[BACKEND] Update LLVM version to llvm/llvm-project@91c5de7

makslevental · makslevental · commit 8eb000b26485 · 2025-01-06T12:33:48.000-05:00
diff --git a/cmake/llvm-hash.txt b/cmake/llvm-hash.txt
@@ -1 +1 @@
-86b69c31642e98f8357df62c09d118ad1da4e16a
+91c5de7fb8f95132043ed08056e58238383cfcb9
diff --git a/lib/Dialect/Triton/Transforms/Combine.cpp b/lib/Dialect/Triton/Transforms/Combine.cpp
@@ -204,7 +204,7 @@ class CombineOpsPass : public TritonCombineOpsBase<CombineOpsPass> {
     patterns.add<CombineAddPtrPattern>(context);
     patterns.add<CombineBroadcastMulReducePattern>(context);
 
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())
+    if (applyPatternsGreedily(m, std::move(patterns)).failed())
       signalPassFailure();
   }
 };
diff --git a/lib/Dialect/Triton/Transforms/ReorderBroadcast.cpp b/lib/Dialect/Triton/Transforms/ReorderBroadcast.cpp
@@ -221,7 +221,7 @@ class ReorderBroadcastPass
     // elementwise(splat(a), splat(b), ...) => splat(elementwise(a, b, ...))
     patterns.add<MoveSplatAfterElementwisePattern>(context);
 
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())
+    if (applyPatternsGreedily(m, std::move(patterns)).failed())
       signalPassFailure();
   }
 };
diff --git a/lib/Dialect/TritonGPU/Transforms/AccelerateMatmul.cpp b/lib/Dialect/TritonGPU/Transforms/AccelerateMatmul.cpp
@@ -797,7 +797,7 @@ class TritonGPUAccelerateMatmulPass
     mlir::RewritePatternSet patterns(context);
     patterns.add<BlockedToMMA, DecomposeScaledBlocked>(context,
                                                        computeCapability);
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) {
+    if (applyPatternsGreedily(m, std::move(patterns)).failed()) {
       signalPassFailure();
     }
     // Now that we have picked the mma type, decompose dot that are not natively
diff --git a/lib/Dialect/TritonGPU/Transforms/CoalesceAsyncCopy.cpp b/lib/Dialect/TritonGPU/Transforms/CoalesceAsyncCopy.cpp
@@ -114,7 +114,7 @@ class CoalesceAsyncCopyPass
     mlir::RewritePatternSet patterns(context);
     patterns.add<ClipAsyncCopySizePerThread>(context);
 
-    if (failed(applyPatternsAndFoldGreedily(m, std::move(patterns))))
+    if (failed(applyPatternsGreedily(m, std::move(patterns))))
       signalPassFailure();
   }
 };
diff --git a/lib/Dialect/TritonGPU/Transforms/F32DotTC.cpp b/lib/Dialect/TritonGPU/Transforms/F32DotTC.cpp
@@ -109,8 +109,7 @@ struct F32DotTCPass : public impl::TritonGPUF32DotTCBase<F32DotTCPass> {
 
     RewritePatternSet decomposePatterns(context);
     decomposePatterns.add<TF32x3>(context);
-    if (applyPatternsAndFoldGreedily(m, std::move(decomposePatterns))
-            .failed()) {
+    if (applyPatternsGreedily(m, std::move(decomposePatterns)).failed()) {
       signalPassFailure();
     }
   }
diff --git a/lib/Dialect/TritonGPU/Transforms/OptimizeDotOperands.cpp b/lib/Dialect/TritonGPU/Transforms/OptimizeDotOperands.cpp
@@ -584,7 +584,7 @@ class TritonGPUOptimizeDotOperandsPass
     patterns.add<FuseTransHopper>(context);
     patterns.add<MMAV3UseRegOperand>(context);
     ConvertLayoutOp::getCanonicalizationPatterns(patterns, context);
-    if (failed(applyPatternsAndFoldGreedily(m, std::move(patterns))))
+    if (failed(applyPatternsGreedily(m, std::move(patterns))))
       signalPassFailure();
   }
 };
diff --git a/lib/Dialect/TritonGPU/Transforms/OptimizeThreadLocality.cpp b/lib/Dialect/TritonGPU/Transforms/OptimizeThreadLocality.cpp
@@ -256,7 +256,7 @@ class TritonGPUOptimizeThreadLocalityPass
     mlir::RewritePatternSet layoutPatterns(&getContext());
     layoutPatterns.add<OptimizeReshapeLayoutPattern>(&getContext());
     layoutPatterns.add<OptimizeGatherLayoutPattern>(&getContext());
-    if (mlir::applyPatternsAndFoldGreedily(mod, std::move(layoutPatterns))
+    if (mlir::applyPatternsGreedily(mod, std::move(layoutPatterns))
             .failed()) {
       signalPassFailure();
     }
diff --git a/lib/Dialect/TritonGPU/Transforms/Pipeliner/SoftwarePipeliner.cpp b/lib/Dialect/TritonGPU/Transforms/Pipeliner/SoftwarePipeliner.cpp
@@ -110,7 +110,7 @@ struct PipelinePass : public impl::TritonGPUPipelineBase<PipelinePass> {
         getOperation().getContext()->getLoadedDialect<arith::ArithDialect>();
     RewritePatternSet patterns(getOperation().getContext());
     arithDialect->getCanonicalizationPatterns(patterns);
-    if (applyPatternsAndFoldGreedily(getOperation(), std::move(patterns))
+    if (applyPatternsGreedily(getOperation(), std::move(patterns))
             .failed())
       return signalPassFailure();
 
diff --git a/lib/Dialect/TritonGPU/Transforms/Prefetch.cpp b/lib/Dialect/TritonGPU/Transforms/Prefetch.cpp
@@ -422,7 +422,7 @@ struct PrefetchPass : public impl::TritonGPUPrefetchBase<PrefetchPass> {
     RewritePatternSet cleanUpPatterns(&getContext());
     triton::gpu::ConvertLayoutOp::getCanonicalizationPatterns(cleanUpPatterns,
                                                               &getContext());
-    if (mlir::applyPatternsAndFoldGreedily(getOperation(),
+    if (mlir::applyPatternsGreedily(getOperation(),
                                            std::move(cleanUpPatterns))
             .failed()) {
       signalPassFailure();
diff --git a/lib/Dialect/TritonGPU/Transforms/RemoveLayoutConversions.cpp b/lib/Dialect/TritonGPU/Transforms/RemoveLayoutConversions.cpp
@@ -1198,7 +1198,7 @@ class TritonGPURemoveLayoutConversionsPass
 
     RewritePatternSet cleanUpPatterns(context);
     ConvertLayoutOp::getCanonicalizationPatterns(cleanUpPatterns, context);
-    if (applyPatternsAndFoldGreedily(m, std::move(cleanUpPatterns)).failed()) {
+    if (applyPatternsGreedily(m, std::move(cleanUpPatterns)).failed()) {
       signalPassFailure();
     }
 
@@ -1230,7 +1230,7 @@ class TritonGPURemoveLayoutConversionsPass
     scf::ForOp::getCanonicalizationPatterns(cleanUpPatterns2, context);
     scf::IfOp::getCanonicalizationPatterns(cleanUpPatterns2, context);
     ConvertLayoutOp::getCanonicalizationPatterns(cleanUpPatterns2, context);
-    if (applyPatternsAndFoldGreedily(m, std::move(cleanUpPatterns2)).failed()) {
+    if (applyPatternsGreedily(m, std::move(cleanUpPatterns2)).failed()) {
       signalPassFailure();
     }
     LLVM_DEBUG({
diff --git a/lib/Dialect/TritonNvidiaGPU/Transforms/TMALowering.cpp b/lib/Dialect/TritonNvidiaGPU/Transforms/TMALowering.cpp
@@ -140,7 +140,7 @@ class TritonNvidiaGPUTMALoweringPass
     mlir::RewritePatternSet patterns(context);
     patterns.add<TMALoadLowering, TMAStoreLowering, TMACreateDescLowering>(
         context);
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())
+    if (applyPatternsGreedily(m, std::move(patterns)).failed())
       signalPassFailure();
   }
 };
diff --git a/third_party/amd/lib/TritonAMDGPUToLLVM/BuiltinFuncToLLVM.cpp b/third_party/amd/lib/TritonAMDGPUToLLVM/BuiltinFuncToLLVM.cpp
@@ -229,7 +229,7 @@ struct ConvertBuiltinFuncToLLVM
     RewritePatternSet patterns(context);
     patterns.add<CallOpConversion>(context, this->ftz);
 
-    if (mlir::applyPatternsAndFoldGreedily(mod, std::move(patterns), config)
+    if (mlir::applyPatternsGreedily(mod, std::move(patterns), config)
             .failed()) {
       signalPassFailure();
     }
diff --git a/third_party/amd/lib/TritonAMDGPUTransforms/AccelerateAMDMatmul.cpp b/third_party/amd/lib/TritonAMDGPUTransforms/AccelerateAMDMatmul.cpp
@@ -857,7 +857,7 @@ class TritonAMDGPUAccelerateMatmulPass
     default:
       break;
     }
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) {
+    if (applyPatternsGreedily(m, std::move(patterns)).failed()) {
       signalPassFailure();
     }
     decomposeMixedModeDotOp(m);
diff --git a/third_party/amd/lib/TritonAMDGPUTransforms/ConvertToBufferOps.cpp b/third_party/amd/lib/TritonAMDGPUTransforms/ConvertToBufferOps.cpp
@@ -345,7 +345,7 @@ class TritonAMDGPUConvertToBufferOpsPass
 
     patterns.add<ConvertTritonLoadToBufferLoad>(context, assumptions);
     patterns.add<ConvertTritonStoreToBufferStore>(context, assumptions);
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())
+    if (applyPatternsGreedily(m, std::move(patterns)).failed())
       signalPassFailure();
   }
 };
diff --git a/third_party/amd/lib/TritonAMDGPUTransforms/OptimizeEpilogue.cpp b/third_party/amd/lib/TritonAMDGPUTransforms/OptimizeEpilogue.cpp
@@ -179,7 +179,7 @@ class TritonAMDGPUOptimizeEpiloguePass
 
     patterns.add<BypassEpilogueSMEM>(context);
 
-    if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) {
+    if (applyPatternsGreedily(m, std::move(patterns)).failed()) {
       signalPassFailure();
     }
   }
diff --git a/third_party/nvidia/lib/NVGPUToLLVM/NVGPUToLLVMPass.cpp b/third_party/nvidia/lib/NVGPUToLLVM/NVGPUToLLVMPass.cpp
@@ -609,7 +609,7 @@ class ConvertNVGPUToLLVM : public ConvertNVGPUToLLVMBase<ConvertNVGPUToLLVM> {
                  StoreMatrixOpPattern, ClusterArriveOpPattern, WGMMAOpPattern,
                  WGMMAWaitGroupOpPattern>(context);
 
-    if (applyPatternsAndFoldGreedily(mod, std::move(patterns)).failed())
+    if (applyPatternsGreedily(mod, std::move(patterns)).failed())
       signalPassFailure();
   }
 };
diff --git a/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/DecomposeUnsupportedConversions.cpp b/third_party/nvidia/lib/TritonNVIDIAGPUToLLVM/DecomposeUnsupportedConversions.cpp
@@ -75,7 +75,7 @@ struct DecomposeUnsupportedConversions
 
     mlir::RewritePatternSet patterns(&getContext());
     patterns.add<DecomposeLocalLoadToDotOperand>(&getContext());
-    if (mlir::applyPatternsAndFoldGreedily(mod, std::move(patterns)).failed()) {
+    if (mlir::applyPatternsGreedily(mod, std::move(patterns)).failed()) {
       signalPassFailure();
     }
   }

Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-86b69c31642e98f8357df62c09d118ad1da4e16a`
	`1`	`+91c5de7fb8f95132043ed08056e58238383cfcb9`
Original file line number	Diff line number	Diff line change
`@@ -204,7 +204,7 @@ class CombineOpsPass : public TritonCombineOpsBase<CombineOpsPass> {`
`204`	`204`	`patterns.add<CombineAddPtrPattern>(context);`
`205`	`205`	`patterns.add<CombineBroadcastMulReducePattern>(context);`
`206`	`206`
`207`		`- if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())`
	`207`	`+ if (applyPatternsGreedily(m, std::move(patterns)).failed())`
`208`	`208`	`signalPassFailure();`
`209`	`209`	`}`
`210`	`210`	`};`
Original file line number	Diff line number	Diff line change
`@@ -221,7 +221,7 @@ class ReorderBroadcastPass`
`221`	`221`	`// elementwise(splat(a), splat(b), ...) => splat(elementwise(a, b, ...))`
`222`	`222`	`patterns.add<MoveSplatAfterElementwisePattern>(context);`
`223`	`223`
`224`		`- if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed())`
	`224`	`+ if (applyPatternsGreedily(m, std::move(patterns)).failed())`
`225`	`225`	`signalPassFailure();`
`226`	`226`	`}`
`227`	`227`	`};`
Original file line number	Diff line number	Diff line change
`@@ -797,7 +797,7 @@ class TritonGPUAccelerateMatmulPass`
`797`	`797`	`mlir::RewritePatternSet patterns(context);`
`798`	`798`	`patterns.add<BlockedToMMA, DecomposeScaledBlocked>(context,`
`799`	`799`	`computeCapability);`
`800`		`- if (applyPatternsAndFoldGreedily(m, std::move(patterns)).failed()) {`
	`800`	`+ if (applyPatternsGreedily(m, std::move(patterns)).failed()) {`
`801`	`801`	`signalPassFailure();`
`802`	`802`	`}`
`803`	`803`	`// Now that we have picked the mma type, decompose dot that are not natively`
Original file line number	Diff line number	Diff line change
`@@ -114,7 +114,7 @@ class CoalesceAsyncCopyPass`
`114`	`114`	`mlir::RewritePatternSet patterns(context);`
`115`	`115`	`patterns.add<ClipAsyncCopySizePerThread>(context);`
`116`	`116`
`117`		`- if (failed(applyPatternsAndFoldGreedily(m, std::move(patterns))))`
	`117`	`+ if (failed(applyPatternsGreedily(m, std::move(patterns))))`
`118`	`118`	`signalPassFailure();`
`119`	`119`	`}`
`120`	`120`	`};`
Original file line number	Diff line number	Diff line change
`@@ -109,8 +109,7 @@ struct F32DotTCPass : public impl::TritonGPUF32DotTCBase<F32DotTCPass> {`
`109`	`109`
`110`	`110`	`RewritePatternSet decomposePatterns(context);`
`111`	`111`	`decomposePatterns.add<TF32x3>(context);`
`112`		`- if (applyPatternsAndFoldGreedily(m, std::move(decomposePatterns))`
`113`		`- .failed()) {`
	`112`	`+ if (applyPatternsGreedily(m, std::move(decomposePatterns)).failed()) {`
`114`	`113`	`signalPassFailure();`
`115`	`114`	`}`
`116`	`115`	`}`
Original file line number	Diff line number	Diff line change
`@@ -584,7 +584,7 @@ class TritonGPUOptimizeDotOperandsPass`
`584`	`584`	`patterns.add<FuseTransHopper>(context);`
`585`	`585`	`patterns.add<MMAV3UseRegOperand>(context);`
`586`	`586`	`ConvertLayoutOp::getCanonicalizationPatterns(patterns, context);`
`587`		`- if (failed(applyPatternsAndFoldGreedily(m, std::move(patterns))))`
	`587`	`+ if (failed(applyPatternsGreedily(m, std::move(patterns))))`
`588`	`588`	`signalPassFailure();`
`589`	`589`	`}`
`590`	`590`	`};`
Original file line number	Diff line number	Diff line change
`@@ -256,7 +256,7 @@ class TritonGPUOptimizeThreadLocalityPass`
`256`	`256`	`mlir::RewritePatternSet layoutPatterns(&getContext());`
`257`	`257`	`layoutPatterns.add<OptimizeReshapeLayoutPattern>(&getContext());`
`258`	`258`	`layoutPatterns.add<OptimizeGatherLayoutPattern>(&getContext());`
`259`		`- if (mlir::applyPatternsAndFoldGreedily(mod, std::move(layoutPatterns))`
	`259`	`+ if (mlir::applyPatternsGreedily(mod, std::move(layoutPatterns))`
`260`	`260`	`.failed()) {`
`261`	`261`	`signalPassFailure();`
`262`	`262`	`}`