GL3: Make gl3drawCmd_t smaller. Step 2: Move transModelMat elsewhere

Most of the time the transformation matrix is the identity matrix
anyway, so let's not waste 64 bytes in every gl3drawCmd_t...

Now it's just an index into a hmm_mat4 array.
Index 0 is always the identity matrix, if a different one is required,
GL3_SetDrawCmdTransMatrix(&drawCmd, mat) can be used to set it - it will
be added to that array and its index is set in drawCmd.

Bonus: This doesn't only reduce the gl3drawCmd_t to 36 bytes but also
makes comparing two of them a lot cheaper: Instead of comparing the
values of the matrices, we can just compare their index.
Also frees up one flag (`flag & DCFlag_IsIdentityMat` is replaced by
`transModelMatIdx == 0`).

src/client/refresh/gl3/gl3_main.c

@@ -136,11 +136,12 @@ static cvar_t *gl_znear;
 DA_TYPEDEF(gl3_3D_vtx_t, Vtx3DArray_t);
 DA_TYPEDEF(GLushort, UShortArray_t);
 DA_TYPEDEF(gl3drawCmd_t, DrawCommandArray_t);
+DA_TYPEDEF(hmm_mat4, Mat4Array_t);
 // dynamic arrays to batch all consecutive 3D draws (with gl3_3D_vtx_t) with same texture to reduce drawcalls
 static Vtx3DArray_t vtxBuf = {0};
 static UShortArray_t idxBuf = {0};
 static DrawCommandArray_t drawCmds = {0};
-
+static Mat4Array_t transModelMats = {0};
 
 // Yaw-Pitch-Roll
 // equivalent to R_z * R_y * R_x where R_x is the trans matrix for rotating around X axis for aroundXdeg
@@ -190,7 +191,7 @@ GL3_RotateUni3DforEntity(entity_t *e)
 // if replaceTransModelMat is true, the existing value of drawCmd->transModelMat
 // is ignored and it's replaced with the one calculated here (otherwise they're multiplied)
 void
-GL3_RotateForEntity(entity_t *e, gl3drawCmd_t* drawCmd, qboolean replaceTransModelMat)
+GL3_RotateForEntity(entity_t *e, gl3drawCmd_t* drawCmd)
 {
 	// TODO: shortcut for "not rotated at all"?
 
@@ -203,11 +204,7 @@ GL3_RotateForEntity(entity_t *e, gl3drawCmd_t* drawCmd, qboolean replaceTransMod
 		transMat.Elements[3][i] = e->origin[i]; // set translation
 	}
 
-	if (replaceTransModelMat)
-		drawCmd->transModelMat = transMat;
-	else
-		drawCmd->transModelMat = HMM_MultiplyMat4(drawCmd->transModelMat, transMat);
-	drawCmd->flags &= ~DCFlag_IsIdentityMat;
+	GL3_SetDrawCmdTransMatrix(drawCmd, transMat);
 }
 
 
@@ -551,6 +548,9 @@ GL3_Init(void)
 	GetPCXPalette (&colormap, d_8to24table);
 	free(colormap);
 
+	da_clear(transModelMats);
+	da_add(transModelMats, gl3_identityMat4); // element 0 is always the identity matrix
+
 	GL3_Register();
 
 	/* set our "safe" mode */
@@ -694,6 +694,7 @@ GL3_Shutdown(void)
 	da_free(vtxBuf);
 	da_free(idxBuf);
 	da_free(drawCmds);
+	da_free(transModelMats);
 
 	/* shutdown OS specific OpenGL stuff like contexts, etc.  */
 	GL3_ShutdownContext();
@@ -765,21 +766,21 @@ void GL3_Draw3DBatchesNow()
 
 	// set curState to something that reflects the actual state, as far as possible,
 	// and otherwise makes sure it'll be set in the loop
-	gl3drawCmd_t curState = GL3_CreateDrawCmd(true);
-	curState.flags = ~drawCmds.p[0].flags; // just the opposite flags of the first element
+	gl3drawCmd_t curState = GL3_CreateDrawCmd();
+
 	curState.texnum = gl3state.currenttexture;
 	curState.lmtexnum = gl3state.currentlightmap;
 	curState.alpha = gl3state.uni3DData.alpha;
 	curState.scroll = gl3state.uni3DData.scroll;
 	curState.lightScaleForTurb = gl3state.uni3DData.lightScaleForTurb;
 
-	gl3ShaderInfo_t* shader = NULL;
+	// for the next two the approach is setting a value that
+	// is different from the one in the first drawCmd, to make sure
+	// the corresponding state is set in the first iteration
+	curState.flags = ~drawCmds.p[0].flags; // just the opposite flags of the first element
+	curState.transModelMatIdx = drawCmds.p[0].transModelMatIdx + 1;
 
-	// .. but at least set the identity mat so DCFlag_IsIdentityMat can be set
-	// as a sane default
-	curState.flags |= DCFlag_IsIdentityMat;
-	gl3state.uni3DData.transModelMat4 = gl3_identityMat4;
-	GL3_UpdateUBO3D();
+	gl3ShaderInfo_t* shader = NULL;
 
 	for(int i=0, n=da_count(drawCmds); i < n; ++i)
 	{
@@ -862,20 +863,11 @@ void GL3_Draw3DBatchesNow()
 			updateUni3D = true;
 		}
 
-		if((flags & DCFlag_IsIdentityMat) != (curFlags & DCFlag_IsIdentityMat))
+		if(cmd->transModelMatIdx != curState.transModelMatIdx)
 		{
-			// one is the identity matrix, the other isn't => we need to update
-			gl3state.uni3DData.transModelMat4 = cmd->transModelMat;
+			gl3state.uni3DData.transModelMat4 = da_get(transModelMats, cmd->transModelMatIdx);
 			updateUni3D = true;
 		}
-		else if( !(flags & DCFlag_IsIdentityMat) )
-		{
-			// neither is an identity matrix
-			// TODO: could check if they're identical and only update if not
-			gl3state.uni3DData.transModelMat4 = cmd->transModelMat;
-			updateUni3D = true;
-		}
-		// else both are the identity matrix - nothing to do
 
 		if(updateUni3D)
 			GL3_UpdateUBO3D();
@@ -891,13 +883,15 @@ void GL3_Draw3DBatchesNow()
 	da_clear(idxBuf);
 	da_clear(drawCmds);
 
+	da_setcount(transModelMats, 1); // keep index 0 (identity matrix)
+
 	// restore sane default for other draw operations (models, particles, 2D)
-	int curFlags = curState.flags;
-	if( !(curFlags & DCFlag_IsIdentityMat) )
+	if(curState.transModelMatIdx != 0)
 	{
 		gl3state.uni3DData.transModelMat4 = gl3_identityMat4;
 		GL3_UpdateUBO3D();
 	}
+	int curFlags = curState.flags;
 	if(curFlags & DCFlag_Blend)
 		glDisable(GL_BLEND);
 	if(curFlags & DCFlag_DisableDepthMask)
@@ -910,7 +904,8 @@ void GL3_Draw3DBatchesNow()
 static qboolean drawStateEqual(const gl3drawCmd_t* a, const gl3drawCmd_t* b)
 {
 	if( a->flags != b->flags || a->shaderIdx != b->shaderIdx
-	   || a->texnum != b->texnum || a->lmtexnum != b->lmtexnum )
+	   || a->texnum != b->texnum || a->lmtexnum != b->lmtexnum
+	   || a->transModelMatIdx != b->transModelMatIdx )
 		return false;
 
 	int flags = a->flags; // at this point we know the flags are identical
@@ -928,24 +923,19 @@ static qboolean drawStateEqual(const gl3drawCmd_t* a, const gl3drawCmd_t* b)
 		return false;
 
 	if((flags & DCFlag_UseLmStyles) && !lmstylesEqual(a->styles, b->styles))
-	{
 		return false;
-	}
-
-	if( !(flags & DCFlag_IsIdentityMat) )
-	{
-		const float* atm = a->transModelMat.Elements[0];
-		const float* btm = b->transModelMat.Elements[0];
-		for(int i=0; i<16; ++i)
-		{
-			if(atm[i] != btm[i]) // TODO: tolerance?
-				return false;
-		}
-	}
 
 	return true;
 }
 
+void
+GL3_SetDrawCmdTransMatrix(gl3drawCmd_t* drawCmd, hmm_mat4 mat)
+{
+	int idx = da_count(transModelMats);
+	da_add(transModelMats, mat);
+	drawCmd->transModelMatIdx = idx;
+}
+
 // assumes gl3state.v[ab]o3D are bound
 // buffers and draws gl3_3D_vtx_t vertices
 // drawMode is something like GL_TRIANGLE_STRIP or GL_TRIANGLE_FAN or whatever
@@ -1033,7 +1023,7 @@ GL3_DrawBeam(entity_t *e)
 	int i;
 	enum { NUM_BEAM_SEGS = 6 };
 
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(true);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 
 	vec3_t perpvec;
 	vec3_t direction, normalized_direction;
@@ -1109,7 +1099,7 @@ GL3_DrawSpriteModel(entity_t *e, gl3model_t *currentmodel)
 	dsprite_t *psprite;
 	gl3image_t *skin;
 
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(true);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 
 	/* don't even bother culling, because it's just
 	   a single polygon without a surface cache */
@@ -1180,7 +1170,7 @@ static void
 GL3_DrawNullModel(entity_t *currententity)
 {
 	vec3_t shadelight;
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(true);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 
 	if(currententity->flags & RF_TRANSLUCENT)
 		drawCmd.flags |= DCFlag_DisableDepthMask;
@@ -1194,7 +1184,7 @@ GL3_DrawNullModel(entity_t *currententity)
 		GL3_LightPoint(currententity, currententity->origin, shadelight);
 	}
 
-	GL3_RotateForEntity(currententity, &drawCmd, true);
+	GL3_RotateForEntity(currententity, &drawCmd);
 
 	drawCmd.alpha = 1.0f;
 	for(int i=0; i<3; ++i)

src/client/refresh/gl3/gl3_surf.c

@@ -164,7 +164,6 @@ GL3_DrawGLPoly(msurface_t *fa, gl3drawCmd_t drawCmd)
 {
 	glpoly_t *p = fa->polys;
 
-
 	GL3_BufferAndDraw3D(p->vertices, p->numverts, GL_TRIANGLE_FAN, drawCmd);
 }
 
@@ -296,7 +295,7 @@ GL3_DrawAlphaSurfaces(void)
 	msurface_t *s;
 
 	/* go back to the world matrix */
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(true);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 	drawCmd.flags |= DCFlag_Blend;
 
 	for (s = gl3_alpha_surfaces; s != NULL; s = s->texturechain)
@@ -344,7 +343,7 @@ DrawTextureChains(entity_t *currententity)
 
 	c_visible_textures = 0;
 
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(true);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 
 	for (i = 0, image = gl3textures; i < numgl3textures; i++, image++)
 	{
@@ -477,7 +476,7 @@ GL3_DrawBrushModel(entity_t *e, gl3model_t *currentmodel)
 		return;
 	}
 
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(false);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 	if(e->flags & RF_TRANSLUCENT)
 		drawCmd.flags |= DCFlag_DisableDepthMask;
 
@@ -527,7 +526,7 @@ GL3_DrawBrushModel(entity_t *e, gl3model_t *currentmodel)
 
 	e->angles[0] = -e->angles[0];
 	e->angles[2] = -e->angles[2];
-	GL3_RotateForEntity(e, &drawCmd, true);
+	GL3_RotateForEntity(e, &drawCmd);
 	e->angles[0] = -e->angles[0];
 	e->angles[2] = -e->angles[2];
 

src/client/refresh/gl3/gl3_warp.c

@@ -253,8 +253,6 @@ GL3_EmitWaterPolys(msurface_t *fa, gl3drawCmd_t drawCmd)
 	drawCmd.lightScaleForTurb = fa->texinfo->image->is_lava ? 1.0f : 0.5f;
 	drawCmd.flags |= DCFlag_UseLightScaleForTurb;
 
-
-
 	GL3_SetDrawCmdShader(&drawCmd, &gl3state.si3Dturb);
 
 	for (bp = fa->polys; bp != NULL; bp = bp->next)
@@ -684,7 +682,7 @@ GL3_DrawSkyBox(void)
 		}
 	}
 
-	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd(false);
+	gl3drawCmd_t drawCmd = GL3_CreateDrawCmd();
 
 	// glPushMatrix();
 
@@ -697,7 +695,8 @@ GL3_DrawSkyBox(void)
 		hmm_vec3 rotAxis = HMM_Vec3(skyaxis[0], skyaxis[1], skyaxis[2]);
 		modMVmat = HMM_MultiplyMat4(modMVmat, HMM_Rotate(r_newrefdef.time * skyrotate, rotAxis));
 	}
-	drawCmd.transModelMat = modMVmat;
+
+	GL3_SetDrawCmdTransMatrix(&drawCmd, modMVmat);
 
 	GL3_SetDrawCmdShader(&drawCmd, &gl3state.si3Dsky);
 

src/client/refresh/gl3/header/local.h

@@ -278,12 +278,13 @@ typedef struct
 
 // drawcommands using gl3_3D_vtx_t, for batching
 typedef struct gl3drawCmd_s {
-	hmm_mat4	transModelMat;
-
 	GLuint		texnum;
 	signed char	lmtexnum;
 	signed char	shaderIdx;
 
+	// index into gl3_main.c transModelMats; 0 always is identity matrix
+	unsigned short transModelMatIdx;
+
 	float		scroll; // for gl3state.uni3DData.scroll
 	float		lightScaleForTurb; // for gl3state.uni3DData.lightScaleForTurb
 	float		alpha; // either part of color or for gl3state.uni3DData.alpha
@@ -305,20 +306,15 @@ enum gl3drawCmd_Flags {
 	DCFlag_UseScroll        = 16,
 	DCFlag_UseLmStyles      = 32,
 	DCFlag_UseLightScaleForTurb = 64,
-	DCFlag_IsIdentityMat    = 128, // avoids comparing the matrix in many cases
 
 	// TODO: DCFlag_SameAsPrevious = 255 for "don't check, just merge into previous command"?
 };
 
 // create an "empty" gl3drawCmd_t with sane defaults
 static inline gl3drawCmd_t
-GL3_CreateDrawCmd(qboolean identityTrans)
+GL3_CreateDrawCmd(void)
 {
 	gl3drawCmd_t ret = {0};
-	if(identityTrans) {
-		ret.transModelMat = gl3_identityMat4;
-		ret.flags = DCFlag_IsIdentityMat;
-	}
 	ret.alpha = 1.0f;
 	ret.styles[0] = 255;
 	ret.lmtexnum = -1;
@@ -468,9 +464,10 @@ GL3_BindEBO(GLuint ebo)
 
 extern void GL3_BufferAndDraw3D(const gl3_3D_vtx_t* verts, int numVerts, GLenum drawMode, gl3drawCmd_t drawCmd);
 extern void GL3_Draw3DBatchesNow(void);
+extern void GL3_SetDrawCmdTransMatrix(gl3drawCmd_t* drawCmd, hmm_mat4 mat);
 
 extern void GL3_RotateUni3DforEntity(entity_t *e);
-extern void GL3_RotateForEntity(entity_t *e, gl3drawCmd_t* drawCmd, qboolean replaceTransModelMat);
+extern void GL3_RotateForEntity(entity_t *e, gl3drawCmd_t* drawCmd);
 
 // gl3_sdl.c
 extern int GL3_InitContext(void* win);