shadps4-emu/ext-cryptopp
1
0
Fork 0
mirror of https://github.com/shadps4-emu/ext-cryptopp.git synced 2024-11-23 09:59:42 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

More Clang workarounds on PowerPC (PR #901)

Browse Source
This commit is contained in:
Jeffrey Walton 2019-10-23 12:40:45 -04:00 committed by GitHub
parent 249eb807bf
commit 1f3e65fc35
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 15 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
.travis.yml
View File

@ -566,6 +566,14 @@ jobs:
- BUILD_MODE=ios
- BUILD_JOBS=2
- BUILD_ARCH=iPhoneSimulator
# We should be OK with PPC64 and GCC until Travis disallows
# access to the build machines. Clang is a differnt story.
# The LLVM devs have their heads so far up their ass I don't
# think we can un-fuck the mess they created on PowerPC. The
# LLVM devs appear to have given up and define everything
# from all the compilers to all the features. It does not
# seem to matter to them the compiler does not actually
# support what LLVM is advertising in the preprocessor.
- os: linux
name: Linux with GCC (all)
arch: ppc64le

23
blake2b_simd.cpp
View File

@ -760,45 +760,36 @@ void BLAKE2_Compress64_NEON(const byte* input, BLAKE2b_State& state)
inline uint64x2_p VecLoad64(const void* p)
{
#if defined(__xlc__) || defined(__xlC__) || defined(__clang__)
return (uint64x2_p)vec_xl(0, (uint8_t*)p);
#else
return (uint64x2_p)vec_vsx_ld(0, (uint8_t*)p);
#endif
return (uint64x2_p)VecLoad((const byte*)p);
}
inline uint64x2_p VecLoad64LE(const void* p)
{
#if __BIG_ENDIAN__
const uint8x16_p m = {7,6,5,4, 3,2,1,0, 15,14,13,12, 11,10,9,8};
const uint64x2_p v = VecLoad64(p);
return VecPermute(v, v, m);
return (uint64x2_p)VecPermute(VecLoad64(p), m, m);
#else
return VecLoad64(p);
return (uint64x2_p)VecLoad64(p);
#endif
}
inline void VecStore64(void* p, const uint64x2_p x)
{
#if defined(__xlc__) || defined(__xlC__) || defined(__clang__)
vec_xst((uint8x16_p)x,0,(uint8_t*)p);
#else
vec_vsx_st((uint8x16_p)x,0,(uint8_t*)p);
#endif
VecStore((uint8x16_p)x, (byte*)p);
}
inline void VecStore64LE(void* p, const uint64x2_p x)
{
#if __BIG_ENDIAN__
const uint8x16_p m = {7,6,5,4, 3,2,1,0, 15,14,13,12, 11,10,9,8};
VecStore64(p, VecPermute(x, x, m));
VecStore64(p, VecPermute(x, m, m));
#else
VecStore64(p, x);
#endif
}
template <unsigned int C>
inline uint64x2_p VecShiftLeftOctet(const uint64x2_p a, const uint64x2_p b)
inline uint64x2_p ShiftLeftOctet(const uint64x2_p a, const uint64x2_p b)
{
#if __BIG_ENDIAN__
return (uint64x2_p)vec_sld((uint8x16_p)a, (uint8x16_p)b, C);
@ -807,7 +798,7 @@ inline uint64x2_p VecShiftLeftOctet(const uint64x2_p a, const uint64x2_p b)
#endif
}
#define vec_shl_octet(a,b,c) VecShiftLeftOctet<c*8>(a, b)
#define vec_shl_octet(a,b,c) ShiftLeftOctet<c*8>(a, b)
// vec_mergeh(a,b) is equivalent to VecPermute(a,b,HH_MASK); and
// vec_mergel(a,b) is equivalent VecPermute(a,b,LL_MASK). Benchmarks