- rewrote vertex buffer code to require GL_ARB_BUFFER_STORAGE extension.

This means it won't work anymore on anything that doesn't support OpenGL 4.0, but I don't think this is a problem. On older NVidia cards performance gains could not be seen and on older AMDs using the vertex buffer was even worse as long as it got mixed with immediate mode rendering.

src/gl/data/gl_vertexbuffer.cpp

@@ -49,20 +49,18 @@
 #include "gl/data/gl_vertexbuffer.h"
 
 
+const int BUFFER_SIZE = 2000000;
+
 CUSTOM_CVAR(Int, gl_usevbo, -1, CVAR_ARCHIVE | CVAR_GLOBALCONFIG | CVAR_NOINITCALL)
 {
-	if (self < -1 || self > 2)
+	if (self < -1 || self > 1 || !(gl.flags & RFL_BUFFER_STORAGE))
 	{
 		self = 0;
 	}
 	else if (self == -1)
 	{
-		if (!(gl.flags & RFL_NVIDIA)) self = 0;
-		else self = 2;
-	}
-	else if (GLRenderer != NULL && GLRenderer->mVBO != NULL && GLRenderer->mVBO->vbo_arg != self)
-	{
-		Printf("Vertex buffer use will be changed for the next level.\n");
+		if (!(gl.flags & RFL_BUFFER_STORAGE)) self = 0;
+		else self = 1;
 	}
 }
 
@@ -96,13 +94,24 @@ FVertexBuffer::~FVertexBuffer()
 FFlatVertexBuffer::FFlatVertexBuffer()
 : FVertexBuffer()
 {
-	vbo_arg = gl_usevbo;
+	if (gl.flags & RFL_BUFFER_STORAGE)
+	{
+		unsigned int bytesize = BUFFER_SIZE * sizeof(FFlatVertex);
+		glBindBuffer(GL_ARRAY_BUFFER, vbo_id);
+		glBufferStorage(GL_ARRAY_BUFFER, bytesize, NULL, GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_COHERENT_BIT);
+		map = (FFlatVertex*)glMapBufferRange(GL_ARRAY_BUFFER, 0, bytesize, GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_COHERENT_BIT);
+	}
+	else
+	{
 		map = NULL;
 	}
+}
 
 FFlatVertexBuffer::~FFlatVertexBuffer()
 {
-	UnmapVBO();
+	glBindBuffer(GL_ARRAY_BUFFER, vbo_id);
+	glUnmapBuffer(GL_ARRAY_BUFFER);
+	glBindBuffer(GL_ARRAY_BUFFER, 0);
 }
 
 //==========================================================================
@@ -118,7 +127,6 @@ void FFlatVertex::SetFlatVertex(vertex_t *vt, const secplane_t & plane)
 	z = plane.ZatPoint(vt->fx, vt->fy);
 	u = vt->fx/64.f;
 	v = -vt->fy/64.f;
-	w = /*dc = df =*/ 0;
 }
 
 //==========================================================================
@@ -260,58 +268,18 @@ void FFlatVertexBuffer::CreateFlatVBO()
 //
 //==========================================================================
 
-void FFlatVertexBuffer::MapVBO()
-{
-	if (map == NULL)
-	{
-		glBindBuffer(GL_ARRAY_BUFFER, vbo_id);
-		map = (FFlatVertex*)glMapBufferRange(GL_ARRAY_BUFFER, 0, vbo_shadowdata.Size() * sizeof(FFlatVertex), 
-			GL_MAP_WRITE_BIT|GL_MAP_FLUSH_EXPLICIT_BIT|GL_MAP_UNSYNCHRONIZED_BIT);
-	}
-}
-
-//==========================================================================
-//
-//
-//
-//==========================================================================
-
-void FFlatVertexBuffer::UnmapVBO()
-{
-	if (map != NULL)
-	{
-		glUnmapBuffer(GL_ARRAY_BUFFER);
-		map = NULL;
-	}
-}
-
-//==========================================================================
-//
-//
-//
-//==========================================================================
-
 void FFlatVertexBuffer::UpdatePlaneVertices(sector_t *sec, int plane)
 {
 	int startvt = sec->vboindex[plane];
 	int countvt = sec->vbocount[plane];
 	secplane_t &splane = sec->GetSecPlane(plane);
 	FFlatVertex *vt = &vbo_shadowdata[startvt];
+	FFlatVertex *mapvt = &map[startvt];
 	for(int i=0; i<countvt; i++, vt++)
 	{
 		vt->z = splane.ZatPoint(vt->x, vt->y);
 		if (plane == sector_t::floor && sec->transdoor) vt->z -= 1;
-	}
-	if (gl.flags & RFL_MAP_BUFFER_RANGE)
-	{
-		MapVBO();
-		if (map == NULL) return;	// Error
-		memcpy(&map[startvt], &vbo_shadowdata[startvt], countvt * sizeof(FFlatVertex));
-		glFlushMappedBufferRange(GL_ARRAY_BUFFER, startvt * sizeof(FFlatVertex), countvt * sizeof(FFlatVertex));
-	}
-	else
-	{
-		glBufferSubData(GL_ARRAY_BUFFER, startvt * sizeof(FFlatVertex), countvt * sizeof(FFlatVertex), &vbo_shadowdata[startvt]);
+		mapvt->z = vt->z;
 	}
 }
 
@@ -324,11 +292,10 @@ void FFlatVertexBuffer::UpdatePlaneVertices(sector_t *sec, int plane)
 void FFlatVertexBuffer::CreateVBO()
 {
 	vbo_shadowdata.Clear();
-	if (vbo_arg > 0)
+	if (gl.flags & RFL_BUFFER_STORAGE)
 	{
 		CreateFlatVBO();
-		glBindBuffer(GL_ARRAY_BUFFER, vbo_id);
-		glBufferData(GL_ARRAY_BUFFER, vbo_shadowdata.Size() * sizeof(FFlatVertex), &vbo_shadowdata[0], GL_DYNAMIC_DRAW);
+		memcpy(map, &vbo_shadowdata[0], vbo_shadowdata.Size() * sizeof(FFlatVertex));
 	}
 	else if (sectors)
 	{
@@ -350,16 +317,14 @@ void FFlatVertexBuffer::CreateVBO()
 
 void FFlatVertexBuffer::BindVBO()
 {
-	if (vbo_arg > 0)
+	if (gl.flags & RFL_BUFFER_STORAGE)
 	{
-		UnmapVBO();
 		glBindBuffer(GL_ARRAY_BUFFER, vbo_id);
 		glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
 		glVertexPointer(3,GL_FLOAT, sizeof(FFlatVertex), &VTO->x);
 		glTexCoordPointer(2,GL_FLOAT, sizeof(FFlatVertex), &VTO->u);
 		glEnableClientState(GL_VERTEX_ARRAY);
 		glEnableClientState(GL_TEXTURE_COORD_ARRAY);
-		glDisableClientState(GL_INDEX_ARRAY);
 	}
 }
 
@@ -370,10 +335,12 @@ void FFlatVertexBuffer::BindVBO()
 //==========================================================================
 
 void FFlatVertexBuffer::CheckPlanes(sector_t *sector)
+{
+	if (gl.flags & RFL_BUFFER_STORAGE)
 	{
 		if (sector->GetPlaneTexZ(sector_t::ceiling) != sector->vboheight[sector_t::ceiling])
 		{
-		if (sector->ceilingdata == NULL) // only update if there's no thinker attached
+			//if (sector->ceilingdata == NULL) // only update if there's no thinker attached
 			{
 				UpdatePlaneVertices(sector, sector_t::ceiling);
 				sector->vboheight[sector_t::ceiling] = sector->GetPlaneTexZ(sector_t::ceiling);
@@ -381,13 +348,14 @@ void FFlatVertexBuffer::CheckPlanes(sector_t *sector)
 		}
 		if (sector->GetPlaneTexZ(sector_t::floor) != sector->vboheight[sector_t::floor])
 		{
-		if (sector->floordata == NULL) // only update if there's no thinker attached
+			//if (sector->floordata == NULL) // only update if there's no thinker attached
 			{
 				UpdatePlaneVertices(sector, sector_t::floor);
 				sector->vboheight[sector_t::floor] = sector->GetPlaneTexZ(sector_t::floor);
 			}
 		}
 	}
+}
 
 //==========================================================================
 //

src/gl/data/gl_vertexbuffer.h

@@ -20,11 +20,10 @@ public:
 	virtual void BindVBO() = 0;
 };
 
-struct FFlatVertex	// exactly 32 bytes large
+struct FFlatVertex
 {
-	float x,z,y,w;	// w only for padding to make one vertex 32 bytes - maybe it will find some use later
+	float x,z,y;	// world position
 	float u,v;		// texture coordinates
-	//float dc, df;	// distance to floor and ceiling on walls - used for glowing
 
 	void SetFlatVertex(vertex_t *vt, const secplane_t &plane);
 };
@@ -35,27 +34,27 @@ struct FFlatVertex	// exactly 32 bytes large
 class FFlatVertexBuffer : public FVertexBuffer
 {
 	FFlatVertex *map;
+	unsigned int mIndex;
 
-	void MapVBO();
 	void CheckPlanes(sector_t *sector);
 
 public:
 	int vbo_arg;
-	TArray<FFlatVertex> vbo_shadowdata;	// this is kept around for non-VBO rendering
+	TArray<FFlatVertex> vbo_shadowdata;	// this is kept around for updating the actual (non-readable) buffer
 
-public:
 	FFlatVertexBuffer();
 	~FFlatVertexBuffer();
 
+	void CreateVBO();
+	void BindVBO();
+	void CheckUpdate(sector_t *sector);
+
+private:
 	int CreateSubsectorVertices(subsector_t *sub, const secplane_t &plane, int floor);
 	int CreateSectorVertices(sector_t *sec, const secplane_t &plane, int floor);
 	int CreateVertices(int h, sector_t *sec, const secplane_t &plane, int floor);
 	void CreateFlatVBO();
-	void CreateVBO();
 	void UpdatePlaneVertices(sector_t *sec, int plane);
-	void BindVBO();
-	void CheckUpdate(sector_t *sector);
-	void UnmapVBO();
 
 };
 

src/gl/models/gl_voxels.cpp

@@ -506,17 +506,18 @@ void FVoxelModel::RenderFrame(FTexture * skin, int frame, int cm, int translatio
 {
 	FMaterial * tex = FMaterial::ValidateTexture(skin);
 	tex->Bind(cm, 0, translation);
-	gl_RenderState.Apply();
 
 	if (mVBO == NULL) MakeGLData();
 	if (mVBO != NULL)
 	{
-		mVBO->BindVBO();
+		gl_RenderState.SetVertexBuffer(mVBO);
+		gl_RenderState.Apply();
 		glDrawElements(GL_QUADS, mIndices.Size(), mVBO->IsInt() ? GL_UNSIGNED_INT : GL_UNSIGNED_SHORT, 0);
-		GLRenderer->mVBO->BindVBO();
-		return;
+		gl_RenderState.SetVertexBuffer(GLRenderer->mVBO);
 	}
-
+	else
+	{
+		gl_RenderState.Apply();
 		glBegin(GL_QUADS);
 		for (unsigned i = 0; i < mIndices.Size(); i++)
 		{
@@ -526,6 +527,7 @@ void FVoxelModel::RenderFrame(FTexture * skin, int frame, int cm, int translatio
 		}
 		glEnd();
 	}
+}
 
 //===========================================================================
 //

src/gl/renderer/gl_renderstate.cpp

@@ -41,6 +41,7 @@
 #include "gl/system/gl_system.h"
 #include "gl/system/gl_interface.h"
 #include "gl/data/gl_data.h"
+#include "gl/data/gl_vertexbuffer.h"
 #include "gl/system/gl_cvars.h"
 #include "gl/shaders/gl_shader.h"
 #include "gl/renderer/gl_renderer.h"
@@ -79,6 +80,7 @@ void FRenderState::Reset()
 	mBlendEquation = GL_FUNC_ADD;
 	glBlendEquation = -1;
 	m2D = true;
+	mVertexBuffer = mCurrentVertexBuffer = NULL;
 }
 
 
@@ -300,6 +302,12 @@ void FRenderState::Apply(bool forcenoshader)
 		}
 	}
 
+	if (mVertexBuffer != mCurrentVertexBuffer)
+	{
+		if (mVertexBuffer == NULL) glBindBuffer(GL_ARRAY_BUFFER, 0);
+		else mVertexBuffer->BindVBO();
+		mCurrentVertexBuffer = mVertexBuffer;
+	}
 	if (forcenoshader || !ApplyShader())
 	{
 		GLRenderer->mShaderManager->SetActiveShader(NULL);

src/gl/renderer/gl_renderstate.h

@@ -5,6 +5,8 @@
 #include "c_cvars.h"
 #include "r_defs.h"
 
+class FVertexBuffer;
+
 EXTERN_CVAR(Bool, gl_direct_state_change)
 
 struct FStateAttr
@@ -104,6 +106,8 @@ class FRenderState
 	int mBlendEquation;
 	bool m2D;
 
+	FVertexBuffer *mVertexBuffer, *mCurrentVertexBuffer;
+
 	FStateVec3 mCameraPos;
 	FStateVec4 mGlowTop, mGlowBottom;
 	FStateVec4 mGlowTopPlane, mGlowBottomPlane;
@@ -140,6 +144,11 @@ public:
 	int SetupShader(bool cameratexture, int &shaderindex, int &cm, float warptime);
 	void Apply(bool forcenoshader = false);
 
+	void SetVertexBuffer(FVertexBuffer *vb)
+	{
+		mVertexBuffer = vb;
+	}
+
 	void SetTextureMode(int mode)
 	{
 		mTextureMode = mode;

src/gl/scene/gl_flats.cpp

@@ -272,7 +272,7 @@ void GLFlat::DrawSubsectors(int pass, bool istrans)
 	}
 	else
 	{
-		if (vboindex >= 0)
+		if (gl_usevbo && vboindex >= 0)
 		{
 			//glColor3f( 1.f,.5f,.5f);
 			int index = vboindex;

src/gl/scene/gl_scene.cpp

@@ -353,7 +353,6 @@ void FGLRenderer::CreateScene()
 	gl_drawinfo->HandleHackedSubsectors();	// open sector hacks for deep water
 	gl_drawinfo->ProcessSectorStacks();		// merge visplanes of sector stacks
 
-	GLRenderer->mVBO->UnmapVBO ();
 	ProcessAll.Unclock();
 
 }
@@ -931,7 +930,7 @@ void FGLRenderer::RenderView (player_t* player)
 		LastCamera=player->camera;
 	}
 
-	mVBO->BindVBO();
+	gl_RenderState.SetVertexBuffer(mVBO);
 
 	// reset statistics counters
 	ResetProfilingData();

src/gl/system/gl_interface.cpp

@@ -155,26 +155,19 @@ void gl_LoadExtensions()
 
 	if (CheckExtension("GL_ARB_texture_compression")) gl.flags|=RFL_TEXTURE_COMPRESSION;
 	if (CheckExtension("GL_EXT_texture_compression_s3tc")) gl.flags|=RFL_TEXTURE_COMPRESSION_S3TC;
-	if (strstr(gl.vendorstring, "NVIDIA")) gl.flags|=RFL_NVIDIA;
-	else if (strstr(gl.vendorstring, "ATI Technologies")) gl.flags|=RFL_ATI;
+	if (CheckExtension("GL_ARB_buffer_storage")) gl.flags |= RFL_BUFFER_STORAGE;
 
-	if (strcmp(version, "2.0") >= 0) gl.flags|=RFL_GL_20;
-	if (strcmp(version, "2.1") >= 0) gl.flags|=RFL_GL_21;
-	if (strcmp(version, "3.0") >= 0) gl.flags|=RFL_GL_30;
+	gl.version = strtod((char*)glGetString(GL_VERSION), NULL);
 
 	glGetIntegerv(GL_MAX_TEXTURE_SIZE,&gl.max_texturesize);
 	glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
 	
-	if (gl.flags & RFL_GL_20)
-	{
 	// Rules:
 	// SM4 will always use shaders. No option to switch them off is needed here.
 	// SM3 has shaders optional but they are off by default (they will have a performance impact
 	// SM2 only uses shaders for colormaps on camera textures and has no option to use them in general.
 	//     On SM2 cards the shaders will be too slow and show visual bugs (at least on GF 6800.)
 	if (strcmp((const char*)glGetString(GL_SHADING_LANGUAGE_VERSION), "1.3") >= 0) gl.shadermodel = 4;
-		else if (CheckExtension("GL_NV_GPU_shader4")) gl.shadermodel = 4;	// for pre-3.0 drivers that support GF8xxx.
-		else if (CheckExtension("GL_EXT_GPU_shader4")) gl.shadermodel = 4;	// for pre-3.0 drivers that support GF8xxx.
 	else if (CheckExtension("GL_NV_vertex_program3")) gl.shadermodel = 3;
 	else if (!strstr(gl.vendorstring, "NVIDIA")) gl.shadermodel = 3;
 	else gl.shadermodel = 2;	// Only for older NVidia cards which had notoriously bad shader support.
@@ -182,14 +175,8 @@ void gl_LoadExtensions()
 	// Command line overrides for testing and problem cases.
 	if (Args->CheckParm("-sm2") && gl.shadermodel > 2) gl.shadermodel = 2;
 	else if (Args->CheckParm("-sm3") && gl.shadermodel > 3) gl.shadermodel = 3;
-	}
 
-	if (CheckExtension("GL_ARB_map_buffer_range")) 
-	{
-		gl.flags|=RFL_MAP_BUFFER_RANGE;
-	}
-
-	if (gl.flags & RFL_GL_30)
+	if (gl.version >= 3.f)
 	{
 		gl.flags|=RFL_FRAMEBUFFER;
 	}

src/gl/system/gl_interface.h

@@ -9,16 +9,8 @@ enum RenderFlags
 	RFL_TEXTURE_COMPRESSION=8,
 	RFL_TEXTURE_COMPRESSION_S3TC=16,
 
-	RFL_MAP_BUFFER_RANGE = 64,
-	RFL_FRAMEBUFFER = 128,
-	RFL_TEXTUREBUFFER = 256,
-	RFL_NVIDIA = 512,
-	RFL_ATI = 1024,
-
-
-	RFL_GL_20 = 0x10000000,
-	RFL_GL_21 = 0x20000000,
-	RFL_GL_30 = 0x40000000,
+	RFL_FRAMEBUFFER = 32,
+	RFL_BUFFER_STORAGE = 64,
 };
 
 enum TexMode
@@ -40,6 +32,7 @@ struct RenderContext
 	unsigned int flags;
 	unsigned int shadermodel;
 	unsigned int maxuniforms;
+	float version;
 	int max_texturesize;
 	char * vendorstring;