Use case insensitive string hashing and compare functions for cvar hashtable

oftentimes lua scripts call CV_FindVar a fuckton of times during runtime, with many cvars this can actually tank performance more than one might think
ideally lua scripts should cache the result, but with so much stuff and servers we cant expect everyone to go through all scripts to do that
using a hashtable for CV_FindNetVar might improve join times aswell, but needs further testing to confirm

src/command.c

@@ -1165,6 +1165,39 @@ void VS_Print(vsbuf_t *buf, const char *data)
 //
 // =========================================================================
 
+#define NAME      cvar_map_t
+#define KEY_TY    const char *
+#define VAL_TY    consvar_t *
+#define HASH_FN   FNV1a_HashLowercaseString
+#define CMPR_FN   vt_cmpr_casestring
+#include "verstable.h"
+
+#define NAME      netvar_map_t
+#define KEY_TY    UINT16
+#define VAL_TY    consvar_t *
+#define HASH_FN   vt_hash_integer
+#define CMPR_FN   vt_cmpr_integer
+#include "verstable.h"
+
+static cvar_map_t cvar_map;
+static netvar_map_t netvar_map;
+
+CONSTRUCTOR static void CV_InitMap(void)
+{
+	// ensure the map is initialized
+	cvar_map_t_init(&cvar_map);
+	cvar_map_t_reserve(&cvar_map, 512);
+
+	netvar_map_t_init(&netvar_map);
+	netvar_map_t_reserve(&netvar_map, 256);
+}
+
+DESTRUCTOR static void CV_DestroyMap(void)
+{
+	cvar_map_t_cleanup(&cvar_map);
+	netvar_map_t_cleanup(&netvar_map);
+}
+
 static const char *cv_null_string = "";
 
 /** Searches if a variable has been registered.
@@ -1175,11 +1208,17 @@ static const char *cv_null_string = "";
   */
 consvar_t *CV_FindVar(const char *name)
 {
-	consvar_t *cvar;
+	cvar_map_t_itr it = cvar_map_t_get(&cvar_map, name);
+	if (!cvar_map_t_is_end(it))
+		return it.data->val;
 
+	// fallback linear search
+	/*
+	consvar_t *cvar;
 	for (cvar = consvar_vars; cvar; cvar = cvar->next)
 		if (fasticmp(name, cvar->name))
 			return cvar;
+	*/
 
 	return NULL;
 }
@@ -1213,11 +1252,17 @@ static inline UINT16 CV_ComputeNetid(const char *s)
   */
 static consvar_t *CV_FindNetVar(UINT16 netid)
 {
+	netvar_map_t_itr it = netvar_map_t_get(&netvar_map, netid);
+	if (!netvar_map_t_is_end(it))
+		return it.data->val;
+
+	/*
 	consvar_t *cvar;
 
 	for (cvar = consvar_vars; cvar; cvar = cvar->next)
 		if (cvar->netid == netid)
 			return cvar;
+	*/
 
 	if (netid == 44542) // ouch this hack
 		return &cv_karteliminatelast;
@@ -1283,6 +1328,11 @@ void CV_RegisterVar(consvar_t *variable)
 
 	// the SetValue will set this bit
 	variable->flags &= ~CV_MODIFIED;
+
+	cvar_map_t_insert(&cvar_map, variable->name, variable);
+
+	if (variable->flags & CV_NETVAR)
+		netvar_map_t_insert(&netvar_map, variable->netid, variable);
 }
 
 /** Finds the string value of a console variable.
@@ -1588,9 +1638,21 @@ size_t CV_LoadNetVars(const UINT8 *bufstart)
 	// prevent "invalid command received"
 	serverloading = true;
 
+	// we can use our netvar map instead of going through all cvars each time
+	netvar_map_t_itr it;
+	for (it = netvar_map_t_first(&netvar_map);
+		 !netvar_map_t_is_end(it);
+		 it = netvar_map_t_next(it))
+	{
+		cvar = it.data->val;
+		Setvalue(cvar, cvar->defaultvalue, true);
+	}
+
+	/*
 	for (cvar = consvar_vars; cvar; cvar = cvar->next)
 		if (cvar->flags & CV_NETVAR)
 			Setvalue(cvar, cvar->defaultvalue, true);
+	*/
 
 	count = READUINT16(p);
 	while (count--)

src/dedicated/i_system.c

@@ -1384,7 +1384,6 @@ static void I_Fork(void)
 
 INT32 I_StartupSystem(void)
 {
-	W_Startup();
 	I_StartupConsole();
 #ifdef NEWSIGNALHANDLER
 	// This is useful when debugging. It lets GDB attach to

src/doomdef.h

@@ -573,6 +573,13 @@ INT32 I_GetKey(void);
 // NOTE: you WILL have nasal troubles if the variable is not initialized
 #define CLEANUP(f) __attribute__((__cleanup__(f)))
 
+// the GNU constructor and destructor attributes
+// The constructor attribute causes the function to be called automatically before execution enters main ().
+// Similarly, the destructor attribute causes the function to be called automatically after main () has completed or exit () has been called.
+// Functions with these attributes are useful for initializing data that will be used implicitly during the execution of the program.
+#define CONSTRUCTOR __attribute__((constructor, used))
+#define DESTRUCTOR __attribute__((destructor, used))
+
 // An assert-type mechanism.
 #ifdef PARANOIA
 #define I_Assert(e) ((e) ? (void)0 : I_Error("assert failed: %s, file %s, line %d", #e, __FILE__, __LINE__))

src/hardware/hw_defs.h

@@ -110,7 +110,7 @@ typedef struct
 typedef struct
 {
 	poly_t *planepoly;  // the generated convex polygon
-} poly_subsector_t;
+} extrasubsector_t;
 
 // Vanilla features
 //#define USE_MODEL_NEXTFRAME

src/hardware/hw_glob.h

@@ -52,12 +52,12 @@ void HWR_ObjectLightLevelPost(gl_vissprite_t *spr, const sector_t *sector, INT32
 // --------
 // hw_bsp.c
 // --------
-extern poly_subsector_t *poly_subsectors;
+extern extrasubsector_t *extrasubsectors;
 #ifdef PARANOIA
-extern size_t num_poly_subsector;
+extern size_t addsubsector;
 #endif
-void HWR_InitPolyPool(void);
-void HWR_FreePolyPool(void);
+
+void HWR_FreeExtraSubsectors(void);
 
 // --------
 // hw_cache.c

src/hardware/hw_main.cpp

@@ -125,7 +125,7 @@ ps_metric_t ps_hw_batchdrawtime = {};
 static void HWR_SplitWall(sector_t *sector, FOutVector *wallVerts, INT32 texnum, boolean noencore, FSurfaceInfo* Surf, INT32 cutflag, ffloor_t *pfloor, FBITFIELD polyflags);
 static void HWR_RenderWall(FOutVector *wallVerts, FSurfaceInfo *pSurf, FBITFIELD blend, boolean fogwall, INT32 lightlevel, extracolormap_t *wallcolormap);
 
-static void HWR_AddTransparentFloor(lumpnum_t lumpnum, poly_subsector_t *xsub, boolean isceiling, fixed_t fixedheight, INT32 lightlevel, INT32 alpha, sector_t *FOFSector, FBITFIELD blend, boolean fogplane, extracolormap_t *planecolormap);
+static void HWR_AddTransparentFloor(lumpnum_t lumpnum, extrasubsector_t *xsub, boolean isceiling, fixed_t fixedheight, INT32 lightlevel, INT32 alpha, sector_t *FOFSector, FBITFIELD blend, boolean fogplane, extracolormap_t *planecolormap);
 static void HWR_AddTransparentWall(FOutVector *wallVerts, FSurfaceInfo *pSurf, INT32 texnum, boolean noencore, FBITFIELD blend, boolean fogwall, INT32 lightlevel, extracolormap_t *wallcolormap);
 static void HWR_AddTransparentPolyobjectFloor(lumpnum_t lumpnum, polyobj_t *polysector, boolean isceiling, fixed_t fixedheight, INT32 lightlevel, INT32 alpha, sector_t *FOFSector, FBITFIELD blend, extracolormap_t *planecolormap);
 
@@ -231,17 +231,6 @@ consvar_t cv_glanisotropicmode = {"gr_anisotropicmode", "1", CV_CALL|CV_SAVE, gl
 
 consvar_t cv_glsolvetjoin = {"gr_solvetjoin", "On", 0, CV_OnOff, NULL, 0, NULL, NULL, 0, 0, NULL};
 
-consvar_t cv_glpolytile = {"gr_polytile", "On", 0, CV_OnOff, NULL, 0, NULL, NULL, 0, 0, NULL};
-
-CV_PossibleValue_t grpolyshape_cons_t[] = {
-	{0, "Subsector"},
-	{1, "Fat"},
-	{2, "Trim"},
-	{3, "NotConvex"},
-	{0, NULL}
-};
-consvar_t cv_glpolyshape = {"gr_polygon_shape", "Trim", CV_SAVE, grpolyshape_cons_t, NULL, 0, NULL, NULL, 0, 0, NULL};
-
 consvar_t cv_glbatching = {"gr_batching", "On", 0, CV_OnOff, NULL, 0, NULL, NULL, 0, 0, NULL};
 
 consvar_t cv_glwireframe = {"gr_wireframe", "Off", 0, CV_OnOff, NULL, 0, NULL, NULL, 0, 0, NULL};
@@ -559,7 +548,7 @@ static FUINT HWR_CalcSlopeLight(FUINT lightnum, pslope_t *slope, const sector_t
 
 // HWR_RenderPlane
 // Render a floor or ceiling convex polygon
-static void HWR_RenderPlane(subsector_t *subsector, poly_subsector_t *xsub, boolean isceiling, fixed_t fixedheight, FBITFIELD PolyFlags, INT32 lightlevel, lumpnum_t lumpnum, sector_t *FOFsector, UINT8 alpha, extracolormap_t *planecolormap)
+static void HWR_RenderPlane(subsector_t *subsector, extrasubsector_t *xsub, boolean isceiling, fixed_t fixedheight, FBITFIELD PolyFlags, INT32 lightlevel, lumpnum_t lumpnum, sector_t *FOFsector, UINT8 alpha, extracolormap_t *planecolormap)
 {
 	FSurfaceInfo Surf = {};
 	FOutVector *v3d;
@@ -2832,7 +2821,7 @@ static void HWR_AddPolyObjectPlanes(void)
 	sector_t *polyobjsector;
 	INT32 light = 0;
 
-	// Polyobject Planes need their own function for drawing because they don't have poly_subsectors by themselves
+	// Polyobject Planes need their own function for drawing because they don't have extrasubsectors by themselves
 	// It should be okay because polyobjects should always be convex anyway
 
 	for (i = 0; i < numpolys; i++)
@@ -2960,9 +2949,9 @@ static void HWR_Subsector(size_t num)
 	ffloor_t *rover;
 
 #ifdef PARANOIA // no risk while developing, enough debugging nights!
-	if (num >= num_poly_subsector)
+	if (num >= addsubsector)
 		I_Error("HWR_Subsector: ss %s with numss = %s, addss = %s\n",
-			sizeu1(num), sizeu2(numsubsectors), sizeu3(num_poly_subsector));
+			sizeu1(num), sizeu2(numsubsectors), sizeu3(addsubsector));
 #endif
 
 	if (num < numsubsectors)
@@ -3057,7 +3046,7 @@ static void HWR_Subsector(size_t num)
 			if (sub->validcount != validcount)
 			{
 				HWR_GetFlat(levelflats[gl_frontsector->floorpic].lumpnum, R_NoEncore(gl_frontsector, false));
-				HWR_RenderPlane(sub, &poly_subsectors[num], false,
+				HWR_RenderPlane(sub, &extrasubsectors[num], false,
 					gl_frontsector->floorheight,
 					// We now return you to your regularly scheduled rendering.
 					PF_Occlude, floorlightlevel, levelflats[gl_frontsector->floorpic].lumpnum, NULL, 255, floorcolormap);
@@ -3072,7 +3061,7 @@ static void HWR_Subsector(size_t num)
 			if (sub->validcount != validcount)
 			{
 				HWR_GetFlat(levelflats[gl_frontsector->ceilingpic].lumpnum, R_NoEncore(gl_frontsector, true));
-				HWR_RenderPlane(sub, &poly_subsectors[num], true,
+				HWR_RenderPlane(sub, &extrasubsectors[num], true,
 					gl_frontsector->ceilingheight,
 					// We now return you to your regularly scheduled rendering.
 					PF_Occlude, ceilinglightlevel, levelflats[gl_frontsector->ceilingpic].lumpnum, NULL, 255, ceilingcolormap);
@@ -3119,7 +3108,7 @@ static void HWR_Subsector(size_t num)
 					alpha = HWR_FogBlockAlpha(*gl_frontsector->lightlist[light].lightlevel, rover->master->frontsector->extra_colormap);
 
 					HWR_AddTransparentFloor(0,
-											&poly_subsectors[num],
+											&extrasubsectors[num],
 											false,
 											*rover->bottomheight,
 											*gl_frontsector->lightlist[light].lightlevel,
@@ -3131,7 +3120,7 @@ static void HWR_Subsector(size_t num)
 					light = R_GetPlaneLight(gl_frontsector, centerHeight, (viewz < bottomCullHeight));
 
 					HWR_AddTransparentFloor(levelflats[*rover->bottompic].lumpnum,
-											&poly_subsectors[num],
+											&extrasubsectors[num],
 											false,
 											*rover->bottomheight,
 											*gl_frontsector->lightlist[light].lightlevel,
@@ -3143,7 +3132,7 @@ static void HWR_Subsector(size_t num)
 					HWR_GetFlat(levelflats[*rover->bottompic].lumpnum, R_NoEncore(gl_frontsector, false));
 					light = R_GetPlaneLight(gl_frontsector, centerHeight, (viewz < bottomCullHeight));
 
-					HWR_RenderPlane(sub, &poly_subsectors[num], false, *rover->bottomheight, HWR_RippleBlend(gl_frontsector, rover, false)|PF_Occlude, *gl_frontsector->lightlist[light].lightlevel, levelflats[*rover->bottompic].lumpnum,
+					HWR_RenderPlane(sub, &extrasubsectors[num], false, *rover->bottomheight, HWR_RippleBlend(gl_frontsector, rover, false)|PF_Occlude, *gl_frontsector->lightlist[light].lightlevel, levelflats[*rover->bottompic].lumpnum,
 									rover->master->frontsector, 255, gl_frontsector->lightlist[light].extra_colormap);
 				}
 			}
@@ -3165,7 +3154,7 @@ static void HWR_Subsector(size_t num)
 					alpha = HWR_FogBlockAlpha(*gl_frontsector->lightlist[light].lightlevel, rover->master->frontsector->extra_colormap);
 
 					HWR_AddTransparentFloor(0,
-											&poly_subsectors[num],
+											&extrasubsectors[num],
 											true,
 											*rover->topheight,
 											*gl_frontsector->lightlist[light].lightlevel,
@@ -3177,7 +3166,7 @@ static void HWR_Subsector(size_t num)
 					light = R_GetPlaneLight(gl_frontsector, centerHeight, (viewz < topCullHeight));
 
 					HWR_AddTransparentFloor(levelflats[*rover->toppic].lumpnum,
-											&poly_subsectors[num],
+											&extrasubsectors[num],
 											true,
 											*rover->topheight,
 											*gl_frontsector->lightlist[light].lightlevel,
@@ -3189,7 +3178,7 @@ static void HWR_Subsector(size_t num)
 					HWR_GetFlat(levelflats[*rover->toppic].lumpnum, R_NoEncore(gl_frontsector, true));
 					light = R_GetPlaneLight(gl_frontsector, centerHeight, (viewz < topCullHeight));
 
-					HWR_RenderPlane(sub, &poly_subsectors[num], true, *rover->topheight, HWR_RippleBlend(gl_frontsector, rover, false)|PF_Occlude, *gl_frontsector->lightlist[light].lightlevel, levelflats[*rover->toppic].lumpnum,
+					HWR_RenderPlane(sub, &extrasubsectors[num], true, *rover->topheight, HWR_RippleBlend(gl_frontsector, rover, false)|PF_Occlude, *gl_frontsector->lightlist[light].lightlevel, levelflats[*rover->toppic].lumpnum,
 									rover->master->frontsector, 255, gl_frontsector->lightlist[light].extra_colormap);
 				}
 			}
@@ -4270,7 +4259,7 @@ typedef struct
 
 typedef struct
 {
-	poly_subsector_t *xsub;
+	extrasubsector_t *xsub;
 	boolean isceiling;
 	fixed_t fixedheight;
 	INT32 lightlevel;
@@ -4354,7 +4343,7 @@ static void HWR_AddTransparentWall(FOutVector *wallVerts, FSurfaceInfo *pSurf, I
 	wallinfo->wallcolormap = wallcolormap;
 }
 
-static void HWR_AddTransparentFloor(lumpnum_t lumpnum, poly_subsector_t *xsub, boolean isceiling, fixed_t fixedheight, INT32 lightlevel, INT32 alpha, sector_t *FOFSector, FBITFIELD blend, boolean fogplane, extracolormap_t *planecolormap)
+static void HWR_AddTransparentFloor(lumpnum_t lumpnum, extrasubsector_t *xsub, boolean isceiling, fixed_t fixedheight, INT32 lightlevel, INT32 alpha, sector_t *FOFSector, FBITFIELD blend, boolean fogplane, extracolormap_t *planecolormap)
 {
 	planeinfo_t *planeinfo = static_cast<planeinfo_t*>(HWR_CreateDrawNode(DRAWNODE_PLANE));
 
@@ -5824,7 +5813,6 @@ void HWR_RenderPlayerView(void)
 
 void HWR_LoadLevel(void)
 {
-	HWR_FreePolyPool();
 	HWR_CreatePlanePolygons((INT32)numnodes - 1);
 
 	// Build the sky dome
@@ -5909,8 +5897,6 @@ void HWR_AddCommands(void)
 	CV_RegisterVar(&cv_glanisotropicmode);
 
 	CV_RegisterVar(&cv_glsolvetjoin);
-	CV_RegisterVar(&cv_glpolytile);
-	CV_RegisterVar(&cv_glpolyshape);
 
 	CV_RegisterVar(&cv_glbatching);
 
@@ -5938,7 +5924,6 @@ void HWR_Startup(void)
 		CONS_Printf("HWR_Startup()...\n");
 		textureformat = patchformat = GL_TEXFMT_RGBA;
 
-		HWR_InitPolyPool();
 		HWR_InitMapTextures();
 		HWR_InitMD2();
 
@@ -6030,7 +6015,7 @@ static void COM_HWR_glinfo(void)
 void HWR_Shutdown(void)
 {
 	CONS_Printf("HWR_Shutdown()\n");
-	HWR_FreePolyPool();
+	HWR_FreeExtraSubsectors();
 	HWR_FreeMapTextures();
 	GL_FlushScreenTextures();
 	HWR_ClearAllTextures();

src/hardware/hw_main.h

@@ -195,7 +195,7 @@ extern consvar_t cv_gllightdither;
 extern consvar_t cv_glfiltermode;
 extern consvar_t cv_glanisotropicmode;
 
-extern consvar_t cv_glsolvetjoin, cv_glpolytile, cv_glpolyshape;
+extern consvar_t cv_glsolvetjoin;
 
 extern consvar_t cv_glbatching;
 

src/sdl/i_system.cpp

@@ -1887,8 +1887,6 @@ INT32 I_StartupSystem(void)
 	SDL_VERSION(&SDLcompiled)
 	SDL_GetVersion(&SDLlinked);
 
-	W_Startup();
-
 	I_StartupConsole();
 #ifdef NEWSIGNALHANDLER
 	// This is useful when debugging. It lets GDB attach to

src/verstable.h

@@ -462,6 +462,8 @@ License (MIT):
 #include <stdbool.h>
 #include <string.h>
 
+#include "fastcmp.h"
+
 // Two-way concatenation macro.
 #define VT_CAT_( a, b ) a##b
 #define VT_CAT( a, b ) VT_CAT_( a, b )
@@ -705,7 +707,12 @@ static inline FORCE_INLINE bool vt_cmpr_integer( uint64_t key_1, uint64_t key_2
 
 static inline FORCE_INLINE bool vt_cmpr_string( const char *key_1, const char *key_2 )
 {
-  return strcmp( key_1, key_2 ) == 0;
+  return fastcmp( key_1, key_2 );
+}
+
+static inline FORCE_INLINE bool vt_cmpr_casestring( const char *key_1, const char *key_2 )
+{
+  return fasticmp( key_1, key_2 );
 }
 
 // Default allocation and free functions.

src/w_wad.c

@@ -158,8 +158,7 @@ void W_Shutdown(void)
 }
 
 // the lumpnum cache needs to be initialized before use
-// call this as early as possible!
-void W_Startup(void)
+CONSTRUCTOR static void W_Startup(void)
 {
 	lumpnum_map_init(&lumpnumcache);
 	lumpnum_map_reserve(&lumpnumcache, LUMPNUMCACHESIZE);
@@ -367,7 +366,7 @@ static inline INT32 W_MakeFileMD5(const char *filename, void *resblock)
 	{
 		precise_t t = I_GetPreciseTime();
 
-		CONS_Debug(DBG_SETUP, "Making MD5 for %s\n",filename);
+		CONS_Debug(DBG_SETUP, "Making MD5 for %s\n", filename);
 
 		if (md5_stream(fhandle, resblock) == 1)
 		{

src/w_wad.h

@@ -154,7 +154,6 @@ extern wadfile_t *wadfiles[MAX_WADFILES];
 
 // =========================================================================
 
-void W_Startup(void);
 void W_Shutdown(void);
 
 // Opens a WAD file. Returns the FILE * handle for the file, or NULL if not found or could not be opened