[wip, win32, canvas] porting new canvas renderer to opengl

build.bat

@@ -1,7 +1,10 @@
 
+@echo off
+setlocal EnableDelayedExpansion
+
 if not exist bin mkdir bin
 
-set glsl_shaders=src\glsl_shaders\common.glsl src\glsl_shaders\blit_vertex.glsl src\glsl_shaders\blit_fragment.glsl src\glsl_shaders\clear_counters.glsl src\glsl_shaders\tile.glsl src\glsl_shaders\sort.glsl src\glsl_shaders\draw.glsl
+set glsl_shaders=src\glsl_shaders\common.glsl src\glsl_shaders\blit_vertex.glsl src\glsl_shaders\blit_fragment.glsl src\glsl_shaders\path_setup.glsl src\glsl_shaders\segment_setup.glsl src\glsl_shaders\backprop.glsl src\glsl_shaders\merge.glsl src\glsl_shaders\raster.glsl
 
 call python3 scripts\embed_text.py %glsl_shaders% --prefix=glsl_ --output src\glsl_shaders.h
 

src/glsl_shaders/backprop.glsl

@@ -0,0 +1,20 @@
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+
+precision mediump float;
+layout(std430) buffer;
+
+layout(binding = 0) restrict readonly buffer pathQueueBufferSSBO
+{
+	mg_gl_path_queue elements[];
+} pathQueueBuffer;
+
+layout(binding = 1) restrict writeonly buffer tileQueueBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} tileQueueBuffer;
+
+void main()
+{
+
+}

src/glsl_shaders/clear_counters.glsl

@@ -1,15 +0,0 @@
-
-layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
-
-precision mediump float;
-layout(std430) buffer;
-
-layout(binding = 0) coherent restrict writeonly buffer tileCounterBufferSSBO {
-	uint elements[];
-} tileCounterBuffer ;
-
-void main()
-{
-	uint tileIndex = gl_WorkGroupID.x;
-	tileCounterBuffer.elements[tileIndex] = 0u;
-}

src/glsl_shaders/common.glsl

@@ -1,14 +1,74 @@
 
 layout(std430) buffer;
 
-struct vertex {
-	vec4 cubic;
-	vec2 pos;
-	int shapeIndex;
+// command
+#define MG_GL_FILL 0
+#define MG_GL_STROKE 1
+
+// element kind
+#define MG_GL_LINE 1
+#define MG_GL_QUADRATIC 2
+#define MG_GL_CUBIC 3
+
+// curve config
+#define MG_GL_BL 1  /* curve on bottom left */
+#define MG_GL_BR 2 /* curve on bottom right */
+#define MG_GL_TL 3 /* curve on top left */
+#define	MG_GL_TR 4 /* curve on top right */
+
+// Operations
+#define MG_GL_OP_START 0
+#define MG_GL_OP_SEGMENT 1
+
+struct mg_gl_path
+{
+	int cmd;
+	float uvTransform[9];
+	vec4 color;
+	vec4 box;
+	vec4 clip;
 };
 
-struct shape {
-	vec4 color;
-	vec4 clip;
-	float uvTransform[6];
+struct mg_gl_path_elt
+{
+	int pathIndex;
+	int localEltIndex;
+	int kind;
+	vec2 p[4];
+};
+
+struct mg_gl_segment
+{
+	int kind;
+	int pathIndex;
+	int config; //TODO pack these
+	int windingIncrement;
+	vec4 box;
+	float hullMatrix[9];
+	float implicitMatrix[9];
+	float sign;
+	vec2 hullVertex;
+	int debugID;
+};
+
+struct mg_gl_path_queue
+{
+	vec4 area;
+	int tileQueues;
+};
+
+struct mg_gl_tile_op
+{
+	int kind;
+	int index;
+	int next;
+	bool crossRight;
+	int windingOffset;
+};
+
+struct mg_gl_tile_queue
+{
+	int windingOffset;
+	int first;
+	int last;
 };

src/glsl_shaders/draw.glsl

@@ -1,242 +0,0 @@
-
-#extension GL_ARB_gpu_shader_int64 : require
-layout(local_size_x = 16, local_size_y = 16, local_size_z = 1) in;
-
-precision mediump float;
-//precision mediump image2D;
-
-layout(binding = 0) restrict readonly buffer vertexBufferSSBO {
-	vertex elements[];
-} vertexBuffer ;
-
-layout(binding = 1) restrict readonly buffer shapeBufferSSBO {
-	shape elements[];
-} shapeBuffer ;
-
-layout(binding = 2) restrict readonly buffer indexBufferSSBO {
-	uint elements[];
-} indexBuffer ;
-
-layout(binding = 3) restrict readonly buffer tileCounterBufferSSBO {
-	uint elements[];
-} tileCounterBuffer ;
-
-layout(binding = 4) restrict readonly buffer tileArrayBufferSSBO {
-	uint elements[];
-} tileArrayBuffer ;
-
-layout(location = 0) uniform uint indexCount;
-layout(location = 1) uniform uvec2 tileCount;
-layout(location = 2) uniform uint tileSize;
-layout(location = 3) uniform uint tileArraySize;
-layout(location = 4) uniform vec2 scaling;
-layout(location = 5) uniform uint useTexture;
-
-layout(rgba8, binding = 0) uniform restrict writeonly image2D outTexture;
-
-layout(binding = 1) uniform sampler2D srcTexture;
-
-
-bool is_top_left(ivec2 a, ivec2 b)
-{
-	return( (a.y == b.y && b.x < a.x)
-	      ||(b.y < a.y));
-}
-
-//////////////////////////////////////////////////////////////////////////////
-//TODO: we should do these computations on 64bits, because otherwise
-//      we might overflow for values > 2048.
-//		Unfortunately this is costly.
-//	    Another way is to precompute triangle edges (b - a) in full precision
-//      once to avoid doing it all the time...
-//////////////////////////////////////////////////////////////////////////////
-int orient2d(ivec2 a, ivec2 b, ivec2 p)
-{
-	return((b.x-a.x)*(p.y-a.y) - (b.y-a.y)*(p.x-a.x));
-}
-
-int is_clockwise(ivec2 p0, ivec2 p1, ivec2 p2)
-{
-	return((p1 - p0).x*(p2 - p0).y - (p1 - p0).y*(p2 - p0).x);
-}
-
-void main()
-{
-	ivec2 pixelCoord = ivec2(gl_WorkGroupID.xy*uvec2(16, 16) + gl_LocalInvocationID.xy);
-	uvec2 tileCoord = uvec2(pixelCoord) / tileSize;
-	uint tileIndex =  tileCoord.y * tileCount.x + tileCoord.x;
-	uint tileCounter = min(tileCounterBuffer.elements[tileIndex], tileArraySize);
-
-	const float subPixelFactor = 16.;
-	ivec2 centerPoint = ivec2((vec2(pixelCoord) + vec2(0.5, 0.5)) * subPixelFactor);
-
-//*
-	const int sampleCount = 8;
-	ivec2 samplePoints[sampleCount] = ivec2[sampleCount](centerPoint + ivec2(1, 3),
-	                                                     centerPoint + ivec2(-1, -3),
-	                                                     centerPoint + ivec2(5, -1),
-	                                                     centerPoint + ivec2(-3, 5),
-	                                                     centerPoint + ivec2(-5, -5),
-	                                                     centerPoint + ivec2(-7, 1),
-	                                                     centerPoint + ivec2(3, -7),
-	                                                     centerPoint + ivec2(7, 7));
-/*/
-	const int sampleCount = 4;
-	ivec2 samplePoints[sampleCount] = ivec2[sampleCount](centerPoint + ivec2(-2, 6),
-	                                                     centerPoint + ivec2(6, 2),
-	                                                     centerPoint + ivec2(-6, -2),
-	                                                     centerPoint + ivec2(2, -6));
-//*/
-	//DEBUG
-/*
-	{
-		vec4 fragColor = vec4(0);
-
-		if( pixelCoord.x % 16 == 0
-	  	  ||pixelCoord.y % 16 == 0)
-		{
-			fragColor = vec4(0, 0, 0, 1);
-		}
-		else if(tileCounterBuffer.elements[tileIndex] == 0xffffu)
-		{
-			fragColor = vec4(1, 0, 1, 1);
-		}
-		else if(tileCounter != 0u)
-		{
-			fragColor = vec4(0, 1, 0, 1);
-		}
-		else
-		{
-			fragColor = vec4(1, 0, 0, 1);
-		}
-		imageStore(outTexture, pixelCoord, fragColor);
-		return;
-	}
-//*/
-	//----
-
-	vec4 sampleColor[sampleCount];
-	vec4 currentColor[sampleCount];
-    int currentShapeIndex[sampleCount];
-    int flipCount[sampleCount];
-
-    for(int i=0; i<sampleCount; i++)
-    {
-		currentShapeIndex[i] = -1;
-		flipCount[i] = 0;
-		sampleColor[i] = vec4(0, 0, 0, 0);
-		currentColor[i] = vec4(0, 0, 0, 0);
-    }
-
-    for(uint tileArrayIndex=0u; tileArrayIndex < tileCounter; tileArrayIndex++)
-    {
-    	uint triangleIndex = tileArrayBuffer.elements[tileArraySize * tileIndex + tileArrayIndex];
-
-		uint i0 = indexBuffer.elements[triangleIndex];
-		uint i1 = indexBuffer.elements[triangleIndex+1u];
-		uint i2 = indexBuffer.elements[triangleIndex+2u];
-
-		ivec2 p0 = ivec2((vertexBuffer.elements[i0].pos * scaling) * subPixelFactor);
-		ivec2 p1 = ivec2((vertexBuffer.elements[i1].pos * scaling) * subPixelFactor);
-		ivec2 p2 = ivec2((vertexBuffer.elements[i2].pos * scaling) * subPixelFactor);
-
-		int shapeIndex = vertexBuffer.elements[i0].shapeIndex;
-		vec4 color = shapeBuffer.elements[shapeIndex].color;
-		color.rgb *= color.a;
-
-		ivec4 clip = ivec4(round((shapeBuffer.elements[shapeIndex].clip * vec4(scaling, scaling) + vec4(0.5, 0.5, 0.5, 0.5)) * subPixelFactor));
-
-		mat3 uvTransform = mat3(shapeBuffer.elements[shapeIndex].uvTransform[0],
-		                        shapeBuffer.elements[shapeIndex].uvTransform[3],
-		                        0.,
-		                        shapeBuffer.elements[shapeIndex].uvTransform[1],
-		                        shapeBuffer.elements[shapeIndex].uvTransform[4],
-		                        0.,
-		                        shapeBuffer.elements[shapeIndex].uvTransform[2],
-		                        shapeBuffer.elements[shapeIndex].uvTransform[5],
-		                        1.);
-
-		//NOTE(martin): reorder triangle counter-clockwise and compute bias for each edge
-		int cw = is_clockwise(p0, p1, p2);
-		if(cw < 0)
-		{
-			uint tmpIndex = i1;
-			i1 = i2;
-			i2 = tmpIndex;
-
-			ivec2 tmpPoint = p1;
-			p1 = p2;
-			p2 = tmpPoint;
-		}
-
-		vec4 cubic0 = vertexBuffer.elements[i0].cubic;
-		vec4 cubic1 = vertexBuffer.elements[i1].cubic;
-		vec4 cubic2 = vertexBuffer.elements[i2].cubic;
-
-		int bias0 = is_top_left(p1, p2) ? 0 : -1;
-		int bias1 = is_top_left(p2, p0) ? 0 : -1;
-		int bias2 = is_top_left(p0, p1) ? 0 : -1;
-
-		for(int sampleIndex = 0; sampleIndex < sampleCount; sampleIndex++)
-		{
-			ivec2 samplePoint = samplePoints[sampleIndex];
-
-			if(  samplePoint.x < clip.x
-			  || samplePoint.x > clip.z
-			  || samplePoint.y < clip.y
-			  || samplePoint.y > clip.w)
-			{
-				continue;
-			}
-
-			int w0 = orient2d(p1, p2, samplePoint);
-			int w1 = orient2d(p2, p0, samplePoint);
-			int w2 = orient2d(p0, p1, samplePoint);
-
-			if((w0+bias0) >= 0 && (w1+bias1) >= 0 && (w2+bias2) >= 0)
-			{
-				vec4 cubic = (cubic0*float(w0) + cubic1*float(w1) + cubic2*float(w2))/(float(w0)+float(w1)+float(w2));
-
-				float eps = 0.0001;
-				if(cubic.w*(cubic.x*cubic.x*cubic.x - cubic.y*cubic.z) <= eps)
-				{
-					if(shapeIndex == currentShapeIndex[sampleIndex])
-					{
-						flipCount[sampleIndex]++;
-					}
-					else
-					{
-						if((flipCount[sampleIndex] & 0x01) != 0)
-						{
-							sampleColor[sampleIndex] = currentColor[sampleIndex];
-						}
-
-						vec4 nextColor = color;
-						if(useTexture)
-						{
-							vec3 sampleFP = vec3(vec2(samplePoint).xy/(subPixelFactor*2.), 1);
-							vec2 uv = (uvTransform * sampleFP).xy;
-							vec4 texColor = texture(srcTexture, uv);
-							texColor.rgb *= texColor.a;
-							nextColor *= texColor;
-						}
-						currentColor[sampleIndex] = sampleColor[sampleIndex]*(1.-nextColor.a) + nextColor;
-						currentShapeIndex[sampleIndex] = shapeIndex;
-						flipCount[sampleIndex] = 1;
-					}
-				}
-			}
-		}
-    }
-    vec4 pixelColor = vec4(0);
-    for(int sampleIndex = 0; sampleIndex < sampleCount; sampleIndex++)
-    {
-    	if((flipCount[sampleIndex] & 0x01) != 0)
-    	{
-			sampleColor[sampleIndex] = currentColor[sampleIndex];
-    	}
-    	pixelColor += sampleColor[sampleIndex];
-	}
-
-	imageStore(outTexture, pixelCoord, pixelColor/float(sampleCount));
-}

src/glsl_shaders/merge.glsl

@@ -0,0 +1,41 @@
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+
+precision mediump float;
+layout(std430) buffer;
+
+layout(binding = 0) restrict readonly buffer pathBufferSSBO
+{
+	mg_gl_path elements[];
+} pathBuffer;
+
+layout(binding = 1) restrict readonly buffer pathQueueBufferSSBO
+{
+	mg_gl_path_queue elements[];
+} pathQueueBuffer;
+
+layout(binding = 2) restrict readonly buffer tileQueueBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} tileQueueBuffer;
+
+layout(binding = 3) restrict readonly buffer tileOpBufferSSBO
+{
+	mg_gl_tile_op elements[];
+} tileOpBuffer;
+
+layout(binding = 4) restrict readonly buffer tileOpCountBufferSSBO
+{
+	int elements[];
+} tileOpCountBuffer;
+
+layout(binding = 5) restrict readonly buffer screenTilesBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} screenTilesBuffer;
+
+
+void main()
+{
+
+}

src/glsl_shaders/path_setup.glsl

@@ -0,0 +1,30 @@
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+
+precision mediump float;
+layout(std430) buffer;
+
+layout(binding = 0) restrict readonly buffer pathBufferSSBO
+{
+	mg_gl_path elements[];
+} pathBuffer;
+
+layout(binding = 1) restrict writeonly buffer pathQueueBufferSSBO
+{
+	mg_gl_path_queue elements[];
+} pathQueueBuffer;
+
+layout(binding = 2) restrict writeonly buffer tileQueueBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} tileQueueBuffer;
+
+layout(binding = 3) restrict writeonly buffer tileQueueCountBufferSSBO
+{
+	int elements[];
+} tileQueueCountBuffer;
+
+void main()
+{
+
+}

src/glsl_shaders/raster.glsl

@@ -0,0 +1,31 @@
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+
+precision mediump float;
+layout(std430) buffer;
+
+layout(binding = 0) restrict readonly buffer screenTilesBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} screenTilesBuffer;
+
+layout(binding = 1) restrict readonly buffer tileOpBufferSSBO
+{
+	mg_gl_tile_op elements[];
+} tileOpBuffer;
+
+layout(binding = 2) restrict readonly buffer pathBufferSSBO
+{
+	mg_gl_path elements[];
+} pathBuffer;
+
+layout(binding = 3) restrict readonly buffer segmentBufferSSBO
+{
+	mg_gl_segment elements[];
+} segmentBuffer;
+
+
+void main()
+{
+
+}

src/glsl_shaders/segment_setup.glsl

@@ -0,0 +1,46 @@
+
+layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+
+precision mediump float;
+layout(std430) buffer;
+
+layout(binding = 0) restrict readonly buffer elementBufferSSBO
+{
+	mg_gl_path_elt elements[];
+} elementBuffer;
+
+layout(binding = 1) restrict buffer segmentCountBufferSSBO
+{
+	int elements[];
+} segmentCountBuffer;
+
+layout(binding = 2) restrict writeonly buffer segmentBufferSSBO
+{
+	mg_gl_segment elements[];
+} segmentBuffer;
+
+layout(binding = 3) restrict readonly buffer pathQueueBufferSSBO
+{
+	mg_gl_path_queue elements[];
+} pathQueueBuffer;
+
+layout(binding = 4) restrict readonly buffer tileQueueBufferSSBO
+{
+	mg_gl_tile_queue elements[];
+} tileQueueBuffer;
+
+layout(binding = 5) restrict readonly buffer tileOpBufferSSBO
+{
+	mg_gl_tile_op elements[];
+} tileOpBuffer;
+
+layout(binding = 6) restrict readonly buffer tileOpCountBufferSSBO
+{
+	int elements[];
+} tileOpCountBuffer;
+
+
+void main()
+{
+
+}

src/glsl_shaders/sort.glsl

@@ -1,61 +0,0 @@
-
-layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
-
-precision mediump float;
-
-layout(binding = 0) restrict readonly buffer vertexBufferSSBO {
-	vertex elements[];
-} vertexBuffer ;
-
-layout(binding = 1) restrict readonly buffer shapeBufferSSBO {
-	shape elements[];
-} shapeBuffer ;
-
-layout(binding = 2) restrict readonly buffer indexBufferSSBO {
-	uint elements[];
-} indexBuffer ;
-
-layout(binding = 3) coherent readonly restrict buffer tileCounterBufferSSBO {
-	uint elements[];
-} tileCounterBuffer ;
-
-layout(binding = 4) coherent restrict buffer tileArrayBufferSSBO {
-	uint elements[];
-} tileArrayBuffer ;
-
-layout(location = 0) uniform uint indexCount;
-layout(location = 1) uniform uvec2 tileCount;
-layout(location = 2) uniform uint tileSize;
-layout(location = 3) uniform uint tileArraySize;
-
-int get_shape_index(uint tileArrayOffset, uint tileArrayIndex)
-{
-	uint triangleIndex = tileArrayBuffer.elements[tileArrayOffset + tileArrayIndex];
-	uint i0 = indexBuffer.elements[triangleIndex];
-	int shapeIndex = vertexBuffer.elements[i0].shapeIndex;
-	return(shapeIndex);
-}
-
-void main()
-{
-	uint tileIndex = gl_WorkGroupID.x;
-	uint tileArrayOffset = tileArraySize * tileIndex;
-	uint tileArrayCount = min(tileCounterBuffer.elements[tileIndex], tileArraySize);
-
-	for(uint tileArrayIndex=1u; tileArrayIndex < tileArrayCount; tileArrayIndex++)
-	{
-		for(uint sortIndex = tileArrayIndex; sortIndex > 0u; sortIndex--)
-		{
-			int shapeIndex = get_shape_index(tileArrayOffset, sortIndex);
-			int prevShapeIndex = get_shape_index(tileArrayOffset, sortIndex-1u);
-
-			if(shapeIndex >= prevShapeIndex)
-			{
-				break;
-			}
-			uint tmp = tileArrayBuffer.elements[tileArrayOffset + sortIndex];
-			tileArrayBuffer.elements[tileArrayOffset + sortIndex] = tileArrayBuffer.elements[tileArrayOffset + sortIndex - 1u];
-			tileArrayBuffer.elements[tileArrayOffset + sortIndex - 1u] = tmp;
-		}
-	}
-}

src/glsl_shaders/tile.glsl

@@ -1,77 +0,0 @@
-
-layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
-
-precision mediump float;
-
-layout(binding = 0) restrict readonly buffer vertexBufferSSBO {
-	vertex elements[];
-} vertexBuffer ;
-
-layout(binding = 1) restrict readonly buffer shapeBufferSSBO {
-	shape elements[];
-} shapeBuffer ;
-
-layout(binding = 2) restrict readonly buffer indexBufferSSBO {
-	uint elements[];
-} indexBuffer ;
-
-layout(binding = 3) coherent restrict buffer tileCounterBufferSSBO {
-	uint elements[];
-} tileCounterBuffer ;
-
-layout(binding = 4) coherent restrict writeonly buffer tileArrayBufferSSBO {
-	uint elements[];
-} tileArrayBuffer ;
-
-layout(location = 0) uniform uint indexCount;
-layout(location = 1) uniform uvec2 tileCount;
-layout(location = 2) uniform uint tileSize;
-layout(location = 3) uniform uint tileArraySize;
-layout(location = 4) uniform vec2 scaling;
-
-void main()
-{
-	uint triangleIndex = (gl_WorkGroupID.x*gl_WorkGroupSize.x + gl_LocalInvocationIndex) * 3u;
-	if(triangleIndex >= indexCount)
-	{
-		return;
-	}
-
-	uint i0 = indexBuffer.elements[triangleIndex];
-	uint i1 = indexBuffer.elements[triangleIndex+1u];
-	uint i2 = indexBuffer.elements[triangleIndex+2u];
-
-	vec2 p0 = vertexBuffer.elements[i0].pos * scaling;
-	vec2 p1 = vertexBuffer.elements[i1].pos * scaling;
-	vec2 p2 = vertexBuffer.elements[i2].pos * scaling;
-
-	int shapeIndex = vertexBuffer.elements[i0].shapeIndex;
-	vec4 clip = shapeBuffer.elements[shapeIndex].clip * vec4(scaling, scaling);
-
-	vec4 fbox = vec4(max(min(min(p0.x, p1.x), p2.x), clip.x),
-		             max(min(min(p0.y, p1.y), p2.y), clip.y),
-		             min(max(max(p0.x, p1.x), p2.x), clip.z),
-		             min(max(max(p0.y, p1.y), p2.y), clip.w));
-
-	ivec4 box = ivec4(floor(fbox))/int(tileSize);
-
-	//NOTE(martin): it's importat to do the computation with signed int, so that we can have negative xMax/yMax
-	//              otherwise all triangles on the left or below the x/y axis are attributed to tiles on row/column 0.
-	int xMin = max(0, box.x);
-	int yMin = max(0, box.y);
-	int xMax = min(box.z, int(tileCount.x) - 1);
-	int yMax = min(box.w, int(tileCount.y) - 1);
-
-	for(int y = yMin; y <= yMax; y++)
-	{
-		for(int x = xMin ; x <= xMax; x++)
-		{
-			uint tileIndex = uint(y)*tileCount.x + uint(x);
-			uint tileCounter = atomicAdd(tileCounterBuffer.elements[tileIndex], 1u);
-			if(tileCounter < tileArraySize)
-			{
-				tileArrayBuffer.elements[tileArraySize*tileIndex + tileCounter] = triangleIndex;
-			}
-		}
-	}
-}