[CIR] Add support for X86 pmovqd512_mask and pmovwb512_mask builtins #173802
Conversation
Add CIR support for pmovqd512_mask and pmovwb512_mask. And a minor type cast fix in CirGenCleanup.cpp that caused compile error on Windows.
DannyDaoBoYang
requested review from
andykaylor,
bcardosolopes,
lanza and
xlauko
as code owners
|
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. |
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-clangir Author: None (DannyDaoBoYang) ChangesAs requested by #167765 This is my first request to this repo. If you noticed anything wrong please comment and I'm looking forward for corrections. Addresses #167765 Full diff: https://github.com/llvm/llvm-project/pull/173802.diff 4 Files Affected:
diff --git a/clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h b/clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h
index cc28941aaa079..481d06091f012 100644
--- a/clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h
+++ b/clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h
@@ -425,6 +425,22 @@ class CIRBaseBuilderTy : public mlir::OpBuilder {
// Cast/Conversion Operators
//===--------------------------------------------------------------------===//
+ /// Create an value truncation to a narrower type.
+ /// Returns the source if types already match. CIR casts do not
+ /// encode NUW/NSW; wrap semantics should be handled by callers.
+ /// Supports both scalar integers and vectors of integers.
+ mlir::Value createTrunc(mlir::Location loc, mlir::Value src,
+ mlir::Type newTy) {
+ auto srcIntTy = mlir::dyn_cast<cir::VectorType>(src.getType());
+ if (newTy == srcIntTy)
+ return src;
+ return createCast(loc, cir::CastKind::integral, src, newTy);
+ }
+
+ mlir::Value createTrunc(mlir::Value src, mlir::Type newTy) {
+ return createTrunc(src.getLoc(), src, newTy);
+ }
+
mlir::Value createCast(mlir::Location loc, cir::CastKind kind,
mlir::Value src, mlir::Type newTy) {
if (newTy == src.getType())
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp
index 1c87e945de846..b553327f676f5 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp
@@ -1274,7 +1274,10 @@ CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID, const CallExpr *expr) {
mask);
}
case X86::BI__builtin_ia32_pmovqd512_mask:
- case X86::BI__builtin_ia32_pmovwb512_mask:
+ case X86::BI__builtin_ia32_pmovwb512_mask: {
+ mlir::Value Res = builder.createTrunc(ops[0], cast<cir::VectorType>(ops[1].getType()));
+ return emitX86Select(builder, getLoc(expr->getExprLoc()), ops[2], Res, ops[1]);
+ }
case X86::BI__builtin_ia32_pblendw128:
case X86::BI__builtin_ia32_blendpd:
case X86::BI__builtin_ia32_blendps:
diff --git a/clang/lib/CIR/CodeGen/CIRGenCleanup.cpp b/clang/lib/CIR/CodeGen/CIRGenCleanup.cpp
index 6c6cb402d1190..8d9ea7c6c22eb 100644
--- a/clang/lib/CIR/CodeGen/CIRGenCleanup.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenCleanup.cpp
@@ -97,7 +97,7 @@ EHScopeStack::getInnermostActiveNormalCleanup() const {
char *EHScopeStack::allocate(size_t size) {
size = llvm::alignTo(size, ScopeStackAlignment);
if (!startOfBuffer) {
- unsigned capacity = llvm::PowerOf2Ceil(std::max(size, 1024ul));
+ unsigned capacity = llvm::PowerOf2Ceil(std::max<size_t>(size, 1024ul));
startOfBuffer = std::make_unique<char[]>(capacity);
startOfData = endOfBuffer = startOfBuffer.get() + capacity;
} else if (static_cast<size_t>(startOfData - startOfBuffer.get()) < size) {
diff --git a/clang/test/CIR/CodeGenBuiltins/X86/pmovqd-mask-builtins.c b/clang/test/CIR/CodeGenBuiltins/X86/pmovqd-mask-builtins.c
new file mode 100644
index 0000000000000..b43d2dc8f050d
--- /dev/null
+++ b/clang/test/CIR/CodeGenBuiltins/X86/pmovqd-mask-builtins.c
@@ -0,0 +1,29 @@
+// RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +avx512f -fclangir -emit-cir -o %t.cir -Wall -Werror -Wsign-conversion
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+
+#include <immintrin.h>
+
+__m256i test_pmovqd_mask(__m512i a, __m256i b, __mmask8 mask) {
+ // CIR-LABEL: test_pmovqd_mask
+ // CIR: %[[TRUNC:.*]] = cir.cast integral {{.*}} : !cir.vector<8 x !s64i> -> !cir.vector<8 x !s32i>
+ // CIR: %[[MASK_VEC:.*]] = cir.cast bitcast {{.*}} : !u8i -> !cir.vector<8 x !cir.int<s, 1>>
+ // CIR: cir.vec.ternary(%[[MASK_VEC]], %[[TRUNC]], {{.*}}) : !cir.vector<8 x !cir.int<s, 1>>, !cir.vector<8 x !s32i>
+ return __builtin_ia32_pmovqd512_mask(a, b, mask);
+}
+
+__m256i test_pmovqd_maskz(__m512i a, __mmask8 mask) {
+ // CIR-LABEL: test_pmovqd_maskz
+ // CIR: %[[TRUNC:.*]] = cir.cast integral {{.*}} : !cir.vector<8 x !s64i> -> !cir.vector<8 x !s32i>
+ // CIR: %[[MASK_VEC:.*]] = cir.cast bitcast {{.*}} : !u8i -> !cir.vector<8 x !cir.int<s, 1>>
+ // CIR: cir.vec.ternary(%[[MASK_VEC]], %[[TRUNC]], {{.*}}) : !cir.vector<8 x !cir.int<s, 1>>, !cir.vector<8 x !s32i>
+ __m256i zero = _mm256_setzero_si256();
+ return __builtin_ia32_pmovqd512_mask(a, zero, mask);
+}
+
+__m256i test_pmovwb_mask(__m512i a, __m256i b, __mmask32 mask) {
+ // CIR-LABEL: test_pmovwb_mask
+ // CIR: %[[TRUNC:.*]] = cir.cast integral {{.*}} : !cir.vector<32 x !s16i> -> !cir.vector<32 x !s8i>
+ // CIR: %[[MASK_VEC:.*]] = cir.cast bitcast {{.*}} : !u32i -> !cir.vector<32 x !cir.int<s, 1>>
+ // CIR: cir.vec.ternary(%[[MASK_VEC]], %[[TRUNC]], {{.*}}) : !cir.vector<32 x !cir.int<s, 1>>, !cir.vector<32 x !s8i>
+ return __builtin_ia32_pmovwb512_mask(a, b, mask);
+}
\ No newline at end of file
|
2 participants
As requested by #167765
-Add CIR support for pmovqd512_mask.
-In addition, I noticed pmovwb512_mask uses identical truncation-and-selection logic X86.cpp., so I added support for pmovwb512_mask using the same logic.
-Added a new LIT test case in clang/test/CIR/CodeGen/pmovqd-mask-builtins.c to verify correct CIR generation for both builtins.
-Added a type cast fix in CirGenCleanup.cpp where std::max encountered a type mismatch between size_t and unsigned long on Win64. This caused compile error on Windows for me while building it.
This is my first request to this repo. If you noticed anything wrong please comment and I'm looking forward for corrections.
Addresses #167765