reg rset_pickrandom(ASMState *as, RegSet rs) +{ + Reg r = rset_pickbot_(rs); + rs >>= r; + if (rs > 1) { /* More than one bit set? */ + while (1) { + /* We need to sample max. the GPR or FPR half of the set. */ + uint32_t d = ra_random_bits(as, RSET_BITS-1); + if ((rs >> d) & 1) { + r += d; + break; + } + } + } + return r; +} +#define rset_picktop(rs) rset_pickrandom(as, rs) +#define rset_pickbot(rs) rset_pickrandom(as, rs) +#else +#define rset_picktop(rs) rset_picktop_(rs) +#define rset_pickbot(rs) rset_pickbot_(rs) +#endif + +/* -- Target-specific instruction emitter --------------------------------- */ + +#if LJ_TARGET_X86ORX64 +#include "lj_emit_x86.h" +#elif LJ_TARGET_ARM +#include "lj_emit_arm.h" +#elif LJ_TARGET_ARM64 +#include "lj_emit_arm64.h" +#elif LJ_TARGET_PPC +#include "lj_emit_ppc.h" +#elif LJ_TARGET_MIPS +#include "lj_emit_mips.h" +#else +#error "Missing instruction emitter for target CPU" +#endif + +/* Generic load/store of register from/to stack slot. */ +#define emit_spload(as, ir, r, ofs) \ + emit_loadofs(as, ir, (r), RID_SP, (ofs)) +#define emit_spstore(as, ir, r, ofs) \ + emit_storeofs(as, ir, (r), RID_SP, (ofs)) + +/* -- Register allocator debugging ---------------------------------------- */ + +/* #define LUAJIT_DEBUG_RA */ + +#ifdef LUAJIT_DEBUG_RA + +#include +#include + +#define RIDNAME(name) #name, +static const char *const ra_regname[] = { + GPRDEF(RIDNAME) + FPRDEF(RIDNAME) + VRIDDEF(RIDNAME) + NULL +}; +#undef RIDNAME + +static char ra_dbg_buf[65536]; +static char *ra_dbg_p; +static char *ra_dbg_merge; +static MCode *ra_dbg_mcp; + +static void ra_dstart(void) +{ + ra_dbg_p = ra_dbg_buf; + ra_dbg_merge = NULL; + ra_dbg_mcp = NULL; +} + +static void ra_dflush(void) +{ + fwrite(ra_dbg_buf, 1, (size_t)(ra_dbg_p-ra_dbg_buf), stdout); + ra_dstart(); +} + +static void ra_dprintf(ASMState *as, const char *fmt, ...) +{ + char *p; + va_list argp; + va_start(argp, fmt); + p = ra_dbg_mcp == as->mcp ? ra_dbg_merge : ra_dbg_p; + ra_dbg_mcp = NULL; + p += sprintf(p, "%08x \e[36m%04d ", (uintptr_t)as->mcp, as->curins-REF_BIAS); + for (;;) { + const char *e = strchr(fmt, '$'); + if (e == NULL) break; + memcpy(p, fmt, (size_t)(e-fmt)); + p += e-fmt; + if (e[1] == 'r') { + Reg r = va_arg(argp, Reg) & RID_MASK; + if (r <= RID_MAX) { + const char *q; + for (q = ra_regname[r]; *q; q++) + *p++ = *q >= 'A' && *q <= 'Z' ? *q + 0x20 : *q; + } else { + *p++ = '?'; + lj_assertA(0, "bad register %d for debug format \"%s\"", r, fmt); + } + } else if (e[1] == 'f' || e[1] == 'i') { + IRRef ref; + if (e[1] == 'f') + ref = va_arg(argp, IRRef); + else + ref = va_arg(argp, IRIns *) - as->ir; + if (ref >= REF_BIAS) + p += sprintf(p, "%04d", ref - REF_BIAS); + else + p += sprintf(p, "K%03d", REF_BIAS - ref); + } else if (e[1] == 's') { + uint32_t slot = va_arg(argp, uint32_t); + p += sprintf(p, "[sp+0x%x]", sps_scale(slot)); + } else if (e[1] == 'x') { + p += sprintf(p, "%08x", va_arg(argp, int32_t)); + } else { + lj_assertA(0, "bad debug format code"); + } + fmt = e+2; + } + va_end(argp); + while (*fmt) + *p++ = *fmt++; + *p++ = '\e'; *p++ = '['; *p++ = 'm'; *p++ = '\n'; + if (p > ra_dbg_buf+sizeof(ra_dbg_buf)-256) { + fwrite(ra_dbg_buf, 1, (size_t)(p-ra_dbg_buf), stdout); + p = ra_dbg_buf; + } + ra_dbg_p = p; +} + +#define RA_DBG_START() ra_dstart() +#define RA_DBG_FLUSH() ra_dflush() +#define RA_DBG_REF() \ + do { char *_p = ra_dbg_p; ra_dprintf(as, ""); \ + ra_dbg_merge = _p; ra_dbg_mcp = as->mcp; } while (0) +#define RA_DBGX(x) ra_dprintf x + +#else +#define RA_DBG_START() ((void)0) +#define RA_DBG_FLUSH() ((void)0) +#define RA_DBG_REF() ((void)0) +#define RA_DBGX(x) ((void)0) +#endif + +/* -- Register allocator -------------------------------------------------- */ + +#define ra_free(as, r) rset_set(as->freeset, (r)) +#define ra_modified(as, r) rset_set(as->modset, (r)) +#define ra_weak(as, r) rset_set(as->weakset, (r)) +#define ra_noweak(as, r) rset_clear(as->weakset, (r)) + +#define ra_used(ir) (ra_hasreg((ir)->r) || ra_hasspill((ir)->s)) + +/* Setup register allocator. */ +static void ra_setup(ASMState *as) +{ + Reg r; + /* Initially all regs (except the stack pointer) are free for use. */ + as->freeset = RSET_INIT; + as->modset = RSET_EMPTY; + as->weakset = RSET_EMPTY; + as->phiset = RSET_EMPTY; + memset(as->phireg, 0, sizeof(as->phireg)); + for (r = RID_MIN_GPR; r < RID_MAX; r++) + as->cost[r] = REGCOST(~0u, 0u); +} + +/* Rematerialize constants. */ +static Reg ra_rematk(ASMState *as, IRRef ref) +{ + IRIns *ir; + Reg r; + if (ra_iskref(ref)) { + r = ra_krefreg(ref); + lj_assertA(!rset_test(as->freeset, r), "rematk of free reg %d", r); + ra_free(as, r); + ra_modified(as, r); +#if LJ_64 + emit_loadu64(as, r, ra_krefk(as, ref)); +#else + emit_loadi(as, r, ra_krefk(as, ref)); +#endif + return r; + } + ir = IR(ref); + r = ir->r; + lj_assertA(ra_hasreg(r), "rematk of K%03d has no reg", REF_BIAS - ref); + lj_assertA(!ra_hasspill(ir->s), + "rematk of K%03d has spill slot [%x]", REF_BIAS - ref, ir->s); + ra_free(as, r); + ra_modified(as, r); + ir->r = RID_INIT; /* Do not keep any hint. */ + RA_DBGX((as, "remat $i $r", ir, r)); +#if !LJ_SOFTFP32 + if (ir->o == IR_KNUM) { + emit_loadk64(as, r, ir); + } else +#endif + if (emit_canremat(REF_BASE) && ir->o == IR_BASE) { + ra_sethint(ir->r, RID_BASE); /* Restore BASE register hint. */ + emit_getgl(as, r, jit_base); + } else if (emit_canremat(ASMREF_L) && ir->o == IR_KPRI) { + /* REF_NIL stores ASMREF_L register. */ + lj_assertA(irt_isnil(ir->t), "rematk of bad ASMREF_L"); + emit_getgl(as, r, cur_L); +#if LJ_64 + } else if (ir->o == IR_KINT64) { + emit_loadu64(as, r, ir_kint64(ir)->u64); +#if LJ_GC64 + } else if (ir->o == IR_KGC) { + emit_loadu64(as, r, (uintptr_t)ir_kgc(ir)); + } else if (ir->o == IR_KPTR || ir->o == IR_KKPTR) { + emit_loadu64(as, r, (uintptr_t)ir_kptr(ir)); +#endif +#endif + } else { + lj_assertA(ir->o == IR_KINT || ir->o == IR_KGC || + ir->o == IR_KPTR || ir->o == IR_KKPTR || ir->o == IR_KNULL, + "rematk of bad IR op %d", ir->o); + emit_loadi(as, r, ir->i); + } + return r; +} + +/* Force a spill. Allocate a new spill slot if needed. */ +static int32_t ra_spill(ASMState *as, IRIns *ir) +{ + int32_t slot = ir->s; + lj_assertA(ir >= as->ir + REF_TRUE, + "spill of K%03d", REF_BIAS - (int)(ir - as->ir)); + if (!ra_hasspill(slot)) { + if (irt_is64(ir->t)) { + slot = as->evenspill; + as->evenspill += 2; + } else if (as->oddspill) { + slot = as->oddspill; + as->oddspill = 0; + } else { + slot = as->evenspill; + as->oddspill = slot+1; + as->evenspill += 2; + } + if (as->evenspill > 256) + lj_trace_err(as->J, LJ_TRERR_SPILLOV); + ir->s = (uint8_t)slot; + } + return sps_scale(slot); +} + +/* Release the temporarily allocated register in ASMREF_TMP1/ASMREF_TMP2. */ +static Reg ra_releasetmp(ASMState *as, IRRef ref) +{ + IRIns *ir = IR(ref); + Reg r = ir->r; + lj_assertA(ra_hasreg(r), "release of TMP%d has no reg", ref-ASMREF_TMP1+1); + lj_assertA(!ra_hasspill(ir->s), + "release of TMP%d has spill slot [%x]", ref-ASMREF_TMP1+1, ir->s); + ra_free(as, r); + ra_modified(as, r); + ir->r = RID_INIT; + return r; +} + +/* Restore a register (marked as free). Rematerialize or force a spill. */ +static Reg ra_restore(ASMState *as, IRRef ref) +{ + if (emit_canremat(ref)) { + return ra_rematk(as, ref); + } else { + IRIns *ir = IR(ref); + int32_t ofs = ra_spill(as, ir); /* Force a spill slot. */ + Reg r = ir->r; + lj_assertA(ra_hasreg(r), "restore of IR %04d has no reg", ref - REF_BIAS); + ra_sethint(ir->r, r); /* Keep hint. */ + ra_free(as, r); + if (!rset_test(as->weakset, r)) { /* Only restore non-weak references. */ + ra_modified(as, r); + RA_DBGX((as, "restore $i $r", ir, r)); + emit_spload(as, ir, r, ofs); + } + return r; + } +} + +/* Save a register to a spill slot. */ +static void ra_save(ASMState *as, IRIns *ir, Reg r) +{ + RA_DBGX((as, "save $i $r", ir, r)); + emit_spstore(as, ir, r, sps_scale(ir->s)); +} + +#define MINCOST(name) \ + if (rset_test(RSET_ALL, RID_##name) && \ + LJ_LIKELY(allow&RID2RSET(RID_##name)) && as->cost[RID_##name] < cost) \ + cost = as->cost[RID_##name]; + +/* Evict the register with the lowest cost, forcing a restore. */ +static Reg ra_evict(ASMState *as, RegSet allow) +{ + IRRef ref; + RegCost cost = ~(RegCost)0; + lj_assertA(allow != RSET_EMPTY, "evict from empty set"); + if (RID_NUM_FPR == 0 || allow < RID2RSET(RID_MAX_GPR)) { + GPRDEF(MINCOST) + } else { + FPRDEF(MINCOST) + } + ref = regcost_ref(cost); + lj_assertA(ra_iskref(ref) || (ref >= as->T->nk && ref < as->T->nins), + "evict of out-of-range IR %04d", ref - REF_BIAS); + /* Preferably pick any weak ref instead of a non-weak, non-const ref. */ + if (!irref_isk(ref) && (as->weakset & allow)) { + IRIns *ir = IR(ref); + if (!rset_test(as->weakset, ir->r)) + ref = regcost_ref(as->cost[rset_pickbot((as->weakset & allow))]); + } + return ra_restore(as, ref); +} + +/* Pick any register (marked as free). Evict on-demand. */ +static Reg ra_pick(ASMState *as, RegSet allow) +{ + RegSet pick = as->freeset & allow; + if (!pick) + return ra_evict(as, allow); + else + return rset_picktop(pick); +} + +/* Get a scratch register (marked as free). */ +static Reg ra_scratch(ASMState *as, RegSet allow) +{ + Reg r = ra_pick(as, allow); + ra_modified(as, r); + RA_DBGX((as, "scratch $r", r)); + return r; +} + +/* Evict all registers from a set (if not free). */ +static void ra_evictset(ASMState *as, RegSet drop) +{ + RegSet work; + as->modset |= drop; +#if !LJ_SOFTFP + work = (drop & ~as->freeset) & RSET_FPR; + while (work) { + Reg r = rset_pickbot(work); + ra_restore(as, regcost_ref(as->cost[r])); + rset_clear(work, r); + checkmclim(as); + } +#endif + work = (drop & ~as->freeset); + while (work) { + Reg r = rset_pickbot(work); + ra_restore(as, regcost_ref(as->cost[r])); + rset_clear(work, r); + checkmclim(as); + } +} + +/* Evict (rematerialize) all registers allocated to constants. */ +static void ra_evictk(ASMState *as) +{ + RegSet work; +#if !LJ_SOFTFP + work = ~as->freeset & RSET_FPR; + while (work) { + Reg r = rset_pickbot(work); + IRRef ref = regcost_ref(as->cost[r]); + if (emit_canremat(ref) && irref_isk(ref)) { + ra_rematk(as, ref); + checkmclim(as); + } + rset_clear(work, r); + } +#endif + work = ~as->freeset & RSET_GPR; + while (work) { + Reg r = rset_pickbot(work); + IRRef ref = regcost_ref(as->cost[r]); + if (emit_canremat(ref) && irref_isk(ref)) { + ra_rematk(as, ref); + checkmclim(as); + } + rset_clear(work, r); + } +} + +#ifdef RID_NUM_KREF +/* Allocate a register for a constant. */ +static Reg ra_allock(ASMState *as, intptr_t k, RegSet allow) +{ + /* First try to find a register which already holds the same constant. */ + RegSet pick, work = ~as->freeset & RSET_GPR; + Reg r; + while (work) { + IRRef ref; + r = rset_pickbot(work); + ref = regcost_ref(as->cost[r]); +#if LJ_64 + if (ref < ASMREF_L) { + if (ra_iskref(ref)) { + if (k == ra_krefk(as, ref)) + return r; + } else { + IRIns *ir = IR(ref); + if ((ir->o == IR_KINT64 && k == (int64_t)ir_kint64(ir)->u64) || +#if LJ_GC64 +#if LJ_TARGET_ARM64 + (ir->o == IR_KINT && (uint64_t)k == (uint32_t)ir->i) || +#else + (ir->o == IR_KINT && k == ir->i) || +#endif + (ir->o == IR_KGC && k == (intptr_t)ir_kgc(ir)) || + ((ir->o == IR_KPTR || ir->o == IR_KKPTR) && + k == (intptr_t)ir_kptr(ir)) +#else + (ir->o != IR_KINT64 && k == ir->i) +#endif + ) + return r; + } + } +#else + if (ref < ASMREF_L && + k == (ra_iskref(ref) ? ra_krefk(as, ref) : IR(ref)->i)) + return r; +#endif + rset_clear(work, r); + } + pick = as->freeset & allow; + if (pick) { + /* Constants should preferably get unmodified registers. */ + if ((pick & ~as->modset)) + pick &= ~as->modset; + r = rset_pickbot(pick); /* Reduce conflicts with inverse allocation. */ + } else { + r = ra_evict(as, allow); + } + RA_DBGX((as, "allock $x $r", k, r)); + ra_setkref(as, r, k); + rset_clear(as->freeset, r); + ra_noweak(as, r); + return r; +} + +/* Allocate a specific register for a constant. */ +static void ra_allockreg(ASMState *as, intptr_t k, Reg r) +{ + Reg kr = ra_allock(as, k, RID2RSET(r)); + if (kr != r) { + IRIns irdummy; + irdummy.t.irt = IRT_INT; + ra_scratch(as, RID2RSET(r)); + emit_movrr(as, &irdummy, r, kr); + } +} +#else +#define ra_allockreg(as, k, r) emit_loadi(as, (r), (k)) +#endif + +/* Allocate a register for ref from the allowed set of registers. +** Note: this function assumes the ref does NOT have a register yet! +** Picks an optimal register, sets the cost and marks the register as non-free. +*/ +static Reg ra_allocref(ASMState *as, IRRef ref, RegSet allow) +{ + IRIns *ir = IR(ref); + RegSet pick = as->freeset & allow; + Reg r; + lj_assertA(ra_noreg(ir->r), + "IR %04d already has reg %d", ref - REF_BIAS, ir->r); + if (pick) { + /* First check register hint from propagation or PHI. */ + if (ra_hashint(ir->r)) { + r = ra_gethint(ir->r); + if (rset_test(pick, r)) /* Use hint register if possible. */ + goto found; + /* Rematerialization is cheaper than missing a hint. */ + if (rset_test(allow, r) && emit_canremat(regcost_ref(as->cost[r]))) { + ra_rematk(as, regcost_ref(as->cost[r])); + goto found; + } + RA_DBGX((as, "hintmiss $f $r", ref, r)); + } + /* Invariants should preferably get unmodified registers. */ + if (ref < as->loopref && !irt_isphi(ir->t)) { + if ((pick & ~as->modset)) + pick &= ~as->modset; + r = rset_pickbot(pick); /* Reduce conflicts with inverse allocation. */ + } else { + /* We've got plenty of regs, so get callee-save regs if possible. */ + if (RID_NUM_GPR > 8 && (pick & ~RSET_SCRATCH)) + pick &= ~RSET_SCRATCH; + r = rset_picktop(pick); + } + } else { + r = ra_evict(as, allow); + } +found: + RA_DBGX((as, "alloc $f $r", ref, r)); + ir->r = (uint8_t)r; + rset_clear(as->freeset, r); + ra_noweak(as, r); + as->cost[r] = REGCOST_REF_T(ref, irt_t(ir->t)); + return r; +} + +/* Allocate a register on-demand. */ +static Reg ra_alloc1(ASMState *as, IRRef ref, RegSet allow) +{ + Reg r = IR(ref)->r; + /* Note: allow is ignored if the register is already allocated. */ + if (ra_noreg(r)) r = ra_allocref(as, ref, allow); + ra_noweak(as, r); + return r; +} + +/* Add a register rename to the IR. */ +static void ra_addrename(ASMState *as, Reg down, IRRef ref, SnapNo snapno) +{ + IRRef ren; + lj_ir_set(as->J, IRT(IR_RENAME, IRT_NIL), ref, snapno); + ren = tref_ref(lj_ir_emit(as->J)); + as->J->cur.ir[ren].r = (uint8_t)down; + as->J->cur.ir[ren].s = SPS_NONE; +} + +/* Rename register allocation and emit move. */ +static void ra_rename(ASMState *as, Reg down, Reg up) +{ + IRRef ref = regcost_ref(as->cost[up] = as->cost[down]); + IRIns *ir = IR(ref); + ir->r = (uint8_t)up; + as->cost[down] = 0; + lj_assertA((down < RID_MAX_GPR) == (up < RID_MAX_GPR), + "rename between GPR/FPR %d and %d", down, up); + lj_assertA(!rset_test(as->freeset, down), "rename from free reg %d", down); + lj_assertA(rset_test(as->freeset, up), "rename to non-free reg %d", up); + ra_free(as, down); /* 'down' is free ... */ + ra_modified(as, down); + rset_clear(as->freeset, up); /* ... and 'up' is now allocated. */ + ra_noweak(as, up); + RA_DBGX((as, "rename $f $r $r", regcost_ref(as->cost[up]), down, up)); + emit_movrr(as, ir, down, up); /* Backwards codegen needs inverse move. */ + if (!ra_hasspill(IR(ref)->s)) { /* Add the rename to the IR. */ + /* + ** The rename is effective at the subsequent (already emitted) exit + ** branch. This is for the current snapshot (as->snapno). Except if we + ** haven't yet allocated any refs for the snapshot (as->snapalloc == 1), + ** then it belongs to the next snapshot. + ** See also the discussion at asm_snap_checkrename(). + */ + ra_addrename(as, down, ref, as->snapno + as->snapalloc); + } +} + +/* Pick a destination register (marked as free). +** Caveat: allow is ignored if there's already a destination register. +** Use ra_destreg() to get a specific register. +*/ +static Reg ra_dest(ASMState *as, IRIns *ir, RegSet allow) +{ + Reg dest = ir->r; + if (ra_hasreg(dest)) { + ra_free(as, dest); + ra_modified(as, dest); + } else { + if (ra_hashint(dest) && rset_test((as->freeset&allow), ra_gethint(dest))) { + dest = ra_gethint(dest); + ra_modified(as, dest); + RA_DBGX((as, "dest $r", dest)); + } else { + dest = ra_scratch(as, allow); + } + ir->r = dest; + } + if (LJ_UNLIKELY(ra_hasspill(ir->s))) ra_save(as, ir, dest); + return dest; +} + +/* Force a specific destination register (marked as free). */ +static void ra_destreg(ASMState *as, IRIns *ir, Reg r) +{ + Reg dest = ra_dest(as, ir, RID2RSET(r)); + if (dest != r) { + lj_assertA(rset_test(as->freeset, r), "dest reg %d is not free", r); + ra_modified(as, r); + emit_movrr(as, ir, dest, r); + } +} + +#if LJ_TARGET_X86ORX64 +/* Propagate dest register to left reference. Emit moves as needed. +** This is a required fixup step for all 2-operand machine instructions. +*/ +static void ra_left(ASMState *as, Reg dest, IRRef lref) +{ + IRIns *ir = IR(lref); + Reg left = ir->r; + if (ra_noreg(left)) { + if (irref_isk(lref)) { + if (ir->o == IR_KNUM) { + /* FP remat needs a load except for +0. Still better than eviction. */ + if (tvispzero(ir_knum(ir)) || !(as->freeset & RSET_FPR)) { + emit_loadk64(as, dest, ir); + return; + } +#if LJ_64 + } else if (ir->o == IR_KINT64) { + emit_loadk64(as, dest, ir); + return; +#if LJ_GC64 + } else if (ir->o == IR_KGC || ir->o == IR_KPTR || ir->o == IR_KKPTR) { + emit_loadk64(as, dest, ir); + return; +#endif +#endif + } else if (ir->o != IR_KPRI) { + lj_assertA(ir->o == IR_KINT || ir->o == IR_KGC || + ir->o == IR_KPTR || ir->o == IR_KKPTR || ir->o == IR_KNULL, + "K%03d has bad IR op %d", REF_BIAS - lref, ir->o); + emit_loadi(as, dest, ir->i); + return; + } + } + if (!ra_hashint(left) && !iscrossref(as, lref)) + ra_sethint(ir->r, dest); /* Propagate register hint. */ + left = ra_allocref(as, lref, dest < RID_MAX_GPR ? RSET_GPR : RSET_FPR); + } + ra_noweak(as, left); + /* Move needed for true 3-operand instruction: y=a+b ==> y=a; y+=b. */ + if (dest != left) { + /* Use register renaming if dest is the PHI reg. */ + if (irt_isphi(ir->t) && as->phireg[dest] == lref) { + ra_modified(as, left); + ra_rename(as, left, dest); + } else { + emit_movrr(as, ir, dest, left); + } + } +} +#else +/* Similar to ra_left, except we override any hints. */ +static void ra_leftov(ASMState *as, Reg dest, IRRef lref) +{ + IRIns *ir = IR(lref); + Reg left = ir->r; + if (ra_noreg(left)) { + ra_sethint(ir->r, dest); /* Propagate register hint. */ + left = ra_allocref(as, lref, + (LJ_SOFTFP || dest < RID_MAX_GPR) ? RSET_GPR : RSET_FPR); + } + ra_noweak(as, left); + if (dest != left) { + /* Use register renaming if dest is the PHI reg. */ + if (irt_isphi(ir->t) && as->phireg[dest] == lref) { + ra_modified(as, left); + ra_rename(as, left, dest); + } else { + emit_movrr(as, ir, dest, left); + } + } +} +#endif + +/* Force a RID_RETLO/RID_RETHI destination register pair (marked as free). */ +static void ra_destpair(ASMState *as, IRIns *ir) +{ + Reg destlo = ir->r, desthi = (ir+1)->r; + IRIns *irx = (LJ_64 && !irt_is64(ir->t)) ? ir+1 : ir; + /* First spill unrelated refs blocking the destination registers. */ + if (!rset_test(as->freeset, RID_RETLO) && + destlo != RID_RETLO && desthi != RID_RETLO) + ra_restore(as, regcost_ref(as->cost[RID_RETLO])); + if (!rset_test(as->freeset, RID_RETHI) && + destlo != RID_RETHI && desthi != RID_RETHI) + ra_restore(as, regcost_ref(as->cost[RID_RETHI])); + /* Next free the destination registers (if any). */ + if (ra_hasreg(destlo)) { + ra_free(as, destlo); + ra_modified(as, destlo); + } else { + destlo = RID_RETLO; + } + if (ra_hasreg(desthi)) { + ra_free(as, desthi); + ra_modified(as, desthi); + } else { + desthi = RID_RETHI; + } + /* Check for conflicts and shuffle the registers as needed. */ + if (destlo == RID_RETHI) { + if (desthi == RID_RETLO) { +#if LJ_TARGET_X86ORX64 + *--as->mcp = XI_XCHGa + RID_RETHI; + if (LJ_64 && irt_is64(irx->t)) *--as->mcp = 0x48; +#else + emit_movrr(as, irx, RID_RETHI, RID_TMP); + emit_movrr(as, irx, RID_RETLO, RID_RETHI); + emit_movrr(as, irx, RID_TMP, RID_RETLO); +#endif + } else { + emit_movrr(as, irx, RID_RETHI, RID_RETLO); + if (desthi != RID_RETHI) emit_movrr(as, irx, desthi, RID_RETHI); + } + } else if (desthi == RID_RETLO) { + emit_movrr(as, irx, RID_RETLO, RID_RETHI); + if (destlo != RID_RETLO) emit_movrr(as, irx, destlo, RID_RETLO); + } else { + if (desthi != RID_RETHI) emit_movrr(as, irx, desthi, RID_RETHI); + if (destlo != RID_RETLO) emit_movrr(as, irx, destlo, RID_RETLO); + } + /* Restore spill slots (if any). */ + if (ra_hasspill((ir+1)->s)) ra_save(as, ir+1, RID_RETHI); + if (ra_hasspill(ir->s)) ra_save(as, ir, RID_RETLO); +} + +/* -- Snapshot handling --------- ----------------------------------------- */ + +/* Can we rematerialize a KNUM instead of forcing a spill? */ +static int asm_snap_canremat(ASMState *as) +{ + Reg r; + for (r = RID_MIN_FPR; r < RID_MAX_FPR; r++) + if (irref_isk(regcost_ref(as->cost[r]))) + return 1; + return 0; +} + +/* Check whether a sunk store corresponds to an allocation. */ +static int asm_sunk_store(ASMState *as, IRIns *ira, IRIns *irs) +{ + if (irs->s == 255) { + if (irs->o == IR_ASTORE || irs->o == IR_HSTORE || + irs->o == IR_FSTORE || irs->o == IR_XSTORE) { + IRIns *irk = IR(irs->op1); + if (irk->o == IR_AREF || irk->o == IR_HREFK) + irk = IR(irk->op1); + return (IR(irk->op1) == ira); + } + return 0; + } else { + return (ira + irs->s == irs); /* Quick check. */ + } +} + +/* Allocate register or spill slot for a ref that escapes to a snapshot. */ +static void asm_snap_alloc1(ASMState *as, IRRef ref) +{ + IRIns *ir = IR(ref); + if (!irref_isk(ref)) { + bloomset(as->snapfilt1, ref); + bloomset(as->snapfilt2, hashrot(ref, ref + HASH_BIAS)); + if (ra_used(ir)) return; + if (ir->r == RID_SINK || ir->r == RID_SUNK) { + ir->r = RID_SUNK; +#if LJ_HASFFI + if (ir->o == IR_CNEWI) { /* Allocate CNEWI value. */ + asm_snap_alloc1(as, ir->op2); + if (LJ_32 && (ir+1)->o == IR_HIOP) + asm_snap_alloc1(as, (ir+1)->op2); + } else +#endif + { /* Allocate stored values for TNEW, TDUP and CNEW. */ + IRIns *irs; + lj_assertA(ir->o == IR_TNEW || ir->o == IR_TDUP || ir->o == IR_CNEW, + "sink of IR %04d has bad op %d", ref - REF_BIAS, ir->o); + for (irs = IR(as->snapref-1); irs > ir; irs--) + if (irs->r == RID_SINK && asm_sunk_store(as, ir, irs)) { + lj_assertA(irs->o == IR_ASTORE || irs->o == IR_HSTORE || + irs->o == IR_FSTORE || irs->o == IR_XSTORE, + "sunk store IR %04d has bad op %d", + (int)(irs - as->ir) - REF_BIAS, irs->o); + asm_snap_alloc1(as, irs->op2); + if (LJ_32 && (irs+1)->o == IR_HIOP) + asm_snap_alloc1(as, (irs+1)->op2); + } + } + } else { + RegSet allow; + if (ir->o == IR_CONV && ir->op2 == IRCONV_NUM_INT) { + IRIns *irc; + for (irc = IR(as->curins); irc > ir; irc--) + if ((irc->op1 == ref || irc->op2 == ref) && + !(irc->r == RID_SINK || irc->r == RID_SUNK)) + goto nosink; /* Don't sink conversion if result is used. */ + asm_snap_alloc1(as, ir->op1); + return; + } + nosink: + allow = (!LJ_SOFTFP && irt_isfp(ir->t)) ? RSET_FPR : RSET_GPR; + if ((as->freeset & allow) || + (allow == RSET_FPR && asm_snap_canremat(as))) { + /* Get a weak register if we have a free one or can rematerialize. */ + Reg r = ra_allocref(as, ref, allow); /* Allocate a register. */ + if (!irt_isphi(ir->t)) + ra_weak(as, r); /* But mark it as weakly referenced. */ + checkmclim(as); + RA_DBGX((as, "snapreg $f $r", ref, ir->r)); + } else { + ra_spill(as, ir); /* Otherwise force a spill slot. */ + RA_DBGX((as, "snapspill $f $s", ref, ir->s)); + } + } + } +} + +/* Allocate refs escaping to a snapshot. */ +static void asm_snap_alloc(ASMState *as, int snapno) +{ + SnapShot *snap = &as->T->snap[snapno]; + SnapEntry *map = &as->T->snapmap[snap->mapofs]; + MSize n, nent = snap->nent; + as->snapfilt1 = as->snapfilt2 = 0; + for (n = 0; n < nent; n++) { + SnapEntry sn = map[n]; + IRRef ref = snap_ref(sn); + if (!irref_isk(ref)) { + asm_snap_alloc1(as, ref); + if (LJ_SOFTFP && (sn & SNAP_SOFTFPNUM)) { + lj_assertA(irt_type(IR(ref+1)->t) == IRT_SOFTFP, + "snap %d[%d] points to bad SOFTFP IR %04d", + snapno, n, ref - REF_BIAS); + asm_snap_alloc1(as, ref+1); + } + } + } +} + +/* All guards for a snapshot use the same exitno. This is currently the +** same as the snapshot number. Since the exact origin of the exit cannot +** be determined, all guards for the same snapshot must exit with the same +** RegSP mapping. +** A renamed ref which has been used in a prior guard for the same snapshot +** would cause an inconsistency. The easy way out is to force a spill slot. +*/ +static int asm_snap_checkrename(ASMState *as, IRRef ren) +{ + if (bloomtest(as->snapfilt1, ren) && + bloomtest(as->snapfilt2, hashrot(ren, ren + HASH_BIAS))) { + IRIns *ir = IR(ren); + ra_spill(as, ir); /* Register renamed, so force a spill slot. */ + RA_DBGX((as, "snaprensp $f $s", ren, ir->s)); + return 1; /* Found. */ + } + return 0; /* Not found. */ +} + +/* Prepare snapshot for next guard or throwing instruction. */ +static void asm_snap_prep(ASMState *as) +{ + if (as->snapalloc) { + /* Alloc on first invocation for each snapshot. */ + as->snapalloc = 0; + asm_snap_alloc(as, as->snapno); + as->snaprename = as->T->nins; + } else { + /* Check any renames above the highwater mark. */ + for (; as->snaprename < as->T->nins; as->snaprename++) { + IRIns *ir = &as->T->ir[as->snaprename]; + if (asm_snap_checkrename(as, ir->op1)) + ir->op2 = REF_BIAS-1; /* Kill rename. */ + } + } +} + +/* Move to previous snapshot when we cross the current snapshot ref. */ +static void asm_snap_prev(ASMState *as) +{ + if (as->curins < as->snapref) { + uintptr_t ofs = (uintptr_t)(as->mctoporig - as->mcp); + if (ofs >= 0x10000) lj_trace_err(as->J, LJ_TRERR_MCODEOV); + do { + if (as->snapno == 0) return; + as->snapno--; + as->snapref = as->T->snap[as->snapno].ref; + as->T->snap[as->snapno].mcofs = (uint16_t)ofs; /* Remember mcode ofs. */ + } while (as->curins < as->snapref); /* May have no ins inbetween. */ + as->snapalloc = 1; + } +} + +/* Fixup snapshot mcode offsetst. */ +static void asm_snap_fixup_mcofs(ASMState *as) +{ + uint32_t sz = (uint32_t)(as->mctoporig - as->mcp); + SnapShot *snap = as->T->snap; + SnapNo i; + for (i = as->T->nsnap-1; i > 0; i--) { + /* Compute offset from mcode start and store in correct snapshot. */ + snap[i].mcofs = (uint16_t)(sz - snap[i-1].mcofs); + } + snap[0].mcofs = 0; +} + +/* -- Miscellaneous helpers ----------------------------------------------- */ + +/* Calculate stack adjustment. */ +static int32_t asm_stack_adjust(ASMState *as) +{ + if (as->evenspill <= SPS_FIXED) + return 0; + return sps_scale(sps_align(as->evenspill)); +} + +/* Must match with hash*() in lj_tab.c. */ +static uint32_t ir_khash(ASMState *as, IRIns *ir) +{ + uint32_t lo, hi; + UNUSED(as); + if (irt_isstr(ir->t)) { + return ir_kstr(ir)->sid; + } else if (irt_isnum(ir->t)) { + lo = ir_knum(ir)->u32.lo; + hi = ir_knum(ir)->u32.hi << 1; + } else if (irt_ispri(ir->t)) { + lj_assertA(!irt_isnil(ir->t), "hash of nil key"); + return irt_type(ir->t)-IRT_FALSE; + } else { + lj_assertA(irt_isgcv(ir->t), "hash of bad IR type %d", irt_type(ir->t)); + lo = u32ptr(ir_kgc(ir)); +#if LJ_GC64 + hi = (uint32_t)(u64ptr(ir_kgc(ir)) >> 32) | (irt_toitype(ir->t) << 15); +#else + hi = lo + HASH_BIAS; +#endif + } + return hashrot(lo, hi); +} + +/* -- Allocations --------------------------------------------------------- */ + +static void asm_gencall(ASMState *as, const CCallInfo *ci, IRRef *args); +static void asm_setupresult(ASMState *as, IRIns *ir, const CCallInfo *ci); + +static void asm_snew(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_str_new]; + IRRef args[3]; + asm_snap_prep(as); + args[0] = ASMREF_L; /* lua_State *L */ + args[1] = ir->op1; /* const char *str */ + args[2] = ir->op2; /* size_t len */ + as->gcsteps++; + asm_setupresult(as, ir, ci); /* GCstr * */ + asm_gencall(as, ci, args); +} + +static void asm_tnew(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_tab_new1]; + IRRef args[2]; + asm_snap_prep(as); + args[0] = ASMREF_L; /* lua_State *L */ + args[1] = ASMREF_TMP1; /* uint32_t ahsize */ + as->gcsteps++; + asm_setupresult(as, ir, ci); /* GCtab * */ + asm_gencall(as, ci, args); + ra_allockreg(as, ir->op1 | (ir->op2 << 24), ra_releasetmp(as, ASMREF_TMP1)); +} + +static void asm_tdup(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_tab_dup]; + IRRef args[2]; + asm_snap_prep(as); + args[0] = ASMREF_L; /* lua_State *L */ + args[1] = ir->op1; /* const GCtab *kt */ + as->gcsteps++; + asm_setupresult(as, ir, ci); /* GCtab * */ + asm_gencall(as, ci, args); +} + +static void asm_gc_check(ASMState *as); + +/* Explicit GC step. */ +static void asm_gcstep(ASMState *as, IRIns *ir) +{ + IRIns *ira; + for (ira = IR(as->stopins+1); ira < ir; ira++) + if ((ira->o == IR_TNEW || ira->o == IR_TDUP || + (LJ_HASFFI && (ira->o == IR_CNEW || ira->o == IR_CNEWI))) && + ra_used(ira)) + as->gcsteps++; + if (as->gcsteps) + asm_gc_check(as); + as->gcsteps = 0x80000000; /* Prevent implicit GC check further up. */ +} + +/* -- Buffer operations --------------------------------------------------- */ + +static void asm_tvptr(ASMState *as, Reg dest, IRRef ref, MSize mode); +#if LJ_HASBUFFER +static void asm_bufhdr_write(ASMState *as, Reg sb); +#endif + +static void asm_bufhdr(ASMState *as, IRIns *ir) +{ + Reg sb = ra_dest(as, ir, RSET_GPR); + switch (ir->op2) { + case IRBUFHDR_RESET: { + Reg tmp = ra_scratch(as, rset_exclude(RSET_GPR, sb)); + IRIns irbp; + irbp.ot = IRT(0, IRT_PTR); /* Buffer data pointer type. */ + emit_storeofs(as, &irbp, tmp, sb, offsetof(SBuf, w)); + emit_loadofs(as, &irbp, tmp, sb, offsetof(SBuf, b)); + break; + } + case IRBUFHDR_APPEND: { + /* Rematerialize const buffer pointer instead of likely spill. */ + IRIns *irp = IR(ir->op1); + if (!(ra_hasreg(irp->r) || irp == ir-1 || + (irp == ir-2 && !ra_used(ir-1)))) { + while (!(irp->o == IR_BUFHDR && irp->op2 == IRBUFHDR_RESET)) + irp = IR(irp->op1); + if (irref_isk(irp->op1)) { + ra_weak(as, ra_allocref(as, ir->op1, RSET_GPR)); + ir = irp; + } + } + break; + } +#if LJ_HASBUFFER + case IRBUFHDR_WRITE: + asm_bufhdr_write(as, sb); + break; +#endif + default: lj_assertA(0, "bad BUFHDR op2 %d", ir->op2); break; + } +#if LJ_TARGET_X86ORX64 + ra_left(as, sb, ir->op1); +#else + ra_leftov(as, sb, ir->op1); +#endif +} + +static void asm_bufput(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_buf_putstr]; + IRRef args[3]; + IRIns *irs; + int kchar = -129; + args[0] = ir->op1; /* SBuf * */ + args[1] = ir->op2; /* GCstr * */ + irs = IR(ir->op2); + lj_assertA(irt_isstr(irs->t), + "BUFPUT of non-string IR %04d", ir->op2 - REF_BIAS); + if (irs->o == IR_KGC) { + GCstr *s = ir_kstr(irs); + if (s->len == 1) { /* Optimize put of single-char string constant. */ + kchar = (int8_t)strdata(s)[0]; /* Signed! */ + args[1] = ASMREF_TMP1; /* int, truncated to char */ + ci = &lj_ir_callinfo[IRCALL_lj_buf_putchar]; + } + } else if (mayfuse(as, ir->op2) && ra_noreg(irs->r)) { + if (irs->o == IR_TOSTR) { /* Fuse number to string conversions. */ + if (irs->op2 == IRTOSTR_NUM) { + args[1] = ASMREF_TMP1; /* TValue * */ + ci = &lj_ir_callinfo[IRCALL_lj_strfmt_putnum]; + } else { + lj_assertA(irt_isinteger(IR(irs->op1)->t), + "TOSTR of non-numeric IR %04d", irs->op1); + args[1] = irs->op1; /* int */ + if (irs->op2 == IRTOSTR_INT) + ci = &lj_ir_callinfo[IRCALL_lj_strfmt_putint]; + else + ci = &lj_ir_callinfo[IRCALL_lj_buf_putchar]; + } + } else if (irs->o == IR_SNEW) { /* Fuse string allocation. */ + args[1] = irs->op1; /* const void * */ + args[2] = irs->op2; /* MSize */ + ci = &lj_ir_callinfo[IRCALL_lj_buf_putmem]; + } + } + asm_setupresult(as, ir, ci); /* SBuf * */ + asm_gencall(as, ci, args); + if (args[1] == ASMREF_TMP1) { + Reg tmp = ra_releasetmp(as, ASMREF_TMP1); + if (kchar == -129) + asm_tvptr(as, tmp, irs->op1, IRTMPREF_IN1); + else + ra_allockreg(as, kchar, tmp); + } +} + +static void asm_bufstr(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_buf_tostr]; + IRRef args[1]; + args[0] = ir->op1; /* SBuf *sb */ + as->gcsteps++; + asm_setupresult(as, ir, ci); /* GCstr * */ + asm_gencall(as, ci, args); +} + +/* -- Type conversions ---------------------------------------------------- */ + +static void asm_tostr(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci; + IRRef args[2]; + asm_snap_prep(as); + args[0] = ASMREF_L; + as->gcsteps++; + if (ir->op2 == IRTOSTR_NUM) { + args[1] = ASMREF_TMP1; /* cTValue * */ + ci = &lj_ir_callinfo[IRCALL_lj_strfmt_num]; + } else { + args[1] = ir->op1; /* int32_t k */ + if (ir->op2 == IRTOSTR_INT) + ci = &lj_ir_callinfo[IRCALL_lj_strfmt_int]; + else + ci = &lj_ir_callinfo[IRCALL_lj_strfmt_char]; + } + asm_setupresult(as, ir, ci); /* GCstr * */ + asm_gencall(as, ci, args); + if (ir->op2 == IRTOSTR_NUM) + asm_tvptr(as, ra_releasetmp(as, ASMREF_TMP1), ir->op1, IRTMPREF_IN1); +} + +#if LJ_32 && LJ_HASFFI && !LJ_SOFTFP && !LJ_TARGET_X86 +static void asm_conv64(ASMState *as, IRIns *ir) +{ + IRType st = (IRType)((ir-1)->op2 & IRCONV_SRCMASK); + IRType dt = (((ir-1)->op2 & IRCONV_DSTMASK) >> IRCONV_DSH); + IRCallID id; + IRRef args[2]; + lj_assertA((ir-1)->o == IR_CONV && ir->o == IR_HIOP, + "not a CONV/HIOP pair at IR %04d", (int)(ir - as->ir) - REF_BIAS); + args[LJ_BE] = (ir-1)->op1; + args[LJ_LE] = ir->op1; + if (st == IRT_NUM || st == IRT_FLOAT) { + id = IRCALL_fp64_d2l + ((st == IRT_FLOAT) ? 2 : 0) + (dt - IRT_I64); + ir--; + } else { + id = IRCALL_fp64_l2d + ((dt == IRT_FLOAT) ? 2 : 0) + (st - IRT_I64); + } + { +#if LJ_TARGET_ARM && !LJ_ABI_SOFTFP + CCallInfo cim = lj_ir_callinfo[id], *ci = &cim; + cim.flags |= CCI_VARARG; /* These calls don't use the hard-float ABI! */ +#else + const CCallInfo *ci = &lj_ir_callinfo[id]; +#endif + asm_setupresult(as, ir, ci); + asm_gencall(as, ci, args); + } +} +#endif + +/* -- Memory references --------------------------------------------------- */ + +static void asm_newref(ASMState *as, IRIns *ir) +{ + const CCallInfo *ci = &lj_ir_callinfo[IRCALL_lj_tab_newkey]; + IRRef args[3]; + if (ir->r == RID_SINK) + return; + asm_snap_prep(as); + args[0] = ASMREF_L; /* lua_State *L */ + args[1] = ir->op1; /* GCtab *t */ + args[2] = ASMREF_TMP1; /* cTValue *key */ + asm_setupresult(as, ir, ci); /* TValue * */ + asm_gencall(as, ci, args); + asm_tvptr(as, ra_releasetmp(as, ASMREF_TMP1), ir->op2, IRTMPREF_IN1); +} + +static void asm_tmpref(ASMState *as, IRIns *ir) +{ + Reg r = ra_dest(as, ir, RSET_GPR); + asm_tvptr(as, r, ir->op1, ir->op2); +} + +static void asm_lref(ASMState *as, IRIns *ir) +{ + Reg r = ra_dest(as, ir, RSET_GPR); +#if LJ_TARGET_X86ORX64 + ra_left(as, r, ASMREF_L); +#else + ra_leftov(as, r, ASMREF_L); +#endif +} + +/* -- Calls --------------------------------------------------------------- */ + +/* Collect arguments from CALL* and CARG instructions. */ +static void asm_collectargs(ASMState *as, IRIns *ir, + const CCallInfo *ci, IRRef *args) +{ + uint32_t n = CCI_XNARGS(ci); + /* Account for split args. */ + lj_assertA(n <= CCI_NARGS_MAX*2, "too many args %d to collect", n); + if ((ci->flags & CCI_L)) { *args++ = ASMREF_L; n--; } + while (n-- > 1) { + ir = IR(ir->op1); + lj_assertA(ir->o == IR_CARG, "malformed CALL arg tree"); + args[n] = ir->op2 == REF_NIL ? 0 : ir->op2; + } + args[0] = ir->op1 == REF_NIL ? 0 : ir->op1; + lj_assertA(IR(ir->op1)->o != IR_CARG, "malformed CALL arg tree"); +} + +/* Reconstruct CCallInfo flags for CALLX*. */ +static uint32_t asm_callx_flags(ASMState *as, IRIns *ir) +{ + uint32_t nargs = 0; + if (ir->op1 != REF_NIL) { /* Count number of arguments first. */ + IRIns *ira = IR(ir->op1); + nargs++; + while (ira->o == IR_CARG) { nargs++; ira = IR(ira->op1); } + } +#if LJ_HASFFI + if (IR(ir->op2)->o == IR_CARG) { /* Copy calling convention info. */ + CTypeID id = (CTypeID)IR(IR(ir->op2)->op2)->i; + CType *ct = ctype_get(ctype_ctsG(J2G(as->J)), id); + nargs |= ((ct->info & CTF_VARARG) ? CCI_VARARG : 0); +#if LJ_TARGET_X86 + nargs |= (ctype_cconv(ct->info) << CCI_CC_SHIFT); +#endif + } +#endif + return (nargs | (ir->t.irt << CCI_OTSHIFT)); +} + +static void asm_callid(ASMState *as, IRIns *ir, IRCallID id) +{ + const CCallInfo *ci = &lj_ir_callinfo[id]; + IRRef args[2]; + args[0] = ir->op1; + args[1] = ir->op2; + asm_setupresult(as, ir, ci); + asm_gencall(as, ci, args); +} + +static void asm_call(ASMState *as, IRIns *ir) +{ + IRRef args[CCI_NARGS_MAX]; + const CCallInfo *ci = &lj_ir_callinfo[ir->op2]; + asm_collectargs(as, ir, ci, args); + asm_setupresult(as, ir, ci); + asm_gencall(as, ci, args); +} + +/* -- PHI and loop handling ----------------------------------------------- */ + +/* Break a PHI cycle by renaming to a free register (evict if needed). */ +static void asm_phi_break(ASMState *as, RegSet blocked, RegSet blockedby, + RegSet allow) +{ + RegSet candidates = blocked & allow; + if (candidates) { /* If this register file has candidates. */ + /* Note: the set for ra_pick cannot be empty, since each register file + ** has some registers never allocated to PHIs. + */ + Reg down, up = ra_pick(as, ~blocked & allow); /* Get a free register. */ + if (candidates & ~blockedby) /* Optimize shifts, else it's a cycle. */ + candidates = candidates & ~blockedby; + down = rset_picktop(candidates); /* Pick candidate PHI register. */ + ra_rename(as, down, up); /* And rename it to the free register. */ + } +} + +/* PHI register shuffling. +** +** The allocator tries hard to preserve PHI register assignments across +** the loop body. Most of the time this loop does nothing, since there +** are no register mismatches. +** +** If a register mismatch is detected and ... +** - the register is currently free: rename it. +** - the register is blocked by an invariant: restore/remat and rename it. +** - Otherwise the register is used by another PHI, so mark it as blocked. +** +** The renames are order-sensitive, so just retry the loop if a register +** is marked as blocked, but has been freed in the meantime. A cycle is +** detected if all of the blocked registers are allocated. To break the +** cycle rename one of them to a free register and retry. +** +** Note that PHI spill slots are kept in sync and don't need to be shuffled. +*/ +static void asm_phi_shuffle(ASMState *as) +{ + RegSet work; + + /* Find and resolve PHI register mismatches. */ + for (;;) { + RegSet blocked = RSET_EMPTY; + RegSet blockedby = RSET_EMPTY; + RegSet phiset = as->phiset; + while (phiset) { /* Check all left PHI operand registers. */ + Reg r = rset_pickbot(phiset); + IRIns *irl = IR(as->phireg[r]); + Reg left = irl->r; + if (r != left) { /* Mismatch? */ + if (!rset_test(as->freeset, r)) { /* PHI register blocked? */ + IRRef ref = regcost_ref(as->cost[r]); + /* Blocked by other PHI (w/reg)? */ + if (!ra_iskref(ref) && irt_ismarked(IR(ref)->t)) { + rset_set(blocked, r); + if (ra_hasreg(left)) + rset_set(blockedby, left); + left = RID_NONE; + } else { /* Otherwise grab register from invariant. */ + ra_restore(as, ref); + checkmclim(as); + } + } + if (ra_hasreg(left)) { + ra_rename(as, left, r); + checkmclim(as); + } + } + rset_clear(phiset, r); + } + if (!blocked) break; /* Finished. */ + if (!(as->freeset & blocked)) { /* Break cycles if none are free. */ + asm_phi_break(as, blocked, blockedby, RSET_GPR); + if (!LJ_SOFTFP) asm_phi_break(as, blocked, blockedby, RSET_FPR); + checkmclim(as); + } /* Else retry some more renames. */ + } + + /* Restore/remat invariants whose registers are modified inside the loop. */ +#if !LJ_SOFTFP + work = as->modset & ~(as->freeset | as->phiset) & RSET_FPR; + while (work) { + Reg r = rset_pickbot(work); + ra_restore(as, regcost_ref(as->cost[r])); + rset_clear(work, r); + checkmclim(as); + } +#endif + work = as->modset & ~(as->freeset | as->phiset); + while (work) { + Reg r = rset_pickbot(work); + ra_restore(as, regcost_ref(as->cost[r])); + rset_clear(work, r); + checkmclim(as); + } + + /* Allocate and save all unsaved PHI regs and clear marks. */ + work = as->phiset; + while (work) { + Reg r = rset_picktop(work); + IRRef lref = as->phireg[r]; + IRIns *ir = IR(lref); + if (ra_hasspill(ir->s)) { /* Left PHI gained a spill slot? */ + irt_clearmark(ir->t); /* Handled here, so clear marker now. */ + ra_alloc1(as, lref, RID2RSET(r)); + ra_save(as, ir, r); /* Save to spill slot inside the loop. */ + checkmclim(as); + } + rset_clear(work, r); + } +} + +/* Copy unsynced left/right PHI spill slots. Rarely needed. */ +static void asm_phi_copyspill(ASMState *as) +{ + int need = 0; + IRIns *ir; + for (ir = IR(as->orignins-1); ir->o == IR_PHI; ir--) + if (ra_hasspill(ir->s) && ra_hasspill(IR(ir->op1)->s)) + need |= irt_isfp(ir->t) ? 2 : 1; /* Unsynced spill slot? */ + if ((need & 1)) { /* Copy integer spill slots. */ +#if !LJ_TARGET_X86ORX64 + Reg r = RID_TMP; +#else + Reg r = RID_RET; + if ((as->freeset & RSET_GPR)) + r = rset_pickbot((as->freeset & RSET_GPR)); + else + emit_spload(as, IR(regcost_ref(as->cost[r])), r, SPOFS_TMP); +#endif + for (ir = IR(as->orignins-1); ir->o == IR_PHI; ir--) { + if (ra_hasspill(ir->s)) { + IRIns *irl = IR(ir->op1); + if (ra_hasspill(irl->s) && !irt_isfp(ir->t)) { + emit_spstore(as, irl, r, sps_scale(irl->s)); + emit_spload(as, ir, r, sps_scale(ir->s)); + checkmclim(as); + } + } + } +#if LJ_TARGET_X86ORX64 + if (!rset_test(as->freeset, r)) + emit_spstore(as, IR(regcost_ref(as->cost[r])), r, SPOFS_TMP); +#endif + } +#if !LJ_SOFTFP + if ((need & 2)) { /* Copy FP spill slots. */ +#if LJ_TARGET_X86 + Reg r = RID_XMM0; +#else + Reg r = RID_FPRET; +#endif + if ((as->freeset & RSET_FPR)) + r = rset_pickbot((as->freeset & RSET_FPR)); + if (!rset_test(as->freeset, r)) + emit_spload(as, IR(regcost_ref(as->cost[r])), r, SPOFS_TMP); + for (ir = IR(as->orignins-1); ir->o == IR_PHI; ir--) { + if (ra_hasspill(ir->s)) { + IRIns *irl = IR(ir->op1); + if (ra_hasspill(irl->s) && irt_isfp(ir->t)) { + emit_spstore(as, irl, r, sps_scale(irl->s)); + emit_spload(as, ir, r, sps_scale(ir->s)); + checkmclim(as); + } + } + } + if (!rset_test(as->freeset, r)) + emit_spstore(as, IR(regcost_ref(as->cost[r])), r, SPOFS_TMP); + } +#endif +} + +/* Emit renames for left PHIs which are only spilled outside the loop. */ +static void asm_phi_fixup(ASMState *as) +{ + RegSet work = as->phiset; + while (work) { + Reg r = rset_picktop(work); + IRRef lref = as->phireg[r]; + IRIns *ir = IR(lref); + if (irt_ismarked(ir->t)) { + irt_clearmark(ir->t); + /* Left PHI gained a spill slot before the loop? */ + if (ra_hasspill(ir->s)) { + ra_addrename(as, r, lref, as->loopsnapno); + } + } + rset_clear(work, r); + } +} + +/* Setup right PHI reference. */ +static void asm_phi(ASMState *as, IRIns *ir) +{ + RegSet allow = ((!LJ_SOFTFP && irt_isfp(ir->t)) ? RSET_FPR : RSET_GPR) & + ~as->phiset; + RegSet afree = (as->freeset & allow); + IRIns *irl = IR(ir->op1); + IRIns *irr = IR(ir->op2); + if (ir->r == RID_SINK) /* Sink PHI. */ + return; + /* Spill slot shuffling is not implemented yet (but rarely needed). */ + if (ra_hasspill(irl->s) || ra_hasspill(irr->s)) + lj_trace_err(as->J, LJ_TRERR_NYIPHI); + /* Leave at least one register free for non-PHIs (and PHI cycle breaking). */ + if ((afree & (afree-1))) { /* Two or more free registers? */ + Reg r; + if (ra_noreg(irr->r)) { /* Get a register for the right PHI. */ + r = ra_allocref(as, ir->op2, allow); + } else { /* Duplicate right PHI, need a copy (rare). */ + r = ra_scratch(as, allow); + emit_movrr(as, irr, r, irr->r); + } + ir->r = (uint8_t)r; + rset_set(as->phiset, r); + as->phireg[r] = (IRRef1)ir->op1; + irt_setmark(irl->t); /* Marks left PHIs _with_ register. */ + if (ra_noreg(irl->r)) + ra_sethint(irl->r, r); /* Set register hint for left PHI. */ + } else { /* Otherwise allocate a spill slot. */ + /* This is overly restrictive, but it triggers only on synthetic code. */ + if (ra_hasreg(irl->r) || ra_hasreg(irr->r)) + lj_trace_err(as->J, LJ_TRERR_NYIPHI); + ra_spill(as, ir); + irr->s = ir->s; /* Set right PHI spill slot. Sync left slot later. */ + } +} + +static void asm_loop_fixup(ASMState *as); + +/* Middle part of a loop. */ +static void asm_loop(ASMState *as) +{ + MCode *mcspill; + /* LOOP is a guard, so the snapno is up to date. */ + as->loopsnapno = as->snapno; + if (as->gcsteps) + asm_gc_check(as); + /* LOOP marks the transition from the variant to the invariant part. */ + as->flagmcp = as->invmcp = NULL; + as->sectref = 0; + if (!neverfuse(as)) as->fuseref = 0; + asm_phi_shuffle(as); + mcspill = as->mcp; + asm_phi_copyspill(as); + asm_loop_fixup(as); + as->mcloop = as->mcp; + RA_DBGX((as, "===== LOOP =====")); + if (!as->realign) RA_DBG_FLUSH(); + if (as->mcp != mcspill) + emit_jmp(as, mcspill); +} + +/* -- Target-specific assembler ------------------------------------------- */ + +#if LJ_TARGET_X86ORX64 +#include "lj_asm_x86.h" +#elif LJ_TARGET_ARM +#include "lj_asm_arm.h" +#elif LJ_TARGET_ARM64 +#include "lj_asm_arm64.h" +#elif LJ_TARGET_PPC +#include "lj_asm_ppc.h" +#elif LJ_TARGET_MIPS +#include "lj_asm_mips.h" +#else +#error "Missing assembler for target CPU" +#endif + +/* -- Common instruction helpers ------------------------------------------ */ + +#if !LJ_SOFTFP32 +#if !LJ_TARGET_X86ORX64 +#define asm_ldexp(as, ir) asm_callid(as, ir, IRCALL_ldexp) +#endif + +static void asm_pow(ASMState *as, IRIns *ir) +{ +#if LJ_64 && LJ_HASFFI + if (!irt_isnum(ir->t)) + asm_callid(as, ir, irt_isi64(ir->t) ? IRCALL_lj_carith_powi64 : + IRCALL_lj_carith_powu64); + else +#endif + asm_callid(as, ir, IRCALL_pow); +} + +static void asm_div(ASMState *as, IRIns *ir) +{ +#if LJ_64 && LJ_HASFFI + if (!irt_isnum(ir->t)) + asm_callid(as, ir, irt_isi64(ir->t) ? IRCALL_lj_carith_divi64 : + IRCALL_lj_carith_divu64); + else +#endif + asm_fpdiv(as, ir); +} +#endif + +static void asm_mod(ASMState *as, IRIns *ir) +{ +#if LJ_64 && LJ_HASFFI + if (!irt_isint(ir->t)) + asm_callid(as, ir, irt_isi64(ir->t) ? IRCALL_lj_carith_modi64 : + IRCALL_lj_carith_modu64); + else +#endif + asm_callid(as, ir, IRCALL_lj_vm_modi); +} + +static void asm_fuseequal(ASMState *as, IRIns *ir) +{ + /* Fuse HREF + EQ/NE. */ + if ((ir-1)->o == IR_HREF && ir->op1 == as->curins-1) { + as->curins--; + asm_href(as, ir-1, (IROp)ir->o); + } else { + asm_equal(as, ir); + } +} + +static void asm_alen(ASMState *as, IRIns *ir) +{ + asm_callid(as, ir, ir->op2 == REF_NIL ? IRCALL_lj_tab_len : + IRCALL_lj_tab_len_hint); +} + +/* -- Instruction dispatch ------------------------------------------------ */ + +/* Assemble a single instruction. */ +static void asm_ir(ASMState *as, IRIns *ir) +{ + switch ((IROp)ir->o) { + /* Miscellaneous ops. */ + case IR_LOOP: asm_loop(as); break; + case IR_NOP: case IR_XBAR: + lj_assertA(!ra_used(ir), + "IR %04d not unused", (int)(ir - as->ir) - REF_BIAS); + break; + case IR_USE: + ra_alloc1(as, ir->op1, irt_isfp(ir->t) ? RSET_FPR : RSET_GPR); break; + case IR_PHI: asm_phi(as, ir); break; + case IR_HIOP: asm_hiop(as, ir); break; + case IR_GCSTEP: asm_gcstep(as, ir); break; + case IR_PROF: asm_prof(as, ir); break; + + /* Guarded assertions. */ + case IR_LT: case IR_GE: case IR_LE: case IR_GT: + case IR_ULT: case IR_UGE: case IR_ULE: case IR_UGT: + case IR_ABC: + asm_comp(as, ir); + break; + case IR_EQ: case IR_NE: asm_fuseequal(as, ir); break; + + case IR_RETF: asm_retf(as, ir); break; + + /* Bit ops. */ + case IR_BNOT: asm_bnot(as, ir); break; + case IR_BSWAP: asm_bswap(as, ir); break; + case IR_BAND: asm_band(as, ir); break; + case IR_BOR: asm_bor(as, ir); break; + case IR_BXOR: asm_bxor(as, ir); break; + case IR_BSHL: asm_bshl(as, ir); break; + case IR_BSHR: asm_bshr(as, ir); break; + case IR_BSAR: asm_bsar(as, ir); break; + case IR_BROL: asm_brol(as, ir); break; + case IR_BROR: asm_bror(as, ir); break; + + /* Arithmetic ops. */ + case IR_ADD: asm_add(as, ir); break; + case IR_SUB: asm_sub(as, ir); break; + case IR_MUL: asm_mul(as, ir); break; + case IR_MOD: asm_mod(as, ir); break; + case IR_NEG: asm_neg(as, ir); break; +#if LJ_SOFTFP32 + case IR_DIV: case IR_POW: case IR_ABS: + case IR_LDEXP: case IR_FPMATH: case IR_TOBIT: + /* Unused for LJ_SOFTFP32. */ + lj_assertA(0, "IR %04d with unused op %d", + (int)(ir - as->ir) - REF_BIAS, ir->o); + break; +#else + case IR_DIV: asm_div(as, ir); break; + case IR_POW: asm_pow(as, ir); break; + case IR_ABS: asm_abs(as, ir); break; + case IR_LDEXP: asm_ldexp(as, ir); break; + case IR_FPMATH: asm_fpmath(as, ir); break; + case IR_TOBIT: asm_tobit(as, ir); break; +#endif + case IR_MIN: asm_min(as, ir); break; + case IR_MAX: asm_max(as, ir); break; + + /* Overflow-checking arithmetic ops. */ + case IR_ADDOV: asm_addov(as, ir); break; + case IR_SUBOV: asm_subov(as, ir); break; + case IR_MULOV: asm_mulov(as, ir); break; + + /* Memory references. */ + case IR_AREF: asm_aref(as, ir); break; + case IR_HREF: asm_href(as, ir, 0); break; + case IR_HREFK: asm_hrefk(as, ir); break; + case IR_NEWREF: asm_newref(as, ir); break; + case IR_UREFO: case IR_UREFC: asm_uref(as, ir); break; + case IR_FREF: asm_fref(as, ir); break; + case IR_TMPREF: asm_tmpref(as, ir); break; + case IR_STRREF: asm_strref(as, ir); break; + case IR_LREF: asm_lref(as, ir); break; + + /* Loads and stores. */ + case IR_ALOAD: case IR_HLOAD: case IR_ULOAD: case IR_VLOAD: + asm_ahuvload(as, ir); + break; + case IR_FLOAD: asm_fload(as, ir); break; + case IR_XLOAD: asm_xload(as, ir); break; + case IR_SLOAD: asm_sload(as, ir); break; + case IR_ALEN: asm_alen(as, ir); break; + + case IR_ASTORE: case IR_HSTORE: case IR_USTORE: asm_ahustore(as, ir); break; + case IR_FSTORE: asm_fstore(as, ir); break; + case IR_XSTORE: asm_xstore(as, ir); break; + + /* Allocations. */ + case IR_SNEW: case IR_XSNEW: asm_snew(as, ir); break; + case IR_TNEW: asm_tnew(as, ir); break; + case IR_TDUP: asm_tdup(as, ir); break; + case IR_CNEW: case IR_CNEWI: +#if LJ_HASFFI + asm_cnew(as, ir); +#else + lj_assertA(0, "IR %04d with unused op %d", + (int)(ir - as->ir) - REF_BIAS, ir->o); +#endif + break; + + /* Buffer operations. */ + case IR_BUFHDR: asm_bufhdr(as, ir); break; + case IR_BUFPUT: asm_bufput(as, ir); break; + case IR_BUFSTR: asm_bufstr(as, ir); break; + + /* Write barriers. */ + case IR_TBAR: asm_tbar(as, ir); break; + case IR_OBAR: asm_obar(as, ir); break; + + /* Type conversions. */ + case IR_CONV: asm_conv(as, ir); break; + case IR_TOSTR: asm_tostr(as, ir); break; + case IR_STRTO: asm_strto(as, ir); break; + + /* Calls. */ + case IR_CALLA: + as->gcsteps++; + /* fallthrough */ + case IR_CALLN: case IR_CALLL: case IR_CALLS: asm_call(as, ir); break; + case IR_CALLXS: asm_callx(as, ir); break; + case IR_CARG: break; + + default: + setintV(&as->J->errinfo, ir->o); + lj_trace_err_info(as->J, LJ_TRERR_NYIIR); + break; + } +} + +/* -- Head of trace ------------------------------------------------------- */ + +/* Head of a root trace. */ +static void asm_head_root(ASMState *as) +{ + int32_t spadj; + asm_head_root_base(as); + emit_setvmstate(as, (int32_t)as->T->traceno); + spadj = asm_stack_adjust(as); + as->T->spadjust = (uint16_t)spadj; + emit_spsub(as, spadj); + /* Root traces assume a checked stack for the starting proto. */ + as->T->topslot = gcref(as->T->startpt)->pt.framesize; +} + +/* Head of a side trace. +** +** The current simplistic algorithm requires that all slots inherited +** from the parent are live in a register between pass 2 and pass 3. This +** avoids the complexity of stack slot shuffling. But of course this may +** overflow the register set in some cases and cause the dreaded error: +** "NYI: register coalescing too complex". A refined algorithm is needed. +*/ +static void asm_head_side(ASMState *as) +{ + IRRef1 sloadins[RID_MAX]; + RegSet allow = RSET_ALL; /* Inverse of all coalesced registers. */ + RegSet live = RSET_EMPTY; /* Live parent registers. */ + RegSet pallow = RSET_GPR; /* Registers needed by the parent stack check. */ + Reg pbase; + IRIns *irp = &as->parent->ir[REF_BASE]; /* Parent base. */ + int32_t spadj, spdelta; + int pass2 = 0; + int pass3 = 0; + IRRef i; + + if (as->snapno && as->topslot > as->parent->topslot) { + /* Force snap #0 alloc to prevent register overwrite in stack check. */ + asm_snap_alloc(as, 0); + } + pbase = asm_head_side_base(as, irp); + if (pbase != RID_NONE) { + rset_clear(allow, pbase); + rset_clear(pallow, pbase); + } + + /* Scan all parent SLOADs and collect register dependencies. */ + for (i = as->stopins; i > REF_BASE; i--) { + IRIns *ir = IR(i); + RegSP rs; + lj_assertA((ir->o == IR_SLOAD && (ir->op2 & IRSLOAD_PARENT)) || + (LJ_SOFTFP && ir->o == IR_HIOP) || ir->o == IR_PVAL, + "IR %04d has bad parent op %d", + (int)(ir - as->ir) - REF_BIAS, ir->o); + rs = as->parentmap[i - REF_FIRST]; + if (ra_hasreg(ir->r)) { + rset_clear(allow, ir->r); + if (ra_hasspill(ir->s)) { + ra_save(as, ir, ir->r); + checkmclim(as); + } + } else if (ra_hasspill(ir->s)) { + irt_setmark(ir->t); + pass2 = 1; + } + if (ir->r == rs) { /* Coalesce matching registers right now. */ + ra_free(as, ir->r); + } else if (ra_hasspill(regsp_spill(rs))) { + if (ra_hasreg(ir->r)) + pass3 = 1; + } else if (ra_used(ir)) { + sloadins[rs] = (IRRef1)i; + rset_set(live, rs); /* Block live parent register. */ + } + if (!ra_hasspill(regsp_spill(rs))) rset_clear(pallow, regsp_reg(rs)); + } + + /* Calculate stack frame adjustment. */ + spadj = asm_stack_adjust(as); + spdelta = spadj - (int32_t)as->parent->spadjust; + if (spdelta < 0) { /* Don't shrink the stack frame. */ + spadj = (int32_t)as->parent->spadjust; + spdelta = 0; + } + as->T->spadjust = (uint16_t)spadj; + + /* Reload spilled target registers. */ + if (pass2) { + for (i = as->stopins; i > REF_BASE; i--) { + IRIns *ir = IR(i); + if (irt_ismarked(ir->t)) { + RegSet mask; + Reg r; + RegSP rs; + irt_clearmark(ir->t); + rs = as->parentmap[i - REF_FIRST]; + if (!ra_hasspill(regsp_spill(rs))) + ra_sethint(ir->r, rs); /* Hint may be gone, set it again. */ + else if (sps_scale(regsp_spill(rs))+spdelta == sps_scale(ir->s)) + continue; /* Same spill slot, do nothing. */ + mask = ((!LJ_SOFTFP && irt_isfp(ir->t)) ? RSET_FPR : RSET_GPR) & allow; + if (mask == RSET_EMPTY) + lj_trace_err(as->J, LJ_TRERR_NYICOAL); + r = ra_allocref(as, i, mask); + ra_save(as, ir, r); + rset_clear(allow, r); + if (r == rs) { /* Coalesce matching registers right now. */ + ra_free(as, r); + rset_clear(live, r); + } else if (ra_hasspill(regsp_spill(rs))) { + pass3 = 1; + } + checkmclim(as); + } + } + } + + /* Store trace number and adjust stack frame relative to the parent. */ + emit_setvmstate(as, (int32_t)as->T->traceno); + emit_spsub(as, spdelta); + +#if !LJ_TARGET_X86ORX64 + /* Restore BASE register from parent spill slot. */ + if (ra_hasspill(irp->s)) + emit_spload(as, IR(REF_BASE), IR(REF_BASE)->r, sps_scale(irp->s)); +#endif + + /* Restore target registers from parent spill slots. */ + if (pass3) { + RegSet work = ~as->freeset & RSET_ALL; + while (work) { + Reg r = rset_pickbot(work); + IRRef ref = regcost_ref(as->cost[r]); + RegSP rs = as->parentmap[ref - REF_FIRST]; + rset_clear(work, r); + if (ra_hasspill(regsp_spill(rs))) { + int32_t ofs = sps_scale(regsp_spill(rs)); + ra_free(as, r); + emit_spload(as, IR(ref), r, ofs); + checkmclim(as); + } + } + } + + /* Shuffle registers to match up target regs with parent regs. */ + for (;;) { + RegSet work; + + /* Repeatedly coalesce free live registers by moving to their target. */ + while ((work = as->freeset & live) != RSET_EMPTY) { + Reg rp = rset_pickbot(work); + IRIns *ir = IR(sloadins[rp]); + rset_clear(live, rp); + rset_clear(allow, rp); + ra_free(as, ir->r); + emit_movrr(as, ir, ir->r, rp); + checkmclim(as); + } + + /* We're done if no live registers remain. */ + if (live == RSET_EMPTY) + break; + + /* Break cycles by renaming one target to a temp. register. */ + if (live & RSET_GPR) { + RegSet tmpset = as->freeset & ~live & allow & RSET_GPR; + if (tmpset == RSET_EMPTY) + lj_trace_err(as->J, LJ_TRERR_NYICOAL); + ra_rename(as, rset_pickbot(live & RSET_GPR), rset_pickbot(tmpset)); + } + if (!LJ_SOFTFP && (live & RSET_FPR)) { + RegSet tmpset = as->freeset & ~live & allow & RSET_FPR; + if (tmpset == RSET_EMPTY) + lj_trace_err(as->J, LJ_TRERR_NYICOAL); + ra_rename(as, rset_pickbot(live & RSET_FPR), rset_pickbot(tmpset)); + } + checkmclim(as); + /* Continue with coalescing to fix up the broken cycle(s). */ + } + + /* Inherit top stack slot already checked by parent trace. */ + as->T->topslot = as->parent->topslot; + if (as->topslot > as->T->topslot) { /* Need to check for higher slot? */ +#ifdef EXITSTATE_CHECKEXIT + /* Highest exit + 1 indicates stack check. */ + ExitNo exitno = as->T->nsnap; +#else + /* Reuse the parent exit in the context of the parent trace. */ + ExitNo exitno = as->J->exitno; +#endif + as->T->topslot = (uint8_t)as->topslot; /* Remember for child traces. */ + asm_stack_check(as, as->topslot, irp, pallow, exitno); + } +} + +/* -- Tail of trace ------------------------------------------------------- */ + +/* Get base slot for a snapshot. */ +static BCReg asm_baseslot(ASMState *as, SnapShot *snap, int *gotframe) +{ + SnapEntry *map = &as->T->snapmap[snap->mapofs]; + MSize n; + for (n = snap->nent; n > 0; n--) { + SnapEntry sn = map[n-1]; + if ((sn & SNAP_FRAME)) { + *gotframe = 1; + return snap_slot(sn) - LJ_FR2; + } + } + return 0; +} + +/* Link to another trace. */ +static void asm_tail_link(ASMState *as) +{ + SnapNo snapno = as->T->nsnap-1; /* Last snapshot. */ + SnapShot *snap = &as->T->snap[snapno]; + int gotframe = 0; + BCReg baseslot = asm_baseslot(as, snap, &gotframe); + + as->topslot = snap->topslot; + checkmclim(as); + ra_allocref(as, REF_BASE, RID2RSET(RID_BASE)); + + if (as->T->link == 0) { + /* Setup fixed registers for exit to interpreter. */ + const BCIns *pc = snap_pc(&as->T->snapmap[snap->mapofs + snap->nent]); + int32_t mres; + if (bc_op(*pc) == BC_JLOOP) { /* NYI: find a better way to do this. */ + BCIns *retpc = &traceref(as->J, bc_d(*pc))->startins; + if (bc_isret(bc_op(*retpc))) + pc = retpc; + } +#if LJ_GC64 + emit_loadu64(as, RID_LPC, u64ptr(pc)); +#else + ra_allockreg(as, i32ptr(J2GG(as->J)->dispatch), RID_DISPATCH); + ra_allockreg(as, i32ptr(pc), RID_LPC); +#endif + mres = (int32_t)(snap->nslots - baseslot - LJ_FR2); + switch (bc_op(*pc)) { + case BC_CALLM: case BC_CALLMT: + mres -= (int32_t)(1 + LJ_FR2 + bc_a(*pc) + bc_c(*pc)); break; + case BC_RETM: mres -= (int32_t)(bc_a(*pc) + bc_d(*pc)); break; + case BC_TSETM: mres -= (int32_t)bc_a(*pc); break; + default: if (bc_op(*pc) < BC_FUNCF) mres = 0; break; + } + ra_allockreg(as, mres, RID_RET); /* Return MULTRES or 0. */ + } else if (baseslot) { + /* Save modified BASE for linking to trace with higher start frame. */ + emit_setgl(as, RID_BASE, jit_base); + } + emit_addptr(as, RID_BASE, 8*(int32_t)baseslot); + + if (as->J->ktrace) { /* Patch ktrace slot with the final GCtrace pointer. */ + setgcref(IR(as->J->ktrace)[LJ_GC64].gcr, obj2gco(as->J->curfinal)); + IR(as->J->ktrace)->o = IR_KGC; + } + + /* Sync the interpreter state with the on-trace state. */ + asm_stack_restore(as, snap); + + /* Root traces that add frames need to check the stack at the end. */ + if (!as->parent && gotframe) + asm_stack_check(as, as->topslot, NULL, as->freeset & RSET_GPR, snapno); +} + +/* -- Trace setup --------------------------------------------------------- */ + +/* Clear reg/sp for all instructions and add register hints. */ +static void asm_setup_regsp(ASMState *as) +{ + GCtrace *T = as->T; + int sink = T->sinktags; + IRRef nins = T->nins; + IRIns *ir, *lastir; + int inloop; +#if LJ_TARGET_ARM + uint32_t rload = 0xa6402a64; +#endif + + ra_setup(as); +#if LJ_TARGET_ARM64 + ra_setkref(as, RID_GL, (intptr_t)J2G(as->J)); +#endif + + /* Clear reg/sp for constants. */ + for (ir = IR(T->nk), lastir = IR(REF_BASE); ir < lastir; ir++) { + ir->prev = REGSP_INIT; + if (irt_is64(ir->t) && ir->o != IR_KNULL) { +#if LJ_GC64 + /* The false-positive of irt_is64() for ASMREF_L (REF_NIL) is OK here. */ + ir->i = 0; /* Will become non-zero only for RIP-relative addresses. */ +#else + /* Make life easier for backends by putting address of constant in i. */ + ir->i = (int32_t)(intptr_t)(ir+1); +#endif + ir++; + } + } + + /* REF_BASE is used for implicit references to the BASE register. */ + lastir->prev = REGSP_HINT(RID_BASE); + + as->snaprename = nins; + as->snapref = nins; + as->snapno = T->nsnap; + as->snapalloc = 0; + + as->stopins = REF_BASE; + as->orignins = nins; + as->curins = nins; + + /* Setup register hints for parent link instructions. */ + ir = IR(REF_FIRST); + if (as->parent) { + uint16_t *p; + lastir = lj_snap_regspmap(as->J, as->parent, as->J->exitno, ir); + if (lastir - ir > LJ_MAX_JSLOTS) + lj_trace_err(as->J, LJ_TRERR_NYICOAL); + as->stopins = (IRRef)((lastir-1) - as->ir); + for (p = as->parentmap; ir < lastir; ir++) { + RegSP rs = ir->prev; + *p++ = (uint16_t)rs; /* Copy original parent RegSP to parentmap. */ + if (!ra_hasspill(regsp_spill(rs))) + ir->prev = (uint16_t)REGSP_HINT(regsp_reg(rs)); + else + ir->prev = REGSP_INIT; + } + } + + inloop = 0; + as->evenspill = SPS_FIRST; + for (lastir = IR(nins); ir < lastir; ir++) { + if (sink) { + if (ir->r == RID_SINK) + continue; + if (ir->r == RID_SUNK) { /* Revert after ASM restart. */ + ir->r = RID_SINK; + continue; + } + } + switch (ir->o) { + case IR_LOOP: + inloop = 1; + break; +#if LJ_TARGET_ARM + case IR_SLOAD: + if (!((ir->op2 & IRSLOAD_TYPECHECK) || (ir+1)->o == IR_HIOP)) + break; + /* fallthrough */ + case IR_ALOAD: case IR_HLOAD: case IR_ULOAD: case IR_VLOAD: + if (!LJ_SOFTFP && irt_isnum(ir->t)) break; + ir->prev = (uint16_t)REGSP_HINT((rload & 15)); + rload = lj_ror(rload, 4); + continue; + case IR_TMPREF: + if ((ir->op2 & IRTMPREF_OUT2) && as->evenspill < 4) + as->evenspill = 4; /* TMPREF OUT2 needs two TValues on the stack. */ + break; +#endif + case IR_CALLXS: { + CCallInfo ci; + ci.flags = asm_callx_flags(as, ir); + ir->prev = asm_setup_call_slots(as, ir, &ci); + if (inloop) + as->modset |= RSET_SCRATCH; + continue; + } + case IR_CALLL: + /* lj_vm_next needs two TValues on the stack. */ +#if LJ_TARGET_X64 && LJ_ABI_WIN + if (ir->op2 == IRCALL_lj_vm_next && as->evenspill < SPS_FIRST + 4) + as->evenspill = SPS_FIRST + 4; +#else + if (SPS_FIRST < 4 && ir->op2 == IRCALL_lj_vm_next && as->evenspill < 4) + as->evenspill = 4; +#endif + /* fallthrough */ + case IR_CALLN: case IR_CALLA: case IR_CALLS: { + const CCallInfo *ci = &lj_ir_callinfo[ir->op2]; + ir->prev = asm_setup_call_slots(as, ir, ci); + if (inloop) + as->modset |= (ci->flags & CCI_NOFPRCLOBBER) ? + (RSET_SCRATCH & ~RSET_FPR) : RSET_SCRATCH; + continue; + } + case IR_HIOP: + switch ((ir-1)->o) { +#if LJ_SOFTFP && LJ_TARGET_ARM + case IR_SLOAD: case IR_ALOAD: case IR_HLOAD: case IR_ULOAD: case IR_VLOAD: + if (ra_hashint((ir-1)->r)) { + ir->prev = (ir-1)->prev + 1; + continue; + } + break; +#endif +#if !LJ_SOFTFP && LJ_NEED_FP64 && LJ_32 && LJ_HASFFI + case IR_CONV: + if (irt_isfp((ir-1)->t)) { + ir->prev = REGSP_HINT(RID_FPRET); + continue; + } +#endif + /* fallthrough */ + case IR_CALLN: case IR_CALLL: case IR_CALLS: case IR_CALLXS: +#if LJ_SOFTFP + case IR_MIN: case IR_MAX: +#endif + (ir-1)->prev = REGSP_HINT(RID_RETLO); + ir->prev = REGSP_HINT(RID_RETHI); + continue; + default: + break; + } + break; +#if LJ_SOFTFP + case IR_MIN: case IR_MAX: + if ((ir+1)->o != IR_HIOP) break; +#endif + /* fallthrough */ + /* C calls evict all scratch regs and return results in RID_RET. */ + case IR_SNEW: case IR_XSNEW: case IR_NEWREF: case IR_BUFPUT: + if (REGARG_NUMGPR < 3 && as->evenspill < 3) + as->evenspill = 3; /* lj_str_new and lj_tab_newkey need 3 args. */ +#if LJ_TARGET_X86 && LJ_HASFFI + if (0) { + case IR_CNEW: + if (ir->op2 != REF_NIL && as->evenspill < 4) + as->evenspill = 4; /* lj_cdata_newv needs 4 args. */ + } + /* fallthrough */ +#else + /* fallthrough */ + case IR_CNEW: +#endif + /* fallthrough */ + case IR_TNEW: case IR_TDUP: case IR_CNEWI: case IR_TOSTR: + case IR_BUFSTR: + ir->prev = REGSP_HINT(RID_RET); + if (inloop) + as->modset = RSET_SCRATCH; + continue; + case IR_STRTO: case IR_OBAR: + if (inloop) + as->modset = RSET_SCRATCH; + break; +#if !LJ_SOFTFP +#if !LJ_TARGET_X86ORX64 + case IR_LDEXP: +#endif +#endif + /* fallthrough */ + case IR_POW: + if (!LJ_SOFTFP && irt_isnum(ir->t)) { + if (inloop) + as->modset |= RSET_SCRATCH; +#if LJ_TARGET_X86 + if (irt_isnum(IR(ir->op2)->t)) { + if (as->evenspill < 4) /* Leave room to call pow(). */ + as->evenspill = 4; + } + break; +#else + ir->prev = REGSP_HINT(RID_FPRET); + continue; +#endif + } + /* fallthrough */ /* for integer POW */ + case IR_DIV: case IR_MOD: + if ((LJ_64 && LJ_SOFTFP) || !irt_isnum(ir->t)) { + ir->prev = REGSP_HINT(RID_RET); + if (inloop) + as->modset |= (RSET_SCRATCH & RSET_GPR); + continue; + } + break; +#if LJ_64 && LJ_SOFTFP + case IR_ADD: case IR_SUB: case IR_MUL: + if (irt_isnum(ir->t)) { + ir->prev = REGSP_HINT(RID_RET); + if (inloop) + as->modset |= (RSET_SCRATCH & RSET_GPR); + continue; + } + break; +#endif + case IR_FPMATH: +#if LJ_TARGET_X86ORX64 + if (ir->op2 <= IRFPM_TRUNC) { + if (!(as->flags & JIT_F_SSE4_1)) { + ir->prev = REGSP_HINT(RID_XMM0); + if (inloop) + as->modset |= RSET_RANGE(RID_XMM0, RID_XMM3+1)|RID2RSET(RID_EAX); + continue; + } + break; + } +#endif + if (inloop) + as->modset |= RSET_SCRATCH; +#if LJ_TARGET_X86 + break; +#else + ir->prev = REGSP_HINT(RID_FPRET); + continue; +#endif +#if LJ_TARGET_X86ORX64 + /* Non-constant shift counts need to be in RID_ECX on x86/x64. */ + case IR_BSHL: case IR_BSHR: case IR_BSAR: + if ((as->flags & JIT_F_BMI2)) /* Except if BMI2 is available. */ + break; + /* fallthrough */ + case IR_BROL: case IR_BROR: + if (!irref_isk(ir->op2) && !ra_hashint(IR(ir->op2)->r)) { + IR(ir->op2)->r = REGSP_HINT(RID_ECX); + if (inloop) + rset_set(as->modset, RID_ECX); + } + break; +#endif + /* Do not propagate hints across type conversions or loads. */ + case IR_TOBIT: + case IR_XLOAD: +#if !LJ_TARGET_ARM + case IR_ALOAD: case IR_HLOAD: case IR_ULOAD: case IR_VLOAD: +#endif + break; + case IR_CONV: + if (irt_isfp(ir->t) || (ir->op2 & IRCONV_SRCMASK) == IRT_NUM || + (ir->op2 & IRCONV_SRCMASK) == IRT_FLOAT) + break; + /* fallthrough */ + default: + /* Propagate hints across likely 'op reg, imm' or 'op reg'. */ + if (irref_isk(ir->op2) && !irref_isk(ir->op1) && + ra_hashint(regsp_reg(IR(ir->op1)->prev))) { + ir->prev = IR(ir->op1)->prev; + continue; + } + break; + } + ir->prev = REGSP_INIT; + } + if ((as->evenspill & 1)) + as->oddspill = as->evenspill++; + else + as->oddspill = 0; +} + +/* -- Assembler core ------------------------------------------------------ */ + +/* Assemble a trace. */ +void lj_asm_trace(jit_State *J, GCtrace *T) +{ + ASMState as_; + ASMState *as = &as_; + + /* Remove nops/renames left over from ASM restart due to LJ_TRERR_MCODELM. */ + { + IRRef nins = T->nins; + IRIns *ir = &T->ir[nins-1]; + if (ir->o == IR_NOP || ir->o == IR_RENAME) { + do { ir--; nins--; } while (ir->o == IR_NOP || ir->o == IR_RENAME); + T->nins = nins; + } + } + + /* Ensure an initialized instruction beyond the last one for HIOP checks. */ + /* This also allows one RENAME to be added without reallocating curfinal. */ + as->orignins = lj_ir_nextins(J); + lj_ir_nop(&J->cur.ir[as->orignins]); + + /* Setup initial state. Copy some fields to reduce indirections. */ + as->J = J; + as->T = T; + J->curfinal = lj_trace_alloc(J->L, T); /* This copies the IR, too. */ + as->flags = J->flags; + as->loopref = J->loopref; + as->realign = NULL; + as->loopinv = 0; + as->parent = J->parent ? traceref(J, J->parent) : NULL; +#ifdef LUAJIT_RANDOM_RA + (void)lj_prng_u64(&J2G(J)->prng); /* Ensure PRNG step between traces. */ +#endif + + /* Reserve MCode memory. */ + as->mctop = as->mctoporig = lj_mcode_reserve(J, &as->mcbot); + as->mcp = as->mctop; + as->mclim = as->mcbot + MCLIM_REDZONE; + asm_setup_target(as); + + /* + ** This is a loop, because the MCode may have to be (re-)assembled + ** multiple times: + ** + ** 1. as->realign is set (and the assembly aborted), if the arch-specific + ** backend wants the MCode to be aligned differently. + ** + ** This is currently only the case on x86/x64, where small loops get + ** an aligned loop body plus a short branch. Not much effort is wasted, + ** because the abort happens very quickly and only once. + ** + ** 2. The IR is immovable, since the MCode embeds pointers to various + ** constants inside the IR. But RENAMEs may need to be added to the IR + ** during assembly, which might grow and reallocate the IR. We check + ** at the end if the IR (in J->cur.ir) has actually grown, resize the + ** copy (in J->curfinal.ir) and try again. + ** + ** 95% of all traces have zero RENAMEs, 3% have one RENAME, 1.5% have + ** 2 RENAMEs and only 0.5% have more than that. That's why we opt to + ** always have one spare slot in the IR (see above), which means we + ** have to redo the assembly for only ~2% of all traces. + ** + ** Very, very rarely, this needs to be done repeatedly, since the + ** location of constants inside the IR (actually, reachability from + ** a global pointer) may affect register allocation and thus the + ** number of RENAMEs. + */ + for (;;) { + as->mcp = as->mctop; +#ifdef LUA_USE_ASSERT + as->mcp_prev = as->mcp; +#endif + as->ir = J->curfinal->ir; /* Use the copied IR. */ + as->curins = J->cur.nins = as->orignins; +#ifdef LUAJIT_RANDOM_RA + as->prngstate = J2G(J)->prng; /* Must (re)start from identical state. */ + as->prngbits = 0; +#endif + + RA_DBG_START(); + RA_DBGX((as, "===== STOP =====")); + + /* General trace setup. Emit tail of trace. */ + asm_tail_prep(as); + as->mcloop = NULL; + as->flagmcp = NULL; + as->topslot = 0; + as->gcsteps = 0; + as->sectref = as->loopref; + as->fuseref = (as->flags & JIT_F_OPT_FUSE) ? as->loopref : FUSE_DISABLED; + asm_setup_regsp(as); + if (!as->loopref) + asm_tail_link(as); + + /* Assemble a trace in linear backwards order. */ + for (as->curins--; as->curins > as->stopins; as->curins--) { + IRIns *ir = IR(as->curins); + /* 64 bit types handled by SPLIT for 32 bit archs. */ + lj_assertA(!(LJ_32 && irt_isint64(ir->t)), + "IR %04d has unsplit 64 bit type", + (int)(ir - as->ir) - REF_BIAS); + asm_snap_prev(as); + if (!ra_used(ir) && !ir_sideeff(ir) && (as->flags & JIT_F_OPT_DCE)) + continue; /* Dead-code elimination can be soooo easy. */ + if (irt_isguard(ir->t)) + asm_snap_prep(as); + RA_DBG_REF(); + checkmclim(as); + asm_ir(as, ir); + } + + if (as->realign && J->curfinal->nins >= T->nins) + continue; /* Retry in case only the MCode needs to be realigned. */ + + /* Emit head of trace. */ + RA_DBG_REF(); + checkmclim(as); + if (as->gcsteps > 0) { + as->curins = as->T->snap[0].ref; + asm_snap_prep(as); /* The GC check is a guard. */ + asm_gc_check(as); + as->curins = as->stopins; + } + ra_evictk(as); + if (as->parent) + asm_head_side(as); + else + asm_head_root(as); + asm_phi_fixup(as); + + if (J->curfinal->nins >= T->nins) { /* IR didn't grow? */ + lj_assertA(J->curfinal->nk == T->nk, "unexpected IR constant growth"); + memcpy(J->curfinal->ir + as->orignins, T->ir + as->orignins, + (T->nins - as->orignins) * sizeof(IRIns)); /* Copy RENAMEs. */ + T->nins = J->curfinal->nins; + /* Fill mcofs of any unprocessed snapshots. */ + as->curins = REF_FIRST; + asm_snap_prev(as); + break; /* Done. */ + } + + /* Otherwise try again with a bigger IR. */ + lj_trace_free(J2G(J), J->curfinal); + J->curfinal = NULL; /* In case lj_trace_alloc() OOMs. */ + J->curfinal = lj_trace_alloc(J->L, T); + as->realign = NULL; + } + + RA_DBGX((as, "===== START ====")); + RA_DBG_FLUSH(); + if (as->freeset != RSET_ALL) + lj_trace_err(as->J, LJ_TRERR_BADRA); /* Ouch! Should never happen. */ + + /* Set trace entry point before fixing up tail to allow link to self. */ + T->mcode = as->mcp; + T->mcloop = as->mcloop ? (MSize)((char *)as->mcloop - (char *)as->mcp) : 0; + if (as->loopref) + asm_loop_tail_fixup(as); + else + asm_tail_fixup(as, T->link); /* Note: this may change as->mctop! */ + T->szmcode = (MSize)((char *)as->mctop - (char *)as->mcp); + asm_snap_fixup_mcofs(as); +#if LJ_TARGET_MCODE_FIXUP + asm_mcode_fixup(T->mcode, T->szmcode); +#endif + lj_mcode_sync(T->mcode, as->mctoporig); +} + +#undef IR + +#endif diff --git a/src/bluajit/lj_bcread.c b/src/bluajit/lj_bcread.c new file mode 100644 index 000000000..54f483c71 --- /dev/null +++ b/src/bluajit/lj_bcread.c @@ -0,0 +1,456 @@ +/* +** Bytecode reader. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#define lj_bcread_c +#define LUA_CORE + +#include "lj_obj.h" +#include "lj_gc.h" +#include "lj_err.h" +#include "lj_buf.h" +#include "lj_str.h" +#include "lj_tab.h" +#include "lj_bc.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#include "lj_cdata.h" +#include "lualib.h" +#endif +#include "lj_lex.h" +#include "lj_bcdump.h" +#include "lj_state.h" +#include "lj_strfmt.h" + +/* Reuse some lexer fields for our own purposes. */ +#define bcread_flags(ls) ls->level +#define bcread_swap(ls) \ + ((bcread_flags(ls) & BCDUMP_F_BE) != LJ_BE*BCDUMP_F_BE) +#define bcread_oldtop(L, ls) restorestack(L, ls->lastline) +#define bcread_savetop(L, ls, top) \ + ls->lastline = (BCLine)savestack(L, (top)) + +/* -- Input buffer handling ----------------------------------------------- */ + +/* Throw reader error. */ +static LJ_NOINLINE void bcread_error(LexState *ls, ErrMsg em) +{ + lua_State *L = ls->L; + const char *name = ls->chunkarg; + if (*name == BCDUMP_HEAD1) name = "(binary)"; + else if (*name == '@' || *name == '=') name++; + lj_strfmt_pushf(L, "%s: %s", name, err2msg(em)); + lj_err_throw(L, LUA_ERRSYNTAX); +} + +/* Refill buffer. */ +static LJ_NOINLINE void bcread_fill(LexState *ls, MSize len, int need) +{ + lj_assertLS(len != 0, "empty refill"); + if (len > LJ_MAX_BUF || ls->c < 0) + bcread_error(ls, LJ_ERR_BCBAD); + do { + const char *buf; + size_t sz; + char *p = ls->sb.b; + MSize n = (MSize)(ls->pe - ls->p); + if (n) { /* Copy remainder to buffer. */ + if (sbuflen(&ls->sb)) { /* Move down in buffer. */ + lj_assertLS(ls->pe == ls->sb.w, "bad buffer pointer"); + if (ls->p != p) memmove(p, ls->p, n); + } else { /* Copy from buffer provided by reader. */ + p = lj_buf_need(&ls->sb, len); + memcpy(p, ls->p, n); + } + ls->p = p; + ls->pe = p + n; + } + ls->sb.w = p + n; + buf = ls->rfunc(ls->L, ls->rdata, &sz); /* Get more data from reader. */ + if (buf == NULL || sz == 0) { /* EOF? */ + if (need) bcread_error(ls, LJ_ERR_BCBAD); + ls->c = -1; /* Only bad if we get called again. */ + break; + } + if (sz >= LJ_MAX_BUF - n) lj_err_mem(ls->L); + if (n) { /* Append to buffer. */ + n += (MSize)sz; + p = lj_buf_need(&ls->sb, n < len ? len : n); + memcpy(ls->sb.w, buf, sz); + ls->sb.w = p + n; + ls->p = p; + ls->pe = p + n; + } else { /* Return buffer provided by reader. */ + ls->p = buf; + ls->pe = buf + sz; + } + } while ((MSize)(ls->pe - ls->p) < len); +} + +/* Need a certain number of bytes. */ +static LJ_AINLINE void bcread_need(LexState *ls, MSize len) +{ + if (LJ_UNLIKELY((MSize)(ls->pe - ls->p) < len)) + bcread_fill(ls, len, 1); +} + +/* Want to read up to a certain number of bytes, but may need less. */ +static LJ_AINLINE void bcread_want(LexState *ls, MSize len) +{ + if (LJ_UNLIKELY((MSize)(ls->pe - ls->p) < len)) + bcread_fill(ls, len, 0); +} + +/* Return memory block from buffer. */ +static LJ_AINLINE uint8_t *bcread_mem(LexState *ls, MSize len) +{ + uint8_t *p = (uint8_t *)ls->p; + ls->p += len; + lj_assertLS(ls->p <= ls->pe, "buffer read overflow"); + return p; +} + +/* Copy memory block from buffer. */ +static void bcread_block(LexState *ls, void *q, MSize len) +{ + memcpy(q, bcread_mem(ls, len), len); +} + +/* Read byte from buffer. */ +static LJ_AINLINE uint32_t bcread_byte(LexState *ls) +{ + lj_assertLS(ls->p < ls->pe, "buffer read overflow"); + return (uint32_t)(uint8_t)*ls->p++; +} + +/* Read ULEB128 value from buffer. */ +static LJ_AINLINE uint32_t bcread_uleb128(LexState *ls) +{ + uint32_t v = lj_buf_ruleb128(&ls->p); + lj_assertLS(ls->p <= ls->pe, "buffer read overflow"); + return v; +} + +/* Read top 32 bits of 33 bit ULEB128 value from buffer. */ +static uint32_t bcread_uleb128_33(LexState *ls) +{ + const uint8_t *p = (const uint8_t *)ls->p; + uint32_t v = (*p++ >> 1); + if (LJ_UNLIKELY(v >= 0x40)) { + int sh = -1; + v &= 0x3f; + do { + v |= ((*p & 0x7f) << (sh += 7)); + } while (*p++ >= 0x80); + } + ls->p = (char *)p; + lj_assertLS(ls->p <= ls->pe, "buffer read overflow"); + return v; +} + +/* -- Bytecode reader ----------------------------------------------------- */ + +/* Read debug info of a prototype. */ +static void bcread_dbg(LexState *ls, GCproto *pt, MSize sizedbg) +{ + void *lineinfo = (void *)proto_lineinfo(pt); + bcread_block(ls, lineinfo, sizedbg); + /* Swap lineinfo if the endianess differs. */ + if (bcread_swap(ls) && pt->numline >= 256) { + MSize i, n = pt->sizebc-1; + if (pt->numline < 65536) { + uint16_t *p = (uint16_t *)lineinfo; + for (i = 0; i < n; i++) p[i] = (uint16_t)((p[i] >> 8)|(p[i] << 8)); + } else { + uint32_t *p = (uint32_t *)lineinfo; + for (i = 0; i < n; i++) p[i] = lj_bswap(p[i]); + } + } +} + +/* Find pointer to varinfo. */ +static const void *bcread_varinfo(GCproto *pt) +{ + const uint8_t *p = proto_uvinfo(pt); + MSize n = pt->sizeuv; + if (n) while (*p++ || --n) ; + return p; +} + +/* Read a single constant key/value of a template table. */ +static void bcread_ktabk(LexState *ls, TValue *o) +{ + MSize tp = bcread_uleb128(ls); + if (tp >= BCDUMP_KTAB_STR) { + MSize len = tp - BCDUMP_KTAB_STR; + const char *p = (const char *)bcread_mem(ls, len); + setstrV(ls->L, o, lj_str_new(ls->L, p, len)); + } else if (tp == BCDUMP_KTAB_INT) { + setintV(o, (int32_t)bcread_uleb128(ls)); + } else if (tp == BCDUMP_KTAB_NUM) { + o->u32.lo = bcread_uleb128(ls); + o->u32.hi = bcread_uleb128(ls); + } else { + lj_assertLS(tp <= BCDUMP_KTAB_TRUE, "bad constant type %d", tp); + setpriV(o, ~tp); + } +} + +/* Read a template table. */ +static GCtab *bcread_ktab(LexState *ls) +{ + MSize narray = bcread_uleb128(ls); + MSize nhash = bcread_uleb128(ls); + GCtab *t = lj_tab_new(ls->L, narray, hsize2hbits(nhash)); + if (narray) { /* Read array entries. */ + MSize i; + TValue *o = tvref(t->array); + for (i = 0; i < narray; i++, o++) + bcread_ktabk(ls, o); + } + if (nhash) { /* Read hash entries. */ + MSize i; + for (i = 0; i < nhash; i++) { + TValue key; + bcread_ktabk(ls, &key); + lj_assertLS(!tvisnil(&key), "nil key"); + bcread_ktabk(ls, lj_tab_set(ls->L, t, &key)); + } + } + return t; +} + +/* Read GC constants of a prototype. */ +static void bcread_kgc(LexState *ls, GCproto *pt, MSize sizekgc) +{ + MSize i; + GCRef *kr = mref(pt->k, GCRef) - (ptrdiff_t)sizekgc; + for (i = 0; i < sizekgc; i++, kr++) { + MSize tp = bcread_uleb128(ls); + if (tp >= BCDUMP_KGC_STR) { + MSize len = tp - BCDUMP_KGC_STR; + const char *p = (const char *)bcread_mem(ls, len); + setgcref(*kr, obj2gco(lj_str_new(ls->L, p, len))); + } else if (tp == BCDUMP_KGC_TAB) { + setgcref(*kr, obj2gco(bcread_ktab(ls))); +#if LJ_HASFFI + } else if (tp != BCDUMP_KGC_CHILD) { + CTypeID id = tp == BCDUMP_KGC_COMPLEX ? CTID_COMPLEX_DOUBLE : + tp == BCDUMP_KGC_I64 ? CTID_INT64 : CTID_UINT64; + CTSize sz = tp == BCDUMP_KGC_COMPLEX ? 16 : 8; + GCcdata *cd = lj_cdata_new_(ls->L, id, sz); + TValue *p = (TValue *)cdataptr(cd); + setgcref(*kr, obj2gco(cd)); + p[0].u32.lo = bcread_uleb128(ls); + p[0].u32.hi = bcread_uleb128(ls); + if (tp == BCDUMP_KGC_COMPLEX) { + p[1].u32.lo = bcread_uleb128(ls); + p[1].u32.hi = bcread_uleb128(ls); + } +#endif + } else { + lua_State *L = ls->L; + lj_assertLS(tp == BCDUMP_KGC_CHILD, "bad constant type %d", tp); + if (L->top <= bcread_oldtop(L, ls)) /* Stack underflow? */ + bcread_error(ls, LJ_ERR_BCBAD); + L->top--; + setgcref(*kr, obj2gco(protoV(L->top))); + } + } +} + +/* Read number constants of a prototype. */ +static void bcread_knum(LexState *ls, GCproto *pt, MSize sizekn) +{ + MSize i; + TValue *o = mref(pt->k, TValue); + for (i = 0; i < sizekn; i++, o++) { + int isnum = (ls->p[0] & 1); + uint32_t lo = bcread_uleb128_33(ls); + if (isnum) { + o->u32.lo = lo; + o->u32.hi = bcread_uleb128(ls); + } else { + setintV(o, lo); + } + } +} + +/* Read bytecode instructions. */ +static void bcread_bytecode(LexState *ls, GCproto *pt, MSize sizebc) +{ + BCIns *bc = proto_bc(pt); + BCIns op; + if (ls->fr2 != LJ_FR2) op = BC_NOT; /* Mark non-native prototype. */ + else if ((pt->flags & PROTO_VARARG)) op = BC_FUNCV; + else op = BC_FUNCF; + bc[0] = BCINS_AD(op, pt->framesize, 0); + bcread_block(ls, bc+1, (sizebc-1)*(MSize)sizeof(BCIns)); + /* Swap bytecode instructions if the endianess differs. */ + if (bcread_swap(ls)) { + MSize i; + for (i = 1; i < sizebc; i++) bc[i] = lj_bswap(bc[i]); + } +} + +/* Read upvalue refs. */ +static void bcread_uv(LexState *ls, GCproto *pt, MSize sizeuv) +{ + if (sizeuv) { + uint16_t *uv = proto_uv(pt); + bcread_block(ls, uv, sizeuv*2); + /* Swap upvalue refs if the endianess differs. */ + if (bcread_swap(ls)) { + MSize i; + for (i = 0; i < sizeuv; i++) + uv[i] = (uint16_t)((uv[i] >> 8)|(uv[i] << 8)); + } + } +} + +/* Read a prototype. */ +GCproto *lj_bcread_proto(LexState *ls) +{ + GCproto *pt; + MSize framesize, numparams, flags, sizeuv, sizekgc, sizekn, sizebc, sizept; + MSize ofsk, ofsuv, ofsdbg; + MSize sizedbg = 0; + BCLine firstline = 0, numline = 0; + + /* Read prototype header. */ + flags = bcread_byte(ls); + numparams = bcread_byte(ls); + framesize = bcread_byte(ls); + sizeuv = bcread_byte(ls); + sizekgc = bcread_uleb128(ls); + sizekn = bcread_uleb128(ls); + sizebc = bcread_uleb128(ls) + 1; + if (!(bcread_flags(ls) & BCDUMP_F_STRIP)) { + sizedbg = bcread_uleb128(ls); + if (sizedbg) { + firstline = bcread_uleb128(ls); + numline = bcread_uleb128(ls); + } + } + + /* Calculate total size of prototype including all colocated arrays. */ + sizept = (MSize)sizeof(GCproto) + + sizebc*(MSize)sizeof(BCIns) + + sizekgc*(MSize)sizeof(GCRef); + sizept = (sizept + (MSize)sizeof(TValue)-1) & ~((MSize)sizeof(TValue)-1); + ofsk = sizept; sizept += sizekn*(MSize)sizeof(TValue); + ofsuv = sizept; sizept += ((sizeuv+1)&~1)*2; + ofsdbg = sizept; sizept += sizedbg; + + /* Allocate prototype object and initialize its fields. */ + pt = (GCproto *)lj_mem_newgco(ls->L, (MSize)sizept); + pt->gct = ~LJ_TPROTO; + pt->numparams = (uint8_t)numparams; + pt->framesize = (uint8_t)framesize; + pt->sizebc = sizebc; + setmref(pt->k, (char *)pt + ofsk); + setmref(pt->uv, (char *)pt + ofsuv); + pt->sizekgc = 0; /* Set to zero until fully initialized. */ + pt->sizekn = sizekn; + pt->sizept = sizept; + pt->sizeuv = (uint8_t)sizeuv; + pt->flags = (uint8_t)flags; + pt->trace = 0; + setgcref(pt->chunkname, obj2gco(ls->chunkname)); + + /* Close potentially uninitialized gap between bc and kgc. */ + *(uint32_t *)((char *)pt + ofsk - sizeof(GCRef)*(sizekgc+1)) = 0; + + /* Read bytecode instructions and upvalue refs. */ + bcread_bytecode(ls, pt, sizebc); + bcread_uv(ls, pt, sizeuv); + + /* Read constants. */ + bcread_kgc(ls, pt, sizekgc); + pt->sizekgc = sizekgc; + bcread_knum(ls, pt, sizekn); + + /* Read and initialize debug info. */ + pt->firstline = firstline; + pt->numline = numline; + if (sizedbg) { + MSize sizeli = (sizebc-1) << (numline < 256 ? 0 : numline < 65536 ? 1 : 2); + setmref(pt->lineinfo, (char *)pt + ofsdbg); + setmref(pt->uvinfo, (char *)pt + ofsdbg + sizeli); + bcread_dbg(ls, pt, sizedbg); + setmref(pt->varinfo, bcread_varinfo(pt)); + } else { + setmref(pt->lineinfo, NULL); + setmref(pt->uvinfo, NULL); + setmref(pt->varinfo, NULL); + } + return pt; +} + +/* Read and check header of bytecode dump. */ +static int bcread_header(LexState *ls) +{ + uint32_t flags; + bcread_want(ls, 3+5+5); + if (bcread_byte(ls) != BCDUMP_HEAD2 || + bcread_byte(ls) != BCDUMP_HEAD3 || + bcread_byte(ls) != BCDUMP_VERSION) return 0; + bcread_flags(ls) = flags = bcread_uleb128(ls); + if ((flags & ~(BCDUMP_F_KNOWN)) != 0) return 0; + if ((flags & BCDUMP_F_FR2) != (uint32_t)ls->fr2*BCDUMP_F_FR2) return 0; + if ((flags & BCDUMP_F_FFI)) { +#if LJ_HASFFI + lua_State *L = ls->L; + ctype_loadffi(L); +#else + return 0; +#endif + } + if ((flags & BCDUMP_F_STRIP)) { + ls->chunkname = lj_str_newz(ls->L, *ls->chunkarg == BCDUMP_HEAD1 ? "=?" : ls->chunkarg); + } else { + MSize len = bcread_uleb128(ls); + bcread_need(ls, len); + ls->chunkname = lj_str_new(ls->L, (const char *)bcread_mem(ls, len), len); + } + return 1; /* Ok. */ +} + +/* Read a bytecode dump. */ +GCproto *lj_bcread(LexState *ls) +{ + lua_State *L = ls->L; + lj_assertLS(ls->c == BCDUMP_HEAD1, "bad bytecode header"); + bcread_savetop(L, ls, L->top); + lj_buf_reset(&ls->sb); + /* Check for a valid bytecode dump header. */ + if (!bcread_header(ls)) + bcread_error(ls, LJ_ERR_BCFMT); + for (;;) { /* Process all prototypes in the bytecode dump. */ + GCproto *pt; + MSize len; + const char *startp; + /* Read length. */ + if (ls->p < ls->pe && ls->p[0] == 0) { /* Shortcut EOF. */ + ls->p++; + break; + } + bcread_want(ls, 5); + len = bcread_uleb128(ls); + if (!len) break; /* EOF */ + bcread_need(ls, len); + startp = ls->p; + pt = lj_bcread_proto(ls); + if (ls->p != startp + len) + bcread_error(ls, LJ_ERR_BCBAD); + setprotoV(L, L->top, pt); + incr_top(L); + } + if ((ls->pe != ls->p && !ls->endmark) || L->top-1 != bcread_oldtop(L, ls)) + bcread_error(ls, LJ_ERR_BCBAD); + /* Pop off last prototype. */ + L->top--; + return protoV(L->top); +} + diff --git a/src/bluajit/lj_parse.c b/src/bluajit/lj_parse.c new file mode 100644 index 000000000..d5eec92b4 --- /dev/null +++ b/src/bluajit/lj_parse.c @@ -0,0 +1,3056 @@ +/* +** Lua parser (source code -> bytecode). +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +** +** Major portions taken verbatim or adapted from the Lua interpreter. +** Copyright (C) 1994-2008 Lua.org, PUC-Rio. See Copyright Notice in lua.h +*/ + +#define lj_parse_c +#define LUA_CORE + +#include "lj_obj.h" +#include "lj_gc.h" +#include "lj_err.h" +#include "lj_debug.h" +#include "lj_buf.h" +#include "lj_str.h" +#include "lj_tab.h" +#include "lj_func.h" +#include "lj_state.h" +#include "lj_bc.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#endif +#include "lj_strfmt.h" +#include "lj_lex.h" +#include "lj_parse.h" +#include "lj_vm.h" +#include "lj_vmevent.h" + +/* -- Parser structures and definitions ----------------------------------- */ + +/* Expression kinds. */ +typedef enum { + /* Constant expressions must be first and in this order: */ + VKNIL, + VKFALSE, + VKTRUE, + VKSTR, /* sval = string value */ + VKNUM, /* nval = number value */ + VKLAST = VKNUM, + VKCDATA, /* nval = cdata value, not treated as a constant expression */ + /* Non-constant expressions follow: */ + VLOCAL, /* info = local register, aux = vstack index */ + VUPVAL, /* info = upvalue index, aux = vstack index */ + VGLOBAL, /* sval = string value */ + VINDEXED, /* info = table register, aux = index reg/byte/string const */ + VJMP, /* info = instruction PC */ + VRELOCABLE, /* info = instruction PC */ + VNONRELOC, /* info = result register */ + VCALL, /* info = instruction PC, aux = base */ + VVOID +} ExpKind; + +/* Expression descriptor. */ +typedef struct ExpDesc { + union { + struct { + uint32_t info; /* Primary info. */ + uint32_t aux; /* Secondary info. */ + } s; + TValue nval; /* Number value. */ + GCstr *sval; /* String value. */ + } u; + ExpKind k; + BCPos t; /* True condition jump list. */ + BCPos f; /* False condition jump list. */ +} ExpDesc; + +/* Macros for expressions. */ +#define expr_hasjump(e) ((e)->t != (e)->f) + +#define expr_isk(e) ((e)->k <= VKLAST) +#define expr_isk_nojump(e) (expr_isk(e) && !expr_hasjump(e)) +#define expr_isnumk(e) ((e)->k == VKNUM) +#define expr_isnumk_nojump(e) (expr_isnumk(e) && !expr_hasjump(e)) +#define expr_isstrk(e) ((e)->k == VKSTR) + +#define expr_numtv(e) check_exp(expr_isnumk((e)), &(e)->u.nval) +#define expr_numberV(e) numberVnum(expr_numtv((e))) +#define expr_intV(e) intV(expr_numtv((e))) + +/* Initialize expression. */ +static LJ_AINLINE void expr_init(ExpDesc *e, ExpKind k, uint32_t info) +{ + e->k = k; + e->u.s.info = info; + e->f = e->t = NO_JMP; +} + +/* Check number constant for +-0. */ +static int expr_numiszero(ExpDesc *e) +{ + TValue *o = expr_numtv(e); + return tvisint(o) ? (intV(o) == 0) : tviszero(o); +} + +/* Per-function linked list of scope blocks. */ +typedef struct FuncScope { + struct FuncScope *prev; /* Link to outer scope. */ + MSize vstart; /* Start of block-local variables. */ + uint8_t nactvar; /* Number of active vars outside the scope. */ + uint8_t flags; /* Scope flags. */ +} FuncScope; + +#define FSCOPE_LOOP 0x01 /* Scope is a (breakable) loop. */ +#define FSCOPE_BREAK 0x02 /* Break used in scope. */ +#define FSCOPE_GOLA 0x04 /* Goto or label used in scope. */ +#define FSCOPE_UPVAL 0x08 /* Upvalue in scope. */ +#define FSCOPE_NOCLOSE 0x10 /* Do not close upvalues. */ + +/* Predefined named for special labels.*/ +/** `break` label. */ +#define NAME_BREAK ((GCstr *)(uintptr_t)1) + +#if BLUAJIT_CONTINUE_STATEMENT + /** `continue` label. */ + #define NAME_CONTINUE ((GCstr *)(uintptr_t)2) + + #define FSCOPE_CONTINUE 0x20 /* Continue used in scope. */ +#endif + +/* Index into variable stack. */ +typedef uint16_t VarIndex; +#define LJ_MAX_VSTACK (65536 - LJ_MAX_UPVAL) + +/* Variable/goto/label info. */ +#define VSTACK_VAR_RW 0x01 /* R/W variable. */ +#define VSTACK_GOTO 0x02 /* Pending goto. */ +#define VSTACK_LABEL 0x04 /* Label. */ + +/* List of LHS variables. */ +typedef struct LHSVarList { + ExpDesc v; /* LHS variable. */ + struct LHSVarList *prev; /* Link to previous LHS variable. */ +} LHSVarList; + +/* Per-function state. */ +typedef struct FuncState { + GCtab *kt; /* Hash table for constants. */ + LexState *ls; /* Lexer state. */ + lua_State *L; /* Lua state. */ + FuncScope *bl; /* Current scope. */ + struct FuncState *prev; /* Enclosing function. */ + LHSVarList *lhs; /* Chain of left hand sides during assign. */ + BCReg nlhs; /* Number of LHS in chain. */ + BCReg nrhs; /* Number of RHS in assignment. */ + BCPos pc; /* Next bytecode position. */ + BCPos lasttarget; /* Bytecode position of last jump target. */ + BCPos jpc; /* Pending jump list to next bytecode. */ + BCReg freereg; /* First free register. */ + BCReg nactvar; /* Number of active local variables. */ + BCReg nkn, nkgc; /* Number of lua_Number/GCobj constants */ + BCLine linedefined; /* First line of the function definition. */ + BCInsLine *bcbase; /* Base of bytecode stack. */ + BCPos bclim; /* Limit of bytecode stack. */ + MSize vbase; /* Base of variable stack for this function. */ + uint8_t flags; /* Prototype flags. */ + uint8_t numparams; /* Number of parameters. */ + uint8_t framesize; /* Fixed frame size. */ + uint8_t nuv; /* Number of upvalues */ + VarIndex varmap[LJ_MAX_LOCVAR]; /* Map from register to variable idx. */ + VarIndex uvmap[LJ_MAX_UPVAL]; /* Map from upvalue to variable idx. */ + VarIndex uvtmp[LJ_MAX_UPVAL]; /* Temporary upvalue map. */ +} FuncState; + +/* Binary and unary operators. ORDER OPR */ +typedef enum BinOpr { + OPR_ADD, OPR_SUB, OPR_MUL, OPR_DIV, OPR_MOD, OPR_POW, /* ORDER ARITH */ + OPR_BAND, OPR_BOR, OPR_BXOR, OPR_SHL, OPR_SHR, /* ORDER BITWISE */ + OPR_CONCAT, + OPR_NE, OPR_EQ, + OPR_LT, OPR_GE, OPR_LE, OPR_GT, + OPR_AND, OPR_OR, + OPR_NOBINOPR +} BinOpr; + +LJ_STATIC_ASSERT((int)BC_ISGE-(int)BC_ISLT == (int)OPR_GE-(int)OPR_LT); +LJ_STATIC_ASSERT((int)BC_ISLE-(int)BC_ISLT == (int)OPR_LE-(int)OPR_LT); +LJ_STATIC_ASSERT((int)BC_ISGT-(int)BC_ISLT == (int)OPR_GT-(int)OPR_LT); +LJ_STATIC_ASSERT((int)BC_SUBVV-(int)BC_ADDVV == (int)OPR_SUB-(int)OPR_ADD); +LJ_STATIC_ASSERT((int)BC_MULVV-(int)BC_ADDVV == (int)OPR_MUL-(int)OPR_ADD); +LJ_STATIC_ASSERT((int)BC_DIVVV-(int)BC_ADDVV == (int)OPR_DIV-(int)OPR_ADD); +LJ_STATIC_ASSERT((int)BC_MODVV-(int)BC_ADDVV == (int)OPR_MOD-(int)OPR_ADD); +LJ_STATIC_ASSERT((int)BC_POW-(int)BC_ADDVV == (int)OPR_POW-(int)OPR_ADD); +LJ_STATIC_ASSERT((int)BC_BORVV-(int)BC_BANDVV == (int)OPR_BOR-(int)OPR_BAND); +LJ_STATIC_ASSERT((int)BC_BXORVV-(int)BC_BANDVV == (int)OPR_BXOR-(int)OPR_BAND); +LJ_STATIC_ASSERT((int)BC_SHLVV-(int)BC_BANDVV == (int)OPR_SHL-(int)OPR_BAND); +LJ_STATIC_ASSERT((int)BC_SHRVV-(int)BC_BANDVV == (int)OPR_SHR-(int)OPR_BAND); + +#ifdef LUA_USE_ASSERT +#define lj_assertFS(c, ...) (lj_assertG_(G(fs->L), (c), __VA_ARGS__)) +#else +#define lj_assertFS(c, ...) ((void)fs) +#endif + +/* -- Error handling ------------------------------------------------------ */ + +LJ_NORET LJ_NOINLINE static void err_syntax(LexState *ls, ErrMsg em) +{ + lj_lex_error(ls, ls->tok, em); +} + +LJ_NORET LJ_NOINLINE static void err_token(LexState *ls, LexToken tok) +{ + lj_lex_error(ls, ls->tok, LJ_ERR_XTOKEN, lj_lex_token2str(ls, tok)); +} + +LJ_NORET static void err_limit(FuncState *fs, uint32_t limit, const char *what) +{ + if (fs->linedefined == 0) + lj_lex_error(fs->ls, 0, LJ_ERR_XLIMM, limit, what); + else + lj_lex_error(fs->ls, 0, LJ_ERR_XLIMF, fs->linedefined, limit, what); +} + +#define checklimit(fs, v, l, m) if ((v) >= (l)) err_limit(fs, l, m) +#define checklimitgt(fs, v, l, m) if ((v) > (l)) err_limit(fs, l, m) +#define checkcond(ls, c, em) { if (!(c)) err_syntax(ls, em); } + +/* -- Management of constants --------------------------------------------- */ + +/* Return bytecode encoding for primitive constant. */ +#define const_pri(e) check_exp((e)->k <= VKTRUE, (e)->k) + +#define tvhaskslot(o) ((o)->u32.hi == 0) +#define tvkslot(o) ((o)->u32.lo) + +/* Add a number constant. */ +static BCReg const_num(FuncState *fs, ExpDesc *e) +{ + lua_State *L = fs->L; + TValue *o; + lj_assertFS(expr_isnumk(e), "bad usage"); + o = lj_tab_set(L, fs->kt, &e->u.nval); + if (tvhaskslot(o)) + return tvkslot(o); + o->u64 = fs->nkn; + return fs->nkn++; +} + +/* Add a GC object constant. */ +static BCReg const_gc(FuncState *fs, GCobj *gc, uint32_t itype) +{ + lua_State *L = fs->L; + TValue key, *o; + setgcV(L, &key, gc, itype); + /* NOBARRIER: the key is new or kept alive. */ + o = lj_tab_set(L, fs->kt, &key); + if (tvhaskslot(o)) + return tvkslot(o); + o->u64 = fs->nkgc; + return fs->nkgc++; +} + +/* Add a string constant. */ +static BCReg const_str(FuncState *fs, ExpDesc *e) +{ + lj_assertFS(expr_isstrk(e) || e->k == VGLOBAL, "bad usage"); + return const_gc(fs, obj2gco(e->u.sval), LJ_TSTR); +} + +/* Anchor string constant to avoid GC. */ +GCstr *lj_parse_keepstr(LexState *ls, const char *str, size_t len) +{ + /* NOBARRIER: the key is new or kept alive. */ + lua_State *L = ls->L; + GCstr *s = lj_str_new(L, str, len); + TValue *tv = lj_tab_setstr(L, ls->fs->kt, s); + if (tvisnil(tv)) setboolV(tv, 1); + lj_gc_check(L); + return s; +} + +#if LJ_HASFFI +/* Anchor cdata to avoid GC. */ +void lj_parse_keepcdata(LexState *ls, TValue *tv, GCcdata *cd) +{ + /* NOBARRIER: the key is new or kept alive. */ + lua_State *L = ls->L; + setcdataV(L, tv, cd); + setboolV(lj_tab_set(L, ls->fs->kt, tv), 1); +} +#endif + +/* -- Jump list handling -------------------------------------------------- */ + +/* Get next element in jump list. */ +static BCPos jmp_next(FuncState *fs, BCPos pc) +{ + ptrdiff_t delta = bc_j(fs->bcbase[pc].ins); + if ((BCPos)delta == NO_JMP) + return NO_JMP; + else + return (BCPos)(((ptrdiff_t)pc+1)+delta); +} + +/* Check if any of the instructions on the jump list produce no value. */ +static int jmp_novalue(FuncState *fs, BCPos list) +{ + for (; list != NO_JMP; list = jmp_next(fs, list)) { + BCIns p = fs->bcbase[list >= 1 ? list-1 : list].ins; + if (!(bc_op(p) == BC_ISTC || bc_op(p) == BC_ISFC || bc_a(p) == NO_REG)) + return 1; + } + return 0; +} + +/* Patch register of test instructions. */ +static int jmp_patchtestreg(FuncState *fs, BCPos pc, BCReg reg) +{ + BCInsLine *ilp = &fs->bcbase[pc >= 1 ? pc-1 : pc]; + BCOp op = bc_op(ilp->ins); + if (op == BC_ISTC || op == BC_ISFC) { + if (reg != NO_REG && reg != bc_d(ilp->ins)) { + setbc_a(&ilp->ins, reg); + } else { /* Nothing to store or already in the right register. */ + setbc_op(&ilp->ins, op+(BC_IST-BC_ISTC)); + setbc_a(&ilp->ins, 0); + } + } else if (bc_a(ilp->ins) == NO_REG) { + if (reg == NO_REG) { + ilp->ins = BCINS_AJ(BC_JMP, bc_a(fs->bcbase[pc].ins), 0); + } else { + setbc_a(&ilp->ins, reg); + if (reg >= bc_a(ilp[1].ins)) + setbc_a(&ilp[1].ins, reg+1); + } + } else { + return 0; /* Cannot patch other instructions. */ + } + return 1; +} + +/* Drop values for all instructions on jump list. */ +static void jmp_dropval(FuncState *fs, BCPos list) +{ + for (; list != NO_JMP; list = jmp_next(fs, list)) + jmp_patchtestreg(fs, list, NO_REG); +} + +/* Patch jump instruction to target. */ +static void jmp_patchins(FuncState *fs, BCPos pc, BCPos dest) +{ + BCIns *jmp = &fs->bcbase[pc].ins; + BCPos offset = dest-(pc+1)+BCBIAS_J; + lj_assertFS(dest != NO_JMP, "uninitialized jump target"); + if (offset > BCMAX_D) + err_syntax(fs->ls, LJ_ERR_XJUMP); + setbc_d(jmp, offset); +} + +/* Append to jump list. */ +static void jmp_append(FuncState *fs, BCPos *l1, BCPos l2) +{ + if (l2 == NO_JMP) { + return; + } else if (*l1 == NO_JMP) { + *l1 = l2; + } else { + BCPos list = *l1; + BCPos next; + while ((next = jmp_next(fs, list)) != NO_JMP) /* Find last element. */ + list = next; + jmp_patchins(fs, list, l2); + } +} + +/* Patch jump list and preserve produced values. */ +static void jmp_patchval(FuncState *fs, BCPos list, BCPos vtarget, + BCReg reg, BCPos dtarget) +{ + while (list != NO_JMP) { + BCPos next = jmp_next(fs, list); + if (jmp_patchtestreg(fs, list, reg)) + jmp_patchins(fs, list, vtarget); /* Jump to target with value. */ + else + jmp_patchins(fs, list, dtarget); /* Jump to default target. */ + list = next; + } +} + +/* Jump to following instruction. Append to list of pending jumps. */ +static void jmp_tohere(FuncState *fs, BCPos list) +{ + fs->lasttarget = fs->pc; + jmp_append(fs, &fs->jpc, list); +} + +/* Patch jump list to target. */ +static void jmp_patch(FuncState *fs, BCPos list, BCPos target) +{ + if (target == fs->pc) { + jmp_tohere(fs, list); + } else { + lj_assertFS(target < fs->pc, "bad jump target"); + jmp_patchval(fs, list, target, NO_REG, target); + } +} + +/* -- Bytecode register allocator ----------------------------------------- */ + +/* Bump frame size. */ +static void bcreg_bump(FuncState *fs, BCReg n) +{ + BCReg sz = fs->freereg + n; + if (sz > fs->framesize) { + if (sz >= LJ_MAX_SLOTS) + err_syntax(fs->ls, LJ_ERR_XSLOTS); + fs->framesize = (uint8_t)sz; + } +} + +/* Reserve registers. */ +static void bcreg_reserve(FuncState *fs, BCReg n) +{ + bcreg_bump(fs, n); + fs->freereg += n; +} + +/* Free register. */ +static void bcreg_free(FuncState *fs, BCReg reg) +{ + if (reg >= fs->nactvar) { + fs->freereg--; + lj_assertFS(reg == fs->freereg, "bad regfree"); + } +} + +/* Free register for expression. */ +static void expr_free(FuncState *fs, ExpDesc *e) +{ + if (e->k == VNONRELOC) + bcreg_free(fs, e->u.s.info); +} + +/* -- Bytecode emitter ---------------------------------------------------- */ + +/* Emit bytecode instruction. */ +static BCPos bcemit_INS(FuncState *fs, BCIns ins) +{ + BCPos pc = fs->pc; + LexState *ls = fs->ls; + jmp_patchval(fs, fs->jpc, pc, NO_REG, pc); + fs->jpc = NO_JMP; + if (LJ_UNLIKELY(pc >= fs->bclim)) { + ptrdiff_t base = fs->bcbase - ls->bcstack; + checklimit(fs, ls->sizebcstack, LJ_MAX_BCINS, "bytecode instructions"); + lj_mem_growvec(fs->L, ls->bcstack, ls->sizebcstack, LJ_MAX_BCINS,BCInsLine); + fs->bclim = (BCPos)(ls->sizebcstack - base); + fs->bcbase = ls->bcstack + base; + } + fs->bcbase[pc].ins = ins; + fs->bcbase[pc].line = ls->lastline; + fs->pc = pc+1; + return pc; +} + +#define bcemit_ABC(fs, o, a, b, c) bcemit_INS(fs, BCINS_ABC(o, a, b, c)) +#define bcemit_AD(fs, o, a, d) bcemit_INS(fs, BCINS_AD(o, a, d)) +#define bcemit_AJ(fs, o, a, j) bcemit_INS(fs, BCINS_AJ(o, a, j)) + +#define bcptr(fs, e) (&(fs)->bcbase[(e)->u.s.info].ins) + +/* -- Bytecode emitter for expressions ------------------------------------ */ + +/* Discharge non-constant expression to any register. */ +static void expr_discharge(FuncState *fs, ExpDesc *e) +{ + BCIns ins; + if (e->k == VUPVAL) { + ins = BCINS_AD(BC_UGET, 0, e->u.s.info); + } else if (e->k == VGLOBAL) { + ins = BCINS_AD(BC_GGET, 0, const_str(fs, e)); + } else if (e->k == VINDEXED) { + BCReg rc = e->u.s.aux; + if ((int32_t)rc < 0) { + ins = BCINS_ABC(BC_TGETS, 0, e->u.s.info, ~rc); + } else if (rc > BCMAX_C) { + ins = BCINS_ABC(BC_TGETB, 0, e->u.s.info, rc-(BCMAX_C+1)); + } else { + bcreg_free(fs, rc); + ins = BCINS_ABC(BC_TGETV, 0, e->u.s.info, rc); + } + bcreg_free(fs, e->u.s.info); + } else if (e->k == VCALL) { + e->u.s.info = e->u.s.aux; + e->k = VNONRELOC; + return; + } else if (e->k == VLOCAL) { + e->k = VNONRELOC; + return; + } else { + return; + } + e->u.s.info = bcemit_INS(fs, ins); + e->k = VRELOCABLE; +} + +/* Emit bytecode to set a range of registers to nil. */ +static void bcemit_nil(FuncState *fs, BCReg from, BCReg n) +{ + if (fs->pc > fs->lasttarget) { /* No jumps to current position? */ + BCIns *ip = &fs->bcbase[fs->pc-1].ins; + BCReg pto, pfrom = bc_a(*ip); + switch (bc_op(*ip)) { /* Try to merge with the previous instruction. */ + case BC_KPRI: + if (bc_d(*ip) != ~LJ_TNIL) break; + if (from == pfrom) { + if (n == 1) return; + } else if (from == pfrom+1) { + from = pfrom; + n++; + } else { + break; + } + *ip = BCINS_AD(BC_KNIL, from, from+n-1); /* Replace KPRI. */ + return; + case BC_KNIL: + pto = bc_d(*ip); + if (pfrom <= from && from <= pto+1) { /* Can we connect both ranges? */ + if (from+n-1 > pto) + setbc_d(ip, from+n-1); /* Patch previous instruction range. */ + return; + } + break; + default: + break; + } + } + /* Emit new instruction or replace old instruction. */ + bcemit_INS(fs, n == 1 ? BCINS_AD(BC_KPRI, from, VKNIL) : + BCINS_AD(BC_KNIL, from, from+n-1)); +} + +/* Discharge an expression to a specific register. Ignore branches. */ +static void expr_toreg_nobranch(FuncState *fs, ExpDesc *e, BCReg reg) +{ + BCIns ins; + expr_discharge(fs, e); + if (e->k == VKSTR) { + ins = BCINS_AD(BC_KSTR, reg, const_str(fs, e)); + } else if (e->k == VKNUM) { +#if LJ_DUALNUM + cTValue *tv = expr_numtv(e); + if (tvisint(tv) && checki16(intV(tv))) + ins = BCINS_AD(BC_KSHORT, reg, (BCReg)(uint16_t)intV(tv)); + else +#else + lua_Number n = expr_numberV(e); + int32_t k = lj_num2int(n); + if (checki16(k) && n == (lua_Number)k) + ins = BCINS_AD(BC_KSHORT, reg, (BCReg)(uint16_t)k); + else +#endif + ins = BCINS_AD(BC_KNUM, reg, const_num(fs, e)); +#if LJ_HASFFI + } else if (e->k == VKCDATA) { + fs->flags |= PROTO_FFI; + ins = BCINS_AD(BC_KCDATA, reg, + const_gc(fs, obj2gco(cdataV(&e->u.nval)), LJ_TCDATA)); +#endif + } else if (e->k == VRELOCABLE) { + setbc_a(bcptr(fs, e), reg); + goto noins; + } else if (e->k == VNONRELOC) { + if (reg == e->u.s.info) + goto noins; + ins = BCINS_AD(BC_MOV, reg, e->u.s.info); + } else if (e->k == VKNIL) { + bcemit_nil(fs, reg, 1); + goto noins; + } else if (e->k <= VKTRUE) { + ins = BCINS_AD(BC_KPRI, reg, const_pri(e)); + } else { + lj_assertFS(e->k == VVOID || e->k == VJMP, "bad expr type %d", e->k); + return; + } + bcemit_INS(fs, ins); +noins: + e->u.s.info = reg; + e->k = VNONRELOC; +} + +/* Forward declaration. */ +static BCPos bcemit_jmp(FuncState *fs); + +/* Discharge an expression to a specific register. */ +static void expr_toreg(FuncState *fs, ExpDesc *e, BCReg reg) +{ + expr_toreg_nobranch(fs, e, reg); + if (e->k == VJMP) + jmp_append(fs, &e->t, e->u.s.info); /* Add it to the true jump list. */ + if (expr_hasjump(e)) { /* Discharge expression with branches. */ + BCPos jend, jfalse = NO_JMP, jtrue = NO_JMP; + if (jmp_novalue(fs, e->t) || jmp_novalue(fs, e->f)) { + BCPos jval = (e->k == VJMP) ? NO_JMP : bcemit_jmp(fs); + jfalse = bcemit_AD(fs, BC_KPRI, reg, VKFALSE); + bcemit_AJ(fs, BC_JMP, fs->freereg, 1); + jtrue = bcemit_AD(fs, BC_KPRI, reg, VKTRUE); + jmp_tohere(fs, jval); + } + jend = fs->pc; + fs->lasttarget = jend; + jmp_patchval(fs, e->f, jend, reg, jfalse); + jmp_patchval(fs, e->t, jend, reg, jtrue); + } + e->f = e->t = NO_JMP; + e->u.s.info = reg; + e->k = VNONRELOC; +} + +/* Discharge an expression to the next free register. */ +static void expr_tonextreg(FuncState *fs, ExpDesc *e) +{ + expr_discharge(fs, e); + expr_free(fs, e); + bcreg_reserve(fs, 1); + expr_toreg(fs, e, fs->freereg - 1); +} + +/* Discharge an expression to any register. */ +static BCReg expr_toanyreg(FuncState *fs, ExpDesc *e) +{ + expr_discharge(fs, e); + if (e->k == VNONRELOC) { + if (!expr_hasjump(e)) return e->u.s.info; /* Already in a register. */ + if (e->u.s.info >= fs->nactvar) { + expr_toreg(fs, e, e->u.s.info); /* Discharge to temp. register. */ + return e->u.s.info; + } + } + expr_tonextreg(fs, e); /* Discharge to next register. */ + return e->u.s.info; +} + +/* Partially discharge expression to a value. */ +static void expr_toval(FuncState *fs, ExpDesc *e) +{ + if (expr_hasjump(e)) + expr_toanyreg(fs, e); + else + expr_discharge(fs, e); +} + +/* Emit store for LHS expression. */ +static void bcemit_store(FuncState *fs, ExpDesc *var, ExpDesc *e) +{ + BCIns ins; + if (var->k == VLOCAL) { + fs->ls->vstack[var->u.s.aux].info |= VSTACK_VAR_RW; + expr_free(fs, e); + expr_toreg(fs, e, var->u.s.info); + return; + } else if (var->k == VUPVAL) { + fs->ls->vstack[var->u.s.aux].info |= VSTACK_VAR_RW; + expr_toval(fs, e); + if (e->k <= VKTRUE) + ins = BCINS_AD(BC_USETP, var->u.s.info, const_pri(e)); + else if (e->k == VKSTR) + ins = BCINS_AD(BC_USETS, var->u.s.info, const_str(fs, e)); + else if (e->k == VKNUM) + ins = BCINS_AD(BC_USETN, var->u.s.info, const_num(fs, e)); + else + ins = BCINS_AD(BC_USETV, var->u.s.info, expr_toanyreg(fs, e)); + } else if (var->k == VGLOBAL) { + BCReg ra = expr_toanyreg(fs, e); + ins = BCINS_AD(BC_GSET, ra, const_str(fs, var)); + } else { + BCReg ra, rc; + lj_assertFS(var->k == VINDEXED, "bad expr type %d", var->k); + ra = expr_toanyreg(fs, e); + rc = var->u.s.aux; + if ((int32_t)rc < 0) { + ins = BCINS_ABC(BC_TSETS, ra, var->u.s.info, ~rc); + } else if (rc > BCMAX_C) { + ins = BCINS_ABC(BC_TSETB, ra, var->u.s.info, rc-(BCMAX_C+1)); + } else { +#ifdef LUA_USE_ASSERT + /* Free late alloced key reg to avoid assert on free of value reg. */ + /* This can only happen when called from expr_table(). */ + if (e->k == VNONRELOC && ra >= fs->nactvar && rc >= ra) + bcreg_free(fs, rc); +#endif + ins = BCINS_ABC(BC_TSETV, ra, var->u.s.info, rc); + } + } + bcemit_INS(fs, ins); + expr_free(fs, e); +} + +/* Emit method lookup expression. */ +static void bcemit_method(FuncState *fs, ExpDesc *e, ExpDesc *key) +{ + BCReg idx, func, fr2, obj = expr_toanyreg(fs, e); + expr_free(fs, e); + func = fs->freereg; + fr2 = fs->ls->fr2; + bcemit_AD(fs, BC_MOV, func+1+fr2, obj); /* Copy object to 1st argument. */ + lj_assertFS(expr_isstrk(key), "bad usage"); + idx = const_str(fs, key); + if (idx <= BCMAX_C) { + bcreg_reserve(fs, 2+fr2); + bcemit_ABC(fs, BC_TGETS, func, obj, idx); + } else { + bcreg_reserve(fs, 3+fr2); + bcemit_AD(fs, BC_KSTR, func+2+fr2, idx); + bcemit_ABC(fs, BC_TGETV, func, obj, func+2+fr2); + fs->freereg--; + } + e->u.s.info = func; + e->k = VNONRELOC; +} + +/* -- Bytecode emitter for branches --------------------------------------- */ + +/* Emit unconditional branch. */ +static BCPos bcemit_jmp(FuncState *fs) +{ + BCPos jpc = fs->jpc; + BCPos j = fs->pc - 1; + BCIns *ip = &fs->bcbase[j].ins; + fs->jpc = NO_JMP; + if ((int32_t)j >= (int32_t)fs->lasttarget && bc_op(*ip) == BC_UCLO) { + setbc_j(ip, NO_JMP); + fs->lasttarget = j+1; + } else { + j = bcemit_AJ(fs, BC_JMP, fs->freereg, NO_JMP); + } + jmp_append(fs, &j, jpc); + return j; +} + +/* Invert branch condition of bytecode instruction. */ +static void invertcond(FuncState *fs, ExpDesc *e) +{ + BCIns *ip = &fs->bcbase[e->u.s.info - 1].ins; + setbc_op(ip, bc_op(*ip)^1); +} + +/* Emit conditional branch. */ +static BCPos bcemit_branch(FuncState *fs, ExpDesc *e, int cond) +{ + BCPos pc; + if (e->k == VRELOCABLE) { + BCIns *ip = bcptr(fs, e); + if (bc_op(*ip) == BC_NOT) { + *ip = BCINS_AD(cond ? BC_ISF : BC_IST, 0, bc_d(*ip)); + return bcemit_jmp(fs); + } + } + if (e->k != VNONRELOC) { + bcreg_reserve(fs, 1); + expr_toreg_nobranch(fs, e, fs->freereg-1); + } + bcemit_AD(fs, cond ? BC_ISTC : BC_ISFC, NO_REG, e->u.s.info); + pc = bcemit_jmp(fs); + expr_free(fs, e); + return pc; +} + +/* Emit branch on true condition. */ +static void bcemit_branch_t(FuncState *fs, ExpDesc *e) +{ + BCPos pc; + expr_discharge(fs, e); + if (e->k == VKSTR || e->k == VKNUM || e->k == VKTRUE) + pc = NO_JMP; /* Never jump. */ + else if (e->k == VJMP) + invertcond(fs, e), pc = e->u.s.info; + else if (e->k == VKFALSE || e->k == VKNIL) + expr_toreg_nobranch(fs, e, NO_REG), pc = bcemit_jmp(fs); + else + pc = bcemit_branch(fs, e, 0); + jmp_append(fs, &e->f, pc); + jmp_tohere(fs, e->t); + e->t = NO_JMP; +} + +/* Emit branch on false condition. */ +static void bcemit_branch_f(FuncState *fs, ExpDesc *e) +{ + BCPos pc; + expr_discharge(fs, e); + if (e->k == VKNIL || e->k == VKFALSE) + pc = NO_JMP; /* Never jump. */ + else if (e->k == VJMP) + pc = e->u.s.info; + else if (e->k == VKSTR || e->k == VKNUM || e->k == VKTRUE) + expr_toreg_nobranch(fs, e, NO_REG), pc = bcemit_jmp(fs); + else + pc = bcemit_branch(fs, e, 1); + jmp_append(fs, &e->t, pc); + jmp_tohere(fs, e->f); + e->f = NO_JMP; +} + +/* -- Bytecode emitter for operators -------------------------------------- */ + +/* Try constant-folding of arithmetic operators. */ +static int foldarith(BinOpr opr, ExpDesc *e1, ExpDesc *e2) +{ + if (!expr_isnumk_nojump(e1) || !expr_isnumk_nojump(e2)) return 0; +#if LJ_INTONLY + /* Don't fold division by zero. */ + if ((opr == OPR_DIV || opr == OPR_MOD) && !expr_intV(e2)) return 0; + setintV(&e1->u.nval, + lj_vm_foldarith_int(expr_intV(e1), expr_intV(e2), (int)opr-OPR_ADD)); +#else + lua_Number n; + TValue o; + n = lj_vm_foldarith(expr_numberV(e1), expr_numberV(e2), (int)opr-OPR_ADD); + setnumV(&o, n); + if (tvisnan(&o) || tvismzero(&o)) return 0; /* Avoid NaN and -0 as consts. */ + if (LJ_DUALNUM) { + int32_t k = lj_num2int(n); + if ((lua_Number)k == n) { + setintV(&e1->u.nval, k); + return 1; + } + } + setnumV(&e1->u.nval, n); +#endif + return 1; +} + +/* Try constant-folding of bitwise operators. */ +static int foldbitwise(FuncState *fs, BinOpr opr, ExpDesc *e1, ExpDesc *e2) +{ + int64_t v1, v2, n; + if (expr_hasjump(e1) || expr_hasjump(e2)) return 0; +#if LJ_HASFFI + if (e1->k == VKCDATA) { + GCcdata *cd1 = cdataV(&e1->u.nval); + if (cd1->ctypeid != CTID_INT64 && cd1->ctypeid != CTID_UINT64) + return 0; + v1 = *(int64_t *)cdataptr(cd1); + } else +#endif + if (expr_isnumk(e1)) { + TValue *o1 = expr_numtv(e1); + v1 = numberVnum(o1); + if ((lua_Number)v1 != numberVnum(o1)) + lj_err_msg(fs->L, LJ_ERR_NOINT); + } + else return 0; +#if LJ_HASFFI + if (e2->k == VKCDATA) { + GCcdata *cd2 = cdataV(&e2->u.nval); + if (cd2->ctypeid != CTID_INT64 && cd2->ctypeid != CTID_UINT64) + return 0; + v2 = *(int64_t *)cdataptr(cd2); + } else +#endif + if (expr_isnumk(e2)) { + TValue *o2 = expr_numtv(e2); + v2 = numberVnum(o2); + if ((lua_Number)v2 != numberVnum(o2)) + lj_err_msg(fs->L, LJ_ERR_NOINT); + } + else return 0; + + n = lj_vm_foldbitwise(v1, v2, (int)opr-OPR_BAND); +#if LJ_HASFFI + if (e1->k == VKCDATA) { + GCcdata *cd1 = cdataV(&e1->u.nval); + *(int64_t *)cdataptr(cd1) = n; + } else if (e2->k == VKCDATA) { + GCcdata *cd2 = cdataV(&e1->u.nval); + *(int64_t *)cdataptr(cd2) = n; + memcpy(e1, e2, sizeof(ExpDesc)); + } else +#endif + { + int32_t k = lj_num2int(n); + if ((lua_Number)k == n) + setintV(&e1->u.nval, k); + else + setnumV(&e1->u.nval, n); + } + return 1; +} + +/* Emit arithmetic operator. */ +static void bcemit_arith(FuncState *fs, BinOpr opr, ExpDesc *e1, ExpDesc *e2) +{ + BCReg rb, rc, t; + uint32_t op; + if (opr >= OPR_BAND && opr <= OPR_SHR ? foldbitwise(fs, opr, e1, e2) : foldarith(opr, e1, e2)) + return; + if (opr == OPR_POW) { + op = BC_POW; + rc = expr_toanyreg(fs, e2); + rb = expr_toanyreg(fs, e1); + } else { + if (opr >= OPR_BAND && opr <= OPR_SHR) + op = opr-OPR_BAND+BC_BANDVV; + else + op = opr-OPR_ADD+BC_ADDVV; + /* Must discharge 2nd operand first since VINDEXED might free regs. */ + expr_toval(fs, e2); + if (expr_isnumk(e2) && (rc = const_num(fs, e2)) <= BCMAX_C) + op -= BC_ADDVV-BC_ADDVN; + else + rc = expr_toanyreg(fs, e2); + /* 1st operand discharged by bcemit_binop_left, but need KNUM/KSHORT. */ + lj_assertFS(expr_isnumk(e1) || e1->k == VNONRELOC, + "bad expr type %d", e1->k); + expr_toval(fs, e1); + /* Avoid two consts to satisfy bytecode constraints. */ + if (expr_isnumk(e1) && !expr_isnumk(e2) && + (t = const_num(fs, e1)) <= BCMAX_B) { + rb = rc; rc = t; op -= BC_ADDVV-BC_ADDNV; + } else { + rb = expr_toanyreg(fs, e1); + } + } + /* Using expr_free might cause asserts if the order is wrong. */ + if (e1->k == VNONRELOC && e1->u.s.info >= fs->nactvar) fs->freereg--; + if (e2->k == VNONRELOC && e2->u.s.info >= fs->nactvar) fs->freereg--; + e1->u.s.info = bcemit_ABC(fs, op, 0, rb, rc); + e1->k = VRELOCABLE; +} + +/* Emit comparison operator. */ +static void bcemit_comp(FuncState *fs, BinOpr opr, ExpDesc *e1, ExpDesc *e2) +{ + ExpDesc *eret = e1; + BCIns ins; + expr_toval(fs, e1); + if (opr == OPR_EQ || opr == OPR_NE) { + BCOp op = opr == OPR_EQ ? BC_ISEQV : BC_ISNEV; + BCReg ra; + if (expr_isk(e1)) { e1 = e2; e2 = eret; } /* Need constant in 2nd arg. */ + ra = expr_toanyreg(fs, e1); /* First arg must be in a reg. */ + expr_toval(fs, e2); + switch (e2->k) { + case VKNIL: case VKFALSE: case VKTRUE: + ins = BCINS_AD(op+(BC_ISEQP-BC_ISEQV), ra, const_pri(e2)); + break; + case VKSTR: + ins = BCINS_AD(op+(BC_ISEQS-BC_ISEQV), ra, const_str(fs, e2)); + break; + case VKNUM: + ins = BCINS_AD(op+(BC_ISEQN-BC_ISEQV), ra, const_num(fs, e2)); + break; + default: + ins = BCINS_AD(op, ra, expr_toanyreg(fs, e2)); + break; + } + } else { + uint32_t op = opr-OPR_LT+BC_ISLT; + BCReg ra, rd; + if ((op-BC_ISLT) & 1) { /* GT -> LT, GE -> LE */ + e1 = e2; e2 = eret; /* Swap operands. */ + op = ((op-BC_ISLT)^3)+BC_ISLT; + expr_toval(fs, e1); + ra = expr_toanyreg(fs, e1); + rd = expr_toanyreg(fs, e2); + } else { + rd = expr_toanyreg(fs, e2); + ra = expr_toanyreg(fs, e1); + } + ins = BCINS_AD(op, ra, rd); + } + /* Using expr_free might cause asserts if the order is wrong. */ + if (e1->k == VNONRELOC && e1->u.s.info >= fs->nactvar) fs->freereg--; + if (e2->k == VNONRELOC && e2->u.s.info >= fs->nactvar) fs->freereg--; + bcemit_INS(fs, ins); + eret->u.s.info = bcemit_jmp(fs); + eret->k = VJMP; +} + +/* Fixup left side of binary operator. */ +static void bcemit_binop_left(FuncState *fs, BinOpr op, ExpDesc *e) +{ + if (op == OPR_AND) { + bcemit_branch_t(fs, e); + } else if (op == OPR_OR) { + bcemit_branch_f(fs, e); + } else if (op == OPR_CONCAT) { + expr_tonextreg(fs, e); + } else if (op == OPR_EQ || op == OPR_NE) { + if (!expr_isk_nojump(e)) expr_toanyreg(fs, e); + } else { + if (!expr_isnumk_nojump(e)) expr_toanyreg(fs, e); + } +} + +/* Emit binary operator. */ +static void bcemit_binop(FuncState *fs, BinOpr op, ExpDesc *e1, ExpDesc *e2) +{ + if (op <= OPR_SHR) { + bcemit_arith(fs, op, e1, e2); + } else if (op == OPR_AND) { + lj_assertFS(e1->t == NO_JMP, "jump list not closed"); + expr_discharge(fs, e2); + jmp_append(fs, &e2->f, e1->f); + *e1 = *e2; + } else if (op == OPR_OR) { + lj_assertFS(e1->f == NO_JMP, "jump list not closed"); + expr_discharge(fs, e2); + jmp_append(fs, &e2->t, e1->t); + *e1 = *e2; + } else if (op == OPR_CONCAT) { + expr_toval(fs, e2); + if (e2->k == VRELOCABLE && bc_op(*bcptr(fs, e2)) == BC_CAT) { + lj_assertFS(e1->u.s.info == bc_b(*bcptr(fs, e2))-1, + "bad CAT stack layout"); + expr_free(fs, e1); + setbc_b(bcptr(fs, e2), e1->u.s.info); + e1->u.s.info = e2->u.s.info; + } else { + expr_tonextreg(fs, e2); + expr_free(fs, e2); + expr_free(fs, e1); + e1->u.s.info = bcemit_ABC(fs, BC_CAT, 0, e1->u.s.info, e2->u.s.info); + } + e1->k = VRELOCABLE; + } else { + lj_assertFS(op == OPR_NE || op == OPR_EQ || + op == OPR_LT || op == OPR_GE || op == OPR_LE || op == OPR_GT, + "bad binop %d", op); + bcemit_comp(fs, op, e1, e2); + } +} + +/* Emit unary operator. */ +static void bcemit_unop(FuncState *fs, BCOp op, ExpDesc *e) +{ + if (op == BC_NOT) { + /* Swap true and false lists. */ + { BCPos temp = e->f; e->f = e->t; e->t = temp; } + jmp_dropval(fs, e->f); + jmp_dropval(fs, e->t); + expr_discharge(fs, e); + if (e->k == VKNIL || e->k == VKFALSE) { + e->k = VKTRUE; + return; + } else if (expr_isk(e) || (LJ_HASFFI && e->k == VKCDATA)) { + e->k = VKFALSE; + return; + } else if (e->k == VJMP) { + invertcond(fs, e); + return; + } else if (e->k == VRELOCABLE) { + bcreg_reserve(fs, 1); + setbc_a(bcptr(fs, e), fs->freereg-1); + e->u.s.info = fs->freereg-1; + e->k = VNONRELOC; + } else { + lj_assertFS(e->k == VNONRELOC, "bad expr type %d", e->k); + } + } else { + lj_assertFS(op == BC_UNM || op == BC_LEN || op == BC_BNOT, "bad unop %d", op); + if (op == BC_UNM && !expr_hasjump(e)) { /* Constant-fold negations. */ +#if LJ_HASFFI + if (e->k == VKCDATA) { /* Fold in-place since cdata is not interned. */ + GCcdata *cd = cdataV(&e->u.nval); + uint64_t *p = (uint64_t *)cdataptr(cd); + if (cd->ctypeid == CTID_COMPLEX_DOUBLE) + p[1] ^= U64x(80000000,00000000); + else + *p = ~*p+1u; + return; + } else +#endif + if (expr_isnumk(e) && !expr_numiszero(e)) { /* Avoid folding to -0. */ + TValue *o = expr_numtv(e); + if (tvisint(o)) { + int32_t k = intV(o), negk = (int32_t)(~(uint32_t)k+1u); +#if !LJ_INTONLY + if (k == negk) + setnumV(o, -(lua_Number)k); + else +#endif + setintV(o, negk); + return; + } else { + o->u64 ^= U64x(80000000,00000000); + return; + } + } + } + if (op == BC_BNOT && !expr_hasjump(e)) { /* Constant-fold. */ +#if LJ_HASFFI + if (e->k == VKCDATA) { /* Fold in-place since cdata is not interned. */ + GCcdata *cd = cdataV(&e->u.nval); + int64_t *p = (int64_t *)cdataptr(cd); + if (cd->ctypeid == CTID_INT64 || cd->ctypeid == CTID_UINT64) { + *p = ~*p; + return; + } + } else +#endif + if (expr_isnumk(e)) { + TValue *o = expr_numtv(e); + int32_t k = numberVint(o); + if ((lua_Number)k != numberVnum(o)) + lj_err_msg(fs->L, LJ_ERR_NOINT); + setintV(o, ~k); + return; + } + } + expr_toanyreg(fs, e); + } + expr_free(fs, e); + e->u.s.info = bcemit_AD(fs, op, 0, e->u.s.info); + e->k = VRELOCABLE; +} + +/* -- Lexer support ------------------------------------------------------- */ + +/* Check and consume optional token. */ +static int lex_opt(LexState *ls, LexToken tok) +{ + if (ls->tok == tok) { + lj_lex_next(ls); + return 1; + } + return 0; +} + +/* Check and consume token. */ +static void lex_check(LexState *ls, LexToken tok) +{ + if (ls->tok != tok) + err_token(ls, tok); + lj_lex_next(ls); +} + +/* Check for matching token. */ +static void lex_match(LexState *ls, LexToken what, LexToken who, BCLine line) +{ + if (!lex_opt(ls, what)) { + if (line == ls->linenumber) { + err_token(ls, what); + } else { + const char *swhat = lj_lex_token2str(ls, what); + const char *swho = lj_lex_token2str(ls, who); + lj_lex_error(ls, ls->tok, LJ_ERR_XMATCH, swhat, swho, line); + } + } +} + +/* Check for string token. */ +static GCstr *lex_str(LexState *ls) +{ + GCstr *s; + if (ls->tok != TK_name && (LJ_52 || ls->tok != TK_goto)) + err_token(ls, TK_name); + s = strV(&ls->tokval); + lj_lex_next(ls); + return s; +} + +/* -- Variable handling --------------------------------------------------- */ + +#define var_get(ls, fs, i) ((ls)->vstack[(fs)->varmap[(i)]]) + +/* Define a new local variable. */ +static void var_new(LexState *ls, BCReg n, GCstr *name) +{ + FuncState *fs = ls->fs; + MSize vtop = ls->vtop; + checklimit(fs, fs->nactvar+n, LJ_MAX_LOCVAR, "local variables"); + if (LJ_UNLIKELY(vtop >= ls->sizevstack)) { + if (ls->sizevstack >= LJ_MAX_VSTACK) + lj_lex_error(ls, 0, LJ_ERR_XLIMC, LJ_MAX_VSTACK); + lj_mem_growvec(ls->L, ls->vstack, ls->sizevstack, LJ_MAX_VSTACK, VarInfo); + } + lj_assertFS((uintptr_t)name < VARNAME__MAX || + lj_tab_getstr(fs->kt, name) != NULL, + "unanchored variable name"); + /* NOBARRIER: name is anchored in fs->kt and ls->vstack is not a GCobj. */ + setgcref(ls->vstack[vtop].name, obj2gco(name)); + fs->varmap[fs->nactvar+n] = (uint16_t)vtop; + ls->vtop = vtop+1; +} + +#define var_new_lit(ls, n, v) \ + var_new(ls, (n), lj_parse_keepstr(ls, "" v, sizeof(v)-1)) + +#define var_new_fixed(ls, n, vn) \ + var_new(ls, (n), (GCstr *)(uintptr_t)(vn)) + +/* Add local variables. */ +static void var_add(LexState *ls, BCReg nvars) +{ + FuncState *fs = ls->fs; + BCReg nactvar = fs->nactvar; + while (nvars--) { + VarInfo *v = &var_get(ls, fs, nactvar); + v->startpc = fs->pc; + v->slot = nactvar++; + v->info = 0; + } + fs->nactvar = nactvar; +} + +/* Remove local variables. */ +static void var_remove(LexState *ls, BCReg tolevel) +{ + FuncState *fs = ls->fs; + while (fs->nactvar > tolevel) + var_get(ls, fs, --fs->nactvar).endpc = fs->pc; +} + +/* Lookup local variable name. */ +static BCReg var_lookup_local(FuncState *fs, GCstr *n) +{ + int i; + for (i = fs->nactvar-1; i >= 0; i--) { + if (n == strref(var_get(fs->ls, fs, i).name)) + return (BCReg)i; + } + return (BCReg)-1; /* Not found. */ +} + +/* Lookup or add upvalue index. */ +static MSize var_lookup_uv(FuncState *fs, MSize vidx, ExpDesc *e) +{ + MSize i, n = fs->nuv; + for (i = 0; i < n; i++) + if (fs->uvmap[i] == vidx) + return i; /* Already exists. */ + /* Otherwise create a new one. */ + checklimit(fs, fs->nuv, LJ_MAX_UPVAL, "upvalues"); + lj_assertFS(e->k == VLOCAL || e->k == VUPVAL, "bad expr type %d", e->k); + fs->uvmap[n] = (uint16_t)vidx; + fs->uvtmp[n] = (uint16_t)(e->k == VLOCAL ? vidx : LJ_MAX_VSTACK+e->u.s.info); + fs->nuv = n+1; + return n; +} + +/* Forward declaration. */ +static void fscope_uvmark(FuncState *fs, BCReg level); + +/* Recursively lookup variables in enclosing functions. */ +static MSize var_lookup_(FuncState *fs, GCstr *name, ExpDesc *e, int first) +{ + if (fs) { + BCReg reg = var_lookup_local(fs, name); + if ((int32_t)reg >= 0) { /* Local in this function? */ + expr_init(e, VLOCAL, reg); + if (!first) + fscope_uvmark(fs, reg); /* Scope now has an upvalue. */ + return (MSize)(e->u.s.aux = (uint32_t)fs->varmap[reg]); + } else { + MSize vidx = var_lookup_(fs->prev, name, e, 0); /* Var in outer func? */ + if ((int32_t)vidx >= 0) { /* Yes, make it an upvalue here. */ + e->u.s.info = (uint8_t)var_lookup_uv(fs, vidx, e); + e->k = VUPVAL; + return vidx; + } + } + } else { /* Not found in any function, must be a global. */ + expr_init(e, VGLOBAL, 0); + e->u.sval = name; + } + return (MSize)-1; /* Global. */ +} + +/* Lookup variable name. */ +#define var_lookup(ls, e) \ + var_lookup_((ls)->fs, lex_str(ls), (e), 1) + +/* -- Goto an label handling ---------------------------------------------- */ + +/* Add a new goto or label. */ +static MSize gola_new(LexState *ls, GCstr *name, uint8_t info, BCPos pc) +{ + FuncState *fs = ls->fs; + MSize vtop = ls->vtop; + if (LJ_UNLIKELY(vtop >= ls->sizevstack)) { + if (ls->sizevstack >= LJ_MAX_VSTACK) + lj_lex_error(ls, 0, LJ_ERR_XLIMC, LJ_MAX_VSTACK); + lj_mem_growvec(ls->L, ls->vstack, ls->sizevstack, LJ_MAX_VSTACK, VarInfo); + } + lj_assertFS(name == NAME_BREAK || name == NAME_CONTINUE || lj_tab_getstr(fs->kt, name) != NULL, + "unanchored label name"); + /* NOBARRIER: name is anchored in fs->kt and ls->vstack is not a GCobj. */ + setgcref(ls->vstack[vtop].name, obj2gco(name)); + ls->vstack[vtop].startpc = pc; + ls->vstack[vtop].slot = (uint8_t)fs->nactvar; + ls->vstack[vtop].info = info; + ls->vstack[vtop].breakn = 0; + ls->vtop = vtop+1; + return vtop; +} + +#define gola_isgoto(v) ((v)->info & VSTACK_GOTO) +#define gola_islabel(v) ((v)->info & VSTACK_LABEL) +#define gola_isgotolabel(v) ((v)->info & (VSTACK_GOTO|VSTACK_LABEL)) + +/* Patch goto to jump to label. */ +static void gola_patch(LexState *ls, VarInfo *vg, VarInfo *vl) +{ + FuncState *fs = ls->fs; + BCPos pc = vg->startpc; + setgcrefnull(vg->name); /* Invalidate pending goto. */ + setbc_a(&fs->bcbase[pc].ins, vl->slot); + jmp_patch(fs, pc, vl->startpc); +} + +/* Patch goto to close upvalues. */ +static void gola_close(LexState *ls, VarInfo *vg) +{ + FuncState *fs = ls->fs; + BCPos pc = vg->startpc; + BCIns *ip = &fs->bcbase[pc].ins; + lj_assertFS(gola_isgoto(vg), "expected goto"); + lj_assertFS(bc_op(*ip) == BC_JMP || bc_op(*ip) == BC_UCLO, + "bad bytecode op %d", bc_op(*ip)); + setbc_a(ip, vg->slot); + if (bc_op(*ip) == BC_JMP) { + BCPos next = jmp_next(fs, pc); + if (next != NO_JMP) jmp_patch(fs, next, pc); /* Jump to UCLO. */ + setbc_op(ip, BC_UCLO); /* Turn into UCLO. */ + setbc_j(ip, NO_JMP); + } +} + +/* Resolve pending forward gotos for label. */ +static void gola_resolve(LexState *ls, FuncScope *bl, MSize idx) +{ + VarInfo *vg = ls->vstack + bl->vstart; + VarInfo *vl = ls->vstack + idx; + for (; vg < vl; vg++) + if (gcrefeq(vg->name, vl->name) && gola_isgoto(vg)) { + if (vg->breakn && --vg->breakn && bl->prev) { + /* Still waiting for the correct scope... keep propagating. */ + bl->prev->flags |= FSCOPE_BREAK; + vg->slot = bl->nactvar; + if ((bl->flags & FSCOPE_UPVAL)) + gola_close(ls, vg); + continue; + } + if (vg->slot < vl->slot) { + GCstr *name = strref(var_get(ls, ls->fs, vg->slot).name); + lj_assertLS((uintptr_t)name >= VARNAME__MAX, "expected goto name"); + ls->linenumber = ls->fs->bcbase[vg->startpc].line; + lj_assertLS(strref(vg->name) != NAME_BREAK, "unexpected break"); + lj_lex_error(ls, 0, LJ_ERR_XGSCOPE, + strdata(strref(vg->name)), strdata(name)); + } + gola_patch(ls, vg, vl); + } +} + +/* Resolve label name to current PC. */ +static void gola_resolvehere(LexState *ls, FuncScope *bl, GCstr *name) +{ + MSize idx = gola_new(ls, name, VSTACK_LABEL, ls->fs->pc); + ls->vtop = idx; /* Drop label immediately. */ + gola_resolve(ls, bl, idx); +} + +/* Fixup remaining gotos and labels for scope. */ +static void gola_fixup(LexState *ls, FuncScope *bl) +{ + VarInfo *v = ls->vstack + bl->vstart; + VarInfo *ve = ls->vstack + ls->vtop; + for (; v < ve; v++) { + GCstr *name = strref(v->name); + if (name != NULL) { /* Only consider remaining valid gotos/labels. */ + if (gola_islabel(v)) { + VarInfo *vg; + setgcrefnull(v->name); /* Invalidate label that goes out of scope. */ + for (vg = v+1; vg < ve; vg++) /* Resolve pending backward gotos. */ + if (strref(vg->name) == name && gola_isgoto(vg)) { + if ((bl->flags&FSCOPE_UPVAL) && vg->slot > v->slot) + gola_close(ls, vg); + gola_patch(ls, vg, v); + } + } else if (gola_isgoto(v)) { + if (bl->prev) { /* Propagate goto or break to outer scope. */ + bl->prev->flags |= name == NAME_BREAK ? FSCOPE_BREAK : (name == NAME_CONTINUE ? FSCOPE_CONTINUE : FSCOPE_GOLA); + v->slot = bl->nactvar; + if ((bl->flags & FSCOPE_UPVAL)) + gola_close(ls, v); + } else { /* No outer scope: undefined goto label or no loop. */ + ls->linenumber = ls->fs->bcbase[v->startpc].line; + if (name == NAME_BREAK) + lj_lex_error(ls, 0, LJ_ERR_XBREAK); + else if (name == NAME_CONTINUE) + lj_lex_error(ls, 0, LJ_ERR_XCONT); + else + lj_lex_error(ls, 0, LJ_ERR_XLUNDEF, strdata(name)); + } + } + } + } +} + +/* Find existing label. */ +static VarInfo *gola_findlabel(LexState *ls, GCstr *name) +{ + VarInfo *v = ls->vstack + ls->fs->bl->vstart; + VarInfo *ve = ls->vstack + ls->vtop; + for (; v < ve; v++) + if (strref(v->name) == name && gola_islabel(v)) + return v; + return NULL; +} + +/* -- Scope handling ------------------------------------------------------ */ + +/* Begin a scope. */ +static void fscope_begin(FuncState *fs, FuncScope *bl, int flags) +{ + bl->nactvar = (uint8_t)fs->nactvar; + bl->flags = flags; + bl->vstart = fs->ls->vtop; + bl->prev = fs->bl; + fs->bl = bl; + lj_assertFS(fs->freereg == fs->nactvar, "bad regalloc"); +} + +/* End a scope. */ +static void fscope_end(FuncState *fs) +{ + FuncScope *bl = fs->bl; + LexState *ls = fs->ls; + fs->bl = bl->prev; + var_remove(ls, bl->nactvar); + fs->freereg = fs->nactvar; + lj_assertFS(bl->nactvar == fs->nactvar, "bad regalloc"); + if ((bl->flags & (FSCOPE_UPVAL|FSCOPE_NOCLOSE)) == FSCOPE_UPVAL) + bcemit_AJ(fs, BC_UCLO, bl->nactvar, 0); + if ((bl->flags & FSCOPE_BREAK)) { + if ((bl->flags & FSCOPE_LOOP)) { + gola_resolvehere(ls, bl, NAME_BREAK); + } else { /* Need the fixup step to propagate the breaks. */ + gola_fixup(ls, bl); + return; + } + } + if ((bl->flags & FSCOPE_GOLA) || (bl->flags & (FSCOPE_CONTINUE|FSCOPE_LOOP)) == FSCOPE_CONTINUE) { + gola_fixup(ls, bl); + } +} + +/* Mark scope as having an upvalue. */ +static void fscope_uvmark(FuncState *fs, BCReg level) +{ + FuncScope *bl; + for (bl = fs->bl; bl && bl->nactvar > level; bl = bl->prev) + ; + if (bl) + bl->flags |= FSCOPE_UPVAL; +} + +/* -- Function state management ------------------------------------------- */ + +/* Fixup bytecode for prototype. */ +static void fs_fixup_bc(FuncState *fs, GCproto *pt, BCIns *bc, MSize n) +{ + BCInsLine *base = fs->bcbase; + MSize i; + BCIns op; + pt->sizebc = n; + if (fs->ls->fr2 != LJ_FR2) op = BC_NOT; /* Mark non-native prototype. */ + else if ((fs->flags & PROTO_VARARG)) op = BC_FUNCV; + else op = BC_FUNCF; + bc[0] = BCINS_AD(op, fs->framesize, 0); + for (i = 1; i < n; i++) + bc[i] = base[i].ins; +} + +/* Fixup upvalues for child prototype, step #2. */ +static void fs_fixup_uv2(FuncState *fs, GCproto *pt) +{ + VarInfo *vstack = fs->ls->vstack; + uint16_t *uv = proto_uv(pt); + MSize i, n = pt->sizeuv; + for (i = 0; i < n; i++) { + VarIndex vidx = uv[i]; + if (vidx >= LJ_MAX_VSTACK) + uv[i] = vidx - LJ_MAX_VSTACK; + else if ((vstack[vidx].info & VSTACK_VAR_RW)) + uv[i] = vstack[vidx].slot | PROTO_UV_LOCAL; + else + uv[i] = vstack[vidx].slot | PROTO_UV_LOCAL | PROTO_UV_IMMUTABLE; + } +} + +/* Fixup constants for prototype. */ +static void fs_fixup_k(FuncState *fs, GCproto *pt, void *kptr) +{ + GCtab *kt; + TValue *array; + Node *node; + MSize i, hmask; + checklimitgt(fs, fs->nkn, BCMAX_D+1, "constants"); + checklimitgt(fs, fs->nkgc, BCMAX_D+1, "constants"); + setmref(pt->k, kptr); + pt->sizekn = fs->nkn; + pt->sizekgc = fs->nkgc; + kt = fs->kt; + array = tvref(kt->array); + for (i = 0; i < kt->asize; i++) + if (tvhaskslot(&array[i])) { + TValue *tv = &((TValue *)kptr)[tvkslot(&array[i])]; + if (LJ_DUALNUM) + setintV(tv, (int32_t)i); + else + setnumV(tv, (lua_Number)i); + } + node = noderef(kt->node); + hmask = kt->hmask; + for (i = 0; i <= hmask; i++) { + Node *n = &node[i]; + if (tvhaskslot(&n->val)) { + ptrdiff_t kidx = (ptrdiff_t)tvkslot(&n->val); + lj_assertFS(!tvisint(&n->key), "unexpected integer key"); + if (tvisnum(&n->key)) { + TValue *tv = &((TValue *)kptr)[kidx]; + if (LJ_DUALNUM) { + lua_Number nn = numV(&n->key); + int32_t k = lj_num2int(nn); + lj_assertFS(!tvismzero(&n->key), "unexpected -0 key"); + if ((lua_Number)k == nn) + setintV(tv, k); + else + *tv = n->key; + } else { + *tv = n->key; + } + } else { + GCobj *o = gcV(&n->key); + setgcref(((GCRef *)kptr)[~kidx], o); + lj_gc_objbarrier(fs->L, pt, o); + if (tvisproto(&n->key)) + fs_fixup_uv2(fs, gco2pt(o)); + } + } + } +} + +/* Fixup upvalues for prototype, step #1. */ +static void fs_fixup_uv1(FuncState *fs, GCproto *pt, uint16_t *uv) +{ + setmref(pt->uv, uv); + pt->sizeuv = fs->nuv; + memcpy(uv, fs->uvtmp, fs->nuv*sizeof(VarIndex)); +} + +#ifndef LUAJIT_DISABLE_DEBUGINFO +/* Prepare lineinfo for prototype. */ +static size_t fs_prep_line(FuncState *fs, BCLine numline) +{ + return (fs->pc-1) << (numline < 256 ? 0 : numline < 65536 ? 1 : 2); +} + +/* Fixup lineinfo for prototype. */ +static void fs_fixup_line(FuncState *fs, GCproto *pt, + void *lineinfo, BCLine numline) +{ + BCInsLine *base = fs->bcbase + 1; + BCLine first = fs->linedefined; + MSize i = 0, n = fs->pc-1; + pt->firstline = fs->linedefined; + pt->numline = numline; + setmref(pt->lineinfo, lineinfo); + if (LJ_LIKELY(numline < 256)) { + uint8_t *li = (uint8_t *)lineinfo; + do { + BCLine delta = base[i].line - first; + lj_assertFS(delta >= 0 && delta < 256, "bad line delta"); + li[i] = (uint8_t)delta; + } while (++i < n); + } else if (LJ_LIKELY(numline < 65536)) { + uint16_t *li = (uint16_t *)lineinfo; + do { + BCLine delta = base[i].line - first; + lj_assertFS(delta >= 0 && delta < 65536, "bad line delta"); + li[i] = (uint16_t)delta; + } while (++i < n); + } else { + uint32_t *li = (uint32_t *)lineinfo; + do { + BCLine delta = base[i].line - first; + lj_assertFS(delta >= 0, "bad line delta"); + li[i] = (uint32_t)delta; + } while (++i < n); + } +} + +/* Prepare variable info for prototype. */ +static size_t fs_prep_var(LexState *ls, FuncState *fs, size_t *ofsvar) +{ + VarInfo *vs =ls->vstack, *ve; + MSize i, n; + BCPos lastpc; + lj_buf_reset(&ls->sb); /* Copy to temp. string buffer. */ + /* Store upvalue names. */ + for (i = 0, n = fs->nuv; i < n; i++) { + GCstr *s = strref(vs[fs->uvmap[i]].name); + MSize len = s->len+1; + char *p = lj_buf_more(&ls->sb, len); + p = lj_buf_wmem(p, strdata(s), len); + ls->sb.w = p; + } + *ofsvar = sbuflen(&ls->sb); + lastpc = 0; + /* Store local variable names and compressed ranges. */ + for (ve = vs + ls->vtop, vs += fs->vbase; vs < ve; vs++) { + if (!gola_isgotolabel(vs)) { + GCstr *s = strref(vs->name); + BCPos startpc; + char *p; + if ((uintptr_t)s < VARNAME__MAX) { + p = lj_buf_more(&ls->sb, 1 + 2*5); + *p++ = (char)(uintptr_t)s; + } else { + MSize len = s->len+1; + p = lj_buf_more(&ls->sb, len + 2*5); + p = lj_buf_wmem(p, strdata(s), len); + } + startpc = vs->startpc; + p = lj_strfmt_wuleb128(p, startpc-lastpc); + p = lj_strfmt_wuleb128(p, vs->endpc-startpc); + ls->sb.w = p; + lastpc = startpc; + } + } + lj_buf_putb(&ls->sb, '\0'); /* Terminator for varinfo. */ + return sbuflen(&ls->sb); +} + +/* Fixup variable info for prototype. */ +static void fs_fixup_var(LexState *ls, GCproto *pt, uint8_t *p, size_t ofsvar) +{ + setmref(pt->uvinfo, p); + setmref(pt->varinfo, (char *)p + ofsvar); + memcpy(p, ls->sb.b, sbuflen(&ls->sb)); /* Copy from temp. buffer. */ +} +#else + +/* Initialize with empty debug info, if disabled. */ +#define fs_prep_line(fs, numline) (UNUSED(numline), 0) +#define fs_fixup_line(fs, pt, li, numline) \ + pt->firstline = pt->numline = 0, setmref((pt)->lineinfo, NULL) +#define fs_prep_var(ls, fs, ofsvar) (UNUSED(ofsvar), 0) +#define fs_fixup_var(ls, pt, p, ofsvar) \ + setmref((pt)->uvinfo, NULL), setmref((pt)->varinfo, NULL) + +#endif + +/* Check if bytecode op returns. */ +static int bcopisret(BCOp op) +{ + switch (op) { + case BC_CALLMT: case BC_CALLT: + case BC_RETM: case BC_RET: case BC_RET0: case BC_RET1: + return 1; + default: + return 0; + } +} + +/* Fixup return instruction for prototype. */ +static void fs_fixup_ret(FuncState *fs) +{ + BCPos lastpc = fs->pc; + if (lastpc <= fs->lasttarget || !bcopisret(bc_op(fs->bcbase[lastpc-1].ins))) { + if ((fs->bl->flags & FSCOPE_UPVAL)) + bcemit_AJ(fs, BC_UCLO, 0, 0); + bcemit_AD(fs, BC_RET0, 0, 1); /* Need final return. */ + } + fs->bl->flags |= FSCOPE_NOCLOSE; /* Handled above. */ + fscope_end(fs); + lj_assertFS(fs->bl == NULL, "bad scope nesting"); + /* May need to fixup returns encoded before first function was created. */ + if (fs->flags & PROTO_FIXUP_RETURN) { + BCPos pc; + for (pc = 1; pc < lastpc; pc++) { + BCIns ins = fs->bcbase[pc].ins; + BCPos offset; + switch (bc_op(ins)) { + case BC_CALLMT: case BC_CALLT: + case BC_RETM: case BC_RET: case BC_RET0: case BC_RET1: + offset = bcemit_INS(fs, ins); /* Copy original instruction. */ + fs->bcbase[offset].line = fs->bcbase[pc].line; + offset = offset-(pc+1)+BCBIAS_J; + if (offset > BCMAX_D) + err_syntax(fs->ls, LJ_ERR_XFIXUP); + /* Replace with UCLO plus branch. */ + fs->bcbase[pc].ins = BCINS_AD(BC_UCLO, 0, offset); + break; + case BC_FNEW: + return; /* We're done. */ + default: + break; + } + } + } +} + +/* Finish a FuncState and return the new prototype. */ +static GCproto *fs_finish(LexState *ls, BCLine line) +{ + lua_State *L = ls->L; + FuncState *fs = ls->fs; + BCLine numline = line - fs->linedefined; + size_t sizept, ofsk, ofsuv, ofsli, ofsdbg, ofsvar; + GCproto *pt; + + /* Apply final fixups. */ + fs_fixup_ret(fs); + + /* Calculate total size of prototype including all colocated arrays. */ + sizept = sizeof(GCproto) + fs->pc*sizeof(BCIns) + fs->nkgc*sizeof(GCRef); + sizept = (sizept + sizeof(TValue)-1) & ~(sizeof(TValue)-1); + ofsk = sizept; sizept += fs->nkn*sizeof(TValue); + ofsuv = sizept; sizept += ((fs->nuv+1)&~1)*2; + ofsli = sizept; sizept += fs_prep_line(fs, numline); + ofsdbg = sizept; sizept += fs_prep_var(ls, fs, &ofsvar); + + /* Allocate prototype and initialize its fields. */ + pt = (GCproto *)lj_mem_newgco(L, (MSize)sizept); + pt->gct = ~LJ_TPROTO; + pt->sizept = (MSize)sizept; + pt->trace = 0; + pt->flags = (uint8_t)(fs->flags & ~(PROTO_HAS_RETURN|PROTO_FIXUP_RETURN)); + pt->numparams = fs->numparams; + pt->framesize = fs->framesize; + setgcref(pt->chunkname, obj2gco(ls->chunkname)); + + /* Close potentially uninitialized gap between bc and kgc. */ + *(uint32_t *)((char *)pt + ofsk - sizeof(GCRef)*(fs->nkgc+1)) = 0; + fs_fixup_bc(fs, pt, (BCIns *)((char *)pt + sizeof(GCproto)), fs->pc); + fs_fixup_k(fs, pt, (void *)((char *)pt + ofsk)); + fs_fixup_uv1(fs, pt, (uint16_t *)((char *)pt + ofsuv)); + fs_fixup_line(fs, pt, (void *)((char *)pt + ofsli), numline); + fs_fixup_var(ls, pt, (uint8_t *)((char *)pt + ofsdbg), ofsvar); + + lj_vmevent_send(L, BC, + setprotoV(L, L->top++, pt); + ); + + L->top--; /* Pop table of constants. */ + ls->vtop = fs->vbase; /* Reset variable stack. */ + ls->fs = fs->prev; + lj_assertL(ls->fs != NULL || ls->tok == TK_eof, "bad parser state"); + return pt; +} + +/* Initialize a new FuncState. */ +static void fs_init(LexState *ls, FuncState *fs) +{ + lua_State *L = ls->L; + fs->prev = ls->fs; ls->fs = fs; /* Append to list. */ + fs->ls = ls; + fs->vbase = ls->vtop; + fs->L = L; + fs->lhs = NULL; + fs->nlhs = 0; + fs->nrhs = 0; + fs->pc = 0; + fs->lasttarget = 0; + fs->jpc = NO_JMP; + fs->freereg = 0; + fs->nkgc = 0; + fs->nkn = 0; + fs->nactvar = 0; + fs->nuv = 0; + fs->bl = NULL; + fs->flags = 0; + fs->framesize = 1; /* Minimum frame size. */ + fs->kt = lj_tab_new(L, 0, 0); + /* Anchor table of constants in stack to avoid being collected. */ + settabV(L, L->top, fs->kt); + incr_top(L); +} + +/* -- Expressions --------------------------------------------------------- */ + +/* Forward declaration. */ +static void expr(LexState *ls, ExpDesc *v); + +/* Return string expression. */ +static void expr_str(LexState *ls, ExpDesc *e) +{ + expr_init(e, VKSTR, 0); + e->u.sval = lex_str(ls); +} + +/* Return index expression. */ +static void expr_index(FuncState *fs, ExpDesc *t, ExpDesc *e) +{ + /* Already called: expr_toval(fs, e). */ + t->k = VINDEXED; + if (expr_isnumk(e)) { +#if LJ_DUALNUM + if (tvisint(expr_numtv(e))) { + int32_t k = intV(expr_numtv(e)); + if (checku8(k)) { + t->u.s.aux = BCMAX_C+1+(uint32_t)k; /* 256..511: const byte key */ + return; + } + } +#else + lua_Number n = expr_numberV(e); + int32_t k = lj_num2int(n); + if (checku8(k) && n == (lua_Number)k) { + t->u.s.aux = BCMAX_C+1+(uint32_t)k; /* 256..511: const byte key */ + return; + } +#endif + } else if (expr_isstrk(e)) { + BCReg idx = const_str(fs, e); + if (idx <= BCMAX_C) { + t->u.s.aux = ~idx; /* -256..-1: const string key */ + return; + } + } + t->u.s.aux = expr_toanyreg(fs, e); /* 0..255: register */ +} + +/* Parse index expression with named field. */ +static void expr_field(LexState *ls, ExpDesc *v) +{ + FuncState *fs = ls->fs; + ExpDesc key; + expr_toanyreg(fs, v); + lj_lex_next(ls); /* Skip dot or colon. */ + expr_str(ls, &key); + expr_index(fs, v, &key); +} + +/* Parse index expression with brackets. */ +static void expr_bracket(LexState *ls, ExpDesc *v) +{ + lj_lex_next(ls); /* Skip '['. */ + expr(ls, v); + expr_toval(ls->fs, v); + lex_check(ls, ']'); +} + +/* Get value of constant expression. */ +static void expr_kvalue(FuncState *fs, TValue *v, ExpDesc *e) +{ + UNUSED(fs); + if (e->k <= VKTRUE) { + setpriV(v, ~(uint32_t)e->k); + } else if (e->k == VKSTR) { + setgcVraw(v, obj2gco(e->u.sval), LJ_TSTR); + } else { + lj_assertFS(tvisnumber(expr_numtv(e)), "bad number constant"); + *v = *expr_numtv(e); + } +} + +/* Parse table constructor expression. */ +static void expr_table(LexState *ls, ExpDesc *e) +{ + FuncState *fs = ls->fs; + BCLine line = ls->linenumber; + GCtab *t = NULL; + int vcall = 0, needarr = 0, fixt = 0; + uint32_t narr = 1; /* First array index. */ + uint32_t nhash = 0; /* Number of hash entries. */ + BCReg freg = fs->freereg; + BCPos pc = bcemit_AD(fs, BC_TNEW, freg, 0); + expr_init(e, VNONRELOC, freg); + bcreg_reserve(fs, 1); + freg++; + lex_check(ls, '{'); + while (ls->tok != '}') { + ExpDesc key, val; + vcall = 0; + if (ls->tok == '[') { + expr_bracket(ls, &key); /* Already calls expr_toval. */ + if (!expr_isk(&key)) expr_index(fs, e, &key); + if (expr_isnumk(&key) && expr_numiszero(&key)) needarr = 1; else nhash++; + lex_check(ls, '='); + } else if ((ls->tok == TK_name || (!LJ_52 && ls->tok == TK_goto)) && + lj_lex_lookahead(ls) == '=') { + expr_str(ls, &key); + lex_check(ls, '='); + nhash++; + } else { + expr_init(&key, VKNUM, 0); + setintV(&key.u.nval, (int)narr); + narr++; + needarr = vcall = 1; + } + expr(ls, &val); + if (expr_isk(&key) && key.k != VKNIL && + (key.k == VKSTR || expr_isk_nojump(&val))) { + TValue k, *v; + if (!t) { /* Create template table on demand. */ + BCReg kidx; + t = lj_tab_new(fs->L, needarr ? narr : 0, hsize2hbits(nhash)); + kidx = const_gc(fs, obj2gco(t), LJ_TTAB); + fs->bcbase[pc].ins = BCINS_AD(BC_TDUP, freg-1, kidx); + } + vcall = 0; + expr_kvalue(fs, &k, &key); + v = lj_tab_set(fs->L, t, &k); + lj_gc_anybarriert(fs->L, t); + if (expr_isk_nojump(&val)) { /* Add const key/value to template table. */ + expr_kvalue(fs, v, &val); + } else { /* Otherwise create dummy string key (avoids lj_tab_newkey). */ + settabV(fs->L, v, t); /* Preserve key with table itself as value. */ + fixt = 1; /* Fix this later, after all resizes. */ + goto nonconst; + } + } else { + nonconst: + if (val.k != VCALL) { expr_toanyreg(fs, &val); vcall = 0; } + if (expr_isk(&key)) expr_index(fs, e, &key); + bcemit_store(fs, e, &val); + } + fs->freereg = freg; + if (!lex_opt(ls, ',') && !lex_opt(ls, ';')) break; + } + lex_match(ls, '}', '{', line); + if (vcall) { + BCInsLine *ilp = &fs->bcbase[fs->pc-1]; + ExpDesc en; + lj_assertFS(bc_a(ilp->ins) == freg && + bc_op(ilp->ins) == (narr > 256 ? BC_TSETV : BC_TSETB), + "bad CALL code generation"); + expr_init(&en, VKNUM, 0); + en.u.nval.u32.lo = narr-1; + en.u.nval.u32.hi = 0x43300000; /* Biased integer to avoid denormals. */ + if (narr > 256) { fs->pc--; ilp--; } + ilp->ins = BCINS_AD(BC_TSETM, freg, const_num(fs, &en)); + setbc_b(&ilp[-1].ins, 0); + } + if (pc == fs->pc-1) { /* Make expr relocable if possible. */ + e->u.s.info = pc; + fs->freereg--; + e->k = VRELOCABLE; + } else { + e->k = VNONRELOC; /* May have been changed by expr_index. */ + } + if (!t) { /* Construct TNEW RD: hhhhhaaaaaaaaaaa. */ + BCIns *ip = &fs->bcbase[pc].ins; + if (!needarr) narr = 0; + else if (narr < 3) narr = 3; + else if (narr > 0x7ff) narr = 0x7ff; + setbc_d(ip, narr|(hsize2hbits(nhash)<<11)); + } else { + if (needarr && t->asize < narr) + lj_tab_reasize(fs->L, t, narr-1); + if (fixt) { /* Fix value for dummy keys in template table. */ + Node *node = noderef(t->node); + uint32_t i, hmask = t->hmask; + for (i = 0; i <= hmask; i++) { + Node *n = &node[i]; + if (tvistab(&n->val)) { + lj_assertFS(tabV(&n->val) == t, "bad dummy key in template table"); + setnilV(&n->val); /* Turn value into nil. */ + } + } + } + lj_gc_check(fs->L); + } +} + +/* Parse function parameters. */ +static BCReg parse_params(LexState *ls, int needself) +{ + FuncState *fs = ls->fs; + BCReg nparams = 0; + lex_check(ls, '('); + if (needself) + var_new_lit(ls, nparams++, "self"); + if (ls->tok != ')') { + do { + if (ls->tok == TK_name || (!LJ_52 && ls->tok == TK_goto)) { + var_new(ls, nparams++, lex_str(ls)); + } else if (ls->tok == TK_dots) { + lj_lex_next(ls); + fs->flags |= PROTO_VARARG; + break; + } else { + err_syntax(ls, LJ_ERR_XPARAM); + } + } while (lex_opt(ls, ',')); + } + var_add(ls, nparams); + lj_assertFS(fs->nactvar == nparams, "bad regalloc"); + bcreg_reserve(fs, nparams); + lex_check(ls, ')'); + return nparams; +} + +/* Forward declaration. */ +static void parse_chunk(LexState *ls); + +#if BLUAJIT_DO_END_ANONYMOUS_FUNCTIONS + /** + * This is a rewrite of whatever is going on in parse_params, except it is always empty. + */ + static void parse_params_empty(LexState *ls) + { + FuncState *fs = ls->fs; + var_add(ls, 0); + lj_assertFS(fs->nactvar == 0, "bad regalloc"); + bcreg_reserve(fs, 0); + } + + /* This is a rewrite of parse_body() except for "chunk END" */ + static void body_no_parameters(LexState *ls, ExpDesc *e, BCLine line) + { + FuncState fs; + FuncState *pfs = ls->fs; + FuncScope bl; + GCproto *pt; + ptrdiff_t oldbase = pfs->bcbase - ls->bcstack; + + fs_init(ls, &fs); + fscope_begin(&fs, &bl, 0); + parse_params_empty(ls); + + fs.linedefined = line; + fs.numparams = 0; + fs.bcbase = pfs->bcbase + pfs->pc; + fs.bclim = pfs->bclim - pfs->pc; + bcemit_AD(&fs, BC_FUNCF, 0, 0); + + parse_chunk(ls); + + if (ls->tok != TK_end) + lex_match(ls, TK_end, TK_do /* TK_function */, line); + + pt = fs_finish(ls, (ls->lastline = ls->linenumber)); + pfs->bcbase = ls->bcstack + oldbase; + pfs->bclim = (BCPos)(ls->sizebcstack - oldbase); + expr_init(e, VRELOCABLE, + bcemit_AD(pfs, BC_FNEW, 0, const_gc(pfs, obj2gco(pt), LJ_TPROTO))); + #if LJ_HASFFI + pfs->flags |= (fs.flags & PROTO_FFI); + #endif + if (!(pfs->flags & PROTO_CHILD)) { + if (pfs->flags & PROTO_HAS_RETURN) + pfs->flags |= PROTO_FIXUP_RETURN; + pfs->flags |= PROTO_CHILD; + } + lj_lex_next(ls); + } +#endif + +/* Parse body of a function. */ +static void parse_body(LexState *ls, ExpDesc *e, int needself, BCLine line) +{ + FuncState fs, *pfs = ls->fs; + FuncScope bl; + GCproto *pt; + ptrdiff_t oldbase = pfs->bcbase - ls->bcstack; + fs_init(ls, &fs); + fscope_begin(&fs, &bl, 0); + fs.linedefined = line; + fs.numparams = (uint8_t)parse_params(ls, needself); + fs.bcbase = pfs->bcbase + pfs->pc; + fs.bclim = pfs->bclim - pfs->pc; + bcemit_AD(&fs, BC_FUNCF, 0, 0); /* Placeholder. */ + parse_chunk(ls); + if (ls->tok != TK_end) lex_match(ls, TK_end, TK_function, line); + pt = fs_finish(ls, (ls->lastline = ls->linenumber)); + pfs->bcbase = ls->bcstack + oldbase; /* May have been reallocated. */ + pfs->bclim = (BCPos)(ls->sizebcstack - oldbase); + /* Store new prototype in the constant array of the parent. */ + expr_init(e, VRELOCABLE, + bcemit_AD(pfs, BC_FNEW, 0, const_gc(pfs, obj2gco(pt), LJ_TPROTO))); +#if LJ_HASFFI + pfs->flags |= (fs.flags & PROTO_FFI); +#endif + if (!(pfs->flags & PROTO_CHILD)) { + if (pfs->flags & PROTO_HAS_RETURN) + pfs->flags |= PROTO_FIXUP_RETURN; + pfs->flags |= PROTO_CHILD; + } + lj_lex_next(ls); +} + +/* Parse expression list. Last expression is left open. */ +static BCReg expr_list(LexState *ls, ExpDesc *v) +{ + BCReg n = 1; + expr(ls, v); + while (lex_opt(ls, ',')) { + expr_tonextreg(ls->fs, v); + expr(ls, v); + n++; + } + return n; +} + +/* Parse function argument list. */ +static void parse_args(LexState *ls, ExpDesc *e) +{ + FuncState *fs = ls->fs; + ExpDesc args; + BCIns ins; + BCReg base; + BCLine line = ls->linenumber; + if (ls->tok == '(') { +#if !LJ_52 + if (line != ls->lastline) + err_syntax(ls, LJ_ERR_XAMBIG); +#endif + lj_lex_next(ls); + if (ls->tok == ')') { /* f(). */ + args.k = VVOID; + } else { + expr_list(ls, &args); + if (args.k == VCALL) /* f(a, b, g()) or f(a, b, ...). */ + setbc_b(bcptr(fs, &args), 0); /* Pass on multiple results. */ + } + lex_match(ls, ')', '(', line); + } else if (ls->tok == '{') { + expr_table(ls, &args); + } else if (ls->tok == TK_string) { + expr_init(&args, VKSTR, 0); + args.u.sval = strV(&ls->tokval); + lj_lex_next(ls); + } else { + err_syntax(ls, LJ_ERR_XFUNARG); + return; /* Silence compiler. */ + } + lj_assertFS(e->k == VNONRELOC, "bad expr type %d", e->k); + base = e->u.s.info; /* Base register for call. */ + if (args.k == VCALL) { + ins = BCINS_ABC(BC_CALLM, base, 2, args.u.s.aux - base - 1 - ls->fr2); + } else { + if (args.k != VVOID) + expr_tonextreg(fs, &args); + ins = BCINS_ABC(BC_CALL, base, 2, fs->freereg - base - ls->fr2); + } + expr_init(e, VCALL, bcemit_INS(fs, ins)); + e->u.s.aux = base; + fs->bcbase[fs->pc - 1].line = line; + fs->freereg = base+1; /* Leave one result by default. */ +} + +/* Parse primary expression. */ +static void expr_primary(LexState *ls, ExpDesc *v) +{ + FuncState *fs = ls->fs; + /* Parse prefix expression. */ + if (ls->tok == '(') { + BCLine line = ls->linenumber; + lj_lex_next(ls); + expr(ls, v); + lex_match(ls, ')', '(', line); + expr_discharge(ls->fs, v); + } else if (ls->tok == TK_name || (!LJ_52 && ls->tok == TK_goto)) { + var_lookup(ls, v); + } else if (ls->tok == '$') { + uint64_t i = ls->tokval.u64; + if (i == 0) i = ls->fs->nrhs; + checkcond(ls, i > 0 && i <= ls->fs->nlhs, LJ_ERR_XPSEUDO); + LHSVarList *lhs = ls->fs->lhs; + i = ls->fs->nlhs - i; + while (i--) lhs = lhs->prev; + *v = lhs->v; + /* If this is a VINDEXED, we need to stash the result in a temporary without + * freereg'ing the index locals. Really, it would be better to keep the value + * in case we repeat the $, but there's no way to do that without shifting all + * the temps up one slot to create room for the value. It would be semantically + * correct to also do that for globals, in case I ever get around to importing + * the shift mechanism from comprehensions. + */ + if (v->k == VINDEXED) { + BCIns ins; + BCReg extra = ls->fs->freereg; + BCReg rc = v->u.s.aux; + if ((int32_t)rc < 0) { + ins = BCINS_ABC(BC_TGETS, extra, v->u.s.info, ~rc); + } else if (rc > BCMAX_C) { + ins = BCINS_ABC(BC_TGETB, extra, v->u.s.info, rc-(BCMAX_C+1)); + } else { + ins = BCINS_ABC(BC_TGETV, extra, v->u.s.info, rc); + } + bcemit_INS(ls->fs, ins); + v->k = VNONRELOC; + v->u.s.info = extra; + bcreg_reserve(ls->fs, 1); + } + lj_lex_next(ls); + } else { + err_syntax(ls, LJ_ERR_XSYMBOL); + } + for (;;) { /* Parse multiple expression suffixes. */ + if (ls->tok == '.') { + expr_field(ls, v); + } else if (ls->tok == '[') { + ExpDesc key; + expr_toanyreg(fs, v); + expr_bracket(ls, &key); + expr_index(fs, v, &key); + } else if (ls->tok == ':') { + ExpDesc key; + lj_lex_next(ls); + expr_str(ls, &key); + bcemit_method(fs, v, &key); + parse_args(ls, v); + } else if (ls->tok == '(' || ls->tok == TK_string || ls->tok == '{') { + expr_tonextreg(fs, v); + if (ls->fr2) bcreg_reserve(fs, 1); + parse_args(ls, v); + } else { + break; + } + } +} + +/* Parse simple expression. */ +static void expr_simple(LexState *ls, ExpDesc *v) +{ + switch (ls->tok) { + case TK_number: + expr_init(v, (LJ_HASFFI && tviscdata(&ls->tokval)) ? VKCDATA : VKNUM, 0); + copyTV(ls->L, &v->u.nval, &ls->tokval); + break; + case TK_string: + expr_init(v, VKSTR, 0); + v->u.sval = strV(&ls->tokval); + break; + case TK_nil: + expr_init(v, VKNIL, 0); + break; + case TK_true: + expr_init(v, VKTRUE, 0); + break; + case TK_false: + expr_init(v, VKFALSE, 0); + break; + case TK_dots: { /* Vararg. */ + FuncState *fs = ls->fs; + BCReg base; + checkcond(ls, fs->flags & PROTO_VARARG, LJ_ERR_XDOTS); + bcreg_reserve(fs, 1); + base = fs->freereg-1; + expr_init(v, VCALL, bcemit_ABC(fs, BC_VARG, base, 2, fs->numparams)); + v->u.s.aux = base; + break; + } + case '{': /* Table constructor. */ + expr_table(ls, v); + return; + #if BLUAJIT_DO_END_ANONYMOUS_FUNCTIONS + case TK_do: + lj_lex_next(ls); + body_no_parameters(ls, v, ls->linenumber); + return; + #endif + case TK_function: + lj_lex_next(ls); + parse_body(ls, v, 0, ls->linenumber); + return; + default: + expr_primary(ls, v); + return; + } + lj_lex_next(ls); +} + +/* Manage syntactic levels to avoid blowing up the stack. */ +static void synlevel_begin(LexState *ls) +{ + if (++ls->level >= LJ_MAX_XLEVEL) + lj_lex_error(ls, 0, LJ_ERR_XLEVELS); +} + +#define synlevel_end(ls) ((ls)->level--) + +/* Convert token to binary operator. */ +static BinOpr token2binop(LexToken tok) +{ + switch (tok) { + case '+': return OPR_ADD; + case '-': return OPR_SUB; + case '*': return OPR_MUL; + case '/': return OPR_DIV; + case '%': return OPR_MOD; + case '^': return OPR_POW; + case '&': return OPR_BAND; + case '|': return OPR_BOR; + case TK_bxor: return OPR_BXOR; + case TK_shl: return OPR_SHL; + case TK_shr: return OPR_SHR; + case TK_concat: return OPR_CONCAT; + case TK_ne: return OPR_NE; + case TK_eq: return OPR_EQ; + case '<': return OPR_LT; + case TK_le: return OPR_LE; + case '>': return OPR_GT; + case TK_ge: return OPR_GE; + case TK_and: return OPR_AND; + case TK_or: return OPR_OR; + default: return OPR_NOBINOPR; + } +} + +/* Priorities for each binary operator. ORDER OPR. */ +static const struct { + uint8_t left; /* Left priority. */ + uint8_t right; /* Right priority. */ +} priority[] = { + {6, 6}, {6, 6}, /* ADD SUB */ + {7, 7}, {7, 7}, {7, 7}, /* MUL DIV MOD */ + {10, 9}, /* POW (right associative) */ + {6,6}, {6,6}, {6,6}, /* BAND BOR BXOR */ + {7,7}, {7,7}, /* SHL SHR */ + {5, 4}, /* CONCAT (right associative) */ + {3,3}, {3,3}, /* EQ NE */ + {3,3}, {3,3}, {3,3}, {3,3}, /* LT GE GT LE */ + {2,2}, {1,1} /* AND OR */ +}; + +#define UNARY_PRIORITY 8 /* Priority for unary operators. */ + +/* Forward declaration. */ +static BinOpr expr_binop(LexState *ls, ExpDesc *v, uint32_t limit); + +/* Parse unary expression. */ +static void expr_unop(LexState *ls, ExpDesc *v) +{ + BCOp op; + if (ls->tok == TK_not) { + op = BC_NOT; + } else if (ls->tok == '-') { + op = BC_UNM; + } else if (ls->tok == '#') { + op = BC_LEN; + } else if (ls->tok == '~') { + op = BC_BNOT; + } else { + expr_simple(ls, v); + return; + } + lj_lex_next(ls); + expr_binop(ls, v, UNARY_PRIORITY); + bcemit_unop(ls->fs, op, v); +} + +/* Parse binary expressions with priority higher than the limit. */ +static BinOpr expr_binop(LexState *ls, ExpDesc *v, uint32_t limit) +{ + BinOpr op; + synlevel_begin(ls); + expr_unop(ls, v); + op = token2binop(ls->tok); + while (op != OPR_NOBINOPR && priority[op].left > limit) { + ExpDesc v2; + BinOpr nextop; + lj_lex_next(ls); + bcemit_binop_left(ls->fs, op, v); + /* Parse binary expression with higher priority. */ + nextop = expr_binop(ls, &v2, priority[op].right); + bcemit_binop(ls->fs, op, v, &v2); + op = nextop; + } + synlevel_end(ls); + return op; /* Return unconsumed binary operator (if any). */ +} + +/* Parse expression. */ +static void expr(LexState *ls, ExpDesc *v) +{ + expr_binop(ls, v, 0); /* Priority 0: parse whole expression. */ +} + +/* Assign expression to the next register. */ +static void expr_next(LexState *ls) +{ + ExpDesc e; + expr(ls, &e); + expr_tonextreg(ls->fs, &e); +} + +/* Parse conditional expression. */ +static BCPos expr_cond(LexState *ls) +{ + ExpDesc v; + expr(ls, &v); + if (v.k == VKNIL) v.k = VKFALSE; + bcemit_branch_t(ls->fs, &v); + return v.f; +} + +/* -- Assignments --------------------------------------------------------- */ + +/* Eliminate write-after-read hazards for local variable assignment. */ +static void assign_hazard(LexState *ls, LHSVarList *lh, const ExpDesc *v) +{ + FuncState *fs = ls->fs; + BCReg reg = v->u.s.info; /* Check against this variable. */ + BCReg tmp = fs->freereg; /* Rename to this temp. register (if needed). */ + int hazard = 0; + for (; lh; lh = lh->prev) { + if (lh->v.k == VINDEXED) { + if (lh->v.u.s.info == reg) { /* t[i], t = 1, 2 */ + hazard = 1; + lh->v.u.s.info = tmp; + } + if (lh->v.u.s.aux == reg) { /* t[i], i = 1, 2 */ + hazard = 1; + lh->v.u.s.aux = tmp; + } + } + } + if (hazard) { + bcemit_AD(fs, BC_MOV, tmp, reg); /* Rename conflicting variable. */ + bcreg_reserve(fs, 1); + } +} + +/* Adjust LHS/RHS of an assignment. */ +static void assign_adjust(LexState *ls, BCReg nvars, BCReg nexps, ExpDesc *e) +{ + FuncState *fs = ls->fs; + int32_t extra = (int32_t)nvars - (int32_t)nexps; + if (e->k == VCALL) { + extra++; /* Compensate for the VCALL itself. */ + if (extra < 0) extra = 0; + setbc_b(bcptr(fs, e), extra+1); /* Fixup call results. */ + if (extra > 1) bcreg_reserve(fs, (BCReg)extra-1); + } else { + if (e->k != VVOID) + expr_tonextreg(fs, e); /* Close last expression. */ + if (extra > 0) { /* Leftover LHS are set to nil. */ + BCReg reg = fs->freereg; + bcreg_reserve(fs, (BCReg)extra); + bcemit_nil(fs, reg, (BCReg)extra); + } + } + if (nexps > nvars) + ls->fs->freereg -= nexps - nvars; /* Drop leftover regs. */ +} + +static void pushlhs (FuncState *fs, LHSVarList *v) { + fs->lhs = v; + fs->nlhs++; + fs->nrhs++; +} + +static void poplhs (FuncState *fs) { + lj_assertFS(fs->lhs != NULL, "bad lhs"); + fs->lhs = fs->lhs->prev; + fs->nlhs--; +} + +/* Recursively parse assignment statement. */ +static void parse_assignment(LexState *ls) +{ + ExpDesc e; + LHSVarList *lh = ls->fs->lhs; + checkcond(ls, VLOCAL <= lh->v.k && lh->v.k <= VINDEXED, LJ_ERR_XSYNTAX); + if (lex_opt(ls, ',')) { /* Collect LHS list and recurse upwards. */ + LHSVarList vl; + vl.prev = lh; + expr_primary(ls, &vl.v); + if (vl.v.k == VLOCAL) + assign_hazard(ls, lh, &vl.v); + checklimit(ls->fs, ls->level + ls->fs->nlhs, LJ_MAX_XLEVEL, "variable names"); + pushlhs(ls->fs, &vl); + parse_assignment(ls); + poplhs(ls->fs); + } else { /* Parse RHS. */ + lex_check(ls, '='); + /* G: nrhs MUST be incremented thorought expression parsing */ + ls->fs->nrhs = 1; + expr(ls, &e); + while (lex_opt(ls, ',')) { + expr_tonextreg(ls->fs, &e); + ls->fs->nrhs++; + expr(ls, &e); + } + if (ls->fs->nrhs == ls->fs->nlhs) { + if (e.k == VCALL) { + if (bc_op(*bcptr(ls->fs, &e)) == BC_VARG) { /* Vararg assignment. */ + ls->fs->freereg--; + e.k = VRELOCABLE; + } else { /* Multiple call results. */ + e.u.s.info = e.u.s.aux; /* Base of call is not relocatable. */ + e.k = VNONRELOC; + } + } + bcemit_store(ls->fs, &lh->v, &e); + return; + } + assign_adjust(ls, ls->fs->nlhs, ls->fs->nrhs, &e); + } + /* Assign RHS to LHS and recurse downwards. */ + expr_init(&e, VNONRELOC, ls->fs->freereg-1); + bcemit_store(ls->fs, &lh->v, &e); +} + +/* Parse call statement or assignment. */ +static void parse_call_assign(LexState *ls) +{ + FuncState *fs = ls->fs; + LHSVarList vl; + expr_primary(ls, &vl.v); + if (vl.v.k == VCALL) { /* Function call statement. */ + setbc_b(bcptr(fs, &vl.v), 1); /* No results. */ + } else { /* Start of an assignment. */ + vl.prev = NULL; + lj_assertFS(ls->fs->lhs == NULL && ls->fs->nlhs == 0 && ls->fs->nrhs == 0, "bad lhs"); + pushlhs(ls->fs, &vl); + parse_assignment(ls); + poplhs(ls->fs); + ls->fs->nrhs = 0; + lj_assertFS(ls->fs->lhs == NULL && ls->fs->nlhs == 0, "bad lhs"); + } +} + +/* Parse 'local' statement. */ +static void parse_local(LexState *ls) +{ + if (lex_opt(ls, TK_function)) { /* Local function declaration. */ + ExpDesc v, b; + FuncState *fs = ls->fs; + var_new(ls, 0, lex_str(ls)); + expr_init(&v, VLOCAL, fs->freereg); + v.u.s.aux = fs->varmap[fs->freereg]; + bcreg_reserve(fs, 1); + var_add(ls, 1); + parse_body(ls, &b, 0, ls->linenumber); + /* bcemit_store(fs, &v, &b) without setting VSTACK_VAR_RW. */ + expr_free(fs, &b); + expr_toreg(fs, &b, v.u.s.info); + /* The upvalue is in scope, but the local is only valid after the store. */ + var_get(ls, fs, fs->nactvar - 1).startpc = fs->pc; + } else { /* Local variable declaration. */ + ExpDesc e; + BCReg nexps, nvars = 0; + do { /* Collect LHS. */ + var_new(ls, nvars++, lex_str(ls)); + } while (lex_opt(ls, ',')); + if (lex_opt(ls, '=')) { /* Optional RHS. */ + nexps = expr_list(ls, &e); + } else { /* Or implicitly set to nil. */ + e.k = VVOID; + nexps = 0; + } + assign_adjust(ls, nvars, nexps, &e); + var_add(ls, nvars); + } +} + +/* Parse 'function' statement. */ +static void parse_func(LexState *ls, BCLine line) +{ + FuncState *fs; + ExpDesc v, b; + int needself = 0; + lj_lex_next(ls); /* Skip 'function'. */ + /* Parse function name. */ + var_lookup(ls, &v); + while (ls->tok == '.') /* Multiple dot-separated fields. */ + expr_field(ls, &v); + if (ls->tok == ':') { /* Optional colon to signify method call. */ + needself = 1; + expr_field(ls, &v); + } + parse_body(ls, &b, needself, line); + fs = ls->fs; + bcemit_store(fs, &v, &b); + fs->bcbase[fs->pc - 1].line = line; /* Set line for the store. */ +} + +/* -- Control transfer statements ----------------------------------------- */ + +/* Check for end of block. */ +static int parse_isend(LexToken tok) +{ + switch (tok) { + case TK_else: case TK_elseif: case TK_end: case TK_until: case TK_eof: + return 1; + default: + return 0; + } +} + +/* Parse 'return' statement. */ +static void parse_return(LexState *ls) +{ + BCIns ins; + FuncState *fs = ls->fs; + lj_lex_next(ls); /* Skip 'return'. */ + fs->flags |= PROTO_HAS_RETURN; + if (parse_isend(ls->tok) || ls->tok == ';') { /* Bare return. */ + ins = BCINS_AD(BC_RET0, 0, 1); + } else { /* Return with one or more values. */ + ExpDesc e; /* Receives the _last_ expression in the list. */ + BCReg nret = expr_list(ls, &e); + if (nret == 1) { /* Return one result. */ + if (e.k == VCALL) { /* Check for tail call. */ +#ifdef LUAJIT_DISABLE_TAILCALL + goto notailcall; +#else + BCIns *ip = bcptr(fs, &e); + /* It doesn't pay off to add BC_VARGT just for 'return ...'. */ + if (bc_op(*ip) == BC_VARG) goto notailcall; + fs->pc--; + ins = BCINS_AD(bc_op(*ip)-BC_CALL+BC_CALLT, bc_a(*ip), bc_c(*ip)); +#endif + } else { /* Can return the result from any register. */ + ins = BCINS_AD(BC_RET1, expr_toanyreg(fs, &e), 2); + } + } else { + if (e.k == VCALL) { /* Append all results from a call. */ + notailcall: + setbc_b(bcptr(fs, &e), 0); + ins = BCINS_AD(BC_RETM, fs->nactvar, e.u.s.aux - fs->nactvar); + } else { + expr_tonextreg(fs, &e); /* Force contiguous registers. */ + ins = BCINS_AD(BC_RET, fs->nactvar, nret+1); + } + } + } + if (fs->flags & PROTO_CHILD) + bcemit_AJ(fs, BC_UCLO, 0, 0); /* May need to close upvalues first. */ + bcemit_INS(fs, ins); +} + +/* Parse 'break' statement. */ +static void parse_break(LexState *ls, int levels) +{ + ls->fs->bl->flags |= FSCOPE_BREAK; + MSize idx = gola_new(ls, NAME_BREAK, VSTACK_GOTO, bcemit_jmp(ls->fs)); + VarInfo *v = ls->vstack + idx; + if (levels <= 0 || levels > 255) { + /* HACK: reject invalid levels here, since breakn is uint8 */ + ls->linenumber = ls->fs->bcbase[v->startpc].line; + lj_lex_error(ls, 0, LJ_ERR_XBREAK); + } + v->breakn = (uint8_t)levels; +} + +static void parse_continue(LexState *ls) +{ + ls->fs->bl->flags |= FSCOPE_CONTINUE; + gola_new(ls, NAME_CONTINUE, VSTACK_GOTO, bcemit_jmp(ls->fs)); +} + +/* Parse 'goto' statement. */ +static void parse_goto(LexState *ls) +{ + FuncState *fs = ls->fs; + GCstr *name = lex_str(ls); + VarInfo *vl = gola_findlabel(ls, name); + if (vl) /* Treat backwards goto within same scope like a loop. */ + bcemit_AJ(fs, BC_LOOP, vl->slot, -1); /* No BC range check. */ + fs->bl->flags |= FSCOPE_GOLA; + gola_new(ls, name, VSTACK_GOTO, bcemit_jmp(fs)); +} + +/* Parse label. */ +static void parse_label(LexState *ls) +{ + FuncState *fs = ls->fs; + GCstr *name; + MSize idx; + fs->lasttarget = fs->pc; + fs->bl->flags |= FSCOPE_GOLA; + lj_lex_next(ls); /* Skip '::'. */ + name = lex_str(ls); + if (gola_findlabel(ls, name)) + lj_lex_error(ls, 0, LJ_ERR_XLDUP, strdata(name)); + idx = gola_new(ls, name, VSTACK_LABEL, fs->pc); + lex_check(ls, TK_label); + /* Recursively parse trailing statements: labels and ';' (Lua 5.2 only). */ + for (;;) { + if (ls->tok == TK_label) { + synlevel_begin(ls); + parse_label(ls); + synlevel_end(ls); + } else if (LJ_52 && ls->tok == ';') { + lj_lex_next(ls); + } else { + break; + } + } + /* Trailing label is considered to be outside of scope. */ + if (parse_isend(ls->tok) && ls->tok != TK_until) + ls->vstack[idx].slot = fs->bl->nactvar; + gola_resolve(ls, fs->bl, idx); +} + +/* -- Blocks, loops and conditional statements ---------------------------- */ + +/* Parse a block. */ +static void parse_block(LexState *ls) +{ + FuncState *fs = ls->fs; + FuncScope bl; + fscope_begin(fs, &bl, 0); + parse_chunk(ls); + fscope_end(fs); +} + +/* Parse 'while' statement. */ +static void parse_while(LexState *ls, BCLine line) +{ + FuncState *fs = ls->fs; + BCPos start, loop, condexit; + FuncScope bl; + lj_lex_next(ls); /* Skip 'while'. */ + start = fs->lasttarget = fs->pc; + condexit = expr_cond(ls); + fscope_begin(fs, &bl, FSCOPE_LOOP); + #if BLUAJIT_OPTIONAL_DO_TOKEN + lex_opt(ls, TK_do); + #else + lex_check(ls, TK_do); + #endif + loop = bcemit_AD(fs, BC_LOOP, fs->nactvar, 0); + parse_block(ls); + gola_resolvehere(ls, &bl, NAME_CONTINUE); + jmp_patch(fs, bcemit_jmp(fs), start); + lex_match(ls, TK_end, TK_while, line); + fscope_end(fs); + jmp_tohere(fs, condexit); + jmp_patchins(fs, loop, fs->pc); +} + +/* Parse 'repeat' statement. */ +static void parse_repeat(LexState *ls, BCLine line) +{ + FuncState *fs = ls->fs; + BCPos loop = fs->lasttarget = fs->pc; + BCPos condexit; + FuncScope bl1, bl2; + fscope_begin(fs, &bl1, FSCOPE_LOOP); /* Breakable loop scope. */ + fscope_begin(fs, &bl2, 0); /* Inner scope. */ + lj_lex_next(ls); /* Skip 'repeat'. */ + bcemit_AD(fs, BC_LOOP, fs->nactvar, 0); + parse_chunk(ls); + gola_resolvehere(ls, &bl2, NAME_CONTINUE); + lex_match(ls, TK_until, TK_repeat, line); + condexit = expr_cond(ls); /* Parse condition (still inside inner scope). */ + if (!(bl2.flags & FSCOPE_UPVAL)) { /* No upvalues? Just end inner scope. */ + fscope_end(fs); + } else { /* Otherwise generate: cond: UCLO+JMP out, !cond: UCLO+JMP loop. */ + parse_break(ls, 1); /* Break from loop and close upvalues. */ + jmp_tohere(fs, condexit); + fscope_end(fs); /* End inner scope and close upvalues. */ + condexit = bcemit_jmp(fs); + } + jmp_patch(fs, condexit, loop); /* Jump backwards if !cond. */ + jmp_patchins(fs, loop, fs->pc); + fscope_end(fs); /* End loop scope. */ +} + +/* Parse numeric 'for'. */ +static void parse_for_num(LexState *ls, GCstr *varname, BCLine line) +{ + FuncState *fs = ls->fs; + BCReg base = fs->freereg; + FuncScope bl; + BCPos loop, loopend; + /* Hidden control variables. */ + var_new_fixed(ls, FORL_IDX, VARNAME_FOR_IDX); + var_new_fixed(ls, FORL_STOP, VARNAME_FOR_STOP); + var_new_fixed(ls, FORL_STEP, VARNAME_FOR_STEP); + /* Visible copy of index variable. */ + var_new(ls, FORL_EXT, varname); + lex_check(ls, '='); + expr_next(ls); + lex_check(ls, ','); + expr_next(ls); + if (lex_opt(ls, ',')) { + expr_next(ls); + } else { + bcemit_AD(fs, BC_KSHORT, fs->freereg, 1); /* Default step is 1. */ + bcreg_reserve(fs, 1); + } + var_add(ls, 3); /* Hidden control variables. */ + #if BLUAJIT_OPTIONAL_DO_TOKEN + lex_opt(ls, TK_do); + #else + lex_check(ls, TK_do); + #endif + loop = bcemit_AJ(fs, BC_FORI, base, NO_JMP); + fscope_begin(fs, &bl, 0); /* Scope for visible variables. */ + var_add(ls, 1); + bcreg_reserve(fs, 1); + parse_block(ls); + gola_resolvehere(ls, &bl, NAME_CONTINUE); + fscope_end(fs); + /* Perform loop inversion. Loop control instructions are at the end. */ + loopend = bcemit_AJ(fs, BC_FORL, base, NO_JMP); + fs->bcbase[loopend].line = line; /* Fix line for control ins. */ + jmp_patchins(fs, loopend, loop+1); + jmp_patchins(fs, loop, fs->pc); +} + +/* Try to predict whether the iterator is next() and specialize the bytecode. +** Detecting next() and pairs() by name is simplistic, but quite effective. +** The interpreter backs off if the check for the closure fails at runtime. +*/ +static int predict_next(LexState *ls, FuncState *fs, BCPos pc) +{ + BCIns ins = fs->bcbase[pc].ins; + GCstr *name; + cTValue *o; + switch (bc_op(ins)) { + case BC_MOV: + if (bc_d(ins) >= fs->nactvar) return 0; + name = gco2str(gcref(var_get(ls, fs, bc_d(ins)).name)); + break; + case BC_UGET: + name = gco2str(gcref(ls->vstack[fs->uvmap[bc_d(ins)]].name)); + break; + case BC_GGET: + /* There's no inverse index (yet), so lookup the strings. */ + o = lj_tab_getstr(fs->kt, lj_str_newlit(ls->L, "pairs")); + if (o && tvhaskslot(o) && tvkslot(o) == bc_d(ins)) + return 1; + o = lj_tab_getstr(fs->kt, lj_str_newlit(ls->L, "next")); + if (o && tvhaskslot(o) && tvkslot(o) == bc_d(ins)) + return 1; + return 0; + default: + return 0; + } + return (name->len == 5 && !strcmp(strdata(name), "pairs")) || + (name->len == 4 && !strcmp(strdata(name), "next")); +} + +/* Parse 'for' iterator. */ +static void parse_for_iter(LexState *ls, GCstr *indexname) +{ + FuncState *fs = ls->fs; + ExpDesc e; + BCReg nvars = 0; + BCLine line; + BCReg base = fs->freereg + 3; + BCPos loop, loopend, exprpc = fs->pc; + FuncScope bl; + int isnext; + /* Hidden control variables. */ + var_new_fixed(ls, nvars++, VARNAME_FOR_GEN); + var_new_fixed(ls, nvars++, VARNAME_FOR_STATE); + var_new_fixed(ls, nvars++, VARNAME_FOR_CTL); + /* Visible variables returned from iterator. */ + var_new(ls, nvars++, indexname); + while (lex_opt(ls, ',')) + var_new(ls, nvars++, lex_str(ls)); + lex_check(ls, TK_in); + line = ls->linenumber; + assign_adjust(ls, 3, expr_list(ls, &e), &e); + /* The iterator needs another 3 [4] slots (func [pc] | state ctl). */ + bcreg_bump(fs, 3+ls->fr2); + isnext = (nvars <= 5 && fs->pc > exprpc && predict_next(ls, fs, exprpc)); + var_add(ls, 3); /* Hidden control variables. */ + #if BLUAJIT_OPTIONAL_DO_TOKEN + lex_opt(ls, TK_do); + #else + lex_check(ls, TK_do); + #endif + loop = bcemit_AJ(fs, isnext ? BC_ISNEXT : BC_JMP, base, NO_JMP); + fscope_begin(fs, &bl, 0); /* Scope for visible variables. */ + var_add(ls, nvars-3); + bcreg_reserve(fs, nvars-3); + parse_block(ls); + gola_resolvehere(ls, &bl, NAME_CONTINUE); + fscope_end(fs); + /* Perform loop inversion. Loop control instructions are at the end. */ + jmp_patchins(fs, loop, fs->pc); + bcemit_ABC(fs, isnext ? BC_ITERN : BC_ITERC, base, nvars-3+1, 2+1); + loopend = bcemit_AJ(fs, BC_ITERL, base, NO_JMP); + fs->bcbase[loopend-1].line = line; /* Fix line for control ins. */ + fs->bcbase[loopend].line = line; + jmp_patchins(fs, loopend, loop+1); +} + +/* Parse 'for' statement. */ +static void parse_for(LexState *ls, BCLine line) +{ + FuncState *fs = ls->fs; + GCstr *varname; + FuncScope bl; + fscope_begin(fs, &bl, FSCOPE_LOOP); + lj_lex_next(ls); /* Skip 'for'. */ + varname = lex_str(ls); /* Get first variable name. */ + if (ls->tok == '=') + parse_for_num(ls, varname, line); + else if (ls->tok == ',' || ls->tok == TK_in) + parse_for_iter(ls, varname); + else + err_syntax(ls, LJ_ERR_XFOR); + lex_match(ls, TK_end, TK_for, line); + fscope_end(fs); /* Resolve break list. */ +} + +/* Parse condition and 'then' block. */ +static BCPos parse_then(LexState *ls) +{ + BCPos condexit; + lj_lex_next(ls); /* Skip 'if' or 'elseif'. */ + condexit = expr_cond(ls); + #if BLUAJIT_OPTIONAL_THEN_TOKEN + lex_opt(ls, TK_then); + #else + lex_check(ls, TK_then); + #endif + parse_block(ls); + return condexit; +} + +/* Parse 'if' statement. */ +static void parse_if(LexState *ls, BCLine line) +{ + FuncState *fs = ls->fs; + BCPos flist; + BCPos escapelist = NO_JMP; + flist = parse_then(ls); + while (ls->tok == TK_elseif) { /* Parse multiple 'elseif' blocks. */ + jmp_append(fs, &escapelist, bcemit_jmp(fs)); + jmp_tohere(fs, flist); + flist = parse_then(ls); + } + if (ls->tok == TK_else) { /* Parse optional 'else' block. */ + jmp_append(fs, &escapelist, bcemit_jmp(fs)); + jmp_tohere(fs, flist); + lj_lex_next(ls); /* Skip 'else'. */ + parse_block(ls); + } else { + jmp_append(fs, &escapelist, flist); + } + jmp_tohere(fs, escapelist); + lex_match(ls, TK_end, TK_if, line); +} + +/* -- Parse statements ---------------------------------------------------- */ + +/* Parse a statement. Returns 1 if it must be the last one in a chunk. */ +static int parse_stmt(LexState *ls) +{ + BCLine line = ls->linenumber; + switch (ls->tok) { + case TK_if: + parse_if(ls, line); + break; + case TK_while: + parse_while(ls, line); + break; + case TK_do: + lj_lex_next(ls); + parse_block(ls); + lex_match(ls, TK_end, TK_do, line); + break; + case TK_for: + parse_for(ls, line); + break; + case TK_repeat: + parse_repeat(ls, line); + break; + case TK_function: + parse_func(ls, line); + break; + case TK_local: + lj_lex_next(ls); + parse_local(ls); + break; + case TK_return: + parse_return(ls); + return 1; /* Must be last. */ + case TK_break: + lj_lex_next(ls); + #if BLUAJIT_MULTILEVEL_BREAK + if (ls->tok == TK_number) { + int levels = numberVint(&ls->tokval); + lj_lex_next(ls); + parse_break(ls, levels); + } else + #endif + parse_break(ls, 1); + return !LJ_52; /* Must be last in Lua 5.1. */ + case TK_continue: + lj_lex_next(ls); + parse_continue(ls); + return !LJ_52; /* Must be last in Lua 5.1. */ +#if LJ_52 + case ';': + lj_lex_next(ls); + break; +#endif + case TK_label: + parse_label(ls); + break; + case TK_goto: + if (LJ_52 || lj_lex_lookahead(ls) == TK_name) { + lj_lex_next(ls); + parse_goto(ls); + break; + } + /* fallthrough */ + default: + parse_call_assign(ls); + break; + } + return 0; +} + +/* A chunk is a list of statements optionally separated by semicolons. */ +static void parse_chunk(LexState *ls) +{ + int islast = 0; + synlevel_begin(ls); + while (!islast && !parse_isend(ls->tok)) { + islast = parse_stmt(ls); + lex_opt(ls, ';'); + lj_assertLS(ls->fs->framesize >= ls->fs->freereg && + ls->fs->freereg >= ls->fs->nactvar, + "bad regalloc"); + ls->fs->freereg = ls->fs->nactvar; /* Free registers after each stmt. */ + } + synlevel_end(ls); +} + +/* Entry point of bytecode parser. */ +GCproto *lj_parse(LexState *ls) +{ + FuncState fs; + FuncScope bl; + GCproto *pt; + lua_State *L = ls->L; +#ifdef LUAJIT_DISABLE_DEBUGINFO + ls->chunkname = lj_str_newlit(L, "="); +#else + ls->chunkname = lj_str_newz(L, ls->chunkarg); +#endif + setstrV(L, L->top, ls->chunkname); /* Anchor chunkname string. */ + incr_top(L); + ls->level = 0; + fs_init(ls, &fs); + fs.linedefined = 0; + fs.numparams = 0; + fs.bcbase = NULL; + fs.bclim = 0; + fs.flags |= PROTO_VARARG; /* Main chunk is always a vararg func. */ + fscope_begin(&fs, &bl, 0); + bcemit_AD(&fs, BC_FUNCV, 0, 0); /* Placeholder. */ + lj_lex_next(ls); /* Read-ahead first token. */ + parse_chunk(ls); + if (ls->tok != TK_eof) + err_token(ls, TK_eof); + pt = fs_finish(ls, ls->linenumber); + L->top--; /* Drop chunkname. */ + lj_assertL(fs.prev == NULL && ls->fs == NULL, "mismatched frame nesting"); + lj_assertL(pt->sizeuv == 0, "toplevel proto has upvalues"); + return pt; +} + diff --git a/src/bluajit/lj_record.c b/src/bluajit/lj_record.c new file mode 100644 index 000000000..81c4f2635 --- /dev/null +++ b/src/bluajit/lj_record.c @@ -0,0 +1,2960 @@ +/* +** Trace recorder (bytecode -> SSA IR). +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#define lj_record_c +#define LUA_CORE + +#include "lj_obj.h" + +#if LJ_HASJIT + +#include "lj_err.h" +#include "lj_str.h" +#include "lj_tab.h" +#include "lj_meta.h" +#include "lj_frame.h" +#if LJ_HASFFI +#include "lj_ctype.h" +#endif +#include "lj_bc.h" +#include "lj_ff.h" +#if LJ_HASPROFILE +#include "lj_debug.h" +#endif +#include "lj_ir.h" +#include "lj_jit.h" +#include "lj_ircall.h" +#include "lj_iropt.h" +#include "lj_trace.h" +#include "lj_record.h" +#include "lj_ffrecord.h" +#include "lj_snap.h" +#include "lj_dispatch.h" +#include "lj_vm.h" +#include "lj_prng.h" + +/* Some local macros to save typing. Undef'd at the end. */ +#define IR(ref) (&J->cur.ir[(ref)]) + +/* Pass IR on to next optimization in chain (FOLD). */ +#define emitir(ot, a, b) (lj_ir_set(J, (ot), (a), (b)), lj_opt_fold(J)) + +/* Emit raw IR without passing through optimizations. */ +#define emitir_raw(ot, a, b) (lj_ir_set(J, (ot), (a), (b)), lj_ir_emit(J)) + +/* -- Sanity checks ------------------------------------------------------- */ + +#ifdef LUA_USE_ASSERT +/* Sanity check the whole IR -- sloooow. */ +static void rec_check_ir(jit_State *J) +{ + IRRef i, nins = J->cur.nins, nk = J->cur.nk; + lj_assertJ(nk <= REF_BIAS && nins >= REF_BIAS && nins < 65536, + "inconsistent IR layout"); + for (i = nk; i < nins; i++) { + IRIns *ir = IR(i); + uint32_t mode = lj_ir_mode[ir->o]; + IRRef op1 = ir->op1; + IRRef op2 = ir->op2; + const char *err = NULL; + switch (irm_op1(mode)) { + case IRMnone: + if (op1 != 0) err = "IRMnone op1 used"; + break; + case IRMref: + if (op1 < nk || (i >= REF_BIAS ? op1 >= i : op1 <= i)) + err = "IRMref op1 out of range"; + break; + case IRMlit: break; + case IRMcst: + if (i >= REF_BIAS) { err = "constant in IR range"; break; } + if (irt_is64(ir->t) && ir->o != IR_KNULL) + i++; + continue; + } + switch (irm_op2(mode)) { + case IRMnone: + if (op2) err = "IRMnone op2 used"; + break; + case IRMref: + if (op2 < nk || (i >= REF_BIAS ? op2 >= i : op2 <= i)) + err = "IRMref op2 out of range"; + break; + case IRMlit: break; + case IRMcst: err = "IRMcst op2"; break; + } + if (!err && ir->prev) { + if (ir->prev < nk || (i >= REF_BIAS ? ir->prev >= i : ir->prev <= i)) + err = "chain out of range"; + else if (ir->o != IR_NOP && IR(ir->prev)->o != ir->o) + err = "chain to different op"; + } + lj_assertJ(!err, "bad IR %04d op %d(%04d,%04d): %s", + i-REF_BIAS, + ir->o, + irm_op1(mode) == IRMref ? op1-REF_BIAS : op1, + irm_op2(mode) == IRMref ? op2-REF_BIAS : op2, + err); + } +} + +/* Compare stack slots and frames of the recorder and the VM. */ +static void rec_check_slots(jit_State *J) +{ + BCReg s, nslots = J->baseslot + J->maxslot; + int32_t depth = 0; + cTValue *base = J->L->base - J->baseslot; + lj_assertJ(J->baseslot >= 1+LJ_FR2, "bad baseslot"); + lj_assertJ(J->baseslot == 1+LJ_FR2 || (J->slot[J->baseslot-1] & TREF_FRAME), + "baseslot does not point to frame"); + lj_assertJ(nslots <= LJ_MAX_JSLOTS, "slot overflow"); + for (s = 0; s < nslots; s++) { + TRef tr = J->slot[s]; + if (tr) { + cTValue *tv = &base[s]; + IRRef ref = tref_ref(tr); + IRIns *ir = NULL; /* Silence compiler. */ + lj_assertJ(tv < J->L->top, "slot %d above top of Lua stack", s); + if (!LJ_FR2 || ref || !(tr & (TREF_FRAME | TREF_CONT))) { + lj_assertJ(ref >= J->cur.nk && ref < J->cur.nins, + "slot %d ref %04d out of range", s, ref - REF_BIAS); + ir = IR(ref); + lj_assertJ(irt_t(ir->t) == tref_t(tr), "slot %d IR type mismatch", s); + } + if (s == 0) { + lj_assertJ(tref_isfunc(tr), "frame slot 0 is not a function"); +#if LJ_FR2 + } else if (s == 1) { + lj_assertJ((tr & ~TREF_FRAME) == 0, "bad frame slot 1"); +#endif + } else if ((tr & TREF_FRAME)) { + GCfunc *fn = gco2func(frame_gc(tv)); + BCReg delta = (BCReg)(tv - frame_prev(tv)); +#if LJ_FR2 + lj_assertJ(!ref || ir_knum(ir)->u64 == tv->u64, + "frame slot %d PC mismatch", s); + tr = J->slot[s-1]; + ir = IR(tref_ref(tr)); +#endif + lj_assertJ(tref_isfunc(tr), + "frame slot %d is not a function", s-LJ_FR2); + lj_assertJ(!tref_isk(tr) || fn == ir_kfunc(ir), + "frame slot %d function mismatch", s-LJ_FR2); + lj_assertJ(s > delta + LJ_FR2 ? (J->slot[s-delta] & TREF_FRAME) + : (s == delta + LJ_FR2), + "frame slot %d broken chain", s-LJ_FR2); + depth++; + } else if ((tr & TREF_CONT)) { +#if LJ_FR2 + lj_assertJ(!ref || ir_knum(ir)->u64 == tv->u64, + "cont slot %d continuation mismatch", s); +#else + lj_assertJ(ir_kptr(ir) == gcrefp(tv->gcr, void), + "cont slot %d continuation mismatch", s); +#endif + lj_assertJ((J->slot[s+1+LJ_FR2] & TREF_FRAME), + "cont slot %d not followed by frame", s); + depth++; + } else if ((tr & TREF_KEYINDEX)) { + lj_assertJ(tref_isint(tr), "keyindex slot %d bad type %d", + s, tref_type(tr)); + } else { + /* Number repr. may differ, but other types must be the same. */ + lj_assertJ(tvisnumber(tv) ? tref_isnumber(tr) : + itype2irt(tv) == tref_type(tr), + "slot %d type mismatch: stack type %d vs IR type %d", + s, itypemap(tv), tref_type(tr)); + if (tref_isk(tr)) { /* Compare constants. */ + TValue tvk; + lj_ir_kvalue(J->L, &tvk, ir); + lj_assertJ((tvisnum(&tvk) && tvisnan(&tvk)) ? + (tvisnum(tv) && tvisnan(tv)) : + lj_obj_equal(tv, &tvk), + "slot %d const mismatch: stack %016llx vs IR %016llx", + s, tv->u64, tvk.u64); + } + } + } + } + lj_assertJ(J->framedepth == depth, + "frame depth mismatch %d vs %d", J->framedepth, depth); +} +#endif + +/* -- Type handling and specialization ------------------------------------ */ + +/* Note: these functions return tagged references (TRef). */ + +/* Specialize a slot to a specific type. Note: slot can be negative! */ +static TRef sloadt(jit_State *J, int32_t slot, IRType t, int mode) +{ + /* Caller may set IRT_GUARD in t. */ + TRef ref = emitir_raw(IRT(IR_SLOAD, t), (int32_t)J->baseslot+slot, mode); + J->base[slot] = ref; + return ref; +} + +/* Specialize a slot to the runtime type. Note: slot can be negative! */ +static TRef sload(jit_State *J, int32_t slot) +{ + IRType t = itype2irt(&J->L->base[slot]); + TRef ref = emitir_raw(IRTG(IR_SLOAD, t), (int32_t)J->baseslot+slot, + IRSLOAD_TYPECHECK); + if (irtype_ispri(t)) ref = TREF_PRI(t); /* Canonicalize primitive refs. */ + J->base[slot] = ref; + return ref; +} + +/* Get TRef from slot. Load slot and specialize if not done already. */ +#define getslot(J, s) (J->base[(s)] ? J->base[(s)] : sload(J, (int32_t)(s))) + +/* Get TRef for current function. */ +static TRef getcurrf(jit_State *J) +{ + if (J->base[-1-LJ_FR2]) + return J->base[-1-LJ_FR2]; + /* Non-base frame functions ought to be loaded already. */ + lj_assertJ(J->baseslot == 1+LJ_FR2, "bad baseslot"); + return sloadt(J, -1-LJ_FR2, IRT_FUNC, IRSLOAD_READONLY); +} + +/* Compare for raw object equality. +** Returns 0 if the objects are the same. +** Returns 1 if they are different, but the same type. +** Returns 2 for two different types. +** Comparisons between primitives always return 1 -- no caller cares about it. +*/ +int lj_record_objcmp(jit_State *J, TRef a, TRef b, cTValue *av, cTValue *bv) +{ + int diff = !lj_obj_equal(av, bv); + if (!tref_isk2(a, b)) { /* Shortcut, also handles primitives. */ + IRType ta = tref_isinteger(a) ? IRT_INT : tref_type(a); + IRType tb = tref_isinteger(b) ? IRT_INT : tref_type(b); + if (ta != tb) { + /* Widen mixed number/int comparisons to number/number comparison. */ + if (ta == IRT_INT && tb == IRT_NUM) { + a = emitir(IRTN(IR_CONV), a, IRCONV_NUM_INT); + ta = IRT_NUM; + } else if (ta == IRT_NUM && tb == IRT_INT) { + b = emitir(IRTN(IR_CONV), b, IRCONV_NUM_INT); + } else { + return 2; /* Two different types are never equal. */ + } + } + emitir(IRTG(diff ? IR_NE : IR_EQ, ta), a, b); + } + return diff; +} + +/* Constify a value. Returns 0 for non-representable object types. */ +TRef lj_record_constify(jit_State *J, cTValue *o) +{ + if (tvisgcv(o)) + return lj_ir_kgc(J, gcV(o), itype2irt(o)); + else if (tvisint(o)) + return lj_ir_kint(J, intV(o)); + else if (tvisnum(o)) + return lj_ir_knumint(J, numV(o)); + else if (tvisbool(o)) + return TREF_PRI(itype2irt(o)); + else + return 0; /* Can't represent lightuserdata (pointless). */ +} + +/* Emit a VLOAD with the correct type. */ +TRef lj_record_vload(jit_State *J, TRef ref, MSize idx, IRType t) +{ + TRef tr = emitir(IRTG(IR_VLOAD, t), ref, idx); + if (irtype_ispri(t)) tr = TREF_PRI(t); /* Canonicalize primitives. */ + return tr; +} + +/* -- Record loop ops ----------------------------------------------------- */ + +/* Loop event. */ +typedef enum { + LOOPEV_LEAVE, /* Loop is left or not entered. */ + LOOPEV_ENTERLO, /* Loop is entered with a low iteration count left. */ + LOOPEV_ENTER /* Loop is entered. */ +} LoopEvent; + +/* Canonicalize slots: convert integers to numbers. */ +static void canonicalize_slots(jit_State *J) +{ + BCReg s; + if (LJ_DUALNUM) return; + for (s = J->baseslot+J->maxslot-1; s >= 1; s--) { + TRef tr = J->slot[s]; + if (tref_isinteger(tr) && !(tr & TREF_KEYINDEX)) { + IRIns *ir = IR(tref_ref(tr)); + if (!(ir->o == IR_SLOAD && (ir->op2 & (IRSLOAD_READONLY)))) + J->slot[s] = emitir(IRTN(IR_CONV), tr, IRCONV_NUM_INT); + } + } +} + +/* Stop recording. */ +void lj_record_stop(jit_State *J, TraceLink linktype, TraceNo lnk) +{ +#ifdef LUAJIT_ENABLE_TABLE_BUMP + if (J->retryrec) + lj_trace_err(J, LJ_TRERR_RETRY); +#endif + lj_trace_end(J); + J->cur.linktype = (uint8_t)linktype; + J->cur.link = (uint16_t)lnk; + /* Looping back at the same stack level? */ + if (lnk == J->cur.traceno && J->framedepth + J->retdepth == 0) { + if ((J->flags & JIT_F_OPT_LOOP)) /* Shall we try to create a loop? */ + goto nocanon; /* Do not canonicalize or we lose the narrowing. */ + if (J->cur.root) /* Otherwise ensure we always link to the root trace. */ + J->cur.link = J->cur.root; + } + canonicalize_slots(J); +nocanon: + /* Note: all loop ops must set J->pc to the following instruction! */ + lj_snap_add(J); /* Add loop snapshot. */ + J->needsnap = 0; + J->mergesnap = 1; /* In case recording continues. */ +} + +/* Search bytecode backwards for a int/num constant slot initializer. */ +static TRef find_kinit(jit_State *J, const BCIns *endpc, BCReg slot, IRType t) +{ + /* This algorithm is rather simplistic and assumes quite a bit about + ** how the bytecode is generated. It works fine for FORI initializers, + ** but it won't necessarily work in other cases (e.g. iterator arguments). + ** It doesn't do anything fancy, either (like backpropagating MOVs). + */ + const BCIns *pc, *startpc = proto_bc(J->pt); + for (pc = endpc-1; pc > startpc; pc--) { + BCIns ins = *pc; + BCOp op = bc_op(ins); + /* First try to find the last instruction that stores to this slot. */ + if (bcmode_a(op) == BCMbase && bc_a(ins) <= slot) { + return 0; /* Multiple results, e.g. from a CALL or KNIL. */ + } else if (bcmode_a(op) == BCMdst && bc_a(ins) == slot) { + if (op == BC_KSHORT || op == BC_KNUM) { /* Found const. initializer. */ + /* Now try to verify there's no forward jump across it. */ + const BCIns *kpc = pc; + for (; pc > startpc; pc--) + if (bc_op(*pc) == BC_JMP) { + const BCIns *target = pc+bc_j(*pc)+1; + if (target > kpc && target <= endpc) + return 0; /* Conditional assignment. */ + } + if (op == BC_KSHORT) { + int32_t k = (int32_t)(int16_t)bc_d(ins); + return t == IRT_INT ? lj_ir_kint(J, k) : lj_ir_knum(J, (lua_Number)k); + } else { + cTValue *tv = proto_knumtv(J->pt, bc_d(ins)); + if (t == IRT_INT) { + int32_t k = numberVint(tv); + if (tvisint(tv) || numV(tv) == (lua_Number)k) /* -0 is ok here. */ + return lj_ir_kint(J, k); + return 0; /* Type mismatch. */ + } else { + return lj_ir_knum(J, numberVnum(tv)); + } + } + } + return 0; /* Non-constant initializer. */ + } + } + return 0; /* No assignment to this slot found? */ +} + +/* Load and optionally convert a FORI argument from a slot. */ +static TRef fori_load(jit_State *J, BCReg slot, IRType t, int mode) +{ + int conv = (tvisint(&J->L->base[slot]) != (t==IRT_INT)) ? IRSLOAD_CONVERT : 0; + return sloadt(J, (int32_t)slot, + t + (((mode & IRSLOAD_TYPECHECK) || + (conv && t == IRT_INT && !(mode >> 16))) ? + IRT_GUARD : 0), + mode + conv); +} + +/* Peek before FORI to find a const initializer. Otherwise load from slot. */ +static TRef fori_arg(jit_State *J, const BCIns *fori, BCReg slot, + IRType t, int mode) +{ + TRef tr = J->base[slot]; + if (!tr) { + tr = find_kinit(J, fori, slot, t); + if (!tr) + tr = fori_load(J, slot, t, mode); + } + return tr; +} + +/* Return the direction of the FOR loop iterator. +** It's important to exactly reproduce the semantics of the interpreter. +*/ +static int rec_for_direction(cTValue *o) +{ + return (tvisint(o) ? intV(o) : (int32_t)o->u32.hi) >= 0; +} + +/* Simulate the runtime behavior of the FOR loop iterator. */ +static LoopEvent rec_for_iter(IROp *op, cTValue *o, int isforl) +{ + lua_Number stopv = numberVnum(&o[FORL_STOP]); + lua_Number idxv = numberVnum(&o[FORL_IDX]); + lua_Number stepv = numberVnum(&o[FORL_STEP]); + if (isforl) + idxv += stepv; + if (rec_for_direction(&o[FORL_STEP])) { + if (idxv <= stopv) { + *op = IR_LE; + return idxv + 2*stepv > stopv ? LOOPEV_ENTERLO : LOOPEV_ENTER; + } + *op = IR_GT; return LOOPEV_LEAVE; + } else { + if (stopv <= idxv) { + *op = IR_GE; + return idxv + 2*stepv < stopv ? LOOPEV_ENTERLO : LOOPEV_ENTER; + } + *op = IR_LT; return LOOPEV_LEAVE; + } +} + +/* Record checks for FOR loop overflow and step direction. */ +static void rec_for_check(jit_State *J, IRType t, int dir, + TRef stop, TRef step, int init) +{ + if (!tref_isk(step)) { + /* Non-constant step: need a guard for the direction. */ + TRef zero = (t == IRT_INT) ? lj_ir_kint(J, 0) : lj_ir_knum_zero(J); + emitir(IRTG(dir ? IR_GE : IR_LT, t), step, zero); + /* Add hoistable overflow checks for a narrowed FORL index. */ + if (init && t == IRT_INT) { + if (tref_isk(stop)) { + /* Constant stop: optimize check away or to a range check for step. */ + int32_t k = IR(tref_ref(stop))->i; + if (dir) { + if (k > 0) + emitir(IRTGI(IR_LE), step, lj_ir_kint(J, (int32_t)0x7fffffff-k)); + } else { + if (k < 0) + emitir(IRTGI(IR_GE), step, lj_ir_kint(J, (int32_t)0x80000000-k)); + } + } else { + /* Stop+step variable: need full overflow check. */ + TRef tr = emitir(IRTGI(IR_ADDOV), step, stop); + emitir(IRTI(IR_USE), tr, 0); /* ADDOV is weak. Avoid dead result. */ + } + } + } else if (init && t == IRT_INT && !tref_isk(stop)) { + /* Constant step: optimize overflow check to a range check for stop. */ + int32_t k = IR(tref_ref(step))->i; + k = (int32_t)(dir ? 0x7fffffff : 0x80000000) - k; + emitir(IRTGI(dir ? IR_LE : IR_GE), stop, lj_ir_kint(J, k)); + } +} + +/* Record a FORL instruction. */ +static void rec_for_loop(jit_State *J, const BCIns *fori, ScEvEntry *scev, + int init) +{ + BCReg ra = bc_a(*fori); + cTValue *tv = &J->L->base[ra]; + TRef idx = J->base[ra+FORL_IDX]; + IRType t = idx ? tref_type(idx) : + (init || LJ_DUALNUM) ? lj_opt_narrow_forl(J, tv) : IRT_NUM; + int mode = IRSLOAD_INHERIT + + ((!LJ_DUALNUM || tvisint(tv) == (t == IRT_INT)) ? IRSLOAD_READONLY : 0); + TRef stop = fori_arg(J, fori, ra+FORL_STOP, t, mode); + TRef step = fori_arg(J, fori, ra+FORL_STEP, t, mode); + int tc, dir = rec_for_direction(&tv[FORL_STEP]); + lj_assertJ(bc_op(*fori) == BC_FORI || bc_op(*fori) == BC_JFORI, + "bad bytecode %d instead of FORI/JFORI", bc_op(*fori)); + scev->t.irt = t; + scev->dir = dir; + scev->stop = tref_ref(stop); + scev->step = tref_ref(step); + rec_for_check(J, t, dir, stop, step, init); + scev->start = tref_ref(find_kinit(J, fori, ra+FORL_IDX, IRT_INT)); + tc = (LJ_DUALNUM && + !(scev->start && irref_isk(scev->stop) && irref_isk(scev->step) && + tvisint(&tv[FORL_IDX]) == (t == IRT_INT))) ? + IRSLOAD_TYPECHECK : 0; + if (tc) { + J->base[ra+FORL_STOP] = stop; + J->base[ra+FORL_STEP] = step; + } + if (!idx) + idx = fori_load(J, ra+FORL_IDX, t, + IRSLOAD_INHERIT + tc + (J->scev.start << 16)); + if (!init) + J->base[ra+FORL_IDX] = idx = emitir(IRT(IR_ADD, t), idx, step); + J->base[ra+FORL_EXT] = idx; + scev->idx = tref_ref(idx); + setmref(scev->pc, fori); + J->maxslot = ra+FORL_EXT+1; +} + +/* Record FORL/JFORL or FORI/JFORI. */ +static LoopEvent rec_for(jit_State *J, const BCIns *fori, int isforl) +{ + BCReg ra = bc_a(*fori); + TValue *tv = &J->L->base[ra]; + TRef *tr = &J->base[ra]; + IROp op; + LoopEvent ev; + TRef stop; + IRType t; + if (isforl) { /* Handle FORL/JFORL opcodes. */ + TRef idx = tr[FORL_IDX]; + if (mref(J->scev.pc, const BCIns) == fori && tref_ref(idx) == J->scev.idx) { + t = J->scev.t.irt; + stop = J->scev.stop; + idx = emitir(IRT(IR_ADD, t), idx, J->scev.step); + tr[FORL_EXT] = tr[FORL_IDX] = idx; + } else { + ScEvEntry scev; + rec_for_loop(J, fori, &scev, 0); + t = scev.t.irt; + stop = scev.stop; + } + } else { /* Handle FORI/JFORI opcodes. */ + BCReg i; + lj_meta_for(J->L, tv); + t = (LJ_DUALNUM || tref_isint(tr[FORL_IDX])) ? lj_opt_narrow_forl(J, tv) : + IRT_NUM; + for (i = FORL_IDX; i <= FORL_STEP; i++) { + if (!tr[i]) sload(J, ra+i); + lj_assertJ(tref_isnumber_str(tr[i]), "bad FORI argument type"); + if (tref_isstr(tr[i])) + tr[i] = emitir(IRTG(IR_STRTO, IRT_NUM), tr[i], 0); + if (t == IRT_INT) { + if (!tref_isinteger(tr[i])) + tr[i] = emitir(IRTGI(IR_CONV), tr[i], IRCONV_INT_NUM|IRCONV_CHECK); + } else { + if (!tref_isnum(tr[i])) + tr[i] = emitir(IRTN(IR_CONV), tr[i], IRCONV_NUM_INT); + } + } + tr[FORL_EXT] = tr[FORL_IDX]; + stop = tr[FORL_STOP]; + rec_for_check(J, t, rec_for_direction(&tv[FORL_STEP]), + stop, tr[FORL_STEP], 1); + } + + ev = rec_for_iter(&op, tv, isforl); + if (ev == LOOPEV_LEAVE) { + J->maxslot = ra+FORL_EXT+1; + J->pc = fori+1; + } else { + J->maxslot = ra; + J->pc = fori+bc_j(*fori)+1; + } + lj_snap_add(J); + + emitir(IRTG(op, t), tr[FORL_IDX], stop); + + if (ev == LOOPEV_LEAVE) { + J->maxslot = ra; + J->pc = fori+bc_j(*fori)+1; + } else { + J->maxslot = ra+FORL_EXT+1; + J->pc = fori+1; + } + J->needsnap = 1; + return ev; +} + +/* Record ITERL/JITERL. */ +static LoopEvent rec_iterl(jit_State *J, const BCIns iterins) +{ + BCReg ra = bc_a(iterins); + if (!tref_isnil(getslot(J, ra))) { /* Looping back? */ + J->base[ra-1] = J->base[ra]; /* Copy result of ITERC to control var. */ + J->maxslot = ra-1+bc_b(J->pc[-1]); + J->pc += bc_j(iterins)+1; + return LOOPEV_ENTER; + } else { + J->maxslot = ra-3; + J->pc++; + return LOOPEV_LEAVE; + } +} + +/* Record LOOP/JLOOP. Now, that was easy. */ +static LoopEvent rec_loop(jit_State *J, BCReg ra, int skip) +{ + if (ra < J->maxslot) J->maxslot = ra; + J->pc += skip; + return LOOPEV_ENTER; +} + +/* Check if a loop repeatedly failed to trace because it didn't loop back. */ +static int innerloopleft(jit_State *J, const BCIns *pc) +{ + ptrdiff_t i; + for (i = 0; i < PENALTY_SLOTS; i++) + if (mref(J->penalty[i].pc, const BCIns) == pc) { + if ((J->penalty[i].reason == LJ_TRERR_LLEAVE || + J->penalty[i].reason == LJ_TRERR_LINNER) && + J->penalty[i].val >= 2*PENALTY_MIN) + return 1; + break; + } + return 0; +} + +/* Handle the case when an interpreted loop op is hit. */ +static void rec_loop_interp(jit_State *J, const BCIns *pc, LoopEvent ev) +{ + if (J->parent == 0 && J->exitno == 0) { + if (pc == J->startpc && J->framedepth + J->retdepth == 0) { + if (bc_op(J->cur.startins) == BC_ITERN) return; /* See rec_itern(). */ + /* Same loop? */ + if (ev == LOOPEV_LEAVE) /* Must loop back to form a root trace. */ + lj_trace_err(J, LJ_TRERR_LLEAVE); + lj_record_stop(J, LJ_TRLINK_LOOP, J->cur.traceno); /* Looping trace. */ + } else if (ev != LOOPEV_LEAVE) { /* Entering inner loop? */ + /* It's usually better to abort here and wait until the inner loop + ** is traced. But if the inner loop repeatedly didn't loop back, + ** this indicates a low trip count. In this case try unrolling + ** an inner loop even in a root trace. But it's better to be a bit + ** more conservative here and only do it for very short loops. + */ + if (bc_j(*pc) != -1 && !innerloopleft(J, pc)) + lj_trace_err(J, LJ_TRERR_LINNER); /* Root trace hit an inner loop. */ + if ((ev != LOOPEV_ENTERLO && + J->loopref && J->cur.nins - J->loopref > 24) || --J->loopunroll < 0) + lj_trace_err(J, LJ_TRERR_LUNROLL); /* Limit loop unrolling. */ + J->loopref = J->cur.nins; + } + } else if (ev != LOOPEV_LEAVE) { /* Side trace enters an inner loop. */ + J->loopref = J->cur.nins; + if (--J->loopunroll < 0) + lj_trace_err(J, LJ_TRERR_LUNROLL); /* Limit loop unrolling. */ + } /* Side trace continues across a loop that's left or not entered. */ +} + +/* Handle the case when an already compiled loop op is hit. */ +static void rec_loop_jit(jit_State *J, TraceNo lnk, LoopEvent ev) +{ + if (J->parent == 0 && J->exitno == 0) { /* Root trace hit an inner loop. */ + /* Better let the inner loop spawn a side trace back here. */ + lj_trace_err(J, LJ_TRERR_LINNER); + } else if (ev != LOOPEV_LEAVE) { /* Side trace enters a compiled loop. */ + J->instunroll = 0; /* Cannot continue across a compiled loop op. */ + if (J->pc == J->startpc && J->framedepth + J->retdepth == 0) + lj_record_stop(J, LJ_TRLINK_LOOP, J->cur.traceno); /* Form extra loop. */ + else + lj_record_stop(J, LJ_TRLINK_ROOT, lnk); /* Link to the loop. */ + } /* Side trace continues across a loop that's left or not entered. */ +} + +/* Record ITERN. */ +static LoopEvent rec_itern(jit_State *J, BCReg ra, BCReg rb) +{ +#if LJ_BE + /* YAGNI: Disabled on big-endian due to issues with lj_vm_next, + ** IR_HIOP, RID_RETLO/RID_RETHI and ra_destpair. + */ + UNUSED(ra); UNUSED(rb); + setintV(&J->errinfo, (int32_t)BC_ITERN); + lj_trace_err_info(J, LJ_TRERR_NYIBC); +#else + RecordIndex ix; + /* Since ITERN is recorded at the start, we need our own loop detection. */ + if (J->pc == J->startpc && + J->framedepth + J->retdepth == 0 && J->parent == 0 && J->exitno == 0) { + IRRef ref = REF_FIRST + LJ_HASPROFILE; +#ifdef LUAJIT_ENABLE_CHECKHOOK + ref += 3; +#endif + if (J->cur.nins > ref || + (LJ_HASPROFILE && J->cur.nins == ref && J->cur.ir[ref-1].o != IR_PROF)) { + J->instunroll = 0; /* Cannot continue unrolling across an ITERN. */ + lj_record_stop(J, LJ_TRLINK_LOOP, J->cur.traceno); /* Looping trace. */ + return LOOPEV_ENTER; + } + } + J->maxslot = ra; + lj_snap_add(J); /* Required to make JLOOP the first ins in a side-trace. */ + ix.tab = getslot(J, ra-2); + ix.key = J->base[ra-1] ? J->base[ra-1] : + sloadt(J, (int32_t)(ra-1), IRT_GUARD|IRT_INT, + IRSLOAD_TYPECHECK|IRSLOAD_KEYINDEX); + copyTV(J->L, &ix.tabv, &J->L->base[ra-2]); + copyTV(J->L, &ix.keyv, &J->L->base[ra-1]); + ix.idxchain = (rb < 3); /* Omit value type check, if unused. */ + ix.mobj = 1; /* We need the next index, too. */ + J->maxslot = ra + lj_record_next(J, &ix); + J->needsnap = 1; + if (!tref_isnil(ix.key)) { /* Looping back? */ + J->base[ra-1] = ix.mobj | TREF_KEYINDEX; /* Control var has next index. */ + J->base[ra] = ix.key; + J->base[ra+1] = ix.val; + J->pc += bc_j(J->pc[1])+2; + return LOOPEV_ENTER; + } else { + J->maxslot = ra-3; + J->pc += 2; + return LOOPEV_LEAVE; + } +#endif +} + +/* Record ISNEXT. */ +static void rec_isnext(jit_State *J, BCReg ra) +{ + cTValue *b = &J->L->base[ra-3]; + if (tvisfunc(b) && funcV(b)->c.ffid == FF_next && + tvistab(b+1) && tvisnil(b+2)) { + /* These checks are folded away for a compiled pairs(). */ + TRef func = getslot(J, ra-3); + TRef trid = emitir(IRT(IR_FLOAD, IRT_U8), func, IRFL_FUNC_FFID); + emitir(IRTGI(IR_EQ), trid, lj_ir_kint(J, FF_next)); + (void)getslot(J, ra-2); /* Type check for table. */ + (void)getslot(J, ra-1); /* Type check for nil key. */ + J->base[ra-1] = lj_ir_kint(J, 0) | TREF_KEYINDEX; + J->maxslot = ra; + } else { /* Abort trace. Interpreter will despecialize bytecode. */ + lj_trace_err(J, LJ_TRERR_RECERR); + } +} + +/* -- Record profiler hook checks ----------------------------------------- */ + +#if LJ_HASPROFILE + +/* Need to insert profiler hook check? */ +static int rec_profile_need(jit_State *J, GCproto *pt, const BCIns *pc) +{ + GCproto *ppt; + lj_assertJ(J->prof_mode == 'f' || J->prof_mode == 'l', + "bad profiler mode %c", J->prof_mode); + if (!pt) + return 0; + ppt = J->prev_pt; + J->prev_pt = pt; + if (pt != ppt && ppt) { + J->prev_line = -1; + return 1; + } + if (J->prof_mode == 'l') { + BCLine line = lj_debug_line(pt, proto_bcpos(pt, pc)); + BCLine pline = J->prev_line; + J->prev_line = line; + if (pline != line) + return 1; + } + return 0; +} + +static void rec_profile_ins(jit_State *J, const BCIns *pc) +{ + if (J->prof_mode && rec_profile_need(J, J->pt, pc)) { + emitir(IRTG(IR_PROF, IRT_NIL), 0, 0); + lj_snap_add(J); + } +} + +static void rec_profile_ret(jit_State *J) +{ + if (J->prof_mode == 'f') { + emitir(IRTG(IR_PROF, IRT_NIL), 0, 0); + J->prev_pt = NULL; + lj_snap_add(J); + } +} + +#endif + +/* -- Record calls and returns -------------------------------------------- */ + +/* Specialize to the runtime value of the called function or its prototype. */ +static TRef rec_call_specialize(jit_State *J, GCfunc *fn, TRef tr) +{ + TRef kfunc; + if (isluafunc(fn)) { + GCproto *pt = funcproto(fn); + /* Too many closures created? Probably not a monomorphic function. */ + if (pt->flags >= PROTO_CLC_POLY) { /* Specialize to prototype instead. */ + TRef trpt = emitir(IRT(IR_FLOAD, IRT_PGC), tr, IRFL_FUNC_PC); + emitir(IRTG(IR_EQ, IRT_PGC), trpt, lj_ir_kptr(J, proto_bc(pt))); + (void)lj_ir_kgc(J, obj2gco(pt), IRT_PROTO); /* Prevent GC of proto. */ + return tr; + } + } else { + /* Don't specialize to non-monomorphic builtins. */ + switch (fn->c.ffid) { + case FF_coroutine_wrap_aux: + case FF_string_gmatch_aux: + /* NYI: io_file_iter doesn't have an ffid, yet. */ + { /* Specialize to the ffid. */ + TRef trid = emitir(IRT(IR_FLOAD, IRT_U8), tr, IRFL_FUNC_FFID); + emitir(IRTGI(IR_EQ), trid, lj_ir_kint(J, fn->c.ffid)); + } + return tr; + default: + /* NYI: don't specialize to non-monomorphic C functions. */ + break; + } + } + /* Otherwise specialize to the function (closure) value itself. */ + kfunc = lj_ir_kfunc(J, fn); + emitir(IRTG(IR_EQ, IRT_FUNC), tr, kfunc); + return kfunc; +} + +/* Record call setup. */ +static void rec_call_setup(jit_State *J, BCReg func, ptrdiff_t nargs) +{ + RecordIndex ix; + TValue *functv = &J->L->base[func]; + TRef kfunc, *fbase = &J->base[func]; + ptrdiff_t i; + (void)getslot(J, func); /* Ensure func has a reference. */ + for (i = 1; i <= nargs; i++) + (void)getslot(J, func+LJ_FR2+i); /* Ensure all args have a reference. */ + if (!tref_isfunc(fbase[0])) { /* Resolve __call metamethod. */ + ix.tab = fbase[0]; + copyTV(J->L, &ix.tabv, functv); + if (!lj_record_mm_lookup(J, &ix, MM_call) || !tref_isfunc(ix.mobj)) + lj_trace_err(J, LJ_TRERR_NOMM); + for (i = ++nargs; i > LJ_FR2; i--) /* Shift arguments up. */ + fbase[i+LJ_FR2] = fbase[i+LJ_FR2-1]; +#if LJ_FR2 + fbase[2] = fbase[0]; +#endif + fbase[0] = ix.mobj; /* Replace function. */ + functv = &ix.mobjv; + } + kfunc = rec_call_specialize(J, funcV(functv), fbase[0]); +#if LJ_FR2 + fbase[0] = kfunc; + fbase[1] = TREF_FRAME; +#else + fbase[0] = kfunc | TREF_FRAME; +#endif + J->maxslot = (BCReg)nargs; +} + +/* Record call. */ +void lj_record_call(jit_State *J, BCReg func, ptrdiff_t nargs) +{ + rec_call_setup(J, func, nargs); + /* Bump frame. */ + J->framedepth++; + J->base += func+1+LJ_FR2; + J->baseslot += func+1+LJ_FR2; + if (J->baseslot + J->maxslot >= LJ_MAX_JSLOTS) + lj_trace_err(J, LJ_TRERR_STACKOV); +} + +/* Record tail call. */ +void lj_record_tailcall(jit_State *J, BCReg func, ptrdiff_t nargs) +{ + rec_call_setup(J, func, nargs); + if (frame_isvarg(J->L->base - 1)) { + BCReg cbase = (BCReg)frame_delta(J->L->base - 1); + if (--J->framedepth < 0) + lj_trace_err(J, LJ_TRERR_NYIRETL); + J->baseslot -= (BCReg)cbase; + J->base -= cbase; + func += cbase; + } + /* Move func + args down. */ + if (LJ_FR2 && J->baseslot == 2) + J->base[func+1] = TREF_FRAME; + memmove(&J->base[-1-LJ_FR2], &J->base[func], sizeof(TRef)*(J->maxslot+1+LJ_FR2)); + /* Note: the new TREF_FRAME is now at J->base[-1] (even for slot #0). */ + /* Tailcalls can form a loop, so count towards the loop unroll limit. */ + if (++J->tailcalled > J->loopunroll) + lj_trace_err(J, LJ_TRERR_LUNROLL); +} + +/* Check unroll limits for down-recursion. */ +static int check_downrec_unroll(jit_State *J, GCproto *pt) +{ + IRRef ptref; + for (ptref = J->chain[IR_KGC]; ptref; ptref = IR(ptref)->prev) + if (ir_kgc(IR(ptref)) == obj2gco(pt)) { + int count = 0; + IRRef ref; + for (ref = J->chain[IR_RETF]; ref; ref = IR(ref)->prev) + if (IR(ref)->op1 == ptref) + count++; + if (count) { + if (J->pc == J->startpc) { + if (count + J->tailcalled > J->param[JIT_P_recunroll]) + return 1; + } else { + lj_trace_err(J, LJ_TRERR_DOWNREC); + } + } + } + return 0; +} + +static TRef rec_cat(jit_State *J, BCReg baseslot, BCReg topslot); + +/* Record return. */ +void lj_record_ret(jit_State *J, BCReg rbase, ptrdiff_t gotresults) +{ + TValue *frame = J->L->base - 1; + ptrdiff_t i; + BCReg baseadj = 0; + for (i = 0; i < gotresults; i++) + (void)getslot(J, rbase+i); /* Ensure all results have a reference. */ + while (frame_ispcall(frame)) { /* Immediately resolve pcall() returns. */ + BCReg cbase = (BCReg)frame_delta(frame); + if (--J->framedepth <= 0) + lj_trace_err(J, LJ_TRERR_NYIRETL); + lj_assertJ(J->baseslot > 1+LJ_FR2, "bad baseslot for return"); + gotresults++; + baseadj += cbase; + rbase += cbase; + J->baseslot -= (BCReg)cbase; + J->base -= cbase; + J->base[--rbase] = TREF_TRUE; /* Prepend true to results. */ + frame = frame_prevd(frame); + J->needsnap = 1; /* Stop catching on-trace errors. */ + } + /* Return to lower frame via interpreter for unhandled cases. */ + if (J->framedepth == 0 && J->pt && bc_isret(bc_op(*J->pc)) && + (!frame_islua(frame) || + (J->parent == 0 && J->exitno == 0 && + !bc_isret(bc_op(J->cur.startins))))) { + /* NYI: specialize to frame type and return directly, not via RET*. */ + for (i = 0; i < (ptrdiff_t)rbase; i++) + J->base[i] = 0; /* Purge dead slots. */ + J->maxslot = rbase + (BCReg)gotresults; + lj_record_stop(J, LJ_TRLINK_RETURN, 0); /* Return to interpreter. */ + return; + } + if (frame_isvarg(frame)) { + BCReg cbase = (BCReg)frame_delta(frame); + if (--J->framedepth < 0) /* NYI: return of vararg func to lower frame. */ + lj_trace_err(J, LJ_TRERR_NYIRETL); + lj_assertJ(J->baseslot > 1+LJ_FR2, "bad baseslot for return"); + baseadj += cbase; + rbase += cbase; + J->baseslot -= (BCReg)cbase; + J->base -= cbase; + frame = frame_prevd(frame); + } + if (frame_islua(frame)) { /* Return to Lua frame. */ + BCIns callins = *(frame_pc(frame)-1); + ptrdiff_t nresults = bc_b(callins) ? (ptrdiff_t)bc_b(callins)-1 :gotresults; + BCReg cbase = bc_a(callins); + GCproto *pt = funcproto(frame_func(frame - (cbase+1+LJ_FR2))); + if ((pt->flags & PROTO_NOJIT)) + lj_trace_err(J, LJ_TRERR_CJITOFF); + if (J->framedepth == 0 && J->pt && frame == J->L->base - 1) { + if (!J->cur.root && check_downrec_unroll(J, pt)) { + J->maxslot = (BCReg)(rbase + gotresults); + lj_snap_purge(J); + lj_record_stop(J, LJ_TRLINK_DOWNREC, J->cur.traceno); /* Down-rec. */ + return; + } + lj_snap_add(J); + } + for (i = 0; i < nresults; i++) /* Adjust results. */ + J->base[i-1-LJ_FR2] = i < gotresults ? J->base[rbase+i] : TREF_NIL; + J->maxslot = cbase+(BCReg)nresults; + if (J->framedepth > 0) { /* Return to a frame that is part of the trace. */ + J->framedepth--; + lj_assertJ(J->baseslot > cbase+1+LJ_FR2, "bad baseslot for return"); + J->baseslot -= cbase+1+LJ_FR2; + J->base -= cbase+1+LJ_FR2; + } else if (J->parent == 0 && J->exitno == 0 && + !bc_isret(bc_op(J->cur.startins))) { + /* Return to lower frame would leave the loop in a root trace. */ + lj_trace_err(J, LJ_TRERR_LLEAVE); + } else if (J->needsnap) { /* Tailcalled to ff with side-effects. */ + lj_trace_err(J, LJ_TRERR_NYIRETL); /* No way to insert snapshot here. */ + } else if (1 + pt->framesize >= LJ_MAX_JSLOTS) { + lj_trace_err(J, LJ_TRERR_STACKOV); + } else { /* Return to lower frame. Guard for the target we return to. */ + TRef trpt = lj_ir_kgc(J, obj2gco(pt), IRT_PROTO); + TRef trpc = lj_ir_kptr(J, (void *)frame_pc(frame)); + emitir(IRTG(IR_RETF, IRT_PGC), trpt, trpc); + J->retdepth++; + J->needsnap = 1; + J->scev.idx = REF_NIL; + lj_assertJ(J->baseslot == 1+LJ_FR2, "bad baseslot for return"); + /* Shift result slots up and clear the slots of the new frame below. */ + memmove(J->base + cbase, J->base-1-LJ_FR2, sizeof(TRef)*nresults); + memset(J->base-1-LJ_FR2, 0, sizeof(TRef)*(cbase+1+LJ_FR2)); + } + } else if (frame_iscont(frame)) { /* Return to continuation frame. */ + ASMFunction cont = frame_contf(frame); + BCReg cbase = (BCReg)frame_delta(frame); + if ((J->framedepth -= 2) < 0) + lj_trace_err(J, LJ_TRERR_NYIRETL); + J->baseslot -= (BCReg)cbase; + J->base -= cbase; + J->maxslot = cbase-(2<base[dst] = gotresults ? J->base[cbase+rbase] : TREF_NIL; + if (dst >= J->maxslot) { + J->maxslot = dst+1; + } + } else if (cont == lj_cont_nop) { + /* Nothing to do here. */ + } else if (cont == lj_cont_cat) { + BCReg bslot = bc_b(*(frame_contpc(frame)-1)); + TRef tr = gotresults ? J->base[cbase+rbase] : TREF_NIL; + if (bslot != J->maxslot) { /* Concatenate the remainder. */ + /* Simulate lower frame and result. */ + TValue *b = J->L->base - baseadj, save; + /* Can't handle MM_concat + CALLT + fast func side-effects. */ + if (J->postproc != LJ_POST_NONE) + lj_trace_err(J, LJ_TRERR_NYIRETL); + J->base[J->maxslot] = tr; + copyTV(J->L, &save, b-(2<L, b-(2<L->base = b - cbase; + tr = rec_cat(J, bslot, cbase-(2<L->base + cbase; /* Undo. */ + J->L->base = b + baseadj; + copyTV(J->L, b-(2<= 0xffffff00) { + lj_err_throw(J->L, -(int32_t)tr); /* Propagate errors. */ + } else if (tr) { /* Store final result. */ + BCReg dst = bc_a(*(frame_contpc(frame)-1)); + J->base[dst] = tr; + if (dst >= J->maxslot) { + J->maxslot = dst+1; + } + } /* Otherwise continue with another __concat call. */ + } else { + /* Result type already specialized. */ + lj_assertJ(cont == lj_cont_condf || cont == lj_cont_condt, + "bad continuation type"); + } + } else { + lj_trace_err(J, LJ_TRERR_NYIRETL); /* NYI: handle return to C frame. */ + } + lj_assertJ(J->baseslot >= 1+LJ_FR2, "bad baseslot for return"); +} + +/* -- Metamethod handling ------------------------------------------------- */ + +/* Prepare to record call to metamethod. */ +static BCReg rec_mm_prep(jit_State *J, ASMFunction cont) +{ + BCReg s, top = cont == lj_cont_cat ? J->maxslot : curr_proto(J->L)->framesize; +#if LJ_FR2 + J->base[top] = lj_ir_k64(J, IR_KNUM, u64ptr(contptr(cont))); + J->base[top+1] = TREF_CONT; +#else + J->base[top] = lj_ir_kptr(J, contptr(cont)) | TREF_CONT; +#endif + J->framedepth++; + for (s = J->maxslot; s < top; s++) + J->base[s] = 0; /* Clear frame gap to avoid resurrecting previous refs. */ + return top+1+LJ_FR2; +} + +/* Record metamethod lookup. */ +int lj_record_mm_lookup(jit_State *J, RecordIndex *ix, MMS mm) +{ + RecordIndex mix; + GCtab *mt; + if (tref_istab(ix->tab)) { + mt = tabref(tabV(&ix->tabv)->metatable); + mix.tab = emitir(IRT(IR_FLOAD, IRT_TAB), ix->tab, IRFL_TAB_META); + } else if (tref_isudata(ix->tab)) { + int udtype = udataV(&ix->tabv)->udtype; + mt = tabref(udataV(&ix->tabv)->metatable); + mix.tab = emitir(IRT(IR_FLOAD, IRT_TAB), ix->tab, IRFL_UDATA_META); + /* The metatables of special userdata objects are treated as immutable. */ + if (udtype != UDTYPE_USERDATA) { + cTValue *mo; + if (LJ_HASFFI && udtype == UDTYPE_FFI_CLIB) { + /* Specialize to the C library namespace object. */ + emitir(IRTG(IR_EQ, IRT_PGC), ix->tab, lj_ir_kptr(J, udataV(&ix->tabv))); + } else { + /* Specialize to the type of userdata. */ + TRef tr = emitir(IRT(IR_FLOAD, IRT_U8), ix->tab, IRFL_UDATA_UDTYPE); + emitir(IRTGI(IR_EQ), tr, lj_ir_kint(J, udtype)); + } + immutable_mt: + mo = lj_tab_getstr(mt, mmname_str(J2G(J), mm)); + ix->mt = mix.tab; + ix->mtv = mt; + if (!mo || tvisnil(mo)) + return 0; /* No metamethod. */ + /* Treat metamethod or index table as immutable, too. */ + if (!(tvisfunc(mo) || tvistab(mo))) + lj_trace_err(J, LJ_TRERR_BADTYPE); + copyTV(J->L, &ix->mobjv, mo); + ix->mobj = lj_ir_kgc(J, gcV(mo), tvisfunc(mo) ? IRT_FUNC : IRT_TAB); + return 1; /* Got metamethod or index table. */ + } + } else { + /* Specialize to base metatable. Must flush mcode in lua_setmetatable(). */ + mt = tabref(basemt_obj(J2G(J), &ix->tabv)); + if (mt == NULL) { + ix->mt = TREF_NIL; + return 0; /* No metamethod. */ + } + /* The cdata metatable is treated as immutable. */ + if (LJ_HASFFI && tref_iscdata(ix->tab)) goto immutable_mt; + ix->mt = mix.tab = lj_ir_ggfload(J, IRT_TAB, + GG_OFS(g.gcroot[GCROOT_BASEMT+itypemap(&ix->tabv)])); + goto nocheck; + } + ix->mt = mt ? mix.tab : TREF_NIL; + emitir(IRTG(mt ? IR_NE : IR_EQ, IRT_TAB), mix.tab, lj_ir_knull(J, IRT_TAB)); +nocheck: + if (mt) { + GCstr *mmstr = mmname_str(J2G(J), mm); + cTValue *mo = lj_tab_getstr(mt, mmstr); + if (mo && !tvisnil(mo)) + copyTV(J->L, &ix->mobjv, mo); + ix->mtv = mt; + settabV(J->L, &mix.tabv, mt); + setstrV(J->L, &mix.keyv, mmstr); + mix.key = lj_ir_kstr(J, mmstr); + mix.val = 0; + mix.idxchain = 0; + ix->mobj = lj_record_idx(J, &mix); + return !tref_isnil(ix->mobj); /* 1 if metamethod found, 0 if not. */ + } + return 0; /* No metamethod. */ +} + +/* Record call to arithmetic metamethod. */ +static TRef rec_mm_arith(jit_State *J, RecordIndex *ix, MMS mm) +{ + /* Set up metamethod call first to save ix->tab and ix->tabv. */ + BCReg func = rec_mm_prep(J, mm == MM_concat ? lj_cont_cat : lj_cont_ra); + TRef *base = J->base + func; + TValue *basev = J->L->base + func; + base[1+LJ_FR2] = ix->tab; base[2+LJ_FR2] = ix->key; + copyTV(J->L, basev+1+LJ_FR2, &ix->tabv); + copyTV(J->L, basev+2+LJ_FR2, &ix->keyv); + if (!lj_record_mm_lookup(J, ix, mm)) { /* Lookup mm on 1st operand. */ + if (mm != MM_unm) { + ix->tab = ix->key; + copyTV(J->L, &ix->tabv, &ix->keyv); + if (lj_record_mm_lookup(J, ix, mm)) /* Lookup mm on 2nd operand. */ + goto ok; + } + lj_trace_err(J, LJ_TRERR_NOMM); + } +ok: + base[0] = ix->mobj; +#if LJ_FR2 + base[1] = 0; +#endif + copyTV(J->L, basev+0, &ix->mobjv); + lj_record_call(J, func, 2); + return 0; /* No result yet. */ +} + +/* Record call to __len metamethod. */ +static TRef rec_mm_len(jit_State *J, TRef tr, TValue *tv) +{ + RecordIndex ix; + ix.tab = tr; + copyTV(J->L, &ix.tabv, tv); + if (lj_record_mm_lookup(J, &ix, MM_len)) { + BCReg func = rec_mm_prep(J, lj_cont_ra); + TRef *base = J->base + func; + TValue *basev = J->L->base + func; + base[0] = ix.mobj; copyTV(J->L, basev+0, &ix.mobjv); + base += LJ_FR2; + basev += LJ_FR2; + base[1] = tr; copyTV(J->L, basev+1, tv); +#if LJ_52 + base[2] = tr; copyTV(J->L, basev+2, tv); +#else + base[2] = TREF_NIL; setnilV(basev+2); +#endif + lj_record_call(J, func, 2); + } else { + if (LJ_52 && tref_istab(tr)) + return emitir(IRTI(IR_ALEN), tr, TREF_NIL); + lj_trace_err(J, LJ_TRERR_NOMM); + } + return 0; /* No result yet. */ +} + +/* Call a comparison metamethod. */ +static void rec_mm_callcomp(jit_State *J, RecordIndex *ix, int op) +{ + BCReg func = rec_mm_prep(J, (op&1) ? lj_cont_condf : lj_cont_condt); + TRef *base = J->base + func + LJ_FR2; + TValue *tv = J->L->base + func + LJ_FR2; + base[-LJ_FR2] = ix->mobj; base[1] = ix->val; base[2] = ix->key; + copyTV(J->L, tv-LJ_FR2, &ix->mobjv); + copyTV(J->L, tv+1, &ix->valv); + copyTV(J->L, tv+2, &ix->keyv); + lj_record_call(J, func, 2); +} + +/* Record call to equality comparison metamethod (for tab and udata only). */ +static void rec_mm_equal(jit_State *J, RecordIndex *ix, int op) +{ + ix->tab = ix->val; + copyTV(J->L, &ix->tabv, &ix->valv); + if (lj_record_mm_lookup(J, ix, MM_eq)) { /* Lookup mm on 1st operand. */ + cTValue *bv; + TRef mo1 = ix->mobj; + TValue mo1v; + copyTV(J->L, &mo1v, &ix->mobjv); + /* Avoid the 2nd lookup and the objcmp if the metatables are equal. */ + bv = &ix->keyv; + if (tvistab(bv) && tabref(tabV(bv)->metatable) == ix->mtv) { + TRef mt2 = emitir(IRT(IR_FLOAD, IRT_TAB), ix->key, IRFL_TAB_META); + emitir(IRTG(IR_EQ, IRT_TAB), mt2, ix->mt); + } else if (tvisudata(bv) && tabref(udataV(bv)->metatable) == ix->mtv) { + TRef mt2 = emitir(IRT(IR_FLOAD, IRT_TAB), ix->key, IRFL_UDATA_META); + emitir(IRTG(IR_EQ, IRT_TAB), mt2, ix->mt); + } else { /* Lookup metamethod on 2nd operand and compare both. */ + ix->tab = ix->key; + copyTV(J->L, &ix->tabv, bv); + if (!lj_record_mm_lookup(J, ix, MM_eq) || + lj_record_objcmp(J, mo1, ix->mobj, &mo1v, &ix->mobjv)) + return; + } + rec_mm_callcomp(J, ix, op); + } +} + +/* Record call to ordered comparison metamethods (for arbitrary objects). */ +static void rec_mm_comp(jit_State *J, RecordIndex *ix, int op) +{ + ix->tab = ix->val; + copyTV(J->L, &ix->tabv, &ix->valv); + while (1) { + MMS mm = (op & 2) ? MM_le : MM_lt; /* Try __le + __lt or only __lt. */ +#if LJ_52 + if (!lj_record_mm_lookup(J, ix, mm)) { /* Lookup mm on 1st operand. */ + ix->tab = ix->key; + copyTV(J->L, &ix->tabv, &ix->keyv); + if (!lj_record_mm_lookup(J, ix, mm)) /* Lookup mm on 2nd operand. */ + goto nomatch; + } + rec_mm_callcomp(J, ix, op); + return; +#else + if (lj_record_mm_lookup(J, ix, mm)) { /* Lookup mm on 1st operand. */ + cTValue *bv; + TRef mo1 = ix->mobj; + TValue mo1v; + copyTV(J->L, &mo1v, &ix->mobjv); + /* Avoid the 2nd lookup and the objcmp if the metatables are equal. */ + bv = &ix->keyv; + if (tvistab(bv) && tabref(tabV(bv)->metatable) == ix->mtv) { + TRef mt2 = emitir(IRT(IR_FLOAD, IRT_TAB), ix->key, IRFL_TAB_META); + emitir(IRTG(IR_EQ, IRT_TAB), mt2, ix->mt); + } else if (tvisudata(bv) && tabref(udataV(bv)->metatable) == ix->mtv) { + TRef mt2 = emitir(IRT(IR_FLOAD, IRT_TAB), ix->key, IRFL_UDATA_META); + emitir(IRTG(IR_EQ, IRT_TAB), mt2, ix->mt); + } else { /* Lookup metamethod on 2nd operand and compare both. */ + ix->tab = ix->key; + copyTV(J->L, &ix->tabv, bv); + if (!lj_record_mm_lookup(J, ix, mm) || + lj_record_objcmp(J, mo1, ix->mobj, &mo1v, &ix->mobjv)) + goto nomatch; + } + rec_mm_callcomp(J, ix, op); + return; + } +#endif + nomatch: + /* Lookup failed. Retry with __lt and swapped operands. */ + if (!(op & 2)) break; /* Already at __lt. Interpreter will throw. */ + ix->tab = ix->key; ix->key = ix->val; ix->val = ix->tab; + copyTV(J->L, &ix->tabv, &ix->keyv); + copyTV(J->L, &ix->keyv, &ix->valv); + copyTV(J->L, &ix->valv, &ix->tabv); + op ^= 3; + } +} + +#if LJ_HASFFI +/* Setup call to cdata comparison metamethod. */ +static void rec_mm_comp_cdata(jit_State *J, RecordIndex *ix, int op, MMS mm) +{ + lj_snap_add(J); + if (tref_iscdata(ix->val)) { + ix->tab = ix->val; + copyTV(J->L, &ix->tabv, &ix->valv); + } else { + lj_assertJ(tref_iscdata(ix->key), "cdata expected"); + ix->tab = ix->key; + copyTV(J->L, &ix->tabv, &ix->keyv); + } + lj_record_mm_lookup(J, ix, mm); + rec_mm_callcomp(J, ix, op); +} +#endif + +/* -- Indexed access ------------------------------------------------------ */ + +#ifdef LUAJIT_ENABLE_TABLE_BUMP +/* Bump table allocations in bytecode when they grow during recording. */ +static void rec_idx_bump(jit_State *J, RecordIndex *ix) +{ + RBCHashEntry *rbc = &J->rbchash[(ix->tab & (RBCHASH_SLOTS-1))]; + if (tref_ref(ix->tab) == rbc->ref) { + const BCIns *pc = mref(rbc->pc, const BCIns); + GCtab *tb = tabV(&ix->tabv); + uint32_t nhbits; + IRIns *ir; + if (!tvisnil(&ix->keyv)) + (void)lj_tab_set(J->L, tb, &ix->keyv); /* Grow table right now. */ + nhbits = tb->hmask > 0 ? lj_fls(tb->hmask)+1 : 0; + ir = IR(tref_ref(ix->tab)); + if (ir->o == IR_TNEW) { + uint32_t ah = bc_d(*pc); + uint32_t asize = ah & 0x7ff, hbits = ah >> 11; + if (nhbits > hbits) hbits = nhbits; + if (tb->asize > asize) { + asize = tb->asize <= 0x7ff ? tb->asize : 0x7ff; + } + if ((asize | (hbits<<11)) != ah) { /* Has the size changed? */ + /* Patch bytecode, but continue recording (for more patching). */ + setbc_d(pc, (asize | (hbits<<11))); + /* Patching TNEW operands is only safe if the trace is aborted. */ + ir->op1 = asize; ir->op2 = hbits; + J->retryrec = 1; /* Abort the trace at the end of recording. */ + } + } else if (ir->o == IR_TDUP) { + GCtab *tpl = gco2tab(proto_kgc(&gcref(rbc->pt)->pt, ~(ptrdiff_t)bc_d(*pc))); + /* Grow template table, but preserve keys with nil values. */ + if ((tb->asize > tpl->asize && (1u << nhbits)-1 == tpl->hmask) || + (tb->asize == tpl->asize && (1u << nhbits)-1 > tpl->hmask)) { + Node *node = noderef(tpl->node); + uint32_t i, hmask = tpl->hmask, asize; + TValue *array; + for (i = 0; i <= hmask; i++) { + if (!tvisnil(&node[i].key) && tvisnil(&node[i].val)) + settabV(J->L, &node[i].val, tpl); + } + if (!tvisnil(&ix->keyv) && tref_isk(ix->key)) { + TValue *o = lj_tab_set(J->L, tpl, &ix->keyv); + if (tvisnil(o)) settabV(J->L, o, tpl); + } + lj_tab_resize(J->L, tpl, tb->asize, nhbits); + node = noderef(tpl->node); + hmask = tpl->hmask; + for (i = 0; i <= hmask; i++) { + /* This is safe, since template tables only hold immutable values. */ + if (tvistab(&node[i].val)) + setnilV(&node[i].val); + } + /* The shape of the table may have changed. Clean up array part, too. */ + asize = tpl->asize; + array = tvref(tpl->array); + for (i = 0; i < asize; i++) { + if (tvistab(&array[i])) + setnilV(&array[i]); + } + J->retryrec = 1; /* Abort the trace at the end of recording. */ + } + } + } +} +#endif + +/* Record bounds-check. */ +static void rec_idx_abc(jit_State *J, TRef asizeref, TRef ikey, uint32_t asize) +{ + /* Try to emit invariant bounds checks. */ + if ((J->flags & (JIT_F_OPT_LOOP|JIT_F_OPT_ABC)) == + (JIT_F_OPT_LOOP|JIT_F_OPT_ABC)) { + IRRef ref = tref_ref(ikey); + IRIns *ir = IR(ref); + int32_t ofs = 0; + IRRef ofsref = 0; + /* Handle constant offsets. */ + if (ir->o == IR_ADD && irref_isk(ir->op2)) { + ofsref = ir->op2; + ofs = IR(ofsref)->i; + ref = ir->op1; + ir = IR(ref); + } + /* Got scalar evolution analysis results for this reference? */ + if (ref == J->scev.idx) { + int32_t stop; + lj_assertJ(irt_isint(J->scev.t) && ir->o == IR_SLOAD, + "only int SCEV supported"); + stop = numberVint(&(J->L->base - J->baseslot)[ir->op1 + FORL_STOP]); + /* Runtime value for stop of loop is within bounds? */ + if ((uint64_t)stop + ofs < (uint64_t)asize) { + /* Emit invariant bounds check for stop. */ + uint32_t abc = IRTG(IR_ABC, tref_isk(asizeref) ? IRT_U32 : IRT_P32); + emitir(abc, asizeref, ofs == 0 ? J->scev.stop : + emitir(IRTI(IR_ADD), J->scev.stop, ofsref)); + /* Emit invariant bounds check for start, if not const or negative. */ + if (!(J->scev.dir && J->scev.start && + (int64_t)IR(J->scev.start)->i + ofs >= 0)) + emitir(abc, asizeref, ikey); + return; + } + } + } + emitir(IRTGI(IR_ABC), asizeref, ikey); /* Emit regular bounds check. */ +} + +/* Record indexed key lookup. */ +static TRef rec_idx_key(jit_State *J, RecordIndex *ix, IRRef *rbref, + IRType1 *rbguard) +{ + TRef key; + GCtab *t = tabV(&ix->tabv); + ix->oldv = lj_tab_get(J->L, t, &ix->keyv); /* Lookup previous value. */ + *rbref = 0; + rbguard->irt = 0; + + /* Integer keys are looked up in the array part first. */ + key = ix->key; + if (tref_isnumber(key)) { + int32_t k = numberVint(&ix->keyv); + if (!tvisint(&ix->keyv) && numV(&ix->keyv) != (lua_Number)k) + k = LJ_MAX_ASIZE; + if ((MSize)k < LJ_MAX_ASIZE) { /* Potential array key? */ + TRef ikey = lj_opt_narrow_index(J, key); + TRef asizeref = emitir(IRTI(IR_FLOAD), ix->tab, IRFL_TAB_ASIZE); + if ((MSize)k < t->asize) { /* Currently an array key? */ + TRef arrayref; + rec_idx_abc(J, asizeref, ikey, t->asize); + arrayref = emitir(IRT(IR_FLOAD, IRT_PGC), ix->tab, IRFL_TAB_ARRAY); + return emitir(IRT(IR_AREF, IRT_PGC), arrayref, ikey); + } else { /* Currently not in array (may be an array extension)? */ + emitir(IRTGI(IR_ULE), asizeref, ikey); /* Inv. bounds check. */ + if (k == 0 && tref_isk(key)) + key = lj_ir_knum_zero(J); /* Canonicalize 0 or +-0.0 to +0.0. */ + /* And continue with the hash lookup. */ + } + } else if (!tref_isk(key)) { + /* We can rule out const numbers which failed the integerness test + ** above. But all other numbers are potential array keys. + */ + if (t->asize == 0) { /* True sparse tables have an empty array part. */ + /* Guard that the array part stays empty. */ + TRef tmp = emitir(IRTI(IR_FLOAD), ix->tab, IRFL_TAB_ASIZE); + emitir(IRTGI(IR_EQ), tmp, lj_ir_kint(J, 0)); + } else { + lj_trace_err(J, LJ_TRERR_NYITMIX); + } + } + } + + /* Otherwise the key is located in the hash part. */ + if (t->hmask == 0) { /* Shortcut for empty hash part. */ + /* Guard that the hash part stays empty. */ + TRef tmp = emitir(IRTI(IR_FLOAD), ix->tab, IRFL_TAB_HMASK); + emitir(IRTGI(IR_EQ), tmp, lj_ir_kint(J, 0)); + return lj_ir_kkptr(J, niltvg(J2G(J))); + } + if (tref_isinteger(key)) /* Hash keys are based on numbers, not ints. */ + key = emitir(IRTN(IR_CONV), key, IRCONV_NUM_INT); + if (tref_isk(key)) { + /* Optimize lookup of constant hash keys. */ + GCSize hslot = (GCSize)((char *)ix->oldv-(char *)&noderef(t->node)[0].val); + if (hslot <= t->hmask*(GCSize)sizeof(Node) && + hslot <= 65535*(GCSize)sizeof(Node)) { + TRef node, kslot, hm; + *rbref = J->cur.nins; /* Mark possible rollback point. */ + *rbguard = J->guardemit; + hm = emitir(IRTI(IR_FLOAD), ix->tab, IRFL_TAB_HMASK); + emitir(IRTGI(IR_EQ), hm, lj_ir_kint(J, (int32_t)t->hmask)); + node = emitir(IRT(IR_FLOAD, IRT_PGC), ix->tab, IRFL_TAB_NODE); + kslot = lj_ir_kslot(J, key, (IRRef)(hslot / sizeof(Node))); + return emitir(IRTG(IR_HREFK, IRT_PGC), node, kslot); + } + } + /* Fall back to a regular hash lookup. */ + return emitir(IRT(IR_HREF, IRT_PGC), ix->tab, key); +} + +/* Determine whether a key is NOT one of the fast metamethod names. */ +static int nommstr(jit_State *J, TRef key) +{ + if (tref_isstr(key)) { + if (tref_isk(key)) { + GCstr *str = ir_kstr(IR(tref_ref(key))); + uint32_t mm; + for (mm = 0; mm <= MM_FAST; mm++) + if (mmname_str(J2G(J), mm) == str) + return 0; /* MUST be one the fast metamethod names. */ + } else { + return 0; /* Variable string key MAY be a metamethod name. */ + } + } + return 1; /* CANNOT be a metamethod name. */ +} + +/* Record indexed load/store. */ +TRef lj_record_idx(jit_State *J, RecordIndex *ix) +{ + TRef xref; + IROp xrefop, loadop; + IRRef rbref; + IRType1 rbguard; + cTValue *oldv; + + while (!tref_istab(ix->tab)) { /* Handle non-table lookup. */ + /* Never call raw lj_record_idx() on non-table. */ + lj_assertJ(ix->idxchain != 0, "bad usage"); + if (!lj_record_mm_lookup(J, ix, ix->val ? MM_newindex : MM_index)) + lj_trace_err(J, LJ_TRERR_NOMM); + handlemm: + if (tref_isfunc(ix->mobj)) { /* Handle metamethod call. */ + BCReg func = rec_mm_prep(J, ix->val ? lj_cont_nop : lj_cont_ra); + TRef *base = J->base + func + LJ_FR2; + TValue *tv = J->L->base + func + LJ_FR2; + base[-LJ_FR2] = ix->mobj; base[1] = ix->tab; base[2] = ix->key; + setfuncV(J->L, tv-LJ_FR2, funcV(&ix->mobjv)); + copyTV(J->L, tv+1, &ix->tabv); + copyTV(J->L, tv+2, &ix->keyv); + if (ix->val) { + base[3] = ix->val; + copyTV(J->L, tv+3, &ix->valv); + lj_record_call(J, func, 3); /* mobj(tab, key, val) */ + return 0; + } else { + lj_record_call(J, func, 2); /* res = mobj(tab, key) */ + return 0; /* No result yet. */ + } + } +#if LJ_HASBUFFER + /* The index table of buffer objects is treated as immutable. */ + if (ix->mt == TREF_NIL && !ix->val && + tref_isudata(ix->tab) && udataV(&ix->tabv)->udtype == UDTYPE_BUFFER && + tref_istab(ix->mobj) && tref_isstr(ix->key) && tref_isk(ix->key)) { + cTValue *val = lj_tab_getstr(tabV(&ix->mobjv), strV(&ix->keyv)); + TRef tr = lj_record_constify(J, val); + if (tr) return tr; /* Specialize to the value, i.e. a method. */ + } +#endif + /* Otherwise retry lookup with metaobject. */ + ix->tab = ix->mobj; + copyTV(J->L, &ix->tabv, &ix->mobjv); + if (--ix->idxchain == 0) + lj_trace_err(J, LJ_TRERR_IDXLOOP); + } + + /* First catch nil and NaN keys for tables. */ + if (tvisnil(&ix->keyv) || (tvisnum(&ix->keyv) && tvisnan(&ix->keyv))) { + if (ix->val) /* Better fail early. */ + lj_trace_err(J, LJ_TRERR_STORENN); + if (tref_isk(ix->key)) { + if (ix->idxchain && lj_record_mm_lookup(J, ix, MM_index)) + goto handlemm; + return TREF_NIL; + } + } + + /* Record the key lookup. */ + xref = rec_idx_key(J, ix, &rbref, &rbguard); + xrefop = IR(tref_ref(xref))->o; + loadop = xrefop == IR_AREF ? IR_ALOAD : IR_HLOAD; + /* The lj_meta_tset() inconsistency is gone, but better play safe. */ + oldv = xrefop == IR_KKPTR ? (cTValue *)ir_kptr(IR(tref_ref(xref))) : ix->oldv; + + if (ix->val == 0) { /* Indexed load */ + IRType t = itype2irt(oldv); + TRef res; + if (oldv == niltvg(J2G(J))) { + emitir(IRTG(IR_EQ, IRT_PGC), xref, lj_ir_kkptr(J, niltvg(J2G(J)))); + res = TREF_NIL; + } else { + res = emitir(IRTG(loadop, t), xref, 0); + } + if (tref_ref(res) < rbref) { /* HREFK + load forwarded? */ + lj_ir_rollback(J, rbref); /* Rollback to eliminate hmask guard. */ + J->guardemit = rbguard; + } + if (t == IRT_NIL && ix->idxchain && lj_record_mm_lookup(J, ix, MM_index)) + goto handlemm; + if (irtype_ispri(t)) res = TREF_PRI(t); /* Canonicalize primitives. */ + return res; + } else { /* Indexed store. */ + GCtab *mt = tabref(tabV(&ix->tabv)->metatable); + int keybarrier = tref_isgcv(ix->key) && !tref_isnil(ix->val); + if (tref_ref(xref) < rbref) { /* HREFK forwarded? */ + lj_ir_rollback(J, rbref); /* Rollback to eliminate hmask guard. */ + J->guardemit = rbguard; + } + if (tvisnil(oldv)) { /* Previous value was nil? */ + /* Need to duplicate the hasmm check for the early guards. */ + int hasmm = 0; + if (ix->idxchain && mt) { + cTValue *mo = lj_tab_getstr(mt, mmname_str(J2G(J), MM_newindex)); + hasmm = mo && !tvisnil(mo); + } + if (hasmm) + emitir(IRTG(loadop, IRT_NIL), xref, 0); /* Guard for nil value. */ + else if (xrefop == IR_HREF) + emitir(IRTG(oldv == niltvg(J2G(J)) ? IR_EQ : IR_NE, IRT_PGC), + xref, lj_ir_kkptr(J, niltvg(J2G(J)))); + if (ix->idxchain && lj_record_mm_lookup(J, ix, MM_newindex)) { + lj_assertJ(hasmm, "inconsistent metamethod handling"); + goto handlemm; + } + lj_assertJ(!hasmm, "inconsistent metamethod handling"); + if (oldv == niltvg(J2G(J))) { /* Need to insert a new key. */ + TRef key = ix->key; + if (tref_isinteger(key)) { /* NEWREF needs a TValue as a key. */ + key = emitir(IRTN(IR_CONV), key, IRCONV_NUM_INT); + } else if (tref_isnum(key)) { + if (tref_isk(key)) { + if (tvismzero(&ix->keyv)) + key = lj_ir_knum_zero(J); /* Canonicalize -0.0 to +0.0. */ + } else { + emitir(IRTG(IR_EQ, IRT_NUM), key, key); /* Check for !NaN. */ + } + } + xref = emitir(IRT(IR_NEWREF, IRT_PGC), ix->tab, key); + keybarrier = 0; /* NEWREF already takes care of the key barrier. */ +#ifdef LUAJIT_ENABLE_TABLE_BUMP + if ((J->flags & JIT_F_OPT_SINK)) /* Avoid a separate flag. */ + rec_idx_bump(J, ix); +#endif + } + } else if (!lj_opt_fwd_wasnonnil(J, loadop, tref_ref(xref))) { + /* Cannot derive that the previous value was non-nil, must do checks. */ + if (xrefop == IR_HREF) /* Guard against store to niltv. */ + emitir(IRTG(IR_NE, IRT_PGC), xref, lj_ir_kkptr(J, niltvg(J2G(J)))); + if (ix->idxchain) { /* Metamethod lookup required? */ + /* A check for NULL metatable is cheaper (hoistable) than a load. */ + if (!mt) { + TRef mtref = emitir(IRT(IR_FLOAD, IRT_TAB), ix->tab, IRFL_TAB_META); + emitir(IRTG(IR_EQ, IRT_TAB), mtref, lj_ir_knull(J, IRT_TAB)); + } else { + IRType t = itype2irt(oldv); + emitir(IRTG(loadop, t), xref, 0); /* Guard for non-nil value. */ + } + if (lj_record_mm_lookup(J, ix, MM_usedindex)) + goto handlemm; + } + } else { + keybarrier = 0; /* Previous non-nil value kept the key alive. */ + if (ix->idxchain && lj_record_mm_lookup(J, ix, MM_usedindex)) + goto handlemm; + } + /* Convert int to number before storing. */ + if (!LJ_DUALNUM && tref_isinteger(ix->val)) + ix->val = emitir(IRTN(IR_CONV), ix->val, IRCONV_NUM_INT); + emitir(IRT(loadop+IRDELTA_L2S, tref_type(ix->val)), xref, ix->val); + if (keybarrier || tref_isgcv(ix->val)) + emitir(IRT(IR_TBAR, IRT_NIL), ix->tab, 0); + /* Invalidate neg. metamethod cache for stores with certain string keys. */ + if (!nommstr(J, ix->key)) { + TRef fref = emitir(IRT(IR_FREF, IRT_PGC), ix->tab, IRFL_TAB_NOMM); + emitir(IRT(IR_FSTORE, IRT_U8), fref, lj_ir_kint(J, 0)); + } + J->needsnap = 1; + return 0; + } +} + +/* Determine result type of table traversal. */ +static IRType rec_next_types(GCtab *t, uint32_t idx) +{ + for (; idx < t->asize; idx++) { + cTValue *a = arrayslot(t, idx); + if (LJ_LIKELY(!tvisnil(a))) + return (LJ_DUALNUM ? IRT_INT : IRT_NUM) + (itype2irt(a) << 8); + } + idx -= t->asize; + for (; idx <= t->hmask; idx++) { + Node *n = &noderef(t->node)[idx]; + if (!tvisnil(&n->val)) + return itype2irt(&n->key) + (itype2irt(&n->val) << 8); + } + return IRT_NIL + (IRT_NIL << 8); +} + +/* Record a table traversal step aka next(). */ +int lj_record_next(jit_State *J, RecordIndex *ix) +{ + IRType t, tkey, tval; + TRef trvk; + t = rec_next_types(tabV(&ix->tabv), ix->keyv.u32.lo); + tkey = (t & 0xff); tval = (t >> 8); + trvk = lj_ir_call(J, IRCALL_lj_vm_next, ix->tab, ix->key); + if (ix->mobj || tkey == IRT_NIL) { + TRef idx = emitir(IRTI(IR_HIOP), trvk, trvk); + /* Always check for invalid key from next() for nil result. */ + if (!ix->mobj) emitir(IRTGI(IR_NE), idx, lj_ir_kint(J, -1)); + ix->mobj = idx; + } + ix->key = lj_record_vload(J, trvk, 1, tkey); + if (tkey == IRT_NIL || ix->idxchain) { /* Omit value type check. */ + ix->val = TREF_NIL; + return 1; + } else { /* Need value. */ + ix->val = lj_record_vload(J, trvk, 0, tval); + return 2; + } +} + +static void rec_tsetm(jit_State *J, BCReg ra, BCReg rn, int32_t i) +{ + RecordIndex ix; + cTValue *basev = J->L->base; + GCtab *t = tabV(&basev[ra-1]); + settabV(J->L, &ix.tabv, t); + ix.tab = getslot(J, ra-1); + ix.idxchain = 0; +#ifdef LUAJIT_ENABLE_TABLE_BUMP + if ((J->flags & JIT_F_OPT_SINK)) { + if (t->asize < i+rn-ra) + lj_tab_reasize(J->L, t, i+rn-ra); + setnilV(&ix.keyv); + rec_idx_bump(J, &ix); + } +#endif + for (; ra < rn; i++, ra++) { + setintV(&ix.keyv, i); + ix.key = lj_ir_kint(J, i); + copyTV(J->L, &ix.valv, &basev[ra]); + ix.val = getslot(J, ra); + lj_record_idx(J, &ix); + } +} + +/* -- Upvalue access ------------------------------------------------------ */ + +/* Check whether upvalue is immutable and ok to constify. */ +static int rec_upvalue_constify(jit_State *J, GCupval *uvp) +{ + if (uvp->immutable) { + cTValue *o = uvval(uvp); + /* Don't constify objects that may retain large amounts of memory. */ +#if LJ_HASFFI + if (tviscdata(o)) { + GCcdata *cd = cdataV(o); + if (!cdataisv(cd) && !(cd->marked & LJ_GC_CDATA_FIN)) { + CType *ct = ctype_raw(ctype_ctsG(J2G(J)), cd->ctypeid); + if (!ctype_hassize(ct->info) || ct->size <= 16) + return 1; + } + return 0; + } +#else + UNUSED(J); +#endif + if (!(tvistab(o) || tvisudata(o) || tvisthread(o))) + return 1; + } + return 0; +} + +/* Record upvalue load/store. */ +static TRef rec_upvalue(jit_State *J, uint32_t uv, TRef val) +{ + GCupval *uvp = &gcref(J->fn->l.uvptr[uv])->uv; + TRef fn = getcurrf(J); + IRRef uref; + int needbarrier = 0; + if (rec_upvalue_constify(J, uvp)) { /* Try to constify immutable upvalue. */ + TRef tr, kfunc; + lj_assertJ(val == 0, "bad usage"); + if (!tref_isk(fn)) { /* Late specialization of current function. */ + if (J->pt->flags >= PROTO_CLC_POLY) + goto noconstify; + kfunc = lj_ir_kfunc(J, J->fn); + emitir(IRTG(IR_EQ, IRT_FUNC), fn, kfunc); +#if LJ_FR2 + J->base[-2] = kfunc; +#else + J->base[-1] = kfunc | TREF_FRAME; +#endif + fn = kfunc; + } + tr = lj_record_constify(J, uvval(uvp)); + if (tr) + return tr; + } +noconstify: + /* Note: this effectively limits LJ_MAX_UPVAL to 127. */ + uv = (uv << 8) | (hashrot(uvp->dhash, uvp->dhash + HASH_BIAS) & 0xff); + if (!uvp->closed) { + /* In current stack? */ + if (uvval(uvp) >= tvref(J->L->stack) && + uvval(uvp) < tvref(J->L->maxstack)) { + int32_t slot = (int32_t)(uvval(uvp) - (J->L->base - J->baseslot)); + if (slot >= 0) { /* Aliases an SSA slot? */ + uref = tref_ref(emitir(IRT(IR_UREFO, IRT_PGC), fn, uv)); + emitir(IRTG(IR_EQ, IRT_PGC), + REF_BASE, + emitir(IRT(IR_ADD, IRT_PGC), uref, + lj_ir_kintpgc(J, (slot - 1 - LJ_FR2) * -8))); + slot -= (int32_t)J->baseslot; /* Note: slot number may be negative! */ + if (val == 0) { + return getslot(J, slot); + } else { + J->base[slot] = val; + if (slot >= (int32_t)J->maxslot) J->maxslot = (BCReg)(slot+1); + return 0; + } + } + } + /* IR_UREFO+IRT_IGC is not checked for open-ness at runtime. + ** Always marked as a guard, since it might get promoted to IRT_PGC later. + */ + uref = emitir(IRTG(IR_UREFO, tref_isgcv(val) ? IRT_PGC : IRT_IGC), fn, uv); + uref = tref_ref(uref); + emitir(IRTG(IR_UGT, IRT_PGC), + emitir(IRT(IR_SUB, IRT_PGC), uref, REF_BASE), + lj_ir_kintpgc(J, (J->baseslot + J->maxslot) * 8)); + } else { + /* If fn is constant, then so is the GCupval*, and the upvalue cannot + ** transition back to open, so no guard is required in this case. + */ + IRType t = (tref_isk(fn) ? 0 : IRT_GUARD) | IRT_PGC; + uref = tref_ref(emitir(IRT(IR_UREFC, t), fn, uv)); + needbarrier = 1; + } + if (val == 0) { /* Upvalue load */ + IRType t = itype2irt(uvval(uvp)); + TRef res = emitir(IRTG(IR_ULOAD, t), uref, 0); + if (irtype_ispri(t)) res = TREF_PRI(t); /* Canonicalize primitive refs. */ + return res; + } else { /* Upvalue store. */ + /* Convert int to number before storing. */ + if (!LJ_DUALNUM && tref_isinteger(val)) + val = emitir(IRTN(IR_CONV), val, IRCONV_NUM_INT); + emitir(IRT(IR_USTORE, tref_type(val)), uref, val); + if (needbarrier && tref_isgcv(val)) + emitir(IRT(IR_OBAR, IRT_NIL), uref, val); + J->needsnap = 1; + return 0; + } +} + +/* -- Record calls to Lua functions --------------------------------------- */ + +/* Check unroll limits for calls. */ +static void check_call_unroll(jit_State *J, TraceNo lnk) +{ + cTValue *frame = J->L->base - 1; + void *pc = mref(frame_func(frame)->l.pc, void); + int32_t depth = J->framedepth; + int32_t count = 0; + if ((J->pt->flags & PROTO_VARARG)) depth--; /* Vararg frame still missing. */ + for (; depth > 0; depth--) { /* Count frames with same prototype. */ + if (frame_iscont(frame)) depth--; + frame = frame_prev(frame); + if (mref(frame_func(frame)->l.pc, void) == pc) + count++; + } + if (J->pc == J->startpc) { + if (count + J->tailcalled > J->param[JIT_P_recunroll]) { + J->pc++; + if (J->framedepth + J->retdepth == 0) + lj_record_stop(J, LJ_TRLINK_TAILREC, J->cur.traceno); /* Tail-rec. */ + else + lj_record_stop(J, LJ_TRLINK_UPREC, J->cur.traceno); /* Up-recursion. */ + } + } else { + if (count > J->param[JIT_P_callunroll]) { + if (lnk) { /* Possible tail- or up-recursion. */ + lj_trace_flush(J, lnk); /* Flush trace that only returns. */ + /* Set a small, pseudo-random hotcount for a quick retry of JFUNC*. */ + hotcount_set(J2GG(J), J->pc+1, lj_prng_u64(&J2G(J)->prng) & 15u); + } + lj_trace_err(J, LJ_TRERR_CUNROLL); + } + } +} + +/* Record Lua function setup. */ +static void rec_func_setup(jit_State *J) +{ + GCproto *pt = J->pt; + BCReg s, numparams = pt->numparams; + if ((pt->flags & PROTO_NOJIT)) + lj_trace_err(J, LJ_TRERR_CJITOFF); + if (J->baseslot + pt->framesize >= LJ_MAX_JSLOTS) + lj_trace_err(J, LJ_TRERR_STACKOV); + /* Fill up missing parameters with nil. */ + for (s = J->maxslot; s < numparams; s++) + J->base[s] = TREF_NIL; + /* The remaining slots should never be read before they are written. */ + J->maxslot = numparams; +} + +/* Record Lua vararg function setup. */ +static void rec_func_vararg(jit_State *J) +{ + GCproto *pt = J->pt; + BCReg s, fixargs, vframe = J->maxslot+1+LJ_FR2; + lj_assertJ((pt->flags & PROTO_VARARG), "FUNCV in non-vararg function"); + if (J->baseslot + vframe + pt->framesize >= LJ_MAX_JSLOTS) + lj_trace_err(J, LJ_TRERR_STACKOV); + J->base[vframe-1-LJ_FR2] = J->base[-1-LJ_FR2]; /* Copy function up. */ +#if LJ_FR2 + J->base[vframe-1] = TREF_FRAME; +#endif + /* Copy fixarg slots up and set their original slots to nil. */ + fixargs = pt->numparams < J->maxslot ? pt->numparams : J->maxslot; + for (s = 0; s < fixargs; s++) { + J->base[vframe+s] = J->base[s]; + J->base[s] = TREF_NIL; + } + J->maxslot = fixargs; + J->framedepth++; + J->base += vframe; + J->baseslot += vframe; +} + +/* Record entry to a Lua function. */ +static void rec_func_lua(jit_State *J) +{ + rec_func_setup(J); + check_call_unroll(J, 0); +} + +/* Record entry to an already compiled function. */ +static void rec_func_jit(jit_State *J, TraceNo lnk) +{ + GCtrace *T; + rec_func_setup(J); + T = traceref(J, lnk); + if (T->linktype == LJ_TRLINK_RETURN) { /* Trace returns to interpreter? */ + check_call_unroll(J, lnk); + /* Temporarily unpatch JFUNC* to continue recording across function. */ + J->patchins = *J->pc; + J->patchpc = (BCIns *)J->pc; + *J->patchpc = T->startins; + return; + } + J->instunroll = 0; /* Cannot continue across a compiled function. */ + if (J->pc == J->startpc && J->framedepth + J->retdepth == 0) + lj_record_stop(J, LJ_TRLINK_TAILREC, J->cur.traceno); /* Extra tail-rec. */ + else + lj_record_stop(J, LJ_TRLINK_ROOT, lnk); /* Link to the function. */ +} + +/* -- Vararg handling ----------------------------------------------------- */ + +/* Detect y = select(x, ...) idiom. */ +static int select_detect(jit_State *J) +{ + BCIns ins = J->pc[1]; + if (bc_op(ins) == BC_CALLM && bc_b(ins) == 2 && bc_c(ins) == 1) { + cTValue *func = &J->L->base[bc_a(ins)]; + if (tvisfunc(func) && funcV(func)->c.ffid == FF_select) { + TRef kfunc = lj_ir_kfunc(J, funcV(func)); + emitir(IRTG(IR_EQ, IRT_FUNC), getslot(J, bc_a(ins)), kfunc); + return 1; + } + } + return 0; +} + +/* Record vararg instruction. */ +static void rec_varg(jit_State *J, BCReg dst, ptrdiff_t nresults) +{ + int32_t numparams = J->pt->numparams; + ptrdiff_t nvararg = frame_delta(J->L->base-1) - numparams - 1 - LJ_FR2; + lj_assertJ(frame_isvarg(J->L->base-1), "VARG in non-vararg frame"); + if (LJ_FR2 && dst > J->maxslot) + J->base[dst-1] = 0; /* Prevent resurrection of unrelated slot. */ + if (J->framedepth > 0) { /* Simple case: varargs defined on-trace. */ + ptrdiff_t i; + if (nvararg < 0) nvararg = 0; + if (nresults != 1) { + if (nresults == -1) nresults = nvararg; + J->maxslot = dst + (BCReg)nresults; + } else if (dst >= J->maxslot) { + J->maxslot = dst + 1; + } + if (J->baseslot + J->maxslot >= LJ_MAX_JSLOTS) + lj_trace_err(J, LJ_TRERR_STACKOV); + for (i = 0; i < nresults; i++) + J->base[dst+i] = i < nvararg ? getslot(J, i - nvararg - 1 - LJ_FR2) : TREF_NIL; + } else { /* Unknown number of varargs passed to trace. */ + TRef fr = emitir(IRTI(IR_SLOAD), LJ_FR2, IRSLOAD_READONLY|IRSLOAD_FRAME); + int32_t frofs = 8*(1+LJ_FR2+numparams)+FRAME_VARG; + if (nresults >= 0) { /* Known fixed number of results. */ + ptrdiff_t i; + if (nvararg > 0) { + ptrdiff_t nload = nvararg >= nresults ? nresults : nvararg; + TRef vbase; + if (nvararg >= nresults) + emitir(IRTGI(IR_GE), fr, lj_ir_kint(J, frofs+8*(int32_t)nresults)); + else + emitir(IRTGI(IR_EQ), fr, + lj_ir_kint(J, (int32_t)frame_ftsz(J->L->base-1))); + vbase = emitir(IRT(IR_SUB, IRT_IGC), REF_BASE, fr); + vbase = emitir(IRT(IR_ADD, IRT_PGC), vbase, + lj_ir_kintpgc(J, frofs-8*(1+LJ_FR2))); + for (i = 0; i < nload; i++) { + IRType t = itype2irt(&J->L->base[i-1-LJ_FR2-nvararg]); + J->base[dst+i] = lj_record_vload(J, vbase, (MSize)i, t); + } + } else { + emitir(IRTGI(IR_LE), fr, lj_ir_kint(J, frofs)); + nvararg = 0; + } + for (i = nvararg; i < nresults; i++) + J->base[dst+i] = TREF_NIL; + if (nresults != 1 || dst >= J->maxslot) { + J->maxslot = dst + (BCReg)nresults; + } + } else if (select_detect(J)) { /* y = select(x, ...) */ + TRef tridx = J->base[dst-1]; + TRef tr = TREF_NIL; + ptrdiff_t idx = lj_ffrecord_select_mode(J, tridx, &J->L->base[dst-1]); + if (idx < 0) goto nyivarg; + if (idx != 0 && !tref_isinteger(tridx)) { + if (tref_isstr(tridx)) + tridx = emitir(IRTG(IR_STRTO, IRT_NUM), tridx, 0); + tridx = emitir(IRTGI(IR_CONV), tridx, IRCONV_INT_NUM|IRCONV_INDEX); + } + if (idx != 0 && tref_isk(tridx)) { + emitir(IRTGI(idx <= nvararg ? IR_GE : IR_LT), + fr, lj_ir_kint(J, frofs+8*(int32_t)idx)); + frofs -= 8; /* Bias for 1-based index. */ + } else if (idx <= nvararg) { /* Compute size. */ + TRef tmp = emitir(IRTI(IR_ADD), fr, lj_ir_kint(J, -frofs)); + if (numparams) + emitir(IRTGI(IR_GE), tmp, lj_ir_kint(J, 0)); + tr = emitir(IRTI(IR_BSHR), tmp, lj_ir_kint(J, 3)); + if (idx != 0) { + tridx = emitir(IRTI(IR_ADD), tridx, lj_ir_kint(J, -1)); + rec_idx_abc(J, tr, tridx, (uint32_t)nvararg); + } + } else { + TRef tmp = lj_ir_kint(J, frofs); + if (idx != 0) { + TRef tmp2 = emitir(IRTI(IR_BSHL), tridx, lj_ir_kint(J, 3)); + tmp = emitir(IRTI(IR_ADD), tmp2, tmp); + } else { + tr = lj_ir_kint(J, 0); + } + emitir(IRTGI(IR_LT), fr, tmp); + } + if (idx != 0 && idx <= nvararg) { + IRType t; + TRef aref, vbase = emitir(IRT(IR_SUB, IRT_IGC), REF_BASE, fr); + vbase = emitir(IRT(IR_ADD, IRT_PGC), vbase, + lj_ir_kintpgc(J, frofs-(8<L->base[idx-2-LJ_FR2-nvararg]); + aref = emitir(IRT(IR_AREF, IRT_PGC), vbase, tridx); + tr = lj_record_vload(J, aref, 0, t); + } + J->base[dst-2-LJ_FR2] = tr; + J->maxslot = dst-1-LJ_FR2; + J->bcskip = 2; /* Skip CALLM + select. */ + } else { + nyivarg: + setintV(&J->errinfo, BC_VARG); + lj_trace_err_info(J, LJ_TRERR_NYIBC); + } + } +} + +/* -- Record allocations -------------------------------------------------- */ + +static TRef rec_tnew(jit_State *J, uint32_t ah) +{ + uint32_t asize = ah & 0x7ff; + uint32_t hbits = ah >> 11; + TRef tr; + if (asize == 0x7ff) asize = 0x801; + tr = emitir(IRTG(IR_TNEW, IRT_TAB), asize, hbits); +#ifdef LUAJIT_ENABLE_TABLE_BUMP + J->rbchash[(tr & (RBCHASH_SLOTS-1))].ref = tref_ref(tr); + setmref(J->rbchash[(tr & (RBCHASH_SLOTS-1))].pc, J->pc); + setgcref(J->rbchash[(tr & (RBCHASH_SLOTS-1))].pt, obj2gco(J->pt)); +#endif + return tr; +} + +/* -- Concatenation ------------------------------------------------------- */ + +typedef struct RecCatDataCP { + jit_State *J; + BCReg baseslot, topslot; + TRef tr; +} RecCatDataCP; + +static TValue *rec_mm_concat_cp(lua_State *L, lua_CFunction dummy, void *ud) +{ + RecCatDataCP *rcd = (RecCatDataCP *)ud; + jit_State *J = rcd->J; + BCReg baseslot = rcd->baseslot, topslot = rcd->topslot; + TRef *top = &J->base[topslot]; + BCReg s; + RecordIndex ix; + UNUSED(L); UNUSED(dummy); + lj_assertJ(baseslot < topslot, "bad CAT arg"); + for (s = baseslot; s <= topslot; s++) + (void)getslot(J, s); /* Ensure all arguments have a reference. */ + if (tref_isnumber_str(top[0]) && tref_isnumber_str(top[-1])) { + TRef tr, hdr, *trp, *xbase, *base = &J->base[baseslot]; + /* First convert numbers to strings. */ + for (trp = top; trp >= base; trp--) { + if (tref_isnumber(*trp)) + *trp = emitir(IRT(IR_TOSTR, IRT_STR), *trp, + tref_isnum(*trp) ? IRTOSTR_NUM : IRTOSTR_INT); + else if (!tref_isstr(*trp)) + break; + } + xbase = ++trp; + tr = hdr = emitir(IRT(IR_BUFHDR, IRT_PGC), + lj_ir_kptr(J, &J2G(J)->tmpbuf), IRBUFHDR_RESET); + do { + tr = emitir(IRTG(IR_BUFPUT, IRT_PGC), tr, *trp++); + } while (trp <= top); + tr = emitir(IRTG(IR_BUFSTR, IRT_STR), tr, hdr); + J->maxslot = (BCReg)(xbase - J->base); + if (xbase == base) { + rcd->tr = tr; /* Return simple concatenation result. */ + return NULL; + } + /* Pass partial result. */ + topslot = J->maxslot--; + *xbase = tr; + top = xbase; + setstrV(J->L, &ix.keyv, &J2G(J)->strempty); /* Simulate string result. */ + } else { + J->maxslot = topslot-1; + copyTV(J->L, &ix.keyv, &J->L->base[topslot]); + } + copyTV(J->L, &ix.tabv, &J->L->base[topslot-1]); + ix.tab = top[-1]; + ix.key = top[0]; + rec_mm_arith(J, &ix, MM_concat); /* Call __concat metamethod. */ + rcd->tr = 0; /* No result yet. */ + return NULL; +} + +static TRef rec_cat(jit_State *J, BCReg baseslot, BCReg topslot) +{ + lua_State *L = J->L; + ptrdiff_t delta = L->top - L->base; + TValue savetv[5+LJ_FR2], errobj; + RecCatDataCP rcd; + int errcode; + rcd.J = J; + rcd.baseslot = baseslot; + rcd.topslot = topslot; + memcpy(savetv, &L->base[topslot-1], sizeof(savetv)); /* Save slots. */ + errcode = lj_vm_cpcall(L, NULL, &rcd, rec_mm_concat_cp); + if (errcode) copyTV(L, &errobj, L->top-1); + memcpy(&L->base[topslot-1], savetv, sizeof(savetv)); /* Restore slots. */ + if (errcode) { + L->top = L->base + delta; + copyTV(L, L->top++, &errobj); + return (TRef)(-errcode); + } + return rcd.tr; +} + +/* -- Record bytecode ops ------------------------------------------------- */ + +/* Prepare for comparison. */ +static void rec_comp_prep(jit_State *J) +{ + /* Prevent merging with snapshot #0 (GC exit) since we fixup the PC. */ + if (J->cur.nsnap == 1 && J->cur.snap[0].ref == J->cur.nins) + emitir_raw(IRT(IR_NOP, IRT_NIL), 0, 0); + lj_snap_add(J); +} + +/* Fixup comparison. */ +static void rec_comp_fixup(jit_State *J, const BCIns *pc, int cond) +{ + BCIns jmpins = pc[1]; + const BCIns *npc = pc + 2 + (cond ? bc_j(jmpins) : 0); + SnapShot *snap = &J->cur.snap[J->cur.nsnap-1]; + /* Set PC to opposite target to avoid re-recording the comp. in side trace. */ +#if LJ_FR2 + SnapEntry *flink = &J->cur.snapmap[snap->mapofs + snap->nent]; + uint64_t pcbase; + memcpy(&pcbase, flink, sizeof(uint64_t)); + pcbase = (pcbase & 0xff) | (u64ptr(npc) << 8); + memcpy(flink, &pcbase, sizeof(uint64_t)); +#else + J->cur.snapmap[snap->mapofs + snap->nent] = SNAP_MKPC(npc); +#endif + J->needsnap = 1; + if (bc_a(jmpins) < J->maxslot) J->maxslot = bc_a(jmpins); + lj_snap_shrink(J); /* Shrink last snapshot if possible. */ +} + +/* Check for falsy zero. */ +static void rec_falsy_zero(jit_State *J, TRef tr, cTValue *tv, int op) +{ + if (tref_isnumber(tr)) { + rec_comp_prep(J); + IROp cond = tvistruecond(tv) ? IR_NE : IR_EQ; + emitir(IRTG(cond, tref_isinteger(tr) ? IRT_INT : IRT_NUM), + tr, tref_isinteger(tr) ? lj_ir_kint(J, 0) : lj_ir_knum_zero(J)); +#if 0 /* this does not work no matter what I try */ + if (op == BC_IST || op == BC_ISF) + rec_comp_fixup(J, J->pc, cond & 1); + else +#endif + J->needsnap = 1; + } +} + +/* Record the next bytecode instruction (_before_ it's executed). */ +void lj_record_ins(jit_State *J) +{ + cTValue *lbase; + RecordIndex ix; + const BCIns *pc; + BCIns ins; + BCOp op; + TRef ra, rb, rc; + + /* Perform post-processing action before recording the next instruction. */ + if (LJ_UNLIKELY(J->postproc != LJ_POST_NONE)) { + switch (J->postproc) { + case LJ_POST_FIXCOMP: /* Fixup comparison. */ + pc = (const BCIns *)(uintptr_t)J2G(J)->tmptv.u64; + rec_comp_fixup(J, pc, (!tvistruecond(&J2G(J)->tmptv2) ^ (bc_op(*pc)&1))); + /* fallthrough */ + case LJ_POST_FIXGUARD: /* Fixup and emit pending guard. */ + case LJ_POST_FIXGUARDSNAP: /* Fixup and emit pending guard and snapshot. */ + if (!tvistruecond(&J2G(J)->tmptv2)) { + J->fold.ins.o ^= 1; /* Flip guard to opposite. */ + if (J->postproc == LJ_POST_FIXGUARDSNAP) { + SnapShot *snap = &J->cur.snap[J->cur.nsnap-1]; + J->cur.snapmap[snap->mapofs+snap->nent-1]--; /* False -> true. */ + } + } + lj_opt_fold(J); /* Emit pending guard. */ + /* fallthrough */ + case LJ_POST_FIXBOOL: + if (!tvistruecond(&J2G(J)->tmptv2)) { + BCReg s; + TValue *tv = J->L->base; + for (s = 0; s < J->maxslot; s++) /* Fixup stack slot (if any). */ + if (J->base[s] == TREF_TRUE && tvisfalse(&tv[s])) { + J->base[s] = TREF_FALSE; + break; + } + } + break; + case LJ_POST_FIXCONST: + { + BCReg s; + TValue *tv = J->L->base; + for (s = 0; s < J->maxslot; s++) /* Constify stack slots (if any). */ + if (J->base[s] == TREF_NIL && !tvisnil(&tv[s])) + J->base[s] = lj_record_constify(J, &tv[s]); + } + break; + case LJ_POST_FFRETRY: /* Suppress recording of retried fast function. */ + if (bc_op(*J->pc) >= BC__MAX) + return; + break; + default: lj_assertJ(0, "bad post-processing mode"); break; + } + J->postproc = LJ_POST_NONE; + } + + /* Need snapshot before recording next bytecode (e.g. after a store). */ + if (J->needsnap) { + J->needsnap = 0; + if (J->pt) lj_snap_purge(J); + lj_snap_add(J); + J->mergesnap = 1; + } + + /* Skip some bytecodes. */ + if (LJ_UNLIKELY(J->bcskip > 0)) { + J->bcskip--; + return; + } + + /* Record only closed loops for root traces. */ + pc = J->pc; + if (J->framedepth == 0 && + (MSize)((char *)pc - (char *)J->bc_min) >= J->bc_extent) + lj_trace_err(J, LJ_TRERR_LLEAVE); + +#ifdef LUA_USE_ASSERT + rec_check_slots(J); + rec_check_ir(J); +#endif + +#if LJ_HASPROFILE + rec_profile_ins(J, pc); +#endif + + /* Keep a copy of the runtime values of var/num/str operands. */ +#define rav (&ix.valv) +#define rbv (&ix.tabv) +#define rcv (&ix.keyv) + + lbase = J->L->base; + ins = *pc; + op = bc_op(ins); + ra = bc_a(ins); + ix.val = 0; + switch (bcmode_a(op)) { + case BCMvar: + copyTV(J->L, rav, &lbase[ra]); ix.val = ra = getslot(J, ra); break; + default: break; /* Handled later. */ + } + rb = bc_b(ins); + rc = bc_c(ins); + switch (bcmode_b(op)) { + case BCMnone: rb = 0; rc = bc_d(ins); break; /* Upgrade rc to 'rd'. */ + case BCMvar: + copyTV(J->L, rbv, &lbase[rb]); ix.tab = rb = getslot(J, rb); break; + default: break; /* Handled later. */ + } + switch (bcmode_c(op)) { + case BCMvar: + copyTV(J->L, rcv, &lbase[rc]); ix.key = rc = getslot(J, rc); break; + case BCMpri: setpriV(rcv, ~rc); ix.key = rc = TREF_PRI(IRT_NIL+rc); break; + case BCMnum: { cTValue *tv = proto_knumtv(J->pt, rc); + copyTV(J->L, rcv, tv); ix.key = rc = tvisint(tv) ? lj_ir_kint(J, intV(tv)) : + tv->u32.hi == LJ_KEYINDEX ? (lj_ir_kint(J, 0) | TREF_KEYINDEX) : + lj_ir_knumint(J, numV(tv)); } break; + case BCMstr: { GCstr *s = gco2str(proto_kgc(J->pt, ~(ptrdiff_t)rc)); + setstrV(J->L, rcv, s); ix.key = rc = lj_ir_kstr(J, s); } break; + default: break; /* Handled later. */ + } + + switch (op) { + + /* -- Comparison ops ---------------------------------------------------- */ + + case BC_ISLT: case BC_ISGE: case BC_ISLE: case BC_ISGT: +#if LJ_HASFFI + if (tref_iscdata(ra) || tref_iscdata(rc)) { + rec_mm_comp_cdata(J, &ix, op, ((int)op & 2) ? MM_le : MM_lt); + break; + } +#endif + /* Emit nothing for two numeric or string consts. */ + if (!(tref_isk2(ra,rc) && tref_isnumber_str(ra) && tref_isnumber_str(rc))) { + IRType ta = tref_isinteger(ra) ? IRT_INT : tref_type(ra); + IRType tc = tref_isinteger(rc) ? IRT_INT : tref_type(rc); + int irop; + if (ta != tc) { + /* Widen mixed number/int comparisons to number/number comparison. */ + if (ta == IRT_INT && tc == IRT_NUM) { + ra = emitir(IRTN(IR_CONV), ra, IRCONV_NUM_INT); + ta = IRT_NUM; + } else if (ta == IRT_NUM && tc == IRT_INT) { + rc = emitir(IRTN(IR_CONV), rc, IRCONV_NUM_INT); + } else if (LJ_52) { + ta = IRT_NIL; /* Force metamethod for different types. */ + } else if (!((ta == IRT_FALSE || ta == IRT_TRUE) && + (tc == IRT_FALSE || tc == IRT_TRUE))) { + break; /* Interpreter will throw for two different types. */ + } + } + rec_comp_prep(J); + irop = (int)op - (int)BC_ISLT + (int)IR_LT; + if (ta == IRT_NUM) { + if ((irop & 1)) irop ^= 4; /* ISGE/ISGT are unordered. */ + if (!lj_ir_numcmp(numberVnum(rav), numberVnum(rcv), (IROp)irop)) + irop ^= 5; + } else if (ta == IRT_INT) { + if (!lj_ir_numcmp(numberVnum(rav), numberVnum(rcv), (IROp)irop)) + irop ^= 1; + } else if (ta == IRT_STR) { + if (!lj_ir_strcmp(strV(rav), strV(rcv), (IROp)irop)) irop ^= 1; + ra = lj_ir_call(J, IRCALL_lj_str_cmp, ra, rc); + rc = lj_ir_kint(J, 0); + ta = IRT_INT; + } else { + rec_mm_comp(J, &ix, (int)op); + break; + } + emitir(IRTG(irop, ta), ra, rc); + rec_comp_fixup(J, J->pc, ((int)op ^ irop) & 1); + } + break; + + case BC_ISEQV: case BC_ISNEV: + case BC_ISEQS: case BC_ISNES: + case BC_ISEQN: case BC_ISNEN: + case BC_ISEQP: case BC_ISNEP: +#if LJ_HASFFI + if (tref_iscdata(ra) || tref_iscdata(rc)) { + rec_mm_comp_cdata(J, &ix, op, MM_eq); + break; + } +#endif + /* Emit nothing for two non-table, non-udata consts. */ + if (!(tref_isk2(ra, rc) && !(tref_istab(ra) || tref_isudata(ra)))) { + int diff; + rec_comp_prep(J); + diff = lj_record_objcmp(J, ra, rc, rav, rcv); + if (diff == 2 || !(tref_istab(ra) || tref_isudata(ra))) + rec_comp_fixup(J, J->pc, ((int)op & 1) == !diff); + else if (diff == 1) /* Only check __eq if different, but same type. */ + rec_mm_equal(J, &ix, (int)op); + } + break; + + /* -- Unary test and copy ops ------------------------------------------- */ + + case BC_ISTC: case BC_ISFC: + if ((op & 1) == tref_istruecond(rc) && !tref_isnumber(rc)) + rc = 0; /* Don't store if condition is not true. */ + /* fallthrough */ + case BC_IST: case BC_ISF: /* Type specialization suffices. */ + rec_falsy_zero(J, rc, rcv, op); + if (bc_a(pc[1]) < J->maxslot) + J->maxslot = bc_a(pc[1]); /* Shrink used slots. */ + break; + + case BC_ISTYPE: case BC_ISNUM: + /* These coercions need to correspond with lj_meta_istype(). */ + if (LJ_DUALNUM && rc == ~LJ_TNUMX+1) + ra = lj_opt_narrow_toint(J, ra); + else if (rc == ~LJ_TNUMX+2) + ra = lj_ir_tonum(J, ra); + else if (rc == ~LJ_TSTR+1) + ra = lj_ir_tostr(J, ra); + /* else: type specialization suffices. */ + J->base[bc_a(ins)] = ra; + break; + + /* -- Unary ops --------------------------------------------------------- */ + + case BC_NOT: + rec_falsy_zero(J, rc, rcv, op); + /* Type specialization already forces const result. */ + rc = tvistruecond(rcv) ? TREF_FALSE : TREF_TRUE; + break; + + case BC_LEN: + if (tref_isstr(rc)) + rc = emitir(IRTI(IR_FLOAD), rc, IRFL_STR_LEN); + else if (!LJ_52 && tref_istab(rc)) + rc = emitir(IRTI(IR_ALEN), rc, TREF_NIL); + else + rc = rec_mm_len(J, rc, rcv); + break; + + /* -- Arithmetic ops ---------------------------------------------------- */ + + case BC_UNM: + if (tref_isnumber_str(rc)) { + rc = lj_opt_narrow_unm(J, rc, rcv); + } else { + ix.tab = rc; + copyTV(J->L, &ix.tabv, rcv); + rc = rec_mm_arith(J, &ix, MM_unm); + } + break; + + case BC_ADDNV: case BC_SUBNV: case BC_MULNV: case BC_DIVNV: case BC_MODNV: + /* Swap rb/rc and rbv/rcv. rav is temp. */ + ix.tab = rc; ix.key = rc = rb; rb = ix.tab; + copyTV(J->L, rav, rbv); + copyTV(J->L, rbv, rcv); + copyTV(J->L, rcv, rav); + if (op == BC_MODNV) + goto recmod; + /* fallthrough */ + case BC_ADDVN: case BC_SUBVN: case BC_MULVN: case BC_DIVVN: + case BC_ADDVV: case BC_SUBVV: case BC_MULVV: case BC_DIVVV: { + MMS mm = bcmode_mm(op); + if (tref_isnumber_str(rb) && tref_isnumber_str(rc)) + rc = lj_opt_narrow_arith(J, rb, rc, rbv, rcv, + (int)mm - (int)MM_add + (int)IR_ADD); + else + rc = rec_mm_arith(J, &ix, mm); + break; + } + + case BC_MODVN: case BC_MODVV: + recmod: + if (tref_isnumber_str(rb) && tref_isnumber_str(rc)) + rc = lj_opt_narrow_mod(J, rb, rc, rbv, rcv); + else + rc = rec_mm_arith(J, &ix, MM_mod); + break; + + case BC_POW: + if (tref_isnumber_str(rb) && tref_isnumber_str(rc)) + rc = lj_opt_narrow_arith(J, rb, rc, rbv, rcv, IR_POW); + else + rc = rec_mm_arith(J, &ix, MM_pow); + break; + + /* -- Bitwise ops ---------------------------------------------------- */ + + case BC_BNOT: + if (tref_isnumber_str(rc)) { + rc = lj_opt_narrow_bnot(J, rc, rcv); + } else { + ix.tab = rc; + copyTV(J->L, &ix.tabv, rcv); + rc = rec_mm_arith(J, &ix, MM_not); + } + break; + + case BC_BANDNV: case BC_BORNV: case BC_BXORNV: case BC_SHLNV: case BC_SHRNV: + /* Swap rb/rc and rbv/rcv. rav is temp. */ + ix.tab = rc; ix.key = rc = rb; rb = ix.tab; + copyTV(J->L, rav, rbv); + copyTV(J->L, rbv, rcv); + copyTV(J->L, rcv, rav); + /* fallthrough */ + case BC_BANDVN: case BC_BORVN: case BC_BXORVN: case BC_SHLVN: case BC_SHRVN: + case BC_BANDVV: case BC_BORVV: case BC_BXORVV: case BC_SHLVV: case BC_SHRVV: { + MMS mm = bcmode_mm(op); + if (tref_isnumber_str(rb) && tref_isnumber_str(rc)) + rc = lj_opt_narrow_bitwise(J, rb, rc, rbv, rcv, + (int)mm - (int)MM_and + (int)IR_BAND); + else + rc = rec_mm_arith(J, &ix, mm); + break; + } + + /* -- Miscellaneous ops ------------------------------------------------- */ + + case BC_CAT: + rc = rec_cat(J, rb, rc); + if (rc >= 0xffffff00) + lj_err_throw(J->L, -(int32_t)rc); /* Propagate errors. */ + break; + + /* -- Constant and move ops --------------------------------------------- */ + + case BC_MOV: + /* Clear gap of method call to avoid resurrecting previous refs. */ + if (ra > J->maxslot) { +#if LJ_FR2 + memset(J->base + J->maxslot, 0, (ra - J->maxslot) * sizeof(TRef)); +#else + J->base[ra-1] = 0; +#endif + } + break; + case BC_KSTR: case BC_KNUM: case BC_KPRI: + break; + case BC_KSHORT: + rc = lj_ir_kint(J, (int32_t)(int16_t)rc); + break; + case BC_KNIL: + if (LJ_FR2 && ra > J->maxslot) + J->base[ra-1] = 0; + while (ra <= rc) + J->base[ra++] = TREF_NIL; + if (rc >= J->maxslot) J->maxslot = rc+1; + break; +#if LJ_HASFFI + case BC_KCDATA: + rc = lj_ir_kgc(J, proto_kgc(J->pt, ~(ptrdiff_t)rc), IRT_CDATA); + break; +#endif + + /* -- Upvalue and function ops ------------------------------------------ */ + + case BC_UGET: + rc = rec_upvalue(J, rc, 0); + break; + case BC_USETV: case BC_USETS: case BC_USETN: case BC_USETP: + rec_upvalue(J, ra, rc); + break; + + /* -- Table ops --------------------------------------------------------- */ + + case BC_GGET: case BC_GSET: + settabV(J->L, &ix.tabv, tabref(J->fn->l.env)); + ix.tab = emitir(IRT(IR_FLOAD, IRT_TAB), getcurrf(J), IRFL_FUNC_ENV); + ix.idxchain = LJ_MAX_IDXCHAIN; + rc = lj_record_idx(J, &ix); + break; + + case BC_TGETB: case BC_TSETB: + setintV(&ix.keyv, (int32_t)rc); + ix.key = lj_ir_kint(J, (int32_t)rc); + /* fallthrough */ + case BC_TGETV: case BC_TGETS: case BC_TSETV: case BC_TSETS: + ix.idxchain = LJ_MAX_IDXCHAIN; + rc = lj_record_idx(J, &ix); + break; + case BC_TGETR: case BC_TSETR: + ix.idxchain = 0; + rc = lj_record_idx(J, &ix); + break; + + case BC_TSETM: + rec_tsetm(J, ra, (BCReg)(J->L->top - J->L->base), (int32_t)rcv->u32.lo); + J->maxslot = ra; /* The table slot at ra-1 is the highest used slot. */ + break; + + case BC_TNEW: + rc = rec_tnew(J, rc); + break; + case BC_TDUP: + rc = emitir(IRTG(IR_TDUP, IRT_TAB), + lj_ir_ktab(J, gco2tab(proto_kgc(J->pt, ~(ptrdiff_t)rc))), 0); +#ifdef LUAJIT_ENABLE_TABLE_BUMP + J->rbchash[(rc & (RBCHASH_SLOTS-1))].ref = tref_ref(rc); + setmref(J->rbchash[(rc & (RBCHASH_SLOTS-1))].pc, pc); + setgcref(J->rbchash[(rc & (RBCHASH_SLOTS-1))].pt, obj2gco(J->pt)); +#endif + break; + + /* -- Calls and vararg handling ----------------------------------------- */ + + case BC_ITERC: + J->base[ra] = getslot(J, ra-3); + J->base[ra+1+LJ_FR2] = getslot(J, ra-2); + J->base[ra+2+LJ_FR2] = getslot(J, ra-1); + { /* Do the actual copy now because lj_record_call needs the values. */ + TValue *b = &J->L->base[ra]; + copyTV(J->L, b, b-3); + copyTV(J->L, b+1+LJ_FR2, b-2); + copyTV(J->L, b+2+LJ_FR2, b-1); + } + lj_record_call(J, ra, (ptrdiff_t)rc-1); + break; + + /* L->top is set to L->base+ra+rc+NARGS-1+1. See lj_dispatch_ins(). */ + case BC_CALLM: + rc = (BCReg)(J->L->top - J->L->base) - ra - LJ_FR2; + /* fallthrough */ + case BC_CALL: + lj_record_call(J, ra, (ptrdiff_t)rc-1); + break; + + case BC_CALLMT: + rc = (BCReg)(J->L->top - J->L->base) - ra - LJ_FR2; + /* fallthrough */ + case BC_CALLT: + lj_record_tailcall(J, ra, (ptrdiff_t)rc-1); + break; + + case BC_VARG: + rec_varg(J, ra, (ptrdiff_t)rb-1); + break; + + /* -- Returns ----------------------------------------------------------- */ + + case BC_RETM: + /* L->top is set to L->base+ra+rc+NRESULTS-1, see lj_dispatch_ins(). */ + rc = (BCReg)(J->L->top - J->L->base) - ra + 1; + /* fallthrough */ + case BC_RET: case BC_RET0: case BC_RET1: +#if LJ_HASPROFILE + rec_profile_ret(J); +#endif + lj_record_ret(J, ra, (ptrdiff_t)rc-1); + break; + + /* -- Loops and branches ------------------------------------------------ */ + + case BC_FORI: + if (rec_for(J, pc, 0) != LOOPEV_LEAVE) + J->loopref = J->cur.nins; + break; + case BC_JFORI: + lj_assertJ(bc_op(pc[(ptrdiff_t)rc-BCBIAS_J]) == BC_JFORL, + "JFORI does not point to JFORL"); + if (rec_for(J, pc, 0) != LOOPEV_LEAVE) /* Link to existing loop. */ + lj_record_stop(J, LJ_TRLINK_ROOT, bc_d(pc[(ptrdiff_t)rc-BCBIAS_J])); + /* Continue tracing if the loop is not entered. */ + break; + + case BC_FORL: + rec_loop_interp(J, pc, rec_for(J, pc+((ptrdiff_t)rc-BCBIAS_J), 1)); + break; + case BC_ITERL: + rec_loop_interp(J, pc, rec_iterl(J, *pc)); + break; + case BC_ITERN: + rec_loop_interp(J, pc, rec_itern(J, ra, rb)); + break; + case BC_LOOP: + rec_loop_interp(J, pc, rec_loop(J, ra, 1)); + break; + + case BC_JFORL: + rec_loop_jit(J, rc, rec_for(J, pc+bc_j(traceref(J, rc)->startins), 1)); + break; + case BC_JITERL: + rec_loop_jit(J, rc, rec_iterl(J, traceref(J, rc)->startins)); + break; + case BC_JLOOP: + rec_loop_jit(J, rc, rec_loop(J, ra, + !bc_isret(bc_op(traceref(J, rc)->startins)) && + bc_op(traceref(J, rc)->startins) != BC_ITERN)); + break; + + case BC_IFORL: + case BC_IITERL: + case BC_ILOOP: + case BC_IFUNCF: + case BC_IFUNCV: + lj_trace_err(J, LJ_TRERR_BLACKL); + break; + + case BC_JMP: + if (ra < J->maxslot) + J->maxslot = ra; /* Shrink used slots. */ + break; + + case BC_ISNEXT: + rec_isnext(J, ra); + break; + + /* -- Function headers -------------------------------------------------- */ + + case BC_FUNCF: + rec_func_lua(J); + break; + case BC_JFUNCF: + rec_func_jit(J, rc); + break; + + case BC_FUNCV: + rec_func_vararg(J); + rec_func_lua(J); + break; + case BC_JFUNCV: + /* Cannot happen. No hotcall counting for varag funcs. */ + lj_assertJ(0, "unsupported vararg hotcall"); + break; + + case BC_FUNCC: + case BC_FUNCCW: + lj_ffrecord_func(J); + break; + + default: + if (op >= BC__MAX) { + lj_ffrecord_func(J); + break; + } + /* fallthrough */ + case BC_UCLO: + case BC_FNEW: + setintV(&J->errinfo, (int32_t)op); + lj_trace_err_info(J, LJ_TRERR_NYIBC); + break; + } + + /* rc == 0 if we have no result yet, e.g. pending __index metamethod call. */ + if (bcmode_a(op) == BCMdst && rc) { + J->base[ra] = rc; + if (ra >= J->maxslot) { +#if LJ_FR2 + if (ra > J->maxslot) J->base[ra-1] = 0; +#endif + J->maxslot = ra+1; + } + } + +#undef rav +#undef rbv +#undef rcv + + /* Limit the number of recorded IR instructions and constants. */ + if (J->cur.nins > REF_FIRST+(IRRef)J->param[JIT_P_maxrecord] || + J->cur.nk < REF_BIAS-(IRRef)J->param[JIT_P_maxirconst]) + lj_trace_err(J, LJ_TRERR_TRACEOV); +} + +/* -- Recording setup ----------------------------------------------------- */ + +/* Setup recording for a root trace started by a hot loop. */ +static const BCIns *rec_setup_root(jit_State *J) +{ + /* Determine the next PC and the bytecode range for the loop. */ + const BCIns *pcj, *pc = J->pc; + BCIns ins = *pc; + BCReg ra = bc_a(ins); + switch (bc_op(ins)) { + case BC_FORL: + J->bc_extent = (MSize)(-bc_j(ins))*sizeof(BCIns); + pc += 1+bc_j(ins); + J->bc_min = pc; + break; + case BC_ITERL: + if (bc_op(pc[-1]) == BC_JLOOP) + lj_trace_err(J, LJ_TRERR_LINNER); + lj_assertJ(bc_op(pc[-1]) == BC_ITERC, "no ITERC before ITERL"); + J->maxslot = ra + bc_b(pc[-1]) - 1; + J->bc_extent = (MSize)(-bc_j(ins))*sizeof(BCIns); + pc += 1+bc_j(ins); + lj_assertJ(bc_op(pc[-1]) == BC_JMP, "ITERL does not point to JMP+1"); + J->bc_min = pc; + break; + case BC_ITERN: + lj_assertJ(bc_op(pc[1]) == BC_ITERL, "no ITERL after ITERN"); + J->maxslot = ra; + J->bc_extent = (MSize)(-bc_j(pc[1]))*sizeof(BCIns); + J->bc_min = pc+2 + bc_j(pc[1]); + J->state = LJ_TRACE_RECORD_1ST; /* Record the first ITERN, too. */ + break; + case BC_LOOP: + /* Only check BC range for real loops, but not for "repeat until true". */ + pcj = pc + bc_j(ins); + ins = *pcj; + if (bc_op(ins) == BC_JMP && bc_j(ins) < 0) { + J->bc_min = pcj+1 + bc_j(ins); + J->bc_extent = (MSize)(-bc_j(ins))*sizeof(BCIns); + } + J->maxslot = ra; + pc++; + break; + case BC_RET: + case BC_RET0: + case BC_RET1: + /* No bytecode range check for down-recursive root traces. */ + J->maxslot = ra + bc_d(ins) - 1; + break; + case BC_FUNCF: + /* No bytecode range check for root traces started by a hot call. */ + J->maxslot = J->pt->numparams; + pc++; + break; + case BC_CALLM: + case BC_CALL: + case BC_ITERC: + /* No bytecode range check for stitched traces. */ + pc++; + break; + default: + lj_assertJ(0, "bad root trace start bytecode %d", bc_op(ins)); + break; + } + return pc; +} + +/* Setup for recording a new trace. */ +void lj_record_setup(jit_State *J) +{ + uint32_t i; + + /* Initialize state related to current trace. */ + memset(J->slot, 0, sizeof(J->slot)); + memset(J->chain, 0, sizeof(J->chain)); +#ifdef LUAJIT_ENABLE_TABLE_BUMP + memset(J->rbchash, 0, sizeof(J->rbchash)); +#endif + memset(J->bpropcache, 0, sizeof(J->bpropcache)); + J->scev.idx = REF_NIL; + setmref(J->scev.pc, NULL); + + J->baseslot = 1+LJ_FR2; /* Invoking function is at base[-1-LJ_FR2]. */ + J->base = J->slot + J->baseslot; + J->maxslot = 0; + J->framedepth = 0; + J->retdepth = 0; + + J->instunroll = J->param[JIT_P_instunroll]; + J->loopunroll = J->param[JIT_P_loopunroll]; + J->tailcalled = 0; + J->loopref = 0; + + J->bc_min = NULL; /* Means no limit. */ + J->bc_extent = ~(MSize)0; + + /* Emit instructions for fixed references. Also triggers initial IR alloc. */ + emitir_raw(IRT(IR_BASE, IRT_PGC), J->parent, J->exitno); + for (i = 0; i <= 2; i++) { + IRIns *ir = IR(REF_NIL-i); + ir->i = 0; + ir->t.irt = (uint8_t)(IRT_NIL+i); + ir->o = IR_KPRI; + ir->prev = 0; + } + J->cur.nk = REF_TRUE; + + J->startpc = J->pc; + setmref(J->cur.startpc, J->pc); + if (J->parent) { /* Side trace. */ + GCtrace *T = traceref(J, J->parent); + TraceNo root = T->root ? T->root : J->parent; + J->cur.root = (uint16_t)root; + J->cur.startins = BCINS_AD(BC_JMP, 0, 0); + /* Check whether we could at least potentially form an extra loop. */ + if (J->exitno == 0 && T->snap[0].nent == 0) { + /* We can narrow a FORL for some side traces, too. */ + if (J->pc > proto_bc(J->pt) && bc_op(J->pc[-1]) == BC_JFORI && + bc_d(J->pc[bc_j(J->pc[-1])-1]) == root) { + lj_snap_add(J); + rec_for_loop(J, J->pc-1, &J->scev, 1); + goto sidecheck; + } + } else { + J->startpc = NULL; /* Prevent forming an extra loop. */ + } + lj_snap_replay(J, T); + sidecheck: + if ((traceref(J, J->cur.root)->nchild >= J->param[JIT_P_maxside] || + T->snap[J->exitno].count >= J->param[JIT_P_hotexit] + + J->param[JIT_P_tryside])) { + if (bc_op(*J->pc) == BC_JLOOP) { + BCIns startins = traceref(J, bc_d(*J->pc))->startins; + if (bc_op(startins) == BC_ITERN) + rec_itern(J, bc_a(startins), bc_b(startins)); + } + lj_record_stop(J, LJ_TRLINK_INTERP, 0); + } + } else { /* Root trace. */ + J->cur.root = 0; + J->cur.startins = *J->pc; + J->pc = rec_setup_root(J); + /* Note: the loop instruction itself is recorded at the end and not + ** at the start! So snapshot #0 needs to point to the *next* instruction. + ** The one exception is BC_ITERN, which sets LJ_TRACE_RECORD_1ST. + */ + lj_snap_add(J); + if (bc_op(J->cur.startins) == BC_FORL) + rec_for_loop(J, J->pc-1, &J->scev, 1); + else if (bc_op(J->cur.startins) == BC_ITERC) + J->startpc = NULL; + if (1 + J->pt->framesize >= LJ_MAX_JSLOTS) + lj_trace_err(J, LJ_TRERR_STACKOV); + } +#if LJ_HASPROFILE + J->prev_pt = NULL; + J->prev_line = -1; +#endif +#ifdef LUAJIT_ENABLE_CHECKHOOK + /* Regularly check for instruction/line hooks from compiled code and + ** exit to the interpreter if the hooks are set. + ** + ** This is a compile-time option and disabled by default, since the + ** hook checks may be quite expensive in tight loops. + ** + ** Note this is only useful if hooks are *not* set most of the time. + ** Use this only if you want to *asynchronously* interrupt the execution. + ** + ** You can set the instruction hook via lua_sethook() with a count of 1 + ** from a signal handler or another native thread. Please have a look + ** at the first few functions in luajit.c for an example (Ctrl-C handler). + */ + { + TRef tr = emitir(IRT(IR_XLOAD, IRT_U8), + lj_ir_kptr(J, &J2G(J)->hookmask), IRXLOAD_VOLATILE); + tr = emitir(IRTI(IR_BAND), tr, lj_ir_kint(J, (LUA_MASKLINE|LUA_MASKCOUNT))); + emitir(IRTGI(IR_EQ), tr, lj_ir_kint(J, 0)); + } +#endif +} + +#undef IR +#undef emitir_raw +#undef emitir + +#endif diff --git a/src/bluajit/lj_tab.c b/src/bluajit/lj_tab.c new file mode 100644 index 000000000..27b3021dd --- /dev/null +++ b/src/bluajit/lj_tab.c @@ -0,0 +1,692 @@ +/* +** Table handling. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +** +** Major portions taken verbatim or adapted from the Lua interpreter. +** Copyright (C) 1994-2008 Lua.org, PUC-Rio. See Copyright Notice in lua.h +*/ + +#define lj_tab_c +#define LUA_CORE + +#include "lj_obj.h" +#include "lj_gc.h" +#include "lj_err.h" +#include "lj_tab.h" + +/* -- Object hashing ------------------------------------------------------ */ + +/* Hash an arbitrary key and return its anchor position in the hash table. */ +static Node *hashkey(const GCtab *t, cTValue *key) +{ + lj_assertX(!tvisint(key), "attempt to hash integer"); + if (tvisstr(key)) + return hashstr(t, strV(key)); + else if (tvisnum(key)) + return hashnum(t, key); + else if (tvisbool(key)) + return hashmask(t, boolV(key)); + else + return hashgcref(t, key->gcr); + /* Only hash 32 bits of lightuserdata on a 64 bit CPU. Good enough? */ +} + +/* -- Table creation and destruction -------------------------------------- */ + +/* Create new hash part for table. */ +static LJ_AINLINE void newhpart(lua_State *L, GCtab *t, uint32_t hbits) +{ + uint32_t hsize; + Node *node; + lj_assertL(hbits != 0, "zero hash size"); + if (hbits > LJ_MAX_HBITS) + lj_err_msg(L, LJ_ERR_TABOV); + hsize = 1u << hbits; + node = lj_mem_newvec(L, hsize, Node); + setmref(t->node, node); + setfreetop(t, node, &node[hsize]); + t->hmask = hsize-1; +} + +/* +** Q: Why all of these copies of t->hmask, t->node etc. to local variables? +** A: Because alias analysis for C is _really_ tough. +** Even state-of-the-art C compilers won't produce good code without this. +*/ + +/* Clear hash part of table. */ +static LJ_AINLINE void clearhpart(GCtab *t) +{ + uint32_t i, hmask = t->hmask; + Node *node = noderef(t->node); + lj_assertX(t->hmask != 0, "empty hash part"); + for (i = 0; i <= hmask; i++) { + Node *n = &node[i]; + setmref(n->next, NULL); + setnilV(&n->key); + setnilV(&n->val); + } +} + +/* Clear array part of table. */ +static LJ_AINLINE void clearapart(GCtab *t) +{ + uint32_t i, asize = t->asize; + TValue *array = tvref(t->array); + for (i = 0; i < asize; i++) + setnilV(&array[i]); +} + +/* Create a new table. Note: the slots are not initialized (yet). */ +static GCtab *newtab(lua_State *L, uint32_t asize, uint32_t hbits) +{ + GCtab *t; + /* First try to colocate the array part. */ + if (LJ_MAX_COLOSIZE != 0 && asize > 0 && asize <= LJ_MAX_COLOSIZE) { + Node *nilnode; + lj_assertL((sizeof(GCtab) & 7) == 0, "bad GCtab size"); + t = (GCtab *)lj_mem_newgco(L, sizetabcolo(asize)); + t->gct = ~LJ_TTAB; + t->nomm = (uint8_t)~0; + t->colo = (int8_t)asize; + setmref(t->array, (TValue *)((char *)t + sizeof(GCtab))); + setgcrefnull(t->metatable); + t->asize = asize; + t->hmask = 0; + nilnode = &G(L)->nilnode; + setmref(t->node, nilnode); +#if LJ_GC64 + setmref(t->freetop, nilnode); +#endif + } else { /* Otherwise separately allocate the array part. */ + Node *nilnode; + t = lj_mem_newobj(L, GCtab); + t->gct = ~LJ_TTAB; + t->nomm = (uint8_t)~0; + t->colo = 0; + setmref(t->array, NULL); + setgcrefnull(t->metatable); + t->asize = 0; /* In case the array allocation fails. */ + t->hmask = 0; + nilnode = &G(L)->nilnode; + setmref(t->node, nilnode); +#if LJ_GC64 + setmref(t->freetop, nilnode); +#endif + if (asize > 0) { + if (asize > LJ_MAX_ASIZE) + lj_err_msg(L, LJ_ERR_TABOV); + setmref(t->array, lj_mem_newvec(L, asize, TValue)); + t->asize = asize; + } + } + if (hbits) + newhpart(L, t, hbits); + return t; +} + +/* Create a new table. +** +** IMPORTANT NOTE: The API differs from lua_createtable()! +** +** The array size is non-inclusive. E.g. asize=128 creates array slots +** for 0..127, but not for 128. If you need slots 1..128, pass asize=129 +** (slot 0 is wasted in this case). +** +** The hash size is given in hash bits. hbits=0 means no hash part. +** hbits=1 creates 2 hash slots, hbits=2 creates 4 hash slots and so on. +*/ +GCtab *lj_tab_new(lua_State *L, uint32_t asize, uint32_t hbits) +{ + GCtab *t = newtab(L, asize, hbits); + clearapart(t); + if (t->hmask > 0) clearhpart(t); + return t; +} + +/* The API of this function conforms to lua_createtable(). */ +GCtab *lj_tab_new_ah(lua_State *L, int32_t a, int32_t h) +{ + return lj_tab_new(L, (uint32_t)(a > 0 ? a+1 : 0), hsize2hbits(h)); +} + +#if LJ_HASJIT +GCtab * LJ_FASTCALL lj_tab_new1(lua_State *L, uint32_t ahsize) +{ + GCtab *t = newtab(L, ahsize & 0xffffff, ahsize >> 24); + clearapart(t); + if (t->hmask > 0) clearhpart(t); + return t; +} +#endif + +/* Duplicate a table. */ +GCtab * LJ_FASTCALL lj_tab_dup(lua_State *L, const GCtab *kt) +{ + GCtab *t; + uint32_t asize, hmask; + t = newtab(L, kt->asize, kt->hmask > 0 ? lj_fls(kt->hmask)+1 : 0); + lj_assertL(kt->asize == t->asize && kt->hmask == t->hmask, + "mismatched size of table and template"); + t->nomm = 0; /* Keys with metamethod names may be present. */ + asize = kt->asize; + if (asize > 0) { + TValue *array = tvref(t->array); + TValue *karray = tvref(kt->array); + if (asize < 64) { /* An inlined loop beats memcpy for < 512 bytes. */ + uint32_t i; + for (i = 0; i < asize; i++) + copyTV(L, &array[i], &karray[i]); + } else { + memcpy(array, karray, asize*sizeof(TValue)); + } + } + hmask = kt->hmask; + if (hmask > 0) { + uint32_t i; + Node *node = noderef(t->node); + Node *knode = noderef(kt->node); + ptrdiff_t d = (char *)node - (char *)knode; + setfreetop(t, node, (Node *)((char *)getfreetop(kt, knode) + d)); + for (i = 0; i <= hmask; i++) { + Node *kn = &knode[i]; + Node *n = &node[i]; + Node *next = nextnode(kn); + /* Don't use copyTV here, since it asserts on a copy of a dead key. */ + n->val = kn->val; n->key = kn->key; + setmref(n->next, next == NULL? next : (Node *)((char *)next + d)); + } + } + return t; +} + +/* Clear a table. */ +void LJ_FASTCALL lj_tab_clear(GCtab *t) +{ + clearapart(t); + if (t->hmask > 0) { + Node *node = noderef(t->node); + setfreetop(t, node, &node[t->hmask+1]); + clearhpart(t); + } +} + +/* Free a table. */ +void LJ_FASTCALL lj_tab_free(global_State *g, GCtab *t) +{ + if (t->hmask > 0) + lj_mem_freevec(g, noderef(t->node), t->hmask+1, Node); + if (t->asize > 0 && LJ_MAX_COLOSIZE != 0 && t->colo <= 0) + lj_mem_freevec(g, tvref(t->array), t->asize, TValue); + if (LJ_MAX_COLOSIZE != 0 && t->colo) + lj_mem_free(g, t, sizetabcolo((uint32_t)t->colo & 0x7f)); + else + lj_mem_freet(g, t); +} + +/* -- Table resizing ------------------------------------------------------ */ + +/* Resize a table to fit the new array/hash part sizes. */ +void lj_tab_resize(lua_State *L, GCtab *t, uint32_t asize, uint32_t hbits) +{ + Node *oldnode = noderef(t->node); + uint32_t oldasize = t->asize; + uint32_t oldhmask = t->hmask; + if (asize > oldasize) { /* Array part grows? */ + TValue *array; + uint32_t i; + if (asize > LJ_MAX_ASIZE) + lj_err_msg(L, LJ_ERR_TABOV); + if (LJ_MAX_COLOSIZE != 0 && t->colo > 0) { + /* A colocated array must be separated and copied. */ + TValue *oarray = tvref(t->array); + array = lj_mem_newvec(L, asize, TValue); + t->colo = (int8_t)(t->colo | 0x80); /* Mark as separated (colo < 0). */ + for (i = 0; i < oldasize; i++) + copyTV(L, &array[i], &oarray[i]); + } else { + array = (TValue *)lj_mem_realloc(L, tvref(t->array), + oldasize*sizeof(TValue), asize*sizeof(TValue)); + } + setmref(t->array, array); + t->asize = asize; + for (i = oldasize; i < asize; i++) /* Clear newly allocated slots. */ + setnilV(&array[i]); + } + /* Create new (empty) hash part. */ + if (hbits) { + newhpart(L, t, hbits); + clearhpart(t); + } else { + global_State *g = G(L); + setmref(t->node, &g->nilnode); +#if LJ_GC64 + setmref(t->freetop, &g->nilnode); +#endif + t->hmask = 0; + } + if (asize < oldasize) { /* Array part shrinks? */ + TValue *array = tvref(t->array); + uint32_t i; + t->asize = asize; /* Note: This 'shrinks' even colocated arrays. */ + for (i = asize; i < oldasize; i++) /* Reinsert old array values. */ + if (!tvisnil(&array[i])) + copyTV(L, lj_tab_setinth(L, t, (int32_t)i), &array[i]); + /* Physically shrink only separated arrays. */ + if (LJ_MAX_COLOSIZE != 0 && t->colo <= 0) + setmref(t->array, lj_mem_realloc(L, array, + oldasize*sizeof(TValue), asize*sizeof(TValue))); + } + if (oldhmask > 0) { /* Reinsert pairs from old hash part. */ + global_State *g; + uint32_t i; + for (i = 0; i <= oldhmask; i++) { + Node *n = &oldnode[i]; + if (!tvisnil(&n->val)) + copyTV(L, lj_tab_set(L, t, &n->key), &n->val); + } + g = G(L); + lj_mem_freevec(g, oldnode, oldhmask+1, Node); + } +} + +static uint32_t countint(cTValue *key, uint32_t *bins) +{ + lj_assertX(!tvisint(key), "bad integer key"); + if (tvisnum(key)) { + lua_Number nk = numV(key); + int32_t k = lj_num2int(nk); + if ((uint32_t)k < LJ_MAX_ASIZE && nk == (lua_Number)k) { + bins[(k > 2 ? lj_fls((uint32_t)(k-1)) : 0)]++; + return 1; + } + } + return 0; +} + +static uint32_t countarray(const GCtab *t, uint32_t *bins) +{ + uint32_t na, b, i; + if (t->asize == 0) return 0; + for (na = i = b = 0; b < LJ_MAX_ABITS; b++) { + uint32_t n, top = 2u << b; + TValue *array; + if (top >= t->asize) { + top = t->asize-1; + if (i > top) + break; + } + array = tvref(t->array); + for (n = 0; i <= top; i++) + if (!tvisnil(&array[i])) + n++; + bins[b] += n; + na += n; + } + return na; +} + +static uint32_t counthash(const GCtab *t, uint32_t *bins, uint32_t *narray) +{ + uint32_t total, na, i, hmask = t->hmask; + Node *node = noderef(t->node); + for (total = na = 0, i = 0; i <= hmask; i++) { + Node *n = &node[i]; + if (!tvisnil(&n->val)) { + na += countint(&n->key, bins); + total++; + } + } + *narray += na; + return total; +} + +static uint32_t bestasize(uint32_t bins[], uint32_t *narray) +{ + uint32_t b, sum, na = 0, sz = 0, nn = *narray; + for (b = 0, sum = 0; 2*nn > (1u< 0 && 2*(sum += bins[b]) > (1u<hmask > 0 ? lj_fls(t->hmask)+1 : 0); +} + +/* -- Table getters ------------------------------------------------------- */ + +cTValue * LJ_FASTCALL lj_tab_getinth(GCtab *t, int32_t key) +{ + TValue k; + Node *n; + k.n = (lua_Number)key; + n = hashnum(t, &k); + do { + if (tvisnum(&n->key) && n->key.n == k.n) + return &n->val; + } while ((n = nextnode(n))); + return NULL; +} + +cTValue *lj_tab_getstr(GCtab *t, const GCstr *key) +{ + Node *n = hashstr(t, key); + do { + if (tvisstr(&n->key) && strV(&n->key) == key) + return &n->val; + } while ((n = nextnode(n))); + return NULL; +} + +cTValue *lj_tab_get(lua_State *L, GCtab *t, cTValue *key) +{ + if (tvisstr(key)) { + cTValue *tv = lj_tab_getstr(t, strV(key)); + if (tv) + return tv; + } else if (tvisint(key)) { + cTValue *tv = lj_tab_getint(t, intV(key)); + if (tv) + return tv; + } else if (tvisnum(key)) { + lua_Number nk = numV(key); + int32_t k = lj_num2int(nk); + if (nk == (lua_Number)k) { + cTValue *tv = lj_tab_getint(t, k); + if (tv) + return tv; + } else { + goto genlookup; /* Else use the generic lookup. */ + } + } else if (!tvisnil(key)) { + Node *n; + genlookup: + n = hashkey(t, key); + do { + if (lj_obj_equal(&n->key, key)) + return &n->val; + } while ((n = nextnode(n))); + } + return niltv(L); +} + +/* -- Table setters ------------------------------------------------------- */ + +/* Insert new key. Use Brent's variation to optimize the chain length. */ +TValue *lj_tab_newkey(lua_State *L, GCtab *t, cTValue *key) +{ + Node *n = hashkey(t, key); + if (!tvisnil(&n->val) || t->hmask == 0) { + Node *nodebase = noderef(t->node); + Node *collide, *freenode = getfreetop(t, nodebase); + lj_assertL(freenode >= nodebase && freenode <= nodebase+t->hmask+1, + "bad freenode"); + do { + if (freenode == nodebase) { /* No free node found? */ + rehashtab(L, t, key); /* Rehash table. */ + return lj_tab_set(L, t, key); /* Retry key insertion. */ + } + } while (!tvisnil(&(--freenode)->key)); + setfreetop(t, nodebase, freenode); + lj_assertL(freenode != &G(L)->nilnode, "store to fallback hash"); + collide = hashkey(t, &n->key); + if (collide != n) { /* Colliding node not the main node? */ + while (noderef(collide->next) != n) /* Find predecessor. */ + collide = nextnode(collide); + setmref(collide->next, freenode); /* Relink chain. */ + /* Copy colliding node into free node and free main node. */ + freenode->val = n->val; + freenode->key = n->key; + freenode->next = n->next; + setmref(n->next, NULL); + setnilV(&n->val); + /* Rechain pseudo-resurrected string keys with colliding hashes. */ + while (nextnode(freenode)) { + Node *nn = nextnode(freenode); + if (!tvisnil(&nn->val) && hashkey(t, &nn->key) == n) { + freenode->next = nn->next; + nn->next = n->next; + setmref(n->next, nn); + /* + ** Rechaining a resurrected string key creates a new dilemma: + ** Another string key may have originally been resurrected via + ** _any_ of the previous nodes as a chain anchor. Including + ** a node that had to be moved, which makes them unreachable. + ** It's not feasible to check for all previous nodes, so rechain + ** any string key that's currently in a non-main positions. + */ + while ((nn = nextnode(freenode))) { + if (!tvisnil(&nn->val)) { + Node *mn = hashkey(t, &nn->key); + if (mn != freenode && mn != nn) { + freenode->next = nn->next; + nn->next = mn->next; + setmref(mn->next, nn); + } else { + freenode = nn; + } + } else { + freenode = nn; + } + } + break; + } else { + freenode = nn; + } + } + } else { /* Otherwise use free node. */ + setmrefr(freenode->next, n->next); /* Insert into chain. */ + setmref(n->next, freenode); + n = freenode; + } + } + n->key.u64 = key->u64; + if (LJ_UNLIKELY(tvismzero(&n->key))) + n->key.u64 = 0; + lj_gc_anybarriert(L, t); + lj_assertL(tvisnil(&n->val), "new hash slot is not empty"); + return &n->val; +} + +TValue *lj_tab_setinth(lua_State *L, GCtab *t, int32_t key) +{ + TValue k; + Node *n; + k.n = (lua_Number)key; + n = hashnum(t, &k); + do { + if (tvisnum(&n->key) && n->key.n == k.n) + return &n->val; + } while ((n = nextnode(n))); + return lj_tab_newkey(L, t, &k); +} + +TValue *lj_tab_setstr(lua_State *L, GCtab *t, const GCstr *key) +{ + TValue k; + Node *n = hashstr(t, key); + do { + if (tvisstr(&n->key) && strV(&n->key) == key) + return &n->val; + } while ((n = nextnode(n))); + setstrV(L, &k, key); + return lj_tab_newkey(L, t, &k); +} + +TValue *lj_tab_set(lua_State *L, GCtab *t, cTValue *key) +{ + Node *n; + t->nomm = 0; /* Invalidate negative metamethod cache. */ + if (tvisstr(key)) { + return lj_tab_setstr(L, t, strV(key)); + } else if (tvisint(key)) { + return lj_tab_setint(L, t, intV(key)); + } else if (tvisnum(key)) { + lua_Number nk = numV(key); + int32_t k = lj_num2int(nk); + if (nk == (lua_Number)k) + return lj_tab_setint(L, t, k); + if (tvisnan(key)) + lj_err_msg(L, LJ_ERR_NANIDX); + /* Else use the generic lookup. */ + } else if (tvisnil(key)) { + lj_err_msg(L, LJ_ERR_NILIDX); + } + n = hashkey(t, key); + do { + if (lj_obj_equal(&n->key, key)) + return &n->val; + } while ((n = nextnode(n))); + return lj_tab_newkey(L, t, key); +} + +/* -- Table traversal ----------------------------------------------------- */ + +/* Table traversal indexes: +** +** Array key index: [0 .. t->asize-1] +** Hash key index: [t->asize .. t->asize+t->hmask] +** Invalid key: ~0 +*/ + +/* Get the successor traversal index of a key. */ +uint32_t LJ_FASTCALL lj_tab_keyindex(GCtab *t, cTValue *key) +{ + TValue tmp; + if (tvisint(key)) { + int32_t k = intV(key); + if ((uint32_t)k < t->asize) + return (uint32_t)k + 1; + setnumV(&tmp, (lua_Number)k); + key = &tmp; + } else if (tvisnum(key)) { + lua_Number nk = numV(key); + int32_t k = lj_num2int(nk); + if ((uint32_t)k < t->asize && nk == (lua_Number)k) + return (uint32_t)k + 1; + } + if (!tvisnil(key)) { + Node *n = hashkey(t, key); + do { + if (lj_obj_equal(&n->key, key)) + return t->asize + (uint32_t)((n+1) - noderef(t->node)); + } while ((n = nextnode(n))); + if (key->u32.hi == LJ_KEYINDEX) /* Despecialized ITERN while running. */ + return key->u32.lo; + return ~0u; /* Invalid key to next. */ + } + return 0; /* A nil key starts the traversal. */ +} + +/* Get the next key/value pair of a table traversal. */ +int lj_tab_next(GCtab *t, cTValue *key, TValue *o) +{ + uint32_t idx = lj_tab_keyindex(t, key); /* Find successor index of key. */ + /* First traverse the array part. */ + for (; idx < t->asize; idx++) { + cTValue *a = arrayslot(t, idx); + if (LJ_LIKELY(!tvisnil(a))) { + setintV(o, idx); + o[1] = *a; + return 1; + } + } + idx -= t->asize; + /* Then traverse the hash part. */ + for (; idx <= t->hmask; idx++) { + Node *n = &noderef(t->node)[idx]; + if (!tvisnil(&n->val)) { + o[0] = n->key; + o[1] = n->val; + return 1; + } + } + return (int32_t)idx < 0 ? -1 : 0; /* Invalid key or end of traversal. */ +} + +/* -- Table length calculation -------------------------------------------- */ + +/* Compute table length. Slow path with mixed array/hash lookups. */ +LJ_NOINLINE static MSize tab_len_slow(GCtab *t, size_t hi) +{ + cTValue *tv; + size_t lo = hi; + hi++; + /* Widening search for an upper bound. */ + while ((tv = lj_tab_getint(t, (int32_t)hi)) && !tvisnil(tv)) { + lo = hi; + hi += hi; + if (hi > (size_t)(0x7fffffff - 2)) { /* Punt and do a linear search. */ + lo = 1; + while ((tv = lj_tab_getint(t, (int32_t)lo)) && !tvisnil(tv)) lo++; + return (MSize)(lo - 1); + } + } + /* Binary search to find a non-nil to nil transition. */ + while (hi - lo > 1) { + size_t mid = (lo+hi) >> 1; + cTValue *tvb = lj_tab_getint(t, (int32_t)mid); + if (tvb && !tvisnil(tvb)) lo = mid; else hi = mid; + } + return (MSize)lo; +} + +/* Compute table length. Fast path. */ +MSize LJ_FASTCALL lj_tab_len(GCtab *t) +{ + size_t hi = (size_t)t->asize; + if (hi) hi--; + /* In a growing array the last array element is very likely nil. */ + if (hi > 0 && LJ_LIKELY(tvisnil(arrayslot(t, hi)))) { + /* Binary search to find a non-nil to nil transition in the array. */ + size_t lo = 0; + while (hi - lo > 1) { + size_t mid = (lo+hi) >> 1; + if (tvisnil(arrayslot(t, mid))) hi = mid; else lo = mid; + } + return (MSize)lo; + } + /* Without a hash part, there's an implicit nil after the last element. */ + return t->hmask ? tab_len_slow(t, hi) : (MSize)hi; +} + +#if LJ_HASJIT +/* Verify hinted table length or compute it. */ +MSize LJ_FASTCALL lj_tab_len_hint(GCtab *t, size_t hint) +{ + size_t asize = (size_t)t->asize; + cTValue *tv = arrayslot(t, hint); + if (LJ_LIKELY(hint+1 < asize)) { + if (LJ_LIKELY(!tvisnil(tv) && tvisnil(tv+1))) return (MSize)hint; + } else if (hint+1 <= asize && LJ_LIKELY(t->hmask == 0) && !tvisnil(tv)) { + return (MSize)hint; + } + return lj_tab_len(t); +} +#endif + diff --git a/src/bluajit/lj_vmmath.c b/src/bluajit/lj_vmmath.c new file mode 100644 index 000000000..9e8f1e0e6 --- /dev/null +++ b/src/bluajit/lj_vmmath.c @@ -0,0 +1,158 @@ +/* +** Math helper functions for assembler VM. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#define lj_vmmath_c +#define LUA_CORE + +#include +#include + +#include "lj_obj.h" +#include "lj_ir.h" +#include "lj_vm.h" + +/* -- Wrapper functions --------------------------------------------------- */ + +#if LJ_TARGET_X86 && __ELF__ && __PIC__ +/* Wrapper functions to deal with the ELF/x86 PIC disaster. */ +LJ_FUNCA double lj_wrap_log(double x) { return log(x); } +LJ_FUNCA double lj_wrap_log10(double x) { return log10(x); } +LJ_FUNCA double lj_wrap_exp(double x) { return exp(x); } +LJ_FUNCA double lj_wrap_sin(double x) { return sin(x); } +LJ_FUNCA double lj_wrap_cos(double x) { return cos(x); } +LJ_FUNCA double lj_wrap_tan(double x) { return tan(x); } +LJ_FUNCA double lj_wrap_asin(double x) { return asin(x); } +LJ_FUNCA double lj_wrap_acos(double x) { return acos(x); } +LJ_FUNCA double lj_wrap_atan(double x) { return atan(x); } +LJ_FUNCA double lj_wrap_sinh(double x) { return sinh(x); } +LJ_FUNCA double lj_wrap_cosh(double x) { return cosh(x); } +LJ_FUNCA double lj_wrap_tanh(double x) { return tanh(x); } +LJ_FUNCA double lj_wrap_atan2(double x, double y) { return atan2(x, y); } +LJ_FUNCA double lj_wrap_pow(double x, double y) { return pow(x, y); } +LJ_FUNCA double lj_wrap_fmod(double x, double y) { return fmod(x, y); } +#endif + +/* -- Helper functions ---------------------------------------------------- */ + +/* Required to prevent the C compiler from applying FMA optimizations. +** +** Yes, there's -ffp-contract and the FP_CONTRACT pragma ... in theory. +** But the current state of C compilers is a mess in this regard. +** Also, this function is not performance sensitive at all. +*/ +LJ_NOINLINE static double lj_vm_floormul(double x, double y) +{ + return lj_vm_floor(x / y) * y; +} + +double lj_vm_foldarith(double x, double y, int op) +{ + switch (op) { + case IR_ADD - IR_ADD: return x+y; break; + case IR_SUB - IR_ADD: return x-y; break; + case IR_MUL - IR_ADD: return x*y; break; + case IR_DIV - IR_ADD: return x/y; break; + case IR_MOD - IR_ADD: return x-lj_vm_floormul(x, y); break; + case IR_POW - IR_ADD: return pow(x, y); break; + case IR_NEG - IR_ADD: return -x; break; + case IR_ABS - IR_ADD: return fabs(x); break; +#if LJ_HASJIT + case IR_LDEXP - IR_ADD: return ldexp(x, (int)y); break; + case IR_MIN - IR_ADD: return x < y ? x : y; break; + case IR_MAX - IR_ADD: return x > y ? x : y; break; +#endif + default: return x; + } +} + +#if LJ_INTONLY +int32_t lj_vm_foldarith_int(int32_t x, int32_t y, int op) +{ + switch (op) { + case IR_ADD - IR_ADD: return x+y; break; + case IR_SUB - IR_ADD: return x-y; break; + case IR_MUL - IR_ADD: return x*y; break; + case IR_DIV - IR_ADD: return x/y; break; + case IR_MOD - IR_ADD: return x%y; break; + case IR_POW - IR_ADD: return pow(x, y); break; + case IR_NEG - IR_ADD: return -x; break; + case IR_ABS - IR_ADD: return abs(x); break; +#if LJ_HASJIT + case IR_LDEXP - IR_ADD: return ldexp(x, (int)y); break; + case IR_MIN - IR_ADD: return x < y ? x : y; break; + case IR_MAX - IR_ADD: return x > y ? x : y; break; +#endif + default: return x; + } +} + +int32_t lj_vm_foldbitwise(uint32_t x, uint32_t y, int op) +#else +int64_t lj_vm_foldbitwise(int64_t x, int64_t y, int op) +#endif +{ + switch (op) { + case IR_BAND - IR_BAND: return x&y; break; + case IR_BOR - IR_BAND: return x|y; break; + case IR_BXOR - IR_BAND: return x^y; break; + case IR_BSHL - IR_BAND: return x<>y; break; + default: return ~x; /* IR_BNOT */ + } +} + +/* -- Helper functions for generated machine code ------------------------- */ + +#if (LJ_HASJIT && !(LJ_TARGET_ARM || LJ_TARGET_ARM64 || LJ_TARGET_PPC)) || LJ_TARGET_MIPS +int32_t LJ_FASTCALL lj_vm_modi(int32_t a, int32_t b) +{ + /* This must be checked before using this function. */ + lj_assertX(b != 0, "modulo with zero divisor"); +#if LJ_INTONLY + return a % b; +#else + uint32_t y, ua, ub; + ua = a < 0 ? ~(uint32_t)a+1u : (uint32_t)a; + ub = b < 0 ? ~(uint32_t)b+1u : (uint32_t)b; + y = ua % ub; + if (y != 0 && (a^b) < 0) y = y - ub; + if (((int32_t)y^b) < 0) y = ~y+1u; + return (int32_t)y; +#endif +} +#endif + +#if LJ_HASJIT + +#ifdef LUAJIT_NO_LOG2 +double lj_vm_log2(double a) +{ + return log(a) * 1.4426950408889634074; +} +#endif + +/* Computes fpm(x) for extended math functions. */ +double lj_vm_foldfpm(double x, int fpm) +{ + switch (fpm) { + case IRFPM_FLOOR: return lj_vm_floor(x); + case IRFPM_CEIL: return lj_vm_ceil(x); + case IRFPM_TRUNC: return lj_vm_trunc(x); + case IRFPM_SQRT: return sqrt(x); + case IRFPM_LOG: return log(x); + case IRFPM_LOG2: return lj_vm_log2(x); + default: lj_assertX(0, "bad fpm %d", fpm); + } + return 0; +} + +#if LJ_HASFFI +int lj_vm_errno(void) +{ + return errno; +} +#endif + +#endif diff --git a/src/bluajit/luaconf.h b/src/bluajit/luaconf.h new file mode 100644 index 000000000..f59cc1b5f --- /dev/null +++ b/src/bluajit/luaconf.h @@ -0,0 +1,206 @@ +/* +** Configuration header. +** Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +*/ + +#ifndef luaconf_h +#define luaconf_h + +#ifndef WINVER +#define WINVER 0x0501 +#endif +#include + +/* Default path for loading Lua and C modules with require(). */ +#if defined(_WIN32) +/* +** In Windows, any exclamation mark ('!') in the path is replaced by the +** path of the directory of the executable file of the current process. +*/ +#define LUA_LDIR "!\\lua\\" +#define LUA_CDIR "!\\" +#define LUA_PATH_DEFAULT \ + ".\\?.lua;" LUA_LDIR"?.lua;" LUA_LDIR"?\\init.lua;" +#define LUA_CPATH_DEFAULT \ + ".\\?.dll;" LUA_CDIR"?.dll;" LUA_CDIR"loadall.dll" +#else +/* +** Note to distribution maintainers: do NOT patch the following lines! +** Please read ../doc/install.html#distro and pass PREFIX=/usr instead. +*/ +#ifndef LUA_MULTILIB +#define LUA_MULTILIB "lib" +#endif +#ifndef LUA_LMULTILIB +#define LUA_LMULTILIB "lib" +#endif +#define LUA_LROOT "/usr/local" +#define LUA_LUADIR "/lua/5.1/" + +#ifdef LUA_ROOT +#define LUA_JROOT LUA_ROOT +#define LUA_RLDIR LUA_ROOT "/share" LUA_LUADIR +#define LUA_RCDIR LUA_ROOT "/" LUA_MULTILIB LUA_LUADIR +#define LUA_RLPATH ";" LUA_RLDIR "?.lua;" LUA_RLDIR "?/init.lua" +#define LUA_RCPATH ";" LUA_RCDIR "?.so" +#else +#define LUA_JROOT LUA_LROOT +#define LUA_RLPATH +#define LUA_RCPATH +#endif + +#ifndef LUA_LJDIR +#define LUA_LJDIR LUA_JROOT "/share/luajit-2.1" +#endif + +#define LUA_JPATH ";" LUA_LJDIR "/?.lua" +#define LUA_LLDIR LUA_LROOT "/share" LUA_LUADIR +#define LUA_LCDIR LUA_LROOT "/" LUA_LMULTILIB LUA_LUADIR +#define LUA_LLPATH ";" LUA_LLDIR "?.lua;" LUA_LLDIR "?/init.lua" +#define LUA_LCPATH1 ";" LUA_LCDIR "?.so" +#define LUA_LCPATH2 ";" LUA_LCDIR "loadall.so" + +#define LUA_PATH_DEFAULT "./?.lua" LUA_JPATH LUA_LLPATH LUA_RLPATH +#define LUA_CPATH_DEFAULT "./?.so" LUA_LCPATH1 LUA_RCPATH LUA_LCPATH2 +#endif + +/* Environment variable names for path overrides and initialization code. */ +#define LUA_PATH "LUA_PATH" +#define LUA_CPATH "LUA_CPATH" +#define LUA_INIT "LUA_INIT" + +/* Special file system characters. */ +#if defined(_WIN32) +#define LUA_DIRSEP "\\" +#else +#define LUA_DIRSEP "/" +#endif +#define LUA_PATHSEP ";" +#define LUA_PATH_MARK "?" +#define LUA_EXECDIR "!" +#define LUA_IGMARK "-" +#define LUA_PATH_CONFIG \ + LUA_DIRSEP "\n" LUA_PATHSEP "\n" LUA_PATH_MARK "\n" \ + LUA_EXECDIR "\n" LUA_IGMARK "\n" + +/* Quoting in error messages. */ +#define LUA_QL(x) "'" x "'" +#define LUA_QS LUA_QL("%s") + +/* Various tunables. */ +#define LUAI_MAXSTACK 65500 /* Max. # of stack slots for a thread (<64K). */ +#define LUAI_MAXCSTACK 8000 /* Max. # of stack slots for a C func (<10K). */ +#define LUAI_GCPAUSE 200 /* Pause GC until memory is at 200%. */ +#define LUAI_GCMUL 200 /* Run GC at 200% of allocation speed. */ +#define LUA_MAXCAPTURES 32 /* Max. pattern captures. */ + +/* Configuration for the frontend (the luajit executable). */ +#if defined(luajit_c) +#define LUA_PROGNAME "luajit" /* Fallback frontend name. */ +#define LUA_PROMPT "> " /* Interactive prompt. */ +#define LUA_PROMPT2 ">> " /* Continuation prompt. */ +#define LUA_MAXINPUT 512 /* Max. input line length. */ +#endif + +/* Note: changing the following defines breaks the Lua 5.1 ABI. */ +#ifdef LUAJIT_INTONLY +#include +#define LUA_INTEGER int32_t +#else +#define LUA_INTEGER ptrdiff_t +#endif +#define LUA_IDSIZE 60 /* Size of lua_Debug.short_src. */ +/* +** Size of lauxlib and io.* on-stack buffers. Weird workaround to avoid using +** unreasonable amounts of stack space, but still retain ABI compatibility. +** Blame Lua for depending on BUFSIZ in the ABI, blame **** for wrecking it. +*/ +#define LUAL_BUFFERSIZE (BUFSIZ > 16384 ? 8192 : BUFSIZ) + +/* The following defines are here only for compatibility with luaconf.h +** from the standard Lua distribution. They must not be changed for LuaJIT. +*/ +#define LUA_NUMBER_DOUBLE +#define LUA_NUMBER double +#define LUAI_UACNUMBER double +#define LUA_NUMBER_SCAN "%lf" +#define LUA_NUMBER_FMT "%.14g" +#define lua_number2str(s, n) sprintf((s), LUA_NUMBER_FMT, (n)) +#define LUAI_MAXNUMBER2STR 32 +#define LUA_INTFRMLEN "l" +#define LUA_INTFRM_T long + +/* Linkage of public API functions. */ +#if defined(LUA_BUILD_AS_DLL) +#if defined(LUA_CORE) || defined(LUA_LIB) +#define LUA_API __declspec(dllexport) +#else +#define LUA_API __declspec(dllimport) +#endif +#else +#define LUA_API extern +#endif + +#define LUALIB_API LUA_API + +/* Compatibility support for assertions. */ +#if defined(LUA_USE_ASSERT) || defined(LUA_USE_APICHECK) +#include +#endif +#ifdef LUA_USE_ASSERT +#define lua_assert(x) assert(x) +#endif +#ifdef LUA_USE_APICHECK +#define luai_apicheck(L, o) { (void)L; assert(o); } +#else +#define luai_apicheck(L, o) { (void)L; } +#endif + +/****************************************************************************** + * bluajit configuration * + ******************************************************************************/ + +/** + * [bluajit] Allow != as well as the traditional Lua ~= for inequality comparison. + */ +#ifndef BLUAJIT_BANG_EQUALS_TOKEN + #define BLUAJIT_BANG_EQUALS_TOKEN 1 +#endif + +/** + * [bluajit] Make `then` after `if ` optional. + */ +#ifndef BLUAJIT_OPTIONAL_THEN_TOKEN + #define BLUAJIT_OPTIONAL_THEN_TOKEN 1 +#endif + +/** + * [bluajit] Make `do` after `for ` and `while ` optional. + */ +#ifndef BLUAJIT_OPTIONAL_DO_TOKEN + #define BLUAJIT_OPTIONAL_DO_TOKEN 1 +#endif + +/** + * [bluajit] Make `do` and `end` create an anonymous function body, much like + * `function() end` + */ +#ifndef BLUAJIT_DO_END_ANONYMOUS_FUNCTIONS + #define BLUAJIT_DO_END_ANONYMOUS_FUNCTIONS 1 +#endif + +/** + * [bluajit] Enable multilevel loop breaks with `break n`. + */ +#ifndef BLUAJIT_MULTILEVEL_BREAK + #define BLUAJIT_MULTILEVEL_BREAK 1 +#endif + +/** + * [bluajit] Add `continue` and `continue n` statements. + */ +#ifndef BLUAJIT_CONTINUE_STATEMENT + #define BLUAJIT_CONTINUE_STATEMENT 1 +#endif + +#endif diff --git a/src/bluajit/vm_mips64.dasc b/src/bluajit/vm_mips64.dasc new file mode 100644 index 000000000..1f0118ecf --- /dev/null +++ b/src/bluajit/vm_mips64.dasc @@ -0,0 +1,5566 @@ +|// Low-level VM code for MIPS64 CPUs. +|// Bytecode interpreter, fast functions and helper functions. +|// Copyright (C) 2005-2023 Mike Pall. See Copyright Notice in luajit.h +|// +|// Contributed by Djordje Kovacevic and Stefan Pejic from RT-RK.com. +|// Sponsored by Cisco Systems, Inc. +| +|.arch mips64 +|.section code_op, code_sub +| +|.actionlist build_actionlist +|.globals GLOB_ +|.globalnames globnames +|.externnames extnames +| +|// Note: The ragged indentation of the instructions is intentional. +|// The starting columns indicate data dependencies. +| +|//----------------------------------------------------------------------- +| +|// Fixed register assignments for the interpreter. +|// Don't use: r0 = 0, r26/r27 = reserved, r28 = gp, r29 = sp, r31 = ra +| +|.macro .FPU, a, b +|.if FPU +| a, b +|.endif +|.endmacro +| +|// The following must be C callee-save (but BASE is often refetched). +|.define BASE, r16 // Base of current Lua stack frame. +|.define KBASE, r17 // Constants of current Lua function. +|.define PC, r18 // Next PC. +|.define DISPATCH, r19 // Opcode dispatch table. +|.define LREG, r20 // Register holding lua_State (also in SAVE_L). +|.define MULTRES, r21 // Size of multi-result: (nresults+1)*8. +| +|.define JGL, r30 // On-trace: global_State + 32768. +| +|// Constants for type-comparisons, stores and conversions. C callee-save. +|.define TISNIL, r30 +|.define TISNUM, r22 +|.if FPU +|.define TOBIT, f30 // 2^52 + 2^51. +|.endif +| +|// The following temporaries are not saved across C calls, except for RA. +|.define RA, r23 // Callee-save. +|.define RB, r8 +|.define RC, r9 +|.define RD, r10 +|.define INS, r11 +| +|.define AT, r1 // Assembler temporary. +|.define TMP0, r12 +|.define TMP1, r13 +|.define TMP2, r14 +|.define TMP3, r15 +| +|// MIPS n64 calling convention. +|.define CFUNCADDR, r25 +|.define CARG1, r4 +|.define CARG2, r5 +|.define CARG3, r6 +|.define CARG4, r7 +|.define CARG5, r8 +|.define CARG6, r9 +|.define CARG7, r10 +|.define CARG8, r11 +| +|.define CRET1, r2 +|.define CRET2, r3 +| +|.if FPU +|.define FARG1, f12 +|.define FARG2, f13 +|.define FARG3, f14 +|.define FARG4, f15 +|.define FARG5, f16 +|.define FARG6, f17 +|.define FARG7, f18 +|.define FARG8, f19 +| +|.define FRET1, f0 +|.define FRET2, f2 +| +|.define FTMP0, f20 +|.define FTMP1, f21 +|.define FTMP2, f22 +|.endif +| +|// Stack layout while in interpreter. Must match with lj_frame.h. +|.if FPU // MIPS64 hard-float. +| +|.define CFRAME_SPACE, 192 // Delta for sp. +| +|//----- 16 byte aligned, <-- sp entering interpreter +|.define SAVE_ERRF, 188(sp) // 32 bit values. +|.define SAVE_NRES, 184(sp) +|.define SAVE_CFRAME, 176(sp) // 64 bit values. +|.define SAVE_L, 168(sp) +|.define SAVE_PC, 160(sp) +|//----- 16 byte aligned +|.define SAVE_GPR_, 80 // .. 80+10*8: 64 bit GPR saves. +|.define SAVE_FPR_, 16 // .. 16+8*8: 64 bit FPR saves. +| +|.else // MIPS64 soft-float +| +|.define CFRAME_SPACE, 128 // Delta for sp. +| +|//----- 16 byte aligned, <-- sp entering interpreter +|.define SAVE_ERRF, 124(sp) // 32 bit values. +|.define SAVE_NRES, 120(sp) +|.define SAVE_CFRAME, 112(sp) // 64 bit values. +|.define SAVE_L, 104(sp) +|.define SAVE_PC, 96(sp) +|//----- 16 byte aligned +|.define SAVE_GPR_, 16 // .. 16+10*8: 64 bit GPR saves. +| +|.endif +| +|.define TMPX, 8(sp) // Unused by interpreter, temp for JIT code. +|.define TMPD, 0(sp) +|//----- 16 byte aligned +| +|.define TMPD_OFS, 0 +| +|.define SAVE_MULTRES, TMPD +| +|//----------------------------------------------------------------------- +| +|.macro saveregs +| daddiu sp, sp, -CFRAME_SPACE +| sd ra, SAVE_GPR_+9*8(sp) +| sd r30, SAVE_GPR_+8*8(sp) +| .FPU sdc1 f31, SAVE_FPR_+7*8(sp) +| sd r23, SAVE_GPR_+7*8(sp) +| .FPU sdc1 f30, SAVE_FPR_+6*8(sp) +| sd r22, SAVE_GPR_+6*8(sp) +| .FPU sdc1 f29, SAVE_FPR_+5*8(sp) +| sd r21, SAVE_GPR_+5*8(sp) +| .FPU sdc1 f28, SAVE_FPR_+4*8(sp) +| sd r20, SAVE_GPR_+4*8(sp) +| .FPU sdc1 f27, SAVE_FPR_+3*8(sp) +| sd r19, SAVE_GPR_+3*8(sp) +| .FPU sdc1 f26, SAVE_FPR_+2*8(sp) +| sd r18, SAVE_GPR_+2*8(sp) +| .FPU sdc1 f25, SAVE_FPR_+1*8(sp) +| sd r17, SAVE_GPR_+1*8(sp) +| .FPU sdc1 f24, SAVE_FPR_+0*8(sp) +| sd r16, SAVE_GPR_+0*8(sp) +|.endmacro +| +|.macro restoreregs_ret +| ld ra, SAVE_GPR_+9*8(sp) +| ld r30, SAVE_GPR_+8*8(sp) +| ld r23, SAVE_GPR_+7*8(sp) +| .FPU ldc1 f31, SAVE_FPR_+7*8(sp) +| ld r22, SAVE_GPR_+6*8(sp) +| .FPU ldc1 f30, SAVE_FPR_+6*8(sp) +| ld r21, SAVE_GPR_+5*8(sp) +| .FPU ldc1 f29, SAVE_FPR_+5*8(sp) +| ld r20, SAVE_GPR_+4*8(sp) +| .FPU ldc1 f28, SAVE_FPR_+4*8(sp) +| ld r19, SAVE_GPR_+3*8(sp) +| .FPU ldc1 f27, SAVE_FPR_+3*8(sp) +| ld r18, SAVE_GPR_+2*8(sp) +| .FPU ldc1 f26, SAVE_FPR_+2*8(sp) +| ld r17, SAVE_GPR_+1*8(sp) +| .FPU ldc1 f25, SAVE_FPR_+1*8(sp) +| ld r16, SAVE_GPR_+0*8(sp) +| .FPU ldc1 f24, SAVE_FPR_+0*8(sp) +| jr ra +| daddiu sp, sp, CFRAME_SPACE +|.endmacro +| +|// Type definitions. Some of these are only used for documentation. +|.type L, lua_State, LREG +|.type GL, global_State +|.type TVALUE, TValue +|.type GCOBJ, GCobj +|.type STR, GCstr +|.type TAB, GCtab +|.type LFUNC, GCfuncL +|.type CFUNC, GCfuncC +|.type PROTO, GCproto +|.type UPVAL, GCupval +|.type NODE, Node +|.type NARGS8, int +|.type TRACE, GCtrace +|.type SBUF, SBuf +| +|//----------------------------------------------------------------------- +| +|// Trap for not-yet-implemented parts. +|.macro NYI; .long 0xec1cf0f0; .endmacro +| +|// Macros to mark delay slots. +|.macro ., a; a; .endmacro +|.macro ., a,b; a,b; .endmacro +|.macro ., a,b,c; a,b,c; .endmacro +|.macro ., a,b,c,d; a,b,c,d; .endmacro +| +|.define FRAME_PC, -8 +|.define FRAME_FUNC, -16 +| +|//----------------------------------------------------------------------- +| +|// Endian-specific defines. +|.if ENDIAN_LE +|.define HI, 4 +|.define LO, 0 +|.define OFS_RD, 2 +|.define OFS_RA, 1 +|.define OFS_OP, 0 +|.else +|.define HI, 0 +|.define LO, 4 +|.define OFS_RD, 0 +|.define OFS_RA, 2 +|.define OFS_OP, 3 +|.endif +| +|// Instruction decode. +|.macro decode_OP1, dst, ins; andi dst, ins, 0xff; .endmacro +|.macro decode_OP8a, dst, ins; andi dst, ins, 0xff; .endmacro +|.macro decode_OP8b, dst; sll dst, dst, 3; .endmacro +|.macro decode_RC8a, dst, ins; srl dst, ins, 13; .endmacro +|.macro decode_RC8b, dst; andi dst, dst, 0x7f8; .endmacro +|.macro decode_RD4b, dst; sll dst, dst, 2; .endmacro +|.macro decode_RA8a, dst, ins; srl dst, ins, 5; .endmacro +|.macro decode_RA8b, dst; andi dst, dst, 0x7f8; .endmacro +|.macro decode_RB8a, dst, ins; srl dst, ins, 21; .endmacro +|.macro decode_RB8b, dst; andi dst, dst, 0x7f8; .endmacro +|.macro decode_RD8a, dst, ins; srl dst, ins, 16; .endmacro +|.macro decode_RD8b, dst; sll dst, dst, 3; .endmacro +|.macro decode_RDtoRC8, dst, src; andi dst, src, 0x7f8; .endmacro +| +|// Instruction fetch. +|.macro ins_NEXT1 +| lw INS, 0(PC) +| daddiu PC, PC, 4 +|.endmacro +|// Instruction decode+dispatch. +|.macro ins_NEXT2 +| decode_OP8a TMP1, INS +| decode_OP8b TMP1 +| daddu TMP0, DISPATCH, TMP1 +| decode_RD8a RD, INS +| ld AT, 0(TMP0) +| decode_RA8a RA, INS +| decode_RD8b RD +| jr AT +| decode_RA8b RA +|.endmacro +|.macro ins_NEXT +| ins_NEXT1 +| ins_NEXT2 +|.endmacro +| +|// Instruction footer. +|.if 1 +| // Replicated dispatch. Less unpredictable branches, but higher I-Cache use. +| .define ins_next, ins_NEXT +| .define ins_next_, ins_NEXT +| .define ins_next1, ins_NEXT1 +| .define ins_next2, ins_NEXT2 +|.else +| // Common dispatch. Lower I-Cache use, only one (very) unpredictable branch. +| // Affects only certain kinds of benchmarks (and only with -j off). +| .macro ins_next +| b ->ins_next +| .endmacro +| .macro ins_next1 +| .endmacro +| .macro ins_next2 +| b ->ins_next +| .endmacro +| .macro ins_next_ +| ->ins_next: +| ins_NEXT +| .endmacro +|.endif +| +|// Call decode and dispatch. +|.macro ins_callt +| // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC +| ld PC, LFUNC:RB->pc +| lw INS, 0(PC) +| daddiu PC, PC, 4 +| decode_OP8a TMP1, INS +| decode_RA8a RA, INS +| decode_OP8b TMP1 +| decode_RA8b RA +| daddu TMP0, DISPATCH, TMP1 +| ld TMP0, 0(TMP0) +| jr TMP0 +| daddu RA, RA, BASE +|.endmacro +| +|.macro ins_call +| // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, PC = caller PC +| sd PC, FRAME_PC(BASE) +| ins_callt +|.endmacro +| +|//----------------------------------------------------------------------- +| +|.macro branch_RD +| srl TMP0, RD, 1 +| lui AT, (-(BCBIAS_J*4 >> 16) & 65535) +| addu TMP0, TMP0, AT +| daddu PC, PC, TMP0 +|.endmacro +| +|// Assumes DISPATCH is relative to GL. +#define DISPATCH_GL(field) (GG_DISP2G + (int)offsetof(global_State, field)) +#define DISPATCH_J(field) (GG_DISP2J + (int)offsetof(jit_State, field)) +#define GG_DISP2GOT (GG_OFS(got) - GG_OFS(dispatch)) +#define DISPATCH_GOT(name) (GG_DISP2GOT + sizeof(void*)*LJ_GOT_##name) +| +#define PC2PROTO(field) ((int)offsetof(GCproto, field)-(int)sizeof(GCproto)) +| +|.macro load_got, func +| ld CFUNCADDR, DISPATCH_GOT(func)(DISPATCH) +|.endmacro +|// Much faster. Sadly, there's no easy way to force the required code layout. +|// .macro call_intern, func; bal extern func; .endmacro +|.macro call_intern, func; jalr CFUNCADDR; .endmacro +|.macro call_extern; jalr CFUNCADDR; .endmacro +|.macro jmp_extern; jr CFUNCADDR; .endmacro +| +|.macro hotcheck, delta, target +| dsrl TMP1, PC, 1 +| andi TMP1, TMP1, 126 +| daddu TMP1, TMP1, DISPATCH +| lhu TMP2, GG_DISP2HOT(TMP1) +| addiu TMP2, TMP2, -delta +| bltz TMP2, target +|. sh TMP2, GG_DISP2HOT(TMP1) +|.endmacro +| +|.macro hotloop +| hotcheck HOTCOUNT_LOOP, ->vm_hotloop +|.endmacro +| +|.macro hotcall +| hotcheck HOTCOUNT_CALL, ->vm_hotcall +|.endmacro +| +|// Set current VM state. Uses TMP0. +|.macro li_vmstate, st; li TMP0, ~LJ_VMST_..st; .endmacro +|.macro st_vmstate; sw TMP0, DISPATCH_GL(vmstate)(DISPATCH); .endmacro +| +|// Move table write barrier back. Overwrites mark and tmp. +|.macro barrierback, tab, mark, tmp, target +| ld tmp, DISPATCH_GL(gc.grayagain)(DISPATCH) +| andi mark, mark, ~LJ_GC_BLACK & 255 // black2gray(tab) +| sd tab, DISPATCH_GL(gc.grayagain)(DISPATCH) +| sb mark, tab->marked +| b target +|. sd tmp, tab->gclist +|.endmacro +| +|// Clear type tag. Isolate lowest 14+32+1=47 bits of reg. +|.macro cleartp, reg; dextm reg, reg, 0, 14; .endmacro +|.macro cleartp, dst, reg; dextm dst, reg, 0, 14; .endmacro +| +|// Set type tag: Merge 17 type bits into bits [15+32=47, 31+32+1=64) of dst. +|.macro settp, dst, tp; dinsu dst, tp, 15, 31; .endmacro +| +|// Extract (negative) type tag. +|.macro gettp, dst, src; dsra dst, src, 47; .endmacro +| +|// Macros to check the TValue type and extract the GCobj. Branch on failure. +|.macro checktp, reg, tp, target +| gettp AT, reg +| daddiu AT, AT, tp +| bnez AT, target +|. cleartp reg +|.endmacro +|.macro checktp, dst, reg, tp, target +| gettp AT, reg +| daddiu AT, AT, tp +| bnez AT, target +|. cleartp dst, reg +|.endmacro +|.macro checkstr, reg, target; checktp reg, -LJ_TSTR, target; .endmacro +|.macro checktab, reg, target; checktp reg, -LJ_TTAB, target; .endmacro +|.macro checkfunc, reg, target; checktp reg, -LJ_TFUNC, target; .endmacro +|.macro checkint, reg, target // Caveat: has delay slot! +| gettp AT, reg +| bne AT, TISNUM, target +|.endmacro +|.macro checknum, reg, target // Caveat: has delay slot! +| gettp AT, reg +| sltiu AT, AT, LJ_TISNUM +| beqz AT, target +|.endmacro +| +|.macro mov_false, reg +| lu reg, 0x8000 +| dsll reg, reg, 32 +| not reg, reg +|.endmacro +|.macro mov_true, reg +| li reg, 0x0001 +| dsll reg, reg, 48 +| not reg, reg +|.endmacro +| +|//----------------------------------------------------------------------- + +/* Generate subroutines used by opcodes and other parts of the VM. */ +/* The .code_sub section should be last to help static branch prediction. */ +static void build_subroutines(BuildCtx *ctx) +{ + |.code_sub + | + |//----------------------------------------------------------------------- + |//-- Return handling ---------------------------------------------------- + |//----------------------------------------------------------------------- + | + |->vm_returnp: + | // See vm_return. Also: TMP2 = previous base. + | andi AT, PC, FRAME_P + | beqz AT, ->cont_dispatch + | + | // Return from pcall or xpcall fast func. + |. mov_true TMP1 + | ld PC, FRAME_PC(TMP2) // Fetch PC of previous frame. + | move BASE, TMP2 // Restore caller base. + | // Prepending may overwrite the pcall frame, so do it at the end. + | sd TMP1, -8(RA) // Prepend true to results. + | daddiu RA, RA, -8 + | + |->vm_returnc: + | addiu RD, RD, 8 // RD = (nresults+1)*8. + | andi TMP0, PC, FRAME_TYPE + | beqz RD, ->vm_unwind_c_eh + |. li CRET1, LUA_YIELD + | beqz TMP0, ->BC_RET_Z // Handle regular return to Lua. + |. move MULTRES, RD + | + |->vm_return: + | // BASE = base, RA = resultptr, RD/MULTRES = (nresults+1)*8, PC = return + | // TMP0 = PC & FRAME_TYPE + | li TMP2, -8 + | xori AT, TMP0, FRAME_C + | and TMP2, PC, TMP2 + | bnez AT, ->vm_returnp + | dsubu TMP2, BASE, TMP2 // TMP2 = previous base. + | + | addiu TMP1, RD, -8 + | sd TMP2, L->base + | li_vmstate C + | lw TMP2, SAVE_NRES + | daddiu BASE, BASE, -16 + | st_vmstate + | beqz TMP1, >2 + |. sll TMP2, TMP2, 3 + |1: + | addiu TMP1, TMP1, -8 + | ld CRET1, 0(RA) + | daddiu RA, RA, 8 + | sd CRET1, 0(BASE) + | bnez TMP1, <1 + |. daddiu BASE, BASE, 8 + | + |2: + | bne TMP2, RD, >6 + |3: + |. sd BASE, L->top // Store new top. + | + |->vm_leave_cp: + | ld TMP0, SAVE_CFRAME // Restore previous C frame. + | move CRET1, r0 // Ok return status for vm_pcall. + | sd TMP0, L->cframe + | + |->vm_leave_unw: + | restoreregs_ret + | + |6: + | ld TMP1, L->maxstack + | slt AT, TMP2, RD + | bnez AT, >7 // Less results wanted? + | // More results wanted. Check stack size and fill up results with nil. + |. slt AT, BASE, TMP1 + | beqz AT, >8 + |. nop + | sd TISNIL, 0(BASE) + | addiu RD, RD, 8 + | b <2 + |. daddiu BASE, BASE, 8 + | + |7: // Less results wanted. + | subu TMP0, RD, TMP2 + | dsubu TMP0, BASE, TMP0 // Either keep top or shrink it. + |.if MIPSR6 + | selnez TMP0, TMP0, TMP2 // LUA_MULTRET+1 case? + | seleqz BASE, BASE, TMP2 + | b <3 + |. or BASE, BASE, TMP0 + |.else + | b <3 + |. movn BASE, TMP0, TMP2 // LUA_MULTRET+1 case? + |.endif + | + |8: // Corner case: need to grow stack for filling up results. + | // This can happen if: + | // - A C function grows the stack (a lot). + | // - The GC shrinks the stack in between. + | // - A return back from a lua_call() with (high) nresults adjustment. + | load_got lj_state_growstack + | move MULTRES, RD + | srl CARG2, TMP2, 3 + | call_intern lj_state_growstack // (lua_State *L, int n) + |. move CARG1, L + | lw TMP2, SAVE_NRES + | ld BASE, L->top // Need the (realloced) L->top in BASE. + | move RD, MULTRES + | b <2 + |. sll TMP2, TMP2, 3 + | + |->vm_unwind_c: // Unwind C stack, return from vm_pcall. + | // (void *cframe, int errcode) + | move sp, CARG1 + | move CRET1, CARG2 + |->vm_unwind_c_eh: // Landing pad for external unwinder. + | ld L, SAVE_L + | li TMP0, ~LJ_VMST_C + | ld GL:TMP1, L->glref + | b ->vm_leave_unw + |. sw TMP0, GL:TMP1->vmstate + | + |->vm_unwind_ff: // Unwind C stack, return from ff pcall. + | // (void *cframe) + | li AT, -4 + | and sp, CARG1, AT + |->vm_unwind_ff_eh: // Landing pad for external unwinder. + | ld L, SAVE_L + | .FPU lui TMP3, 0x59c0 // TOBIT = 2^52 + 2^51 (float). + | li TISNIL, LJ_TNIL + | li TISNUM, LJ_TISNUM + | ld BASE, L->base + | ld DISPATCH, L->glref // Setup pointer to dispatch table. + | .FPU mtc1 TMP3, TOBIT + | mov_false TMP1 + | li_vmstate INTERP + | ld PC, FRAME_PC(BASE) // Fetch PC of previous frame. + | .FPU cvt.d.s TOBIT, TOBIT + | daddiu RA, BASE, -8 // Results start at BASE-8. + | daddiu DISPATCH, DISPATCH, GG_G2DISP + | sd TMP1, 0(RA) // Prepend false to error message. + | st_vmstate + | b ->vm_returnc + |. li RD, 16 // 2 results: false + error message. + | + |->vm_unwind_stub: // Jump to exit stub from unwinder. + | jr CARG1 + |. move ra, CARG2 + | + |//----------------------------------------------------------------------- + |//-- Grow stack for calls ----------------------------------------------- + |//----------------------------------------------------------------------- + | + |->vm_growstack_c: // Grow stack for C function. + | b >2 + |. li CARG2, LUA_MINSTACK + | + |->vm_growstack_l: // Grow stack for Lua function. + | // BASE = new base, RA = BASE+framesize*8, RC = nargs*8, PC = first PC + | daddu RC, BASE, RC + | dsubu RA, RA, BASE + | sd BASE, L->base + | daddiu PC, PC, 4 // Must point after first instruction. + | sd RC, L->top + | srl CARG2, RA, 3 + |2: + | // L->base = new base, L->top = top + | load_got lj_state_growstack + | sd PC, SAVE_PC + | call_intern lj_state_growstack // (lua_State *L, int n) + |. move CARG1, L + | ld BASE, L->base + | ld RC, L->top + | ld LFUNC:RB, FRAME_FUNC(BASE) + | dsubu RC, RC, BASE + | cleartp LFUNC:RB + | // BASE = new base, RB = LFUNC/CFUNC, RC = nargs*8, FRAME_PC(BASE) = PC + | ins_callt // Just retry the call. + | + |//----------------------------------------------------------------------- + |//-- Entry points into the assembler VM --------------------------------- + |//----------------------------------------------------------------------- + | + |->vm_resume: // Setup C frame and resume thread. + | // (lua_State *L, TValue *base, int nres1 = 0, ptrdiff_t ef = 0) + | saveregs + | move L, CARG1 + | ld DISPATCH, L->glref // Setup pointer to dispatch table. + | move BASE, CARG2 + | lbu TMP1, L->status + | sd L, SAVE_L + | li PC, FRAME_CP + | daddiu TMP0, sp, CFRAME_RESUME + | daddiu DISPATCH, DISPATCH, GG_G2DISP + | sw r0, SAVE_NRES + | sw r0, SAVE_ERRF + | sd CARG1, SAVE_PC // Any value outside of bytecode is ok. + | sd r0, SAVE_CFRAME + | beqz TMP1, >3 + |. sd TMP0, L->cframe + | + | // Resume after yield (like a return). + | sd L, DISPATCH_GL(cur_L)(DISPATCH) + | move RA, BASE + | ld BASE, L->base + | ld TMP1, L->top + | ld PC, FRAME_PC(BASE) + | .FPU lui TMP3, 0x59c0 // TOBIT = 2^52 + 2^51 (float). + | dsubu RD, TMP1, BASE + | .FPU mtc1 TMP3, TOBIT + | sb r0, L->status + | .FPU cvt.d.s TOBIT, TOBIT + | li_vmstate INTERP + | daddiu RD, RD, 8 + | st_vmstate + | move MULTRES, RD + | andi TMP0, PC, FRAME_TYPE + | li TISNIL, LJ_TNIL + | beqz TMP0, ->BC_RET_Z + |. li TISNUM, LJ_TISNUM + | b ->vm_return + |. nop + | + |->vm_pcall: // Setup protected C frame and enter VM. + | // (lua_State *L, TValue *base, int nres1, ptrdiff_t ef) + | saveregs + | sw CARG4, SAVE_ERRF + | b >1 + |. li PC, FRAME_CP + | + |->vm_call: // Setup C frame and enter VM. + | // (lua_State *L, TValue *base, int nres1) + | saveregs + | li PC, FRAME_C + | + |1: // Entry point for vm_pcall above (PC = ftype). + | ld TMP1, L:CARG1->cframe + | move L, CARG1 + | sw CARG3, SAVE_NRES + | ld DISPATCH, L->glref // Setup pointer to dispatch table. + | sd CARG1, SAVE_L + | move BASE, CARG2 + | daddiu DISPATCH, DISPATCH, GG_G2DISP + | sd CARG1, SAVE_PC // Any value outside of bytecode is ok. + | sd TMP1, SAVE_CFRAME + | sd sp, L->cframe // Add our C frame to cframe chain. + | + |3: // Entry point for vm_cpcall/vm_resume (BASE = base, PC = ftype). + | sd L, DISPATCH_GL(cur_L)(DISPATCH) + | ld TMP2, L->base // TMP2 = old base (used in vmeta_call). + | .FPU lui TMP3, 0x59c0 // TOBIT = 2^52 + 2^51 (float). + | ld TMP1, L->top + | .FPU mtc1 TMP3, TOBIT + | daddu PC, PC, BASE + | dsubu NARGS8:RC, TMP1, BASE + | li TISNUM, LJ_TISNUM + | dsubu PC, PC, TMP2 // PC = frame delta + frame type + | .FPU cvt.d.s TOBIT, TOBIT + | li_vmstate INTERP + | li TISNIL, LJ_TNIL + | st_vmstate + | + |->vm_call_dispatch: + | // TMP2 = old base, BASE = new base, RC = nargs*8, PC = caller PC + | ld LFUNC:RB, FRAME_FUNC(BASE) + | checkfunc LFUNC:RB, ->vmeta_call + | + |->vm_call_dispatch_f: + | ins_call + | // BASE = new base, RB = func, RC = nargs*8, PC = caller PC + | + |->vm_cpcall: // Setup protected C frame, call C. + | // (lua_State *L, lua_CFunction func, void *ud, lua_CPFunction cp) + | saveregs + | move L, CARG1 + | ld TMP0, L:CARG1->stack + | sd CARG1, SAVE_L + | ld TMP1, L->top + | ld DISPATCH, L->glref // Setup pointer to dispatch table. + | sd CARG1, SAVE_PC // Any value outside of bytecode is ok. + | dsubu TMP0, TMP0, TMP1 // Compute -savestack(L, L->top). + | ld TMP1, L->cframe + | daddiu DISPATCH, DISPATCH, GG_G2DISP + | sw TMP0, SAVE_NRES // Neg. delta means cframe w/o frame. + | sw r0, SAVE_ERRF // No error function. + | sd TMP1, SAVE_CFRAME + | sd sp, L->cframe // Add our C frame to cframe chain. + | sd L, DISPATCH_GL(cur_L)(DISPATCH) + | jalr CARG4 // (lua_State *L, lua_CFunction func, void *ud) + |. move CFUNCADDR, CARG4 + | move BASE, CRET1 + | bnez CRET1, <3 // Else continue with the call. + |. li PC, FRAME_CP + | b ->vm_leave_cp // No base? Just remove C frame. + |. nop + | + |//----------------------------------------------------------------------- + |//-- Metamethod handling ------------------------------------------------ + |//----------------------------------------------------------------------- + | + |// The lj_meta_* functions (except for lj_meta_cat) don't reallocate the + |// stack, so BASE doesn't need to be reloaded across these calls. + | + |//-- Continuation dispatch ---------------------------------------------- + | + |->cont_dispatch: + | // BASE = meta base, RA = resultptr, RD = (nresults+1)*8 + | ld TMP0, -32(BASE) // Continuation. + | move RB, BASE + | move BASE, TMP2 // Restore caller BASE. + | ld LFUNC:TMP1, FRAME_FUNC(TMP2) + |.if FFI + | sltiu AT, TMP0, 2 + |.endif + | ld PC, -24(RB) // Restore PC from [cont|PC]. + | cleartp LFUNC:TMP1 + | daddu TMP2, RA, RD + |.if FFI + | bnez AT, >1 + |.endif + |. sd TISNIL, -8(TMP2) // Ensure one valid arg. + | ld TMP1, LFUNC:TMP1->pc + | // BASE = base, RA = resultptr, RB = meta base + | jr TMP0 // Jump to continuation. + |. ld KBASE, PC2PROTO(k)(TMP1) + | + |.if FFI + |1: + | bnez TMP0, ->cont_ffi_callback // cont = 1: return from FFI callback. + | // cont = 0: tailcall from C function. + |. daddiu TMP1, RB, -32 + | b ->vm_call_tail + |. dsubu RC, TMP1, BASE + |.endif + | + |->cont_cat: // RA = resultptr, RB = meta base + | lw INS, -4(PC) + | daddiu CARG2, RB, -32 + | ld CRET1, 0(RA) + | decode_RB8a MULTRES, INS + | decode_RA8a RA, INS + | decode_RB8b MULTRES + | decode_RA8b RA + | daddu TMP1, BASE, MULTRES + | sd BASE, L->base + | dsubu CARG3, CARG2, TMP1 + | bne TMP1, CARG2, ->BC_CAT_Z + |. sd CRET1, 0(CARG2) + | daddu RA, BASE, RA + | b ->cont_nop + |. sd CRET1, 0(RA) + | + |//-- Table indexing metamethods ----------------------------------------- + | + |->vmeta_tgets1: + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv) + | li TMP0, LJ_TSTR + | settp STR:RC, TMP0 + | b >1 + |. sd STR:RC, 0(CARG3) + | + |->vmeta_tgets: + | daddiu CARG2, DISPATCH, DISPATCH_GL(tmptv) + | li TMP0, LJ_TTAB + | li TMP1, LJ_TSTR + | settp TAB:RB, TMP0 + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv2) + | sd TAB:RB, 0(CARG2) + | settp STR:RC, TMP1 + | b >1 + |. sd STR:RC, 0(CARG3) + | + |->vmeta_tgetb: // TMP0 = index + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv) + | settp TMP0, TISNUM + | sd TMP0, 0(CARG3) + | + |->vmeta_tgetv: + |1: + | load_got lj_meta_tget + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_meta_tget // (lua_State *L, TValue *o, TValue *k) + |. move CARG1, L + | // Returns TValue * (finished) or NULL (metamethod). + | beqz CRET1, >3 + |. daddiu TMP1, BASE, -FRAME_CONT + | ld CARG1, 0(CRET1) + | ins_next1 + | sd CARG1, 0(RA) + | ins_next2 + | + |3: // Call __index metamethod. + | // BASE = base, L->top = new base, stack = cont/func/t/k + | ld BASE, L->top + | sd PC, -24(BASE) // [cont|PC] + | dsubu PC, BASE, TMP1 + | ld LFUNC:RB, FRAME_FUNC(BASE) // Guaranteed to be a function here. + | cleartp LFUNC:RB + | b ->vm_call_dispatch_f + |. li NARGS8:RC, 16 // 2 args for func(t, k). + | + |->vmeta_tgetr: + | load_got lj_tab_getinth + | call_intern lj_tab_getinth // (GCtab *t, int32_t key) + |. nop + | // Returns cTValue * or NULL. + | beqz CRET1, ->BC_TGETR_Z + |. move CARG2, TISNIL + | b ->BC_TGETR_Z + |. ld CARG2, 0(CRET1) + | + |//----------------------------------------------------------------------- + | + |->vmeta_tsets1: + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv) + | li TMP0, LJ_TSTR + | settp STR:RC, TMP0 + | b >1 + |. sd STR:RC, 0(CARG3) + | + |->vmeta_tsets: + | daddiu CARG2, DISPATCH, DISPATCH_GL(tmptv) + | li TMP0, LJ_TTAB + | li TMP1, LJ_TSTR + | settp TAB:RB, TMP0 + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv2) + | sd TAB:RB, 0(CARG2) + | settp STR:RC, TMP1 + | b >1 + |. sd STR:RC, 0(CARG3) + | + |->vmeta_tsetb: // TMP0 = index + | daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv) + | settp TMP0, TISNUM + | sd TMP0, 0(CARG3) + | + |->vmeta_tsetv: + |1: + | load_got lj_meta_tset + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_meta_tset // (lua_State *L, TValue *o, TValue *k) + |. move CARG1, L + | // Returns TValue * (finished) or NULL (metamethod). + | beqz CRET1, >3 + |. ld CARG1, 0(RA) + | // NOBARRIER: lj_meta_tset ensures the table is not black. + | ins_next1 + | sd CARG1, 0(CRET1) + | ins_next2 + | + |3: // Call __newindex metamethod. + | // BASE = base, L->top = new base, stack = cont/func/t/k/(v) + | daddiu TMP1, BASE, -FRAME_CONT + | ld BASE, L->top + | sd PC, -24(BASE) // [cont|PC] + | dsubu PC, BASE, TMP1 + | ld LFUNC:RB, FRAME_FUNC(BASE) // Guaranteed to be a function here. + | cleartp LFUNC:RB + | sd CARG1, 16(BASE) // Copy value to third argument. + | b ->vm_call_dispatch_f + |. li NARGS8:RC, 24 // 3 args for func(t, k, v) + | + |->vmeta_tsetr: + | load_got lj_tab_setinth + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_tab_setinth // (lua_State *L, GCtab *t, int32_t key) + |. move CARG1, L + | // Returns TValue *. + | b ->BC_TSETR_Z + |. nop + | + |//-- Comparison metamethods --------------------------------------------- + | + |->vmeta_comp: + | // RA/RD point to o1/o2. + | move CARG2, RA + | move CARG3, RD + | load_got lj_meta_comp + | daddiu PC, PC, -4 + | sd BASE, L->base + | sd PC, SAVE_PC + | decode_OP1 CARG4, INS + | call_intern lj_meta_comp // (lua_State *L, TValue *o1, *o2, int op) + |. move CARG1, L + | // Returns 0/1 or TValue * (metamethod). + |3: + | sltiu AT, CRET1, 2 + | beqz AT, ->vmeta_binop + | negu TMP2, CRET1 + |4: + | lhu RD, OFS_RD(PC) + | daddiu PC, PC, 4 + | lui TMP1, (-(BCBIAS_J*4 >> 16) & 65535) + | sll RD, RD, 2 + | addu RD, RD, TMP1 + | and RD, RD, TMP2 + | daddu PC, PC, RD + |->cont_nop: + | ins_next + | + |->cont_ra: // RA = resultptr + | lbu TMP1, -4+OFS_RA(PC) + | ld CRET1, 0(RA) + | sll TMP1, TMP1, 3 + | daddu TMP1, BASE, TMP1 + | b ->cont_nop + |. sd CRET1, 0(TMP1) + | + |->cont_condt: // RA = resultptr + | ld TMP0, 0(RA) + | gettp TMP0, TMP0 + | sltiu AT, TMP0, LJ_TISTRUECOND + | b <4 + |. negu TMP2, AT // Branch if result is true. + | + |->cont_condf: // RA = resultptr + | ld TMP0, 0(RA) + | gettp TMP0, TMP0 + | sltiu AT, TMP0, LJ_TISTRUECOND + | b <4 + |. addiu TMP2, AT, -1 // Branch if result is false. + | + |->vmeta_equal: + | // CARG1/CARG2 point to o1/o2. TMP0 is set to 0/1. + | load_got lj_meta_equal + | cleartp LFUNC:CARG3, CARG2 + | cleartp LFUNC:CARG2, CARG1 + | move CARG4, TMP0 + | daddiu PC, PC, -4 + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_meta_equal // (lua_State *L, GCobj *o1, *o2, int ne) + |. move CARG1, L + | // Returns 0/1 or TValue * (metamethod). + | b <3 + |. nop + | + |->vmeta_equal_cd: + |.if FFI + | load_got lj_meta_equal_cd + | move CARG2, INS + | daddiu PC, PC, -4 + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_meta_equal_cd // (lua_State *L, BCIns op) + |. move CARG1, L + | // Returns 0/1 or TValue * (metamethod). + | b <3 + |. nop + |.endif + | + |->vmeta_istype: + | load_got lj_meta_istype + | daddiu PC, PC, -4 + | sd BASE, L->base + | srl CARG2, RA, 3 + | srl CARG3, RD, 3 + | sd PC, SAVE_PC + | call_intern lj_meta_istype // (lua_State *L, BCReg ra, BCReg tp) + |. move CARG1, L + | b ->cont_nop + |. nop + | + |//-- Arithmetic metamethods --------------------------------------------- + | + |->vmeta_unm: + | move RC, RB + | + |->vmeta_arith: + | load_got lj_meta_arith + | sd BASE, L->base + | move CARG2, RA + | sd PC, SAVE_PC + | move CARG3, RB + | move CARG4, RC + | decode_OP1 CARG5, INS // CARG5 == RB. + | call_intern lj_meta_arith // (lua_State *L, TValue *ra,*rb,*rc, BCReg op) + |. move CARG1, L + | // Returns NULL (finished) or TValue * (metamethod). + | beqz CRET1, ->cont_nop + |. nop + | + | // Call metamethod for binary op. + |->vmeta_binop: + | // BASE = old base, CRET1 = new base, stack = cont/func/o1/o2 + | dsubu TMP1, CRET1, BASE + | sd PC, -24(CRET1) // [cont|PC] + | move TMP2, BASE + | daddiu PC, TMP1, FRAME_CONT + | move BASE, CRET1 + | b ->vm_call_dispatch + |. li NARGS8:RC, 16 // 2 args for func(o1, o2). + | + |->vmeta_len: + | // CARG2 already set by BC_LEN. +#if LJ_52 + | move MULTRES, CARG1 +#endif + | load_got lj_meta_len + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_meta_len // (lua_State *L, TValue *o) + |. move CARG1, L + | // Returns NULL (retry) or TValue * (metamethod base). +#if LJ_52 + | bnez CRET1, ->vmeta_binop // Binop call for compatibility. + |. nop + | b ->BC_LEN_Z + |. move CARG1, MULTRES +#else + | b ->vmeta_binop // Binop call for compatibility. + |. nop +#endif + | + |//-- Call metamethod ---------------------------------------------------- + | + |->vmeta_call: // Resolve and call __call metamethod. + | // TMP2 = old base, BASE = new base, RC = nargs*8 + | load_got lj_meta_call + | sd TMP2, L->base // This is the callers base! + | daddiu CARG2, BASE, -16 + | sd PC, SAVE_PC + | daddu CARG3, BASE, RC + | move MULTRES, NARGS8:RC + | call_intern lj_meta_call // (lua_State *L, TValue *func, TValue *top) + |. move CARG1, L + | ld LFUNC:RB, FRAME_FUNC(BASE) // Guaranteed to be a function here. + | daddiu NARGS8:RC, MULTRES, 8 // Got one more argument now. + | cleartp LFUNC:RB + | ins_call + | + |->vmeta_callt: // Resolve __call for BC_CALLT. + | // BASE = old base, RA = new base, RC = nargs*8 + | load_got lj_meta_call + | sd BASE, L->base + | daddiu CARG2, RA, -16 + | sd PC, SAVE_PC + | daddu CARG3, RA, RC + | move MULTRES, NARGS8:RC + | call_intern lj_meta_call // (lua_State *L, TValue *func, TValue *top) + |. move CARG1, L + | ld RB, FRAME_FUNC(RA) // Guaranteed to be a function here. + | ld TMP1, FRAME_PC(BASE) + | daddiu NARGS8:RC, MULTRES, 8 // Got one more argument now. + | b ->BC_CALLT_Z + |. cleartp LFUNC:CARG3, RB + | + |//-- Argument coercion for 'for' statement ------------------------------ + | + |->vmeta_for: + | load_got lj_meta_for + | sd BASE, L->base + | move CARG2, RA + | sd PC, SAVE_PC + | move MULTRES, INS + | call_intern lj_meta_for // (lua_State *L, TValue *base) + |. move CARG1, L + |.if JIT + | decode_OP1 TMP0, MULTRES + | li AT, BC_JFORI + |.endif + | decode_RA8a RA, MULTRES + | decode_RD8a RD, MULTRES + | decode_RA8b RA + |.if JIT + | beq TMP0, AT, =>BC_JFORI + |. decode_RD8b RD + | b =>BC_FORI + |. nop + |.else + | b =>BC_FORI + |. decode_RD8b RD + |.endif + | + |//----------------------------------------------------------------------- + |//-- Fast functions ----------------------------------------------------- + |//----------------------------------------------------------------------- + | + |.macro .ffunc, name + |->ff_ .. name: + |.endmacro + | + |.macro .ffunc_1, name + |->ff_ .. name: + | beqz NARGS8:RC, ->fff_fallback + |. ld CARG1, 0(BASE) + |.endmacro + | + |.macro .ffunc_2, name + |->ff_ .. name: + | sltiu AT, NARGS8:RC, 16 + | ld CARG1, 0(BASE) + | bnez AT, ->fff_fallback + |. ld CARG2, 8(BASE) + |.endmacro + | + |.macro .ffunc_n, name // Caveat: has delay slot! + |->ff_ .. name: + | ld CARG1, 0(BASE) + | beqz NARGS8:RC, ->fff_fallback + | // Either ldc1 or the 1st instruction of checknum is in the delay slot. + | .FPU ldc1 FARG1, 0(BASE) + | checknum CARG1, ->fff_fallback + |.endmacro + | + |.macro .ffunc_nn, name // Caveat: has delay slot! + |->ff_ .. name: + | ld CARG1, 0(BASE) + | sltiu AT, NARGS8:RC, 16 + | ld CARG2, 8(BASE) + | bnez AT, ->fff_fallback + |. gettp TMP0, CARG1 + | gettp TMP1, CARG2 + | sltiu TMP0, TMP0, LJ_TISNUM + | sltiu TMP1, TMP1, LJ_TISNUM + | .FPU ldc1 FARG1, 0(BASE) + | and TMP0, TMP0, TMP1 + | .FPU ldc1 FARG2, 8(BASE) + | beqz TMP0, ->fff_fallback + |.endmacro + | + |// Inlined GC threshold check. Caveat: uses TMP0 and TMP1 and has delay slot! + |// MIPSR6: no delay slot, but a forbidden slot. + |.macro ffgccheck + | ld TMP0, DISPATCH_GL(gc.total)(DISPATCH) + | ld TMP1, DISPATCH_GL(gc.threshold)(DISPATCH) + | dsubu AT, TMP0, TMP1 + |.if MIPSR6 + | bgezalc AT, ->fff_gcstep + |.else + | bgezal AT, ->fff_gcstep + |.endif + |.endmacro + | + |//-- Base library: checks ----------------------------------------------- + |.ffunc_1 assert + | gettp AT, CARG1 + | sltiu AT, AT, LJ_TISTRUECOND + | beqz AT, ->fff_fallback + |. daddiu RA, BASE, -16 + | ld PC, FRAME_PC(BASE) + | addiu RD, NARGS8:RC, 8 // Compute (nresults+1)*8. + | daddu TMP2, RA, RD + | daddiu TMP1, BASE, 8 + | beq BASE, TMP2, ->fff_res // Done if exactly 1 argument. + |. sd CARG1, 0(RA) + |1: + | ld CRET1, 0(TMP1) + | sd CRET1, -16(TMP1) + | bne TMP1, TMP2, <1 + |. daddiu TMP1, TMP1, 8 + | b ->fff_res + |. nop + | + |.ffunc_1 type + | gettp TMP0, CARG1 + | sltu TMP1, TISNUM, TMP0 + | not TMP2, TMP0 + | li TMP3, ~LJ_TISNUM + |.if MIPSR6 + | selnez TMP2, TMP2, TMP1 + | seleqz TMP3, TMP3, TMP1 + | or TMP2, TMP2, TMP3 + |.else + | movz TMP2, TMP3, TMP1 + |.endif + | dsll TMP2, TMP2, 3 + | daddu TMP2, CFUNC:RB, TMP2 + | b ->fff_restv + |. ld CARG1, CFUNC:TMP2->upvalue + | + |//-- Base library: getters and setters --------------------------------- + | + |.ffunc_1 getmetatable + | gettp TMP2, CARG1 + | daddiu TMP0, TMP2, -LJ_TTAB + | daddiu TMP1, TMP2, -LJ_TUDATA + |.if MIPSR6 + | selnez TMP0, TMP1, TMP0 + |.else + | movn TMP0, TMP1, TMP0 + |.endif + | bnez TMP0, >6 + |. cleartp TAB:CARG1 + |1: // Field metatable must be at same offset for GCtab and GCudata! + | ld TAB:RB, TAB:CARG1->metatable + |2: + | ld STR:RC, DISPATCH_GL(gcroot[GCROOT_MMNAME+MM_metatable])(DISPATCH) + | beqz TAB:RB, ->fff_restv + |. li CARG1, LJ_TNIL + | lw TMP0, TAB:RB->hmask + | lw TMP1, STR:RC->sid + | ld NODE:TMP2, TAB:RB->node + | and TMP1, TMP1, TMP0 // idx = str->sid & tab->hmask + | dsll TMP0, TMP1, 5 + | dsll TMP1, TMP1, 3 + | dsubu TMP1, TMP0, TMP1 + | daddu NODE:TMP2, NODE:TMP2, TMP1 // node = tab->node + (idx*32-idx*8) + | li CARG4, LJ_TSTR + | settp STR:RC, CARG4 // Tagged key to look for. + |3: // Rearranged logic, because we expect _not_ to find the key. + | ld TMP0, NODE:TMP2->key + | ld CARG1, NODE:TMP2->val + | ld NODE:TMP2, NODE:TMP2->next + | beq RC, TMP0, >5 + |. li AT, LJ_TTAB + | bnez NODE:TMP2, <3 + |. nop + |4: + | move CARG1, RB + | b ->fff_restv // Not found, keep default result. + |. settp CARG1, AT + |5: + | bne CARG1, TISNIL, ->fff_restv + |. nop + | b <4 // Ditto for nil value. + |. nop + | + |6: + | sltiu AT, TMP2, LJ_TISNUM + |.if MIPSR6 + | selnez TMP0, TISNUM, AT + | seleqz AT, TMP2, AT + | or TMP2, TMP0, AT + |.else + | movn TMP2, TISNUM, AT + |.endif + | dsll TMP2, TMP2, 3 + | dsubu TMP0, DISPATCH, TMP2 + | b <2 + |. ld TAB:RB, DISPATCH_GL(gcroot[GCROOT_BASEMT])-8(TMP0) + | + |.ffunc_2 setmetatable + | // Fast path: no mt for table yet and not clearing the mt. + | checktp TMP1, CARG1, -LJ_TTAB, ->fff_fallback + | gettp TMP3, CARG2 + | ld TAB:TMP0, TAB:TMP1->metatable + | lbu TMP2, TAB:TMP1->marked + | daddiu AT, TMP3, -LJ_TTAB + | cleartp TAB:CARG2 + | or AT, AT, TAB:TMP0 + | bnez AT, ->fff_fallback + |. andi AT, TMP2, LJ_GC_BLACK // isblack(table) + | beqz AT, ->fff_restv + |. sd TAB:CARG2, TAB:TMP1->metatable + | barrierback TAB:TMP1, TMP2, TMP0, ->fff_restv + | + |.ffunc rawget + | ld CARG2, 0(BASE) + | sltiu AT, NARGS8:RC, 16 + | load_got lj_tab_get + | gettp TMP0, CARG2 + | cleartp CARG2 + | daddiu TMP0, TMP0, -LJ_TTAB + | or AT, AT, TMP0 + | bnez AT, ->fff_fallback + |. daddiu CARG3, BASE, 8 + | call_intern lj_tab_get // (lua_State *L, GCtab *t, cTValue *key) + |. move CARG1, L + | b ->fff_restv + |. ld CARG1, 0(CRET1) + | + |//-- Base library: conversions ------------------------------------------ + | + |.ffunc tonumber + | // Only handles the number case inline (without a base argument). + | ld CARG1, 0(BASE) + | xori AT, NARGS8:RC, 8 // Exactly one number argument. + | gettp TMP1, CARG1 + | sltu TMP0, TISNUM, TMP1 + | or AT, AT, TMP0 + | bnez AT, ->fff_fallback + |. nop + | b ->fff_restv + |. nop + | + |.ffunc_1 tostring + | // Only handles the string or number case inline. + | gettp TMP0, CARG1 + | daddiu AT, TMP0, -LJ_TSTR + | // A __tostring method in the string base metatable is ignored. + | beqz AT, ->fff_restv // String key? + | // Handle numbers inline, unless a number base metatable is present. + |. ld TMP1, DISPATCH_GL(gcroot[GCROOT_BASEMT_NUM])(DISPATCH) + | sltu TMP0, TISNUM, TMP0 + | or TMP0, TMP0, TMP1 + | bnez TMP0, ->fff_fallback + |. sd BASE, L->base // Add frame since C call can throw. + |.if MIPSR6 + | sd PC, SAVE_PC // Redundant (but a defined value). + | ffgccheck + |.else + | ffgccheck + |. sd PC, SAVE_PC // Redundant (but a defined value). + |.endif + | load_got lj_strfmt_number + | move CARG1, L + | call_intern lj_strfmt_number // (lua_State *L, cTValue *o) + |. move CARG2, BASE + | // Returns GCstr *. + | li AT, LJ_TSTR + | settp CRET1, AT + | b ->fff_restv + |. move CARG1, CRET1 + | + |//-- Base library: iterators ------------------------------------------- + | + |.ffunc_1 next + | checktp CARG1, -LJ_TTAB, ->fff_fallback + | daddu TMP2, BASE, NARGS8:RC + | sd TISNIL, 0(TMP2) // Set missing 2nd arg to nil. + | load_got lj_tab_next + | ld PC, FRAME_PC(BASE) + | daddiu CARG2, BASE, 8 + | call_intern lj_tab_next // (GCtab *t, cTValue *key, TValue *o) + |. daddiu CARG3, BASE, -16 + | // Returns 1=found, 0=end, -1=error. + | daddiu RA, BASE, -16 + | bgtz CRET1, ->fff_res // Found key/value. + |. li RD, (2+1)*8 + | beqz CRET1, ->fff_restv // End of traversal: return nil. + |. move CARG1, TISNIL + | ld CFUNC:RB, FRAME_FUNC(BASE) + | cleartp CFUNC:RB + | b ->fff_fallback // Invalid key. + |. li RC, 2*8 + | + |.ffunc_1 pairs + | checktp TAB:TMP1, CARG1, -LJ_TTAB, ->fff_fallback + | ld PC, FRAME_PC(BASE) +#if LJ_52 + | ld TAB:TMP2, TAB:TMP1->metatable + | ld TMP0, CFUNC:RB->upvalue[0] + | bnez TAB:TMP2, ->fff_fallback +#else + | ld TMP0, CFUNC:RB->upvalue[0] +#endif + |. daddiu RA, BASE, -16 + | sd TISNIL, 0(BASE) + | sd CARG1, -8(BASE) + | sd TMP0, 0(RA) + | b ->fff_res + |. li RD, (3+1)*8 + | + |.ffunc_2 ipairs_aux + | checktab CARG1, ->fff_fallback + | checkint CARG2, ->fff_fallback + |. lw TMP0, TAB:CARG1->asize + | ld TMP1, TAB:CARG1->array + | ld PC, FRAME_PC(BASE) + | sextw TMP2, CARG2 + | addiu TMP2, TMP2, 1 + | sltu AT, TMP2, TMP0 + | daddiu RA, BASE, -16 + | zextw TMP0, TMP2 + | settp TMP0, TISNUM + | beqz AT, >2 // Not in array part? + |. sd TMP0, 0(RA) + | dsll TMP3, TMP2, 3 + | daddu TMP3, TMP1, TMP3 + | ld TMP1, 0(TMP3) + |1: + | beq TMP1, TISNIL, ->fff_res // End of iteration, return 0 results. + |. li RD, (0+1)*8 + | sd TMP1, -8(BASE) + | b ->fff_res + |. li RD, (2+1)*8 + |2: // Check for empty hash part first. Otherwise call C function. + | lw TMP0, TAB:CARG1->hmask + | load_got lj_tab_getinth + | beqz TMP0, ->fff_res + |. li RD, (0+1)*8 + | call_intern lj_tab_getinth // (GCtab *t, int32_t key) + |. move CARG2, TMP2 + | // Returns cTValue * or NULL. + | beqz CRET1, ->fff_res + |. li RD, (0+1)*8 + | b <1 + |. ld TMP1, 0(CRET1) + | + |.ffunc_1 ipairs + | checktp TAB:TMP1, CARG1, -LJ_TTAB, ->fff_fallback + | ld PC, FRAME_PC(BASE) +#if LJ_52 + | ld TAB:TMP2, TAB:TMP1->metatable + | ld CFUNC:TMP0, CFUNC:RB->upvalue[0] + | bnez TAB:TMP2, ->fff_fallback +#else + | ld TMP0, CFUNC:RB->upvalue[0] +#endif + | daddiu RA, BASE, -16 + | dsll AT, TISNUM, 47 + | sd CARG1, -8(BASE) + | sd AT, 0(BASE) + | sd CFUNC:TMP0, 0(RA) + | b ->fff_res + |. li RD, (3+1)*8 + | + |//-- Base library: catch errors ---------------------------------------- + | + |.ffunc pcall + | ld TMP1, L->maxstack + | daddu TMP2, BASE, NARGS8:RC + | sltu AT, TMP1, TMP2 + | bnez AT, ->fff_fallback + |. lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH) + | daddiu NARGS8:TMP0, NARGS8:RC, -8 + | bltz NARGS8:TMP0, ->fff_fallback + |. move TMP2, BASE + | move NARGS8:RC, NARGS8:TMP0 + | daddiu BASE, BASE, 16 + | // Remember active hook before pcall. + | srl TMP3, TMP3, HOOK_ACTIVE_SHIFT + | andi TMP3, TMP3, 1 + | daddiu PC, TMP3, 16+FRAME_PCALL + | beqz NARGS8:RC, ->vm_call_dispatch + |1: + |. daddu TMP0, BASE, NARGS8:RC + |2: + | ld TMP1, -16(TMP0) + | sd TMP1, -8(TMP0) + | daddiu TMP0, TMP0, -8 + | bne TMP0, BASE, <2 + |. nop + | b ->vm_call_dispatch + |. nop + | + |.ffunc xpcall + | ld TMP1, L->maxstack + | daddu TMP2, BASE, NARGS8:RC + | sltu AT, TMP1, TMP2 + | bnez AT, ->fff_fallback + |. ld CARG1, 0(BASE) + | daddiu NARGS8:TMP0, NARGS8:RC, -16 + | ld CARG2, 8(BASE) + | bltz NARGS8:TMP0, ->fff_fallback + |. lbu TMP1, DISPATCH_GL(hookmask)(DISPATCH) + | gettp AT, CARG2 + | daddiu AT, AT, -LJ_TFUNC + | bnez AT, ->fff_fallback // Traceback must be a function. + |. move TMP2, BASE + | move NARGS8:RC, NARGS8:TMP0 + | daddiu BASE, BASE, 24 + | // Remember active hook before pcall. + | srl TMP3, TMP3, HOOK_ACTIVE_SHIFT + | sd CARG2, 0(TMP2) // Swap function and traceback. + | andi TMP3, TMP3, 1 + | sd CARG1, 8(TMP2) + | beqz NARGS8:RC, ->vm_call_dispatch + |. daddiu PC, TMP3, 24+FRAME_PCALL + | b <1 + |. nop + | + |//-- Coroutine library -------------------------------------------------- + | + |.macro coroutine_resume_wrap, resume + |.if resume + |.ffunc_1 coroutine_resume + | checktp CARG1, CARG1, -LJ_TTHREAD, ->fff_fallback + |.else + |.ffunc coroutine_wrap_aux + | ld L:CARG1, CFUNC:RB->upvalue[0].gcr + | cleartp L:CARG1 + |.endif + | lbu TMP0, L:CARG1->status + | ld TMP1, L:CARG1->cframe + | ld CARG2, L:CARG1->top + | ld TMP2, L:CARG1->base + | addiu AT, TMP0, -LUA_YIELD + | daddu CARG3, CARG2, TMP0 + | daddiu TMP3, CARG2, 8 + |.if MIPSR6 + | seleqz CARG2, CARG2, AT + | selnez TMP3, TMP3, AT + | bgtz AT, ->fff_fallback // st > LUA_YIELD? + |. or CARG2, TMP3, CARG2 + |.else + | bgtz AT, ->fff_fallback // st > LUA_YIELD? + |. movn CARG2, TMP3, AT + |.endif + | xor TMP2, TMP2, CARG3 + | bnez TMP1, ->fff_fallback // cframe != 0? + |. or AT, TMP2, TMP0 + | ld TMP0, L:CARG1->maxstack + | beqz AT, ->fff_fallback // base == top && st == 0? + |. ld PC, FRAME_PC(BASE) + | daddu TMP2, CARG2, NARGS8:RC + | sltu AT, TMP0, TMP2 + | bnez AT, ->fff_fallback // Stack overflow? + |. sd PC, SAVE_PC + | sd BASE, L->base + |1: + |.if resume + | daddiu BASE, BASE, 8 // Keep resumed thread in stack for GC. + | daddiu NARGS8:RC, NARGS8:RC, -8 + | daddiu TMP2, TMP2, -8 + |.endif + | sd TMP2, L:CARG1->top + | daddu TMP1, BASE, NARGS8:RC + | move CARG3, CARG2 + | sd BASE, L->top + |2: // Move args to coroutine. + | ld CRET1, 0(BASE) + | sltu AT, BASE, TMP1 + | beqz AT, >3 + |. daddiu BASE, BASE, 8 + | sd CRET1, 0(CARG3) + | b <2 + |. daddiu CARG3, CARG3, 8 + |3: + | bal ->vm_resume // (lua_State *L, TValue *base, 0, 0) + |. move L:RA, L:CARG1 + | // Returns thread status. + |4: + | ld TMP2, L:RA->base + | sltiu AT, CRET1, LUA_YIELD+1 + | ld TMP3, L:RA->top + | li_vmstate INTERP + | ld BASE, L->base + | sd L, DISPATCH_GL(cur_L)(DISPATCH) + | st_vmstate + | beqz AT, >8 + |. dsubu RD, TMP3, TMP2 + | ld TMP0, L->maxstack + | beqz RD, >6 // No results? + |. daddu TMP1, BASE, RD + | sltu AT, TMP0, TMP1 + | bnez AT, >9 // Need to grow stack? + |. daddu TMP3, TMP2, RD + | sd TMP2, L:RA->top // Clear coroutine stack. + | move TMP1, BASE + |5: // Move results from coroutine. + | ld CRET1, 0(TMP2) + | daddiu TMP2, TMP2, 8 + | sltu AT, TMP2, TMP3 + | sd CRET1, 0(TMP1) + | bnez AT, <5 + |. daddiu TMP1, TMP1, 8 + |6: + | andi TMP0, PC, FRAME_TYPE + |.if resume + | mov_true TMP1 + | daddiu RA, BASE, -8 + | sd TMP1, -8(BASE) // Prepend true to results. + | daddiu RD, RD, 16 + |.else + | move RA, BASE + | daddiu RD, RD, 8 + |.endif + |7: + | sd PC, SAVE_PC + | beqz TMP0, ->BC_RET_Z + |. move MULTRES, RD + | b ->vm_return + |. nop + | + |8: // Coroutine returned with error (at co->top-1). + |.if resume + | daddiu TMP3, TMP3, -8 + | mov_false TMP1 + | ld CRET1, 0(TMP3) + | sd TMP3, L:RA->top // Remove error from coroutine stack. + | li RD, (2+1)*8 + | sd TMP1, -8(BASE) // Prepend false to results. + | daddiu RA, BASE, -8 + | sd CRET1, 0(BASE) // Copy error message. + | b <7 + |. andi TMP0, PC, FRAME_TYPE + |.else + | load_got lj_ffh_coroutine_wrap_err + | move CARG2, L:RA + | call_intern lj_ffh_coroutine_wrap_err // (lua_State *L, lua_State *co) + |. move CARG1, L + |.endif + | + |9: // Handle stack expansion on return from yield. + | load_got lj_state_growstack + | srl CARG2, RD, 3 + | call_intern lj_state_growstack // (lua_State *L, int n) + |. move CARG1, L + | b <4 + |. li CRET1, 0 + |.endmacro + | + | coroutine_resume_wrap 1 // coroutine.resume + | coroutine_resume_wrap 0 // coroutine.wrap + | + |.ffunc coroutine_yield + | ld TMP0, L->cframe + | daddu TMP1, BASE, NARGS8:RC + | sd BASE, L->base + | andi TMP0, TMP0, CFRAME_RESUME + | sd TMP1, L->top + | beqz TMP0, ->fff_fallback + |. li CRET1, LUA_YIELD + | sd r0, L->cframe + | b ->vm_leave_unw + |. sb CRET1, L->status + | + |//-- Math library ------------------------------------------------------- + | + |.ffunc_1 math_abs + | gettp CARG2, CARG1 + | daddiu AT, CARG2, -LJ_TISNUM + | bnez AT, >1 + |. sextw TMP1, CARG1 + | sra TMP0, TMP1, 31 // Extract sign. + | xor TMP1, TMP1, TMP0 + | dsubu CARG1, TMP1, TMP0 + | dsll TMP3, CARG1, 32 + | bgez TMP3, ->fff_restv + |. settp CARG1, TISNUM + | li CARG1, 0x41e0 // 2^31 as a double. + | b ->fff_restv + |. dsll CARG1, CARG1, 48 + |1: + | sltiu AT, CARG2, LJ_TISNUM + | beqz AT, ->fff_fallback + |. dextm CARG1, CARG1, 0, 30 + |// fallthrough + | + |->fff_restv: + | // CARG1 = TValue result. + | ld PC, FRAME_PC(BASE) + | daddiu RA, BASE, -16 + | sd CARG1, -16(BASE) + |->fff_res1: + | // RA = results, PC = return. + | li RD, (1+1)*8 + |->fff_res: + | // RA = results, RD = (nresults+1)*8, PC = return. + | andi TMP0, PC, FRAME_TYPE + | bnez TMP0, ->vm_return + |. move MULTRES, RD + | lw INS, -4(PC) + | decode_RB8a RB, INS + | decode_RB8b RB + |5: + | sltu AT, RD, RB + | bnez AT, >6 // More results expected? + |. decode_RA8a TMP0, INS + | decode_RA8b TMP0 + | ins_next1 + | // Adjust BASE. KBASE is assumed to be set for the calling frame. + | dsubu BASE, RA, TMP0 + | ins_next2 + | + |6: // Fill up results with nil. + | daddu TMP1, RA, RD + | daddiu RD, RD, 8 + | b <5 + |. sd TISNIL, -8(TMP1) + | + |.macro math_extern, func + | .ffunc_n math_ .. func + | load_got func + | call_extern + |. nop + | b ->fff_resn + |. nop + |.endmacro + | + |.macro math_extern2, func + | .ffunc_nn math_ .. func + |. load_got func + | call_extern + |. nop + | b ->fff_resn + |. nop + |.endmacro + | + |// TODO: Return integer type if result is integer (own sf implementation). + |.macro math_round, func + |->ff_math_ .. func: + | ld CARG1, 0(BASE) + | beqz NARGS8:RC, ->fff_fallback + |. gettp TMP0, CARG1 + | beq TMP0, TISNUM, ->fff_restv + |. sltu AT, TMP0, TISNUM + | beqz AT, ->fff_fallback + |.if FPU + |. ldc1 FARG1, 0(BASE) + | bal ->vm_ .. func + |. nop + |.else + |. load_got func + | call_extern + |. nop + |.endif + | b ->fff_resn + |. nop + |.endmacro + | + | math_round floor + | math_round ceil + | + |.ffunc math_log + | li AT, 8 + | bne NARGS8:RC, AT, ->fff_fallback // Exactly 1 argument. + |. ld CARG1, 0(BASE) + | checknum CARG1, ->fff_fallback + |. load_got log + |.if FPU + | call_extern + |. ldc1 FARG1, 0(BASE) + |.else + | call_extern + |. nop + |.endif + | b ->fff_resn + |. nop + | + | math_extern log10 + | math_extern exp + | math_extern sin + | math_extern cos + | math_extern tan + | math_extern asin + | math_extern acos + | math_extern atan + | math_extern sinh + | math_extern cosh + | math_extern tanh + | math_extern2 pow + | math_extern2 atan2 + | math_extern2 fmod + | + |.if FPU + |.ffunc_n math_sqrt + |. sqrt.d FRET1, FARG1 + |// fallthrough to ->fff_resn + |.else + | math_extern sqrt + |.endif + | + |->fff_resn: + | ld PC, FRAME_PC(BASE) + | daddiu RA, BASE, -16 + | b ->fff_res1 + |.if FPU + |. sdc1 FRET1, 0(RA) + |.else + |. sd CRET1, 0(RA) + |.endif + | + | + |.ffunc_2 math_ldexp + | checknum CARG1, ->fff_fallback + | checkint CARG2, ->fff_fallback + |. load_got ldexp + | .FPU ldc1 FARG1, 0(BASE) + | call_extern + |. lw CARG2, 8+LO(BASE) + | b ->fff_resn + |. nop + | + |.ffunc_n math_frexp + | load_got frexp + | ld PC, FRAME_PC(BASE) + | call_extern + |. daddiu CARG2, DISPATCH, DISPATCH_GL(tmptv) + | lw TMP1, DISPATCH_GL(tmptv)(DISPATCH) + | daddiu RA, BASE, -16 + |.if FPU + | mtc1 TMP1, FARG2 + | sdc1 FRET1, 0(RA) + | cvt.d.w FARG2, FARG2 + | sdc1 FARG2, 8(RA) + |.else + | sd CRET1, 0(RA) + | zextw TMP1, TMP1 + | settp TMP1, TISNUM + | sd TMP1, 8(RA) + |.endif + | b ->fff_res + |. li RD, (2+1)*8 + | + |.ffunc_n math_modf + | load_got modf + | ld PC, FRAME_PC(BASE) + | call_extern + |. daddiu CARG2, BASE, -16 + | daddiu RA, BASE, -16 + |.if FPU + | sdc1 FRET1, -8(BASE) + |.else + | sd CRET1, -8(BASE) + |.endif + | b ->fff_res + |. li RD, (2+1)*8 + | + |.macro math_minmax, name, intins, intinsc, fpins + | .ffunc_1 name + | daddu TMP3, BASE, NARGS8:RC + | checkint CARG1, >5 + |. daddiu TMP2, BASE, 8 + |1: // Handle integers. + | beq TMP2, TMP3, ->fff_restv + |. ld CARG2, 0(TMP2) + | checkint CARG2, >3 + |. sextw CARG1, CARG1 + | lw CARG2, LO(TMP2) + |. slt AT, CARG1, CARG2 + |.if MIPSR6 + | intins TMP1, CARG2, AT + | intinsc CARG1, CARG1, AT + | or CARG1, CARG1, TMP1 + |.else + | intins CARG1, CARG2, AT + |.endif + | daddiu TMP2, TMP2, 8 + | zextw CARG1, CARG1 + | b <1 + |. settp CARG1, TISNUM + | + |3: // Convert intermediate result to number and continue with number loop. + | checknum CARG2, ->fff_fallback + |.if FPU + |. mtc1 CARG1, FRET1 + | cvt.d.w FRET1, FRET1 + | b >7 + |. ldc1 FARG1, 0(TMP2) + |.else + |. nop + | bal ->vm_sfi2d_1 + |. nop + | b >7 + |. nop + |.endif + | + |5: + | .FPU ldc1 FRET1, 0(BASE) + | checknum CARG1, ->fff_fallback + |6: // Handle numbers. + |. ld CARG2, 0(TMP2) + | beq TMP2, TMP3, ->fff_resn + |.if FPU + | ldc1 FARG1, 0(TMP2) + |.else + | move CRET1, CARG1 + |.endif + | checknum CARG2, >8 + |. nop + |7: + |.if FPU + |.if MIPSR6 + | fpins FRET1, FRET1, FARG1 + |.else + |.if fpins // ismax + | c.olt.d FARG1, FRET1 + |.else + | c.olt.d FRET1, FARG1 + |.endif + | movf.d FRET1, FARG1 + |.endif + |.else + |.if fpins // ismax + | bal ->vm_sfcmpogt + |.else + | bal ->vm_sfcmpolt + |.endif + |. nop + |.if MIPSR6 + | seleqz AT, CARG2, CRET1 + | selnez CARG1, CARG1, CRET1 + | or CARG1, CARG1, AT + |.else + | movz CARG1, CARG2, CRET1 + |.endif + |.endif + | b <6 + |. daddiu TMP2, TMP2, 8 + | + |8: // Convert integer to number and continue with number loop. + | checkint CARG2, ->fff_fallback + |.if FPU + |. lwc1 FARG1, LO(TMP2) + | b <7 + |. cvt.d.w FARG1, FARG1 + |.else + |. lw CARG2, LO(TMP2) + | bal ->vm_sfi2d_2 + |. nop + | b <7 + |. nop + |.endif + | + |.endmacro + | + |.if MIPSR6 + | math_minmax math_min, seleqz, selnez, min.d + | math_minmax math_max, selnez, seleqz, max.d + |.else + | math_minmax math_min, movz, _, 0 + | math_minmax math_max, movn, _, 1 + |.endif + | + |//-- String library ----------------------------------------------------- + | + |.ffunc string_byte // Only handle the 1-arg case here. + | ld CARG1, 0(BASE) + | gettp TMP0, CARG1 + | xori AT, NARGS8:RC, 8 + | daddiu TMP0, TMP0, -LJ_TSTR + | or AT, AT, TMP0 + | bnez AT, ->fff_fallback // Need exactly 1 string argument. + |. cleartp STR:CARG1 + | lw TMP0, STR:CARG1->len + | daddiu RA, BASE, -16 + | ld PC, FRAME_PC(BASE) + | sltu RD, r0, TMP0 + | lbu TMP1, STR:CARG1[1] // Access is always ok (NUL at end). + | addiu RD, RD, 1 + | sll RD, RD, 3 // RD = ((str->len != 0)+1)*8 + | settp TMP1, TISNUM + | b ->fff_res + |. sd TMP1, 0(RA) + | + |.ffunc string_char // Only handle the 1-arg case here. + | ffgccheck + |.if not MIPSR6 + |. nop + |.endif + | ld CARG1, 0(BASE) + | gettp TMP0, CARG1 + | xori AT, NARGS8:RC, 8 // Exactly 1 argument. + | daddiu TMP0, TMP0, -LJ_TISNUM // Integer. + | li TMP1, 255 + | sextw CARG1, CARG1 + | or AT, AT, TMP0 + | sltu TMP1, TMP1, CARG1 // !(255 < n). + | or AT, AT, TMP1 + | bnez AT, ->fff_fallback + |. li CARG3, 1 + | daddiu CARG2, sp, TMPD_OFS + | sb CARG1, TMPD + |->fff_newstr: + | load_got lj_str_new + | sd BASE, L->base + | sd PC, SAVE_PC + | call_intern lj_str_new // (lua_State *L, char *str, size_t l) + |. move CARG1, L + | // Returns GCstr *. + | ld BASE, L->base + |->fff_resstr: + | li AT, LJ_TSTR + | settp CRET1, AT + | b ->fff_restv + |. move CARG1, CRET1 + | + |.ffunc string_sub + | ffgccheck + |.if not MIPSR6 + |. nop + |.endif + | addiu AT, NARGS8:RC, -16 + | ld TMP0, 0(BASE) + | bltz AT, ->fff_fallback + |. gettp TMP3, TMP0 + | cleartp STR:CARG1, TMP0 + | ld CARG2, 8(BASE) + | beqz AT, >1 + |. li CARG4, -1 + | ld CARG3, 16(BASE) + | checkint CARG3, ->fff_fallback + |. sextw CARG4, CARG3 + |1: + | checkint CARG2, ->fff_fallback + |. li AT, LJ_TSTR + | bne TMP3, AT, ->fff_fallback + |. sextw CARG3, CARG2 + | lw CARG2, STR:CARG1->len + | // STR:CARG1 = str, CARG2 = str->len, CARG3 = start, CARG4 = end + | slt AT, CARG4, r0 + | addiu TMP0, CARG2, 1 + | addu TMP1, CARG4, TMP0 + | slt TMP3, CARG3, r0 + |.if MIPSR6 + | seleqz CARG4, CARG4, AT + | selnez TMP1, TMP1, AT + | or CARG4, TMP1, CARG4 // if (end < 0) end += len+1 + |.else + | movn CARG4, TMP1, AT // if (end < 0) end += len+1 + |.endif + | addu TMP1, CARG3, TMP0 + |.if MIPSR6 + | selnez TMP1, TMP1, TMP3 + | seleqz CARG3, CARG3, TMP3 + | or CARG3, TMP1, CARG3 // if (start < 0) start += len+1 + | li TMP2, 1 + | slt AT, CARG4, r0 + | slt TMP3, r0, CARG3 + | seleqz CARG4, CARG4, AT // if (end < 0) end = 0 + | selnez CARG3, CARG3, TMP3 + | seleqz TMP2, TMP2, TMP3 + | or CARG3, TMP2, CARG3 // if (start < 1) start = 1 + | slt AT, CARG2, CARG4 + | seleqz CARG4, CARG4, AT + | selnez CARG2, CARG2, AT + | or CARG4, CARG2, CARG4 // if (end > len) end = len + |.else + | movn CARG3, TMP1, TMP3 // if (start < 0) start += len+1 + | li TMP2, 1 + | slt AT, CARG4, r0 + | slt TMP3, r0, CARG3 + | movn CARG4, r0, AT // if (end < 0) end = 0 + | movz CARG3, TMP2, TMP3 // if (start < 1) start = 1 + | slt AT, CARG2, CARG4 + | movn CARG4, CARG2, AT // if (end > len) end = len + |.endif + | daddu CARG2, STR:CARG1, CARG3 + | subu CARG3, CARG4, CARG3 // len = end - start + | daddiu CARG2, CARG2, sizeof(GCstr)-1 + | bgez CARG3, ->fff_newstr + |. addiu CARG3, CARG3, 1 // len++ + |->fff_emptystr: // Return empty string. + | li AT, LJ_TSTR + | daddiu STR:CARG1, DISPATCH, DISPATCH_GL(strempty) + | b ->fff_restv + |. settp CARG1, AT + | + |.macro ffstring_op, name + | .ffunc string_ .. name + | ffgccheck + |. nop + | beqz NARGS8:RC, ->fff_fallback + |. ld CARG2, 0(BASE) + | checkstr STR:CARG2, ->fff_fallback + | daddiu SBUF:CARG1, DISPATCH, DISPATCH_GL(tmpbuf) + | load_got lj_buf_putstr_ .. name + | ld TMP0, SBUF:CARG1->b + | sd L, SBUF:CARG1->L + | sd BASE, L->base + | sd TMP0, SBUF:CARG1->w + | call_intern extern lj_buf_putstr_ .. name + |. sd PC, SAVE_PC + | load_got lj_buf_tostr + | call_intern lj_buf_tostr + |. move SBUF:CARG1, SBUF:CRET1 + | b ->fff_resstr + |. ld BASE, L->base + |.endmacro + | + |ffstring_op reverse + |ffstring_op lower + |ffstring_op upper + | + |//-- Bit library -------------------------------------------------------- + | + |->vm_tobit_fb: + | beqz TMP1, ->fff_fallback + |.if FPU + |. ldc1 FARG1, 0(BASE) + | add.d FARG1, FARG1, TOBIT + | mfc1 CRET1, FARG1 + | jr ra + |. zextw CRET1, CRET1 + |.else + |// FP number to bit conversion for soft-float. + |->vm_tobit: + | dsll TMP0, CARG1, 1 + | li CARG3, 1076 + | dsrl AT, TMP0, 53 + | dsubu CARG3, CARG3, AT + | sltiu AT, CARG3, 54 + | beqz AT, >1 + |. dextm TMP0, TMP0, 0, 20 + | dinsu TMP0, AT, 21, 21 + | slt AT, CARG1, r0 + | dsrlv CRET1, TMP0, CARG3 + | dsubu TMP0, r0, CRET1 + |.if MIPSR6 + | selnez TMP0, TMP0, AT + | seleqz CRET1, CRET1, AT + | or CRET1, CRET1, TMP0 + |.else + | movn CRET1, TMP0, AT + |.endif + | jr ra + |. zextw CRET1, CRET1 + |1: + | jr ra + |. move CRET1, r0 + | + |// FP number to int conversion with a check for soft-float. + |// Modifies CARG1, CRET1, CRET2, TMP0, AT. + |->vm_tointg: + |.if JIT + | dsll CRET2, CARG1, 1 + | beqz CRET2, >2 + |. li TMP0, 1076 + | dsrl AT, CRET2, 53 + | dsubu TMP0, TMP0, AT + | sltiu AT, TMP0, 54 + | beqz AT, >1 + |. dextm CRET2, CRET2, 0, 20 + | dinsu CRET2, AT, 21, 21 + | slt AT, CARG1, r0 + | dsrlv CRET1, CRET2, TMP0 + | dsubu CARG1, r0, CRET1 + |.if MIPSR6 + | seleqz CRET1, CRET1, AT + | selnez CARG1, CARG1, AT + | or CRET1, CRET1, CARG1 + |.else + | movn CRET1, CARG1, AT + |.endif + | li CARG1, 64 + | subu TMP0, CARG1, TMP0 + | dsllv CRET2, CRET2, TMP0 // Integer check. + | sextw AT, CRET1 + | xor AT, CRET1, AT // Range check. + |.if MIPSR6 + | seleqz AT, AT, CRET2 + | selnez CRET2, CRET2, CRET2 + | jr ra + |. or CRET2, AT, CRET2 + |.else + | jr ra + |. movz CRET2, AT, CRET2 + |.endif + |1: + | jr ra + |. li CRET2, 1 + |2: + | jr ra + |. move CRET1, r0 + |.endif + |.endif + | + |.macro .ffunc_bit, name + | .ffunc_1 bit_..name + | gettp TMP0, CARG1 + | beq TMP0, TISNUM, >6 + |. zextw CRET1, CARG1 + | bal ->vm_tobit_fb + |. sltiu TMP1, TMP0, LJ_TISNUM + |6: + |.endmacro + | + |.macro .ffunc_bit_op, name, bins + | .ffunc_bit name + | daddiu TMP2, BASE, 8 + | daddu TMP3, BASE, NARGS8:RC + |1: + | beq TMP2, TMP3, ->fff_resi + |. ld CARG1, 0(TMP2) + | gettp TMP0, CARG1 + |.if FPU + | bne TMP0, TISNUM, >2 + |. daddiu TMP2, TMP2, 8 + | zextw CARG1, CARG1 + | b <1 + |. bins CRET1, CRET1, CARG1 + |2: + | ldc1 FARG1, -8(TMP2) + | sltiu AT, TMP0, LJ_TISNUM + | beqz AT, ->fff_fallback + |. add.d FARG1, FARG1, TOBIT + | mfc1 CARG1, FARG1 + | zextw CARG1, CARG1 + | b <1 + |. bins CRET1, CRET1, CARG1 + |.else + | beq TMP0, TISNUM, >2 + |. move CRET2, CRET1 + | bal ->vm_tobit_fb + |. sltiu TMP1, TMP0, LJ_TISNUM + | move CARG1, CRET2 + |2: + | zextw CARG1, CARG1 + | bins CRET1, CRET1, CARG1 + | b <1 + |. daddiu TMP2, TMP2, 8 + |.endif + |.endmacro + | + |.ffunc_bit_op band, and + |.ffunc_bit_op bor, or + |.ffunc_bit_op bxor, xor + | + |.ffunc_bit bswap + | dsrl TMP0, CRET1, 8 + | dsrl TMP1, CRET1, 24 + | andi TMP2, TMP0, 0xff00 + | dins TMP1, CRET1, 24, 31 + | dins TMP2, TMP0, 16, 23 + | b ->fff_resi + |. or CRET1, TMP1, TMP2 + | + |.ffunc_bit bnot + | not CRET1, CRET1 + | b ->fff_resi + |. zextw CRET1, CRET1 + | + |.macro .ffunc_bit_sh, name, shins, shmod + | .ffunc_2 bit_..name + | gettp TMP0, CARG1 + | beq TMP0, TISNUM, >1 + |. nop + | bal ->vm_tobit_fb + |. sltiu TMP1, TMP0, LJ_TISNUM + | move CARG1, CRET1 + |1: + | gettp TMP0, CARG2 + | bne TMP0, TISNUM, ->fff_fallback + |. zextw CARG2, CARG2 + | sextw CARG1, CARG1 + |.if shmod == 1 + | negu CARG2, CARG2 + |.endif + | shins CRET1, CARG1, CARG2 + | b ->fff_resi + |. zextw CRET1, CRET1 + |.endmacro + | + |.ffunc_bit_sh lshift, sllv, 0 + |.ffunc_bit_sh rshift, srlv, 0 + |.ffunc_bit_sh arshift, srav, 0 + |.ffunc_bit_sh rol, rotrv, 1 + |.ffunc_bit_sh ror, rotrv, 0 + | + |.ffunc_bit tobit + |->fff_resi: + | ld PC, FRAME_PC(BASE) + | daddiu RA, BASE, -16 + | settp CRET1, TISNUM + | b ->fff_res1 + |. sd CRET1, -16(BASE) + | + |//----------------------------------------------------------------------- + |->fff_fallback: // Call fast function fallback handler. + | // BASE = new base, RB = CFUNC, RC = nargs*8 + | ld TMP3, CFUNC:RB->f + | daddu TMP1, BASE, NARGS8:RC + | ld PC, FRAME_PC(BASE) // Fallback may overwrite PC. + | daddiu TMP0, TMP1, 8*LUA_MINSTACK + | ld TMP2, L->maxstack + | sd PC, SAVE_PC // Redundant (but a defined value). + | sltu AT, TMP2, TMP0 + | sd BASE, L->base + | sd TMP1, L->top + | bnez AT, >5 // Need to grow stack. + |. move CFUNCADDR, TMP3 + | jalr TMP3 // (lua_State *L) + |. move CARG1, L + | // Either throws an error, or recovers and returns -1, 0 or nresults+1. + | ld BASE, L->base + | sll RD, CRET1, 3 + | bgtz CRET1, ->fff_res // Returned nresults+1? + |. daddiu RA, BASE, -16 + |1: // Returned 0 or -1: retry fast path. + | ld LFUNC:RB, FRAME_FUNC(BASE) + | ld TMP0, L->top + | cleartp LFUNC:RB + | bnez CRET1, ->vm_call_tail // Returned -1? + |. dsubu NARGS8:RC, TMP0, BASE + | ins_callt // Returned 0: retry fast path. + | + |// Reconstruct previous base for vmeta_call during tailcall. + |->vm_call_tail: + | andi TMP0, PC, FRAME_TYPE + | li AT, -4 + | bnez TMP0, >3 + |. and TMP1, PC, AT + | lbu TMP1, OFS_RA(PC) + | sll TMP1, TMP1, 3 + | addiu TMP1, TMP1, 16 + |3: + | b ->vm_call_dispatch // Resolve again for tailcall. + |. dsubu TMP2, BASE, TMP1 + | + |5: // Grow stack for fallback handler. + | load_got lj_state_growstack + | li CARG2, LUA_MINSTACK + | call_intern lj_state_growstack // (lua_State *L, int n) + |. move CARG1, L + | ld BASE, L->base + | b <1 + |. li CRET1, 0 // Force retry. + | + |->fff_gcstep: // Call GC step function. + | // BASE = new base, RC = nargs*8 + | move MULTRES, ra + | load_got lj_gc_step + | sd BASE, L->base + | daddu TMP0, BASE, NARGS8:RC + | sd PC, SAVE_PC // Redundant (but a defined value). + | sd TMP0, L->top + | call_intern lj_gc_step // (lua_State *L) + |. move CARG1, L + | ld BASE, L->base + | move ra, MULTRES + | ld TMP0, L->top + | ld CFUNC:RB, FRAME_FUNC(BASE) + | cleartp CFUNC:RB + | jr ra + |. dsubu NARGS8:RC, TMP0, BASE + | + |//----------------------------------------------------------------------- + |//-- Special dispatch targets ------------------------------------------- + |//----------------------------------------------------------------------- + | + |->vm_record: // Dispatch target for recording phase. + |.if JIT + | lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH) + | andi AT, TMP3, HOOK_VMEVENT // No recording while in vmevent. + | bnez AT, >5 + | // Decrement the hookcount for consistency, but always do the call. + |. lw TMP2, DISPATCH_GL(hookcount)(DISPATCH) + | andi AT, TMP3, HOOK_ACTIVE + | bnez AT, >1 + |. addiu TMP2, TMP2, -1 + | andi AT, TMP3, LUA_MASKLINE|LUA_MASKCOUNT + | beqz AT, >1 + |. nop + | b >1 + |. sw TMP2, DISPATCH_GL(hookcount)(DISPATCH) + |.endif + | + |->vm_rethook: // Dispatch target for return hooks. + | lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH) + | andi AT, TMP3, HOOK_ACTIVE // Hook already active? + | beqz AT, >1 + |5: // Re-dispatch to static ins. + |. ld AT, GG_DISP2STATIC(TMP0) // Assumes TMP0 holds DISPATCH+OP*4. + | jr AT + |. nop + | + |->vm_inshook: // Dispatch target for instr/line hooks. + | lbu TMP3, DISPATCH_GL(hookmask)(DISPATCH) + | lw TMP2, DISPATCH_GL(hookcount)(DISPATCH) + | andi AT, TMP3, HOOK_ACTIVE // Hook already active? + | bnez AT, <5 + |. andi AT, TMP3, LUA_MASKLINE|LUA_MASKCOUNT + | beqz AT, <5 + |. addiu TMP2, TMP2, -1 + | beqz TMP2, >1 + |. sw TMP2, DISPATCH_GL(hookcount)(DISPATCH) + | andi AT, TMP3, LUA_MASKLINE + | beqz AT, <5 + |1: + |. load_got lj_dispatch_ins + | sw MULTRES, SAVE_MULTRES + | move CARG2, PC + | sd BASE, L->base + | // SAVE_PC must hold the _previous_ PC. The callee updates it with PC. + | call_intern lj_dispatch_ins // (lua_State *L, const BCIns *pc) + |. move CARG1, L + |3: + | ld BASE, L->base + |4: // Re-dispatch to static ins. + | lw INS, -4(PC) + | decode_OP8a TMP1, INS + | decode_OP8b TMP1 + | daddu TMP0, DISPATCH, TMP1 + | decode_RD8a RD, INS + | ld AT, GG_DISP2STATIC(TMP0) + | decode_RA8a RA, INS + | decode_RD8b RD + | jr AT + | decode_RA8b RA + | + |->cont_hook: // Continue from hook yield. + | daddiu PC, PC, 4 + | b <4 + |. lw MULTRES, -24+LO(RB) // Restore MULTRES for *M ins. + | + |->vm_hotloop: // Hot loop counter underflow. + |.if JIT + | ld LFUNC:TMP1, FRAME_FUNC(BASE) + | daddiu CARG1, DISPATCH, GG_DISP2J + | cleartp LFUNC:TMP1 + | sd PC, SAVE_PC + | ld TMP1, LFUNC:TMP1->pc + | move CARG2, PC + | sd L, DISPATCH_J(L)(DISPATCH) + | lbu TMP1, PC2PROTO(framesize)(TMP1) + | load_got lj_trace_hot + | sd BASE, L->base + | dsll TMP1, TMP1, 3 + | daddu TMP1, BASE, TMP1 + | call_intern lj_trace_hot // (jit_State *J, const BCIns *pc) + |. sd TMP1, L->top + | b <3 + |. nop + |.endif + | + | + |->vm_callhook: // Dispatch target for call hooks. + |.if JIT + | b >1 + |.endif + |. move CARG2, PC + | + |->vm_hotcall: // Hot call counter underflow. + |.if JIT + | ori CARG2, PC, 1 + |1: + |.endif + | load_got lj_dispatch_call + | daddu TMP0, BASE, RC + | sd PC, SAVE_PC + | sd BASE, L->base + | dsubu RA, RA, BASE + | sd TMP0, L->top + | call_intern lj_dispatch_call // (lua_State *L, const BCIns *pc) + |. move CARG1, L + | // Returns ASMFunction. + | ld BASE, L->base + | ld TMP0, L->top + | sd r0, SAVE_PC // Invalidate for subsequent line hook. + | dsubu NARGS8:RC, TMP0, BASE + | daddu RA, BASE, RA + | ld LFUNC:RB, FRAME_FUNC(BASE) + | cleartp LFUNC:RB + | jr CRET1 + |. lw INS, -4(PC) + | + |->cont_stitch: // Trace stitching. + |.if JIT + | // RA = resultptr, RB = meta base + | lw INS, -4(PC) + | ld TRACE:TMP2, -40(RB) // Save previous trace. + | decode_RA8a RC, INS + | daddiu AT, MULTRES, -8 + | cleartp TRACE:TMP2 + | decode_RA8b RC + | beqz AT, >2 + |. daddu RC, BASE, RC // Call base. + |1: // Move results down. + | ld CARG1, 0(RA) + | daddiu AT, AT, -8 + | daddiu RA, RA, 8 + | sd CARG1, 0(RC) + | bnez AT, <1 + |. daddiu RC, RC, 8 + |2: + | decode_RA8a RA, INS + | decode_RB8a RB, INS + | decode_RA8b RA + | decode_RB8b RB + | daddu RA, RA, RB + | daddu RA, BASE, RA + |3: + | sltu AT, RC, RA + | bnez AT, >9 // More results wanted? + |. nop + | + | lhu TMP3, TRACE:TMP2->traceno + | lhu RD, TRACE:TMP2->link + | beq RD, TMP3, ->cont_nop // Blacklisted. + |. load_got lj_dispatch_stitch + | bnez RD, =>BC_JLOOP // Jump to stitched trace. + |. sll RD, RD, 3 + | + | // Stitch a new trace to the previous trace. + | sw TMP3, DISPATCH_J(exitno)(DISPATCH) + | sd L, DISPATCH_J(L)(DISPATCH) + | sd BASE, L->base + | daddiu CARG1, DISPATCH, GG_DISP2J + | call_intern lj_dispatch_stitch // (jit_State *J, const BCIns *pc) + |. move CARG2, PC + | b ->cont_nop + |. ld BASE, L->base + | + |9: + | sd TISNIL, 0(RC) + | b <3 + |. daddiu RC, RC, 8 + |.endif + | + |->vm_profhook: // Dispatch target for profiler hook. +#if LJ_HASPROFILE + | load_got lj_dispatch_profile + | sw MULTRES, SAVE_MULTRES + | move CARG2, PC + | sd BASE, L->base + | call_intern lj_dispatch_profile // (lua_State *L, const BCIns *pc) + |. move CARG1, L + | // HOOK_PROFILE is off again, so re-dispatch to dynamic instruction. + | daddiu PC, PC, -4 + | b ->cont_nop + |. ld BASE, L->base +#endif + | + |//----------------------------------------------------------------------- + |//-- Trace exit handler ------------------------------------------------- + |//----------------------------------------------------------------------- + | + |.macro savex_, a, b + |.if FPU + | sdc1 f..a, a*8(sp) + | sdc1 f..b, b*8(sp) + | sd r..a, 32*8+a*8(sp) + | sd r..b, 32*8+b*8(sp) + |.else + | sd r..a, a*8(sp) + | sd r..b, b*8(sp) + |.endif + |.endmacro + | + |->vm_exit_handler: + |.if JIT + |.if FPU + | daddiu sp, sp, -(32*8+32*8) + |.else + | daddiu sp, sp, -(32*8) + |.endif + | savex_ 0, 1 + | savex_ 2, 3 + | savex_ 4, 5 + | savex_ 6, 7 + | savex_ 8, 9 + | savex_ 10, 11 + | savex_ 12, 13 + | savex_ 14, 15 + | savex_ 16, 17 + | savex_ 18, 19 + | savex_ 20, 21 + | savex_ 22, 23 + | savex_ 24, 25 + | savex_ 26, 27 + | savex_ 28, 30 + |.if FPU + | sdc1 f29, 29*8(sp) + | sdc1 f31, 31*8(sp) + | sd r0, 32*8+31*8(sp) // Clear RID_TMP. + | daddiu TMP2, sp, 32*8+32*8 // Recompute original value of sp. + | sd TMP2, 32*8+29*8(sp) // Store sp in RID_SP + |.else + | sd r0, 31*8(sp) // Clear RID_TMP. + | daddiu TMP2, sp, 32*8 // Recompute original value of sp. + | sd TMP2, 29*8(sp) // Store sp in RID_SP + |.endif + | li_vmstate EXIT + | daddiu DISPATCH, JGL, -GG_DISP2G-32768 + | lw TMP1, 0(TMP2) // Load exit number. + | st_vmstate + | ld L, DISPATCH_GL(cur_L)(DISPATCH) + | ld BASE, DISPATCH_GL(jit_base)(DISPATCH) + | load_got lj_trace_exit + | sd L, DISPATCH_J(L)(DISPATCH) + | sw ra, DISPATCH_J(parent)(DISPATCH) // Store trace number. + | sd BASE, L->base + | sw TMP1, DISPATCH_J(exitno)(DISPATCH) // Store exit number. + | daddiu CARG1, DISPATCH, GG_DISP2J + | sd r0, DISPATCH_GL(jit_base)(DISPATCH) + | call_intern lj_trace_exit // (jit_State *J, ExitState *ex) + |. move CARG2, sp + | // Returns MULTRES (unscaled) or negated error code. + | ld TMP1, L->cframe + | li AT, -4 + | ld BASE, L->base + | and sp, TMP1, AT + | ld PC, SAVE_PC // Get SAVE_PC. + | b >1 + |. sd L, SAVE_L // Set SAVE_L (on-trace resume/yield). + |.endif + |->vm_exit_interp: + |.if JIT + | // CRET1 = MULTRES or negated error code, BASE, PC and JGL set. + | ld L, SAVE_L + | daddiu DISPATCH, JGL, -GG_DISP2G-32768 + | sd BASE, L->base + |1: + | sltiu TMP0, CRET1, -LUA_ERRERR // Check for error from exit. + | beqz TMP0, >9 + |. ld LFUNC:RB, FRAME_FUNC(BASE) + | .FPU lui TMP3, 0x59c0 // TOBIT = 2^52 + 2^51 (float). + | dsll MULTRES, CRET1, 3 + | cleartp LFUNC:RB + | sw MULTRES, SAVE_MULTRES + | li TISNIL, LJ_TNIL + | li TISNUM, LJ_TISNUM // Setup type comparison constants. + | .FPU mtc1 TMP3, TOBIT + | ld TMP1, LFUNC:RB->pc + | sd r0, DISPATCH_GL(jit_base)(DISPATCH) + | ld KBASE, PC2PROTO(k)(TMP1) + | .FPU cvt.d.s TOBIT, TOBIT + | // Modified copy of ins_next which handles function header dispatch, too. + | lw INS, 0(PC) + | addiu CRET1, CRET1, 17 // Static dispatch? + | // Assumes TISNIL == ~LJ_VMST_INTERP == -1 + | sw TISNIL, DISPATCH_GL(vmstate)(DISPATCH) + | decode_RD8a RD, INS + | beqz CRET1, >5 + |. daddiu PC, PC, 4 + | decode_OP8a TMP1, INS + | decode_OP8b TMP1 + | daddu TMP0, DISPATCH, TMP1 + | sltiu TMP2, TMP1, BC_FUNCF*8 + | ld AT, 0(TMP0) + | decode_RA8a RA, INS + | beqz TMP2, >2 + |. decode_RA8b RA + | jr AT + |. decode_RD8b RD + |2: + | sltiu TMP2, TMP1, (BC_FUNCC+2)*8 // Fast function? + | bnez TMP2, >3 + |. ld TMP1, FRAME_PC(BASE) + | // Check frame below fast function. + | andi TMP0, TMP1, FRAME_TYPE + | bnez TMP0, >3 // Trace stitching continuation? + |. nop + | // Otherwise set KBASE for Lua function below fast function. + | lw TMP2, -4(TMP1) + | decode_RA8a TMP0, TMP2 + | decode_RA8b TMP0 + | dsubu TMP1, BASE, TMP0 + | ld LFUNC:TMP2, -32(TMP1) + | cleartp LFUNC:TMP2 + | ld TMP1, LFUNC:TMP2->pc + | ld KBASE, PC2PROTO(k)(TMP1) + |3: + | daddiu RC, MULTRES, -8 + | jr AT + |. daddu RA, RA, BASE + | + |5: // Dispatch to static entry of original ins replaced by BC_JLOOP. + | ld TMP0, DISPATCH_J(trace)(DISPATCH) + | decode_RD8b RD + | daddu TMP0, TMP0, RD + | ld TRACE:TMP2, 0(TMP0) + | lw INS, TRACE:TMP2->startins + | decode_OP8a TMP1, INS + | decode_OP8b TMP1 + | daddu TMP0, DISPATCH, TMP1 + | decode_RD8a RD, INS + | ld AT, GG_DISP2STATIC(TMP0) + | decode_RA8a RA, INS + | decode_RD8b RD + | jr AT + |. decode_RA8b RA + | + |9: // Rethrow error from the right C frame. + | load_got lj_err_trace + | sub CARG2, r0, CRET1 + | call_intern lj_err_trace // (lua_State *L, int errcode) + |. move CARG1, L + |.endif + | + |//----------------------------------------------------------------------- + |//-- Math helper functions ---------------------------------------------- + |//----------------------------------------------------------------------- + | + |// Hard-float round to integer. + |// Modifies AT, TMP0, FRET1, FRET2, f4. Keeps all others incl. FARG1. + |// MIPSR6: Modifies FTMP1, too. + |.macro vm_round_hf, func + | lui TMP0, 0x4330 // Hiword of 2^52 (double). + | dsll TMP0, TMP0, 32 + | dmtc1 TMP0, f4 + | abs.d FRET2, FARG1 // |x| + | dmfc1 AT, FARG1 + |.if MIPSR6 + | cmp.lt.d FTMP1, FRET2, f4 + | add.d FRET1, FRET2, f4 // (|x| + 2^52) - 2^52 + | bc1eqz FTMP1, >1 // Truncate only if |x| < 2^52. + |.else + | c.olt.d 0, FRET2, f4 + | add.d FRET1, FRET2, f4 // (|x| + 2^52) - 2^52 + | bc1f 0, >1 // Truncate only if |x| < 2^52. + |.endif + |. sub.d FRET1, FRET1, f4 + | slt AT, AT, r0 + |.if "func" == "ceil" + | lui TMP0, 0xbff0 // Hiword of -1 (double). Preserves -0. + |.else + | lui TMP0, 0x3ff0 // Hiword of +1 (double). + |.endif + |.if "func" == "trunc" + | dsll TMP0, TMP0, 32 + | dmtc1 TMP0, f4 + |.if MIPSR6 + | cmp.lt.d FTMP1, FRET2, FRET1 // |x| < result? + | sub.d FRET2, FRET1, f4 + | sel.d FTMP1, FRET1, FRET2 // If yes, subtract +1. + | dmtc1 AT, FRET1 + | neg.d FRET2, FTMP1 + | jr ra + |. sel.d FRET1, FTMP1, FRET2 // Merge sign bit back in. + |.else + | c.olt.d 0, FRET2, FRET1 // |x| < result? + | sub.d FRET2, FRET1, f4 + | movt.d FRET1, FRET2, 0 // If yes, subtract +1. + | neg.d FRET2, FRET1 + | jr ra + |. movn.d FRET1, FRET2, AT // Merge sign bit back in. + |.endif + |.else + | neg.d FRET2, FRET1 + | dsll TMP0, TMP0, 32 + | dmtc1 TMP0, f4 + |.if MIPSR6 + | dmtc1 AT, FTMP1 + | sel.d FTMP1, FRET1, FRET2 + |.if "func" == "ceil" + | cmp.lt.d FRET1, FTMP1, FARG1 // x > result? + |.else + | cmp.lt.d FRET1, FARG1, FTMP1 // x < result? + |.endif + | sub.d FRET2, FTMP1, f4 // If yes, subtract +-1. + | jr ra + |. sel.d FRET1, FTMP1, FRET2 + |.else + | movn.d FRET1, FRET2, AT // Merge sign bit back in. + |.if "func" == "ceil" + | c.olt.d 0, FRET1, FARG1 // x > result? + |.else + | c.olt.d 0, FARG1, FRET1 // x < result? + |.endif + | sub.d FRET2, FRET1, f4 // If yes, subtract +-1. + | jr ra + |. movt.d FRET1, FRET2, 0 + |.endif + |.endif + |1: + | jr ra + |. mov.d FRET1, FARG1 + |.endmacro + | + |.macro vm_round, func + |.if FPU + | vm_round_hf, func + |.endif + |.endmacro + | + |->vm_floor: + | vm_round floor + |->vm_ceil: + | vm_round ceil + |->vm_trunc: + |.if JIT + | vm_round trunc + |.endif + | + |// Soft-float integer to number conversion. + |.macro sfi2d, ARG + |.if not FPU + | beqz ARG, >9 // Handle zero first. + |. sra TMP0, ARG, 31 + | xor TMP1, ARG, TMP0 + | dsubu TMP1, TMP1, TMP0 // Absolute value in TMP1. + | dclz ARG, TMP1 + | addiu ARG, ARG, -11 + | li AT, 0x3ff+63-11-1 + | dsllv TMP1, TMP1, ARG // Align mantissa left with leading 1. + | subu ARG, AT, ARG // Exponent - 1. + | ins ARG, TMP0, 11, 11 // Sign | Exponent. + | dsll ARG, ARG, 52 // Align left. + | jr ra + |. daddu ARG, ARG, TMP1 // Add mantissa, increment exponent. + |9: + | jr ra + |. nop + |.endif + |.endmacro + | + |// Input CARG1. Output: CARG1. Temporaries: AT, TMP0, TMP1. + |->vm_sfi2d_1: + | sfi2d CARG1 + | + |// Input CARG2. Output: CARG2. Temporaries: AT, TMP0, TMP1. + |->vm_sfi2d_2: + | sfi2d CARG2 + | + |// Soft-float comparison. Equivalent to c.eq.d. + |// Input: CARG*. Output: CRET1. Temporaries: AT, TMP0, TMP1. + |->vm_sfcmpeq: + |.if not FPU + | dsll AT, CARG1, 1 + | dsll TMP0, CARG2, 1 + | or TMP1, AT, TMP0 + | beqz TMP1, >8 // Both args +-0: return 1. + |. lui TMP1, 0xffe0 + | dsll TMP1, TMP1, 32 + | sltu AT, TMP1, AT + | sltu TMP0, TMP1, TMP0 + | or TMP1, AT, TMP0 + | bnez TMP1, >9 // Either arg is NaN: return 0; + |. xor AT, CARG1, CARG2 + | jr ra + |. sltiu CRET1, AT, 1 // Same values: return 1. + |8: + | jr ra + |. li CRET1, 1 + |9: + | jr ra + |. li CRET1, 0 + |.endif + | + |// Soft-float comparison. Equivalent to c.ult.d and c.olt.d. + |// Input: CARG1, CARG2. Output: CRET1. Temporaries: AT, TMP0, TMP1, CRET2. + |->vm_sfcmpult: + |.if not FPU + | b >1 + |. li CRET2, 1 + |.endif + | + |->vm_sfcmpolt: + |.if not FPU + | li CRET2, 0 + |1: + | dsll AT, CARG1, 1 + | dsll TMP0, CARG2, 1 + | or TMP1, AT, TMP0 + | beqz TMP1, >8 // Both args +-0: return 0. + |. lui TMP1, 0xffe0 + | dsll TMP1, TMP1, 32 + | sltu AT, TMP1, AT + | sltu TMP0, TMP1, TMP0 + | or TMP1, AT, TMP0 + | bnez TMP1, >9 // Either arg is NaN: return 0 or 1; + |. and AT, CARG1, CARG2 + | bltz AT, >5 // Both args negative? + |. nop + | jr ra + |. slt CRET1, CARG1, CARG2 + |5: // Swap conditions if both operands are negative. + | jr ra + |. slt CRET1, CARG2, CARG1 + |8: + | jr ra + |. li CRET1, 0 + |9: + | jr ra + |. move CRET1, CRET2 + |.endif + | + |->vm_sfcmpogt: + |.if not FPU + | dsll AT, CARG2, 1 + | dsll TMP0, CARG1, 1 + | or TMP1, AT, TMP0 + | beqz TMP1, >8 // Both args +-0: return 0. + |. lui TMP1, 0xffe0 + | dsll TMP1, TMP1, 32 + | sltu AT, TMP1, AT + | sltu TMP0, TMP1, TMP0 + | or TMP1, AT, TMP0 + | bnez TMP1, >9 // Either arg is NaN: return 0 or 1; + |. and AT, CARG2, CARG1 + | bltz AT, >5 // Both args negative? + |. nop + | jr ra + |. slt CRET1, CARG2, CARG1 + |5: // Swap conditions if both operands are negative. + | jr ra + |. slt CRET1, CARG1, CARG2 + |8: + | jr ra + |. li CRET1, 0 + |9: + | jr ra + |. li CRET1, 0 + |.endif + | + |// Soft-float comparison. Equivalent to c.ole.d a, b or c.ole.d b, a. + |// Input: CARG1, CARG2, TMP3. Output: CRET1. Temporaries: AT, TMP0, TMP1. + |->vm_sfcmpolex: + |.if not FPU + | dsll AT, CARG1, 1 + | dsll TMP0, CARG2, 1 + | or TMP1, AT, TMP0 + | beqz TMP1, >8 // Both args +-0: return 1. + |. lui TMP1, 0xffe0 + | dsll TMP1, TMP1, 32 + | sltu AT, TMP1, AT + | sltu TMP0, TMP1, TMP0 + | or TMP1, AT, TMP0 + | bnez TMP1, >9 // Either arg is NaN: return 0; + |. and AT, CARG1, CARG2 + | xor AT, AT, TMP3 + | bltz AT, >5 // Both args negative? + |. nop + | jr ra + |. slt CRET1, CARG2, CARG1 + |5: // Swap conditions if both operands are negative. + | jr ra + |. slt CRET1, CARG1, CARG2 + |8: + | jr ra + |. li CRET1, 1 + |9: + | jr ra + |. li CRET1, 0 + |.endif + | + |.macro sfmin_max, name, fpcall + |->vm_sf .. name: + |.if JIT and not FPU + | move TMP2, ra + | bal ->fpcall + |. nop + | move ra, TMP2 + | move TMP0, CRET1 + | move CRET1, CARG1 + |.if MIPSR6 + | selnez CRET1, CRET1, TMP0 + | seleqz TMP0, CARG2, TMP0 + | jr ra + |. or CRET1, CRET1, TMP0 + |.else + | jr ra + |. movz CRET1, CARG2, TMP0 + |.endif + |.endif + |.endmacro + | + | sfmin_max min, vm_sfcmpolt + | sfmin_max max, vm_sfcmpogt + | + |//----------------------------------------------------------------------- + |//-- Miscellaneous functions -------------------------------------------- + |//----------------------------------------------------------------------- + | + |.define NEXT_TAB, TAB:CARG1 + |.define NEXT_IDX, CARG2 + |.define NEXT_ASIZE, CARG3 + |.define NEXT_NIL, CARG4 + |.define NEXT_TMP0, r12 + |.define NEXT_TMP1, r13 + |.define NEXT_TMP2, r14 + |.define NEXT_RES_VK, CRET1 + |.define NEXT_RES_IDX, CRET2 + |.define NEXT_RES_PTR, sp + |.define NEXT_RES_VAL, 0(sp) + |.define NEXT_RES_KEY, 8(sp) + | + |// TValue *lj_vm_next(GCtab *t, uint32_t idx) + |// Next idx returned in CRET2. + |->vm_next: + |.if JIT and ENDIAN_LE + | lw NEXT_ASIZE, NEXT_TAB->asize + | ld NEXT_TMP0, NEXT_TAB->array + | li NEXT_NIL, LJ_TNIL + |1: // Traverse array part. + | sltu AT, NEXT_IDX, NEXT_ASIZE + | sll NEXT_TMP1, NEXT_IDX, 3 + | beqz AT, >5 + |. daddu NEXT_TMP1, NEXT_TMP0, NEXT_TMP1 + | li AT, LJ_TISNUM + | ld NEXT_TMP2, 0(NEXT_TMP1) + | dsll AT, AT, 47 + | or NEXT_TMP1, NEXT_IDX, AT + | beq NEXT_TMP2, NEXT_NIL, <1 + |. addiu NEXT_IDX, NEXT_IDX, 1 + | sd NEXT_TMP2, NEXT_RES_VAL + | sd NEXT_TMP1, NEXT_RES_KEY + | move NEXT_RES_VK, NEXT_RES_PTR + | jr ra + |. move NEXT_RES_IDX, NEXT_IDX + | + |5: // Traverse hash part. + | subu NEXT_RES_IDX, NEXT_IDX, NEXT_ASIZE + | ld NODE:NEXT_RES_VK, NEXT_TAB->node + | sll NEXT_TMP2, NEXT_RES_IDX, 5 + | lw NEXT_TMP0, NEXT_TAB->hmask + | sll AT, NEXT_RES_IDX, 3 + | subu AT, NEXT_TMP2, AT + | daddu NODE:NEXT_RES_VK, NODE:NEXT_RES_VK, AT + |6: + | sltu AT, NEXT_TMP0, NEXT_RES_IDX + | bnez AT, >8 + |. nop + | ld NEXT_TMP2, NODE:NEXT_RES_VK->val + | bne NEXT_TMP2, NEXT_NIL, >9 + |. addiu NEXT_RES_IDX, NEXT_RES_IDX, 1 + | // Skip holes in hash part. + | b <6 + |. daddiu NODE:NEXT_RES_VK, NODE:NEXT_RES_VK, sizeof(Node) + | + |8: // End of iteration. Set the key to nil (not the value). + | sd NEXT_NIL, NEXT_RES_KEY + | move NEXT_RES_VK, NEXT_RES_PTR + |9: + | jr ra + |. addu NEXT_RES_IDX, NEXT_RES_IDX, NEXT_ASIZE + |.endif + | + |//----------------------------------------------------------------------- + |//-- FFI helper functions ----------------------------------------------- + |//----------------------------------------------------------------------- + | + |// Handler for callback functions. Callback slot number in r1, g in r2. + |->vm_ffi_callback: + |.if FFI + |.type CTSTATE, CTState, PC + | saveregs + | ld CTSTATE, GL:r2->ctype_state + | daddiu DISPATCH, r2, GG_G2DISP + | load_got lj_ccallback_enter + | sw r1, CTSTATE->cb.slot + | sd CARG1, CTSTATE->cb.gpr[0] + | .FPU sdc1 FARG1, CTSTATE->cb.fpr[0] + | sd CARG2, CTSTATE->cb.gpr[1] + | .FPU sdc1 FARG2, CTSTATE->cb.fpr[1] + | sd CARG3, CTSTATE->cb.gpr[2] + | .FPU sdc1 FARG3, CTSTATE->cb.fpr[2] + | sd CARG4, CTSTATE->cb.gpr[3] + | .FPU sdc1 FARG4, CTSTATE->cb.fpr[3] + | sd CARG5, CTSTATE->cb.gpr[4] + | .FPU sdc1 FARG5, CTSTATE->cb.fpr[4] + | sd CARG6, CTSTATE->cb.gpr[5] + | .FPU sdc1 FARG6, CTSTATE->cb.fpr[5] + | sd CARG7, CTSTATE->cb.gpr[6] + | .FPU sdc1 FARG7, CTSTATE->cb.fpr[6] + | sd CARG8, CTSTATE->cb.gpr[7] + | .FPU sdc1 FARG8, CTSTATE->cb.fpr[7] + | daddiu TMP0, sp, CFRAME_SPACE + | sd TMP0, CTSTATE->cb.stack + | sd r0, SAVE_PC // Any value outside of bytecode is ok. + | move CARG2, sp + | call_intern lj_ccallback_enter // (CTState *cts, void *cf) + |. move CARG1, CTSTATE + | // Returns lua_State *. + | ld BASE, L:CRET1->base + | ld RC, L:CRET1->top + | move L, CRET1 + | .FPU lui TMP3, 0x59c0 // TOBIT = 2^52 + 2^51 (float). + | ld LFUNC:RB, FRAME_FUNC(BASE) + | .FPU mtc1 TMP3, TOBIT + | li TISNIL, LJ_TNIL + | li TISNUM, LJ_TISNUM + | li_vmstate INTERP + | subu RC, RC, BASE + | cleartp LFUNC:RB + | st_vmstate + | .FPU cvt.d.s TOBIT, TOBIT + | ins_callt + |.endif + | + |->cont_ffi_callback: // Return from FFI callback. + |.if FFI + | load_got lj_ccallback_leave + | ld CTSTATE, DISPATCH_GL(ctype_state)(DISPATCH) + | sd BASE, L->base + | sd RB, L->top + | sd L, CTSTATE->L + | move CARG2, RA + | call_intern lj_ccallback_leave // (CTState *cts, TValue *o) + |. move CARG1, CTSTATE + | .FPU ldc1 FRET1, CTSTATE->cb.fpr[0] + | ld CRET1, CTSTATE->cb.gpr[0] + | .FPU ldc1 FRET2, CTSTATE->cb.fpr[1] + | b ->vm_leave_unw + |. ld CRET2, CTSTATE->cb.gpr[1] + |.endif + | + |->vm_ffi_call: // Call C function via FFI. + | // Caveat: needs special frame unwinding, see below. + |.if FFI + | .type CCSTATE, CCallState, CARG1 + | lw TMP1, CCSTATE->spadj + | lbu CARG2, CCSTATE->nsp + | move TMP2, sp + | dsubu sp, sp, TMP1 + | sd ra, -8(TMP2) + | sd r16, -16(TMP2) + | sd CCSTATE, -24(TMP2) + | move r16, TMP2 + | daddiu TMP1, CCSTATE, offsetof(CCallState, stack) + | move TMP2, sp + | beqz CARG2, >2 + |. daddu TMP3, TMP1, CARG2 + |1: + | ld TMP0, 0(TMP1) + | daddiu TMP1, TMP1, 8 + | sltu AT, TMP1, TMP3 + | sd TMP0, 0(TMP2) + | bnez AT, <1 + |. daddiu TMP2, TMP2, 8 + |2: + | ld CFUNCADDR, CCSTATE->func + | .FPU ldc1 FARG1, CCSTATE->gpr[0] + | ld CARG2, CCSTATE->gpr[1] + | .FPU ldc1 FARG2, CCSTATE->gpr[1] + | ld CARG3, CCSTATE->gpr[2] + | .FPU ldc1 FARG3, CCSTATE->gpr[2] + | ld CARG4, CCSTATE->gpr[3] + | .FPU ldc1 FARG4, CCSTATE->gpr[3] + | ld CARG5, CCSTATE->gpr[4] + | .FPU ldc1 FARG5, CCSTATE->gpr[4] + | ld CARG6, CCSTATE->gpr[5] + | .FPU ldc1 FARG6, CCSTATE->gpr[5] + | ld CARG7, CCSTATE->gpr[6] + | .FPU ldc1 FARG7, CCSTATE->gpr[6] + | ld CARG8, CCSTATE->gpr[7] + | .FPU ldc1 FARG8, CCSTATE->gpr[7] + | jalr CFUNCADDR + |. ld CARG1, CCSTATE->gpr[0] // Do this last, since CCSTATE is CARG1. + | ld CCSTATE:TMP1, -24(r16) + | ld TMP2, -16(r16) + | ld ra, -8(r16) + | sd CRET1, CCSTATE:TMP1->gpr[0] + | sd CRET2, CCSTATE:TMP1->gpr[1] + |.if FPU + | sdc1 FRET1, CCSTATE:TMP1->fpr[0] + | sdc1 FRET2, CCSTATE:TMP1->fpr[1] + |.else + | sd CARG1, CCSTATE:TMP1->gpr[2] // 2nd FP struct field for soft-float. + |.endif + | move sp, r16 + | jr ra + |. move r16, TMP2 + |.endif + |// Note: vm_ffi_call must be the last function in this object file! + | + |//----------------------------------------------------------------------- +} + +/* Generate the code for a single instruction. */ +static void build_ins(BuildCtx *ctx, BCOp op, int defop) +{ + int vk = 0; + |=>defop: + + switch (op) { + + /* -- Comparison ops ---------------------------------------------------- */ + + /* Remember: all ops branch for a true comparison, fall through otherwise. */ + + case BC_ISLT: case BC_ISGE: case BC_ISLE: case BC_ISGT: + | // RA = src1*8, RD = src2*8, JMP with RD = target + |.macro bc_comp, FRA, FRD, ARGRA, ARGRD, movop, fmovop, fcomp, sfcomp + | daddu RA, BASE, RA + | daddu RD, BASE, RD + | ld ARGRA, 0(RA) + | ld ARGRD, 0(RD) + | lhu TMP2, OFS_RD(PC) + | gettp CARG3, ARGRA + | gettp CARG4, ARGRD + | bne CARG3, TISNUM, >2 + |. daddiu PC, PC, 4 + | bne CARG4, TISNUM, >5 + |. decode_RD4b TMP2 + | sextw ARGRA, ARGRA + | sextw ARGRD, ARGRD + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | slt AT, CARG1, CARG2 + | addu TMP2, TMP2, TMP3 + |.if MIPSR6 + | movop TMP2, TMP2, AT + |.else + | movop TMP2, r0, AT + |.endif + |1: + | daddu PC, PC, TMP2 + | ins_next + | + |2: // RA is not an integer. + | sltiu AT, CARG3, LJ_TISNUM + | beqz AT, ->vmeta_comp + |. lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | sltiu AT, CARG4, LJ_TISNUM + | beqz AT, >4 + |. decode_RD4b TMP2 + |.if FPU + | ldc1 FRA, 0(RA) + | ldc1 FRD, 0(RD) + |.endif + |3: // RA and RD are both numbers. + |.if FPU + |.if MIPSR6 + | fcomp FTMP0, FTMP0, FTMP2 + | addu TMP2, TMP2, TMP3 + | mfc1 TMP3, FTMP0 + | b <1 + |. fmovop TMP2, TMP2, TMP3 + |.else + | fcomp FTMP0, FTMP2 + | addu TMP2, TMP2, TMP3 + | b <1 + |. fmovop TMP2, r0 + |.endif + |.else + | bal sfcomp + |. addu TMP2, TMP2, TMP3 + | b <1 + |.if MIPSR6 + |. movop TMP2, TMP2, CRET1 + |.else + |. movop TMP2, r0, CRET1 + |.endif + |.endif + | + |4: // RA is a number, RD is not a number. + | bne CARG4, TISNUM, ->vmeta_comp + | // RA is a number, RD is an integer. Convert RD to a number. + |.if FPU + |. lwc1 FRD, LO(RD) + | ldc1 FRA, 0(RA) + | b <3 + |. cvt.d.w FRD, FRD + |.else + |.if "ARGRD" == "CARG1" + |. sextw CARG1, CARG1 + | bal ->vm_sfi2d_1 + |. nop + |.else + |. sextw CARG2, CARG2 + | bal ->vm_sfi2d_2 + |. nop + |.endif + | b <3 + |. nop + |.endif + | + |5: // RA is an integer, RD is not an integer + | sltiu AT, CARG4, LJ_TISNUM + | beqz AT, ->vmeta_comp + |. lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | // RA is an integer, RD is a number. Convert RA to a number. + |.if FPU + | lwc1 FRA, LO(RA) + | ldc1 FRD, 0(RD) + | b <3 + | cvt.d.w FRA, FRA + |.else + |.if "ARGRA" == "CARG1" + | bal ->vm_sfi2d_1 + |. sextw CARG1, CARG1 + |.else + | bal ->vm_sfi2d_2 + |. sextw CARG2, CARG2 + |.endif + | b <3 + |. nop + |.endif + |.endmacro + | + |.if MIPSR6 + if (op == BC_ISLT) { + | bc_comp FTMP0, FTMP2, CARG1, CARG2, selnez, selnez, cmp.lt.d, ->vm_sfcmpolt + } else if (op == BC_ISGE) { + | bc_comp FTMP0, FTMP2, CARG1, CARG2, seleqz, seleqz, cmp.lt.d, ->vm_sfcmpolt + } else if (op == BC_ISLE) { + | bc_comp FTMP2, FTMP0, CARG2, CARG1, seleqz, seleqz, cmp.ult.d, ->vm_sfcmpult + } else { + | bc_comp FTMP2, FTMP0, CARG2, CARG1, selnez, selnez, cmp.ult.d, ->vm_sfcmpult + } + |.else + if (op == BC_ISLT) { + | bc_comp FTMP0, FTMP2, CARG1, CARG2, movz, movf, c.olt.d, ->vm_sfcmpolt + } else if (op == BC_ISGE) { + | bc_comp FTMP0, FTMP2, CARG1, CARG2, movn, movt, c.olt.d, ->vm_sfcmpolt + } else if (op == BC_ISLE) { + | bc_comp FTMP2, FTMP0, CARG2, CARG1, movn, movt, c.ult.d, ->vm_sfcmpult + } else { + | bc_comp FTMP2, FTMP0, CARG2, CARG1, movz, movf, c.ult.d, ->vm_sfcmpult + } + |.endif + break; + + case BC_ISEQV: case BC_ISNEV: + vk = op == BC_ISEQV; + | // RA = src1*8, RD = src2*8, JMP with RD = target + | daddu RA, BASE, RA + | daddiu PC, PC, 4 + | daddu RD, BASE, RD + | ld CARG1, 0(RA) + | lhu TMP2, -4+OFS_RD(PC) + | ld CARG2, 0(RD) + | gettp CARG3, CARG1 + | gettp CARG4, CARG2 + | sltu AT, TISNUM, CARG3 + | sltu TMP1, TISNUM, CARG4 + | or AT, AT, TMP1 + if (vk) { + | beqz AT, ->BC_ISEQN_Z + } else { + | beqz AT, ->BC_ISNEN_Z + } + | // Either or both types are not numbers. + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + |.if FFI + |. li AT, LJ_TCDATA + | beq CARG3, AT, ->vmeta_equal_cd + |.endif + | decode_RD4b TMP2 + |.if FFI + | beq CARG4, AT, ->vmeta_equal_cd + |. nop + |.endif + | bne CARG1, CARG2, >2 + |. addu TMP2, TMP2, TMP3 + | // Tag and value are equal. + if (vk) { + |->BC_ISEQV_Z: + | daddu PC, PC, TMP2 + } + |1: + | ins_next + | + |2: // Check if the tags are the same and it's a table or userdata. + | xor AT, CARG3, CARG4 // Same type? + | sltiu TMP0, CARG3, LJ_TISTABUD+1 // Table or userdata? + |.if MIPSR6 + | seleqz TMP0, TMP0, AT + |.else + | movn TMP0, r0, AT + |.endif + if (vk) { + | beqz TMP0, <1 + } else { + | beqz TMP0, ->BC_ISEQV_Z // Reuse code from opposite instruction. + } + | // Different tables or userdatas. Need to check __eq metamethod. + | // Field metatable must be at same offset for GCtab and GCudata! + |. cleartp TAB:TMP1, CARG1 + | ld TAB:TMP3, TAB:TMP1->metatable + if (vk) { + | beqz TAB:TMP3, <1 // No metatable? + |. nop + | lbu TMP3, TAB:TMP3->nomm + | andi TMP3, TMP3, 1<1 // Or 'no __eq' flag set? + } else { + | beqz TAB:TMP3,->BC_ISEQV_Z // No metatable? + |. nop + | lbu TMP3, TAB:TMP3->nomm + | andi TMP3, TMP3, 1<BC_ISEQV_Z // Or 'no __eq' flag set? + } + |. nop + | b ->vmeta_equal // Handle __eq metamethod. + |. li TMP0, 1-vk // ne = 0 or 1. + break; + + case BC_ISEQS: case BC_ISNES: + vk = op == BC_ISEQS; + | // RA = src*8, RD = str_const*8 (~), JMP with RD = target + | daddu RA, BASE, RA + | daddiu PC, PC, 4 + | ld CARG1, 0(RA) + | dsubu RD, KBASE, RD + | lhu TMP2, -4+OFS_RD(PC) + | ld CARG2, -8(RD) // KBASE-8-str_const*8 + |.if FFI + | gettp TMP0, CARG1 + | li AT, LJ_TCDATA + |.endif + | li TMP1, LJ_TSTR + | decode_RD4b TMP2 + |.if FFI + | beq TMP0, AT, ->vmeta_equal_cd + |.endif + |. settp CARG2, TMP1 + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | xor TMP1, CARG1, CARG2 + | addu TMP2, TMP2, TMP3 + |.if MIPSR6 + if (vk) { + | seleqz TMP2, TMP2, TMP1 + } else { + | selnez TMP2, TMP2, TMP1 + } + |.else + if (vk) { + | movn TMP2, r0, TMP1 + } else { + | movz TMP2, r0, TMP1 + } + |.endif + | daddu PC, PC, TMP2 + | ins_next + break; + + case BC_ISEQN: case BC_ISNEN: + vk = op == BC_ISEQN; + | // RA = src*8, RD = num_const*8, JMP with RD = target + | daddu RA, BASE, RA + | daddu RD, KBASE, RD + | ld CARG1, 0(RA) + | ld CARG2, 0(RD) + | lhu TMP2, OFS_RD(PC) + | gettp CARG3, CARG1 + | gettp CARG4, CARG2 + | daddiu PC, PC, 4 + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + if (vk) { + |->BC_ISEQN_Z: + } else { + |->BC_ISNEN_Z: + } + | bne CARG3, TISNUM, >3 + |. decode_RD4b TMP2 + | bne CARG4, TISNUM, >6 + |. addu TMP2, TMP2, TMP3 + | xor AT, CARG1, CARG2 + |.if MIPSR6 + if (vk) { + | seleqz TMP2, TMP2, AT + |1: + | daddu PC, PC, TMP2 + |2: + } else { + | selnez TMP2, TMP2, AT + |1: + |2: + | daddu PC, PC, TMP2 + } + |.else + if (vk) { + | movn TMP2, r0, AT + |1: + | daddu PC, PC, TMP2 + |2: + } else { + | movz TMP2, r0, AT + |1: + |2: + | daddu PC, PC, TMP2 + } + |.endif + | ins_next + | + |3: // RA is not an integer. + | sltu AT, CARG3, TISNUM + |.if FFI + | beqz AT, >8 + |.else + | beqz AT, <2 + |.endif + |. addu TMP2, TMP2, TMP3 + | sltu AT, CARG4, TISNUM + |.if FPU + | ldc1 FTMP0, 0(RA) + | ldc1 FTMP2, 0(RD) + |.endif + | beqz AT, >5 + |. nop + |4: // RA and RD are both numbers. + |.if FPU + |.if MIPSR6 + | cmp.eq.d FTMP0, FTMP0, FTMP2 + | dmfc1 TMP1, FTMP0 + | b <1 + if (vk) { + |. selnez TMP2, TMP2, TMP1 + } else { + |. seleqz TMP2, TMP2, TMP1 + } + |.else + | c.eq.d FTMP0, FTMP2 + | b <1 + if (vk) { + |. movf TMP2, r0 + } else { + |. movt TMP2, r0 + } + |.endif + |.else + | bal ->vm_sfcmpeq + |. nop + | b <1 + |.if MIPSR6 + if (vk) { + |. selnez TMP2, TMP2, CRET1 + } else { + |. seleqz TMP2, TMP2, CRET1 + } + |.else + if (vk) { + |. movz TMP2, r0, CRET1 + } else { + |. movn TMP2, r0, CRET1 + } + |.endif + |.endif + | + |5: // RA is a number, RD is not a number. + |.if FFI + | bne CARG4, TISNUM, >9 + |.else + | bne CARG4, TISNUM, <2 + |.endif + | // RA is a number, RD is an integer. Convert RD to a number. + |.if FPU + |. lwc1 FTMP2, LO(RD) + | b <4 + |. cvt.d.w FTMP2, FTMP2 + |.else + |. sextw CARG2, CARG2 + | bal ->vm_sfi2d_2 + |. nop + | b <4 + |. nop + |.endif + | + |6: // RA is an integer, RD is not an integer + | sltu AT, CARG4, TISNUM + |.if FFI + | beqz AT, >9 + |.else + | beqz AT, <2 + |.endif + | // RA is an integer, RD is a number. Convert RA to a number. + |.if FPU + |. lwc1 FTMP0, LO(RA) + | ldc1 FTMP2, 0(RD) + | b <4 + | cvt.d.w FTMP0, FTMP0 + |.else + |. sextw CARG1, CARG1 + | bal ->vm_sfi2d_1 + |. nop + | b <4 + |. nop + |.endif + | + |.if FFI + |8: + | li AT, LJ_TCDATA + | bne CARG3, AT, <2 + |. nop + | b ->vmeta_equal_cd + |. nop + |9: + | li AT, LJ_TCDATA + | bne CARG4, AT, <2 + |. nop + | b ->vmeta_equal_cd + |. nop + |.endif + break; + + case BC_ISEQP: case BC_ISNEP: + vk = op == BC_ISEQP; + | // RA = src*8, RD = primitive_type*8 (~), JMP with RD = target + | daddu RA, BASE, RA + | srl TMP1, RD, 3 + | ld TMP0, 0(RA) + | lhu TMP2, OFS_RD(PC) + | not TMP1, TMP1 + | gettp TMP0, TMP0 + | daddiu PC, PC, 4 + |.if FFI + | li AT, LJ_TCDATA + | beq TMP0, AT, ->vmeta_equal_cd + |.endif + |. xor TMP0, TMP0, TMP1 + | decode_RD4b TMP2 + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | addu TMP2, TMP2, TMP3 + |.if MIPSR6 + if (vk) { + | seleqz TMP2, TMP2, TMP0 + } else { + | selnez TMP2, TMP2, TMP0 + } + |.else + if (vk) { + | movn TMP2, r0, TMP0 + } else { + | movz TMP2, r0, TMP0 + } + |.endif + | daddu PC, PC, TMP2 + | ins_next + break; + + /* -- Unary test and copy ops ------------------------------------------- */ + + case BC_ISTC: case BC_ISFC: case BC_IST: case BC_ISF: + | // RA = dst*8 or unused, RD = src*8, JMP with RD = target + | daddu RD, BASE, RD + | lhu TMP2, OFS_RD(PC) + | ld TMP0, 0(RD) + | daddiu PC, PC, 4 + | gettp TMP0, TMP0 + | sltiu TMP0, TMP0, LJ_TISTRUECOND + if (op == BC_IST || op == BC_ISF) { + | decode_RD4b TMP2 + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | addu TMP2, TMP2, TMP3 + |.if MIPSR6 + if (op == BC_IST) { + | selnez TMP2, TMP2, TMP0; + } else { + | seleqz TMP2, TMP2, TMP0; + } + |.else + if (op == BC_IST) { + | movz TMP2, r0, TMP0 + } else { + | movn TMP2, r0, TMP0 + } + |.endif + | daddu PC, PC, TMP2 + } else { + | ld CRET1, 0(RD) + if (op == BC_ISTC) { + | beqz TMP0, >1 + } else { + | bnez TMP0, >1 + } + |. daddu RA, BASE, RA + | decode_RD4b TMP2 + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | addu TMP2, TMP2, TMP3 + | sd CRET1, 0(RA) + | daddu PC, PC, TMP2 + |1: + } + | ins_next + break; + + case BC_ISTYPE: + | // RA = src*8, RD = -type*8 + | daddu TMP2, BASE, RA + | srl TMP1, RD, 3 + | ld TMP0, 0(TMP2) + | ins_next1 + | gettp TMP0, TMP0 + | daddu AT, TMP0, TMP1 + | bnez AT, ->vmeta_istype + |. ins_next2 + break; + case BC_ISNUM: + | // RA = src*8, RD = -(TISNUM-1)*8 + | daddu TMP2, BASE, RA + | ld TMP0, 0(TMP2) + | ins_next1 + | checknum TMP0, ->vmeta_istype + |. ins_next2 + break; + + /* -- Unary ops --------------------------------------------------------- */ + + case BC_MOV: + | // RA = dst*8, RD = src*8 + | daddu RD, BASE, RD + | daddu RA, BASE, RA + | ld CRET1, 0(RD) + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + break; + case BC_NOT: + | // RA = dst*8, RD = src*8 + | daddu RD, BASE, RD + | daddu RA, BASE, RA + | ld TMP0, 0(RD) + | li AT, LJ_TTRUE + | gettp TMP0, TMP0 + | sltu TMP0, AT, TMP0 + | addiu TMP0, TMP0, 1 + | dsll TMP0, TMP0, 47 + | not TMP0, TMP0 + | ins_next1 + | sd TMP0, 0(RA) + | ins_next2 + break; + case BC_UNM: + | // RA = dst*8, RD = src*8 + | daddu RB, BASE, RD + | ld CARG1, 0(RB) + | daddu RA, BASE, RA + | gettp CARG3, CARG1 + | bne CARG3, TISNUM, >2 + |. lui TMP1, 0x8000 + | sextw CARG1, CARG1 + | beq CARG1, TMP1, ->vmeta_unm // Meta handler deals with -2^31. + |. negu CARG1, CARG1 + | zextw CARG1, CARG1 + | settp CARG1, TISNUM + |1: + | ins_next1 + | sd CARG1, 0(RA) + | ins_next2 + |2: + | sltiu AT, CARG3, LJ_TISNUM + | beqz AT, ->vmeta_unm + |. dsll TMP1, TMP1, 32 + | b <1 + |. xor CARG1, CARG1, TMP1 + break; + case BC_LEN: + | // RA = dst*8, RD = src*8 + | daddu CARG2, BASE, RD + | daddu RA, BASE, RA + | ld TMP0, 0(CARG2) + | gettp TMP1, TMP0 + | daddiu AT, TMP1, -LJ_TSTR + | bnez AT, >2 + |. cleartp STR:CARG1, TMP0 + | lw CRET1, STR:CARG1->len + |1: + | settp CRET1, TISNUM + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + |2: + | daddiu AT, TMP1, -LJ_TTAB + | bnez AT, ->vmeta_len + |. nop +#if LJ_52 + | ld TAB:TMP2, TAB:CARG1->metatable + | bnez TAB:TMP2, >9 + |. nop + |3: +#endif + |->BC_LEN_Z: + | load_got lj_tab_len + | call_intern lj_tab_len // (GCtab *t) + |. nop + | // Returns uint32_t (but less than 2^31). + | b <1 + |. nop +#if LJ_52 + |9: + | lbu TMP0, TAB:TMP2->nomm + | andi TMP0, TMP0, 1<vmeta_len + |. nop +#endif + break; + + /* -- Binary ops -------------------------------------------------------- */ + + |.macro fpmod, a, b, c + | bal ->vm_floor // floor(b/c) + |. div.d FARG1, b, c + | mul.d a, FRET1, c + | sub.d a, b, a // b - floor(b/c)*c + |.endmacro + + |.macro sfpmod + | daddiu sp, sp, -16 + | + | load_got __divdf3 + | sd CARG1, 0(sp) + | call_extern + |. sd CARG2, 8(sp) + | + | load_got floor + | call_extern + |. move CARG1, CRET1 + | + | load_got __muldf3 + | move CARG1, CRET1 + | call_extern + |. ld CARG2, 8(sp) + | + | load_got __subdf3 + | ld CARG1, 0(sp) + | call_extern + |. move CARG2, CRET1 + | + | daddiu sp, sp, 16 + |.endmacro + + |.macro ins_arithpre, label + ||vk = ((int)op - BC_ADDVN) / (BC_ADDNV-BC_ADDVN); + | // RA = dst*8, RB = src1*8, RC = src2*8 | num_const*8 + ||switch (vk) { + ||case 0: + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | // RA = dst*8, RB = src1*8, RC = num_const*8 + | daddu RB, BASE, RB + |.if "label" ~= "none" + | b label + |.endif + |. daddu RC, KBASE, RC + || break; + ||case 1: + | decode_RB8a RC, INS + | decode_RB8b RC + | decode_RDtoRC8 RB, RD + | // RA = dst*8, RB = num_const*8, RC = src1*8 + | daddu RC, BASE, RC + |.if "label" ~= "none" + | b label + |.endif + |. daddu RB, KBASE, RB + || break; + ||default: + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | // RA = dst*8, RB = src1*8, RC = src2*8 + | daddu RB, BASE, RB + |.if "label" ~= "none" + | b label + |.endif + |. daddu RC, BASE, RC + || break; + ||} + |.endmacro + | + |.macro ins_arith, intins, fpins, fpcall, label + | ins_arithpre none + | + |.if "label" ~= "none" + |label: + |.endif + | + |// Used in 5. + | ld CARG1, 0(RB) + | ld CARG2, 0(RC) + | gettp TMP0, CARG1 + | gettp TMP1, CARG2 + | + |.if "intins" ~= "div" + | + | // Check for two integers. + | sextw CARG3, CARG1 + | bne TMP0, TISNUM, >5 + |. sextw CARG4, CARG2 + | bne TMP1, TISNUM, >5 + | + |.if "intins" == "addu" + |. intins CRET1, CARG3, CARG4 + | xor TMP1, CRET1, CARG3 // ((y^a) & (y^b)) < 0: overflow. + | xor TMP2, CRET1, CARG4 + | and TMP1, TMP1, TMP2 + | bltz TMP1, ->vmeta_arith + |. daddu RA, BASE, RA + |.elif "intins" == "subu" + |. intins CRET1, CARG3, CARG4 + | xor TMP1, CRET1, CARG3 // ((y^a) & (a^b)) < 0: overflow. + | xor TMP2, CARG3, CARG4 + | and TMP1, TMP1, TMP2 + | bltz TMP1, ->vmeta_arith + |. daddu RA, BASE, RA + |.elif "intins" == "mult" + |.if MIPSR6 + |. nop + | mul CRET1, CARG3, CARG4 + | muh TMP2, CARG3, CARG4 + |.else + |. intins CARG3, CARG4 + | mflo CRET1 + | mfhi TMP2 + |.endif + | sra TMP1, CRET1, 31 + | bne TMP1, TMP2, ->vmeta_arith + |. daddu RA, BASE, RA + |.else + |. load_got lj_vm_modi + | beqz CARG4, ->vmeta_arith + |. daddu RA, BASE, RA + | move CARG1, CARG3 + | call_extern + |. move CARG2, CARG4 + |.endif + | + | zextw CRET1, CRET1 + | settp CRET1, TISNUM + | ins_next1 + | sd CRET1, 0(RA) + |3: + | ins_next2 + | + |.endif + | + |5: // Check for two numbers. + | .FPU ldc1 FTMP0, 0(RB) + | sltu AT, TMP0, TISNUM + | sltu TMP0, TMP1, TISNUM + | .FPU ldc1 FTMP2, 0(RC) + | and AT, AT, TMP0 + | beqz AT, ->vmeta_arith + |. daddu RA, BASE, RA + | + |.if FPU + | fpins FRET1, FTMP0, FTMP2 + |.elif "fpcall" == "sfpmod" + | sfpmod + |.else + | load_got fpcall + | call_extern + |. nop + |.endif + | + | ins_next1 + |.if "intins" ~= "div" + | b <3 + |.endif + |.if FPU + |. sdc1 FRET1, 0(RA) + |.else + |. sd CRET1, 0(RA) + |.endif + |.if "intins" == "div" + | ins_next2 + |.endif + | + |.endmacro + + case BC_ADDVN: case BC_ADDNV: case BC_ADDVV: + | ins_arith addu, add.d, __adddf3, none + break; + case BC_SUBVN: case BC_SUBNV: case BC_SUBVV: + | ins_arith subu, sub.d, __subdf3, none + break; + case BC_MULVN: case BC_MULNV: case BC_MULVV: + | ins_arith mult, mul.d, __muldf3, none + break; + case BC_DIVVN: + | ins_arith div, div.d, __divdf3, ->BC_DIVVN_Z + break; + case BC_DIVNV: case BC_DIVVV: + | ins_arithpre ->BC_DIVVN_Z + break; + case BC_MODVN: + | ins_arith modi, fpmod, sfpmod, ->BC_MODVN_Z + break; + case BC_MODNV: case BC_MODVV: + | ins_arithpre ->BC_MODVN_Z + break; + case BC_POW: + | ins_arithpre none + | ld CARG1, 0(RB) + | ld CARG2, 0(RC) + | gettp TMP0, CARG1 + | gettp TMP1, CARG2 + | sltiu TMP0, TMP0, LJ_TISNUM + | sltiu TMP1, TMP1, LJ_TISNUM + | and AT, TMP0, TMP1 + | load_got pow + | beqz AT, ->vmeta_arith + |. daddu RA, BASE, RA + |.if FPU + | ldc1 FARG1, 0(RB) + | ldc1 FARG2, 0(RC) + |.endif + | call_extern + |. nop + | ins_next1 + |.if FPU + | sdc1 FRET1, 0(RA) + |.else + | sd CRET1, 0(RA) + |.endif + | ins_next2 + break; + + case BC_CAT: + | // RA = dst*8, RB = src_start*8, RC = src_end*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | dsubu CARG3, RC, RB + | sd BASE, L->base + | daddu CARG2, BASE, RC + | move MULTRES, RB + |->BC_CAT_Z: + | load_got lj_meta_cat + | srl CARG3, CARG3, 3 + | sd PC, SAVE_PC + | call_intern lj_meta_cat // (lua_State *L, TValue *top, int left) + |. move CARG1, L + | // Returns NULL (finished) or TValue * (metamethod). + | bnez CRET1, ->vmeta_binop + |. ld BASE, L->base + | daddu RB, BASE, MULTRES + | ld CRET1, 0(RB) + | daddu RA, BASE, RA + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + break; + + /* -- Constant ops ------------------------------------------------------ */ + + case BC_KSTR: + | // RA = dst*8, RD = str_const*8 (~) + | dsubu TMP1, KBASE, RD + | ins_next1 + | li TMP2, LJ_TSTR + | ld TMP0, -8(TMP1) // KBASE-8-str_const*8 + | daddu RA, BASE, RA + | settp TMP0, TMP2 + | sd TMP0, 0(RA) + | ins_next2 + break; + case BC_KCDATA: + |.if FFI + | // RA = dst*8, RD = cdata_const*8 (~) + | dsubu TMP1, KBASE, RD + | ins_next1 + | ld TMP0, -8(TMP1) // KBASE-8-cdata_const*8 + | li TMP2, LJ_TCDATA + | daddu RA, BASE, RA + | settp TMP0, TMP2 + | sd TMP0, 0(RA) + | ins_next2 + |.endif + break; + case BC_KSHORT: + | // RA = dst*8, RD = int16_literal*8 + | sra RD, INS, 16 + | daddu RA, BASE, RA + | zextw RD, RD + | ins_next1 + | settp RD, TISNUM + | sd RD, 0(RA) + | ins_next2 + break; + case BC_KNUM: + | // RA = dst*8, RD = num_const*8 + | daddu RD, KBASE, RD + | daddu RA, BASE, RA + | ld CRET1, 0(RD) + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + break; + case BC_KPRI: + | // RA = dst*8, RD = primitive_type*8 (~) + | daddu RA, BASE, RA + | dsll TMP0, RD, 44 + | not TMP0, TMP0 + | ins_next1 + | sd TMP0, 0(RA) + | ins_next2 + break; + case BC_KNIL: + | // RA = base*8, RD = end*8 + | daddu RA, BASE, RA + | sd TISNIL, 0(RA) + | daddiu RA, RA, 8 + | daddu RD, BASE, RD + |1: + | sd TISNIL, 0(RA) + | slt AT, RA, RD + | bnez AT, <1 + |. daddiu RA, RA, 8 + | ins_next_ + break; + + /* -- Upvalue and function ops ------------------------------------------ */ + + case BC_UGET: + | // RA = dst*8, RD = uvnum*8 + | ld LFUNC:RB, FRAME_FUNC(BASE) + | daddu RA, BASE, RA + | cleartp LFUNC:RB + | daddu RD, RD, LFUNC:RB + | ld UPVAL:RB, LFUNC:RD->uvptr + | ins_next1 + | ld TMP1, UPVAL:RB->v + | ld CRET1, 0(TMP1) + | sd CRET1, 0(RA) + | ins_next2 + break; + case BC_USETV: + | // RA = uvnum*8, RD = src*8 + | ld LFUNC:RB, FRAME_FUNC(BASE) + | daddu RD, BASE, RD + | cleartp LFUNC:RB + | daddu RA, RA, LFUNC:RB + | ld UPVAL:RB, LFUNC:RA->uvptr + | ld CRET1, 0(RD) + | lbu TMP3, UPVAL:RB->marked + | ld CARG2, UPVAL:RB->v + | andi TMP3, TMP3, LJ_GC_BLACK // isblack(uv) + | lbu TMP0, UPVAL:RB->closed + | gettp TMP2, CRET1 + | sd CRET1, 0(CARG2) + | li AT, LJ_GC_BLACK|1 + | or TMP3, TMP3, TMP0 + | beq TMP3, AT, >2 // Upvalue is closed and black? + |. daddiu TMP2, TMP2, -(LJ_TNUMX+1) + |1: + | ins_next + | + |2: // Check if new value is collectable. + | sltiu AT, TMP2, LJ_TISGCV - (LJ_TNUMX+1) + | beqz AT, <1 // tvisgcv(v) + |. cleartp GCOBJ:CRET1, CRET1 + | lbu TMP3, GCOBJ:CRET1->gch.marked + | andi TMP3, TMP3, LJ_GC_WHITES // iswhite(v) + | beqz TMP3, <1 + |. load_got lj_gc_barrieruv + | // Crossed a write barrier. Move the barrier forward. + | call_intern lj_gc_barrieruv // (global_State *g, TValue *tv) + |. daddiu CARG1, DISPATCH, GG_DISP2G + | b <1 + |. nop + break; + case BC_USETS: + | // RA = uvnum*8, RD = str_const*8 (~) + | ld LFUNC:RB, FRAME_FUNC(BASE) + | dsubu TMP1, KBASE, RD + | cleartp LFUNC:RB + | daddu RA, RA, LFUNC:RB + | ld UPVAL:RB, LFUNC:RA->uvptr + | ld STR:TMP1, -8(TMP1) // KBASE-8-str_const*8 + | lbu TMP2, UPVAL:RB->marked + | ld CARG2, UPVAL:RB->v + | lbu TMP3, STR:TMP1->marked + | andi AT, TMP2, LJ_GC_BLACK // isblack(uv) + | lbu TMP2, UPVAL:RB->closed + | li TMP0, LJ_TSTR + | settp TMP1, TMP0 + | bnez AT, >2 + |. sd TMP1, 0(CARG2) + |1: + | ins_next + | + |2: // Check if string is white and ensure upvalue is closed. + | beqz TMP2, <1 + |. andi AT, TMP3, LJ_GC_WHITES // iswhite(str) + | beqz AT, <1 + |. load_got lj_gc_barrieruv + | // Crossed a write barrier. Move the barrier forward. + | call_intern lj_gc_barrieruv // (global_State *g, TValue *tv) + |. daddiu CARG1, DISPATCH, GG_DISP2G + | b <1 + |. nop + break; + case BC_USETN: + | // RA = uvnum*8, RD = num_const*8 + | ld LFUNC:RB, FRAME_FUNC(BASE) + | daddu RD, KBASE, RD + | cleartp LFUNC:RB + | daddu RA, RA, LFUNC:RB + | ld UPVAL:RB, LFUNC:RA->uvptr + | ld CRET1, 0(RD) + | ld TMP1, UPVAL:RB->v + | ins_next1 + | sd CRET1, 0(TMP1) + | ins_next2 + break; + case BC_USETP: + | // RA = uvnum*8, RD = primitive_type*8 (~) + | ld LFUNC:RB, FRAME_FUNC(BASE) + | dsll TMP0, RD, 44 + | cleartp LFUNC:RB + | daddu RA, RA, LFUNC:RB + | not TMP0, TMP0 + | ld UPVAL:RB, LFUNC:RA->uvptr + | ins_next1 + | ld TMP1, UPVAL:RB->v + | sd TMP0, 0(TMP1) + | ins_next2 + break; + + case BC_UCLO: + | // RA = level*8, RD = target + | ld TMP2, L->openupval + | branch_RD // Do this first since RD is not saved. + | load_got lj_func_closeuv + | sd BASE, L->base + | beqz TMP2, >1 + |. move CARG1, L + | call_intern lj_func_closeuv // (lua_State *L, TValue *level) + |. daddu CARG2, BASE, RA + | ld BASE, L->base + |1: + | ins_next + break; + + case BC_FNEW: + | // RA = dst*8, RD = proto_const*8 (~) (holding function prototype) + | load_got lj_func_newL_gc + | dsubu TMP1, KBASE, RD + | ld CARG3, FRAME_FUNC(BASE) + | ld CARG2, -8(TMP1) // KBASE-8-tab_const*8 + | sd BASE, L->base + | sd PC, SAVE_PC + | cleartp CARG3 + | // (lua_State *L, GCproto *pt, GCfuncL *parent) + | call_intern lj_func_newL_gc + |. move CARG1, L + | // Returns GCfuncL *. + | li TMP0, LJ_TFUNC + | ld BASE, L->base + | ins_next1 + | settp CRET1, TMP0 + | daddu RA, BASE, RA + | sd CRET1, 0(RA) + | ins_next2 + break; + + /* -- Table ops --------------------------------------------------------- */ + + case BC_TNEW: + case BC_TDUP: + | // RA = dst*8, RD = (hbits|asize)*8 | tab_const*8 (~) + | ld TMP0, DISPATCH_GL(gc.total)(DISPATCH) + | ld TMP1, DISPATCH_GL(gc.threshold)(DISPATCH) + | sd BASE, L->base + | sd PC, SAVE_PC + | sltu AT, TMP0, TMP1 + | beqz AT, >5 + |1: + if (op == BC_TNEW) { + | load_got lj_tab_new + | srl CARG2, RD, 3 + | andi CARG2, CARG2, 0x7ff + | li TMP0, 0x801 + | addiu AT, CARG2, -0x7ff + | srl CARG3, RD, 14 + |.if MIPSR6 + | seleqz TMP0, TMP0, AT + | selnez CARG2, CARG2, AT + | or CARG2, CARG2, TMP0 + |.else + | movz CARG2, TMP0, AT + |.endif + | // (lua_State *L, int32_t asize, uint32_t hbits) + | call_intern lj_tab_new + |. move CARG1, L + | // Returns Table *. + } else { + | load_got lj_tab_dup + | dsubu TMP1, KBASE, RD + | move CARG1, L + | call_intern lj_tab_dup // (lua_State *L, Table *kt) + |. ld CARG2, -8(TMP1) // KBASE-8-str_const*8 + | // Returns Table *. + } + | li TMP0, LJ_TTAB + | ld BASE, L->base + | ins_next1 + | daddu RA, BASE, RA + | settp CRET1, TMP0 + | sd CRET1, 0(RA) + | ins_next2 + |5: + | load_got lj_gc_step_fixtop + | move MULTRES, RD + | call_intern lj_gc_step_fixtop // (lua_State *L) + |. move CARG1, L + | b <1 + |. move RD, MULTRES + break; + + case BC_GGET: + | // RA = dst*8, RD = str_const*8 (~) + case BC_GSET: + | // RA = src*8, RD = str_const*8 (~) + | ld LFUNC:TMP2, FRAME_FUNC(BASE) + | dsubu TMP1, KBASE, RD + | ld STR:RC, -8(TMP1) // KBASE-8-str_const*8 + | cleartp LFUNC:TMP2 + | ld TAB:RB, LFUNC:TMP2->env + if (op == BC_GGET) { + | b ->BC_TGETS_Z + } else { + | b ->BC_TSETS_Z + } + |. daddu RA, BASE, RA + break; + + case BC_TGETV: + | // RA = dst*8, RB = table*8, RC = key*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | daddu CARG2, BASE, RB + | daddu CARG3, BASE, RC + | ld TAB:RB, 0(CARG2) + | ld TMP2, 0(CARG3) + | daddu RA, BASE, RA + | checktab TAB:RB, ->vmeta_tgetv + | gettp TMP3, TMP2 + | bne TMP3, TISNUM, >5 // Integer key? + |. lw TMP0, TAB:RB->asize + | sextw TMP2, TMP2 + | ld TMP1, TAB:RB->array + | sltu AT, TMP2, TMP0 + | sll TMP2, TMP2, 3 + | beqz AT, ->vmeta_tgetv // Integer key and in array part? + |. daddu TMP2, TMP1, TMP2 + | ld AT, 0(TMP2) + | beq AT, TISNIL, >2 + |. ld CRET1, 0(TMP2) + |1: + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + | + |2: // Check for __index if table value is nil. + | ld TAB:TMP2, TAB:RB->metatable + | beqz TAB:TMP2, <1 // No metatable: done. + |. nop + | lbu TMP0, TAB:TMP2->nomm + | andi TMP0, TMP0, 1<vmeta_tgetv + |. nop + | + |5: + | li AT, LJ_TSTR + | bne TMP3, AT, ->vmeta_tgetv + |. cleartp RC, TMP2 + | b ->BC_TGETS_Z // String key? + |. nop + break; + case BC_TGETS: + | // RA = dst*8, RB = table*8, RC = str_const*8 (~) + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RC8a RC, INS + | daddu CARG2, BASE, RB + | decode_RC8b RC + | ld TAB:RB, 0(CARG2) + | dsubu CARG3, KBASE, RC + | daddu RA, BASE, RA + | ld STR:RC, -8(CARG3) // KBASE-8-str_const*8 + | checktab TAB:RB, ->vmeta_tgets1 + |->BC_TGETS_Z: + | // TAB:RB = GCtab *, STR:RC = GCstr *, RA = dst*8 + | lw TMP0, TAB:RB->hmask + | lw TMP1, STR:RC->sid + | ld NODE:TMP2, TAB:RB->node + | and TMP1, TMP1, TMP0 // idx = str->sid & tab->hmask + | sll TMP0, TMP1, 5 + | sll TMP1, TMP1, 3 + | subu TMP1, TMP0, TMP1 + | li TMP3, LJ_TSTR + | daddu NODE:TMP2, NODE:TMP2, TMP1 // node = tab->node + (idx*32-idx*8) + | settp STR:RC, TMP3 // Tagged key to look for. + |1: + | ld CARG1, NODE:TMP2->key + | ld CRET1, NODE:TMP2->val + | ld NODE:TMP1, NODE:TMP2->next + | bne CARG1, RC, >4 + |. ld TAB:TMP3, TAB:RB->metatable + | beq CRET1, TISNIL, >5 // Key found, but nil value? + |. nop + |3: + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + | + |4: // Follow hash chain. + | bnez NODE:TMP1, <1 + |. move NODE:TMP2, NODE:TMP1 + | // End of hash chain: key not found, nil result. + | + |5: // Check for __index if table value is nil. + | beqz TAB:TMP3, <3 // No metatable: done. + |. move CRET1, TISNIL + | lbu TMP0, TAB:TMP3->nomm + | andi TMP0, TMP0, 1<vmeta_tgets + |. nop + break; + case BC_TGETB: + | // RA = dst*8, RB = table*8, RC = index*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | daddu CARG2, BASE, RB + | decode_RDtoRC8 RC, RD + | ld TAB:RB, 0(CARG2) + | daddu RA, BASE, RA + | srl TMP0, RC, 3 + | checktab TAB:RB, ->vmeta_tgetb + | lw TMP1, TAB:RB->asize + | ld TMP2, TAB:RB->array + | sltu AT, TMP0, TMP1 + | beqz AT, ->vmeta_tgetb + |. daddu RC, TMP2, RC + | ld AT, 0(RC) + | beq AT, TISNIL, >5 + |. ld CRET1, 0(RC) + |1: + | ins_next1 + | sd CRET1, 0(RA) + | ins_next2 + | + |5: // Check for __index if table value is nil. + | ld TAB:TMP2, TAB:RB->metatable + | beqz TAB:TMP2, <1 // No metatable: done. + |. nop + | lbu TMP1, TAB:TMP2->nomm + | andi TMP1, TMP1, 1<vmeta_tgetb // Caveat: preserve TMP0 and CARG2! + |. nop + break; + case BC_TGETR: + | // RA = dst*8, RB = table*8, RC = key*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | daddu RB, BASE, RB + | daddu RC, BASE, RC + | ld TAB:CARG1, 0(RB) + | lw CARG2, LO(RC) + | daddu RA, BASE, RA + | cleartp TAB:CARG1 + | lw TMP0, TAB:CARG1->asize + | ld TMP1, TAB:CARG1->array + | sltu AT, CARG2, TMP0 + | sll TMP2, CARG2, 3 + | beqz AT, ->vmeta_tgetr // In array part? + |. daddu CRET1, TMP1, TMP2 + | ld CARG2, 0(CRET1) + |->BC_TGETR_Z: + | ins_next1 + | sd CARG2, 0(RA) + | ins_next2 + break; + + case BC_TSETV: + | // RA = src*8, RB = table*8, RC = key*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | daddu CARG2, BASE, RB + | daddu CARG3, BASE, RC + | ld RB, 0(CARG2) + | ld TMP2, 0(CARG3) + | daddu RA, BASE, RA + | checktab RB, ->vmeta_tsetv + | checkint TMP2, >5 + |. sextw RC, TMP2 + | lw TMP0, TAB:RB->asize + | ld TMP1, TAB:RB->array + | sltu AT, RC, TMP0 + | sll TMP2, RC, 3 + | beqz AT, ->vmeta_tsetv // Integer key and in array part? + |. daddu TMP1, TMP1, TMP2 + | ld TMP0, 0(TMP1) + | lbu TMP3, TAB:RB->marked + | beq TMP0, TISNIL, >3 + |. ld CRET1, 0(RA) + |1: + | andi AT, TMP3, LJ_GC_BLACK // isblack(table) + | bnez AT, >7 + |. sd CRET1, 0(TMP1) + |2: + | ins_next + | + |3: // Check for __newindex if previous value is nil. + | ld TAB:TMP2, TAB:RB->metatable + | beqz TAB:TMP2, <1 // No metatable: done. + |. nop + | lbu TMP2, TAB:TMP2->nomm + | andi TMP2, TMP2, 1<vmeta_tsetv + |. nop + | + |5: + | gettp AT, TMP2 + | daddiu AT, AT, -LJ_TSTR + | bnez AT, ->vmeta_tsetv + |. nop + | b ->BC_TSETS_Z // String key? + |. cleartp STR:RC, TMP2 + | + |7: // Possible table write barrier for the value. Skip valiswhite check. + | barrierback TAB:RB, TMP3, TMP0, <2 + break; + case BC_TSETS: + | // RA = src*8, RB = table*8, RC = str_const*8 (~) + | decode_RB8a RB, INS + | decode_RB8b RB + | daddu CARG2, BASE, RB + | decode_RC8a RC, INS + | ld TAB:RB, 0(CARG2) + | decode_RC8b RC + | dsubu CARG3, KBASE, RC + | ld RC, -8(CARG3) // KBASE-8-str_const*8 + | daddu RA, BASE, RA + | cleartp STR:RC + | checktab TAB:RB, ->vmeta_tsets1 + |->BC_TSETS_Z: + | // TAB:RB = GCtab *, STR:RC = GCstr *, RA = BASE+src*8 + | lw TMP0, TAB:RB->hmask + | lw TMP1, STR:RC->sid + | ld NODE:TMP2, TAB:RB->node + | sb r0, TAB:RB->nomm // Clear metamethod cache. + | and TMP1, TMP1, TMP0 // idx = str->sid & tab->hmask + | sll TMP0, TMP1, 5 + | sll TMP1, TMP1, 3 + | subu TMP1, TMP0, TMP1 + | li TMP3, LJ_TSTR + | daddu NODE:TMP2, NODE:TMP2, TMP1 // node = tab->node + (idx*32-idx*8) + | settp STR:RC, TMP3 // Tagged key to look for. + |.if FPU + | ldc1 FTMP0, 0(RA) + |.else + | ld CRET1, 0(RA) + |.endif + |1: + | ld TMP0, NODE:TMP2->key + | ld CARG2, NODE:TMP2->val + | ld NODE:TMP1, NODE:TMP2->next + | bne TMP0, RC, >5 + |. lbu TMP3, TAB:RB->marked + | beq CARG2, TISNIL, >4 // Key found, but nil value? + |. ld TAB:TMP0, TAB:RB->metatable + |2: + | andi AT, TMP3, LJ_GC_BLACK // isblack(table) + | bnez AT, >7 + |.if FPU + |. sdc1 FTMP0, NODE:TMP2->val + |.else + |. sd CRET1, NODE:TMP2->val + |.endif + |3: + | ins_next + | + |4: // Check for __newindex if previous value is nil. + | beqz TAB:TMP0, <2 // No metatable: done. + |. nop + | lbu TMP0, TAB:TMP0->nomm + | andi TMP0, TMP0, 1<vmeta_tsets + |. nop + | + |5: // Follow hash chain. + | bnez NODE:TMP1, <1 + |. move NODE:TMP2, NODE:TMP1 + | // End of hash chain: key not found, add a new one + | + | // But check for __newindex first. + | ld TAB:TMP2, TAB:RB->metatable + | beqz TAB:TMP2, >6 // No metatable: continue. + |. daddiu CARG3, DISPATCH, DISPATCH_GL(tmptv) + | lbu TMP0, TAB:TMP2->nomm + | andi TMP0, TMP0, 1<vmeta_tsets // 'no __newindex' flag NOT set: check. + |6: + | load_got lj_tab_newkey + | sd RC, 0(CARG3) + | sd BASE, L->base + | move CARG2, TAB:RB + | sd PC, SAVE_PC + | call_intern lj_tab_newkey // (lua_State *L, GCtab *t, TValue *k + |. move CARG1, L + | // Returns TValue *. + | ld BASE, L->base + |.if FPU + | b <3 // No 2nd write barrier needed. + |. sdc1 FTMP0, 0(CRET1) + |.else + | ld CARG1, 0(RA) + | b <3 // No 2nd write barrier needed. + |. sd CARG1, 0(CRET1) + |.endif + | + |7: // Possible table write barrier for the value. Skip valiswhite check. + | barrierback TAB:RB, TMP3, TMP0, <3 + break; + case BC_TSETB: + | // RA = src*8, RB = table*8, RC = index*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | daddu CARG2, BASE, RB + | decode_RDtoRC8 RC, RD + | ld TAB:RB, 0(CARG2) + | daddu RA, BASE, RA + | srl TMP0, RC, 3 + | checktab RB, ->vmeta_tsetb + | lw TMP1, TAB:RB->asize + | ld TMP2, TAB:RB->array + | sltu AT, TMP0, TMP1 + | beqz AT, ->vmeta_tsetb + |. daddu RC, TMP2, RC + | ld TMP1, 0(RC) + | lbu TMP3, TAB:RB->marked + | beq TMP1, TISNIL, >5 + |1: + |. ld CRET1, 0(RA) + | andi AT, TMP3, LJ_GC_BLACK // isblack(table) + | bnez AT, >7 + |. sd CRET1, 0(RC) + |2: + | ins_next + | + |5: // Check for __newindex if previous value is nil. + | ld TAB:TMP2, TAB:RB->metatable + | beqz TAB:TMP2, <1 // No metatable: done. + |. nop + | lbu TMP1, TAB:TMP2->nomm + | andi TMP1, TMP1, 1<vmeta_tsetb // Caveat: preserve TMP0 and CARG2! + |. nop + | + |7: // Possible table write barrier for the value. Skip valiswhite check. + | barrierback TAB:RB, TMP3, TMP0, <2 + break; + case BC_TSETR: + | // RA = dst*8, RB = table*8, RC = key*8 + | decode_RB8a RB, INS + | decode_RB8b RB + | decode_RDtoRC8 RC, RD + | daddu CARG1, BASE, RB + | daddu CARG3, BASE, RC + | ld TAB:CARG2, 0(CARG1) + | lw CARG3, LO(CARG3) + | cleartp TAB:CARG2 + | lbu TMP3, TAB:CARG2->marked + | lw TMP0, TAB:CARG2->asize + | ld TMP1, TAB:CARG2->array + | andi AT, TMP3, LJ_GC_BLACK // isblack(table) + | bnez AT, >7 + |. daddu RA, BASE, RA + |2: + | sltu AT, CARG3, TMP0 + | sll TMP2, CARG3, 3 + | beqz AT, ->vmeta_tsetr // In array part? + |. daddu CRET1, TMP1, TMP2 + |->BC_TSETR_Z: + | ld CARG1, 0(RA) + | ins_next1 + | sd CARG1, 0(CRET1) + | ins_next2 + | + |7: // Possible table write barrier for the value. Skip valiswhite check. + | barrierback TAB:CARG2, TMP3, CRET1, <2 + break; + + case BC_TSETM: + | // RA = base*8 (table at base-1), RD = num_const*8 (start index) + | daddu RA, BASE, RA + |1: + | daddu TMP3, KBASE, RD + | ld TAB:CARG2, -8(RA) // Guaranteed to be a table. + | addiu TMP0, MULTRES, -8 + | lw TMP3, LO(TMP3) // Integer constant is in lo-word. + | beqz TMP0, >4 // Nothing to copy? + |. srl CARG3, TMP0, 3 + | cleartp CARG2 + | addu CARG3, CARG3, TMP3 + | lw TMP2, TAB:CARG2->asize + | sll TMP1, TMP3, 3 + | lbu TMP3, TAB:CARG2->marked + | ld CARG1, TAB:CARG2->array + | sltu AT, TMP2, CARG3 + | bnez AT, >5 + |. daddu TMP2, RA, TMP0 + | daddu TMP1, TMP1, CARG1 + | andi TMP0, TMP3, LJ_GC_BLACK // isblack(table) + |3: // Copy result slots to table. + | ld CRET1, 0(RA) + | daddiu RA, RA, 8 + | sltu AT, RA, TMP2 + | sd CRET1, 0(TMP1) + | bnez AT, <3 + |. daddiu TMP1, TMP1, 8 + | bnez TMP0, >7 + |. nop + |4: + | ins_next + | + |5: // Need to resize array part. + | load_got lj_tab_reasize + | sd BASE, L->base + | sd PC, SAVE_PC + | move BASE, RD + | call_intern lj_tab_reasize // (lua_State *L, GCtab *t, int nasize) + |. move CARG1, L + | // Must not reallocate the stack. + | move RD, BASE + | b <1 + |. ld BASE, L->base // Reload BASE for lack of a saved register. + | + |7: // Possible table write barrier for any value. Skip valiswhite check. + | barrierback TAB:CARG2, TMP3, TMP0, <4 + break; + + /* -- Calls and vararg handling ----------------------------------------- */ + + case BC_CALLM: + | // RA = base*8, (RB = (nresults+1)*8,) RC = extra_nargs*8 + | decode_RDtoRC8 NARGS8:RC, RD + | b ->BC_CALL_Z + |. addu NARGS8:RC, NARGS8:RC, MULTRES + break; + case BC_CALL: + | // RA = base*8, (RB = (nresults+1)*8,) RC = (nargs+1)*8 + | decode_RDtoRC8 NARGS8:RC, RD + |->BC_CALL_Z: + | move TMP2, BASE + | daddu BASE, BASE, RA + | ld LFUNC:RB, 0(BASE) + | daddiu BASE, BASE, 16 + | addiu NARGS8:RC, NARGS8:RC, -8 + | checkfunc RB, ->vmeta_call + | ins_call + break; + + case BC_CALLMT: + | // RA = base*8, (RB = 0,) RC = extra_nargs*8 + | addu NARGS8:RD, NARGS8:RD, MULTRES // BC_CALLT gets RC from RD. + | // Fall through. Assumes BC_CALLT follows. + break; + case BC_CALLT: + | // RA = base*8, (RB = 0,) RC = (nargs+1)*8 + | daddu RA, BASE, RA + | ld RB, 0(RA) + | move NARGS8:RC, RD + | ld TMP1, FRAME_PC(BASE) + | daddiu RA, RA, 16 + | addiu NARGS8:RC, NARGS8:RC, -8 + | checktp CARG3, RB, -LJ_TFUNC, ->vmeta_callt + |->BC_CALLT_Z: + | andi TMP0, TMP1, FRAME_TYPE // Caveat: preserve TMP0 until the 'or'. + | lbu TMP3, LFUNC:CARG3->ffid + | bnez TMP0, >7 + |. xori TMP2, TMP1, FRAME_VARG + |1: + | sd RB, FRAME_FUNC(BASE) // Copy function down, but keep PC. + | sltiu AT, TMP3, 2 // (> FF_C) Calling a fast function? + | move TMP2, BASE + | move RB, CARG3 + | beqz NARGS8:RC, >3 + |. move TMP3, NARGS8:RC + |2: + | ld CRET1, 0(RA) + | daddiu RA, RA, 8 + | addiu TMP3, TMP3, -8 + | sd CRET1, 0(TMP2) + | bnez TMP3, <2 + |. daddiu TMP2, TMP2, 8 + |3: + | or TMP0, TMP0, AT + | beqz TMP0, >5 + |. nop + |4: + | ins_callt + | + |5: // Tailcall to a fast function with a Lua frame below. + | lw INS, -4(TMP1) + | decode_RA8a RA, INS + | decode_RA8b RA + | dsubu TMP1, BASE, RA + | ld TMP1, -32(TMP1) + | cleartp LFUNC:TMP1 + | ld TMP1, LFUNC:TMP1->pc + | b <4 + |. ld KBASE, PC2PROTO(k)(TMP1) // Need to prepare KBASE. + | + |7: // Tailcall from a vararg function. + | andi AT, TMP2, FRAME_TYPEP + | bnez AT, <1 // Vararg frame below? + |. dsubu TMP2, BASE, TMP2 // Relocate BASE down. + | move BASE, TMP2 + | ld TMP1, FRAME_PC(TMP2) + | b <1 + |. andi TMP0, TMP1, FRAME_TYPE + break; + + case BC_ITERC: + | // RA = base*8, (RB = (nresults+1)*8, RC = (nargs+1)*8 ((2+1)*8)) + | move TMP2, BASE // Save old BASE fir vmeta_call. + | daddu BASE, BASE, RA + | ld RB, -24(BASE) + | ld CARG1, -16(BASE) + | ld CARG2, -8(BASE) + | li NARGS8:RC, 16 // Iterators get 2 arguments. + | sd RB, 0(BASE) // Copy callable. + | sd CARG1, 16(BASE) // Copy state. + | sd CARG2, 24(BASE) // Copy control var. + | daddiu BASE, BASE, 16 + | checkfunc RB, ->vmeta_call + | ins_call + break; + + case BC_ITERN: + |.if JIT and ENDIAN_LE + | hotloop + |.endif + |->vm_IITERN: + | // RA = base*8, (RB = (nresults+1)*8, RC = (nargs+1)*8 (2+1)*8) + | daddu RA, BASE, RA + | ld TAB:RB, -16(RA) + | lw RC, -8+LO(RA) // Get index from control var. + | cleartp TAB:RB + | daddiu PC, PC, 4 + | lw TMP0, TAB:RB->asize + | ld TMP1, TAB:RB->array + | dsll CARG3, TISNUM, 47 + |1: // Traverse array part. + | sltu AT, RC, TMP0 + | beqz AT, >5 // Index points after array part? + |. sll TMP3, RC, 3 + | daddu TMP3, TMP1, TMP3 + | ld CARG1, 0(TMP3) + | lhu RD, -4+OFS_RD(PC) + | or TMP2, RC, CARG3 + | beq CARG1, TISNIL, <1 // Skip holes in array part. + |. addiu RC, RC, 1 + | sd TMP2, 0(RA) + | sd CARG1, 8(RA) + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | decode_RD4b RD + | daddu RD, RD, TMP3 + | sw RC, -8+LO(RA) // Update control var. + | daddu PC, PC, RD + |3: + | ins_next + | + |5: // Traverse hash part. + | lw TMP1, TAB:RB->hmask + | subu RC, RC, TMP0 + | ld TMP2, TAB:RB->node + |6: + | sltu AT, TMP1, RC // End of iteration? Branch to ITERL+1. + | bnez AT, <3 + |. sll TMP3, RC, 5 + | sll RB, RC, 3 + | subu TMP3, TMP3, RB + | daddu NODE:TMP3, TMP3, TMP2 + | ld CARG1, 0(NODE:TMP3) + | lhu RD, -4+OFS_RD(PC) + | beq CARG1, TISNIL, <6 // Skip holes in hash part. + |. addiu RC, RC, 1 + | ld CARG2, NODE:TMP3->key + | lui TMP3, (-(BCBIAS_J*4 >> 16) & 65535) + | sd CARG1, 8(RA) + | addu RC, RC, TMP0 + | decode_RD4b RD + | addu RD, RD, TMP3 + | sd CARG2, 0(RA) + | daddu PC, PC, RD + | b <3 + |. sw RC, -8+LO(RA) // Update control var. + break; + + case BC_ISNEXT: + | // RA = base*8, RD = target (points to ITERN) + | daddu RA, BASE, RA + | srl TMP0, RD, 1 + | ld CFUNC:CARG1, -24(RA) + | daddu TMP0, PC, TMP0 + | ld CARG2, -16(RA) + | ld CARG3, -8(RA) + | lui TMP2, (-(BCBIAS_J*4 >> 16) & 65535) + | checkfunc CFUNC:CARG1, >5 + | gettp CARG2, CARG2 + | daddiu CARG2, CARG2, -LJ_TTAB + | lbu TMP1, CFUNC:CARG1->ffid + | daddiu CARG3, CARG3, -LJ_TNIL + | or AT, CARG2, CARG3 + | daddiu TMP1, TMP1, -FF_next_N + | or AT, AT, TMP1 + | bnez AT, >5 + |. lui TMP1, (LJ_KEYINDEX >> 16) + | daddu PC, TMP0, TMP2 + | ori TMP1, TMP1, (LJ_KEYINDEX & 0xffff) + | dsll TMP1, TMP1, 32 + | sd TMP1, -8(RA) + |1: + | ins_next + |5: // Despecialize bytecode if any of the checks fail. + | li TMP3, BC_JMP + | li TMP1, BC_ITERC + | sb TMP3, -4+OFS_OP(PC) + | daddu PC, TMP0, TMP2 + |.if JIT + | lb TMP0, OFS_OP(PC) + | li AT, BC_ITERN + | bne TMP0, AT, >6 + |. lhu TMP2, OFS_RD(PC) + |.endif + | b <1 + |. sb TMP1, OFS_OP(PC) + |.if JIT + |6: // Unpatch JLOOP. + | ld TMP0, DISPATCH_J(trace)(DISPATCH) + | sll TMP2, TMP2, 3 + | daddu TMP0, TMP0, TMP2 + | ld TRACE:TMP2, 0(TMP0) + | lw TMP0, TRACE:TMP2->startins + | li AT, -256 + | and TMP0, TMP0, AT + | or TMP0, TMP0, TMP1 + | b <1 + |. sw TMP0, 0(PC) + |.endif + break; + + case BC_VARG: + | // RA = base*8, RB = (nresults+1)*8, RC = numparams*8 + | ld TMP0, FRAME_PC(BASE) + | decode_RDtoRC8 RC, RD + | decode_RB8a RB, INS + | daddu RC, BASE, RC + | decode_RB8b RB + | daddu RA, BASE, RA + | daddiu RC, RC, FRAME_VARG + | daddu TMP2, RA, RB + | daddiu TMP3, BASE, -16 // TMP3 = vtop + | dsubu RC, RC, TMP0 // RC = vbase + | // Note: RC may now be even _above_ BASE if nargs was < numparams. + | beqz RB, >5 // Copy all varargs? + |. dsubu TMP1, TMP3, RC + | daddiu TMP2, TMP2, -16 + |1: // Copy vararg slots to destination slots. + | ld CARG1, 0(RC) + | sltu AT, RC, TMP3 + | daddiu RC, RC, 8 + |.if MIPSR6 + | selnez CARG1, CARG1, AT + | seleqz AT, TISNIL, AT + | or CARG1, CARG1, AT + |.else + | movz CARG1, TISNIL, AT + |.endif + | sd CARG1, 0(RA) + | sltu AT, RA, TMP2 + | bnez AT, <1 + |. daddiu RA, RA, 8 + |3: + | ins_next + | + |5: // Copy all varargs. + | ld TMP0, L->maxstack + | blez TMP1, <3 // No vararg slots? + |. li MULTRES, 8 // MULTRES = (0+1)*8 + | daddu TMP2, RA, TMP1 + | sltu AT, TMP0, TMP2 + | bnez AT, >7 + |. daddiu MULTRES, TMP1, 8 + |6: + | ld CRET1, 0(RC) + | daddiu RC, RC, 8 + | sd CRET1, 0(RA) + | sltu AT, RC, TMP3 + | bnez AT, <6 // More vararg slots? + |. daddiu RA, RA, 8 + | b <3 + |. nop + | + |7: // Grow stack for varargs. + | load_got lj_state_growstack + | sd RA, L->top + | dsubu RA, RA, BASE + | sd BASE, L->base + | dsubu BASE, RC, BASE // Need delta, because BASE may change. + | sd PC, SAVE_PC + | srl CARG2, TMP1, 3 + | call_intern lj_state_growstack // (lua_State *L, int n) + |. move CARG1, L + | move RC, BASE + | ld BASE, L->base + | daddu RA, BASE, RA + | daddu RC, BASE, RC + | b <6 + |. daddiu TMP3, BASE, -16 + break; + + /* -- Returns ----------------------------------------------------------- */ + + case BC_RETM: + | // RA = results*8, RD = extra_nresults*8 + | addu RD, RD, MULTRES // MULTRES >= 8, so RD >= 8. + | // Fall through. Assumes BC_RET follows. + break; + + case BC_RET: + | // RA = results*8, RD = (nresults+1)*8 + | ld PC, FRAME_PC(BASE) + | daddu RA, BASE, RA + | move MULTRES, RD + |1: + | andi TMP0, PC, FRAME_TYPE + | bnez TMP0, ->BC_RETV_Z + |. xori TMP1, PC, FRAME_VARG + | + |->BC_RET_Z: + | // BASE = base, RA = resultptr, RD = (nresults+1)*8, PC = return + | lw INS, -4(PC) + | daddiu TMP2, BASE, -16 + | daddiu RC, RD, -8 + | decode_RA8a TMP0, INS + | decode_RB8a RB, INS + | decode_RA8b TMP0 + | decode_RB8b RB + | daddu TMP3, TMP2, RB + | beqz RC, >3 + |. dsubu BASE, TMP2, TMP0 + |2: + | ld CRET1, 0(RA) + | daddiu RA, RA, 8 + | daddiu RC, RC, -8 + | sd CRET1, 0(TMP2) + | bnez RC, <2 + |. daddiu TMP2, TMP2, 8 + |3: + | daddiu TMP3, TMP3, -8 + |5: + | sltu AT, TMP2, TMP3 + | bnez AT, >6 + |. ld LFUNC:TMP1, FRAME_FUNC(BASE) + | ins_next1 + | cleartp LFUNC:TMP1 + | ld TMP1, LFUNC:TMP1->pc + | ld KBASE, PC2PROTO(k)(TMP1) + | ins_next2 + | + |6: // Fill up results with nil. + | sd TISNIL, 0(TMP2) + | b <5 + |. daddiu TMP2, TMP2, 8 + | + |->BC_RETV_Z: // Non-standard return case. + | andi TMP2, TMP1, FRAME_TYPEP + | bnez TMP2, ->vm_return + |. nop + | // Return from vararg function: relocate BASE down. + | dsubu BASE, BASE, TMP1 + | b <1 + |. ld PC, FRAME_PC(BASE) + break; + + case BC_RET0: case BC_RET1: + | // RA = results*8, RD = (nresults+1)*8 + | ld PC, FRAME_PC(BASE) + | daddu RA, BASE, RA + | move MULTRES, RD + | andi TMP0, PC, FRAME_TYPE + | bnez TMP0, ->BC_RETV_Z + |. xori TMP1, PC, FRAME_VARG + | lw INS, -4(PC) + | daddiu TMP2, BASE, -16 + if (op == BC_RET1) { + | ld CRET1, 0(RA) + } + | decode_RB8a RB, INS + | decode_RA8a RA, INS + | decode_RB8b RB + | decode_RA8b RA + | dsubu BASE, TMP2, RA + if (op == BC_RET1) { + | sd CRET1, 0(TMP2) + } + |5: + | sltu AT, RD, RB + | bnez AT, >6 + |. ld TMP1, FRAME_FUNC(BASE) + | ins_next1 + | cleartp LFUNC:TMP1 + | ld TMP1, LFUNC:TMP1->pc + | ld KBASE, PC2PROTO(k)(TMP1) + | ins_next2 + | + |6: // Fill up results with nil. + | daddiu TMP2, TMP2, 8 + | daddiu RD, RD, 8 + | b <5 + if (op == BC_RET1) { + |. sd TISNIL, 0(TMP2) + } else { + |. sd TISNIL, -8(TMP2) + } + break; + + /* -- Loops and branches ------------------------------------------------ */ + + case BC_FORL: + |.if JIT + | hotloop + |.endif + | // Fall through. Assumes BC_IFORL follows. + break; + + case BC_JFORI: + case BC_JFORL: +#if !LJ_HASJIT + break; +#endif + case BC_FORI: + case BC_IFORL: + | // RA = base*8, RD = target (after end of loop or start of loop) + vk = (op == BC_IFORL || op == BC_JFORL); + | daddu RA, BASE, RA + | ld CARG1, FORL_IDX*8(RA) // IDX CARG1 - CARG3 type + | gettp CARG3, CARG1 + if (op != BC_JFORL) { + | srl RD, RD, 1 + | lui TMP2, (-(BCBIAS_J*4 >> 16) & 65535) + | daddu TMP2, RD, TMP2 + } + if (!vk) { + | ld CARG2, FORL_STOP*8(RA) // STOP CARG2 - CARG4 type + | ld CRET1, FORL_STEP*8(RA) // STEP CRET1 - CRET2 type + | gettp CARG4, CARG2 + | bne CARG3, TISNUM, >5 + |. gettp CRET2, CRET1 + | bne CARG4, TISNUM, ->vmeta_for + |. sextw CARG3, CARG1 + | bne CRET2, TISNUM, ->vmeta_for + |. sextw CARG2, CARG2 + | dext AT, CRET1, 31, 0 + | slt CRET1, CARG2, CARG3 + | slt TMP1, CARG3, CARG2 + |.if MIPSR6 + | selnez TMP1, TMP1, AT + | seleqz CRET1, CRET1, AT + | or CRET1, CRET1, TMP1 + |.else + | movn CRET1, TMP1, AT + |.endif + } else { + | bne CARG3, TISNUM, >5 + |. ld CARG2, FORL_STEP*8(RA) // STEP CARG2 - CARG4 type + | ld CRET1, FORL_STOP*8(RA) // STOP CRET1 - CRET2 type + | sextw TMP3, CARG1 + | sextw CARG2, CARG2 + | sextw CRET1, CRET1 + | addu CARG1, TMP3, CARG2 + | xor TMP0, CARG1, TMP3 + | xor TMP1, CARG1, CARG2 + | and TMP0, TMP0, TMP1 + | slt TMP1, CARG1, CRET1 + | slt CRET1, CRET1, CARG1 + | slt AT, CARG2, r0 + | slt TMP0, TMP0, r0 // ((y^a) & (y^b)) < 0: overflow. + |.if MIPSR6 + | selnez TMP1, TMP1, AT + | seleqz CRET1, CRET1, AT + | or CRET1, CRET1, TMP1 + |.else + | movn CRET1, TMP1, AT + |.endif + | or CRET1, CRET1, TMP0 + | zextw CARG1, CARG1 + | settp CARG1, TISNUM + } + |1: + if (op == BC_FORI) { + |.if MIPSR6 + | selnez TMP2, TMP2, CRET1 + |.else + | movz TMP2, r0, CRET1 + |.endif + | daddu PC, PC, TMP2 + } else if (op == BC_JFORI) { + | daddu PC, PC, TMP2 + | lhu RD, -4+OFS_RD(PC) + } else if (op == BC_IFORL) { + |.if MIPSR6 + | seleqz TMP2, TMP2, CRET1 + |.else + | movn TMP2, r0, CRET1 + |.endif + | daddu PC, PC, TMP2 + } + if (vk) { + | sd CARG1, FORL_IDX*8(RA) + } + | ins_next1 + | sd CARG1, FORL_EXT*8(RA) + |2: + if (op == BC_JFORI) { + | beqz CRET1, =>BC_JLOOP + |. decode_RD8b RD + } else if (op == BC_JFORL) { + | beqz CRET1, =>BC_JLOOP + } + | ins_next2 + | + |5: // FP loop. + |.if FPU + if (!vk) { + | ldc1 f0, FORL_IDX*8(RA) + | ldc1 f2, FORL_STOP*8(RA) + | sltiu TMP0, CARG3, LJ_TISNUM + | sltiu TMP1, CARG4, LJ_TISNUM + | sltiu AT, CRET2, LJ_TISNUM + | ld TMP3, FORL_STEP*8(RA) + | and TMP0, TMP0, TMP1 + | and AT, AT, TMP0 + | beqz AT, ->vmeta_for + |. slt TMP3, TMP3, r0 + |.if MIPSR6 + | dmtc1 TMP3, FTMP2 + | cmp.lt.d FTMP0, f0, f2 + | cmp.lt.d FTMP1, f2, f0 + | sel.d FTMP2, FTMP1, FTMP0 + | b <1 + |. dmfc1 CRET1, FTMP2 + |.else + | c.ole.d 0, f0, f2 + | c.ole.d 1, f2, f0 + | li CRET1, 1 + | movt CRET1, r0, 0 + | movt AT, r0, 1 + | b <1 + |. movn CRET1, AT, TMP3 + |.endif + } else { + | ldc1 f0, FORL_IDX*8(RA) + | ldc1 f4, FORL_STEP*8(RA) + | ldc1 f2, FORL_STOP*8(RA) + | ld TMP3, FORL_STEP*8(RA) + | add.d f0, f0, f4 + |.if MIPSR6 + | slt TMP3, TMP3, r0 + | dmtc1 TMP3, FTMP2 + | cmp.lt.d FTMP0, f0, f2 + | cmp.lt.d FTMP1, f2, f0 + | sel.d FTMP2, FTMP1, FTMP0 + | dmfc1 CRET1, FTMP2 + if (op == BC_IFORL) { + | seleqz TMP2, TMP2, CRET1 + | daddu PC, PC, TMP2 + } + |.else + | c.ole.d 0, f0, f2 + | c.ole.d 1, f2, f0 + | slt TMP3, TMP3, r0 + | li CRET1, 1 + | li AT, 1 + | movt CRET1, r0, 0 + | movt AT, r0, 1 + | movn CRET1, AT, TMP3 + if (op == BC_IFORL) { + | movn TMP2, r0, CRET1 + | daddu PC, PC, TMP2 + } + |.endif + | sdc1 f0, FORL_IDX*8(RA) + | ins_next1 + | b <2 + |. sdc1 f0, FORL_EXT*8(RA) + } + |.else + if (!vk) { + | sltiu TMP0, CARG3, LJ_TISNUM + | sltiu TMP1, CARG4, LJ_TISNUM + | sltiu AT, CRET2, LJ_TISNUM + | and TMP0, TMP0, TMP1 + | and AT, AT, TMP0 + | beqz AT, ->vmeta_for + |. nop + | bal ->vm_sfcmpolex + |. lw TMP3, FORL_STEP*8+HI(RA) + | b <1 + |. nop + } else { + | load_got __adddf3 + | call_extern + |. sw TMP2, TMPD + | ld CARG2, FORL_STOP*8(RA) + | move CARG1, CRET1 + if ( op == BC_JFORL ) { + | lhu RD, -4+OFS_RD(PC) + | decode_RD8b RD + } + | bal ->vm_sfcmpolex + |. lw TMP3, FORL_STEP*8+HI(RA) + | b <1 + |. lw TMP2, TMPD + } + |.endif + break; + + case BC_ITERL: + |.if JIT + | hotloop + |.endif + | // Fall through. Assumes BC_IITERL follows. + break; + + case BC_JITERL: +#if !LJ_HASJIT + break; +#endif + case BC_IITERL: + | // RA = base*8, RD = target + | daddu RA, BASE, RA + | ld TMP1, 0(RA) + | beq TMP1, TISNIL, >1 // Stop if iterator returned nil. + |. nop + if (op == BC_JITERL) { + | b =>BC_JLOOP + |. sd TMP1, -8(RA) + } else { + | branch_RD // Otherwise save control var + branch. + | sd TMP1, -8(RA) + } + |1: + | ins_next + break; + + case BC_LOOP: + | // RA = base*8, RD = target (loop extent) + | // Note: RA/RD is only used by trace recorder to determine scope/extent + | // This opcode does NOT jump, it's only purpose is to detect a hot loop. + |.if JIT + | hotloop + |.endif + | // Fall through. Assumes BC_ILOOP follows. + break; + + case BC_ILOOP: + | // RA = base*8, RD = target (loop extent) + | ins_next + break; + + case BC_JLOOP: + |.if JIT + | // RA = base*8 (ignored), RD = traceno*8 + | ld TMP1, DISPATCH_J(trace)(DISPATCH) + | li AT, 0 + | daddu TMP1, TMP1, RD + | // Traces on MIPS don't store the trace number, so use 0. + | sd AT, DISPATCH_GL(vmstate)(DISPATCH) + | ld TRACE:TMP2, 0(TMP1) + | sd BASE, DISPATCH_GL(jit_base)(DISPATCH) + | ld TMP2, TRACE:TMP2->mcode + | sd L, DISPATCH_GL(tmpbuf.L)(DISPATCH) + | jr TMP2 + |. daddiu JGL, DISPATCH, GG_DISP2G+32768 + |.endif + break; + + case BC_JMP: + | // RA = base*8 (only used by trace recorder), RD = target + | branch_RD + | ins_next + break; + + /* -- Function headers -------------------------------------------------- */ + + case BC_FUNCF: + |.if JIT + | hotcall + |.endif + case BC_FUNCV: /* NYI: compiled vararg functions. */ + | // Fall through. Assumes BC_IFUNCF/BC_IFUNCV follow. + break; + + case BC_JFUNCF: +#if !LJ_HASJIT + break; +#endif + case BC_IFUNCF: + | // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8 + | ld TMP2, L->maxstack + | lbu TMP1, -4+PC2PROTO(numparams)(PC) + | ld KBASE, -4+PC2PROTO(k)(PC) + | sltu AT, TMP2, RA + | bnez AT, ->vm_growstack_l + |. sll TMP1, TMP1, 3 + if (op != BC_JFUNCF) { + | ins_next1 + } + |2: + | sltu AT, NARGS8:RC, TMP1 // Check for missing parameters. + | bnez AT, >3 + |. daddu AT, BASE, NARGS8:RC + if (op == BC_JFUNCF) { + | decode_RD8a RD, INS + | b =>BC_JLOOP + |. decode_RD8b RD + } else { + | ins_next2 + } + | + |3: // Clear missing parameters. + | sd TISNIL, 0(AT) + | b <2 + |. addiu NARGS8:RC, NARGS8:RC, 8 + break; + + case BC_JFUNCV: +#if !LJ_HASJIT + break; +#endif + | NYI // NYI: compiled vararg functions + break; /* NYI: compiled vararg functions. */ + + case BC_IFUNCV: + | // BASE = new base, RA = BASE+framesize*8, RB = LFUNC, RC = nargs*8 + | li TMP0, LJ_TFUNC + | daddu TMP1, BASE, RC + | ld TMP2, L->maxstack + | settp LFUNC:RB, TMP0 + | daddu TMP0, RA, RC + | sd LFUNC:RB, 0(TMP1) // Store (tagged) copy of LFUNC. + | daddiu TMP2, TMP2, -8 + | daddiu TMP3, RC, 16+FRAME_VARG + | sltu AT, TMP0, TMP2 + | ld KBASE, -4+PC2PROTO(k)(PC) + | beqz AT, ->vm_growstack_l + |. sd TMP3, 8(TMP1) // Store delta + FRAME_VARG. + | lbu TMP2, -4+PC2PROTO(numparams)(PC) + | move RA, BASE + | move RC, TMP1 + | ins_next1 + | beqz TMP2, >3 + |. daddiu BASE, TMP1, 16 + |1: + | ld TMP0, 0(RA) + | sltu AT, RA, RC // Less args than parameters? + | move CARG1, TMP0 + |.if MIPSR6 + | selnez TMP0, TMP0, AT + | seleqz TMP3, TISNIL, AT + | or TMP0, TMP0, TMP3 + | seleqz TMP3, CARG1, AT + | selnez CARG1, TISNIL, AT + | or CARG1, CARG1, TMP3 + |.else + | movz TMP0, TISNIL, AT // Clear missing parameters. + | movn CARG1, TISNIL, AT // Clear old fixarg slot (help the GC). + |.endif + | addiu TMP2, TMP2, -1 + | sd TMP0, 16(TMP1) + | daddiu TMP1, TMP1, 8 + | sd CARG1, 0(RA) + | bnez TMP2, <1 + |. daddiu RA, RA, 8 + |3: + | ins_next2 + break; + + case BC_FUNCC: + case BC_FUNCCW: + | // BASE = new base, RA = BASE+framesize*8, RB = CFUNC, RC = nargs*8 + if (op == BC_FUNCC) { + | ld CFUNCADDR, CFUNC:RB->f + } else { + | ld CFUNCADDR, DISPATCH_GL(wrapf)(DISPATCH) + } + | daddu TMP1, RA, NARGS8:RC + | ld TMP2, L->maxstack + | daddu RC, BASE, NARGS8:RC + | sd BASE, L->base + | sltu AT, TMP2, TMP1 + | sd RC, L->top + | li_vmstate C + if (op == BC_FUNCCW) { + | ld CARG2, CFUNC:RB->f + } + | bnez AT, ->vm_growstack_c // Need to grow stack. + |. move CARG1, L + | jalr CFUNCADDR // (lua_State *L [, lua_CFunction f]) + |. st_vmstate + | // Returns nresults. + | ld BASE, L->base + | sll RD, CRET1, 3 + | ld TMP1, L->top + | li_vmstate INTERP + | ld PC, FRAME_PC(BASE) // Fetch PC of caller. + | dsubu RA, TMP1, RD // RA = L->top - nresults*8 + | sd L, DISPATCH_GL(cur_L)(DISPATCH) + | b ->vm_returnc + |. st_vmstate + break; + + /* ---------------------------------------------------------------------- */ + + default: + fprintf(stderr, "Error: undefined opcode BC_%s\n", bc_names[op]); + exit(2); + break; + } +} + +static int build_backend(BuildCtx *ctx) +{ + int op; + + dasm_growpc(Dst, BC__MAX); + + build_subroutines(ctx); + + |.code_op + for (op = 0; op < BC__MAX; op++) + build_ins(ctx, (BCOp)op, op); + + return BC__MAX; +} + +/* Emit pseudo frame-info for all assembler functions. */ +static void emit_asm_debug(BuildCtx *ctx) +{ + int fcofs = (int)((uint8_t *)ctx->glob[GLOB_vm_ffi_call] - ctx->code); + int i; + switch (ctx->mode) { + case BUILD_elfasm: + fprintf(ctx->fp, "\t.section .debug_frame,\"\",@progbits\n"); + fprintf(ctx->fp, + ".Lframe0:\n" + "\t.4byte .LECIE0-.LSCIE0\n" + ".LSCIE0:\n" + "\t.4byte 0xffffffff\n" + "\t.byte 0x1\n" + "\t.string \"\"\n" + "\t.uleb128 0x1\n" + "\t.sleb128 -4\n" + "\t.byte 31\n" + "\t.byte 0xc\n\t.uleb128 29\n\t.uleb128 0\n" + "\t.align 2\n" + ".LECIE0:\n\n"); + fprintf(ctx->fp, + ".LSFDE0:\n" + "\t.4byte .LEFDE0-.LASFDE0\n" + ".LASFDE0:\n" + "\t.4byte .Lframe0\n" + "\t.8byte .Lbegin\n" + "\t.8byte %d\n" + "\t.byte 0xe\n\t.uleb128 %d\n" + "\t.byte 0x9f\n\t.sleb128 2*5\n" + "\t.byte 0x9e\n\t.sleb128 2*6\n", + fcofs, CFRAME_SIZE); + for (i = 23; i >= 16; i--) + fprintf(ctx->fp, "\t.byte %d\n\t.uleb128 %d\n", 0x80+i, 2*(30-i)); +#if !LJ_SOFTFP + for (i = 31; i >= 24; i--) + fprintf(ctx->fp, "\t.byte %d\n\t.uleb128 %d\n", 0x80+32+i, 2*(46-i)); +#endif + fprintf(ctx->fp, + "\t.align 2\n" + ".LEFDE0:\n\n"); +#if LJ_HASFFI + fprintf(ctx->fp, + ".LSFDE1:\n" + "\t.4byte .LEFDE1-.LASFDE1\n" + ".LASFDE1:\n" + "\t.4byte .Lframe0\n" + "\t.4byte lj_vm_ffi_call\n" + "\t.4byte %d\n" + "\t.byte 0x9f\n\t.uleb128 2*1\n" + "\t.byte 0x90\n\t.uleb128 2*2\n" + "\t.byte 0xd\n\t.uleb128 0x10\n" + "\t.align 2\n" + ".LEFDE1:\n\n", (int)ctx->codesz - fcofs); +#endif +#if !LJ_NO_UNWIND + /* NYI */ +#endif + break; + default: + break; + } +} + diff --git a/src/lua_baselib.c b/src/lua_baselib.c index 059df924c..b15cf2cf0 100644 --- a/src/lua_baselib.c +++ b/src/lua_baselib.c @@ -42,6 +42,123 @@ boolean luaL_checkboolean(lua_State *L, int narg) { return lua_toboolean(L, narg); } +static int ref_open, ref_iowrite, ref_fwrite; // original versions of io functions + +#define FILELIMIT 1024*1024 // Size limit for reading/writing files + +static const char *whitelist[] = { // Allow scripters to write files of these types to SRB2's folder + ".txt", + ".sav2", + ".cfg", + ".png", + ".bmp" +}; + +static int StartsWith(const char *a, const char *b) // this is wolfs being lazy yet again +{ + if(strncmp(a, b, strlen(b)) == 0) return 1; + return 0; +}; + +// Wrapper for opening files +static int lib_open(lua_State *L) +{ + const char *filename = luaL_checkstring(L, 1); + int pass = 0; + size_t i; + int length = strlen(filename); + char *splitter, *forward, *backward; + char *destFilename; + + for (i = 0; i < (sizeof (whitelist) / sizeof(const char *)); i++) + { + if (!stricmp(&filename[length - strlen(whitelist[i])], whitelist[i])) + { + pass = 1; + break; + } + } + if (strstr(filename, "..") || strchr(filename, ':') || StartsWith(filename, "\\") + || StartsWith(filename, "/") || !pass) + { + return luaL_error(L,"access denied to %s", filename); + } + + destFilename = va("%s"PATHSEP"luafiles"PATHSEP"%s", srb2home, filename); + + // Make directories as needed + splitter = destFilename; + + forward = strchr(splitter, '/'); + backward = strchr(splitter, '\\'); + while ((splitter = (forward && backward) ? min(forward, backward) : (forward ?: backward))) + { + *splitter = 0; + I_mkdir(destFilename, 0755); + *splitter = '/'; + splitter++; + + forward = strchr(splitter, '/'); + backward = strchr(splitter, '\\'); + } + + // replace the filename string + lua_pushstring(L, destFilename); + lua_replace(L, 1); + // now call the real io.open + lua_rawgeti(L, LUA_REGISTRYINDEX, ref_open); + lua_insert(L, 1); + lua_call(L, lua_gettop(L)-1, LUA_MULTRET); + return lua_gettop(L); +} + +// Wrapper for writing to files +static int lib_write(lua_State *L, int arg) +{ + int top = lua_gettop(L); + boolean error = true; + size_t len, size = 0; + int ogref = ref_iowrite; + + if (arg != 1) { + ogref = ref_fwrite; + FILE *fp = *(FILE **)luaL_checkudata(L, 1, LUA_FILEHANDLE); + if (fp == NULL) goto call; // real file:write will throw + size = ftell(fp); + } + + // check the size of each argument. if there's too much data, + // cut off the stack and print an error + for (; arg <= top; arg++) { + if (lua_tolstring(L, arg, &len) == NULL) goto call; + if ((size += len) > FILELIMIT) { + // oops! too many bytes! + lua_settop(L, arg-1); + goto call; + } + } + error = false; + +call: + lua_rawgeti(L, LUA_REGISTRYINDEX, ogref); + lua_insert(L, 1); + lua_call(L, lua_gettop(L)-1, LUA_MULTRET); + + if (error) + return luaL_error(L,"write limit bypassed in file. Changes have been discarded."); + return lua_gettop(L); +} + +static int lib_write_f(lua_State *L) +{ + return lib_write(L, 2); +} + +static int lib_write_io(lua_State *L) +{ + return lib_write(L, 1); +} + // String concatination static int lib_concat(lua_State *L) { @@ -3348,6 +3465,28 @@ int LUA_BaseLib(lua_State *L) lua_setfield(L,-2,"__add"); // ... store it as mathematical addition lua_pop(L, 2); // pop metatable and dummy string + // replaces an existing function with another, + // saving the original as a reference +#define REPLACE(name, func, ref) \ + lua_pushliteral(L, name); \ + lua_rawget(L, -2); \ + ref = luaL_ref(L, LUA_REGISTRYINDEX); \ + lua_pushliteral(L, name); \ + lua_pushcclosure(L, &func, 0); \ + lua_rawset(L, -3); + + // replace io.open + lua_getglobal(L, "io"); + REPLACE("open", lib_open, ref_open) + REPLACE("write", lib_write_io, ref_iowrite) + lua_pop(L, 1); + + // replace file:write + luaL_getmetatable(L, LUA_FILEHANDLE); + REPLACE("write", lib_write_f, ref_fwrite) + lua_pop(L, 1); +#undef REPLACE + lua_newtable(L); lua_setfield(L, LUA_REGISTRYINDEX, LREG_EXTVARS); diff --git a/src/lua_consolelib.c b/src/lua_consolelib.c index e1e246d28..2f85c7154 100644 --- a/src/lua_consolelib.c +++ b/src/lua_consolelib.c @@ -324,6 +324,8 @@ static int lib_cvRegisterVar(lua_State *L) const char* category = NULL; const char* menu_name = NULL; + boolean hasfunc = false; + int funcref = 0; luaL_checktype(L, 1, LUA_TTABLE); lua_settop(L, 1); // Clear out all other possible arguments, leaving only the first one. @@ -409,12 +411,11 @@ static int lib_cvRegisterVar(lua_State *L) } else if (i == 5 || (k && fasticmp(k, "func"))) { if (!lua_isfunction(L, 4)) TYPEERROR("func", LUA_TFUNCTION) - lua_getfield(L, LUA_REGISTRYINDEX, "CV_OnChange"); - I_Assert(lua_istable(L, 5)); - lua_pushvalue(L, 4); - lua_setfield(L, 5, cvar->name); - lua_pop(L, 1); - cvar->func = Lua_OnChange; + if (hasfunc) + luaL_unref(L, LUA_REGISTRYINDEX, funcref); // real clever, huh? + funcref = luaL_ref(L, LUA_REGISTRYINDEX); + hasfunc = true; + continue; } else if (((i == 5 && !(cvar->flags & CV_CALL)) || (cvar->flags & CV_CALL && i == 6)) @@ -434,6 +435,16 @@ static int lib_cvRegisterVar(lua_State *L) #undef TYPEERROR // stack: cvar table, cvar userdata + if (hasfunc) { + lua_getfield(L, LUA_REGISTRYINDEX, "CV_OnChange"); + I_Assert(lua_istable(L, 3)); + lua_rawgeti(L, LUA_REGISTRYINDEX, funcref); + lua_setfield(L, 3, cvar->name); + lua_pop(L, 1); + cvar->func = Lua_OnChange; + luaL_unref(L, LUA_REGISTRYINDEX, funcref); + } + lua_getfield(L, LUA_REGISTRYINDEX, "CV_Vars"); I_Assert(lua_istable(L, 3));