diff --git a/examples/polygon/main.c b/examples/polygon/main.c
index 7912cca..f3ea8d7 100644
--- a/examples/polygon/main.c
+++ b/examples/polygon/main.c
@@ -16,8 +16,6 @@
 
 #include"milepost.h"
 
-#define LOG_SUBSYSTEM "Main"
-
 int main()
 {
 	mp_init();
@@ -111,7 +109,6 @@ int main()
 			mg_set_color_rgba(1, 0, 0, 1);
 			mg_fill();
 
-/*
 			mg_move_to(200, 100);
 			mg_line_to(410, 100);
 			mg_line_to(410, 200);
@@ -119,7 +116,7 @@ int main()
 			mg_close_path();
 			mg_set_color_rgba(0, 1, 0, 1);
 			mg_fill();
-
+/*
 			mg_move_to(400, 400);
 			mg_quadratic_to(600, 601, 800, 400);
 			mg_close_path();
diff --git a/src/gl_canvas.c b/src/gl_canvas.c
index bb0c597..f47cd60 100644
--- a/src/gl_canvas.c
+++ b/src/gl_canvas.c
@@ -273,16 +273,13 @@ void mg_gl_render_batch(mg_gl_canvas_backend* backend,
 	glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 
 	//NOTE: backprop pass
-/*
 	glUseProgram(backend->backprop);
 
-	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, backend->tileQueueBuffer);
+	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, backend->pathQueueBuffer);
+	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 1, backend->tileQueueBuffer);
 
-	glUniform2i(0, nTilesX, nTilesY);
-
-	glDispatchCompute(nTilesY, 1, 1);
+	glDispatchCompute(pathCount, 1, 1);
 	glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
-*/
 
 	//NOTE: merge pass
 	glUseProgram(backend->merge);
diff --git a/src/glsl_shaders/backprop.glsl b/src/glsl_shaders/backprop.glsl
index 46e5c46..6cb4741 100644
--- a/src/glsl_shaders/backprop.glsl
+++ b/src/glsl_shaders/backprop.glsl
@@ -1,26 +1,49 @@
 
-layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+layout(local_size_x = 16, local_size_y = 1, local_size_z = 1) in;
 
 precision mediump float;
 layout(std430) buffer;
 
-layout(binding = 0) restrict buffer tileQueueBufferSSBO
+layout(binding = 0) restrict readonly buffer pathQueueBufferSSBO
+{
+	mg_gl_path_queue elements[];
+} pathQueueBuffer;
+
+layout(binding = 1) restrict buffer tileQueueBufferSSBO
 {
 	mg_gl_tile_queue elements[];
 } tileQueueBuffer;
 
-layout(location = 0) uniform ivec2 nTiles;
+shared int nextRowIndex;
 
 void main()
 {
-	int rowIndex = int(gl_WorkGroupID.x);
+	int pathIndex = int(gl_WorkGroupID.x);
+	int localID = int(gl_LocalInvocationID.x);
 
-	int sum = 0;
-	for(int x = nTiles.x-1; x >= 0; x--)
+	if(localID == 0)
 	{
-		int tileIndex = rowIndex * nTiles.x + x;
-		int offset = tileQueueBuffer.elements[tileIndex].windingOffset;
-		tileQueueBuffer.elements[tileIndex].windingOffset = sum;
-		sum += offset;
+		nextRowIndex = 0;
+	}
+	barrier();
+
+	int rowIndex = 0;
+	mg_gl_path_queue pathQueue = pathQueueBuffer.elements[pathIndex];
+	int tileQueueBase = pathQueue.tileQueues;
+	int rowSize = pathQueue.area.z;
+	int rowCount = pathQueue.area.w;
+
+	rowIndex = atomicAdd(nextRowIndex, 1);
+	while(rowIndex < rowCount)
+	{
+		int sum = 0;
+		for(int x = rowSize-1; x >= 0; x--)
+		{
+			int tileIndex = tileQueueBase + rowIndex * rowSize.x + x;
+			int offset = tileQueueBuffer.elements[tileIndex].windingOffset;
+			tileQueueBuffer.elements[tileIndex].windingOffset = sum;
+			sum += offset;
+		}
+		rowIndex = atomicAdd(nextRowIndex, 1);
 	}
 }
diff --git a/src/glsl_shaders/merge.glsl b/src/glsl_shaders/merge.glsl
index 67de3e1..ea1fdda 100644
--- a/src/glsl_shaders/merge.glsl
+++ b/src/glsl_shaders/merge.glsl
@@ -45,7 +45,6 @@ void main()
 	int tileIndex = tileCoord.y * nTiles.x + tileCoord.x;
 
 	screenTilesBuffer.elements[tileIndex] = -1;
-
 	int lastOpIndex = -1;
 
 	for(int pathIndex = 0; pathIndex < pathCount; pathIndex++)
diff --git a/src/glsl_shaders/raster.glsl b/src/glsl_shaders/raster.glsl
index 3a31cf7..d6f5bb0 100644
--- a/src/glsl_shaders/raster.glsl
+++ b/src/glsl_shaders/raster.glsl
@@ -150,8 +150,8 @@ void main()
 
 	while(opIndex >= 0)
 	{
-		//imageStore(outTexture, ivec2(sampleCoord), vec4(0, 1, 0, 1));
-		//return;
+//		imageStore(outTexture, ivec2(sampleCoord), vec4(0, 1, 1, 1));
+//		return;
 
 		mg_gl_tile_op op = tileOpBuffer.elements[opIndex];
 		opIndex = op.next;