[X86] Return illegal vectors in memory #121944

pranavk · 2025-01-07T14:33:42Z

Illegal vectors should be returned in memory via a hidden reference instead of returning them in SIMD registers.

See detailed discussion in https://groups.google.com/g/x86-64-abi/c/BjOOyihHuqg

llvmbot · 2025-01-07T14:34:18Z

@llvm/pr-subscribers-clang

Author: Pranav Kant (pranavk)

Changes

Illegal vectors should be returned in memory via a hidden reference instead of returning them in SIMD registers.

See detailed discussion in https://groups.google.com/g/x86-64-abi/c/BjOOyihHuqg

Full diff: https://github.com/llvm/llvm-project/pull/121944.diff

4 Files Affected:

(modified) clang/include/clang/Basic/LangOptions.h (+2)
(modified) clang/lib/CodeGen/Targets/X86.cpp (+37-3)
(added) clang/test/CodeGen/X86/x86-illegal-vector.c (+22)
(modified) clang/test/CodeGen/X86/x86-vec-i128.c (+20-18)

diff --git a/clang/include/clang/Basic/LangOptions.h b/clang/include/clang/Basic/LangOptions.h
index 949c8f5d448bcf..7dd8251c9da5c2 100644
--- a/clang/include/clang/Basic/LangOptions.h
+++ b/clang/include/clang/Basic/LangOptions.h
@@ -245,6 +245,8 @@ class LangOptionsBase {
     ///   construction vtable because it hasn't added 'type' as a substitution.
     ///   - Skip mangling enclosing class templates of member-like friend
     ///   function templates.
+    ///   - Incorrectly return illegal vectors (size greater than native
+    ///     vector size) to be returned in illegal registers on x86_64.
     Ver19,
 
     /// Conform to the underlying platform's C and C++ ABIs as closely
diff --git a/clang/lib/CodeGen/Targets/X86.cpp b/clang/lib/CodeGen/Targets/X86.cpp
index 7f73bf2a65266e..7fac5b264c668c 100644
--- a/clang/lib/CodeGen/Targets/X86.cpp
+++ b/clang/lib/CodeGen/Targets/X86.cpp
@@ -1298,8 +1298,12 @@ class X86_64ABIInfo : public ABIInfo {
                                            unsigned &NeededSSE,
                                            unsigned &MaxVectorWidth) const;
 
+  // Checks whether vector types for function arguments are illegal
   bool IsIllegalVectorType(QualType Ty) const;
 
+  // Checks whether vector types for returns are illegal
+  bool IsIllegalReturnVectorType(QualType Ty) const;
+
   /// The 0.98 ABI revision clarified a lot of ambiguities,
   /// unfortunately in ways that were not always consistent with
   /// certain previous compilers.  In particular, platforms which
@@ -1334,6 +1338,17 @@ class X86_64ABIInfo : public ABIInfo {
     return T.isOSLinux() || T.isOSNetBSD();
   }
 
+  bool returnIllegalVectorsInMem() const {
+    // Clang <= 19.0 did not do this.
+    if (getContext().getLangOpts().getClangABICompat() <=
+        LangOptions::ClangABI::Ver19)
+      return false;
+
+    const llvm::Triple &T = getTarget().getTriple();
+    return T.isOSLinux() || T.isOSNetBSD();
+  }
+
+
   X86AVXABILevel AVXLevel;
   // Some ABIs (e.g. X32 ABI and Native Client OS) use 32 bit pointers on
   // 64-bit hardware.
@@ -2156,9 +2171,12 @@ void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase, Class &Lo,
 }
 
 ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
+  const bool returnIllegalVectorsIndirectly = (returnIllegalVectorsInMem() &&
+                                               IsIllegalReturnVectorType(Ty));
+
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
   // place naturally.
-  if (!isAggregateTypeForABI(Ty)) {
+  if (!isAggregateTypeForABI(Ty) && !returnIllegalVectorsIndirectly) {
     // Treat an enum type as its underlying type.
     if (const EnumType *EnumTy = Ty->getAs<EnumType>())
       Ty = EnumTy->getDecl()->getIntegerType();
@@ -2173,12 +2191,24 @@ ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
   return getNaturalAlignIndirect(Ty);
 }
 
-bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+static bool IsIllegalVector(QualType Ty,
+                            uint64_t Size,
+                            X86AVXABILevel AVXLevel) {
   if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
-    uint64_t Size = getContext().getTypeSize(VecTy);
     unsigned LargestVector = getNativeVectorSizeForAVXABI(AVXLevel);
     if (Size <= 64 || Size > LargestVector)
       return true;
+  }
+
+  return false;
+}
+
+bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+  if (IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel))
+    return true;
+
+  // Maintain backward compatibility
+  if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
     QualType EltTy = VecTy->getElementType();
     if (passInt128VectorsInMem() &&
         (EltTy->isSpecificBuiltinType(BuiltinType::Int128) ||
@@ -2189,6 +2219,10 @@ bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
   return false;
 }
 
+bool X86_64ABIInfo::IsIllegalReturnVectorType(QualType Ty) const {
+  return IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel);
+}
+
 ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
                                             unsigned freeIntRegs) const {
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
diff --git a/clang/test/CodeGen/X86/x86-illegal-vector.c b/clang/test/CodeGen/X86/x86-illegal-vector.c
new file mode 100644
index 00000000000000..73fef26539fca8
--- /dev/null
+++ b/clang/test/CodeGen/X86/x86-illegal-vector.c
@@ -0,0 +1,22 @@
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,MEMRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+#define __MM_MALLOC_H
+#include <x86intrin.h>
+
+// REGRET128: define{{.*}} <4 x float> @get2()
+__m128 get2() { __m128 r = (__m128){5,6}; return r; }
+
+// MEMRET256: define{{.*}} void @get4(ptr{{.*}} sret(<8 x float>) align 32 %{{.*}})
+// REGRET256: define{{.*}} <8 x float> @get4()
+__m256 get4() { __m256 r = (__m256){7,8,9,10}; return r; }
+
+// MEMRET512: define{{.*}} void @get8(ptr{{.*}} sret(<16 x float>) align 64 %{{.*}})
+// REGRET512: define{{.*}} <16 x float> @get8()
+__m512 get8() { __m512 r = (__m512){7,8,9,10,1,2,3,4}; return r; }
\ No newline at end of file
diff --git a/clang/test/CodeGen/X86/x86-vec-i128.c b/clang/test/CodeGen/X86/x86-vec-i128.c
index ee58cb92da6b10..bdcfaeeea27784 100644
--- a/clang/test/CodeGen/X86/x86-vec-i128.c
+++ b/clang/test/CodeGen/X86/x86-vec-i128.c
@@ -1,16 +1,16 @@
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN16,MEM512ALIGN16
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN32
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN16,MEMARG512ALIGN16
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN32
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
 
 // RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
 // RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
@@ -32,8 +32,9 @@ typedef unsigned long long v32u64 __attribute__((vector_size(32)));
 typedef unsigned __int128 v32u128 __attribute__((vector_size(32)));
 
 v32u64 test_v32u128(v32u64 a, v32u128 b) {
-// MEM256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
-// MEM256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMARG256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
+// MEMARG256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMRETMEMARG256ALIGN32: define{{.*}} void @test_v32u128(ptr{{.*}} sret(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG10ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG9ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, <2 x i128> noundef %{{.*}})
   return a + (v32u64)b;
@@ -43,9 +44,10 @@ typedef unsigned long long v64u64 __attribute__((vector_size(64)));
 typedef unsigned __int128 v64u128 __attribute__((vector_size(64)));
 
 v64u64 test_v64u128(v64u64 a, v64u128 b) {
-// MEM512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
-// MEM512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
-// MEM512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMARG512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
+// MEMARG512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
+// MEMARG512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMRETMEMARG512ALIGN64: define{{.*}} void @test_v64u128(ptr{{.*}} sret(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG10ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG9ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, <4 x i128> noundef %{{.*}})
   return a + (v64u64)b;

llvmbot · 2025-01-07T14:34:18Z

@llvm/pr-subscribers-clang-codegen

Author: Pranav Kant (pranavk)

Changes

Illegal vectors should be returned in memory via a hidden reference instead of returning them in SIMD registers.

See detailed discussion in https://groups.google.com/g/x86-64-abi/c/BjOOyihHuqg

Full diff: https://github.com/llvm/llvm-project/pull/121944.diff

4 Files Affected:

(modified) clang/include/clang/Basic/LangOptions.h (+2)
(modified) clang/lib/CodeGen/Targets/X86.cpp (+37-3)
(added) clang/test/CodeGen/X86/x86-illegal-vector.c (+22)
(modified) clang/test/CodeGen/X86/x86-vec-i128.c (+20-18)

diff --git a/clang/include/clang/Basic/LangOptions.h b/clang/include/clang/Basic/LangOptions.h
index 949c8f5d448bcf..7dd8251c9da5c2 100644
--- a/clang/include/clang/Basic/LangOptions.h
+++ b/clang/include/clang/Basic/LangOptions.h
@@ -245,6 +245,8 @@ class LangOptionsBase {
     ///   construction vtable because it hasn't added 'type' as a substitution.
     ///   - Skip mangling enclosing class templates of member-like friend
     ///   function templates.
+    ///   - Incorrectly return illegal vectors (size greater than native
+    ///     vector size) to be returned in illegal registers on x86_64.
     Ver19,
 
     /// Conform to the underlying platform's C and C++ ABIs as closely
diff --git a/clang/lib/CodeGen/Targets/X86.cpp b/clang/lib/CodeGen/Targets/X86.cpp
index 7f73bf2a65266e..7fac5b264c668c 100644
--- a/clang/lib/CodeGen/Targets/X86.cpp
+++ b/clang/lib/CodeGen/Targets/X86.cpp
@@ -1298,8 +1298,12 @@ class X86_64ABIInfo : public ABIInfo {
                                            unsigned &NeededSSE,
                                            unsigned &MaxVectorWidth) const;
 
+  // Checks whether vector types for function arguments are illegal
   bool IsIllegalVectorType(QualType Ty) const;
 
+  // Checks whether vector types for returns are illegal
+  bool IsIllegalReturnVectorType(QualType Ty) const;
+
   /// The 0.98 ABI revision clarified a lot of ambiguities,
   /// unfortunately in ways that were not always consistent with
   /// certain previous compilers.  In particular, platforms which
@@ -1334,6 +1338,17 @@ class X86_64ABIInfo : public ABIInfo {
     return T.isOSLinux() || T.isOSNetBSD();
   }
 
+  bool returnIllegalVectorsInMem() const {
+    // Clang <= 19.0 did not do this.
+    if (getContext().getLangOpts().getClangABICompat() <=
+        LangOptions::ClangABI::Ver19)
+      return false;
+
+    const llvm::Triple &T = getTarget().getTriple();
+    return T.isOSLinux() || T.isOSNetBSD();
+  }
+
+
   X86AVXABILevel AVXLevel;
   // Some ABIs (e.g. X32 ABI and Native Client OS) use 32 bit pointers on
   // 64-bit hardware.
@@ -2156,9 +2171,12 @@ void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase, Class &Lo,
 }
 
 ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
+  const bool returnIllegalVectorsIndirectly = (returnIllegalVectorsInMem() &&
+                                               IsIllegalReturnVectorType(Ty));
+
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
   // place naturally.
-  if (!isAggregateTypeForABI(Ty)) {
+  if (!isAggregateTypeForABI(Ty) && !returnIllegalVectorsIndirectly) {
     // Treat an enum type as its underlying type.
     if (const EnumType *EnumTy = Ty->getAs<EnumType>())
       Ty = EnumTy->getDecl()->getIntegerType();
@@ -2173,12 +2191,24 @@ ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
   return getNaturalAlignIndirect(Ty);
 }
 
-bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+static bool IsIllegalVector(QualType Ty,
+                            uint64_t Size,
+                            X86AVXABILevel AVXLevel) {
   if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
-    uint64_t Size = getContext().getTypeSize(VecTy);
     unsigned LargestVector = getNativeVectorSizeForAVXABI(AVXLevel);
     if (Size <= 64 || Size > LargestVector)
       return true;
+  }
+
+  return false;
+}
+
+bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+  if (IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel))
+    return true;
+
+  // Maintain backward compatibility
+  if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
     QualType EltTy = VecTy->getElementType();
     if (passInt128VectorsInMem() &&
         (EltTy->isSpecificBuiltinType(BuiltinType::Int128) ||
@@ -2189,6 +2219,10 @@ bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
   return false;
 }
 
+bool X86_64ABIInfo::IsIllegalReturnVectorType(QualType Ty) const {
+  return IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel);
+}
+
 ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
                                             unsigned freeIntRegs) const {
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
diff --git a/clang/test/CodeGen/X86/x86-illegal-vector.c b/clang/test/CodeGen/X86/x86-illegal-vector.c
new file mode 100644
index 00000000000000..73fef26539fca8
--- /dev/null
+++ b/clang/test/CodeGen/X86/x86-illegal-vector.c
@@ -0,0 +1,22 @@
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,MEMRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+#define __MM_MALLOC_H
+#include <x86intrin.h>
+
+// REGRET128: define{{.*}} <4 x float> @get2()
+__m128 get2() { __m128 r = (__m128){5,6}; return r; }
+
+// MEMRET256: define{{.*}} void @get4(ptr{{.*}} sret(<8 x float>) align 32 %{{.*}})
+// REGRET256: define{{.*}} <8 x float> @get4()
+__m256 get4() { __m256 r = (__m256){7,8,9,10}; return r; }
+
+// MEMRET512: define{{.*}} void @get8(ptr{{.*}} sret(<16 x float>) align 64 %{{.*}})
+// REGRET512: define{{.*}} <16 x float> @get8()
+__m512 get8() { __m512 r = (__m512){7,8,9,10,1,2,3,4}; return r; }
\ No newline at end of file
diff --git a/clang/test/CodeGen/X86/x86-vec-i128.c b/clang/test/CodeGen/X86/x86-vec-i128.c
index ee58cb92da6b10..bdcfaeeea27784 100644
--- a/clang/test/CodeGen/X86/x86-vec-i128.c
+++ b/clang/test/CodeGen/X86/x86-vec-i128.c
@@ -1,16 +1,16 @@
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN16,MEM512ALIGN16
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN32
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN16,MEMARG512ALIGN16
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN32
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
 
 // RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
 // RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
@@ -32,8 +32,9 @@ typedef unsigned long long v32u64 __attribute__((vector_size(32)));
 typedef unsigned __int128 v32u128 __attribute__((vector_size(32)));
 
 v32u64 test_v32u128(v32u64 a, v32u128 b) {
-// MEM256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
-// MEM256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMARG256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
+// MEMARG256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMRETMEMARG256ALIGN32: define{{.*}} void @test_v32u128(ptr{{.*}} sret(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG10ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG9ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, <2 x i128> noundef %{{.*}})
   return a + (v32u64)b;
@@ -43,9 +44,10 @@ typedef unsigned long long v64u64 __attribute__((vector_size(64)));
 typedef unsigned __int128 v64u128 __attribute__((vector_size(64)));
 
 v64u64 test_v64u128(v64u64 a, v64u128 b) {
-// MEM512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
-// MEM512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
-// MEM512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMARG512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
+// MEMARG512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
+// MEMARG512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMRETMEMARG512ALIGN64: define{{.*}} void @test_v64u128(ptr{{.*}} sret(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG10ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG9ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, <4 x i128> noundef %{{.*}})
   return a + (v64u64)b;

llvmbot · 2025-01-07T14:34:19Z

@llvm/pr-subscribers-backend-x86

Author: Pranav Kant (pranavk)

Changes

Illegal vectors should be returned in memory via a hidden reference instead of returning them in SIMD registers.

See detailed discussion in https://groups.google.com/g/x86-64-abi/c/BjOOyihHuqg

Full diff: https://github.com/llvm/llvm-project/pull/121944.diff

4 Files Affected:

(modified) clang/include/clang/Basic/LangOptions.h (+2)
(modified) clang/lib/CodeGen/Targets/X86.cpp (+37-3)
(added) clang/test/CodeGen/X86/x86-illegal-vector.c (+22)
(modified) clang/test/CodeGen/X86/x86-vec-i128.c (+20-18)

diff --git a/clang/include/clang/Basic/LangOptions.h b/clang/include/clang/Basic/LangOptions.h
index 949c8f5d448bcf..7dd8251c9da5c2 100644
--- a/clang/include/clang/Basic/LangOptions.h
+++ b/clang/include/clang/Basic/LangOptions.h
@@ -245,6 +245,8 @@ class LangOptionsBase {
     ///   construction vtable because it hasn't added 'type' as a substitution.
     ///   - Skip mangling enclosing class templates of member-like friend
     ///   function templates.
+    ///   - Incorrectly return illegal vectors (size greater than native
+    ///     vector size) to be returned in illegal registers on x86_64.
     Ver19,
 
     /// Conform to the underlying platform's C and C++ ABIs as closely
diff --git a/clang/lib/CodeGen/Targets/X86.cpp b/clang/lib/CodeGen/Targets/X86.cpp
index 7f73bf2a65266e..7fac5b264c668c 100644
--- a/clang/lib/CodeGen/Targets/X86.cpp
+++ b/clang/lib/CodeGen/Targets/X86.cpp
@@ -1298,8 +1298,12 @@ class X86_64ABIInfo : public ABIInfo {
                                            unsigned &NeededSSE,
                                            unsigned &MaxVectorWidth) const;
 
+  // Checks whether vector types for function arguments are illegal
   bool IsIllegalVectorType(QualType Ty) const;
 
+  // Checks whether vector types for returns are illegal
+  bool IsIllegalReturnVectorType(QualType Ty) const;
+
   /// The 0.98 ABI revision clarified a lot of ambiguities,
   /// unfortunately in ways that were not always consistent with
   /// certain previous compilers.  In particular, platforms which
@@ -1334,6 +1338,17 @@ class X86_64ABIInfo : public ABIInfo {
     return T.isOSLinux() || T.isOSNetBSD();
   }
 
+  bool returnIllegalVectorsInMem() const {
+    // Clang <= 19.0 did not do this.
+    if (getContext().getLangOpts().getClangABICompat() <=
+        LangOptions::ClangABI::Ver19)
+      return false;
+
+    const llvm::Triple &T = getTarget().getTriple();
+    return T.isOSLinux() || T.isOSNetBSD();
+  }
+
+
   X86AVXABILevel AVXLevel;
   // Some ABIs (e.g. X32 ABI and Native Client OS) use 32 bit pointers on
   // 64-bit hardware.
@@ -2156,9 +2171,12 @@ void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase, Class &Lo,
 }
 
 ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
+  const bool returnIllegalVectorsIndirectly = (returnIllegalVectorsInMem() &&
+                                               IsIllegalReturnVectorType(Ty));
+
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
   // place naturally.
-  if (!isAggregateTypeForABI(Ty)) {
+  if (!isAggregateTypeForABI(Ty) && !returnIllegalVectorsIndirectly) {
     // Treat an enum type as its underlying type.
     if (const EnumType *EnumTy = Ty->getAs<EnumType>())
       Ty = EnumTy->getDecl()->getIntegerType();
@@ -2173,12 +2191,24 @@ ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
   return getNaturalAlignIndirect(Ty);
 }
 
-bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+static bool IsIllegalVector(QualType Ty,
+                            uint64_t Size,
+                            X86AVXABILevel AVXLevel) {
   if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
-    uint64_t Size = getContext().getTypeSize(VecTy);
     unsigned LargestVector = getNativeVectorSizeForAVXABI(AVXLevel);
     if (Size <= 64 || Size > LargestVector)
       return true;
+  }
+
+  return false;
+}
+
+bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
+  if (IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel))
+    return true;
+
+  // Maintain backward compatibility
+  if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
     QualType EltTy = VecTy->getElementType();
     if (passInt128VectorsInMem() &&
         (EltTy->isSpecificBuiltinType(BuiltinType::Int128) ||
@@ -2189,6 +2219,10 @@ bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
   return false;
 }
 
+bool X86_64ABIInfo::IsIllegalReturnVectorType(QualType Ty) const {
+  return IsIllegalVector(Ty, getContext().getTypeSize(Ty), AVXLevel);
+}
+
 ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
                                             unsigned freeIntRegs) const {
   // If this is a scalar LLVM value then assume LLVM will pass it in the right
diff --git a/clang/test/CodeGen/X86/x86-illegal-vector.c b/clang/test/CodeGen/X86/x86-illegal-vector.c
new file mode 100644
index 00000000000000..73fef26539fca8
--- /dev/null
+++ b/clang/test/CodeGen/X86/x86-illegal-vector.c
@@ -0,0 +1,22 @@
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,MEMRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,MEMRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - -fclang-abi-compat=19 | FileCheck %s --check-prefixes=REGRET128,REGRET256,REGRET512
+
+#define __MM_MALLOC_H
+#include <x86intrin.h>
+
+// REGRET128: define{{.*}} <4 x float> @get2()
+__m128 get2() { __m128 r = (__m128){5,6}; return r; }
+
+// MEMRET256: define{{.*}} void @get4(ptr{{.*}} sret(<8 x float>) align 32 %{{.*}})
+// REGRET256: define{{.*}} <8 x float> @get4()
+__m256 get4() { __m256 r = (__m256){7,8,9,10}; return r; }
+
+// MEMRET512: define{{.*}} void @get8(ptr{{.*}} sret(<16 x float>) align 64 %{{.*}})
+// REGRET512: define{{.*}} <16 x float> @get8()
+__m512 get8() { __m512 r = (__m512){7,8,9,10,1,2,3,4}; return r; }
\ No newline at end of file
diff --git a/clang/test/CodeGen/X86/x86-vec-i128.c b/clang/test/CodeGen/X86/x86-vec-i128.c
index ee58cb92da6b10..bdcfaeeea27784 100644
--- a/clang/test/CodeGen/X86/x86-vec-i128.c
+++ b/clang/test/CodeGen/X86/x86-vec-i128.c
@@ -1,16 +1,16 @@
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN16,MEM512ALIGN16
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEM256ALIGN32,MEM512ALIGN64
-
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN32
-// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
-// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEM512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,MEMRETMEMARG256ALIGN32,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN16,MEMARG512ALIGN16
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +sse2 -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +sse2 -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,MEMARG256ALIGN32,MEMARG512ALIGN64
+
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,MEMRETMEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-apple-darwin %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN32
+// RUN: %clang_cc1 -triple x86_64-scei-ps4 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-unknown-freebsd10.0 %s -target-feature +avx -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
+// RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx -emit-llvm -o - -fclang-abi-compat=9 | FileCheck %s --check-prefixes=CLANG9ABI128,CLANG9ABI256,MEMARG512ALIGN64
 
 // RUN: %clang_cc1 -triple x86_64-linux-gnu %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
 // RUN: %clang_cc1 -triple x86_64-netbsd %s -target-feature +avx512f -emit-llvm -o - | FileCheck %s --check-prefixes=CLANG10ABI128,CLANG10ABI256,CLANG10ABI512
@@ -32,8 +32,9 @@ typedef unsigned long long v32u64 __attribute__((vector_size(32)));
 typedef unsigned __int128 v32u128 __attribute__((vector_size(32)));
 
 v32u64 test_v32u128(v32u64 a, v32u128 b) {
-// MEM256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
-// MEM256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMARG256ALIGN16: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 16 %{{.*}}, ptr noundef byval(<2 x i128>) align 16 %{{.*}})
+// MEMARG256ALIGN32: define{{.*}} <4 x i64> @test_v32u128(ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
+// MEMRETMEMARG256ALIGN32: define{{.*}} void @test_v32u128(ptr{{.*}} sret(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i64>) align 32 %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG10ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, ptr noundef byval(<2 x i128>) align 32 %{{.*}})
 // CLANG9ABI256: define{{.*}} <4 x i64> @test_v32u128(<4 x i64> noundef %{{.*}}, <2 x i128> noundef %{{.*}})
   return a + (v32u64)b;
@@ -43,9 +44,10 @@ typedef unsigned long long v64u64 __attribute__((vector_size(64)));
 typedef unsigned __int128 v64u128 __attribute__((vector_size(64)));
 
 v64u64 test_v64u128(v64u64 a, v64u128 b) {
-// MEM512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
-// MEM512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
-// MEM512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMARG512ALIGN16: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 16 %{{.*}}, ptr noundef byval(<4 x i128>) align 16 %{{.*}})
+// MEMARG512ALIGN32: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 32 %{{.*}}, ptr noundef byval(<4 x i128>) align 32 %{{.*}})
+// MEMARG512ALIGN64: define{{.*}} <8 x i64> @test_v64u128(ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
+// MEMRETMEMARG512ALIGN64: define{{.*}} void @test_v64u128(ptr{{.*}} sret(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<8 x i64>) align 64 %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG10ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, ptr noundef byval(<4 x i128>) align 64 %{{.*}})
 // CLANG9ABI512: define{{.*}} <8 x i64> @test_v64u128(<8 x i64> noundef %{{.*}}, <4 x i128> noundef %{{.*}})
   return a + (v64u64)b;

github-actions · 2025-01-07T14:37:11Z

✅ With the latest revision this PR passed the C/C++ code formatter.

When vector size doesn't fit in native machine vector size, we should return vector via a hidden reference.

pranavk · 2025-01-07T17:40:51Z

Took a look at failing tests -- they need to be modified to conform to the new ABI. Should be easy to do.

phoebewang · 2025-01-08T08:31:08Z

This is a long existing problem and LLVM community doesn't make consensus about it, see https://reviews.llvm.org/D53919

My point is the case of returning illegal vectors is not well defined by psABI, and we have already warned this case in frontend. Considering the backward compatibility risk, I'd like to keep it as it.

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:codegen IR generation bugs: mangling, exceptions, etc. labels Jan 7, 2025

pranavk requested a review from phoebewang January 7, 2025 14:34

[X86] Return illegal vectors in memory

2347ae9

When vector size doesn't fit in native machine vector size, we should return vector via a hidden reference.

pranavk force-pushed the psabi2 branch from 02c9a94 to 2347ae9 Compare January 7, 2025 14:48

phoebewang requested a review from efriedma-quic January 8, 2025 08:31

pranavk closed this May 1, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[X86] Return illegal vectors in memory #121944

[X86] Return illegal vectors in memory #121944

pranavk commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

github-actions bot commented Jan 7, 2025 •

edited

Loading

Uh oh!

pranavk commented Jan 7, 2025

Uh oh!

phoebewang commented Jan 8, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

[X86] Return illegal vectors in memory #121944

[X86] Return illegal vectors in memory #121944

Conversation

pranavk commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

llvmbot commented Jan 7, 2025

Uh oh!

github-actions bot commented Jan 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pranavk commented Jan 7, 2025

Uh oh!

phoebewang commented Jan 8, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions bot commented Jan 7, 2025 •

edited

Loading