[PowerPC] Fix x86 vector intrinsics wrapper compilation under C++

[nemanjai]

LGTM. Thanks for fixing this.

clang/lib/Headers/ppc_wrappers/xmmintrin.h

31

I agree that it would be good to get rid of the extra space. :)

clang/test/CodeGen/ppc-xmmintrin.c

1433

I am not sure if this file has any tests that use the actual enumerators rather than an integer, but it seems like we should.

[bjope]

clang/test/CodeGen/ppc-xmmintrin.c

10

Unfortunately I get some failures with this. Maybe because of an unstandard build setup.

We've started to use -DCLANG_DEFAULT_RTLIB=compiler-rt -DCLANG_DEFAULT_CXX_STDLIB=libc++ when building clang. And we also use -DLLVM_ENABLE_PER_TARGET_RUNTIME_DIR=ON. Not sure if that is a setup that is asking for trouble. Anyway, when running this test case we end up with

: 'RUN: at line 10';   /workspace/llvm/build/bin/clang -x c++ -fsyntax-only -target powerpc64le-unknown-linux-gnu -mcpu=pwr8 -ffreestanding -DNO_WARN_X86_INTRINSICS /workspace/clang/test/CodeGen/ppc-xmmintrin.c    -fno-discard-value-names -mllvm -disable-llvm-optzns
--
Exit Code: 1

Command Output (stderr):
--
In file included from /workspace/clang/test/CodeGen/ppc-xmmintrin.c:13:
In file included from /workspace/llvm/build/lib/clang/13.0.0/include/ppc_wrappers/xmmintrin.h:42:
In file included from /workspace/llvm/build/lib/clang/13.0.0/include/altivec.h:44:
In file included from /workspace/llvm/build/bin/../include/c++/v1/stddef.h:39:
/workspace/llvm/build/bin/../include/c++/v1/__config:13:10: fatal error: '__config_site' file not found
#include <__config_site>
         ^~~~~~~~~~~~~~~
1 error generated.

Not sure really how to solve that.

Maybe we should stop building like that?

Or there is a bug somewhere (such as that we only get the __config_site headers in target specific dirs for targets that we actually build runtimes for, maybe something that was missing in https://reviews.llvm.org/D97572)?
(Maybe @phosek could comment on that?)

Or this test case is missing some options to make it a bit more independent on the runtimes build setup?

[qiucf]

clang/test/CodeGen/ppc-xmmintrin.c

10

The tests relies on some system stuff. Is the failure related to this change? (or exposed by -x c++ option?)

[bjope]

clang/test/CodeGen/ppc-xmmintrin.c

10

Well, i guess it was exposed by -x c++ (in combination with D975729).

I don't really understand how things are supposed to work given the pre-target specific __config_site.

Since I only build libcxx for x86_64-unknown-linux-gnu, I only get a __config_site file for that specific triple inside bin/../include/x86_64-unknown-linux-gnu/c++/v1/__config_site in the build result. But a test case like this one, using a different triple, ends up including bin/../include/c++/v1/stddef.h, that wants wants to include <__config_site>, but it won't find any such include for the powerpc64le-unknown-linux-gnu triple.

[qiucf]

clang/test/CodeGen/ppc-xmmintrin.c

10

Hi, I tried a build with libcxx enabled on ppc64le, by default it still looks for stddef.h inside lib/clang/13.0.0/include. If I specify -isystem include/c++/v1, the error won't happen because __config_site is there.

I didn't try -DLLVM_ENABLE_PER_TARGET_RUNTIME_DIR=ON since that makes build complaining bin/clang-tblgen: error while loading shared libraries: libc++.so.1: cannot open shared object file.

The revision is going to LLVM 12.0.1. And this failure doesn't look like a blocker, since (if I understand correctly) even without this patch, a simple test also triggers that in the condition you provided.

clang/lib/Headers/ppc_wrappers/xmmintrin.h

/* This header file is to help porting code using Intel intrinsics
 
    It should be noted that there's much difference between X86_64's
    MXSCR and PowerISA's FPSCR/VSCR registers. It's recommended to use
    portable <fenv.h> instead of access MXSCR directly.
 
    Most SSE scalar float intrinsic operations can be performed more
    efficiently as C language float scalar operations or optimized to
    use vector SIMD operations. We recommend this for new applications. */
 #error "Please read comment above.  Use -DNO_WARN_X86_INTRINSICS to disable this error."
 #endif
 
 #ifndef _XMMINTRIN_H_INCLUDED
 #define _XMMINTRIN_H_INCLUDED
 
 #if defined(__linux__) && defined(__ppc64__)
 
 /* Define four value permute mask */

 
 /* Get _mm_malloc () and _mm_free ().  */
 #if __STDC_HOSTED__
 #include <mm_malloc.h>
 #endif
 
 /* The Intel API is flexible enough that we must allow aliasing with other
    vector types, and their scalar components.  */
-typedef float __m128 __attribute__ ((__vector_size__ (16), __may_alias__));
+typedef vector float __m128 __attribute__((__may_alias__));
 
 /* Unaligned version of the same type.  */
-typedef float __m128_u __attribute__ ((__vector_size__ (16), __may_alias__,
-               __aligned__ (1)));
+typedef vector float __m128_u __attribute__((__may_alias__, __aligned__(1)));
 
 /* Internal data types for implementing the intrinsics.  */
-typedef float __v4sf __attribute__ ((__vector_size__ (16)));
+typedef vector float __v4sf;
 
 /* Create an undefined vector.  */
 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
 _mm_undefined_ps (void)
 {
   __m128 __Y = __Y;
   return __Y;
 }

clang/test/CodeGen/ppc-xmmintrin.c

 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
 // REQUIRES: powerpc-registered-target
 
 // RUN: %clang -S -emit-llvm -target powerpc64-unknown-linux-gnu -mcpu=pwr8 -ffreestanding -DNO_WARN_X86_INTRINSICS %s \
 // RUN:   -fno-discard-value-names -mllvm -disable-llvm-optzns -o - | llvm-cxxfilt -n | FileCheck %s --check-prefixes=CHECK,CHECK-BE
+// RUN: %clang -x c++ -fsyntax-only -target powerpc64-unknown-linux-gnu -mcpu=pwr8 -ffreestanding -DNO_WARN_X86_INTRINSICS %s \
+// RUN:   -fno-discard-value-names -mllvm -disable-llvm-optzns
 // RUN: %clang -S -emit-llvm -target powerpc64le-unknown-linux-gnu -mcpu=pwr8 -ffreestanding -DNO_WARN_X86_INTRINSICS %s \
 // RUN:   -fno-discard-value-names -mllvm -disable-llvm-optzns -o - | llvm-cxxfilt -n | FileCheck %s --check-prefixes=CHECK,CHECK-LE
+// RUN: %clang -x c++ -fsyntax-only -target powerpc64le-unknown-linux-gnu -mcpu=pwr8 -ffreestanding -DNO_WARN_X86_INTRINSICS %s \
+// RUN:   -fno-discard-value-names -mllvm -disable-llvm-optzns
 
 #include <xmmintrin.h>
 
 __m128 res, m1, m2;
 __m64 res64, ms[2];
 float fs[4];
 int i, i2;
 long long i64;

 // CHECK-NEXT: store i64 [[REG709]], i64* [[REG711:[0-9a-zA-Z_%.]+]], align 8
 // CHECK-NEXT: [[REG712:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG710]], align 8
 // CHECK-NEXT: [[REG713:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG711]], align 8
 // CHECK-NEXT: [[REG714:[0-9a-zA-Z_%.]+]] = call i64 @_mm_mulhi_pu16(i64 [[REG712]], i64 [[REG713]])
 // CHECK-NEXT: ret i64 [[REG714]]
 
 void __attribute__((noinline))
 test_prefetch() {
-  _mm_prefetch(ms, i);
+  _mm_prefetch(ms, (enum _mm_hint)i);
 }
 
 // CHECK-LABEL: @test_prefetch
 
 // CHECK: define available_externally void @_mm_prefetch
 // CHECK: store i8* {{[0-9a-zA-Z_%.]+}}, i8** {{[0-9a-zA-Z_%.]+}}, align 8
 // CHECK-NEXT: store i32 {{[0-9a-zA-Z_%.]+}}, i32* {{[0-9a-zA-Z_%.]+}}, align 4
 // CHECK-NEXT: [[REG715:[0-9a-zA-Z_%.]+]] = load i8*, i8** {{[0-9a-zA-Z_%.]+}}, align 8