[SPIRV] Add logic for OpGenericCastToPtrExplicit rewriting #146596
+150
−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
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-backend-spir-v Author: Justin Cai (jzc) ChangesThis PR adds overrides in Full diff: https://github.com/llvm/llvm-project/pull/146596.diff 4 Files Affected:
diff --git a/llvm/lib/Target/SPIRV/CMakeLists.txt b/llvm/lib/Target/SPIRV/CMakeLists.txt
index 4a2b534b948d6..ba09451104479 100644
--- a/llvm/lib/Target/SPIRV/CMakeLists.txt
+++ b/llvm/lib/Target/SPIRV/CMakeLists.txt
@@ -44,6 +44,7 @@ add_llvm_target(SPIRVCodeGen
SPIRVRegularizer.cpp
SPIRVSubtarget.cpp
SPIRVTargetMachine.cpp
+ SPIRVTargetTransformInfo.cpp
SPIRVUtils.cpp
SPIRVEmitNonSemanticDI.cpp
diff --git a/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.cpp b/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.cpp
new file mode 100644
index 0000000000000..95093d2b3c263
--- /dev/null
+++ b/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.cpp
@@ -0,0 +1,40 @@
+//===- SPIRVTargetTransformInfo.cpp - SPIR-V specific TTI -------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "SPIRVTargetTransformInfo.h"
+#include "llvm/IR/IntrinsicsSPIRV.h"
+
+using namespace llvm;
+
+bool llvm::SPIRVTTIImpl::collectFlatAddressOperands(
+ SmallVectorImpl<int> &OpIndexes, Intrinsic::ID IID) const {
+ switch (IID) {
+ case Intrinsic::spv_generic_cast_to_ptr_explicit:
+ OpIndexes.push_back(0);
+ return true;
+ default:
+ return false;
+ }
+}
+
+Value *llvm::SPIRVTTIImpl::rewriteIntrinsicWithAddressSpace(IntrinsicInst *II,
+ Value *OldV,
+ Value *NewV) const {
+ auto IntrID = II->getIntrinsicID();
+ switch (IntrID) {
+ case Intrinsic::spv_generic_cast_to_ptr_explicit: {
+ unsigned NewAS = NewV->getType()->getPointerAddressSpace();
+ unsigned DstAS = II->getType()->getPointerAddressSpace();
+ return NewAS == DstAS ? NewV
+ : ConstantPointerNull::get(
+ PointerType::get(NewV->getContext(), DstAS));
+ }
+ default:
+ return nullptr;
+ }
+}
diff --git a/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.h b/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.h
index 40e561ba38881..43bf6e9dd2a6e 100644
--- a/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.h
+++ b/llvm/lib/Target/SPIRV/SPIRVTargetTransformInfo.h
@@ -50,14 +50,15 @@ class SPIRVTTIImpl final : public BasicTTIImplBase<SPIRVTTIImpl> {
}
unsigned getFlatAddressSpace() const override {
- if (ST->isShader())
- return 0;
- // FIXME: Clang has 2 distinct address space maps. One where
+ // Clang has 2 distinct address space maps. One where
// default=4=Generic, and one with default=0=Function. This depends on the
- // environment. For OpenCL, we don't need to run the InferAddrSpace pass, so
- // we can return -1, but we might want to fix this.
- return -1;
+ // environment.
+ return ST->isShader() ? 0 : 4;
}
+ bool collectFlatAddressOperands(SmallVectorImpl<int> &OpIndexes,
+ Intrinsic::ID IID) const override;
+ Value *rewriteIntrinsicWithAddressSpace(IntrinsicInst *II, Value *OldV,
+ Value *NewV) const override;
};
} // namespace llvm
diff --git a/llvm/test/Transforms/InferAddressSpaces/SPIRV/generic-cast-explicit.ll b/llvm/test/Transforms/InferAddressSpaces/SPIRV/generic-cast-explicit.ll
new file mode 100644
index 0000000000000..2c8076df868a3
--- /dev/null
+++ b/llvm/test/Transforms/InferAddressSpaces/SPIRV/generic-cast-explicit.ll
@@ -0,0 +1,95 @@
+; This test checks that the address space casts for SPIR-V generic pointer casts
+; are lowered correctly by the infer-address-spaces pass.
+; RUN: opt < %s -passes=infer-address-spaces -S --mtriple=spirv64-unknown-unknown | FileCheck %s
+
+; Casting a global pointer to a global pointer.
+; The uses of c2 will be replaced with %global.
+; CHECK: @kernel1(ptr addrspace(1) %global)
+define i1 @kernel1(ptr addrspace(1) %global) {
+ %c1 = addrspacecast ptr addrspace(1) %global to ptr addrspace(4)
+ %c2 = call ptr addrspace(1) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(1) %global, null
+ %b1 = icmp eq ptr addrspace(1) %c2, null
+ ret i1 %b1
+}
+
+; Casting a global pointer to a local pointer.
+; The uses of c2 will be replaced with null.
+; CHECK: @kernel2(ptr addrspace(1) %global)
+define i1 @kernel2(ptr addrspace(1) %global) {
+ %c1 = addrspacecast ptr addrspace(1) %global to ptr addrspace(4)
+ %c2 = call ptr addrspace(3) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(3) null, null
+ %b1 = icmp eq ptr addrspace(3) %c2, null
+ ret i1 %b1
+}
+
+; Casting a global pointer to a private pointer.
+; CHECK: @kernel3(ptr addrspace(1) %global)
+define i1 @kernel3(ptr addrspace(1) %global) {
+ %c1 = addrspacecast ptr addrspace(1) %global to ptr addrspace(4)
+ %c2 = call ptr @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr null, null
+ %b1 = icmp eq ptr %c2, null
+ ret i1 %b1
+}
+
+; Casting a local pointer to a local pointer.
+; CHECK: @kernel4(ptr addrspace(3) %local)
+define i1 @kernel4(ptr addrspace(3) %local) {
+ %c1 = addrspacecast ptr addrspace(3) %local to ptr addrspace(4)
+ %c2 = call ptr addrspace(3) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(3) %local, null
+ %b1 = icmp eq ptr addrspace(3) %c2, null
+ ret i1 %b1
+}
+
+; Casting a local pointer to a global pointer.
+; CHECK: @kernel5(ptr addrspace(3) %local)
+define i1 @kernel5(ptr addrspace(3) %local) {
+ %c1 = addrspacecast ptr addrspace(3) %local to ptr addrspace(4)
+ %c2 = call ptr addrspace(1) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(1) null, null
+ %b1 = icmp eq ptr addrspace(1) %c2, null
+ ret i1 %b1
+}
+
+; Casting a local pointer to a private pointer.
+; CHECK: @kernel6(ptr addrspace(3) %local)
+define i1 @kernel6(ptr addrspace(3) %local) {
+ %c1 = addrspacecast ptr addrspace(3) %local to ptr addrspace(4)
+ %c2 = call ptr @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr null, null
+ %b1 = icmp eq ptr %c2, null
+ ret i1 %b1
+}
+
+; Casting a private pointer to a private pointer.
+; CHECK: @kernel7(ptr %private)
+define i1 @kernel7(ptr %private) {
+ %c1 = addrspacecast ptr %private to ptr addrspace(4)
+ %c2 = call ptr @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr %private, null
+ %b1 = icmp eq ptr %c2, null
+ ret i1 %b1
+}
+
+; Casting a private pointer to a global pointer.
+; CHECK: @kernel8(ptr %private)
+define i1 @kernel8(ptr %private) {
+ %c1 = addrspacecast ptr %private to ptr addrspace(4)
+ %c2 = call ptr addrspace(1) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(1) null, null
+ %b1 = icmp eq ptr addrspace(1) %c2, null
+ ret i1 %b1
+}
+
+; Casting a private pointer to a local pointer.
+; CHECK: @kernel9(ptr %private)
+define i1 @kernel9(ptr %private) {
+ %c1 = addrspacecast ptr %private to ptr addrspace(4)
+ %c2 = call ptr addrspace(3) @llvm.spv.generic.cast.to.ptr.explicit(ptr addrspace(4) %c1)
+ ; CHECK: %b1 = icmp eq ptr addrspace(3) null, null
+ %b1 = icmp eq ptr addrspace(3) %c2, null
+ ret i1 %b1
+}
|
s-perron
approved these changes
Jul 3, 2025
This works for shaders as well. |
maarquitos14
reviewed
Jul 3, 2025
| } | ||
|
|
||
| Value *llvm::SPIRVTTIImpl::rewriteIntrinsicWithAddressSpace(IntrinsicInst *II, | ||
| Value *OldV, |
There was a problem hiding this comment.
Does the compiler not complain about this being unused?
There was a problem hiding this comment.
It's a virtual function, so I don't think so
| } | ||
|
|
||
| ; Casting a global pointer to a local pointer. | ||
| ; The uses of c2 will be replaced with null. |
There was a problem hiding this comment.
I think this kind of comment is quite helpful, can we add something similar for the test cases below?
Keenuts
approved these changes
Jul 3, 2025
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.
This PR adds overrides in
SPIRVTTIImplforcollectFlatAddressOperandsandrewriteIntrinsicWithAddressSpaceto enableInferAddressSpacesPassto rewrite thellvm.spv.generic.cast.to.ptr.explicitintrinsic (corresponding toOpGenericCastToPtrExplicit) when the address space of the argument can be inferred. When the destination address space of the cast matches the inferred address space of the argument, the call is replaced with that argument. When they do not match, the cast is replaced with a constant null pointer.