shpuld's single-pass clipping optimization

source/host.c

@@ -610,75 +610,6 @@ void Host_ServerFrame (void)
 	SV_SendClientMessages ();
 }
 
-int extshit()
-{
-	u32 size, sizeblock;
-	u8 *ram;
-	// Init variables
-	size = 0;
-	sizeblock = (1024 * 1024);
-	// Check loop
-	while (sizeblock)
-	{
-		// Increment size
-		size += sizeblock;
-		// Allocate ram
-		ram = (u8*)malloc(size);
-		// Check allocate
-		if (!(ram))
-		{
-			// Restore old size
-			size -= sizeblock;
-			// Size block / 2
-			sizeblock >>= 1;
-		}
-		else
-			free(ram);
-	}
-	
-	return size;
-}
-
-int freememory()
-{
-	u8 **ram, **temp;
-	u32 size, count, x;
-	// Init variables
-	ram = NULL;
-	size = 0;
-	count = 0;
-	// Check loop
-	for (;;)
-	{
-		// Check size entries
-		if (!(count % 10))
-		{
-			// Allocate more entries if needed
-			temp = (u8**)realloc(ram,sizeof(u8*) * (count + 10));
-			if (!(temp)) break;
-			// Update entries and size (size contains also size of entries)
-			ram = temp;
-			size += (sizeof(u8 *) * 10);
-		}
-		// Find max lineare size available
-		x = extshit();
-		if (!(x)) break;
-		// Allocate ram
-		ram[count] = (u8*)malloc(x);
-		if (!(ram[count])) break;
-		// Update variables
-		size += x;
-		count++;
-	}
-	// Free ram
-	if (ram)
-	{
-		for (x=0;x<count;x++) free(ram[x]);
-		free(ram);
-	}
-	return (int)size;
-}
-
 /*
 ==================
 Host_Frame

source/psp/video_hardware_model.h

@@ -153,15 +153,11 @@ typedef struct glpoly_s
     struct glpoly_s	*caustics_chain;	// next caustic poly in chain
     vec3_t midpoint;//MHQuake
 	float fxofs;	//MHQuake
-	// This is a variable sized array, and hence must be the last element in
+
-	// this structure.
+	// shpuld: for display list allocated temporary memory for psp rendering
-	//
+	int			numclippedverts;
-	// The array is (numverts * 2) in size. The first half are regular
+	glvert_t	* display_list_verts;
-	// vertices, and the second half have copies of the first half's XYZs but
+	// Variable size array, has to be the last element of the struct. actual size is numverts
-	// keep the light map texture coordinates. This makes the vertices easier
-	// to render on the PSP.
-	// naievil -- changed from verts[1] to verts[2] in order to not have possible out of bounds error during compilation
-	//            and create some sort of condition that causes the verts to not be assigned!
 	glvert_t	verts[2];
 } glpoly_t;
 
@@ -197,6 +193,11 @@ typedef struct msurface_s
 	byte		*samples;		// [numstyles*surfsize]
 } msurface_t;
 
+typedef struct lightmap_face_s {
+	msurface_t *face;
+	struct lightmap_face_s *next;
+} lightmap_face_t;
+
 typedef struct mnode_s
 {
 // common with leaf

source/psp/video_hardware_surface.cpp

@@ -60,10 +60,20 @@ typedef struct glRect_s
 //////////////////////////////////////////////////////////////////////////////
 //For none .lit maps./////////////////////////////////////////////////////////
 //////////////////////////////////////////////////////////////////////////////
-glpoly_t	*lightmap_polys[MAX_LIGHTMAPS];
 qboolean	lightmap_modified[MAX_LIGHTMAPS];
 glRect_t	lightmap_rectchange[MAX_LIGHTMAPS];
 
+#define MAX_VISIBLE_LIGHTMAPPED_FACES 1024
+// contains references to every msurface_t that's getting rendered.
+// in testing 512 is not enough for bigger complicated areas (like parts of dm2).
+// consider doubling to 2048 or somewhere inbetween.
+// when the limit gets exceeded, there will be some surfaces without lightmaps.
+lightmap_face_t lightmap_faces[MAX_VISIBLE_LIGHTMAPPED_FACES];
+// chains of lightmap faces for each lightmap num
+lightmap_face_t	*lightmap_chains[MAX_LIGHTMAPS];
+
+int num_lightmapped_faces;
+
 int			allocated[MAX_LIGHTMAPS][BLOCK_WIDTH];
 
 // the lightmap texture data needs to be kept in
@@ -77,8 +87,6 @@ msurface_t  *waterchain = NULL;
 glpoly_t	*caustics_polys = NULL;
 glpoly_t	*detail_polys   = NULL;
 
-void R_RenderDynamicLightmaps (msurface_t *fa);
-
 void 	VID_SetPaletteLM();
 // switch palette for lightmaps
 
@@ -404,95 +412,41 @@ extern	int		solidskytexture;
 extern	int		alphaskytexture;
 extern	float	speedscale;		// for top sky and bottom sky
 
-static inline void DrawGLPolyLM (glpoly_t *p)
+static inline void DrawGLPolyLM (glpoly_t * poly)
 {
-	// Does this poly need clipped?
+	if (r_showtris.value)
-
-
-	const int				unclipped_vertex_count	= p->numverts;
-	const glvert_t* const	unclipped_vertices		= &(p->verts[p->numverts]);
-
-	if (clipping::is_clipping_required(
-		unclipped_vertices,
-		unclipped_vertex_count))
 	{
-		// Clip the polygon.
+		sceGuDisable(GU_TEXTURE_2D);
-		const glvert_t*	clipped_vertices;
+		sceGuDisable(GU_BLEND);
-		std::size_t		clipped_vertex_count;
-		clipping::clip(
-			unclipped_vertices,
-			unclipped_vertex_count,
-			&clipped_vertices,
-			&clipped_vertex_count);
 
-		// Did we have any vertices left?
+		// Draw the lines directly.
-		if (clipped_vertex_count)
+		sceGumDrawArray(
-		{
+			GU_LINE_STRIP,
-			// Copy the vertices to the display list.
+			GU_TEXTURE_32BITF | GU_VERTEX_32BITF ,
-			const std::size_t buffer_size = clipped_vertex_count * sizeof(glvert_t);
+			poly->numclippedverts, 0, poly->display_list_verts);
-			glvert_t* const display_list_vertices = static_cast<glvert_t*>(sceGuGetMemory(buffer_size));
-			memcpy(display_list_vertices, clipped_vertices, buffer_size);
 
-            // Draw the clipped vertices.
+		sceGuEnable(GU_TEXTURE_2D);
-            sceGuDrawArray(
+		sceGuEnable(GU_BLEND);
-                GU_TRIANGLE_FAN,
-                GU_TEXTURE_32BITF | GU_VERTEX_32BITF ,
-                clipped_vertex_count, 0, display_list_vertices);
-		}
-	}
-	else
-	{
-            // Draw the poly directly.
-			sceGuDrawArray(
-                GU_TRIANGLE_FAN,
-                GU_TEXTURE_32BITF | GU_VERTEX_32BITF ,
-                unclipped_vertex_count, 0, unclipped_vertices);
-    }
-}
 
-static inline void DrawGLPoly (glpoly_t *p)
-{
-	// Does this poly need clipped?
-	const int				unclipped_vertex_count	= p->numverts;
-	const glvert_t* const	unclipped_vertices		= p->verts;
-	if (clipping::is_clipping_required(
-		unclipped_vertices,
-		unclipped_vertex_count))
-	{
-		// Clip the polygon.
-		const glvert_t*	clipped_vertices;
-		std::size_t		clipped_vertex_count;
-		clipping::clip(
-			unclipped_vertices,
-			unclipped_vertex_count,
-			&clipped_vertices,
-			&clipped_vertex_count);
-
-		// Did we have any vertices left?
-		if (clipped_vertex_count)
-		{
-			// Copy the vertices to the display list.
-			const std::size_t buffer_size = clipped_vertex_count * sizeof(glvert_t);
-			glvert_t* const display_list_vertices = static_cast<glvert_t*>(sceGuGetMemory(buffer_size));
-			memcpy(display_list_vertices, clipped_vertices, buffer_size);
-
-			// Draw the clipped vertices.
-			sceGuDrawArray(
-				GU_TRIANGLE_FAN,
-				GU_TEXTURE_32BITF | GU_VERTEX_32BITF,
-				clipped_vertex_count, 0, display_list_vertices);
-		}
 	}
 	else
 	{
 		// Draw the poly directly.
 		sceGuDrawArray(
 			GU_TRIANGLE_FAN,
-			GU_TEXTURE_32BITF | GU_VERTEX_32BITF,
+			GU_TEXTURE_32BITF | GU_VERTEX_32BITF ,
-			unclipped_vertex_count, 0, unclipped_vertices);
+			poly->numclippedverts, 0, poly->display_list_verts);
 	}
 }
 
+static inline void DrawGLPoly (glpoly_t * poly)
+{
+	sceGuDrawArray(
+			GU_TRIANGLE_FAN,
+			GU_TEXTURE_32BITF | GU_VERTEX_32BITF,
+			poly->numclippedverts, 0, poly->display_list_verts);
+}
+
 static inline void DrawTrisPoly (glpoly_t *p) //Crow_bar
 {
     sceGuDisable(GU_TEXTURE_2D);
@@ -815,7 +769,6 @@ R_BlendLightmaps
 static void R_BlendLightmaps (void)
 {
 	int			i;
-	glpoly_t	*p;
 
 	if (r_fullbright.value)
 		return;
@@ -830,10 +783,11 @@ static void R_BlendLightmaps (void)
 	if (r_lightmap.value)
 		sceGuDisable(GU_BLEND);
 
+	lightmap_face_t * lmface;
 	for (i=0 ; i<MAX_LIGHTMAPS ; i++)
 	{
-		p = lightmap_polys[i];
+		lmface = lightmap_chains[i];
-		if (!p)
+		if (!lmface)
 			continue;
 
 		char lm_name[16];
@@ -845,32 +799,83 @@ static void R_BlendLightmaps (void)
 			lightmap_rectchange[i].w = 0;
 			lightmap_rectchange[i].h = 0;
 
-			sprintf(lm_name,"lightmap%d",i);
+			snprintf(lm_name, sizeof(lm_name), "lightmap%d",i);
 			lightmap_index[i] = GL_LoadTextureLM (lm_name, BLOCK_WIDTH, BLOCK_HEIGHT, lightmaps+(i*BLOCK_WIDTH*BLOCK_HEIGHT*LIGHTMAP_BYTES), LIGHTMAP_BYTES, GU_LINEAR, qtrue, qfalse);
 		}
 		GL_BindLM (lightmap_index[i]);
-		for ( ; p ; p=p->chain)
+		for (; lmface; lmface = lmface->next) {
-		{
+			msurface_t * face = lmface->face;
-			if (p->flags & SURF_UNDERWATER)
+			float scale = 0.0625f;
-			{
+			float tscale = face->texinfo->texture->width / (BLOCK_WIDTH * 16.f);
-				DrawGLPolyLM(p);
+			float sscale = face->texinfo->texture->height / (BLOCK_HEIGHT * 16.f);
-				//DrawGLWaterPolyLM(p);
-			}
-			else
-				DrawGLPolyLM(p);
 			
+			sceGuTexScale(tscale, sscale);
+			sceGuTexOffset(
+				tscale * (float)(-1.f * face->texturemins[0] + face->light_s * 16 + 8) / (float)(face->texinfo->texture->width),
+				sscale * (float)(-1.f * face->texturemins[1] + face->light_t * 16 + 8) / (float)(face->texinfo->texture->height)
+			);
+
+			DrawGLPolyLM(face->polys);
 		}
 	}
 
 	if(LIGHTMAP_BYTES == 1)
 	   VID_SetPaletteTX();
 
+	sceGuTexOffset(0, 0);
+	sceGuTexScale(1, 1);
 	sceGuDisable(GU_BLEND);
 	sceGuBlendFunc(GU_ADD, GU_SRC_ALPHA, GU_ONE_MINUS_SRC_ALPHA, 0, 0);
 	sceGuDepthMask (GU_FALSE);
 	sceGuEnable(GU_DEPTH_TEST); // dr_mabuse1981: fix
 }
 
+int ClipFace (msurface_t * fa)
+{
+	// shpuld: moved clipping here to have it in one place only
+	int verts_total = 0;
+	glpoly_t* poly = fa->polys;
+	const int unclipped_vertex_count = poly->numverts;
+	const glvert_t* const unclipped_vertices = poly->verts;
+
+	if (clipping::is_clipping_required(
+		unclipped_vertices,
+		unclipped_vertex_count))
+	{
+		// Clip the polygon.
+		const glvert_t*	clipped_vertices;
+		std::size_t		clipped_vertex_count;
+		clipping::clip(
+			unclipped_vertices,
+			unclipped_vertex_count,
+			&clipped_vertices,
+			&clipped_vertex_count
+		);
+
+		verts_total += clipped_vertex_count;
+
+		// Did we have any vertices left?
+		if (!clipped_vertex_count)
+		{
+			poly->numclippedverts = 0;
+			return verts_total;
+		}
+
+		const std::size_t buffer_size = clipped_vertex_count * sizeof(glvert_t);
+		poly->display_list_verts = static_cast<glvert_t*>(sceGuGetMemory(buffer_size));
+		memcpy(poly->display_list_verts, clipped_vertices, buffer_size);
+		poly->numclippedverts = clipped_vertex_count;
+	} else {
+		verts_total += unclipped_vertex_count;
+
+		const std::size_t buffer_size = unclipped_vertex_count * sizeof(glvert_t);
+		poly->display_list_verts = static_cast<glvert_t*>(sceGuGetMemory(buffer_size));
+		memcpy(poly->display_list_verts, unclipped_vertices, buffer_size);
+		poly->numclippedverts = unclipped_vertex_count;
+	}
+	return verts_total;
+}
+
 /*
 ================
 R_RenderBrushPoly
@@ -887,17 +892,12 @@ void R_RenderBrushPoly (msurface_t *fa)
 
 	c_brush_polys++;
 
-	/*if(r_showtris.value) //Crow_bar
+	// sky and water use multiple polys per surface,
-	{
+	// this makes clipping more tricky, but they don't have LMs so no prob.
-		sceGuDepthMask (GU_TRUE);
-
-		DrawTrisPoly (fa->polys);
-
-		sceGuDepthMask (GU_FALSE);
-	}*/
-
 	if (fa->flags & SURF_DRAWSKY)
-	{	// warp texture, no lightmaps
+	{	
+		// warp texture, no lightmaps
+		// shpuld: replace with scissor pass and then sky, maybe faster?
 		EmitBothSkyLayers (fa);
 		return;
 	}
@@ -911,6 +911,12 @@ void R_RenderBrushPoly (msurface_t *fa)
 		return;
 	}
 
+	// Everything from here only uses 1 poly from fa->polys
+	int verts_count = ClipFace(fa);
+
+	if (verts_count <= 0)
+		return;
+
 	sceGuEnable(GU_ALPHA_TEST);
 	sceGuAlphaFunc(GU_GREATER, 0xaa, 0xff);
 	sceGuTexFunc(GU_TFX_MODULATE, GU_TCC_RGBA);
@@ -931,10 +937,14 @@ void R_RenderBrushPoly (msurface_t *fa)
 	}
 	// motolegacy -- end texflags
 
-	// add the poly to the proper lightmap chain
+	// Manage lightmap chain
-	fa->polys->chain = lightmap_polys[fa->lightmaptexturenum];
+	if (num_lightmapped_faces < MAX_VISIBLE_LIGHTMAPPED_FACES)
-	lightmap_polys[fa->lightmaptexturenum] = fa->polys;
+	{
-
+		lightmap_faces[num_lightmapped_faces].face = fa;
+		lightmap_faces[num_lightmapped_faces].next = lightmap_chains[fa->lightmaptexturenum];
+		lightmap_chains[fa->lightmaptexturenum] = &lightmap_faces[num_lightmapped_faces];
+		num_lightmapped_faces++;
+	}
 
 	// check for lightmap modification
 	for (maps = 0 ; maps < MAXLIGHTMAPS && fa->styles[maps] != 255 ; maps++)
@@ -976,64 +986,6 @@ dynamic:
 	sceGuDisable(GU_ALPHA_TEST);
 }
 
-/*
-================
-R_RenderDynamicLightmaps
-Multitexture
-================
-*/
-void R_RenderDynamicLightmaps (msurface_t *fa)
-{
-	byte		*base;
-	int			maps;
-	glRect_t    *theRect;
-	int smax, tmax;
-
-	c_brush_polys++;
-
-	if (fa->flags & ( SURF_DRAWSKY | SURF_DRAWTURB) )
-		return;
-
-	fa->polys->chain = lightmap_polys[fa->lightmaptexturenum];
-	lightmap_polys[fa->lightmaptexturenum] = fa->polys;
-
-	// check for lightmap modification
-	for (maps = 0 ; maps < MAXLIGHTMAPS && fa->styles[maps] != 255 ;
-		 maps++)
-		if (d_lightstylevalue[fa->styles[maps]] != fa->cached_light[maps])
-			goto dynamic;
-
-	if (fa->dlightframe == r_framecount	// dynamic this frame
-		|| fa->cached_dlight)			// dynamic previously
-	{
-dynamic:
-		if (r_dynamic.value)
-		{
-			lightmap_modified[fa->lightmaptexturenum] = qtrue;
-			theRect = &lightmap_rectchange[fa->lightmaptexturenum];
-			if (fa->light_t < theRect->t) {
-				if (theRect->h)
-					theRect->h += theRect->t - fa->light_t;
-				theRect->t = fa->light_t;
-			}
-			if (fa->light_s < theRect->l) {
-				if (theRect->w)
-					theRect->w += theRect->l - fa->light_s;
-				theRect->l = fa->light_s;
-			}
-			smax = (fa->extents[0]>>4)+1;
-			tmax = (fa->extents[1]>>4)+1;
-			if ((theRect->w + theRect->l) < (fa->light_s + smax))
-				theRect->w = (fa->light_s-theRect->l)+smax;
-			if ((theRect->h + theRect->t) < (fa->light_t + tmax))
-				theRect->h = (fa->light_t-theRect->t)+tmax;
-			base = lightmaps + fa->lightmaptexturenum*LIGHTMAP_BYTES*BLOCK_WIDTH*BLOCK_HEIGHT;
-			base += fa->light_t * BLOCK_WIDTH * LIGHTMAP_BYTES + fa->light_s * LIGHTMAP_BYTES;
-			R_BuildLightMap (fa, base, BLOCK_WIDTH*LIGHTMAP_BYTES);
-		}
-	}
-}
-
 /*
 ================
 R_MirrorChain
@@ -1227,7 +1179,8 @@ void R_DrawBrushModel (entity_t *e)
 	}
 
 
-	memset (lightmap_polys, 0, sizeof(lightmap_polys));
+	memset (lightmap_chains, 0, sizeof(lightmap_chains));
+	num_lightmapped_faces = 0;
 
 	VectorSubtract (r_refdef.vieworg, e->origin, modelorg);
 	if (rotated)
@@ -1526,7 +1479,8 @@ void R_DrawWorld (void)
 	currenttexture = -1;
 
 	/*glColor3f (1,1,1);*/
-	memset (lightmap_polys, 0, sizeof(lightmap_polys));
+	memset (lightmap_chains, 0, sizeof(lightmap_chains));
+	num_lightmapped_faces = 0;
 
 	R_ClearSkyBox ();
 
@@ -1673,7 +1627,7 @@ static void BuildSurfaceDisplayList (msurface_t *fa)
 	//
 	// draw texture
 	//
-    poly = static_cast<glpoly_t*>(Hunk_Alloc (sizeof(glpoly_t) + (lnumverts * 2 - 1) * sizeof(glvert_t)));
+    poly = static_cast<glpoly_t*>(Hunk_Alloc (sizeof(glpoly_t) + (lnumverts - 1) * sizeof(glvert_t)));
 	poly->next = fa->polys;
 	poly->flags = fa->flags;
 	fa->polys = poly;
@@ -1702,25 +1656,6 @@ static void BuildSurfaceDisplayList (msurface_t *fa)
 		VectorCopy(vec, poly->verts[i].xyz);
 		poly->verts[i].st[0] = s;
 		poly->verts[i].st[1] = t;
-
-		//
-		// lightmap texture coordinates
-		//
-		s = DotProduct (vec, fa->texinfo->vecs[0]) + fa->texinfo->vecs[0][3];
-		s -= fa->texturemins[0];
-		s += fa->light_s*16;
-		s += 8;
-		s /= BLOCK_WIDTH*16; //fa->texinfo->texture->width;
-
-		t = DotProduct (vec, fa->texinfo->vecs[1]) + fa->texinfo->vecs[1][3];
-		t -= fa->texturemins[1];
-		t += fa->light_t*16;
-		t += 8;
-		t /= BLOCK_HEIGHT*16; //fa->texinfo->texture->height;
-
-		VectorCopy(vec, poly->verts[i + lnumverts].xyz);
-		poly->verts[i + lnumverts].st[0] = s;
-		poly->verts[i + lnumverts].st[1] = t;
 	}
 
 	//
@@ -1760,7 +1695,6 @@ static void BuildSurfaceDisplayList (msurface_t *fa)
 				for (j = i + 1; j < lnumverts; j = j + 1)
 				{
 					poly->verts[j - 1] = poly->verts[j];
-					poly->verts[poly->numverts + j - 1] = poly->verts[poly->numverts+j];
 				}
 
 				--lnumverts;
@@ -1770,12 +1704,6 @@ static void BuildSurfaceDisplayList (msurface_t *fa)
 				--i;
 			}
 		}
-
-		if (numRemoved > 0) {
-			for (j = poly->numverts; j < poly->numverts + lnumverts; j++) {
-				poly->verts[j - numRemoved] = poly->verts[j];
-			}
-		}
 	}
 
 	// Colinear point removal-end
@@ -1846,6 +1774,7 @@ void GL_BuildLightmaps (void)
 		currentmodel = m;
 		for (i=0 ; i<m->numsurfaces ; i++)
 		{
+			// todo: investigate why this is done even for turb/sky
 			GL_CreateSurfaceLightmap (m->surfaces + i);
 			if ( m->surfaces[i].flags & SURF_DRAWTURB )
 				continue;