[Mlir-commits] [mlir] update tdesc_attr (PR #109144)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Wed Sep 18 07:03:22 PDT 2024
github-actions[bot] wrote:
<!--LLVM CODE FORMAT COMMENT: {clang-format}-->
:warning: C/C++ code formatter, clang-format found issues in your code. :warning:
<details>
<summary>
You can test this locally with the following command:
</summary>
``````````bash
git-clang-format --diff 08a61eb01172054fc5f8c78ff527f01d9768569b 24adc84d0a42f5e7712291ef3a886fa5de044f0f --extensions cpp -- mlir/lib/Dialect/XeGPU/IR/XeGPUDialect.cpp mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
``````````
</details>
<details>
<summary>
View the diff from clang-format here.
</summary>
``````````diff
diff --git a/mlir/lib/Dialect/XeGPU/IR/XeGPUDialect.cpp b/mlir/lib/Dialect/XeGPU/IR/XeGPUDialect.cpp
index 555c232ff1..4573045515 100644
--- a/mlir/lib/Dialect/XeGPU/IR/XeGPUDialect.cpp
+++ b/mlir/lib/Dialect/XeGPU/IR/XeGPUDialect.cpp
@@ -33,8 +33,9 @@ void XeGPUDialect::initialize() {
// XeGPU_BlockTensorDescAttr
//===----------------------------------------------------------------------===//
BlockTensorDescAttr BlockTensorDescAttr::get(mlir::MLIRContext *context,
- xegpu::MemoryScope memory_scope,
- int array_length, bool boundary_check) {
+ xegpu::MemoryScope memory_scope,
+ int array_length,
+ bool boundary_check) {
auto scopeAttr = MemoryScopeAttr::get(context, memory_scope);
auto lengthAttr =
IntegerAttr::get(IntegerType::get(context, 64), array_length);
@@ -45,9 +46,9 @@ BlockTensorDescAttr BlockTensorDescAttr::get(mlir::MLIRContext *context,
//===----------------------------------------------------------------------===//
// XeGPU_ScatterTensorDescAttr
//===----------------------------------------------------------------------===//
-ScatterTensorDescAttr ScatterTensorDescAttr::get(mlir::MLIRContext *context,
- xegpu::MemoryScope memory_scope,
- int chunk_size) {
+ScatterTensorDescAttr
+ScatterTensorDescAttr::get(mlir::MLIRContext *context,
+ xegpu::MemoryScope memory_scope, int chunk_size) {
auto scopeAttr = MemoryScopeAttr::get(context, memory_scope);
auto chunkSizeAttr =
IntegerAttr::get(IntegerType::get(context, 64), chunk_size);
@@ -120,9 +121,11 @@ void TensorDescType::print(::mlir::AsmPrinter &printer) const {
TensorDescType TensorDescType::get(llvm::ArrayRef<int64_t> shape,
mlir::Type elementType, int array_length,
- bool boundary_check, MemoryScope memory_scope) {
+ bool boundary_check,
+ MemoryScope memory_scope) {
auto context = elementType.getContext();
- auto attr = BlockTensorDescAttr::get(context, memory_scope, array_length, boundary_check);
+ auto attr = BlockTensorDescAttr::get(context, memory_scope, array_length,
+ boundary_check);
return Base::get(context, shape, elementType, attr);
}
diff --git a/mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp b/mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
index 0da38df90f..a4e9bbe58c 100644
--- a/mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
+++ b/mlir/lib/Dialect/XeGPU/IR/XeGPUOps.cpp
@@ -307,9 +307,11 @@ LogicalResult CreateDescOp::verify() {
auto chunkSize = tdescTy.getChunkSize();
// check chunk_size
- llvm::SmallVector<int64_t> supportedChunkSizes = {1, 2, 3, 4, 8, 16, 32, 64, 128, 256};
+ llvm::SmallVector<int64_t> supportedChunkSizes = {1, 2, 3, 4, 8,
+ 16, 32, 64, 128, 256};
if (!llvm::is_contained(supportedChunkSizes, chunkSize))
- return emitOpError("Invalid chunk_size. Supported values are 1, 2, 3, 4, 8, 16, 32, 64, 128, or 256.");
+ return emitOpError("Invalid chunk_size. Supported values are 1, 2, 3, 4, "
+ "8, 16, 32, 64, 128, or 256.");
// check total size
auto elemBits = tdescTy.getElementType().getIntOrFloatBitWidth();
@@ -318,13 +320,16 @@ LogicalResult CreateDescOp::verify() {
// For 8-bit and 16-bit data, the hardware only supports chunk size of 1.
// For 32-bit data, the hardware can support larger larger chunk size. So
// we can bitcast 8-bit/16-bit data to 32-bit data for better performance.
- // But this requires the total size is 32 bit aligned to make the optimization work.
- return emitOpError("access size (chunk_size * sizeof(elemTy)) should be 32-bit aligned.");
+ // But this requires the total size is 32 bit aligned to make the
+ // optimization work.
+ return emitOpError(
+ "access size (chunk_size * sizeof(elemTy)) should be 32-bit aligned.");
}
auto lscConstraints = 512 * 8; // each access is upto 512 bytes.
if (elemBits * tdescTy.getNumElements() > lscConstraints)
- return emitOpError("total access size (simd_lanes * chunk_size * sizeof(elemTy)) is upto 512 bytes.");
+ return emitOpError("total access size (simd_lanes * chunk_size * "
+ "sizeof(elemTy)) is upto 512 bytes.");
SmallVector<int64_t> shape({(int64_t)getNumOffsets()});
if (chunkSize != 1)
@@ -397,7 +402,6 @@ LogicalResult LoadGatherOp::verify() {
transpose({1, 0}, tdescShape);
}
-
if (valueShape != tdescShape)
return emitOpError("Unexpected result shape")
<< "(Expected shape: " << makeString(tdescShape)
``````````
</details>
https://github.com/llvm/llvm-project/pull/109144
More information about the Mlir-commits
mailing list