[mlir][OpenACC][OpenMP] Modify atomic capture to allow update/write #167963
+52
−6
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
OpenACC's C++ version has a variant of capture that permits a update followed by a write. Therefore the verifier was overly strict in this case. According to our reading of the OpenMP 6.0 spec, it appears that `atomic captured update` (page 495) also requires this form, so it seems reasonable to allow this for both languages.
Collaborator
Author
|
@razvanlupusoru : can you suggest OMP reviewers for me? |
Member
|
@llvm/pr-subscribers-mlir-openacc @llvm/pr-subscribers-mlir Author: Erich Keane (erichkeane) ChangesOpenACC's C++ version has a variant of capture that permits a update followed by a write. Therefore the verifier was overly strict in this case. According to our reading of the OpenMP 6.0 spec, it appears that Full diff: https://github.com/llvm/llvm-project/pull/167963.diff 4 Files Affected:
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
index 5b89f741e296d..8c33be7ff1747 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
@@ -2999,6 +2999,12 @@ def AtomicCaptureOp : OpenACC_Op<"atomic.capture",
acc.atomic.write ...
acc.terminator
}
+
+ acc.atomic.capture {
+ acc.atomic.update ...
+ acc.atomic.write ...
+ acc.terminator
+ }
```
}];
diff --git a/mlir/include/mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.td b/mlir/include/mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.td
index 223bee9ab1c27..9df6b907eb326 100644
--- a/mlir/include/mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.td
+++ b/mlir/include/mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.td
@@ -239,6 +239,7 @@ def AtomicCaptureOpInterface : OpInterface<"AtomicCaptureOpInterface"> {
implement one of the atomic interfaces. It can be found in one of these
forms:
`{ atomic.update, atomic.read }`
+ `{ atomic.update, atomic.write }`
`{ atomic.read, atomic.update }`
`{ atomic.read, atomic.write }`
}];
@@ -291,12 +292,15 @@ def AtomicCaptureOpInterface : OpInterface<"AtomicCaptureOpInterface"> {
auto secondWriteStmt = dyn_cast<AtomicWriteOpInterface>(secondOp);
if (!((firstUpdateStmt && secondReadStmt) ||
+ (firstUpdateStmt && secondWriteStmt) ||
(firstReadStmt && secondUpdateStmt) ||
(firstReadStmt && secondWriteStmt)))
return ops.front().emitError()
<< "invalid sequence of operations in the capture region";
- if (firstUpdateStmt && secondReadStmt &&
- firstUpdateStmt.getX() != secondReadStmt.getX())
+ if ((firstUpdateStmt && secondReadStmt &&
+ firstUpdateStmt.getX() != secondReadStmt.getX()) ||
+ (firstUpdateStmt && secondWriteStmt &&
+ firstUpdateStmt.getX() != secondWriteStmt.getX()))
return firstUpdateStmt.emitError()
<< "updated variable in atomic.update must be captured in "
"second operation";
diff --git a/mlir/test/Dialect/OpenACC/invalid.mlir b/mlir/test/Dialect/OpenACC/invalid.mlir
index 0e75894eaeceb..173c6a74545f8 100644
--- a/mlir/test/Dialect/OpenACC/invalid.mlir
+++ b/mlir/test/Dialect/OpenACC/invalid.mlir
@@ -690,7 +690,6 @@ func.func @acc_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
func.func @acc_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
acc.atomic.capture {
- // expected-error @below {{invalid sequence of operations in the capture region}}
acc.atomic.update %x : memref<i32> {
^bb0(%xval: i32):
%newval = llvm.add %xval, %expr : i32
@@ -704,6 +703,23 @@ func.func @acc_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
// -----
+func.func @acc_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
+ acc.atomic.capture {
+ // expected-error @below {{updated variable in atomic.update must be captured in second operation}}
+ acc.atomic.update %x : memref<i32> {
+ ^bb0(%xval: i32):
+ %newval = llvm.add %xval, %expr : i32
+ acc.yield %newval : i32
+ }
+ acc.atomic.write %v = %expr : memref<i32>, i32
+
+ acc.terminator
+ }
+ return
+}
+
+// -----
+
func.func @acc_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
acc.atomic.capture {
// expected-error @below {{invalid sequence of operations in the capture region}}
diff --git a/mlir/test/Dialect/OpenMP/invalid.mlir b/mlir/test/Dialect/OpenMP/invalid.mlir
index af24d969064ab..6224a236e7c78 100644
--- a/mlir/test/Dialect/OpenMP/invalid.mlir
+++ b/mlir/test/Dialect/OpenMP/invalid.mlir
@@ -1263,7 +1263,22 @@ func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
omp.atomic.capture {
- // expected-error @below {{invalid sequence of operations in the capture region}}
+ // expected-error @below {{updated variable in atomic.update must be captured in second operation}}
+ omp.atomic.update %x : memref<i32> {
+ ^bb0(%xval: i32):
+ %newval = llvm.add %xval, %expr : i32
+ omp.yield (%newval : i32)
+ }
+ omp.atomic.write %v = %expr : memref<i32>, i32
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
+ omp.atomic.capture {
omp.atomic.update %x : memref<i32> {
^bb0(%xval: i32):
%newval = llvm.add %xval, %expr : i32
@@ -1289,6 +1304,22 @@ func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
// -----
+func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) {
+ omp.atomic.capture {
+ omp.atomic.update %x : memref<i32> {
+ ^bb0(%xval: i32):
+ %newval = llvm.add %xval, %expr : i32
+ omp.yield (%newval : i32)
+ }
+ omp.atomic.write %x = %expr : memref<i32>, i32
+
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
func.func @omp_atomic_capture(%x: memref<i32>, %y: memref<i32>, %v: memref<i32>, %expr: i32) {
omp.atomic.capture {
// expected-error @below {{updated variable in atomic.update must be captured in second operation}}
|
|
|
||
| // ----- | ||
|
|
||
| func.func @omp_atomic_capture(%x: memref<i32>, %v: memref<i32>, %expr: i32) { |
Collaborator
Author
There was a problem hiding this comment.
This is supposed to be a 'valid' config test, so I don't have diagnostics. That said, I seem to have double-added this 'is valid' test.
Contributor
There was a problem hiding this comment.
Can you add the "valid" tests to ops.mlir (mlir/test/Dialect/OpenACC/ops.mlir and mlir/test/Dialect/OpenMP/ops.mlir)?
The invalid tests look great to me.
Collaborator
Author
There was a problem hiding this comment.
Ah! Yes, I did that. Thanks for the help on where to put those!
razvanlupusoru
approved these changes
Nov 13, 2025
Contributor
razvanlupusoru
left a comment
razvanlupusoru
left a comment
There was a problem hiding this comment.
Looks great to me! Can you please allow a bit of time for the omp reviewers to take a look also? Thank you for this!
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
OpenACC's C++ version has a variant of capture that permits a update followed by a write. Therefore the verifier was overly strict in this case.
According to our reading of the OpenMP 6.0 spec, it appears that
atomic captured update(page 495) also requires this form, so it seems reasonable to allow this for both languages.