[mlir][GPU] Expand LLVM function attribute copies #76755
Merged
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
|
@llvm/pr-subscribers-mlir-gpu Author: Krzysztof Drewniak (krzysz00) ChangesExpand the copying of attributes on GPU kernel arguments during LLVM lowering. Support copying attributes from values that are already LLVM pointers. Support copying attributes, like Full diff: https://github.com/llvm/llvm-project/pull/76755.diff 2 Files Affected:
diff --git a/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp b/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
index e79a02f931af2d..69d69356b1f6cf 100644
--- a/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
+++ b/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
@@ -194,18 +194,22 @@ GPUFuncOpLowering::matchAndRewrite(gpu::GPUFuncOp gpuFuncOp, OpAdaptor adaptor,
// Get memref type from function arguments and set the noalias to
// pointer arguments.
for (const auto &en : llvm::enumerate(gpuFuncOp.getArgumentTypes())) {
- auto memrefTy = en.value().dyn_cast<MemRefType>();
+ auto remapping = signatureConversion.getInputMapping(en.index());
NamedAttrList argAttr = argAttrs
? argAttrs[en.index()].cast<DictionaryAttr>()
: NamedAttrList();
-
+ auto copyAttribute = [&](StringRef attrName) {
+ Attribute attr = argAttr.erase(attrName);
+ if (!attr)
+ return;
+ for (size_t i = 0, e = remapping->size; i < e; ++i)
+ llvmFuncOp.setArgAttr(remapping->inputNo + i, attrName, attr);
+ };
auto copyPointerAttribute = [&](StringRef attrName) {
Attribute attr = argAttr.erase(attrName);
- // This is a proxy for the bare pointer calling convention.
if (!attr)
return;
- auto remapping = signatureConversion.getInputMapping(en.index());
if (remapping->size > 1 &&
attrName == LLVM::LLVMDialect::getNoAliasAttrName()) {
emitWarning(llvmFuncOp.getLoc(),
@@ -224,10 +228,23 @@ GPUFuncOpLowering::matchAndRewrite(gpu::GPUFuncOp gpuFuncOp, OpAdaptor adaptor,
if (argAttr.empty())
continue;
- if (memrefTy) {
+ copyAttribute(LLVM::LLVMDialect::getReturnedAttrName());
+ copyAttribute(LLVM::LLVMDialect::getNoUndefAttrName());
+ copyAttribute(LLVM::LLVMDialect::getInRegAttrName());
+ bool lowersToPointer = false;
+ for (size_t i = 0, e = remapping->size; i < e; ++i) {
+ lowersToPointer |= isa<LLVM::LLVMPointerType>(
+ llvmFuncOp.getArgument(remapping->inputNo + i).getType());
+ }
+
+ if (lowersToPointer) {
copyPointerAttribute(LLVM::LLVMDialect::getNoAliasAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getNoCaptureAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getNoFreeAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getAlignAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getReadonlyAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getWriteOnlyAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getReadnoneAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getNonNullAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getDereferenceableAttrName());
copyPointerAttribute(
diff --git a/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir b/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
index 33374984eb7c91..e7c742067b4eb5 100644
--- a/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
+++ b/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
@@ -24,6 +24,17 @@ gpu.module @kernel {
// ROCDL-SAME: !llvm.ptr {llvm.writeonly}
// NVVM-SAME: !llvm.ptr {llvm.writeonly}
+// -----
+
+gpu.module @kernel {
+ gpu.func @test_func_readonly_ptr(%arg0 : !llvm.ptr {llvm.readonly} ) {
+ gpu.return
+ }
+}
+
+// CHECK-LABEL: llvm.func @test_func_readonly_ptr
+// ROCDL-SAME: !llvm.ptr {llvm.readonly}
+// NVVM-SAME: !llvm.ptr {llvm.readonly}
// -----
@@ -62,3 +73,17 @@ gpu.module @kernel {
// CHECK-LABEL: llvm.func @test_func_dereferenceable_or_null
// ROCDL-SAME: !llvm.ptr {llvm.dereferenceable_or_null = 4 : i64}
// NVVM-SAME: !llvm.ptr {llvm.dereferenceable_or_null = 4 : i64}
+
+// -----
+
+gpu.module @kernel {
+ gpu.func @test_func_noundef(%arg0 : memref<f32> {llvm.noundef} ) {
+ gpu.return
+ }
+}
+
+// CHECK-LABEL: llvm.func @test_func_noundef
+// ROCDL-SAME: !llvm.ptr {llvm.noundef}
+// ROCDL-SAME: i64 {llvm.noundef}
+// NVVM-SAME: !llvm.ptr {llvm.noundef}
+// NVVM-SAME: i64 {llvm.noundef}
|
|
@llvm/pr-subscribers-mlir Author: Krzysztof Drewniak (krzysz00) ChangesExpand the copying of attributes on GPU kernel arguments during LLVM lowering. Support copying attributes from values that are already LLVM pointers. Support copying attributes, like Full diff: https://github.com/llvm/llvm-project/pull/76755.diff 2 Files Affected:
diff --git a/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp b/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
index e79a02f931af2d..69d69356b1f6cf 100644
--- a/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
+++ b/mlir/lib/Conversion/GPUCommon/GPUOpsLowering.cpp
@@ -194,18 +194,22 @@ GPUFuncOpLowering::matchAndRewrite(gpu::GPUFuncOp gpuFuncOp, OpAdaptor adaptor,
// Get memref type from function arguments and set the noalias to
// pointer arguments.
for (const auto &en : llvm::enumerate(gpuFuncOp.getArgumentTypes())) {
- auto memrefTy = en.value().dyn_cast<MemRefType>();
+ auto remapping = signatureConversion.getInputMapping(en.index());
NamedAttrList argAttr = argAttrs
? argAttrs[en.index()].cast<DictionaryAttr>()
: NamedAttrList();
-
+ auto copyAttribute = [&](StringRef attrName) {
+ Attribute attr = argAttr.erase(attrName);
+ if (!attr)
+ return;
+ for (size_t i = 0, e = remapping->size; i < e; ++i)
+ llvmFuncOp.setArgAttr(remapping->inputNo + i, attrName, attr);
+ };
auto copyPointerAttribute = [&](StringRef attrName) {
Attribute attr = argAttr.erase(attrName);
- // This is a proxy for the bare pointer calling convention.
if (!attr)
return;
- auto remapping = signatureConversion.getInputMapping(en.index());
if (remapping->size > 1 &&
attrName == LLVM::LLVMDialect::getNoAliasAttrName()) {
emitWarning(llvmFuncOp.getLoc(),
@@ -224,10 +228,23 @@ GPUFuncOpLowering::matchAndRewrite(gpu::GPUFuncOp gpuFuncOp, OpAdaptor adaptor,
if (argAttr.empty())
continue;
- if (memrefTy) {
+ copyAttribute(LLVM::LLVMDialect::getReturnedAttrName());
+ copyAttribute(LLVM::LLVMDialect::getNoUndefAttrName());
+ copyAttribute(LLVM::LLVMDialect::getInRegAttrName());
+ bool lowersToPointer = false;
+ for (size_t i = 0, e = remapping->size; i < e; ++i) {
+ lowersToPointer |= isa<LLVM::LLVMPointerType>(
+ llvmFuncOp.getArgument(remapping->inputNo + i).getType());
+ }
+
+ if (lowersToPointer) {
copyPointerAttribute(LLVM::LLVMDialect::getNoAliasAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getNoCaptureAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getNoFreeAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getAlignAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getReadonlyAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getWriteOnlyAttrName());
+ copyPointerAttribute(LLVM::LLVMDialect::getReadnoneAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getNonNullAttrName());
copyPointerAttribute(LLVM::LLVMDialect::getDereferenceableAttrName());
copyPointerAttribute(
diff --git a/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir b/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
index 33374984eb7c91..e7c742067b4eb5 100644
--- a/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
+++ b/mlir/test/Conversion/GPUCommon/memref-arg-attrs.mlir
@@ -24,6 +24,17 @@ gpu.module @kernel {
// ROCDL-SAME: !llvm.ptr {llvm.writeonly}
// NVVM-SAME: !llvm.ptr {llvm.writeonly}
+// -----
+
+gpu.module @kernel {
+ gpu.func @test_func_readonly_ptr(%arg0 : !llvm.ptr {llvm.readonly} ) {
+ gpu.return
+ }
+}
+
+// CHECK-LABEL: llvm.func @test_func_readonly_ptr
+// ROCDL-SAME: !llvm.ptr {llvm.readonly}
+// NVVM-SAME: !llvm.ptr {llvm.readonly}
// -----
@@ -62,3 +73,17 @@ gpu.module @kernel {
// CHECK-LABEL: llvm.func @test_func_dereferenceable_or_null
// ROCDL-SAME: !llvm.ptr {llvm.dereferenceable_or_null = 4 : i64}
// NVVM-SAME: !llvm.ptr {llvm.dereferenceable_or_null = 4 : i64}
+
+// -----
+
+gpu.module @kernel {
+ gpu.func @test_func_noundef(%arg0 : memref<f32> {llvm.noundef} ) {
+ gpu.return
+ }
+}
+
+// CHECK-LABEL: llvm.func @test_func_noundef
+// ROCDL-SAME: !llvm.ptr {llvm.noundef}
+// ROCDL-SAME: i64 {llvm.noundef}
+// NVVM-SAME: !llvm.ptr {llvm.noundef}
+// NVVM-SAME: i64 {llvm.noundef}
|
joker-eph
approved these changes
Jan 2, 2024
kuhar
reviewed
Jan 2, 2024
Expand the copying of attributes on GPU kernel arguments during LLVM lowering. Support copying attributes from values that are already LLVM pointers. Support copying attributes, like `noundef`, that aren't specific to (the pointer parts of) arguments.
krzysz00
force-pushed
the
mlir-improve-gpu-function-attr-copies
branch
from
b2d261e to
97c7966
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Expand the copying of attributes on GPU kernel arguments during LLVM lowering.
Support copying attributes from values that are already LLVM pointers.
Support copying attributes, like
noundef, that aren't specific to (the pointer parts of) arguments.