format + remove debug prints

Author: alexbaden

third_party/intel/backend/compiler.py

@@ -250,7 +250,6 @@ def make_ttgir(mod, metadata, opt, properties):
             intel.passes.ttgpuir.add_rewrite_tensor_pointer(pm)
         intel.passes.ttgpuir.add_pipeline(pm, opt.num_stages, False)
 
-    
         passes.ttgpuir.add_optimize_thread_locality(pm)
         passes.ttgpuir.add_optimize_dot_operands(pm, True)
         passes.common.add_cse(pm)

third_party/intel/lib/TritonIntelGPUTransforms/AccelerateMatmul.cpp

@@ -123,31 +123,20 @@ class BlockedToDPAS : public OpRewritePattern<tt::DotOp> {
         oldAType.getElementType().isFloat8E4M3FN())
       dpasElemBitWidths = 2 * dpasElemBitWidths;
 
-    // now we can get the order from the a defining op 
-
-    llvm::errs() << "oldAType: " << oldAType << "\n";
-    llvm::errs() << "oldBType: " << oldBType << "\n";
-
-    llvm::errs() << "a: " << a << "\n";
-    llvm::errs() << "a defining op: " << *a.getDefiningOp() << "\n";
-
     SmallVector<unsigned> order;
-    Operation* aOp = a.getDefiningOp();
+    Operation *aOp = a.getDefiningOp();
     if (isa<ttg::ConvertLayoutOp>(aOp)) {
       assert(aOp->getNumOperands() == 1);
       auto aLoad = aOp->getOperand(0);
-      order = triton::gpu::getOrder(cast<RankedTensorType>(aLoad.getType()).getEncoding());
+      order = triton::gpu::getOrder(
+          cast<RankedTensorType>(aLoad.getType()).getEncoding());
     } else {
       assert(isa<tt::LoadOp>(aOp) && "expecting load input to DPAS");
       assert(aOp->getNumResults() == 1);
       auto ret = aOp->getResult(0);
-      order = triton::gpu::getOrder(cast<RankedTensorType>(ret.getType()).getEncoding());
+      order = triton::gpu::getOrder(
+          cast<RankedTensorType>(ret.getType()).getEncoding());
     }
-    llvm::errs() << "a load order: " << order[0] << ", " << order[1] << "\n";
-    #if 0
-    const bool aIsTransposed = order.size() == 2 && order[0] == 0 && order[1] == 1;
-    llvm::errs() << "Transposed? " << aIsTransposed << "\n";
-    #endif
 
     SmallVector<unsigned> warpsPerTile =
         getWarpsPerTile(dotOp, dpasCap, retShape, numWarps, order);