Parallelised compute shader in example 48 (#2906)

* Parallelised compute shader in example 48

* Ex48 - Dispatch numToDraw/64 workgroups of 64 local threads

* fixed vs build (again)

examples/48-drawindirect/cs_drawindirect.sc

@@ -14,19 +14,26 @@ BUFFER_WR(instanceBufferOut, vec4, 2);
 
 uniform vec4 u_drawParams;
 
-NUM_THREADS(1, 1, 1)
+// Use 64*1*1 local threads
+NUM_THREADS(64, 1, 1)
 
 void main()
 {
+	int tId = int(gl_GlobalInvocationID.x);
 	int numDrawItems = int(u_drawParams.x);
 	int sideSize = int(u_drawParams.y);
 	float time = u_drawParams.z;
 	
-	// Prepare draw mtx
-	
+	// Work out the amount of work we're going to do here
 	int maxToDraw = min(sideSize*sideSize, numDrawItems);
 	
-	for (int k = 0; k < maxToDraw; k++) {
+	int numToDrawPerThread = maxToDraw/64 + 1;
+	
+	int idxStart = tId*numToDrawPerThread;
+	int idxMax = min(maxToDraw, (tId+1)*numToDrawPerThread);
+	
+	// Prepare draw mtx
+	for (int k = idxStart; k < idxMax; k++) {
 		int yy = k / sideSize;
 		int xx = k % sideSize;
 
@@ -58,7 +65,7 @@ void main()
 	
 	// Fill indirect buffer
 	
-	for (int k = 0; k < maxToDraw; k++) {
+	for (int k = idxStart; k < idxMax; k++) {
 		drawIndexedIndirect(
 						// Target location params:
 			indirectBuffer,			// target buffer

examples/48-drawindirect/drawindirect.cpp

@@ -344,26 +344,30 @@ public:
 				// The model matrix for each instance is also set on compute
 				// you could modify this to, eg, do frustrum culling on the GPU		
 				float ud[4] = { float(m_nDrawElements), float(m_sideSize), float(time), 0 };
+				uint32_t numToDraw = (m_sideSize*m_sideSize);
+				
 				bgfx::setUniform(u_drawParams, ud);
 						
 				bgfx::setBuffer(0, m_object_list_buffer, bgfx::Access::Read);
 				bgfx::setBuffer(1, m_indirect_buffer_handle, bgfx::Access::Write);
 				bgfx::setBuffer(2, m_instance_buffer, bgfx::Access::Write);
 				
-				bgfx::dispatch(0, m_indirect_program);
+				// Dispatch the call. We are using 64 local threads on the GPU to process the object list
+				// So lets dispatch ceil(numToDraw/64) workgroups of 64 local threads
+				bgfx::dispatch(0, m_indirect_program, uint32_t(numToDraw/64 + 1), 1, 1);
 				
 				// Submit our 1 draw call
 				// Set vertex and index buffer.
 				bgfx::setIndexBuffer(m_ibh);
 				bgfx::setVertexBuffer(0, m_vbh);
-				bgfx::setInstanceDataBuffer(m_instance_buffer, 0, m_sideSize*m_sideSize);
+				bgfx::setInstanceDataBuffer(m_instance_buffer, 0, numToDraw);
 				
 				// Set render states.
 				bgfx::setState(BGFX_STATE_DEFAULT);
 
 				// Submit primitive for rendering to view 0.
 				// note that this submission requires the draw count
-				bgfx::submit(0, m_program, m_indirect_buffer_handle, 0, uint16_t(m_sideSize*m_sideSize));
+				bgfx::submit(0, m_program, m_indirect_buffer_handle, 0, uint16_t(numToDraw));
 			}
 			else
 			{