Migrate internal changes (#12)

- Fix most non-functional style casts in order to conform with upstream
  deprecations
- Fix minor build/documentation issues
- add psutil and build packages

Co-authored-by: Sagar Shelke

Signed-off-by: Christopher Bate <[email protected]>

mlir-tensorrt/README.md

@@ -50,7 +50,7 @@ git apply ../build_tools/llvm-project.patch
 
 # Do the build
 cd ..
-./build_tools/scripts/build_mlir.sh llvm-project build/llvm
+./build_tools/scripts/build_mlir.sh llvm-project build/llvm-project
 ```
 
 2. Build the project and run all tests

mlir-tensorrt/build_tools/scripts/cicd_build.sh

@@ -0,0 +1,32 @@
+#!/usr/bin/env bash
+set -ex
+set -o pipefail
+
+REPO_ROOT=$(pwd)
+BUILD_DIR="${BUILD_DIR:=${REPO_ROOT}/build/mlir-tensorrt}"
+
+ENABLE_NCCL=${ENABLE_NCCL:OFF}
+RUN_LONG_TESTS=${RUN_LONG_TESTS:-False}
+LLVM_LIT_ARGS=${LLVM_LIT_ARGS:-"-v --xunit-xml-output ${BUILD_DIR}/test-results.xml --timeout=1200 --time-tests -Drun_long_tests=${RUN_LONG_TESTS}"}
+DOWNLOAD_TENSORRT_VERSION=${DOWNLOAD_TENSORRT_VERSION:-10.0.0.6}
+ENABLE_ASAN=${ENABLE_ASAN:-OFF}
+
+echo "Using DOWNLOAD_TENSORRT_VERSION=${DOWNLOAD_TENSORRT_VERSION}"
+echo "Using LLVM_LIT_ARGS=${LLVM_LIT_ARGS}"
+
+cmake -GNinja -B "${BUILD_DIR}" -S "${REPO_ROOT}" \
+  -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ \
+  -DMLIR_TRT_USE_LINKER=lld -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+  -DMLIR_TRT_PACKAGE_CACHE_DIR=$PWD/.cache.cpm \
+  -DMLIR_TRT_ENABLE_PYTHON=ON \
+  -DMLIR_TRT_ENABLE_NCCL=${ENABLE_NCCL} \
+  -DMLIR_TRT_DOWNLOAD_TENSORRT_VERSION="$DOWNLOAD_TENSORRT_VERSION" \
+  -DLLVM_LIT_ARGS="${LLVM_LIT_ARGS}" \
+  -DENABLE_ASAN="${ENABLE_ASAN}" \
+  -DMLIR_DIR=${REPO_ROOT}/build/llvm-project/lib/cmake/mlir \
+  -DCMAKE_PLATFORM_NO_VERSIONED_SONAME=ON
+
+echo "==== Running Build ==="
+ninja -C ${BUILD_DIR} -k 0 check-mlir-executor
+ninja -C ${BUILD_DIR} -k 0 check-mlir-tensorrt
+ninja -C ${BUILD_DIR} -k 0 check-mlir-tensorrt-dialect

mlir-tensorrt/compiler/lib/Conversion/CUDAToExecutor/CUDAToExecutor.cpp

@@ -124,10 +124,10 @@ struct CudaBlasRunGemmOpConverter
     SmallVector<Value> newOperands = {adaptor.getHandle(), adaptor.getStream()};
     newOperands.push_back(adaptor.getAlgo());
     auto createMemRefAndExractPtr = [&](Value oldVal, Value newVal) {
-      auto memrefType = oldVal.getType().cast<MemRefType>();
+      auto memrefType = cast<MemRefType>(oldVal.getType());
       if (!memrefType)
         return failure();
-      assert(newVal.getType().isa<TableType>());
+      assert(isa<TableType>(newVal.getType()));
       executor::MemRefDescriptor memref(newVal, memrefType);
       newOperands.push_back(memref.alignedPtr(b));
       return success();

mlir-tensorrt/compiler/lib/Conversion/PlanToExecutor/CMakeLists.txt

@@ -5,4 +5,5 @@ add_mlir_tensorrt_library(MLIRTensorRTPlanToExecutor
   MLIRTensorRTExecutorDialect
   MLIRTensorRTPlanDialect
   MLIRTransforms
+  MLIRSCFTransforms
 )

mlir-tensorrt/compiler/lib/Conversion/PlanToExecutor/PlanToExecutor.cpp

@@ -98,9 +98,8 @@ struct ConstantOpConverter : public OpConversionPattern<arith::ConstantOp> {
   LogicalResult
   matchAndRewrite(arith::ConstantOp op, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
-    auto resultType = getTypeConverter()
-                          ->convertType(op.getType())
-                          .dyn_cast_or_null<RankedTensorType>();
+    auto resultType = dyn_cast_or_null<RankedTensorType>(
+        getTypeConverter()->convertType(op.getType()));
     if (!resultType)
       return failure();
 

mlir-tensorrt/compiler/lib/Conversion/StablehloScalarToArith/StablehloScalarToArith.cpp

@@ -119,7 +119,7 @@ struct StablehloRewriteConcat
   matchAndRewrite(stablehlo::ConcatenateOp op, OpAdaptor adaptor,
                   OneToNPatternRewriter &rewriter) const override {
     if (!llvm::all_of(op->getOperandTypes(), [](Type t) {
-          return t.cast<RankedTensorType>().getRank() == 1;
+          return cast<RankedTensorType>(t).getRank() == 1;
         }))
       return failure();
     rewriter.replaceOp(op, adaptor.getFlatOperands(),
@@ -133,13 +133,13 @@ struct StablehloRewriteConcat
 /// scalar `type`.
 static Attribute getScalarValue(RewriterBase &rewriter, Type type,
                                 int64_t idx) {
-  if (type.isa<FloatType>())
+  if (isa<FloatType>(type))
     return rewriter.getFloatAttr(type, static_cast<double>(idx));
-  if (type.isa<IndexType>())
+  if (isa<IndexType>(type))
     return rewriter.getIndexAttr(idx);
-  if (auto integerType = type.dyn_cast<IntegerType>())
+  if (auto integerType = dyn_cast<IntegerType>(type))
     return rewriter.getIntegerAttr(
-        type, APInt(type.cast<IntegerType>().getWidth(), idx));
+        type, APInt(cast<IntegerType>(type).getWidth(), idx));
   return {};
 }
 

mlir-tensorrt/compiler/lib/Conversion/StablehloToScf/StablehloToScf.cpp

@@ -51,7 +51,7 @@ static void inlineStablehloRegionIntoSCFRegion(PatternRewriter &rewriter,
 static Value extractScalarFromTensorValue(OpBuilder &b, Value tensor) {
   Location loc = tensor.getLoc();
   // If ranked tensor, first collapse shape.
-  if (tensor.getType().cast<RankedTensorType>().getRank() != 0)
+  if (cast<RankedTensorType>(tensor.getType()).getRank() != 0)
     tensor = b.create<tensor::CollapseShapeOp>(
         loc, tensor, SmallVector<ReassociationIndices>());
 
@@ -129,10 +129,10 @@ static scf::IfOp createNestedCases(int currentIdx, stablehlo::CaseOp op,
 
   // Determine if the current index matches the case index.
   auto scalarType = idxValue.getType();
-  auto shapedType = scalarType.cast<ShapedType>();
+  auto shapedType = cast<ShapedType>(scalarType);
   auto constAttr = DenseElementsAttr::get(
       shapedType,
-      {outerBuilder.getI32IntegerAttr(currentIdx).cast<mlir::Attribute>()});
+      {cast<mlir::Attribute>(outerBuilder.getI32IntegerAttr(currentIdx))});
   Value currentIdxVal = outerBuilder.create<stablehlo::ConstantOp>(
       loc, idxValue.getType(), constAttr);
 

mlir-tensorrt/compiler/lib/Conversion/StablehloToTensorRT/ChloToTensorRT.cpp

@@ -41,7 +41,7 @@ struct ConvertChloErfToTensorRT
                   ConversionPatternRewriter &rewriter) const override {
     Location loc = op->getLoc();
     auto operand = adaptor.getOperand();
-    auto operandType = operand.getType().cast<RankedTensorType>();
+    auto operandType = cast<RankedTensorType>(operand.getType());
     Type resultType = typeConverter->convertType(op.getType());
     if (!resultType)
       return failure();
@@ -74,7 +74,7 @@ struct ConvertChloTopKOpToTensorRT
   matchAndRewrite(chlo::TopKOp op, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
     auto operand = adaptor.getOperand();
-    RankedTensorType operandType = operand.getType().cast<RankedTensorType>();
+    RankedTensorType operandType = cast<RankedTensorType>(operand.getType());
 
     int64_t rank = operandType.getRank();
     uint64_t axis = static_cast<uint64_t>(rank) - 1;

mlir-tensorrt/compiler/lib/Conversion/StablehloToTensorRT/ControlFlowOps.cpp

@@ -118,7 +118,7 @@ struct ConvertCaseOp : public ConvertHloOpToTensorRTPattern<stablehlo::CaseOp> {
       if (!isa_and_nonnull<tensorrt::IdentityOp, stablehlo::ConvertOp>(op))
         return false;
       RankedTensorType producerType =
-          op->getOperand(0).getType().cast<RankedTensorType>();
+          cast<RankedTensorType>(op->getOperand(0).getType());
       return isa_and_nonnull<tensorrt::IdentityOp, stablehlo::ConvertOp>(op) &&
              producerType.getElementType().isInteger(1) &&
              producerType.getNumElements() == 1;