[mlir][NFC] update `mlir/Dialect` create APIs (14/n) #149920

makslevental · 2025-07-21T22:17:13Z

See #147168 for more info.

github-actions · 2025-07-21T22:19:39Z

✅ With the latest revision this PR passed the C/C++ code formatter.

See llvm#147168 for more info.

llvmbot · 2025-07-21T22:44:00Z

@llvm/pr-subscribers-mlir-gpu
@llvm/pr-subscribers-mlir-affine
@llvm/pr-subscribers-mlir
@llvm/pr-subscribers-mlir-amx

@llvm/pr-subscribers-backend-amdgpu

Author: Maksim Levental (makslevental)

Changes

See #147168 for more info.

Patch is 136.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/149920.diff

23 Files Affected:

(modified) mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp (+15-14)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp (+28-27)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp (+5-5)
(modified) mlir/lib/Dialect/AMX/IR/AMXDialect.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp (+17-16)
(modified) mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/LoopFusion.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/PipelineDataTransfer.cpp (+5-5)
(modified) mlir/lib/Dialect/Affine/Transforms/ReifyValueBounds.cpp (+3-3)
(modified) mlir/lib/Dialect/Affine/Transforms/SuperVectorize.cpp (+25-26)
(modified) mlir/lib/Dialect/Affine/Utils/LoopUtils.cpp (+69-69)
(modified) mlir/lib/Dialect/Affine/Utils/Utils.cpp (+67-65)
(modified) mlir/lib/Dialect/Arith/IR/ArithDialect.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferDeallocationOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferizableOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateUnsupportedFloats.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateWideInt.cpp (+109-108)
(modified) mlir/lib/Dialect/Arith/Transforms/ExpandOps.cpp (+134-133)
(modified) mlir/lib/Dialect/Arith/Transforms/IntRangeOptimizations.cpp (+5-5)
(modified) mlir/lib/Dialect/Arith/Transforms/ReifyValueBounds.cpp (+14-14)
(modified) mlir/lib/Dialect/Arith/Transforms/ShardingInterfaceImpl.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Utils/Utils.cpp (+49-49)

diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
index 37e0d2af55fe1..6d1f64e94df15 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
@@ -99,8 +99,8 @@ static Value flattenVecToBits(ConversionPatternRewriter &rewriter, Location loc,
       vectorType.getElementTypeBitWidth() * vectorType.getNumElements();
   Type allBitsType = rewriter.getIntegerType(bitwidth);
   auto allBitsVecType = VectorType::get({1}, allBitsType);
-  Value bitcast = rewriter.create<vector::BitCastOp>(loc, allBitsVecType, val);
-  Value scalar = rewriter.create<vector::ExtractOp>(loc, bitcast, 0);
+  Value bitcast = vector::BitCastOp::create(rewriter, loc, allBitsVecType, val);
+  Value scalar = vector::ExtractOp::create(rewriter, loc, bitcast, 0);
   return scalar;
 }
 
@@ -118,27 +118,27 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
 
   SmallVector<NamedAttribute> loadAttrs;
   patchOperandSegmentSizes(origAttrs, loadAttrs, DataArgAction::Drop);
-  Value initialLoad =
-      rewriter.create<RawBufferLoadOp>(loc, dataType, invariantArgs, loadAttrs);
+  Value initialLoad = RawBufferLoadOp::create(rewriter, loc, dataType,
+                                              invariantArgs, loadAttrs);
   Block *currentBlock = rewriter.getInsertionBlock();
   Block *afterAtomic =
       rewriter.splitBlock(currentBlock, rewriter.getInsertionPoint());
   Block *loopBlock = rewriter.createBlock(afterAtomic, {dataType}, {loc});
 
   rewriter.setInsertionPointToEnd(currentBlock);
-  rewriter.create<cf::BranchOp>(loc, loopBlock, initialLoad);
+  cf::BranchOp::create(rewriter, loc, loopBlock, initialLoad);
 
   rewriter.setInsertionPointToEnd(loopBlock);
   Value prevLoad = loopBlock->getArgument(0);
-  Value operated = rewriter.create<ArithOp>(loc, data, prevLoad);
+  Value operated = ArithOp::create(rewriter, loc, data, prevLoad);
   dataType = operated.getType();
 
   SmallVector<NamedAttribute> cmpswapAttrs;
   patchOperandSegmentSizes(origAttrs, cmpswapAttrs, DataArgAction::Duplicate);
   SmallVector<Value> cmpswapArgs = {operated, prevLoad};
   cmpswapArgs.append(invariantArgs.begin(), invariantArgs.end());
-  Value atomicRes = rewriter.create<RawBufferAtomicCmpswapOp>(
-      loc, dataType, cmpswapArgs, cmpswapAttrs);
+  Value atomicRes = RawBufferAtomicCmpswapOp::create(rewriter, loc, dataType,
+                                                     cmpswapArgs, cmpswapAttrs);
 
   // We care about exact bitwise equality here, so do some bitcasts.
   // These will fold away during lowering to the ROCDL dialect, where
@@ -150,14 +150,15 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
   if (auto floatDataTy = dyn_cast<FloatType>(dataType)) {
     Type equivInt = rewriter.getIntegerType(floatDataTy.getWidth());
     prevLoadForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, prevLoad);
+        arith::BitcastOp::create(rewriter, loc, equivInt, prevLoad);
     atomicResForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, atomicRes);
+        arith::BitcastOp::create(rewriter, loc, equivInt, atomicRes);
   }
-  Value canLeave = rewriter.create<arith::CmpIOp>(
-      loc, arith::CmpIPredicate::eq, atomicResForCompare, prevLoadForCompare);
-  rewriter.create<cf::CondBranchOp>(loc, canLeave, afterAtomic, ValueRange{},
-                                    loopBlock, atomicRes);
+  Value canLeave =
+      arith::CmpIOp::create(rewriter, loc, arith::CmpIPredicate::eq,
+                            atomicResForCompare, prevLoadForCompare);
+  cf::CondBranchOp::create(rewriter, loc, canLeave, afterAtomic, ValueRange{},
+                           loopBlock, atomicRes);
   rewriter.eraseOp(atomicOp);
   return success();
 }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
index 60c8660658a95..42f238132069d 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
@@ -56,11 +56,11 @@ static Value createVectorLoadForMaskedLoad(OpBuilder &builder, Location loc,
                                            vector::MaskedLoadOp maskedOp,
                                            bool passthru) {
   VectorType vectorType = maskedOp.getVectorType();
-  Value load = builder.create<vector::LoadOp>(
-      loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
+  Value load = vector::LoadOp::create(
+      builder, loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
   if (passthru)
-    load = builder.create<arith::SelectOp>(loc, vectorType, maskedOp.getMask(),
-                                           load, maskedOp.getPassThru());
+    load = arith::SelectOp::create(builder, loc, vectorType, maskedOp.getMask(),
+                                   load, maskedOp.getPassThru());
   return load;
 }
 
@@ -110,7 +110,7 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
     SmallVector<OpFoldResult> indices = maskedOp.getIndices();
 
     auto stridedMetadata =
-        rewriter.create<memref::ExtractStridedMetadataOp>(loc, src);
+        memref::ExtractStridedMetadataOp::create(rewriter, loc, src);
     SmallVector<OpFoldResult> strides =
         stridedMetadata.getConstifiedMixedStrides();
     SmallVector<OpFoldResult> sizes = stridedMetadata.getConstifiedMixedSizes();
@@ -124,47 +124,47 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
 
     // delta = bufferSize - linearizedOffset
     Value vectorSizeOffset =
-        rewriter.create<arith::ConstantIndexOp>(loc, vectorSize);
+        arith::ConstantIndexOp::create(rewriter, loc, vectorSize);
     Value linearIndex =
         getValueOrCreateConstantIndexOp(rewriter, loc, linearizedIndices);
     Value totalSize = getValueOrCreateConstantIndexOp(
         rewriter, loc, linearizedInfo.linearizedSize);
-    Value delta = rewriter.create<arith::SubIOp>(loc, totalSize, linearIndex);
+    Value delta = arith::SubIOp::create(rewriter, loc, totalSize, linearIndex);
 
     // 1) check if delta < vectorSize
-    Value isOutofBounds = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
+    Value isOutofBounds = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
 
     // 2) check if (detla % elements_per_word != 0)
-    Value elementsPerWord = rewriter.create<arith::ConstantIndexOp>(
-        loc, llvm::divideCeil(32, elementBitWidth));
-    Value isNotWordAligned = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ne,
-        rewriter.create<arith::RemUIOp>(loc, delta, elementsPerWord),
-        rewriter.create<arith::ConstantIndexOp>(loc, 0));
+    Value elementsPerWord = arith::ConstantIndexOp::create(
+        rewriter, loc, llvm::divideCeil(32, elementBitWidth));
+    Value isNotWordAligned = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ne,
+        arith::RemUIOp::create(rewriter, loc, delta, elementsPerWord),
+        arith::ConstantIndexOp::create(rewriter, loc, 0));
 
     // We take the fallback of maskedload default lowering only it is both
     // out-of-bounds and not word aligned. The fallback ensures correct results
     // when loading at the boundary of the buffer since buffer load returns
     // inconsistent zeros for the whole word when boundary is crossed.
     Value ifCondition =
-        rewriter.create<arith::AndIOp>(loc, isOutofBounds, isNotWordAligned);
+        arith::AndIOp::create(rewriter, loc, isOutofBounds, isNotWordAligned);
 
     auto thenBuilder = [&](OpBuilder &builder, Location loc) {
       Operation *read = builder.clone(*maskedOp.getOperation());
       read->setAttr(kMaskedloadNeedsMask, builder.getUnitAttr());
       Value readResult = read->getResult(0);
-      builder.create<scf::YieldOp>(loc, readResult);
+      scf::YieldOp::create(builder, loc, readResult);
     };
 
     auto elseBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, maskedOp,
                                                 /*passthru=*/true);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
 
     auto ifOp =
-        rewriter.create<scf::IfOp>(loc, ifCondition, thenBuilder, elseBuilder);
+        scf::IfOp::create(rewriter, loc, ifCondition, thenBuilder, elseBuilder);
 
     rewriter.replaceOp(maskedOp, ifOp);
 
@@ -187,13 +187,13 @@ struct FullMaskedLoadToConditionalLoad
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, loadOp,
                                                 /*passthru=*/false);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
     auto falseBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<scf::YieldOp>(loc, loadOp.getPassThru());
+      scf::YieldOp::create(rewriter, loc, loadOp.getPassThru());
     };
-    auto ifOp = rewriter.create<scf::IfOp>(loadOp.getLoc(), cond, trueBuilder,
-                                           falseBuilder);
+    auto ifOp = scf::IfOp::create(rewriter, loadOp.getLoc(), cond, trueBuilder,
+                                  falseBuilder);
     rewriter.replaceOp(loadOp, ifOp);
     return success();
   }
@@ -212,11 +212,12 @@ struct FullMaskedStoreToConditionalStore
     Value cond = maybeCond.value();
 
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<vector::StoreOp>(loc, storeOp.getValueToStore(),
-                                       storeOp.getBase(), storeOp.getIndices());
-      rewriter.create<scf::YieldOp>(loc);
+      vector::StoreOp::create(rewriter, loc, storeOp.getValueToStore(),
+                              storeOp.getBase(), storeOp.getIndices());
+      scf::YieldOp::create(rewriter, loc);
     };
-    auto ifOp = rewriter.create<scf::IfOp>(storeOp.getLoc(), cond, trueBuilder);
+    auto ifOp =
+        scf::IfOp::create(rewriter, storeOp.getLoc(), cond, trueBuilder);
     rewriter.replaceOp(storeOp, ifOp);
     return success();
   }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
index 195f59d625554..f8bab8289cbc6 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
@@ -37,8 +37,8 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       return rewriter.notifyMatchFailure(metadataOp,
                                          "not a fat raw buffer cast");
     Location loc = castOp.getLoc();
-    auto sourceMetadata = rewriter.create<memref::ExtractStridedMetadataOp>(
-        loc, castOp.getSource());
+    auto sourceMetadata = memref::ExtractStridedMetadataOp::create(
+        rewriter, loc, castOp.getSource());
     SmallVector<Value> results;
     if (metadataOp.getBaseBuffer().use_empty()) {
       results.push_back(nullptr);
@@ -48,13 +48,13 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       if (baseBufferType == castOp.getResult().getType()) {
         results.push_back(castOp.getResult());
       } else {
-        results.push_back(rewriter.create<memref::ReinterpretCastOp>(
-            loc, baseBufferType, castOp.getResult(), /*offset=*/0,
+        results.push_back(memref::ReinterpretCastOp::create(
+            rewriter, loc, baseBufferType, castOp.getResult(), /*offset=*/0,
             /*sizes=*/ArrayRef<int64_t>{}, /*strides=*/ArrayRef<int64_t>{}));
       }
     }
     if (castOp.getResetOffset())
-      results.push_back(rewriter.create<arith::ConstantIndexOp>(loc, 0));
+      results.push_back(arith::ConstantIndexOp::create(rewriter, loc, 0));
     else
       results.push_back(sourceMetadata.getOffset());
     llvm::append_range(results, sourceMetadata.getSizes());
diff --git a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
index 12b375b373fa9..748ff1edbfeb2 100644
--- a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
+++ b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
@@ -76,8 +76,8 @@ static SmallVector<Value> getTileSizes(Location loc, amx::TileType tType,
   auto mattr = rewriter.getI16IntegerAttr(tType.getDimSize(0));
   auto nattr = rewriter.getI16IntegerAttr(tType.getDimSize(1) * bytes);
   return SmallVector<Value>{
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, mattr),
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, nattr)};
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, mattr),
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, nattr)};
 }
 
 /// Maps the 2-dim memref shape to the 64-bit stride. Note that the buffer
@@ -95,7 +95,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
     // Dynamic stride needs code to compute the stride at runtime.
     MemRefDescriptor memrefDescriptor(base);
     auto attr = rewriter.getI64IntegerAttr(bytes);
-    Value scale = rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr);
+    Value scale = LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr);
     return rewriter
         .create<LLVM::MulOp>(loc, llvmInt64Type, scale,
                              memrefDescriptor.stride(rewriter, loc, preLast))
@@ -103,7 +103,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
   }
   // Use direct constant for static stride.
   auto attr = rewriter.getI64IntegerAttr(strides[preLast] * bytes);
-  return rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr)
+  return LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr)
       .getResult();
 }
 
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
index f18cec5a14fae..df39544aeaa09 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
@@ -202,7 +202,7 @@ void AffineDataCopyGeneration::runOnBlock(Block *block,
 void AffineDataCopyGeneration::runOnOperation() {
   func::FuncOp f = getOperation();
   OpBuilder topBuilder(f.getBody());
-  zeroIndex = topBuilder.create<arith::ConstantIndexOp>(f.getLoc(), 0);
+  zeroIndex = arith::ConstantIndexOp::create(topBuilder, f.getLoc(), 0);
 
   // Nests that are copy-in's or copy-out's; the root AffineForOps of those
   // nests are stored herein.
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
index 5430bdc4ff858..c0d174a04abf9 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
@@ -58,8 +58,9 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
       // Note: basis elements and their products are, definitionally,
       // non-negative, so `nuw` is justified.
       if (dynamicPart)
-        dynamicPart = rewriter.create<arith::MulIOp>(
-            loc, dynamicPart, dynamicBasis[dynamicIndex - 1], ovflags);
+        dynamicPart =
+            arith::MulIOp::create(rewriter, loc, dynamicPart,
+                                  dynamicBasis[dynamicIndex - 1], ovflags);
       else
         dynamicPart = dynamicBasis[dynamicIndex - 1];
       --dynamicIndex;
@@ -74,7 +75,7 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
           rewriter.createOrFold<arith::ConstantIndexOp>(loc, staticPart);
       if (dynamicPart)
         stride =
-            rewriter.create<arith::MulIOp>(loc, dynamicPart, stride, ovflags);
+            arith::MulIOp::create(rewriter, loc, dynamicPart, stride, ovflags);
       result.push_back(stride);
     }
   }
@@ -106,20 +107,20 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
   Value zero = rewriter.createOrFold<arith::ConstantIndexOp>(loc, 0);
 
   Value initialPart =
-      rewriter.create<arith::FloorDivSIOp>(loc, linearIdx, strides.front());
+      arith::FloorDivSIOp::create(rewriter, loc, linearIdx, strides.front());
   results.push_back(initialPart);
 
   auto emitModTerm = [&](Value stride) -> Value {
-    Value remainder = rewriter.create<arith::RemSIOp>(loc, linearIdx, stride);
-    Value remainderNegative = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::slt, remainder, zero);
+    Value remainder = arith::RemSIOp::create(rewriter, loc, linearIdx, stride);
+    Value remainderNegative = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::slt, remainder, zero);
     // If the correction is relevant, this term is <= stride, which is known
     // to be positive in `index`. Otherwise, while 2 * stride might overflow,
     // this branch won't be taken, so the risk of `poison` is fine.
-    Value corrected = rewriter.create<arith::AddIOp>(
-        loc, remainder, stride, arith::IntegerOverflowFlags::nsw);
-    Value mod = rewriter.create<arith::SelectOp>(loc, remainderNegative,
-                                                 corrected, remainder);
+    Value corrected = arith::AddIOp::create(rewriter, loc, remainder, stride,
+                                            arith::IntegerOverflowFlags::nsw);
+    Value mod = arith::SelectOp::create(rewriter, loc, remainderNegative,
+                                        corrected, remainder);
     return mod;
   };
 
@@ -131,7 +132,7 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
     // We know both inputs are positive, so floorDiv == div.
     // This could potentially be a divui, but it's not clear if that would
     // cause issues.
-    Value divided = rewriter.create<arith::DivSIOp>(loc, modulus, nextStride);
+    Value divided = arith::DivSIOp::create(rewriter, loc, modulus, nextStride);
     results.push_back(divided);
   }
 
@@ -167,8 +168,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   // our hands on an `OpOperand&` for the loop invariant counting function.
   for (auto [stride, idxOp] :
        llvm::zip_equal(strides, llvm::drop_end(op.getMultiIndexMutable()))) {
-    Value scaledIdx = rewriter.create<arith::MulIOp>(
-        loc, idxOp.get(), stride, arith::IntegerOverflowFlags::nsw);
+    Value scaledIdx = arith::MulIOp::create(rewriter, loc, idxOp.get(), stride,
+                                            arith::IntegerOverflowFlags::nsw);
     int64_t numHoistableLoops = numEnclosingInvariantLoops(idxOp);
     scaledValues.emplace_back(scaledIdx, numHoistableLoops);
   }
@@ -184,8 +185,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   Value result = scaledValues.front().first;
   for (auto [scaledValue, numHoistableLoops] : llvm::drop_begin(scaledValues)) {
     std::ignore = numHoistableLoops;
-    result = rewriter.create<arith::AddIOp>(loc, result, scaledValue,
-                                            arith::IntegerOverflowFlags::nsw);
+    result = arith::AddIOp::create(rewriter, loc, result, scaledValue,
+                                   arith::IntegerOverflowFlags::nsw);
   }
   rewriter.replaceOp(op, result);
   return success();
diff --git a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
index 4fd0cf9b3cd25..3c00b323473d2 100644
--- a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
@@ -88,8 +88,8 @@ static AffineApplyOp createSubApply(RewriterBase &rewriter,
   auto rhsMap = AffineMap::get(m.getNumDims(), m.getNumSymbols(), expr, ctx);
   SmallVector<Value> rhsOperands = originalOp->getOperands();
   canonicalizeMapAndOperands(&rhsMap, &rhsOperands);
-  return rewriter.create<AffineApplyOp>(originalOp.getLoc(), rhsMap,
-                                        rhsOperands);
+  return AffineApplyOp::create(rewriter, originalOp.getLoc(), rhsMap,
+                               rhsOperands);
 }
 
 FailureOr<AffineApplyOp> mlir::affine::decompose(RewriterBase &rewriter,
@@ -160,8 +160,8 @@ FailureOr<AffineApplyO...
[truncated]

llvmbot · 2025-07-21T22:44:00Z

@llvm/pr-subscribers-mlir-amdgpu

Author: Maksim Levental (makslevental)

Changes

See #147168 for more info.

Patch is 136.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/149920.diff

23 Files Affected:

(modified) mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp (+15-14)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp (+28-27)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp (+5-5)
(modified) mlir/lib/Dialect/AMX/IR/AMXDialect.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp (+17-16)
(modified) mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/LoopFusion.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/PipelineDataTransfer.cpp (+5-5)
(modified) mlir/lib/Dialect/Affine/Transforms/ReifyValueBounds.cpp (+3-3)
(modified) mlir/lib/Dialect/Affine/Transforms/SuperVectorize.cpp (+25-26)
(modified) mlir/lib/Dialect/Affine/Utils/LoopUtils.cpp (+69-69)
(modified) mlir/lib/Dialect/Affine/Utils/Utils.cpp (+67-65)
(modified) mlir/lib/Dialect/Arith/IR/ArithDialect.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferDeallocationOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferizableOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateUnsupportedFloats.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateWideInt.cpp (+109-108)
(modified) mlir/lib/Dialect/Arith/Transforms/ExpandOps.cpp (+134-133)
(modified) mlir/lib/Dialect/Arith/Transforms/IntRangeOptimizations.cpp (+5-5)
(modified) mlir/lib/Dialect/Arith/Transforms/ReifyValueBounds.cpp (+14-14)
(modified) mlir/lib/Dialect/Arith/Transforms/ShardingInterfaceImpl.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Utils/Utils.cpp (+49-49)

diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
index 37e0d2af55fe1..6d1f64e94df15 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
@@ -99,8 +99,8 @@ static Value flattenVecToBits(ConversionPatternRewriter &rewriter, Location loc,
       vectorType.getElementTypeBitWidth() * vectorType.getNumElements();
   Type allBitsType = rewriter.getIntegerType(bitwidth);
   auto allBitsVecType = VectorType::get({1}, allBitsType);
-  Value bitcast = rewriter.create<vector::BitCastOp>(loc, allBitsVecType, val);
-  Value scalar = rewriter.create<vector::ExtractOp>(loc, bitcast, 0);
+  Value bitcast = vector::BitCastOp::create(rewriter, loc, allBitsVecType, val);
+  Value scalar = vector::ExtractOp::create(rewriter, loc, bitcast, 0);
   return scalar;
 }
 
@@ -118,27 +118,27 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
 
   SmallVector<NamedAttribute> loadAttrs;
   patchOperandSegmentSizes(origAttrs, loadAttrs, DataArgAction::Drop);
-  Value initialLoad =
-      rewriter.create<RawBufferLoadOp>(loc, dataType, invariantArgs, loadAttrs);
+  Value initialLoad = RawBufferLoadOp::create(rewriter, loc, dataType,
+                                              invariantArgs, loadAttrs);
   Block *currentBlock = rewriter.getInsertionBlock();
   Block *afterAtomic =
       rewriter.splitBlock(currentBlock, rewriter.getInsertionPoint());
   Block *loopBlock = rewriter.createBlock(afterAtomic, {dataType}, {loc});
 
   rewriter.setInsertionPointToEnd(currentBlock);
-  rewriter.create<cf::BranchOp>(loc, loopBlock, initialLoad);
+  cf::BranchOp::create(rewriter, loc, loopBlock, initialLoad);
 
   rewriter.setInsertionPointToEnd(loopBlock);
   Value prevLoad = loopBlock->getArgument(0);
-  Value operated = rewriter.create<ArithOp>(loc, data, prevLoad);
+  Value operated = ArithOp::create(rewriter, loc, data, prevLoad);
   dataType = operated.getType();
 
   SmallVector<NamedAttribute> cmpswapAttrs;
   patchOperandSegmentSizes(origAttrs, cmpswapAttrs, DataArgAction::Duplicate);
   SmallVector<Value> cmpswapArgs = {operated, prevLoad};
   cmpswapArgs.append(invariantArgs.begin(), invariantArgs.end());
-  Value atomicRes = rewriter.create<RawBufferAtomicCmpswapOp>(
-      loc, dataType, cmpswapArgs, cmpswapAttrs);
+  Value atomicRes = RawBufferAtomicCmpswapOp::create(rewriter, loc, dataType,
+                                                     cmpswapArgs, cmpswapAttrs);
 
   // We care about exact bitwise equality here, so do some bitcasts.
   // These will fold away during lowering to the ROCDL dialect, where
@@ -150,14 +150,15 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
   if (auto floatDataTy = dyn_cast<FloatType>(dataType)) {
     Type equivInt = rewriter.getIntegerType(floatDataTy.getWidth());
     prevLoadForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, prevLoad);
+        arith::BitcastOp::create(rewriter, loc, equivInt, prevLoad);
     atomicResForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, atomicRes);
+        arith::BitcastOp::create(rewriter, loc, equivInt, atomicRes);
   }
-  Value canLeave = rewriter.create<arith::CmpIOp>(
-      loc, arith::CmpIPredicate::eq, atomicResForCompare, prevLoadForCompare);
-  rewriter.create<cf::CondBranchOp>(loc, canLeave, afterAtomic, ValueRange{},
-                                    loopBlock, atomicRes);
+  Value canLeave =
+      arith::CmpIOp::create(rewriter, loc, arith::CmpIPredicate::eq,
+                            atomicResForCompare, prevLoadForCompare);
+  cf::CondBranchOp::create(rewriter, loc, canLeave, afterAtomic, ValueRange{},
+                           loopBlock, atomicRes);
   rewriter.eraseOp(atomicOp);
   return success();
 }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
index 60c8660658a95..42f238132069d 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
@@ -56,11 +56,11 @@ static Value createVectorLoadForMaskedLoad(OpBuilder &builder, Location loc,
                                            vector::MaskedLoadOp maskedOp,
                                            bool passthru) {
   VectorType vectorType = maskedOp.getVectorType();
-  Value load = builder.create<vector::LoadOp>(
-      loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
+  Value load = vector::LoadOp::create(
+      builder, loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
   if (passthru)
-    load = builder.create<arith::SelectOp>(loc, vectorType, maskedOp.getMask(),
-                                           load, maskedOp.getPassThru());
+    load = arith::SelectOp::create(builder, loc, vectorType, maskedOp.getMask(),
+                                   load, maskedOp.getPassThru());
   return load;
 }
 
@@ -110,7 +110,7 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
     SmallVector<OpFoldResult> indices = maskedOp.getIndices();
 
     auto stridedMetadata =
-        rewriter.create<memref::ExtractStridedMetadataOp>(loc, src);
+        memref::ExtractStridedMetadataOp::create(rewriter, loc, src);
     SmallVector<OpFoldResult> strides =
         stridedMetadata.getConstifiedMixedStrides();
     SmallVector<OpFoldResult> sizes = stridedMetadata.getConstifiedMixedSizes();
@@ -124,47 +124,47 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
 
     // delta = bufferSize - linearizedOffset
     Value vectorSizeOffset =
-        rewriter.create<arith::ConstantIndexOp>(loc, vectorSize);
+        arith::ConstantIndexOp::create(rewriter, loc, vectorSize);
     Value linearIndex =
         getValueOrCreateConstantIndexOp(rewriter, loc, linearizedIndices);
     Value totalSize = getValueOrCreateConstantIndexOp(
         rewriter, loc, linearizedInfo.linearizedSize);
-    Value delta = rewriter.create<arith::SubIOp>(loc, totalSize, linearIndex);
+    Value delta = arith::SubIOp::create(rewriter, loc, totalSize, linearIndex);
 
     // 1) check if delta < vectorSize
-    Value isOutofBounds = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
+    Value isOutofBounds = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
 
     // 2) check if (detla % elements_per_word != 0)
-    Value elementsPerWord = rewriter.create<arith::ConstantIndexOp>(
-        loc, llvm::divideCeil(32, elementBitWidth));
-    Value isNotWordAligned = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ne,
-        rewriter.create<arith::RemUIOp>(loc, delta, elementsPerWord),
-        rewriter.create<arith::ConstantIndexOp>(loc, 0));
+    Value elementsPerWord = arith::ConstantIndexOp::create(
+        rewriter, loc, llvm::divideCeil(32, elementBitWidth));
+    Value isNotWordAligned = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ne,
+        arith::RemUIOp::create(rewriter, loc, delta, elementsPerWord),
+        arith::ConstantIndexOp::create(rewriter, loc, 0));
 
     // We take the fallback of maskedload default lowering only it is both
     // out-of-bounds and not word aligned. The fallback ensures correct results
     // when loading at the boundary of the buffer since buffer load returns
     // inconsistent zeros for the whole word when boundary is crossed.
     Value ifCondition =
-        rewriter.create<arith::AndIOp>(loc, isOutofBounds, isNotWordAligned);
+        arith::AndIOp::create(rewriter, loc, isOutofBounds, isNotWordAligned);
 
     auto thenBuilder = [&](OpBuilder &builder, Location loc) {
       Operation *read = builder.clone(*maskedOp.getOperation());
       read->setAttr(kMaskedloadNeedsMask, builder.getUnitAttr());
       Value readResult = read->getResult(0);
-      builder.create<scf::YieldOp>(loc, readResult);
+      scf::YieldOp::create(builder, loc, readResult);
     };
 
     auto elseBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, maskedOp,
                                                 /*passthru=*/true);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
 
     auto ifOp =
-        rewriter.create<scf::IfOp>(loc, ifCondition, thenBuilder, elseBuilder);
+        scf::IfOp::create(rewriter, loc, ifCondition, thenBuilder, elseBuilder);
 
     rewriter.replaceOp(maskedOp, ifOp);
 
@@ -187,13 +187,13 @@ struct FullMaskedLoadToConditionalLoad
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, loadOp,
                                                 /*passthru=*/false);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
     auto falseBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<scf::YieldOp>(loc, loadOp.getPassThru());
+      scf::YieldOp::create(rewriter, loc, loadOp.getPassThru());
     };
-    auto ifOp = rewriter.create<scf::IfOp>(loadOp.getLoc(), cond, trueBuilder,
-                                           falseBuilder);
+    auto ifOp = scf::IfOp::create(rewriter, loadOp.getLoc(), cond, trueBuilder,
+                                  falseBuilder);
     rewriter.replaceOp(loadOp, ifOp);
     return success();
   }
@@ -212,11 +212,12 @@ struct FullMaskedStoreToConditionalStore
     Value cond = maybeCond.value();
 
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<vector::StoreOp>(loc, storeOp.getValueToStore(),
-                                       storeOp.getBase(), storeOp.getIndices());
-      rewriter.create<scf::YieldOp>(loc);
+      vector::StoreOp::create(rewriter, loc, storeOp.getValueToStore(),
+                              storeOp.getBase(), storeOp.getIndices());
+      scf::YieldOp::create(rewriter, loc);
     };
-    auto ifOp = rewriter.create<scf::IfOp>(storeOp.getLoc(), cond, trueBuilder);
+    auto ifOp =
+        scf::IfOp::create(rewriter, storeOp.getLoc(), cond, trueBuilder);
     rewriter.replaceOp(storeOp, ifOp);
     return success();
   }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
index 195f59d625554..f8bab8289cbc6 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
@@ -37,8 +37,8 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       return rewriter.notifyMatchFailure(metadataOp,
                                          "not a fat raw buffer cast");
     Location loc = castOp.getLoc();
-    auto sourceMetadata = rewriter.create<memref::ExtractStridedMetadataOp>(
-        loc, castOp.getSource());
+    auto sourceMetadata = memref::ExtractStridedMetadataOp::create(
+        rewriter, loc, castOp.getSource());
     SmallVector<Value> results;
     if (metadataOp.getBaseBuffer().use_empty()) {
       results.push_back(nullptr);
@@ -48,13 +48,13 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       if (baseBufferType == castOp.getResult().getType()) {
         results.push_back(castOp.getResult());
       } else {
-        results.push_back(rewriter.create<memref::ReinterpretCastOp>(
-            loc, baseBufferType, castOp.getResult(), /*offset=*/0,
+        results.push_back(memref::ReinterpretCastOp::create(
+            rewriter, loc, baseBufferType, castOp.getResult(), /*offset=*/0,
             /*sizes=*/ArrayRef<int64_t>{}, /*strides=*/ArrayRef<int64_t>{}));
       }
     }
     if (castOp.getResetOffset())
-      results.push_back(rewriter.create<arith::ConstantIndexOp>(loc, 0));
+      results.push_back(arith::ConstantIndexOp::create(rewriter, loc, 0));
     else
       results.push_back(sourceMetadata.getOffset());
     llvm::append_range(results, sourceMetadata.getSizes());
diff --git a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
index 12b375b373fa9..748ff1edbfeb2 100644
--- a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
+++ b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
@@ -76,8 +76,8 @@ static SmallVector<Value> getTileSizes(Location loc, amx::TileType tType,
   auto mattr = rewriter.getI16IntegerAttr(tType.getDimSize(0));
   auto nattr = rewriter.getI16IntegerAttr(tType.getDimSize(1) * bytes);
   return SmallVector<Value>{
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, mattr),
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, nattr)};
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, mattr),
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, nattr)};
 }
 
 /// Maps the 2-dim memref shape to the 64-bit stride. Note that the buffer
@@ -95,7 +95,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
     // Dynamic stride needs code to compute the stride at runtime.
     MemRefDescriptor memrefDescriptor(base);
     auto attr = rewriter.getI64IntegerAttr(bytes);
-    Value scale = rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr);
+    Value scale = LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr);
     return rewriter
         .create<LLVM::MulOp>(loc, llvmInt64Type, scale,
                              memrefDescriptor.stride(rewriter, loc, preLast))
@@ -103,7 +103,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
   }
   // Use direct constant for static stride.
   auto attr = rewriter.getI64IntegerAttr(strides[preLast] * bytes);
-  return rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr)
+  return LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr)
       .getResult();
 }
 
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
index f18cec5a14fae..df39544aeaa09 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
@@ -202,7 +202,7 @@ void AffineDataCopyGeneration::runOnBlock(Block *block,
 void AffineDataCopyGeneration::runOnOperation() {
   func::FuncOp f = getOperation();
   OpBuilder topBuilder(f.getBody());
-  zeroIndex = topBuilder.create<arith::ConstantIndexOp>(f.getLoc(), 0);
+  zeroIndex = arith::ConstantIndexOp::create(topBuilder, f.getLoc(), 0);
 
   // Nests that are copy-in's or copy-out's; the root AffineForOps of those
   // nests are stored herein.
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
index 5430bdc4ff858..c0d174a04abf9 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
@@ -58,8 +58,9 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
       // Note: basis elements and their products are, definitionally,
       // non-negative, so `nuw` is justified.
       if (dynamicPart)
-        dynamicPart = rewriter.create<arith::MulIOp>(
-            loc, dynamicPart, dynamicBasis[dynamicIndex - 1], ovflags);
+        dynamicPart =
+            arith::MulIOp::create(rewriter, loc, dynamicPart,
+                                  dynamicBasis[dynamicIndex - 1], ovflags);
       else
         dynamicPart = dynamicBasis[dynamicIndex - 1];
       --dynamicIndex;
@@ -74,7 +75,7 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
           rewriter.createOrFold<arith::ConstantIndexOp>(loc, staticPart);
       if (dynamicPart)
         stride =
-            rewriter.create<arith::MulIOp>(loc, dynamicPart, stride, ovflags);
+            arith::MulIOp::create(rewriter, loc, dynamicPart, stride, ovflags);
       result.push_back(stride);
     }
   }
@@ -106,20 +107,20 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
   Value zero = rewriter.createOrFold<arith::ConstantIndexOp>(loc, 0);
 
   Value initialPart =
-      rewriter.create<arith::FloorDivSIOp>(loc, linearIdx, strides.front());
+      arith::FloorDivSIOp::create(rewriter, loc, linearIdx, strides.front());
   results.push_back(initialPart);
 
   auto emitModTerm = [&](Value stride) -> Value {
-    Value remainder = rewriter.create<arith::RemSIOp>(loc, linearIdx, stride);
-    Value remainderNegative = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::slt, remainder, zero);
+    Value remainder = arith::RemSIOp::create(rewriter, loc, linearIdx, stride);
+    Value remainderNegative = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::slt, remainder, zero);
     // If the correction is relevant, this term is <= stride, which is known
     // to be positive in `index`. Otherwise, while 2 * stride might overflow,
     // this branch won't be taken, so the risk of `poison` is fine.
-    Value corrected = rewriter.create<arith::AddIOp>(
-        loc, remainder, stride, arith::IntegerOverflowFlags::nsw);
-    Value mod = rewriter.create<arith::SelectOp>(loc, remainderNegative,
-                                                 corrected, remainder);
+    Value corrected = arith::AddIOp::create(rewriter, loc, remainder, stride,
+                                            arith::IntegerOverflowFlags::nsw);
+    Value mod = arith::SelectOp::create(rewriter, loc, remainderNegative,
+                                        corrected, remainder);
     return mod;
   };
 
@@ -131,7 +132,7 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
     // We know both inputs are positive, so floorDiv == div.
     // This could potentially be a divui, but it's not clear if that would
     // cause issues.
-    Value divided = rewriter.create<arith::DivSIOp>(loc, modulus, nextStride);
+    Value divided = arith::DivSIOp::create(rewriter, loc, modulus, nextStride);
     results.push_back(divided);
   }
 
@@ -167,8 +168,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   // our hands on an `OpOperand&` for the loop invariant counting function.
   for (auto [stride, idxOp] :
        llvm::zip_equal(strides, llvm::drop_end(op.getMultiIndexMutable()))) {
-    Value scaledIdx = rewriter.create<arith::MulIOp>(
-        loc, idxOp.get(), stride, arith::IntegerOverflowFlags::nsw);
+    Value scaledIdx = arith::MulIOp::create(rewriter, loc, idxOp.get(), stride,
+                                            arith::IntegerOverflowFlags::nsw);
     int64_t numHoistableLoops = numEnclosingInvariantLoops(idxOp);
     scaledValues.emplace_back(scaledIdx, numHoistableLoops);
   }
@@ -184,8 +185,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   Value result = scaledValues.front().first;
   for (auto [scaledValue, numHoistableLoops] : llvm::drop_begin(scaledValues)) {
     std::ignore = numHoistableLoops;
-    result = rewriter.create<arith::AddIOp>(loc, result, scaledValue,
-                                            arith::IntegerOverflowFlags::nsw);
+    result = arith::AddIOp::create(rewriter, loc, result, scaledValue,
+                                   arith::IntegerOverflowFlags::nsw);
   }
   rewriter.replaceOp(op, result);
   return success();
diff --git a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
index 4fd0cf9b3cd25..3c00b323473d2 100644
--- a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
@@ -88,8 +88,8 @@ static AffineApplyOp createSubApply(RewriterBase &rewriter,
   auto rhsMap = AffineMap::get(m.getNumDims(), m.getNumSymbols(), expr, ctx);
   SmallVector<Value> rhsOperands = originalOp->getOperands();
   canonicalizeMapAndOperands(&rhsMap, &rhsOperands);
-  return rewriter.create<AffineApplyOp>(originalOp.getLoc(), rhsMap,
-                                        rhsOperands);
+  return AffineApplyOp::create(rewriter, originalOp.getLoc(), rhsMap,
+                               rhsOperands);
 }
 
 FailureOr<AffineApplyOp> mlir::affine::decompose(RewriterBase &rewriter,
@@ -160,8 +160,8 @@ FailureOr<AffineApplyO...
[truncated]

llvmbot · 2025-07-21T22:44:01Z

@llvm/pr-subscribers-mlir-arith

Author: Maksim Levental (makslevental)

Changes

See #147168 for more info.

Patch is 136.98 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/149920.diff

23 Files Affected:

(modified) mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp (+15-14)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp (+28-27)
(modified) mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp (+5-5)
(modified) mlir/lib/Dialect/AMX/IR/AMXDialect.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp (+17-16)
(modified) mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp (+4-4)
(modified) mlir/lib/Dialect/Affine/Transforms/LoopFusion.cpp (+1-1)
(modified) mlir/lib/Dialect/Affine/Transforms/PipelineDataTransfer.cpp (+5-5)
(modified) mlir/lib/Dialect/Affine/Transforms/ReifyValueBounds.cpp (+3-3)
(modified) mlir/lib/Dialect/Affine/Transforms/SuperVectorize.cpp (+25-26)
(modified) mlir/lib/Dialect/Affine/Utils/LoopUtils.cpp (+69-69)
(modified) mlir/lib/Dialect/Affine/Utils/Utils.cpp (+67-65)
(modified) mlir/lib/Dialect/Arith/IR/ArithDialect.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferDeallocationOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/BufferizableOpInterfaceImpl.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateUnsupportedFloats.cpp (+2-2)
(modified) mlir/lib/Dialect/Arith/Transforms/EmulateWideInt.cpp (+109-108)
(modified) mlir/lib/Dialect/Arith/Transforms/ExpandOps.cpp (+134-133)
(modified) mlir/lib/Dialect/Arith/Transforms/IntRangeOptimizations.cpp (+5-5)
(modified) mlir/lib/Dialect/Arith/Transforms/ReifyValueBounds.cpp (+14-14)
(modified) mlir/lib/Dialect/Arith/Transforms/ShardingInterfaceImpl.cpp (+1-1)
(modified) mlir/lib/Dialect/Arith/Utils/Utils.cpp (+49-49)

diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
index 37e0d2af55fe1..6d1f64e94df15 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/EmulateAtomics.cpp
@@ -99,8 +99,8 @@ static Value flattenVecToBits(ConversionPatternRewriter &rewriter, Location loc,
       vectorType.getElementTypeBitWidth() * vectorType.getNumElements();
   Type allBitsType = rewriter.getIntegerType(bitwidth);
   auto allBitsVecType = VectorType::get({1}, allBitsType);
-  Value bitcast = rewriter.create<vector::BitCastOp>(loc, allBitsVecType, val);
-  Value scalar = rewriter.create<vector::ExtractOp>(loc, bitcast, 0);
+  Value bitcast = vector::BitCastOp::create(rewriter, loc, allBitsVecType, val);
+  Value scalar = vector::ExtractOp::create(rewriter, loc, bitcast, 0);
   return scalar;
 }
 
@@ -118,27 +118,27 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
 
   SmallVector<NamedAttribute> loadAttrs;
   patchOperandSegmentSizes(origAttrs, loadAttrs, DataArgAction::Drop);
-  Value initialLoad =
-      rewriter.create<RawBufferLoadOp>(loc, dataType, invariantArgs, loadAttrs);
+  Value initialLoad = RawBufferLoadOp::create(rewriter, loc, dataType,
+                                              invariantArgs, loadAttrs);
   Block *currentBlock = rewriter.getInsertionBlock();
   Block *afterAtomic =
       rewriter.splitBlock(currentBlock, rewriter.getInsertionPoint());
   Block *loopBlock = rewriter.createBlock(afterAtomic, {dataType}, {loc});
 
   rewriter.setInsertionPointToEnd(currentBlock);
-  rewriter.create<cf::BranchOp>(loc, loopBlock, initialLoad);
+  cf::BranchOp::create(rewriter, loc, loopBlock, initialLoad);
 
   rewriter.setInsertionPointToEnd(loopBlock);
   Value prevLoad = loopBlock->getArgument(0);
-  Value operated = rewriter.create<ArithOp>(loc, data, prevLoad);
+  Value operated = ArithOp::create(rewriter, loc, data, prevLoad);
   dataType = operated.getType();
 
   SmallVector<NamedAttribute> cmpswapAttrs;
   patchOperandSegmentSizes(origAttrs, cmpswapAttrs, DataArgAction::Duplicate);
   SmallVector<Value> cmpswapArgs = {operated, prevLoad};
   cmpswapArgs.append(invariantArgs.begin(), invariantArgs.end());
-  Value atomicRes = rewriter.create<RawBufferAtomicCmpswapOp>(
-      loc, dataType, cmpswapArgs, cmpswapAttrs);
+  Value atomicRes = RawBufferAtomicCmpswapOp::create(rewriter, loc, dataType,
+                                                     cmpswapArgs, cmpswapAttrs);
 
   // We care about exact bitwise equality here, so do some bitcasts.
   // These will fold away during lowering to the ROCDL dialect, where
@@ -150,14 +150,15 @@ LogicalResult RawBufferAtomicByCasPattern<AtomicOp, ArithOp>::matchAndRewrite(
   if (auto floatDataTy = dyn_cast<FloatType>(dataType)) {
     Type equivInt = rewriter.getIntegerType(floatDataTy.getWidth());
     prevLoadForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, prevLoad);
+        arith::BitcastOp::create(rewriter, loc, equivInt, prevLoad);
     atomicResForCompare =
-        rewriter.create<arith::BitcastOp>(loc, equivInt, atomicRes);
+        arith::BitcastOp::create(rewriter, loc, equivInt, atomicRes);
   }
-  Value canLeave = rewriter.create<arith::CmpIOp>(
-      loc, arith::CmpIPredicate::eq, atomicResForCompare, prevLoadForCompare);
-  rewriter.create<cf::CondBranchOp>(loc, canLeave, afterAtomic, ValueRange{},
-                                    loopBlock, atomicRes);
+  Value canLeave =
+      arith::CmpIOp::create(rewriter, loc, arith::CmpIPredicate::eq,
+                            atomicResForCompare, prevLoadForCompare);
+  cf::CondBranchOp::create(rewriter, loc, canLeave, afterAtomic, ValueRange{},
+                           loopBlock, atomicRes);
   rewriter.eraseOp(atomicOp);
   return success();
 }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
index 60c8660658a95..42f238132069d 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/MaskedloadToLoad.cpp
@@ -56,11 +56,11 @@ static Value createVectorLoadForMaskedLoad(OpBuilder &builder, Location loc,
                                            vector::MaskedLoadOp maskedOp,
                                            bool passthru) {
   VectorType vectorType = maskedOp.getVectorType();
-  Value load = builder.create<vector::LoadOp>(
-      loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
+  Value load = vector::LoadOp::create(
+      builder, loc, vectorType, maskedOp.getBase(), maskedOp.getIndices());
   if (passthru)
-    load = builder.create<arith::SelectOp>(loc, vectorType, maskedOp.getMask(),
-                                           load, maskedOp.getPassThru());
+    load = arith::SelectOp::create(builder, loc, vectorType, maskedOp.getMask(),
+                                   load, maskedOp.getPassThru());
   return load;
 }
 
@@ -110,7 +110,7 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
     SmallVector<OpFoldResult> indices = maskedOp.getIndices();
 
     auto stridedMetadata =
-        rewriter.create<memref::ExtractStridedMetadataOp>(loc, src);
+        memref::ExtractStridedMetadataOp::create(rewriter, loc, src);
     SmallVector<OpFoldResult> strides =
         stridedMetadata.getConstifiedMixedStrides();
     SmallVector<OpFoldResult> sizes = stridedMetadata.getConstifiedMixedSizes();
@@ -124,47 +124,47 @@ struct MaskedLoadLowering final : OpRewritePattern<vector::MaskedLoadOp> {
 
     // delta = bufferSize - linearizedOffset
     Value vectorSizeOffset =
-        rewriter.create<arith::ConstantIndexOp>(loc, vectorSize);
+        arith::ConstantIndexOp::create(rewriter, loc, vectorSize);
     Value linearIndex =
         getValueOrCreateConstantIndexOp(rewriter, loc, linearizedIndices);
     Value totalSize = getValueOrCreateConstantIndexOp(
         rewriter, loc, linearizedInfo.linearizedSize);
-    Value delta = rewriter.create<arith::SubIOp>(loc, totalSize, linearIndex);
+    Value delta = arith::SubIOp::create(rewriter, loc, totalSize, linearIndex);
 
     // 1) check if delta < vectorSize
-    Value isOutofBounds = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
+    Value isOutofBounds = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ult, delta, vectorSizeOffset);
 
     // 2) check if (detla % elements_per_word != 0)
-    Value elementsPerWord = rewriter.create<arith::ConstantIndexOp>(
-        loc, llvm::divideCeil(32, elementBitWidth));
-    Value isNotWordAligned = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::ne,
-        rewriter.create<arith::RemUIOp>(loc, delta, elementsPerWord),
-        rewriter.create<arith::ConstantIndexOp>(loc, 0));
+    Value elementsPerWord = arith::ConstantIndexOp::create(
+        rewriter, loc, llvm::divideCeil(32, elementBitWidth));
+    Value isNotWordAligned = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::ne,
+        arith::RemUIOp::create(rewriter, loc, delta, elementsPerWord),
+        arith::ConstantIndexOp::create(rewriter, loc, 0));
 
     // We take the fallback of maskedload default lowering only it is both
     // out-of-bounds and not word aligned. The fallback ensures correct results
     // when loading at the boundary of the buffer since buffer load returns
     // inconsistent zeros for the whole word when boundary is crossed.
     Value ifCondition =
-        rewriter.create<arith::AndIOp>(loc, isOutofBounds, isNotWordAligned);
+        arith::AndIOp::create(rewriter, loc, isOutofBounds, isNotWordAligned);
 
     auto thenBuilder = [&](OpBuilder &builder, Location loc) {
       Operation *read = builder.clone(*maskedOp.getOperation());
       read->setAttr(kMaskedloadNeedsMask, builder.getUnitAttr());
       Value readResult = read->getResult(0);
-      builder.create<scf::YieldOp>(loc, readResult);
+      scf::YieldOp::create(builder, loc, readResult);
     };
 
     auto elseBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, maskedOp,
                                                 /*passthru=*/true);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
 
     auto ifOp =
-        rewriter.create<scf::IfOp>(loc, ifCondition, thenBuilder, elseBuilder);
+        scf::IfOp::create(rewriter, loc, ifCondition, thenBuilder, elseBuilder);
 
     rewriter.replaceOp(maskedOp, ifOp);
 
@@ -187,13 +187,13 @@ struct FullMaskedLoadToConditionalLoad
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
       Value res = createVectorLoadForMaskedLoad(builder, loc, loadOp,
                                                 /*passthru=*/false);
-      rewriter.create<scf::YieldOp>(loc, res);
+      scf::YieldOp::create(rewriter, loc, res);
     };
     auto falseBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<scf::YieldOp>(loc, loadOp.getPassThru());
+      scf::YieldOp::create(rewriter, loc, loadOp.getPassThru());
     };
-    auto ifOp = rewriter.create<scf::IfOp>(loadOp.getLoc(), cond, trueBuilder,
-                                           falseBuilder);
+    auto ifOp = scf::IfOp::create(rewriter, loadOp.getLoc(), cond, trueBuilder,
+                                  falseBuilder);
     rewriter.replaceOp(loadOp, ifOp);
     return success();
   }
@@ -212,11 +212,12 @@ struct FullMaskedStoreToConditionalStore
     Value cond = maybeCond.value();
 
     auto trueBuilder = [&](OpBuilder &builder, Location loc) {
-      rewriter.create<vector::StoreOp>(loc, storeOp.getValueToStore(),
-                                       storeOp.getBase(), storeOp.getIndices());
-      rewriter.create<scf::YieldOp>(loc);
+      vector::StoreOp::create(rewriter, loc, storeOp.getValueToStore(),
+                              storeOp.getBase(), storeOp.getIndices());
+      scf::YieldOp::create(rewriter, loc);
     };
-    auto ifOp = rewriter.create<scf::IfOp>(storeOp.getLoc(), cond, trueBuilder);
+    auto ifOp =
+        scf::IfOp::create(rewriter, storeOp.getLoc(), cond, trueBuilder);
     rewriter.replaceOp(storeOp, ifOp);
     return success();
   }
diff --git a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
index 195f59d625554..f8bab8289cbc6 100644
--- a/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
+++ b/mlir/lib/Dialect/AMDGPU/Transforms/ResolveStridedMetadata.cpp
@@ -37,8 +37,8 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       return rewriter.notifyMatchFailure(metadataOp,
                                          "not a fat raw buffer cast");
     Location loc = castOp.getLoc();
-    auto sourceMetadata = rewriter.create<memref::ExtractStridedMetadataOp>(
-        loc, castOp.getSource());
+    auto sourceMetadata = memref::ExtractStridedMetadataOp::create(
+        rewriter, loc, castOp.getSource());
     SmallVector<Value> results;
     if (metadataOp.getBaseBuffer().use_empty()) {
       results.push_back(nullptr);
@@ -48,13 +48,13 @@ struct ExtractStridedMetadataOnFatRawBufferCastFolder final
       if (baseBufferType == castOp.getResult().getType()) {
         results.push_back(castOp.getResult());
       } else {
-        results.push_back(rewriter.create<memref::ReinterpretCastOp>(
-            loc, baseBufferType, castOp.getResult(), /*offset=*/0,
+        results.push_back(memref::ReinterpretCastOp::create(
+            rewriter, loc, baseBufferType, castOp.getResult(), /*offset=*/0,
             /*sizes=*/ArrayRef<int64_t>{}, /*strides=*/ArrayRef<int64_t>{}));
       }
     }
     if (castOp.getResetOffset())
-      results.push_back(rewriter.create<arith::ConstantIndexOp>(loc, 0));
+      results.push_back(arith::ConstantIndexOp::create(rewriter, loc, 0));
     else
       results.push_back(sourceMetadata.getOffset());
     llvm::append_range(results, sourceMetadata.getSizes());
diff --git a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
index 12b375b373fa9..748ff1edbfeb2 100644
--- a/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
+++ b/mlir/lib/Dialect/AMX/IR/AMXDialect.cpp
@@ -76,8 +76,8 @@ static SmallVector<Value> getTileSizes(Location loc, amx::TileType tType,
   auto mattr = rewriter.getI16IntegerAttr(tType.getDimSize(0));
   auto nattr = rewriter.getI16IntegerAttr(tType.getDimSize(1) * bytes);
   return SmallVector<Value>{
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, mattr),
-      rewriter.create<LLVM::ConstantOp>(loc, llvmInt16Type, nattr)};
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, mattr),
+      LLVM::ConstantOp::create(rewriter, loc, llvmInt16Type, nattr)};
 }
 
 /// Maps the 2-dim memref shape to the 64-bit stride. Note that the buffer
@@ -95,7 +95,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
     // Dynamic stride needs code to compute the stride at runtime.
     MemRefDescriptor memrefDescriptor(base);
     auto attr = rewriter.getI64IntegerAttr(bytes);
-    Value scale = rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr);
+    Value scale = LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr);
     return rewriter
         .create<LLVM::MulOp>(loc, llvmInt64Type, scale,
                              memrefDescriptor.stride(rewriter, loc, preLast))
@@ -103,7 +103,7 @@ static Value getStride(Location loc, MemRefType mType, Value base,
   }
   // Use direct constant for static stride.
   auto attr = rewriter.getI64IntegerAttr(strides[preLast] * bytes);
-  return rewriter.create<LLVM::ConstantOp>(loc, llvmInt64Type, attr)
+  return LLVM::ConstantOp::create(rewriter, loc, llvmInt64Type, attr)
       .getResult();
 }
 
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
index f18cec5a14fae..df39544aeaa09 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineDataCopyGeneration.cpp
@@ -202,7 +202,7 @@ void AffineDataCopyGeneration::runOnBlock(Block *block,
 void AffineDataCopyGeneration::runOnOperation() {
   func::FuncOp f = getOperation();
   OpBuilder topBuilder(f.getBody());
-  zeroIndex = topBuilder.create<arith::ConstantIndexOp>(f.getLoc(), 0);
+  zeroIndex = arith::ConstantIndexOp::create(topBuilder, f.getLoc(), 0);
 
   // Nests that are copy-in's or copy-out's; the root AffineForOps of those
   // nests are stored herein.
diff --git a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
index 5430bdc4ff858..c0d174a04abf9 100644
--- a/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/AffineExpandIndexOps.cpp
@@ -58,8 +58,9 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
       // Note: basis elements and their products are, definitionally,
       // non-negative, so `nuw` is justified.
       if (dynamicPart)
-        dynamicPart = rewriter.create<arith::MulIOp>(
-            loc, dynamicPart, dynamicBasis[dynamicIndex - 1], ovflags);
+        dynamicPart =
+            arith::MulIOp::create(rewriter, loc, dynamicPart,
+                                  dynamicBasis[dynamicIndex - 1], ovflags);
       else
         dynamicPart = dynamicBasis[dynamicIndex - 1];
       --dynamicIndex;
@@ -74,7 +75,7 @@ static SmallVector<Value> computeStrides(Location loc, RewriterBase &rewriter,
           rewriter.createOrFold<arith::ConstantIndexOp>(loc, staticPart);
       if (dynamicPart)
         stride =
-            rewriter.create<arith::MulIOp>(loc, dynamicPart, stride, ovflags);
+            arith::MulIOp::create(rewriter, loc, dynamicPart, stride, ovflags);
       result.push_back(stride);
     }
   }
@@ -106,20 +107,20 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
   Value zero = rewriter.createOrFold<arith::ConstantIndexOp>(loc, 0);
 
   Value initialPart =
-      rewriter.create<arith::FloorDivSIOp>(loc, linearIdx, strides.front());
+      arith::FloorDivSIOp::create(rewriter, loc, linearIdx, strides.front());
   results.push_back(initialPart);
 
   auto emitModTerm = [&](Value stride) -> Value {
-    Value remainder = rewriter.create<arith::RemSIOp>(loc, linearIdx, stride);
-    Value remainderNegative = rewriter.create<arith::CmpIOp>(
-        loc, arith::CmpIPredicate::slt, remainder, zero);
+    Value remainder = arith::RemSIOp::create(rewriter, loc, linearIdx, stride);
+    Value remainderNegative = arith::CmpIOp::create(
+        rewriter, loc, arith::CmpIPredicate::slt, remainder, zero);
     // If the correction is relevant, this term is <= stride, which is known
     // to be positive in `index`. Otherwise, while 2 * stride might overflow,
     // this branch won't be taken, so the risk of `poison` is fine.
-    Value corrected = rewriter.create<arith::AddIOp>(
-        loc, remainder, stride, arith::IntegerOverflowFlags::nsw);
-    Value mod = rewriter.create<arith::SelectOp>(loc, remainderNegative,
-                                                 corrected, remainder);
+    Value corrected = arith::AddIOp::create(rewriter, loc, remainder, stride,
+                                            arith::IntegerOverflowFlags::nsw);
+    Value mod = arith::SelectOp::create(rewriter, loc, remainderNegative,
+                                        corrected, remainder);
     return mod;
   };
 
@@ -131,7 +132,7 @@ affine::lowerAffineDelinearizeIndexOp(RewriterBase &rewriter,
     // We know both inputs are positive, so floorDiv == div.
     // This could potentially be a divui, but it's not clear if that would
     // cause issues.
-    Value divided = rewriter.create<arith::DivSIOp>(loc, modulus, nextStride);
+    Value divided = arith::DivSIOp::create(rewriter, loc, modulus, nextStride);
     results.push_back(divided);
   }
 
@@ -167,8 +168,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   // our hands on an `OpOperand&` for the loop invariant counting function.
   for (auto [stride, idxOp] :
        llvm::zip_equal(strides, llvm::drop_end(op.getMultiIndexMutable()))) {
-    Value scaledIdx = rewriter.create<arith::MulIOp>(
-        loc, idxOp.get(), stride, arith::IntegerOverflowFlags::nsw);
+    Value scaledIdx = arith::MulIOp::create(rewriter, loc, idxOp.get(), stride,
+                                            arith::IntegerOverflowFlags::nsw);
     int64_t numHoistableLoops = numEnclosingInvariantLoops(idxOp);
     scaledValues.emplace_back(scaledIdx, numHoistableLoops);
   }
@@ -184,8 +185,8 @@ LogicalResult affine::lowerAffineLinearizeIndexOp(RewriterBase &rewriter,
   Value result = scaledValues.front().first;
   for (auto [scaledValue, numHoistableLoops] : llvm::drop_begin(scaledValues)) {
     std::ignore = numHoistableLoops;
-    result = rewriter.create<arith::AddIOp>(loc, result, scaledValue,
-                                            arith::IntegerOverflowFlags::nsw);
+    result = arith::AddIOp::create(rewriter, loc, result, scaledValue,
+                                   arith::IntegerOverflowFlags::nsw);
   }
   rewriter.replaceOp(op, result);
   return success();
diff --git a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
index 4fd0cf9b3cd25..3c00b323473d2 100644
--- a/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
+++ b/mlir/lib/Dialect/Affine/Transforms/DecomposeAffineOps.cpp
@@ -88,8 +88,8 @@ static AffineApplyOp createSubApply(RewriterBase &rewriter,
   auto rhsMap = AffineMap::get(m.getNumDims(), m.getNumSymbols(), expr, ctx);
   SmallVector<Value> rhsOperands = originalOp->getOperands();
   canonicalizeMapAndOperands(&rhsMap, &rhsOperands);
-  return rewriter.create<AffineApplyOp>(originalOp.getLoc(), rhsMap,
-                                        rhsOperands);
+  return AffineApplyOp::create(rewriter, originalOp.getLoc(), rhsMap,
+                               rhsOperands);
 }
 
 FailureOr<AffineApplyOp> mlir::affine::decompose(RewriterBase &rewriter,
@@ -160,8 +160,8 @@ FailureOr<AffineApplyO...
[truncated]

[mlir][NFC] update mlir/Dialect create APIs (14/n)

e4a90e2

See llvm#147168 for more info.

makslevental force-pushed the makslevental/update-create-14n branch from 0edc7c6 to e4a90e2 Compare July 21, 2025 22:26

makslevental marked this pull request as ready for review July 21, 2025 22:43

makslevental requested review from aartbik, dcaballe and matthias-springer as code owners July 21, 2025 22:43

llvmbot added backend:AMDGPU mlir:gpu mlir:affine mlir mlir:amdgpu mlir:amx mlir:arith labels Jul 21, 2025

makslevental requested review from ftynse, Hardcode84 and arnab-polymage July 21, 2025 22:45

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[mlir][NFC] update `mlir/Dialect` create APIs (14/n) #149920

[mlir][NFC] update `mlir/Dialect` create APIs (14/n) #149920

Uh oh!

makslevental commented Jul 21, 2025

Uh oh!

github-actions bot commented Jul 21, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jul 21, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jul 21, 2025

Uh oh!

llvmbot commented Jul 21, 2025

Uh oh!

Uh oh!

[mlir][NFC] update mlir/Dialect create APIs (14/n) #149920

Are you sure you want to change the base?

[mlir][NFC] update mlir/Dialect create APIs (14/n) #149920

Uh oh!

Conversation

makslevental commented Jul 21, 2025

Uh oh!

github-actions bot commented Jul 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jul 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jul 21, 2025

Uh oh!

llvmbot commented Jul 21, 2025

Uh oh!

Uh oh!

[mlir][NFC] update `mlir/Dialect` create APIs (14/n) #149920

[mlir][NFC] update `mlir/Dialect` create APIs (14/n) #149920

github-actions bot commented Jul 21, 2025 •

edited

Loading

llvmbot commented Jul 21, 2025 •

edited

Loading