[Mlir-commits] [mlir] [mlir][GPU][NFC] adds TODO messages for amdgpu-specific pattern relocation (PR #165812)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Thu Oct 30 18:16:49 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir
Author: Muzammil (Muzammiluddin-Syed-ECE)
<details>
<summary>Changes</summary>
Adds TODO messages to signal future work to move misplaced patterns, see https://github.com/llvm/llvm-project/issues/165811.
---
Full diff: https://github.com/llvm/llvm-project/pull/165812.diff
2 Files Affected:
- (modified) mlir/lib/Dialect/GPU/Transforms/PromoteShuffleToAMDGPU.cpp (+2)
- (modified) mlir/lib/Dialect/GPU/Transforms/SubgroupReduceLowering.cpp (+2)
``````````diff
diff --git a/mlir/lib/Dialect/GPU/Transforms/PromoteShuffleToAMDGPU.cpp b/mlir/lib/Dialect/GPU/Transforms/PromoteShuffleToAMDGPU.cpp
index 67cef8af1e3b5..f27565e04b48b 100644
--- a/mlir/lib/Dialect/GPU/Transforms/PromoteShuffleToAMDGPU.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/PromoteShuffleToAMDGPU.cpp
@@ -26,6 +26,8 @@ namespace {
constexpr amdgpu::Chipset kGfx950 = amdgpu::Chipset(9, 5, 0);
+// TODO: Move amdgpu specific patterns out of GPU dialect (#165811).
+
/// Try to promote `gpu.shuffle` to `amdgpu.swizzle_bitmode`, width must be 64
/// and offset must be a constant integer in the range [0, 31].
struct PromoteShuffleToSwizzlePattern
diff --git a/mlir/lib/Dialect/GPU/Transforms/SubgroupReduceLowering.cpp b/mlir/lib/Dialect/GPU/Transforms/SubgroupReduceLowering.cpp
index 81c3069cec16e..8c1f54b800089 100644
--- a/mlir/lib/Dialect/GPU/Transforms/SubgroupReduceLowering.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/SubgroupReduceLowering.cpp
@@ -367,6 +367,8 @@ struct VectorSubgroupReduceToShuffles final
bool matchClustered = false;
};
+// TODO: Move amdgpu specific patterns out of GPU dialect (#165811).
+
static FailureOr<Value>
createSubgroupDPPReduction(PatternRewriter &rewriter, gpu::SubgroupReduceOp op,
Value input, gpu::AllReduceOperation mode,
``````````
</details>
https://github.com/llvm/llvm-project/pull/165812
More information about the Mlir-commits
mailing list