- use uniform buffers for dynamic lights everywhere.

Branching on SSBO content does not work that well on NVidia so it is better disabled.
So far only implemented for OpenGL - Vulkan needs more work.

src/common/rendering/gl/gl_shader.cpp

@@ -391,25 +391,17 @@ bool FShader::Load(const char * name, const char * vert_prog_lump, const char *
 	assert(screen->mLights != NULL);
 	assert(screen->mBones != NULL);
 
-	bool lightbuffertype = screen->mLights->GetBufferType();
-	unsigned int lightbuffersize = screen->mLights->GetBlockSize();
-	if (!lightbuffertype)
-	{
-		vp_comb.Format("#version 330 core\n#define NUM_UBO_LIGHTS %d\n#define NUM_UBO_BONES %d\n", lightbuffersize, screen->mBones->GetBlockSize());
-	}
-	else
-	{
-		// This differentiation is for Intel which do not seem to expose the full extension, even if marked as required.
-		if (gl.glslversion < 4.3f)
-			vp_comb = "#version 400 core\n#extension GL_ARB_shader_storage_buffer_object : require\n#define SHADER_STORAGE_LIGHTS\n#define SHADER_STORAGE_BONES\n";
-		else
-			vp_comb = "#version 430 core\n#define SHADER_STORAGE_LIGHTS\n#define SHADER_STORAGE_BONES\n";
-	}
 
 	if ((gl.flags & RFL_SHADER_STORAGE_BUFFER) && screen->allowSSBO())
-	{
+		vp_comb << "#version 430 core\n#define SUPPORTS_SHADOWMAPS\n";
-		vp_comb << "#define SUPPORTS_SHADOWMAPS\n";
+	else 
-	}
+		vp_comb << "#version 330 core\n";
+
+	bool lightbuffertype = screen->mLights->GetBufferType();
+	if (!lightbuffertype)
+		vp_comb.AppendFormat("#define NUM_UBO_LIGHTS %d\n#define NUM_UBO_BONES %d\n", screen->mLights->GetBlockSize(), screen->mBones->GetBlockSize());
+	else
+		vp_comb << "#define SHADER_STORAGE_LIGHTS\n#define SHADER_STORAGE_BONES\n";
 
 	FString fp_comb = vp_comb;
 	vp_comb << defines << i_data.GetChars();

src/common/rendering/gl_load/gl_interface.cpp

@@ -108,6 +108,8 @@ static void InitContext()
 
 void gl_LoadExtensions()
 {
+	int v = 0;
+
 	InitContext();
 	CollectExtensions();
 
@@ -162,32 +164,18 @@ void gl_LoadExtensions()
 #endif
 		gl.glslversion = 3.31f;	// Force GLSL down to 3.3.
 	}
-	else if (gl_version < 4.5f)
+	else if (gl_version >= 4.5f)
-	{
-		// don't use GL 4.x features when running a GL 3.x context.
-		if (CheckExtension("GL_ARB_buffer_storage"))
-		{
-			// work around a problem with older AMD drivers: Their implementation of shader storage buffer objects is piss-poor and does not match uniform buffers even closely.
-			// Recent drivers, GL 4.4 don't have this problem, these can easily be recognized by also supporting the GL_ARB_buffer_storage extension.
-			if (CheckExtension("GL_ARB_shader_storage_buffer_object"))
-			{
-				gl.flags |= RFL_SHADER_STORAGE_BUFFER;
-			}
-			gl.flags |= RFL_BUFFER_STORAGE;
-		}
-	}
-	else
 	{
 		// Assume that everything works without problems on GL 4.5 drivers where these things are core features.
 		gl.flags |= RFL_SHADER_STORAGE_BUFFER | RFL_BUFFER_STORAGE;
+
+		// Mesa implements shader storage only for fragment shaders.
+		// Just disable the feature there. The light buffer may just use a uniform buffer without any adverse effects.
+		glGetIntegerv(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, &v);
+		if (v == 0)
+			gl.flags &= ~RFL_SHADER_STORAGE_BUFFER;
 	}
 
-	// Mesa implements shader storage only for fragment shaders.
-	// Just disable the feature there. The light buffer may just use a uniform buffer without any adverse effects.
-	int v = 0;
-	glGetIntegerv(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, &v);
-	if (v == 0)
-		gl.flags &= ~RFL_SHADER_STORAGE_BUFFER;
 
 
 	if (gl_version >= 4.3f || CheckExtension("GL_ARB_invalidate_subdata")) gl.flags |= RFL_INVALIDATE_BUFFER;

src/common/rendering/hwrenderer/data/hw_bonebuffer.cpp

@@ -33,10 +33,7 @@ BoneBuffer::BoneBuffer(int pipelineNbr) : mPipelineNbr(pipelineNbr)
 	mBufferSize = maxNumberOfBones;
 	mByteSize = mBufferSize * BONE_SIZE;
 
-	// Hack alert: On Intel's GL driver SSBO's perform quite worse than UBOs.
+	if (screen->useSSBO())
-	// We only want to disable using SSBOs for bones but not disable the feature entirely.
-	// Note that using an uniform buffer here will limit the number of bones per model so it isn't done for NVidia and AMD.
-	if (screen->IsVulkan() || screen->IsPoly() || ((screen->hwcaps & RFL_SHADER_STORAGE_BUFFER) && screen->allowSSBO() && !strstr(screen->vendorstring, "Intel")))
 	{
 		mBufferType = true;
 		mBlockAlign = 0;

src/common/rendering/hwrenderer/data/hw_lightbuffer.cpp

@@ -41,10 +41,7 @@ FLightBuffer::FLightBuffer(int pipelineNbr):
 	mBufferSize = maxNumberOfLights * ELEMENTS_PER_LIGHT;
 	mByteSize = mBufferSize * ELEMENT_SIZE;
 
-	// Hack alert: On Intel's GL driver SSBO's perform quite worse than UBOs.
+	if (screen->useSSBO())
-	// We only want to disable using SSBOs for lights but not disable the feature entirely.
-	// Note that using an uniform buffer here will limit the number of lights per surface so it isn't done for NVidia and AMD.
-	if (screen->IsVulkan() || screen->IsPoly() || ((screen->hwcaps & RFL_SHADER_STORAGE_BUFFER) && screen->allowSSBO() && !strstr(screen->vendorstring, "Intel")))
 	{
 		mBufferType = true;
 		mBlockAlign = 0;

src/common/rendering/v_video.h

@@ -170,7 +170,7 @@ public:
 		mShadowMap.SetAABBTree(tree);
 	}
 	virtual void SetLevelMesh(hwrenderer::LevelMesh *mesh) { }
-	bool allowSSBO()
+	bool allowSSBO() const
 	{
 #ifndef HW_BLOCK_SSBO
 		return true;
@@ -179,6 +179,14 @@ public:
 #endif
 	}
 
+	// Hack alert: On Intel's GL driver SSBO's perform quite worse than UBOs.
+	// We only want to disable using SSBOs for lights but not disable the feature entirely.
+	// Note that using an uniform buffer here will limit the number of lights per surface so it isn't done for NVidia and AMD.
+	bool useSSBO() 
+	{
+		return IsVulkan();// || ((hwcaps & RFL_SHADER_STORAGE_BUFFER) && allowSSBO() && !strstr(vendorstring, "Intel"));
+	}
+
 	virtual DCanvas* GetCanvas() { return nullptr; }
 
 	void SetSize(int width, int height);

src/doomdef.h

@@ -249,6 +249,7 @@ enum
 	BCOMPATF_FLOATBOB			= 1 << 8,	// Use Hexen's original method of preventing floatbobbing items from falling down
 	BCOMPATF_NOSLOPEID			= 1 << 9,	// disable line IDs on slopes.
 	BCOMPATF_CLIPMIDTEX			= 1 << 10,	// Always Clip midtex's in the software renderer (required to run certain GZDoom maps, has no effect in the hardware renderer)
+	BCOMPATF_NOSECTIONMERGE		= 1 << 11,	// (for IWAD maps) keep separate sections for sectors with intra-sector linedefs. 
 };
 
 // phares 3/20/98:

src/maploader/compatibility.cpp

@@ -126,6 +126,8 @@ static FCompatOption Options[] =
 	{ "floatbob",				BCOMPATF_FLOATBOB, SLOT_BCOMPAT },
 	{ "noslopeid",				BCOMPATF_NOSLOPEID, SLOT_BCOMPAT },
 	{ "clipmidtex",				BCOMPATF_CLIPMIDTEX, SLOT_BCOMPAT },
+	{ "nosectionmerge",			BCOMPATF_NOSECTIONMERGE, SLOT_BCOMPAT },
+
 
 	// list copied from g_mapinfo.cpp
 	{ "shorttex",				COMPATF_SHORTTEX, SLOT_COMPAT },
@@ -288,12 +290,13 @@ FName MapLoader::CheckCompatibility(MapData *map)
 	Level->ii_compatflags2 = 0;
 	Level->ib_compatflags = 0;
 
-	// When playing Doom IWAD levels force COMPAT_SHORTTEX and COMPATF_LIGHT.
+	// When playing Doom IWAD levels force BCOMPATF_NOSECTIONMERGE, COMPAT_SHORTTEX and COMPATF_LIGHT.
-	// I'm not sure if the IWAD maps actually need COMPATF_LIGHT but it certainly does not hurt.
+	// I'm not sure if the IWAD maps actually need COMPATF_LIGHT but it certainly does not hurt. BCOMPATF_NOSECTIONMERGE is mainly for MAP18's sector 0
 	// TNT's MAP31 also needs COMPATF_STAIRINDEX but that only gets activated for TNT.WAD.
 	if (fileSystem.GetFileContainer(map->lumpnum) == fileSystem.GetIwadNum() && (gameinfo.flags & GI_COMPATSHORTTEX) && Level->maptype == MAPTYPE_DOOM)
 	{
 		Level->ii_compatflags = COMPATF_SHORTTEX|COMPATF_LIGHT;
+		Level->ib_compatflags = BCOMPATF_NOSECTIONMERGE;
 		if (gameinfo.flags & GI_COMPATSTAIRS) Level->ii_compatflags |= COMPATF_STAIRINDEX;
 	}
 

src/r_data/r_sections.cpp

@@ -243,7 +243,8 @@ public:
 				{
 					if (sub->firstline[i].PartnerSeg && sub->firstline[i].Subsector->render_sector == sub->firstline[i].PartnerSeg->Subsector->render_sector)
 					{
-						seglist.Push(sub->firstline[i].PartnerSeg);
+						if (sub->firstline[i].sidedef == nullptr || !(Level->ib_compatflags & BCOMPATF_NOSECTIONMERGE))
+							seglist.Push(sub->firstline[i].PartnerSeg);
 					}
 				}
 			};