Drop revert on LLVM c0b42ec05344707d94805ec795a7bc8d33a09594 (#4368)

This is a follow-up to #4358.

Remove the revert by fixing all deprecated usages of
`OpBuilder::create`, as requested in
llvm/llvm-project@c0b42ec

---------

Signed-off-by: Yu-Zhewen <[email protected]>

Author: Yu-Zhewen

.gitmodules

@@ -1,6 +1,6 @@
 [submodule "externals/llvm-project"]
 	path = externals/llvm-project
-	url = https://github.com/iree-org/llvm-project.git
+	url = https://github.com/llvm/llvm-project.git
 [submodule "externals/stablehlo"]
 	path = externals/stablehlo
 	url = https://github.com/openxla/stablehlo.git

externals/llvm-project

@@ -44,8 +44,8 @@ Torch::ValueTensorType getQTorchTypeFromTorchIntType(Type ty);
 template <typename T>
 Value getItemOp(OpBinder binder, ConversionPatternRewriter &rewriter,
                 Value &ofItem) {
-  return rewriter.create<Torch::AtenItemOp>(binder.getLoc(),
-                                            rewriter.getType<T>(), ofItem);
+  return Torch::AtenItemOp::create(rewriter, binder.getLoc(),
+                                   rewriter.getType<T>(), ofItem);
 }
 
 LogicalResult OnnxLstmExpander(OpBinder binder,

include/torch-mlir/Conversion/TorchOnnxToTorch/Utils.h

@@ -40,8 +40,8 @@ static LogicalResult getOrCreateGlobalVariableForSeed(OpBuilder &b,
   }
 
   b.setInsertionPointToStart(module.getBody());
-  b.create<ml_program::GlobalOp>(
-      UnknownLoc::get(b.getContext()),
+  ml_program::GlobalOp::create(
+      b, UnknownLoc::get(b.getContext()),
       /*sym_name=*/getSeedGobalVarName(),
       /*type=*/tensorType,
       /*is_mutable=*/true,
@@ -71,25 +71,25 @@ class ConvertGetNextSeedOp : public OpConversionPattern<GetNextSeedOp> {
     // Refer to https://en.wikipedia.org/wiki/Linear_congruential_generator.
     // Get the current seed value.
     auto tensorType = RankedTensorType::get({}, rewriter.getI64Type());
-    Value globalVar = rewriter.create<ml_program::GlobalLoadOp>(
-        loc, tensorType,
+    Value globalVar = ml_program::GlobalLoadOp::create(
+        rewriter, loc, tensorType,
         SymbolRefAttr::get(op->getContext(), getSeedGobalVarName()));
-    Value currentSeed = rewriter.create<tensor::ExtractOp>(loc, globalVar);
+    Value currentSeed = tensor::ExtractOp::create(rewriter, loc, globalVar);
 
     // The value of multiplier and incrementStep are referenced from
     // https://en.wikipedia.org/wiki/Linear_congruential_generator for 2^64.
-    Value multiplier = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI64IntegerAttr(6364136223846793005));
-    Value incrementStep = rewriter.create<arith::ConstantOp>(
-        loc, rewriter.getI64IntegerAttr(1442695040888963407));
+    Value multiplier = arith::ConstantOp::create(
+        rewriter, loc, rewriter.getI64IntegerAttr(6364136223846793005));
+    Value incrementStep = arith::ConstantOp::create(
+        rewriter, loc, rewriter.getI64IntegerAttr(1442695040888963407));
     // temp = multiplier * currentSeed + incrementStep
-    Value mul = rewriter.create<arith::MulIOp>(loc, currentSeed, multiplier);
-    Value seed = rewriter.create<arith::AddIOp>(loc, mul, incrementStep);
+    Value mul = arith::MulIOp::create(rewriter, loc, currentSeed, multiplier);
+    Value seed = arith::AddIOp::create(rewriter, loc, mul, incrementStep);
     globalVar =
-        rewriter.create<tensor::InsertOp>(loc, seed, globalVar, ValueRange());
-    rewriter.create<ml_program::GlobalStoreOp>(
-        loc, SymbolRefAttr::get(op->getContext(), getSeedGobalVarName()),
-        globalVar);
+        tensor::InsertOp::create(rewriter, loc, seed, globalVar, ValueRange());
+    ml_program::GlobalStoreOp::create(
+        rewriter, loc,
+        SymbolRefAttr::get(op->getContext(), getSeedGobalVarName()), globalVar);
     rewriter.replaceOp(op, seed);
     return success();
   }

lib/Conversion/TorchConversionToMLProgram/TorchConversionToMLProgram.cpp

@@ -20,11 +20,11 @@ Value mlir::torch::onnx_c::createConstantIntList(
     ArrayRef<int64_t> cstInput) {
   SmallVector<Value> cstValue;
   for (int64_t i : cstInput) {
-    cstValue.push_back(rewriter.create<Torch::ConstantIntOp>(
-        binder.getLoc(), rewriter.getI64IntegerAttr(i)));
+    cstValue.push_back(Torch::ConstantIntOp::create(
+        rewriter, binder.getLoc(), rewriter.getI64IntegerAttr(i)));
   }
-  return rewriter.create<Torch::PrimListConstructOp>(
-      binder.getLoc(),
+  return Torch::PrimListConstructOp::create(
+      rewriter, binder.getLoc(),
       Torch::ListType::get(Torch::IntType::get(binder.op->getContext())),
       cstValue);
 }
@@ -109,12 +109,12 @@ LogicalResult mlir::torch::onnx_c::createTorchTransposeOp(
   if (failed(getTransposedType(cast<Torch::BaseTensorType>(input.getType()),
                                dimA, dimB, transposedType)))
     return failure();
-  Value cstDimA = rewriter.create<Torch::ConstantIntOp>(
-      loc, rewriter.getI64IntegerAttr(dimA));
-  Value cstDimB = rewriter.create<Torch::ConstantIntOp>(
-      loc, rewriter.getI64IntegerAttr(dimB));
-  transposed = rewriter.create<Torch::AtenTransposeIntOp>(
-      loc, transposedType, input, cstDimA, cstDimB);
+  Value cstDimA = Torch::ConstantIntOp::create(
+      rewriter, loc, rewriter.getI64IntegerAttr(dimA));
+  Value cstDimB = Torch::ConstantIntOp::create(
+      rewriter, loc, rewriter.getI64IntegerAttr(dimB));
+  transposed = Torch::AtenTransposeIntOp::create(rewriter, loc, transposedType,
+                                                 input, cstDimA, cstDimB);
   return success();
 }
 
@@ -127,19 +127,19 @@ LogicalResult mlir::torch::onnx_c::createTorchPermuteOp(
                                  permuteDims, permutedType)))
     return failure();
   Value permuteDimsList = createConstantIntList(binder, rewriter, permuteDims);
-  permuted = rewriter.create<Torch::AtenPermuteOp>(loc, permutedType, input,
-                                                   permuteDimsList);
+  permuted = Torch::AtenPermuteOp::create(rewriter, loc, permutedType, input,
+                                          permuteDimsList);
   return success();
 }
 
 Value mlir::torch::onnx_c::createActivationByName(ImplicitLocOpBuilder &b,
                                                   StringRef name, Value input) {
   if (name == "Sigmoid")
-    return b.create<Torch::AtenSigmoidOp>(input.getType(), input);
+    return Torch::AtenSigmoidOp::create(b, input.getType(), input);
   if (name == "Tanh")
-    return b.create<Torch::AtenTanhOp>(input.getType(), input);
+    return Torch::AtenTanhOp::create(b, input.getType(), input);
   if (name == "Relu")
-    return b.create<Torch::AtenReluOp>(input.getType(), input);
+    return Torch::AtenReluOp::create(b, input.getType(), input);
   llvm_unreachable("Unsupported activation function");
 }
 
@@ -158,23 +158,23 @@ LogicalResult mlir::torch::onnx_c::extractPerTensorQuantizationArguments(
   if (!check(inScale) || !check(inZeroPoint))
     return failure();
 
-  Value emptyList = rewriter.create<Torch::PrimListConstructOp>(
-      loc,
+  Value emptyList = Torch::PrimListConstructOp::create(
+      rewriter, loc,
       rewriter.getType<Torch::ListType>(rewriter.getType<Torch::IntType>()),
       ValueRange{});
   auto extract = [&rewriter, &loc, &emptyList](Value v) {
     auto vTy = cast<Torch::ValueTensorType>(v.getType());
     if (!vTy.getSizes().empty()) {
       vTy = rewriter.getType<Torch::ValueTensorType>(ArrayRef<int64_t>({}),
                                                      vTy.getOptionalDtype());
-      v = rewriter.create<Torch::AtenReshapeOp>(loc, vTy, v, emptyList);
+      v = Torch::AtenReshapeOp::create(rewriter, loc, vTy, v, emptyList);
     }
 
     Type extractTy = rewriter.getType<Torch::FloatType>();
     if (isa<IntegerType>(vTy.getDtype()))
       extractTy = rewriter.getType<Torch::IntType>();
 
-    return rewriter.create<Torch::AtenItemOp>(loc, extractTy, v);
+    return Torch::AtenItemOp::create(rewriter, loc, extractTy, v);
   };
 
   outScale = extract(inScale);
@@ -191,14 +191,13 @@ LogicalResult mlir::torch::onnx_c::createDequantizeTensor(
     return failure();
 
   Torch::ValueTensorType makeTensorTy = getQTorchTypeFromTorchIntType(inputTy);
-  Value quantizedInput =
-      rewriter.create<Torch::Aten_MakePerTensorQuantizedTensorOp>(
-          loc, makeTensorTy, input, scale, zeroPoint);
+  Value quantizedInput = Torch::Aten_MakePerTensorQuantizedTensorOp::create(
+      rewriter, loc, makeTensorTy, input, scale, zeroPoint);
 
   Torch::ValueTensorType resultTy = rewriter.getType<Torch::ValueTensorType>(
       inputTy.getSizes(), rewriter.getF32Type());
-  output = rewriter.create<Torch::AtenDequantizeSelfOp>(loc, resultTy,
-                                                        quantizedInput);
+  output = Torch::AtenDequantizeSelfOp::create(rewriter, loc, resultTy,
+                                               quantizedInput);
 
   return success();
 }