[MLIR][LLVM] Propagate alignment attribute from memref to LLVM #151380
Conversation
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-mlir Author: Erick Ochoa Lopez (amd-eochoalo) ChangesPropagate alignment attribute from operations in the memref dialect to the LLVM dialect. @kuhar Possible improvements: maybe the alignment attribute in LLVM's store and load operations should be confined/constrained to i64? I believe that way one can avoid typing the value in the attribute dictionary. I.e., from Full diff: https://github.com/llvm/llvm-project/pull/151380.diff 2 Files Affected:
diff --git a/mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp b/mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp
index dc2035b0700d0..d6bdd342c9045 100644
--- a/mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp
+++ b/mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp
@@ -895,8 +895,8 @@ struct LoadOpLowering : public LoadStoreOpLowering<memref::LoadOp> {
adaptor.getMemref(),
adaptor.getIndices(), kNoWrapFlags);
rewriter.replaceOpWithNewOp<LLVM::LoadOp>(
- loadOp, typeConverter->convertType(type.getElementType()), dataPtr, 0,
- false, loadOp.getNontemporal());
+ loadOp, typeConverter->convertType(type.getElementType()), dataPtr,
+ loadOp.getAlignment().value_or(0), false, loadOp.getNontemporal());
return success();
}
};
@@ -918,7 +918,8 @@ struct StoreOpLowering : public LoadStoreOpLowering<memref::StoreOp> {
getStridedElementPtr(rewriter, op.getLoc(), type, adaptor.getMemref(),
adaptor.getIndices(), kNoWrapFlags);
rewriter.replaceOpWithNewOp<LLVM::StoreOp>(op, adaptor.getValue(), dataPtr,
- 0, false, op.getNontemporal());
+ op.getAlignment().value_or(0),
+ false, op.getNontemporal());
return success();
}
};
diff --git a/mlir/test/Conversion/MemRefToLLVM/memref-to-llvm.mlir b/mlir/test/Conversion/MemRefToLLVM/memref-to-llvm.mlir
index 51d56389dac9e..959cc21acdae2 100644
--- a/mlir/test/Conversion/MemRefToLLVM/memref-to-llvm.mlir
+++ b/mlir/test/Conversion/MemRefToLLVM/memref-to-llvm.mlir
@@ -753,6 +753,18 @@ func.func @load_non_temporal(%arg0 : memref<32xf32, affine_map<(d0) -> (d0)>>) {
// -----
+// CHECK-LABEL: func @load_with_alignment(
+// CHECK-INTERFACE-LABEL: func @load_with_alignment(
+func.func @load_with_alignment(%arg0 : memref<32xf32>) {
+ %1 = arith.constant 7 : index
+ // CHECK: llvm.load %{{.*}} {alignment = 32 : i64} : !llvm.ptr -> f32
+ // CHECK-INTERFACE: llvm.load
+ %2 = memref.load %arg0[%1] {alignment = 32} : memref<32xf32>
+ func.return
+}
+
+// -----
+
// CHECK-LABEL: func @store_non_temporal(
// CHECK-INTERFACE-LABEL: func @store_non_temporal(
func.func @store_non_temporal(%input : memref<32xf32, affine_map<(d0) -> (d0)>>, %output : memref<32xf32, affine_map<(d0) -> (d0)>>) {
@@ -766,6 +778,18 @@ func.func @store_non_temporal(%input : memref<32xf32, affine_map<(d0) -> (d0)>>,
// -----
+// CHECK-LABEL: func @store_with_alignment(
+// CHECK-INTERFACE-LABEL: func @store_with_alignment(
+func.func @store_with_alignment(%arg0 : memref<32xf32>, %arg1 : f32) {
+ %1 = arith.constant 7 : index
+ // CHECK: llvm.store %{{.*}}, %{{.*}} {alignment = 32 : i64} : f32, !llvm.ptr
+ // CHECK-INTERFACE: llvm.store
+ memref.store %arg1, %arg0[%1] {alignment = 32} : memref<32xf32>
+ func.return
+}
+
+// -----
+
// Ensure unconvertable memory space not cause a crash
// CHECK-LABEL: @alloca_unconvertable_memory_space
|
| // CHECK-LABEL: func @load_with_alignment( | ||
| // CHECK-INTERFACE-LABEL: func @load_with_alignment( | ||
| func.func @load_with_alignment(%arg0 : memref<32xf32>) { | ||
| %1 = arith.constant 7 : index |
There was a problem hiding this comment.
[nit] Reduce the number of Ops in the test by turning this into a function argument.
There was a problem hiding this comment.
Thanks! Here are the changes: 7ae0025
|
I don't have merge privileges. Can someone merge for me? Thanks in advance! |
|
@amd-eochoalo Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
…151380) Propagate alignment attribute from operations in the memref dialect to the LLVM dialect. Possible improvements: maybe the alignment attribute in LLVM's store and load operations should be confined/constrained to i64? I believe that way one can avoid typing the value in the attribute dictionary. I.e., from `{ alignment = 32 : i64 }` to `{ alignment = 32}`
Propagate alignment attribute from operations in the memref dialect to the LLVM dialect.
Possible improvements: maybe the alignment attribute in LLVM's store and load operations should be confined/constrained to i64? I believe that way one can avoid typing the value in the attribute dictionary. I.e., from
{ alignment = 32 : i64 }to{ alignment = 32}