address comments and add test

triton-lang · Nov 7, 2024 · 6e49d28 · 6e49d28
1 parent f1b904c
commit 6e49d28
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 1 deletion.
diff --git a/test/TritonGPU/amd/optimize-lds-usage.mlir b/test/TritonGPU/amd/optimize-lds-usage.mlir
@@ -119,6 +119,8 @@ module attributes {"triton_gpu.num-warps" = 8 : i32, "triton_gpu.threads-per-war
   }
 }
 
+// -----
+
 // Checks that optimization do not crash on 1d tensor
 // CHECK-LABEL: convert_1d
 // CHECK: triton_gpu.local_alloc

diff --git a/third_party/amd/lib/TritonAMDGPUToLLVM/OptimizeLDSUtility.cpp b/third_party/amd/lib/TritonAMDGPUToLLVM/OptimizeLDSUtility.cpp
@@ -69,7 +69,8 @@ Attribute createTmpLayout(Attribute layout, ArrayRef<unsigned> warpsPerCTA) {
         src.getKWidth());
   }
   if (auto src = dyn_cast<triton::gpu::SliceEncodingAttr>(layout)) {
-    // TODO: think of a way to construct slice layouts based on warpsPerCTA argument
+    // TODO: think of a way to construct slice layouts based on warpsPerCTA
+    // argument
     auto parentWarpsPerCTA = triton::gpu::getWarpsPerCTA(src.getParent());
     return triton::gpu::SliceEncodingAttr::get(
         ctx, src.getDim(), createTmpLayout(src.getParent(), parentWarpsPerCTA));