From 524168b20b3a32d26cc9ca38f982940eda1a468e Mon Sep 17 00:00:00 2001 From: Alug Date: Thu, 2 Oct 2025 17:15:14 +0200 Subject: [PATCH 1/2] experiment: remove opengl adjusted segs in theory this should improve levelload times and reduce memory usage? gotta see if it really does, visually so far there seems to be no difference at all --- src/hardware/hw_bsp.c | 103 ---------------------------------- src/hardware/hw_main.cpp | 116 ++++++++------------------------------- src/p_setup.c | 11 ++++ src/r_defs.h | 12 ++-- 4 files changed, 43 insertions(+), 199 deletions(-) diff --git a/src/hardware/hw_bsp.c b/src/hardware/hw_bsp.c index 8cad91168..5889dcb44 100644 --- a/src/hardware/hw_bsp.c +++ b/src/hardware/hw_bsp.c @@ -80,14 +80,6 @@ static poly_t *HWR_AllocPoly(INT32 numpts) return p; } -static polyvertex_t *HWR_AllocVertex(void) -{ - polyvertex_t *p; - size_t size = sizeof(polyvertex_t); - p = Z_Malloc(size, PU_HWRPLANE, NULL); - return p; -} - /// \todo polygons should be freed in reverse order for efficiency, /// for now don't free because it doesn't free in reverse order static void HWR_FreePoly(poly_t *poly) @@ -855,100 +847,6 @@ static INT32 SolveTProblem(void) return numsplitpoly; } -#define NEARDIST (0.75f) -#define MYMAX (10000000000000.0f) - -/* Adjust true segs (from the segs lump) to be exactely the same as - * plane polygone segs - * This also convert fixed_t point of segs in float (in moste case - * it share the same vertice - */ -static void AdjustSegs(void) -{ - size_t i, count; - INT32 j; - seg_t *lseg; - poly_t *p; - INT32 v1found = 0, v2found = 0; - float nearv1, nearv2; - - for (i = 0; i < numsubsectors; i++) - { - count = subsectors[i].numlines; - lseg = &segs[subsectors[i].firstline]; - p = extrasubsectors[i].planepoly; - - if (!p) - continue; - - for (; count--; lseg++) - { - float distv1,distv2,tmp; - nearv1 = nearv2 = MYMAX; - - // Don't touch polyobject segs. We'll compensate - // for this when we go about drawing them. - if (lseg->polyseg) - continue; - - for (j = 0; j < p->numpts; j++) - { - distv1 = p->pts[j].x - FIXED_TO_FLOAT(lseg->v1->x); - tmp = p->pts[j].y - FIXED_TO_FLOAT(lseg->v1->y); - distv1 = distv1*distv1+tmp*tmp; - - if (distv1 <= nearv1) - { - v1found = j; - nearv1 = distv1; - } - - // the same with v2 - distv2 = p->pts[j].x - FIXED_TO_FLOAT(lseg->v2->x); - tmp = p->pts[j].y - FIXED_TO_FLOAT(lseg->v2->y); - distv2 = distv2*distv2+tmp*tmp; - - if (distv2 <= nearv2) - { - v2found = j; - nearv2 = distv2; - } - } - - if (nearv1 <= NEARDIST*NEARDIST) - // share vertice with segs - lseg->pv1 = &(p->pts[v1found]); - else - { - // BP: here we can do better, using PointInSeg and compute - // the right point position also split a polygone side to - // solve a T-intersection, but too mush work - - // convert fixed vertex to float vertex - polyvertex_t *pv = HWR_AllocVertex(); - pv->x = FIXED_TO_FLOAT(lseg->v1->x); - pv->y = FIXED_TO_FLOAT(lseg->v1->y); - lseg->pv1 = pv; - } - - if (nearv2 <= NEARDIST*NEARDIST) - lseg->pv2 = &(p->pts[v2found]); - else - { - polyvertex_t *pv = HWR_AllocVertex(); - pv->x = FIXED_TO_FLOAT(lseg->v2->x); - pv->y = FIXED_TO_FLOAT(lseg->v2->y); - lseg->pv2 = pv; - } - - lseg->pv1->x2 = FloatToFixed(lseg->pv1->x); - lseg->pv1->y2 = FloatToFixed(lseg->pv1->y); - lseg->pv2->x2 = FloatToFixed(lseg->pv2->x); - lseg->pv2->y2 = FloatToFixed(lseg->pv2->y); - } - } -} - // call this routine after the BSP of a Doom wad file is loaded, // and it will generate all the convex polys for the hardware renderer @@ -1017,7 +915,6 @@ void HWR_CreatePlanePolygons(INT32 bspnum) #else SolveTProblem(); #endif - AdjustSegs(); #ifdef DEBUG_HWBSP //debug debug.. diff --git a/src/hardware/hw_main.cpp b/src/hardware/hw_main.cpp index c88b075a1..dc8cb55c0 100644 --- a/src/hardware/hw_main.cpp +++ b/src/hardware/hw_main.cpp @@ -741,27 +741,11 @@ static void HWR_RenderPlane(subsector_t *subsector, extrasubsector_t *xsub, bool P_ClosestPointOnLine(viewx, viewy, ld, &v); dist = FixedToFloat(R_PointToDist(v.x, v.y)); - if (line->pv1) - { - x1 = line->pv1->x; - y1 = line->pv1->y; - } - else - { - x1 = FixedToFloat(line->v1->x); - y1 = FixedToFloat(line->v1->y); - } + x1 = (line->fv1.x); + y1 = (line->fv1.y); - if (line->pv2) - { - xd = line->pv2->x - x1; - yd = line->pv2->y - y1; - } - else - { - xd = FixedToFloat(line->v2->x) - x1; - yd = FixedToFloat(line->v2->y) - y1; - } + xd = (line->fv2.x) - x1; + yd = (line->fv2.y) - y1; // Based on the seg length and the distance from the line, split horizon into multiple poly sets to reduce distortion dist = sqrtf((xd*xd) + (yd*yd)) / dist / 16.0f; @@ -825,11 +809,11 @@ static void HWR_DrawSegsSplats(FSurfaceInfo * pSurf) M_ClearBox(segbbox); M_AddToBox(segbbox, - FloatToFixed(gl_curline->pv1->x), - FloatToFixed(gl_curline->pv1->y)); + FloatToFixed(gl_curline->fv1.x), + FloatToFixed(gl_curline->fv1.y)); M_AddToBox(segbbox, - FloatToFixed(gl_curline->pv2->x), - FloatToFixed(gl_curline->pv2->y)); + FloatToFixed(gl_curline->fv2.x), + FloatToFixed(gl_curline->fv2.y)); splat = (wallsplat_t *)gl_curline->linedef->splats; for (; splat; splat = splat->next) @@ -1324,35 +1308,15 @@ void HWR_ProcessSeg(void) // Sort of like GLWall::Process in GZDoom const boolean noencore = (gl_linedef->flags & ML_TFERLINE); - if (LIKELY(gl_curline->pv1)) - { - vs.x = gl_curline->pv1->x; - vs.y = gl_curline->pv1->y; - v1x = gl_curline->pv1->x2; - v1y = gl_curline->pv1->y2; - } - else - { - vs.x = FixedToFloat(gl_curline->v1->x); - vs.y = FixedToFloat(gl_curline->v1->y); - v1x = gl_curline->v1->x; - v1y = gl_curline->v1->y; - } + vs.x = (gl_curline->fv1.x); + vs.y = (gl_curline->fv1.y); + v1x = gl_curline->v1->x; + v1y = gl_curline->v1->y; - if (LIKELY(gl_curline->pv2)) - { - ve.x = gl_curline->pv2->x; - ve.y = gl_curline->pv2->y; - v2x = gl_curline->pv2->x2; - v2y = gl_curline->pv2->y2; - } - else - { - ve.x = FixedToFloat(gl_curline->v2->x); - ve.y = FixedToFloat(gl_curline->v2->y); - v2x = gl_curline->v2->x; - v2y = gl_curline->v2->y; - } + ve.x = (gl_curline->fv2.x); + ve.y = (gl_curline->fv2.y); + v2x = gl_curline->v2->x; + v2y = gl_curline->v2->y; #define SLOPEPARAMS(slope, end1, end2, normalheight) \ end1 = P_GetZAt(slope, v1x, v1y, normalheight); \ @@ -2203,27 +2167,11 @@ static boolean CheckClip(sector_t * afrontsector, sector_t * abacksector) { fixed_t v1x, v1y, v2x, v2y; // the seg's vertexes as fixed_t - if (LIKELY(gl_curline->pv1)) - { - v1x = gl_curline->pv1->x2; - v1y = gl_curline->pv1->y2; - } - else - { - v1x = gl_curline->v1->x; - v1y = gl_curline->v1->y; - } + v1x = gl_curline->v1->x; + v1y = gl_curline->v1->y; - if (LIKELY(gl_curline->pv2)) - { - v2x = gl_curline->pv2->x2; - v2y = gl_curline->pv2->y2; - } - else - { - v2x = gl_curline->v2->x; - v2y = gl_curline->v2->y; - } + v2x = gl_curline->v2->x; + v2y = gl_curline->v2->y; #define SLOPEPARAMS(slope, end1, end2, normalheight) \ end1 = P_GetZAt(slope, v1x, v1y, normalheight); \ @@ -2428,27 +2376,11 @@ static void HWR_AddLine(seg_t *line) gl_curline = line; - if (LIKELY(gl_curline->pv1)) - { - v1x = gl_curline->pv1->x2; - v1y = gl_curline->pv1->y2; - } - else - { - v1x = gl_curline->v1->x; - v1y = gl_curline->v1->y; - } + v1x = gl_curline->v1->x; + v1y = gl_curline->v1->y; - if (LIKELY(gl_curline->pv2)) - { - v2x = gl_curline->pv2->x2; - v2y = gl_curline->pv2->y2; - } - else - { - v2x = gl_curline->v2->x; - v2y = gl_curline->v2->y; - } + v2x = gl_curline->v2->x; + v2y = gl_curline->v2->y; // OPTIMIZE: quickly reject orthogonal back sides. angle1 = R_PointToAngle64(v1x, v1y); diff --git a/src/p_setup.c b/src/p_setup.c index 2ccbd1821..afdda1ded 100644 --- a/src/p_setup.c +++ b/src/p_setup.c @@ -544,7 +544,18 @@ static void P_LoadSegs(UINT8 *data) { li->v1 = &vertexes[SHORT(ml->v1)]; li->v2 = &vertexes[SHORT(ml->v2)]; +#ifdef HWRENDER + if (rendermode == render_opengl) + { + li->fv1.x = FixedToFloat(li->v1->x); + li->fv1.y = FixedToFloat(li->v1->y); + //li->fv1.z = FixedToFloat(li->v1->z); // z is unused in gl + li->fv2.x = FixedToFloat(li->v2->x); + li->fv2.y = FixedToFloat(li->v2->y); + //li->fv2.z = FixedToFloat(li->v2->z); + } +#endif li->length = P_SegLength(li); li->angle = (SHORT(ml->angle))<offset = (SHORT(ml->offset))< Date: Thu, 2 Oct 2025 18:37:05 +0200 Subject: [PATCH 2/2] a thousand curses upon you polyobjects! --- src/hardware/hw_main.cpp | 23 ++++++++++++++++++----- 1 file changed, 18 insertions(+), 5 deletions(-) diff --git a/src/hardware/hw_main.cpp b/src/hardware/hw_main.cpp index dc8cb55c0..1c7234d2c 100644 --- a/src/hardware/hw_main.cpp +++ b/src/hardware/hw_main.cpp @@ -729,6 +729,9 @@ static void HWR_RenderPlane(subsector_t *subsector, extrasubsector_t *xsub, bool for (i = 0; i < subsector->numlines; i++, line++) { + if (line->polyseg) + continue; + line_t* ld = line->linedef; // this check sucks and is a hotspot lel @@ -1308,16 +1311,26 @@ void HWR_ProcessSeg(void) // Sort of like GLWall::Process in GZDoom const boolean noencore = (gl_linedef->flags & ML_TFERLINE); - vs.x = (gl_curline->fv1.x); - vs.y = (gl_curline->fv1.y); v1x = gl_curline->v1->x; v1y = gl_curline->v1->y; - - ve.x = (gl_curline->fv2.x); - ve.y = (gl_curline->fv2.y); v2x = gl_curline->v2->x; v2y = gl_curline->v2->y; + if (gl_curline->polyseg) + { + vs.x = FixedToFloat(v1x); + vs.y = FixedToFloat(v1y); + ve.x = FixedToFloat(v2x); + ve.y = FixedToFloat(v2y); + } + else + { + vs.x = gl_curline->fv1.x; + vs.y = gl_curline->fv1.y; + ve.x = gl_curline->fv2.x; + ve.y = gl_curline->fv2.y; + } + #define SLOPEPARAMS(slope, end1, end2, normalheight) \ end1 = P_GetZAt(slope, v1x, v1y, normalheight); \ end2 = P_GetZAt(slope, v2x, v2y, normalheight);