zeropad_U : loop reordering for performance optimization

leela-zero · Aug 20, 2018 · a3bd3f0 · a3bd3f0
1 parent 0fd4637
commit a3bd3f0
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/src/OpenCLScheduler.cpp b/src/OpenCLScheduler.cpp
@@ -34,10 +34,10 @@ static std::vector<float> zeropad_U(const std::vector<float>& U,
     auto Upad =
         std::vector<float>(WINOGRAD_TILE * outputs_pad * channels_pad);
 
-    for (auto o = 0; o < outputs; o++) {
-        for (auto c = 0; c < channels; c++) {
-            for (auto xi = 0; xi < WINOGRAD_ALPHA; xi++){
-                for (auto nu = 0; nu < WINOGRAD_ALPHA; nu++) {
+    for (auto xi = 0; xi < WINOGRAD_ALPHA; xi++){
+        for (auto nu = 0; nu < WINOGRAD_ALPHA; nu++) {
+            for (auto c = 0; c < channels; c++) {
+                for (auto o = 0; o < outputs; o++) {
                     Upad[xi * (WINOGRAD_ALPHA * outputs_pad * channels_pad)
                          + nu * (outputs_pad * channels_pad)
                          + c * outputs_pad +