optimize dispatch calls

Shaders/Compute.glsl

@@ -2,6 +2,8 @@
 
 layout(rgba32f, binding = 0) uniform writeonly image2D outputImage;
 
+layout(location = 0) uniform uint triangleCount;
+
 layout(std430, binding = 3) buffer layoutName
 {
   vec2 o_Points[];
@@ -26,16 +28,19 @@ layout(std430, binding = 3) buffer layoutName
 // gl_WorkGroupID : This is the current work group for this shader invocation. Each of the XYZ components will be on the half-open range [0, gl_NumWorkGroups.XYZ).
 // gl_LocalInvocationID : This is the current invocation of the shader within the work group. Each of the XYZ components will be on the half-open range [0, gl_WorkGroupSize.XYZ).
 
-layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
+// avant : 1 3 9
+// après : 3 9 27
+// count : 1 3 9
+layout(local_size_x = 64) in;
 void main()
 {
-    uint triangleCount = gl_NumWorkGroups.x;
-    uint currentPoint = gl_WorkGroupID.x;
+    uint currentPoint = gl_GlobalInvocationID.x;
+    if (currentPoint >= triangleCount)
+      return;
 
     uint offset = (triangleCount - 1) / 2;
-
     uint firstPointIndex = offset * 3 + 1;
-    o_Points[firstPointIndex + currentPoint * 3 ] = o_Points[offset + currentPoint] / 2 + vec2(0, 0.36);
+    o_Points[firstPointIndex + currentPoint * 3] = o_Points[offset + currentPoint] / 2 + vec2(0, 0.36);
     o_Points[firstPointIndex + currentPoint * 3 + 1] = o_Points[offset + currentPoint] / 2 + vec2(-0.5, -0.5);
     o_Points[firstPointIndex + currentPoint * 3 + 2] = o_Points[offset + currentPoint] / 2 + vec2(0.5, -0.5);
 }

Shaders/Vertex.glsl

@@ -1,14 +1,10 @@
 #version 460 core
 
-layout(location = 0) in vec2 a_Position;
-
 layout(std430, binding = 3) buffer layoutName
 {
     vec2 data_SSBO[];
 };
 
-out vec2 v_TexCoord;
-
 void main()
 {
 	gl_Position = vec4(data_SSBO[gl_InstanceID], 0.0, 1.0);

src/Main.cpp

@@ -139,7 +139,7 @@ int main()
 	GLuint vertexBuffer;
 	glCreateBuffers(1, &vertexBuffer);
 
-	constexpr int DEPTH = 13;
+	constexpr int DEPTH = 14;
 
 	auto vertices = GetPoints(DEPTH);
 
@@ -155,6 +155,8 @@ int main()
 	glVertexArrayAttribBinding(vertexArray, 0, 0);
 	glVertexArrayAttribBinding(vertexArray, 1, 0);
 
+	Timer timer;
+
 	GLuint ssbo;
 	glGenBuffers(1, &ssbo);
 	glBindBuffer(GL_SHADER_STORAGE_BUFFER, ssbo);
@@ -173,7 +175,13 @@ int main()
 	std::size_t count = 1;
 	for (std::size_t i = 0; i < DEPTH; i++)
 	{
-		glDispatchCompute(count, 1, 1);
+		glUniform1ui(0, count);
+		std::cout << count << std::endl;
+
+		int dispatchCount = (count - 1) / 64 + 1;
+		std::cout << "DC : " <<dispatchCount<< std::endl;
+
+		glDispatchCompute(dispatchCount, 1, 1);
 
 		// Ensure all writes to the image are complete
 		glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
@@ -181,6 +189,8 @@ int main()
 		count *= 3;
 	}
 
+	std::cout << "Generated points in " << timer.ElapsedMillis() << " ms\n";
+
 	while (!glfwWindowShouldClose(window))
 	{
 		// ScopedTimer timer("Main Loop");