quick commit

shaders/A2Task1KernelDecomposition.comp

@@ -9,14 +9,18 @@ in uint  gl_LocalInvocationIndex;
 */
 layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 
-layout(push_constant) uniform PushStruct {
-    uint size;
-} p;
+layout(push_constant) uniform PushStruct
+{
+    uint offset;
+}
+p;
 
-layout(binding = 0) buffer inBuffer {
+layout(binding = 0) buffer inBuffer
+{
     uint v[];
 };
-layout(binding = 1) buffer outBuffer {
+layout(binding = 1) buffer outBuffer
+{
     uint g_v[];
 };
 
@@ -25,22 +29,37 @@ layout(binding = 1) buffer outBuffer {
 const uint bufferSize = 256;
 shared uint[bufferSize] localBuffer;
 
-void main() {
-    // TODO: Kernel implementation
+void main()
+{
+    uint tid    = gl_LocalInvocationID.x;
+    uint gid    = gl_WorkGroupID.x;
+    uint offset = gid * bufferSize;
 
-    for (uint i = p.size / 2; i < 0; i -= 2) {
-        localBuffer[i] = v[i] + v[i + 1];
-    }
+    uint idx1 = offset + tid;
+    uint idx2 = offset + tid + gl_WorkGroupSize.x;
 
-    for (uint j = bufferSize ; j != 0; j / 2) {
-        for (uint i = bufferSize / 2; i < 0; i -= 2) {
-            localBuffer[i] = localBuffer[i] + localBuffer[i + 1];
+    uint val1 = 0;
+    uint val2 = 0;
+
+    if (idx1 < p.offset)
+        val1 = v[idx1];
+    if (idx2 < p.offset)
+        val2 = v[idx2];
+
+    localBuffer[tid] = val1 + val2;
+
+    // Reduction in shared memory
+    for (uint s = gl_WorkGroupSize.x / 2; s > 0; s /= 2)
+    {
+        if (tid < s)
+        {
+            barrier();
+            localBuffer[tid] += localBuffer[tid + s];
         }
     }
 
-    localBuffer[0] = localBuffer[0] + localBuffer[1];
-
-    for (uint i = 0; i < bufferSize; i ++) {
-        g_v[i] = localBuffer[i];
+    if (tid == 0)
+    {
+        g_v[gid] = localBuffer[tid];
     }
-}
+}