Fixes for llvm/llvm-project#144636

Author: umangyadav

mlir/lib/Conversion/RockToGPU/RockToGPU.cpp

@@ -317,7 +317,7 @@ void LowerRockOpsToGPUPass::runOnOperation() {
             auto blockVal = b.create<arith::ConstantIndexOp>(loc, blockSize);
             auto cst1 = b.create<arith::ConstantIndexOp>(loc, 1);
             auto dynamicSharedMemSize =
-                b.create<arith::ConstantIntOp>(loc, 0, b.getI32Type());
+                b.create<arith::ConstantIntOp>(loc, b.getI32Type(), 0);
             gpu::KernelDim3 gridDims{gridVal, cst1, cst1};
             gpu::KernelDim3 blockDims{blockVal, cst1, cst1};
             b.create<gpu::LaunchFuncOp>(loc, gpuFunc, gridDims, blockDims,

mlir/lib/Dialect/Rock/Transforms/ConvToGemm.cpp

@@ -353,7 +353,7 @@ struct ZeroInitKernelRewritePattern final
     Value zeroIndex = b.createOrFold<arith::ConstantIndexOp>(loc, 0);
     Value memref = makeGpuAllocContaining(b, initOp);
     Value trueOp =
-        b.createOrFold<arith::ConstantIntOp>(loc, true, b.getI1Type());
+        b.createOrFold<arith::ConstantIntOp>(loc, b.getI1Type(), true);
     GemmFeatures features = op.getFeatures();
 
     auto loopBody = [&memref, &initVectorLen, &trueOp, &features, &zeroIndex,
@@ -397,7 +397,7 @@ struct ConvertingCopyKernelRewritePattern final
 
     Type loadType = vectorTypeOrSelf(inputDataType, conversionVectorLen);
     Type storeType = vectorTypeOrSelf(outputDataType, conversionVectorLen);
-    Value trueOp = b.create<arith::ConstantIntOp>(loc, true, b.getI1Type());
+    Value trueOp = b.create<arith::ConstantIntOp>(loc, b.getI1Type(), true);
     GemmFeatures features = op.getFeatures();
     bool needs64BitIdx =
         is4GBMemoryType(input.getType()) || is4GBMemoryType(output.getType());

mlir/lib/Dialect/Rock/Transforms/SugarToLoops.cpp

@@ -199,7 +199,7 @@ struct TransformingForRewritePattern
       if (!useDiffs || transforms.empty()) {
         AffineResults computed;
         Value isValid =
-            b.create<arith::ConstantIntOp>(loc, true, b.getI1Type());
+            b.create<arith::ConstantIntOp>(loc, b.getI1Type(), true);
         // Start by offsetting the upper inputs.
         for (auto p : llvm::zip(op.getUpperInits(i), ivs)) {
           computed.push_back(
@@ -227,7 +227,7 @@ struct TransformingForRewritePattern
       } else { // index diff maps
         IndexDiffUpdateOp lastDiff;
         Value isValid =
-            b.create<arith::ConstantIntOp>(loc, true, b.getI1Type());
+            b.create<arith::ConstantIntOp>(loc, b.getI1Type(), true);
         for (const auto &[t, lowerInit] : llvm::zip(
                  transforms.getAsRange<TransformMapAttr>(), lowerInits[i])) {
           if (!lastDiff)

mlir/lib/Dialect/Rock/Transforms/ThreadwiseGemmLowering.cpp

@@ -572,7 +572,8 @@ LogicalResult ThreadwiseReadIntoRewritePattern::matchAndRewrite(
       op.getValidityRecord() && !op.getValidityRecord().use_empty();
   Value validityInit = nullptr;
   if (recordsValidity) {
-    Value trueConst = b.createOrFold<arith::ConstantIntOp>(loc, true, 1);
+    Value trueConst =
+        b.createOrFold<arith::ConstantIntOp>(loc, b.getI1Type(), true);
     validityInit = b.create<vector::SplatOp>(
         loc, op.getValidityRecord().getType(), trueConst);
   }

mlir/lib/Dialect/Rock/utility/transformMapUtils.cpp

@@ -1075,7 +1075,7 @@ Value mlir::rock::updateValidityAfter(OpBuilder &b, Location loc,
                                       TransformMapAttr map,
                                       ValueRange outputs) {
   Value isValid =
-      b.createOrFold<arith::ConstantIntOp>(loc, true, b.getI1Type());
+      b.createOrFold<arith::ConstantIntOp>(loc, b.getI1Type(), true);
   ArrayRef<int64_t> lowerBounds = map.getLowerBounds();
 
   // unsigned < catches both negatives (as all negatives are > the bound)

mlir/tools/rocmlir-gen/rocmlir-gen.cpp

@@ -1361,8 +1361,8 @@ static func::FuncOp createGPUWrapper(ModuleOp module,
   // Emit device selection
   if (deviceNum.getNumOccurrences() > 0)
     b.create<gpu::SetDefaultDeviceOp>(
-        loc, b.create<arith::ConstantIntOp>(loc, deviceNum.getValue(),
-                                            b.getIntegerType(32)));
+        loc, b.create<arith::ConstantIntOp>(loc, b.getIntegerType(32),
+                                            deviceNum.getValue()));
 
   SmallVector<Value, 4> cpuMem;
   SmallVector<Value, 4> gpuMem;
@@ -1569,7 +1569,7 @@ static LogicalResult populateRandomTensorFillLogic(OpBuilder &b, Location loc,
     if (i16vals.find(v) == i16vals.end()) {
       auto i16Type = b.getIntegerType(16);
       i16vals.try_emplace(
-          v, b.createOrFold<arith::ConstantIntOp>(loc, v, i16Type));
+          v, b.createOrFold<arith::ConstantIntOp>(loc, i16Type, v));
     }
     return i16vals[v];
   };
@@ -3924,7 +3924,7 @@ static func::FuncOp createVerifierFunc(ModuleOp module, const KernelIF &kernel,
 
   auto getF32Val = [&](float val) -> Value {
     llvm::APFloat apVal(val);
-    return b.create<arith::ConstantFloatOp>(loc, apVal, floatType);
+    return b.create<arith::ConstantFloatOp>(loc, floatType, apVal);
   };
   // Thresholds for different metrics
   // RMS: 0.00003f for all data types
@@ -4386,7 +4386,7 @@ static LogicalResult populateHostHarnessLogic(
   if (isRandom) {
     auto seedFunc = makeFuncDecl(module, "seedRandomValues", {b.getI32Type()});
     int seed = getRandomSeed();
-    Value seedConst = b.create<arith::ConstantIntOp>(loc, seed, b.getI32Type());
+    Value seedConst = b.create<arith::ConstantIntOp>(loc, b.getI32Type(), seed);
     b.create<func::CallOp>(loc, seedFunc, seedConst);
   }
 
@@ -4458,7 +4458,7 @@ static LogicalResult populateHostHarnessLogic(
       for (auto pair : llvm::enumerate(currentSeqLen)) {
         Value index = b.create<arith::ConstantIndexOp>(loc, pair.index());
         Value value =
-            b.create<arith::ConstantIntOp>(loc, pair.value(), b.getI32Type());
+            b.create<arith::ConstantIntOp>(loc, b.getI32Type(), pair.value());
         b.create<memref::StoreOp>(loc, value, lvar, ValueRange{index});
       }
     } else if (!isRandom) {