[win32, canvas] use glBindBufferBase() and explicit uniform offset to avoid glBindBufferRange() implementation-specific alignment issues

src/gl_canvas.c

@@ -966,6 +966,11 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 	int pathCount = backend->pathCount - backend->pathBatchStart;
 	int eltCount = backend->eltCount - backend->eltBatchStart;
 
+	if(!pathCount || !eltCount)
+	{
+		return;
+	}
+
 	glBindBuffer(GL_SHADER_STORAGE_BUFFER, pathBuffer);
 	glFlushMappedBufferRange(GL_SHADER_STORAGE_BUFFER, pathBufferOffset, pathCount*sizeof(mg_gl_path));
 
@@ -992,6 +997,12 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 
 	glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 
+	int err = glGetError();
+	if(err)
+	{
+		log_error("gl error %i\n", err);
+	}
+
 	//NOTE: path setup pass
 	int maxWorkGroupCount = 0;
 	glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_COUNT, 0, &maxWorkGroupCount);
@@ -1009,17 +1020,27 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 
 	for(int i=0; i<pathCount; i += maxWorkGroupCount)
 	{
-		int pathOffset = pathBufferOffset + i*sizeof(mg_gl_path);
-		int pathQueueOffset = i*sizeof(mg_gl_path_queue);
 		int count = minimum(maxWorkGroupCount, pathCount-i);
 
-		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer, pathOffset, count*sizeof(mg_gl_path));
+		glUniform1i(2, backend->pathBatchStart + i);
-		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 1, backend->pathQueueBuffer, pathQueueOffset, count*sizeof(mg_gl_path_queue));
+		glUniform1i(3, i);
+
+		glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer);
+		glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, backend->pathQueueBuffer);
 
 		glDispatchCompute(count, 1, 1);
 		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 	}
 
+	if(!err)
+	{
+		err = glGetError();
+		if(err)
+		{
+			log_error("gl error %i\n", err);
+		}
+	}
+
 	//NOTE: segment setup pass
 	glUseProgram(backend->segmentSetup);
 
@@ -1038,12 +1059,23 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 		int offset = elementBufferOffset + i*sizeof(mg_gl_path_elt);
 		int count = minimum(maxWorkGroupCount, eltCount-i);
 
-		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, elementBuffer, offset, count*sizeof(mg_gl_path_elt));
+		glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, elementBuffer);
+
+		glUniform1i(2, (backend->eltBatchStart + i));
 
 		glDispatchCompute(count, 1, 1);
 		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 	}
 
+	if(!err)
+	{
+		err = glGetError();
+		if(err)
+		{
+			log_error("gl error %i\n", err);
+		}
+	}
+
 	//NOTE: backprop pass
 	glUseProgram(backend->backprop);
 
@@ -1051,19 +1083,28 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 
 	for(int i=0; i<pathCount; i += maxWorkGroupCount)
 	{
-		int offset = i*sizeof(mg_gl_path_queue);
 		int count = minimum(maxWorkGroupCount, pathCount-i);
 
-		glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, backend->pathQueueBuffer, offset, count*sizeof(mg_gl_path_queue));
+		glUniform1i(0, i);
+		glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, backend->pathQueueBuffer);
 
 		glDispatchCompute(count, 1, 1);
 		glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 	}
 
+	if(!err)
+	{
+		err = glGetError();
+		if(err)
+		{
+			log_error("gl error %i\n", err);
+		}
+	}
+
 	//NOTE: merge pass
 	glUseProgram(backend->merge);
 
-	glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer, pathBufferOffset, pathCount*sizeof(mg_gl_path));
+	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, backend->pathQueueBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, backend->tileQueueBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, backend->tileOpCountBuffer);
@@ -1085,13 +1126,23 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 		glUniform1i(3, 1);
 	}
 
+	glUniform1i(4, backend->pathBatchStart);
+
 	glDispatchCompute(nTilesX, nTilesY, 1);
 	glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 
+	if(!err)
+	{
+		err = glGetError();
+		if(err)
+		{
+			log_error("gl error %i\n", err);
+		}
+	}
 	//NOTE: raster pass
 	glUseProgram(backend->raster);
 
-	glBindBufferRange(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer, pathBufferOffset, pathCount*sizeof(mg_gl_path));
+	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, pathBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, backend->segmentBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 2, backend->tileOpBuffer);
 	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 3, backend->screenTilesBuffer);
@@ -1113,9 +1164,20 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 		glUniform1ui(2, 0);
 	}
 
+	glUniform1i(3, backend->pathBatchStart);
+
 	glBindBuffer(GL_DISPATCH_INDIRECT_BUFFER, backend->rasterDispatchBuffer);
 	glDispatchComputeIndirect(0);
 
+	if(!err)
+	{
+		err = glGetError();
+		if(err)
+		{
+			log_error("gl error %i\n", err);
+		}
+	}
+
 	//NOTE: blit pass
 	glUseProgram(backend->blit);
 	glBindBuffer(GL_ARRAY_BUFFER, backend->dummyVertexBuffer);
@@ -1125,11 +1187,14 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 
 	glDrawArrays(GL_TRIANGLES, 0, 6);
 
-	int err = glGetError();
+	if(!err)
+	{
+		err = glGetError();
 		if(err)
 		{
 			log_error("gl error %i\n", err);
 		}
+	}
 
 	backend->pathBatchStart = backend->pathCount;
 	backend->eltBatchStart = backend->eltCount;

src/glsl_shaders/backprop.glsl

@@ -14,6 +14,8 @@ layout(binding = 1) restrict buffer tileQueueBufferSSBO
 	mg_gl_tile_queue elements[];
 } tileQueueBuffer;
 
+layout(location = 0) uniform int pathQueueBufferStart;
+
 shared int nextRowIndex;
 
 void main()
@@ -28,7 +30,7 @@ void main()
 	barrier();
 
 	int rowIndex = 0;
-	mg_gl_path_queue pathQueue = pathQueueBuffer.elements[pathIndex];
+	mg_gl_path_queue pathQueue = pathQueueBuffer.elements[pathQueueBufferStart + pathIndex];
 	int tileQueueBase = pathQueue.tileQueues;
 	int rowSize = pathQueue.area.z;
 	int rowCount = pathQueue.area.w;

src/glsl_shaders/merge.glsl

@@ -44,6 +44,7 @@ layout(location = 0) uniform int tileSize;
 layout(location = 1) uniform float scale;
 layout(location = 2) uniform int pathCount;
 layout(location = 3) uniform int cullSolidTiles;
+layout(location = 4) uniform int pathBufferStart;
 
 void main()
 {
@@ -60,8 +61,8 @@ void main()
 		mg_gl_path_queue pathQueue = pathQueueBuffer.elements[pathIndex];
 		ivec2 pathTileCoord = tileCoord - pathQueue.area.xy;
 
-		vec4 pathBox = pathBuffer.elements[pathIndex].box;
+		vec4 pathBox = pathBuffer.elements[pathBufferStart + pathIndex].box;
-		vec4 pathClip = pathBuffer.elements[pathIndex].clip;
+		vec4 pathClip = pathBuffer.elements[pathBufferStart + pathIndex].clip;
 
 		float xMax = min(pathBox.z, pathClip.z);
 		int tileMax = int(xMax * scale) / tileSize;
@@ -87,7 +88,7 @@ void main()
 
 			vec4 tileBox = vec4(tileCoord.x, tileCoord.y, tileCoord.x+1, tileCoord.y+1);
 			tileBox *= tileSize;
-			vec4 clip = pathBuffer.elements[pathIndex].clip * scale;
+			vec4 clip = pathBuffer.elements[pathBufferStart + pathIndex].clip * scale;
 
 			if(  tileBox.x >= clip.z
 			  || tileBox.z < clip.x
@@ -126,7 +127,7 @@ void main()
 					{
 						tileOpBuffer.elements[pathOpIndex].kind = MG_GL_OP_FILL;
 
-						if(  pathBuffer.elements[pathIndex].color.a == 1
+						if(  pathBuffer.elements[pathBufferStart + pathIndex].color.a == 1
 						  && cullSolidTiles != 0)
 						{
 							screenTilesBuffer.elements[tileIndex].first = pathOpIndex;

src/glsl_shaders/path_setup.glsl

@@ -26,11 +26,13 @@ layout(binding = 3) restrict writeonly buffer tileQueueBufferSSBO
 
 layout(location = 0) uniform int tileSize;
 layout(location = 1) uniform float scale;
+layout(location = 2) uniform int pathBufferStart;
+layout(location = 3) uniform int pathQueueBufferStart;
 
 void main()
 {
 	uint pathIndex = gl_WorkGroupID.x;
-	const mg_gl_path path = pathBuffer.elements[pathIndex];
+	const mg_gl_path path = pathBuffer.elements[pathIndex + pathBufferStart];
 
 	//NOTE: we don't clip on the right, since we need those tiles to accurately compute
 	//      the prefix sum of winding increments in the backprop pass.
@@ -48,8 +50,8 @@ void main()
 
 	int tileQueuesIndex = atomicAdd(tileQueueCountBuffer.elements[0], tileCount);
 
-	pathQueueBuffer.elements[pathIndex].area = ivec4(firstTile.x, firstTile.y, nTilesX, nTilesY);
+	pathQueueBuffer.elements[pathQueueBufferStart + pathIndex].area = ivec4(firstTile.x, firstTile.y, nTilesX, nTilesY);
-	pathQueueBuffer.elements[pathIndex].tileQueues = tileQueuesIndex;
+	pathQueueBuffer.elements[pathQueueBufferStart + pathIndex].tileQueues = tileQueuesIndex;
 
 	for(int i=0; i<tileCount; i++)
 	{

src/glsl_shaders/raster.glsl

@@ -27,6 +27,7 @@ layout(binding = 3) restrict readonly buffer screenTilesBufferSSBO
 layout(location = 0) uniform float scale;
 layout(location = 1) uniform int msaaSampleCount;
 layout(location = 2) uniform uint useTexture;
+layout(location = 3) uniform int pathBufferStart;
 
 layout(rgba8, binding = 0) uniform restrict writeonly image2D outTexture;
 layout(binding = 1) uniform sampler2D srcTexture;
@@ -129,7 +130,7 @@ void main()
 		{
 			int pathIndex = op.index;
 
-			vec4 nextColor = pathBuffer.elements[pathIndex].color;
+			vec4 nextColor = pathBuffer.elements[pathBufferStart + pathIndex].color;
 			nextColor.rgb *= nextColor.a;
 
 			if(useTexture != 0)
@@ -139,7 +140,7 @@ void main()
 				{
 					vec2 sampleCoord = imgSampleCoords[sampleIndex];
 					vec3 ph = vec3(sampleCoord.xy, 1);
-					vec2 uv = (pathBuffer.elements[pathIndex].uvTransform * ph).xy;
+					vec2 uv = (pathBuffer.elements[pathBufferStart + pathIndex].uvTransform * ph).xy;
 					texColor += texture(srcTexture, uv);
 				}
 				texColor /= srcSampleCount;
@@ -153,7 +154,7 @@ void main()
 			}
 			else
 			{
-				vec4 clip = pathBuffer.elements[pathIndex].clip * scale;
+				vec4 clip = pathBuffer.elements[pathBufferStart + pathIndex].clip * scale;
 				float coverage = 0;
 
 				for(int sampleIndex = 0; sampleIndex<sampleCount; sampleIndex++)
@@ -166,8 +167,10 @@ void main()
 					  && sampleCoord.y < clip.w)
 					{
 						bool filled = op.kind == MG_GL_OP_CLIP_FILL
-						            ||(pathBuffer.elements[pathIndex].cmd == MG_GL_FILL && ((winding[sampleIndex] & 1) != 0))
+						            ||(pathBuffer.elements[pathBufferStart + pathIndex].cmd == MG_GL_FILL
-						            ||(pathBuffer.elements[pathIndex].cmd == MG_GL_STROKE && (winding[sampleIndex] != 0));
+						              && ((winding[sampleIndex] & 1) != 0))
+						            ||(pathBuffer.elements[pathBufferStart + pathIndex].cmd == MG_GL_STROKE
+						              && (winding[sampleIndex] != 0));
 						if(filled)
 						{
 							coverage++;

src/glsl_shaders/segment_setup.glsl

@@ -41,6 +41,7 @@ layout(binding = 6) restrict buffer tileOpBufferSSBO
 
 layout(location = 0) uniform float scale;
 layout(location = 1) uniform uint tileSize;
+layout(location = 2) uniform int elementBufferStart;
 
 void bin_to_tiles(int segIndex)
 {
@@ -824,7 +825,7 @@ void main()
 {
 	int eltIndex = int(gl_WorkGroupID.x);
 
-	mg_gl_path_elt elt = elementBuffer.elements[eltIndex];
+	mg_gl_path_elt elt = elementBuffer.elements[elementBufferStart + eltIndex];
 
 	switch(elt.kind)
 	{