more debug logging

Author: alexbaden

third_party/intel/lib/TritonIntelGPUTransforms/AccelerateMatmul.cpp

@@ -64,6 +64,7 @@ getWarpsPerTile(tt::DotOp dotOp,
       ceil<uint32_t>(dpasCap.repeatCount, dpasCap.executionSize);
   uint32_t colRowRatio =
       ceil<uint32_t>(dpasCap.executionSize, dpasCap.repeatCount);
+  llvm::errs() << "rowColRation: " << rowColRatio << ", colRowRatio: " << colRowRatio << ", ret: " << ret[0] << ", " << ret[1] << "\n";
 
   int rowDim = order[rank - 2], colDim = order[rank - 1];
   do {
@@ -141,7 +142,7 @@ class BlockedToDPAS : public OpRewritePattern<tt::DotOp> {
     } else {
       // llvm::errs() << "no A op for A: " << a << "\n";
     }
-    // llvm::errs() << "order: " << order[0] << ", " << order[1] << "\n";
+    llvm::errs() << "order: " << order[0] << ", " << order[1] << "\n";
 
     SmallVector<unsigned> warpsPerTile =
         getWarpsPerTile(dotOp, dpasCap, retShape, numWarps, order);