[X86] Define __LAHF_SAHF__ if feature 'sahf' is set or 32-bit mode

[MaskRay]

Summary

GCC 11 will define this macro.

In LLVM, the feature flag only applies to 64-bit mode and we always define the
macro in 32-bit mode. This is different from GCC -m32 in which -mno-sahf can
suppress the macro. The discrepancy can unlikely cause trouble.

[craig.topper]

clang/lib/Basic/Targets/X86.cpp

564

It looks like gcc may define this always in 32-bit mode?

[craig.topper]

clang/lib/Basic/Targets/X86.cpp

564

More confusingly, they define it in 32-bit mode unless -mno-sahf is on the command line. Despite 32-bit mode always having LAHF/SAHF instructions.

[MaskRay]

clang/lib/Basic/Targets/X86.cpp

564

I think this means we fail to set sahf for all 32-bit CPUs. GCC gcc/config/i386/i386-options.c sets OPTION_MASK_ISA_SAHF on all 32-bit CPUs unless -mno-sahf.

According to https://git.noc.ruhr-uni-bochum.de/kostea7x/ghidra/-/commit/239106a356ef8555575a64cc2cfd3fd71b415406?view=parallel
some pre-2005 Intel and AMD 64-bit CPUs had problems so 'sahf' is disabled for some CPUs.

[craig.topper]

clang/lib/Basic/Targets/X86.cpp

564

The X86Subtarget and X86.td file calls "sahf" as HasLAHFSAHF64. And contains we have this predicate for codegen uses.

{code}

bool hasLAHFSAHF() const { return HasLAHFSAHF64 || !is64Bit(); }

{code}

Does -mno-sahf prevent gcc from using LAHF/SAHF in 32-bit mode?

[MaskRay]

Always define __LAHF_SAHF__ in 32-bit mode

[MaskRay]

clang/lib/Basic/Targets/X86.cpp

564

From gcc/config/i386/i386-options.c sets OPTION_MASK_ISA_SAHF, I think so, but there isn't any codegen test... (like many other components of GCC)

This discrepancy is not a big problem so let's always define it for 32-bit mode for now.

[craig.topper]

LGTM

clang/lib/Basic/Targets/X86.cpp

if (HasVAES)
     Builder.defineMacro("__VAES__");
 
   if (HasPCLMUL)
     Builder.defineMacro("__PCLMUL__");
 
   if (HasVPCLMULQDQ)
     Builder.defineMacro("__VPCLMULQDQ__");
 
+  // Note, in 32-bit mode, GCC does not define the macro if -mno-sahf. In LLVM,
+  // the feature flag only applies to 64-bit mode.
+  if (HasLAHFSAHF || getTriple().getArch() == llvm::Triple::x86)
+    Builder.defineMacro("__LAHF_SAHF__");
+
   if (HasLZCNT)
     Builder.defineMacro("__LZCNT__");
 
   if (HasRDRND)
     Builder.defineMacro("__RDRND__");
 
   if (HasFSGSBASE)
     Builder.defineMacro("__FSGSBASE__");

clang/test/Preprocessor/predefined-arch-macros.c

 // Begin X86/GCC/Linux tests ----------------
 
 // RUN: %clang -march=i386 -m32 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_I386_M32
+// CHECK_I386_M32: #define __LAHF_SAHF__ 1
 // CHECK_I386_M32: #define __i386 1
 // CHECK_I386_M32: #define __i386__ 1
 // CHECK_I386_M32: #define __tune_i386__ 1
 // CHECK_I386_M32: #define i386 1
 // RUN: not %clang -march=i386 -m64 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_I386_M64
 // CHECK_I386_M64: error: {{.*}}

 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_CORE_AVX2_M32
 // CHECK_CORE_AVX2_M32: #define __AVX2__ 1
 // CHECK_CORE_AVX2_M32: #define __AVX__ 1
 // CHECK_CORE_AVX2_M32: #define __BMI2__ 1
 // CHECK_CORE_AVX2_M32: #define __BMI__ 1
 // CHECK_CORE_AVX2_M32: #define __F16C__ 1
 // CHECK_CORE_AVX2_M32: #define __FMA__ 1
 // CHECK_CORE_AVX2_M32: #define __INVPCID__ 1
+// CHECK_CORE_AVX2_M32: #define __LAHF_SAHF__ 1
 // CHECK_CORE_AVX2_M32: #define __LZCNT__ 1
 // CHECK_CORE_AVX2_M32: #define __MMX__ 1
 // CHECK_CORE_AVX2_M32: #define __MOVBE__ 1
 // CHECK_CORE_AVX2_M32: #define __PCLMUL__ 1
 // CHECK_CORE_AVX2_M32: #define __POPCNT__ 1
 // CHECK_CORE_AVX2_M32: #define __RDRND__ 1
 // CHECK_CORE_AVX2_M32: #define __SSE2__ 1
 // CHECK_CORE_AVX2_M32: #define __SSE3__ 1

 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_CORE_AVX2_M64
 // CHECK_CORE_AVX2_M64: #define __AVX2__ 1
 // CHECK_CORE_AVX2_M64: #define __AVX__ 1
 // CHECK_CORE_AVX2_M64: #define __BMI2__ 1
 // CHECK_CORE_AVX2_M64: #define __BMI__ 1
 // CHECK_CORE_AVX2_M64: #define __F16C__ 1
 // CHECK_CORE_AVX2_M64: #define __FMA__ 1
 // CHECK_CORE_AVX2_M64: #define __INVPCID__ 1
+// CHECK_CORE_AVX2_M64: #define __LAHF_SAHF__ 1
 // CHECK_CORE_AVX2_M64: #define __LZCNT__ 1
 // CHECK_CORE_AVX2_M64: #define __MMX__ 1
 // CHECK_CORE_AVX2_M64: #define __MOVBE__ 1
 // CHECK_CORE_AVX2_M64: #define __PCLMUL__ 1
 // CHECK_CORE_AVX2_M64: #define __POPCNT__ 1
 // CHECK_CORE_AVX2_M64: #define __RDRND__ 1
 // CHECK_CORE_AVX2_M64: #define __SSE2_MATH__ 1
 // CHECK_CORE_AVX2_M64: #define __SSE2__ 1

 // CHECK_BROADWELL_M32: #define __ADX__ 1
 // CHECK_BROADWELL_M32: #define __AVX2__ 1
 // CHECK_BROADWELL_M32: #define __AVX__ 1
 // CHECK_BROADWELL_M32: #define __BMI2__ 1
 // CHECK_BROADWELL_M32: #define __BMI__ 1
 // CHECK_BROADWELL_M32: #define __F16C__ 1
 // CHECK_BROADWELL_M32: #define __FMA__ 1
 // CHECK_BROADWELL_M32: #define __INVPCID__ 1
+// CHECK_BROADWELL_M32: #define __LAHF_SAHF__ 1
 // CHECK_BROADWELL_M32: #define __LZCNT__ 1
 // CHECK_BROADWELL_M32: #define __MMX__ 1
 // CHECK_BROADWELL_M32: #define __MOVBE__ 1
 // CHECK_BROADWELL_M32: #define __PCLMUL__ 1
 // CHECK_BROADWELL_M32: #define __POPCNT__ 1
 // CHECK_BROADWELL_M32: #define __PRFCHW__ 1
 // CHECK_BROADWELL_M32: #define __RDRND__ 1
 // CHECK_BROADWELL_M32: #define __RDSEED__ 1

 // CHECK_BROADWELL_M64: #define __ADX__ 1
 // CHECK_BROADWELL_M64: #define __AVX2__ 1
 // CHECK_BROADWELL_M64: #define __AVX__ 1
 // CHECK_BROADWELL_M64: #define __BMI2__ 1
 // CHECK_BROADWELL_M64: #define __BMI__ 1
 // CHECK_BROADWELL_M64: #define __F16C__ 1
 // CHECK_BROADWELL_M64: #define __FMA__ 1
 // CHECK_BROADWELL_M64: #define __INVPCID__ 1
+// CHECK_BROADWELL_M64: #define __LAHF_SAHF__ 1
 // CHECK_BROADWELL_M64: #define __LZCNT__ 1
 // CHECK_BROADWELL_M64: #define __MMX__ 1
 // CHECK_BROADWELL_M64: #define __MOVBE__ 1
 // CHECK_BROADWELL_M64: #define __PCLMUL__ 1
 // CHECK_BROADWELL_M64: #define __POPCNT__ 1
 // CHECK_BROADWELL_M64: #define __PRFCHW__ 1
 // CHECK_BROADWELL_M64: #define __RDRND__ 1
 // CHECK_BROADWELL_M64: #define __RDSEED__ 1

 // CHECK_ATHLON_FX_M64: #define __x86_64 1
 // CHECK_ATHLON_FX_M64: #define __x86_64__ 1
 
 // RUN: %clang -march=amdfam10 -m32 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_AMDFAM10_M32
 // CHECK_AMDFAM10_M32: #define __3dNOW_A__ 1
 // CHECK_AMDFAM10_M32: #define __3dNOW__ 1
+// CHECK_AMDFAM10_M32: #define __LAHF_SAHF__ 1
 // CHECK_AMDFAM10_M32: #define __LZCNT__ 1
 // CHECK_AMDFAM10_M32: #define __MMX__ 1
 // CHECK_AMDFAM10_M32: #define __POPCNT__ 1
 // CHECK_AMDFAM10_M32: #define __PRFCHW__ 1
 // CHECK_AMDFAM10_M32: #define __SSE2_MATH__ 1
 // CHECK_AMDFAM10_M32: #define __SSE2__ 1
 // CHECK_AMDFAM10_M32: #define __SSE3__ 1
 // CHECK_AMDFAM10_M32: #define __SSE4A__ 1
 // CHECK_AMDFAM10_M32: #define __SSE_MATH__ 1
 // CHECK_AMDFAM10_M32: #define __SSE__ 1
 // CHECK_AMDFAM10_M32-NOT: #define __SSSE3__ 1
 // CHECK_AMDFAM10_M32: #define __amdfam10 1
 // CHECK_AMDFAM10_M32: #define __amdfam10__ 1
 // CHECK_AMDFAM10_M32: #define __i386 1
 // CHECK_AMDFAM10_M32: #define __i386__ 1
 // CHECK_AMDFAM10_M32: #define __tune_amdfam10__ 1
 
 // RUN: %clang -march=amdfam10 -m64 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_AMDFAM10_M64
 // CHECK_AMDFAM10_M64: #define __3dNOW_A__ 1
 // CHECK_AMDFAM10_M64: #define __3dNOW__ 1
 // CHECK_AMDFAM10_M64: #define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16 1
+// CHECK_AMDFAM10_M64: #define __LAHF_SAHF__ 1
 // CHECK_AMDFAM10_M64: #define __LZCNT__ 1
 // CHECK_AMDFAM10_M64: #define __MMX__ 1
 // CHECK_AMDFAM10_M64: #define __POPCNT__ 1
 // CHECK_AMDFAM10_M64: #define __PRFCHW__ 1
 // CHECK_AMDFAM10_M64: #define __SSE2_MATH__ 1
 // CHECK_AMDFAM10_M64: #define __SSE2__ 1
 // CHECK_AMDFAM10_M64: #define __SSE3__ 1
 // CHECK_AMDFAM10_M64: #define __SSE4A__ 1
 // CHECK_AMDFAM10_M64: #define __SSE_MATH__ 1
 // CHECK_AMDFAM10_M64: #define __SSE__ 1
 // CHECK_AMDFAM10_M64-NOT: #define __SSSE3__ 1
 // CHECK_AMDFAM10_M64: #define __amd64 1
 // CHECK_AMDFAM10_M64: #define __amd64__ 1
 // CHECK_AMDFAM10_M64: #define __amdfam10 1
 // CHECK_AMDFAM10_M64: #define __amdfam10__ 1
 // CHECK_AMDFAM10_M64: #define __tune_amdfam10__ 1
 // CHECK_AMDFAM10_M64: #define __x86_64 1
 // CHECK_AMDFAM10_M64: #define __x86_64__ 1
 
 // RUN: %clang -march=btver1 -m32 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_BTVER1_M32
 // CHECK_BTVER1_M32-NOT: #define __3dNOW_A__ 1
 // CHECK_BTVER1_M32-NOT: #define __3dNOW__ 1
+// CHECK_BTVER1_M32: #define __LAHF_SAHF__ 1
 // CHECK_BTVER1_M32: #define __LZCNT__ 1
 // CHECK_BTVER1_M32: #define __MMX__ 1
 // CHECK_BTVER1_M32: #define __POPCNT__ 1
 // CHECK_BTVER1_M32: #define __PRFCHW__ 1
 // CHECK_BTVER1_M32: #define __SSE2_MATH__ 1
 // CHECK_BTVER1_M32: #define __SSE2__ 1
 // CHECK_BTVER1_M32: #define __SSE3__ 1
 // CHECK_BTVER1_M32: #define __SSE4A__ 1
 // CHECK_BTVER1_M32: #define __SSE_MATH__ 1
 // CHECK_BTVER1_M32: #define __SSE__ 1
 // CHECK_BTVER1_M32: #define __SSSE3__ 1
 // CHECK_BTVER1_M32: #define __btver1 1
 // CHECK_BTVER1_M32: #define __btver1__ 1
 // CHECK_BTVER1_M32: #define __i386 1
 // CHECK_BTVER1_M32: #define __i386__ 1
 // CHECK_BTVER1_M32: #define __tune_btver1__ 1
 
 // RUN: %clang -march=btver1 -m64 -E -dM %s -o - 2>&1 \
 // RUN:     -target i386-unknown-linux \
 // RUN:   | FileCheck -match-full-lines %s -check-prefix=CHECK_BTVER1_M64
 // CHECK_BTVER1_M64-NOT: #define __3dNOW_A__ 1
 // CHECK_BTVER1_M64-NOT: #define __3dNOW__ 1
+// CHECK_BTVER1_M64: #define __LAHF_SAHF__ 1
 // CHECK_BTVER1_M64: #define __LZCNT__ 1
 // CHECK_BTVER1_M64: #define __MMX__ 1
 // CHECK_BTVER1_M64: #define __POPCNT__ 1
 // CHECK_BTVER1_M64: #define __PRFCHW__ 1
 // CHECK_BTVER1_M64: #define __SSE2_MATH__ 1
 // CHECK_BTVER1_M64: #define __SSE2__ 1
 // CHECK_BTVER1_M64: #define __SSE3__ 1
 // CHECK_BTVER1_M64: #define __SSE4A__ 1

 // CHECK_ZNVER1_M64: #define __BMI2__ 1
 // CHECK_ZNVER1_M64: #define __BMI__ 1
 // CHECK_ZNVER1_M64: #define __CLFLUSHOPT__ 1
 // CHECK_ZNVER1_M64: #define __CLZERO__ 1
 // CHECK_ZNVER1_M64: #define __F16C__ 1
 // CHECK_ZNVER1_M64-NOT: #define __FMA4__ 1
 // CHECK_ZNVER1_M64: #define __FMA__ 1
 // CHECK_ZNVER1_M64: #define __FSGSBASE__ 1
+// CHECK_ZNVER1_M64: #define __LAHF_SAHF__ 1
 // CHECK_ZNVER1_M64: #define __LZCNT__ 1
 // CHECK_ZNVER1_M64: #define __MMX__ 1
 // CHECK_ZNVER1_M64: #define __MOVBE__ 1
 // CHECK_ZNVER1_M64: #define __PCLMUL__ 1
 // CHECK_ZNVER1_M64: #define __POPCNT__ 1
 // CHECK_ZNVER1_M64: #define __PRFCHW__ 1
 // CHECK_ZNVER1_M64: #define __RDRND__ 1
 // CHECK_ZNVER1_M64: #define __RDSEED__ 1

 // CHECK_ZNVER2_M32: #define __BMI__ 1
 // CHECK_ZNVER2_M32: #define __CLFLUSHOPT__ 1
 // CHECK_ZNVER2_M32: #define __CLWB__ 1
 // CHECK_ZNVER2_M32: #define __CLZERO__ 1
 // CHECK_ZNVER2_M32: #define __F16C__ 1
 // CHECK_ZNVER2_M32-NOT: #define __FMA4__ 1
 // CHECK_ZNVER2_M32: #define __FMA__ 1
 // CHECK_ZNVER2_M32: #define __FSGSBASE__ 1
+// CHECK_ZNVER2_M32: #define __LAHF_SAHF__ 1
 // CHECK_ZNVER2_M32: #define __LZCNT__ 1
 // CHECK_ZNVER2_M32: #define __MMX__ 1
 // CHECK_ZNVER2_M32: #define __PCLMUL__ 1
 // CHECK_ZNVER2_M32: #define __POPCNT__ 1
 // CHECK_ZNVER2_M32: #define __PRFCHW__ 1
 // CHECK_ZNVER2_M32: #define __RDPID__ 1
 // CHECK_ZNVER2_M32: #define __RDRND__ 1
 // CHECK_ZNVER2_M32: #define __RDSEED__ 1

 // CHECK_ZNVER2_M64: #define __BMI__ 1
 // CHECK_ZNVER2_M64: #define __CLFLUSHOPT__ 1
 // CHECK_ZNVER2_M64: #define __CLWB__ 1
 // CHECK_ZNVER2_M64: #define __CLZERO__ 1
 // CHECK_ZNVER2_M64: #define __F16C__ 1
 // CHECK_ZNVER2_M64-NOT: #define __FMA4__ 1
 // CHECK_ZNVER2_M64: #define __FMA__ 1
 // CHECK_ZNVER2_M64: #define __FSGSBASE__ 1
+// CHECK_ZNVER2_M64: #define __LAHF_SAHF__ 1
 // CHECK_ZNVER2_M64: #define __LZCNT__ 1
 // CHECK_ZNVER2_M64: #define __MMX__ 1
 // CHECK_ZNVER2_M64: #define __PCLMUL__ 1
 // CHECK_ZNVER2_M64: #define __POPCNT__ 1
 // CHECK_ZNVER2_M64: #define __PRFCHW__ 1
 // CHECK_ZNVER2_M64: #define __RDPID__ 1
 // CHECK_ZNVER2_M64: #define __RDRND__ 1
 // CHECK_ZNVER2_M64: #define __RDSEED__ 1