mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-17 17:11:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

SPU: Use unaligned AVX instructions for cmp_data_avx

Browse Source
This commit is contained in:
Eladash 2020-09-06 06:15:49 +03:00 committed by Ivan
parent e4abd3dc5a
commit 2688081656
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
rpcs3/Emu/Cell/SPUThread.cpp
View File

@ -84,10 +84,10 @@ static const bool s_tsx_haswell = utils::has_rtm() && !utils::has_mpx();
static FORCE_INLINE bool cmp_rdata_avx(const __m256i* lhs, const __m256i* rhs)
{
#if defined(_MSC_VER) || defined(__AVX__)
const __m256 x0 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_load_si256(lhs + 0)), _mm256_castsi256_ps(_mm256_load_si256(rhs + 0)));
const __m256 x1 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_load_si256(lhs + 1)), _mm256_castsi256_ps(_mm256_load_si256(rhs + 1)));
const __m256 x2 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_load_si256(lhs + 2)), _mm256_castsi256_ps(_mm256_load_si256(rhs + 2)));
const __m256 x3 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_load_si256(lhs + 3)), _mm256_castsi256_ps(_mm256_load_si256(rhs + 3)));
const __m256 x0 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_loadu_si256(lhs + 0)), _mm256_castsi256_ps(_mm256_loadu_si256(rhs + 0)));
const __m256 x1 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_loadu_si256(lhs + 1)), _mm256_castsi256_ps(_mm256_loadu_si256(rhs + 1)));
const __m256 x2 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_loadu_si256(lhs + 2)), _mm256_castsi256_ps(_mm256_loadu_si256(rhs + 2)));
const __m256 x3 = _mm256_xor_ps(_mm256_castsi256_ps(_mm256_loadu_si256(lhs + 3)), _mm256_castsi256_ps(_mm256_loadu_si256(rhs + 3)));
const __m256 c0 = _mm256_or_ps(x0, x1);
const __m256 c1 = _mm256_or_ps(x2, x3);
const __m256 c2 = _mm256_or_ps(c0, c1);
@ -95,10 +95,10 @@ static FORCE_INLINE bool cmp_rdata_avx(const __m256i* lhs, const __m256i* rhs)
#else
bool result = 0;
__asm__(
"vmovaps 0*32(%[lhs]), %%ymm0;" // load
"vmovaps 1*32(%[lhs]), %%ymm1;"
"vmovaps 2*32(%[lhs]), %%ymm2;"
"vmovaps 3*32(%[lhs]), %%ymm3;"
"vmovups 0*32(%[lhs]), %%ymm0;" // load
"vmovups 1*32(%[lhs]), %%ymm1;"
"vmovups 2*32(%[lhs]), %%ymm2;"
"vmovups 3*32(%[lhs]), %%ymm3;"
"vxorps 0*32(%[rhs]), %%ymm0, %%ymm0;" // compare
"vxorps 1*32(%[rhs]), %%ymm1, %%ymm1;"
"vxorps 2*32(%[rhs]), %%ymm2, %%ymm2;"