[Affine] Parallelize loops with MemoryEffects::Free ops. #172388
+39
−2
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
If an operation deallocates memory allocated inside the loop, then it is as safe for parallelization as the allocation, which is already handled by `isLocallyDefined`.
Member
|
@llvm/pr-subscribers-mlir Author: Slava Zakharin (vzakhari) ChangesIf an operation deallocates memory allocated inside the loop, Full diff: https://github.com/llvm/llvm-project/pull/172388.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp b/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
index 3d1a73417d1ea..b126909439b88 100644
--- a/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
+++ b/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
@@ -134,6 +134,25 @@ static bool isLocallyDefined(Value v, Operation *enclosingOp) {
return viewOp && isLocallyDefined(viewOp.getViewSource(), enclosingOp);
}
+/// Returns true if `op` has only Free memory effects on the values
+/// that are locally defined (i.e. they are allocated by operations
+/// nested withing `enclosingOp`).
+static bool isDeallocationOfLocallyDefined(Operation *op,
+ Operation *enclosingOp) {
+ std::optional<SmallVector<MemoryEffects::EffectInstance>> effects =
+ getEffectsRecursively(op);
+ if (!effects)
+ return false;
+
+ for (const MemoryEffects::EffectInstance &effect : *effects) {
+ Value freed = effect.getValue();
+ if (!isa<MemoryEffects::Free>(effect.getEffect()) || !freed ||
+ !isLocallyDefined(freed, enclosingOp))
+ return false;
+ }
+ return true;
+}
+
bool mlir::affine::isLoopMemoryParallel(AffineForOp forOp) {
// Any memref-typed iteration arguments are treated as serializing.
if (llvm::any_of(forOp.getResultTypes(), llvm::IsaPred<BaseMemRefType>))
@@ -152,6 +171,7 @@ bool mlir::affine::isLoopMemoryParallel(AffineForOp forOp) {
loadAndStoreOps.push_back(op);
} else if (!isa<AffineForOp, AffineYieldOp, AffineIfOp>(op) &&
!hasSingleEffect<MemoryEffects::Allocate>(op) &&
+ !isDeallocationOfLocallyDefined(op, forOp) &&
!isMemoryEffectFree(op)) {
// Alloc-like ops inside `forOp` are fine (they don't impact parallelism)
// as long as they don't escape the loop (which has been checked above).
diff --git a/mlir/test/Dialect/Affine/parallelize.mlir b/mlir/test/Dialect/Affine/parallelize.mlir
index bfd1720959861..95cf8f3a45f3f 100644
--- a/mlir/test/Dialect/Affine/parallelize.mlir
+++ b/mlir/test/Dialect/Affine/parallelize.mlir
@@ -272,19 +272,20 @@ func.func @nested_min_max(%m: memref<?xf32>, %lb0: index,
// Test in the presence of locally allocated memrefs.
-// CHECK: func @local_alloc
+// CHECK-LABEL: func @local_alloc
func.func @local_alloc() {
%cst = arith.constant 0.0 : f32
affine.for %i = 0 to 100 {
%m = memref.alloc() : memref<1xf32>
%ma = memref.alloca() : memref<1xf32>
affine.store %cst, %m[0] : memref<1xf32>
+ memref.dealloc %m : memref<1xf32>
}
// CHECK: affine.parallel
return
}
-// CHECK: func @local_alloc_cast
+// CHECK-LABEL: func @local_alloc_cast
func.func @local_alloc_cast() {
%cst = arith.constant 0.0 : f32
affine.for %i = 0 to 100 {
@@ -300,6 +301,7 @@ func.func @local_alloc_cast() {
affine.for %j = 0 to 8 {
affine.store %cst, %r[%j, %j] : memref<8x16xf32>
}
+ memref.dealloc %m : memref<128xf32>
}
// CHECK: affine.parallel
// CHECK: affine.parallel
@@ -313,6 +315,21 @@ func.func @local_alloc_cast() {
return
}
+// When memref.dealloc deallocates out-of-loop allocation,
+// the loop should not be parallelized. This test is quite
+// artificial though.
+// CHECK-LABEL: func @local_dealloc
+func.func @local_dealloc() {
+ %cst = arith.constant 0.0 : f32
+ %m = memref.alloc() : memref<1xf32>
+ affine.for %i = 0 to 1 {
+ affine.store %cst, %m[%i] : memref<1xf32>
+ memref.dealloc %m : memref<1xf32>
+ }
+ // CHECK-NOT: affine.parallel
+ return
+}
+
// CHECK-LABEL: @iter_arg_memrefs
func.func @iter_arg_memrefs(%in: memref<10xf32>) {
%mi = memref.alloc() : memref<f32>
|
Member
|
@llvm/pr-subscribers-mlir-affine Author: Slava Zakharin (vzakhari) ChangesIf an operation deallocates memory allocated inside the loop, Full diff: https://github.com/llvm/llvm-project/pull/172388.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp b/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
index 3d1a73417d1ea..b126909439b88 100644
--- a/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
+++ b/mlir/lib/Dialect/Affine/Analysis/AffineAnalysis.cpp
@@ -134,6 +134,25 @@ static bool isLocallyDefined(Value v, Operation *enclosingOp) {
return viewOp && isLocallyDefined(viewOp.getViewSource(), enclosingOp);
}
+/// Returns true if `op` has only Free memory effects on the values
+/// that are locally defined (i.e. they are allocated by operations
+/// nested withing `enclosingOp`).
+static bool isDeallocationOfLocallyDefined(Operation *op,
+ Operation *enclosingOp) {
+ std::optional<SmallVector<MemoryEffects::EffectInstance>> effects =
+ getEffectsRecursively(op);
+ if (!effects)
+ return false;
+
+ for (const MemoryEffects::EffectInstance &effect : *effects) {
+ Value freed = effect.getValue();
+ if (!isa<MemoryEffects::Free>(effect.getEffect()) || !freed ||
+ !isLocallyDefined(freed, enclosingOp))
+ return false;
+ }
+ return true;
+}
+
bool mlir::affine::isLoopMemoryParallel(AffineForOp forOp) {
// Any memref-typed iteration arguments are treated as serializing.
if (llvm::any_of(forOp.getResultTypes(), llvm::IsaPred<BaseMemRefType>))
@@ -152,6 +171,7 @@ bool mlir::affine::isLoopMemoryParallel(AffineForOp forOp) {
loadAndStoreOps.push_back(op);
} else if (!isa<AffineForOp, AffineYieldOp, AffineIfOp>(op) &&
!hasSingleEffect<MemoryEffects::Allocate>(op) &&
+ !isDeallocationOfLocallyDefined(op, forOp) &&
!isMemoryEffectFree(op)) {
// Alloc-like ops inside `forOp` are fine (they don't impact parallelism)
// as long as they don't escape the loop (which has been checked above).
diff --git a/mlir/test/Dialect/Affine/parallelize.mlir b/mlir/test/Dialect/Affine/parallelize.mlir
index bfd1720959861..95cf8f3a45f3f 100644
--- a/mlir/test/Dialect/Affine/parallelize.mlir
+++ b/mlir/test/Dialect/Affine/parallelize.mlir
@@ -272,19 +272,20 @@ func.func @nested_min_max(%m: memref<?xf32>, %lb0: index,
// Test in the presence of locally allocated memrefs.
-// CHECK: func @local_alloc
+// CHECK-LABEL: func @local_alloc
func.func @local_alloc() {
%cst = arith.constant 0.0 : f32
affine.for %i = 0 to 100 {
%m = memref.alloc() : memref<1xf32>
%ma = memref.alloca() : memref<1xf32>
affine.store %cst, %m[0] : memref<1xf32>
+ memref.dealloc %m : memref<1xf32>
}
// CHECK: affine.parallel
return
}
-// CHECK: func @local_alloc_cast
+// CHECK-LABEL: func @local_alloc_cast
func.func @local_alloc_cast() {
%cst = arith.constant 0.0 : f32
affine.for %i = 0 to 100 {
@@ -300,6 +301,7 @@ func.func @local_alloc_cast() {
affine.for %j = 0 to 8 {
affine.store %cst, %r[%j, %j] : memref<8x16xf32>
}
+ memref.dealloc %m : memref<128xf32>
}
// CHECK: affine.parallel
// CHECK: affine.parallel
@@ -313,6 +315,21 @@ func.func @local_alloc_cast() {
return
}
+// When memref.dealloc deallocates out-of-loop allocation,
+// the loop should not be parallelized. This test is quite
+// artificial though.
+// CHECK-LABEL: func @local_dealloc
+func.func @local_dealloc() {
+ %cst = arith.constant 0.0 : f32
+ %m = memref.alloc() : memref<1xf32>
+ affine.for %i = 0 to 1 {
+ affine.store %cst, %m[%i] : memref<1xf32>
+ memref.dealloc %m : memref<1xf32>
+ }
+ // CHECK-NOT: affine.parallel
+ return
+}
+
// CHECK-LABEL: @iter_arg_memrefs
func.func @iter_arg_memrefs(%in: memref<10xf32>) {
%mi = memref.alloc() : memref<f32>
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
If an operation deallocates memory allocated inside the loop,
then it is as safe for parallelization as the allocation,
which is already handled by
isLocallyDefined.