github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

fix unit-internal release builds using __builtin_constant_p() as well

Browse Source
This commit is contained in:
Konstantinos Margaritis 2021-11-25 15:09:01 +02:00
parent 00384c9e37
commit 7ceca78db4
1 changed files with 49 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

57
src/util/supervector/arch/x86/impl.cpp
View File

@ -520,16 +520,18 @@ really_inline SuperVector<16> SuperVector<16>::loadu_maskz(void const *ptr, uint
return mask & v; return mask & v;
} }
#ifdef HS_OPTIMIZE
template<> template<>
really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset) really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset)
{ {
#if defined(HAVE__BUILTIN_CONSTANT_P)
if (__builtin_constant_p(offset)) {
if (offset == 16) {
return *this;
} else {
return {_mm_alignr_epi8(u.v128[0], other.u.v128[0], offset)}; return {_mm_alignr_epi8(u.v128[0], other.u.v128[0], offset)};
} }
#else }
template<> #endif
really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset)
{
switch(offset) { switch(offset) {
case 0: return other; break; case 0: return other; break;
case 1: return {_mm_alignr_epi8(u.v128[0], other.u.v128[0], 1)}; break; case 1: return {_mm_alignr_epi8(u.v128[0], other.u.v128[0], 1)}; break;
@ -551,7 +553,6 @@ really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, in
} }
return *this; return *this;
} }
#endif
template<> template<>
template<> template<>
@ -1037,10 +1038,11 @@ really_inline SuperVector<32> SuperVector<32>::vshr(uint8_t const N) const
return vshr_256(N); return vshr_256(N);
} }
#ifdef HS_OPTIMIZE
template <> template <>
really_inline SuperVector<32> SuperVector<32>::operator>>(uint8_t const N) const really_inline SuperVector<32> SuperVector<32>::operator>>(uint8_t const N) const
{ {
#if defined(HAVE__BUILTIN_CONSTANT_P)
if (__builtin_constant_p(N)) {
// As found here: https://stackoverflow.com/questions/25248766/emulating-shifts-on-32-bytes-with-avx // As found here: https://stackoverflow.com/questions/25248766/emulating-shifts-on-32-bytes-with-avx
if (N < 16) { if (N < 16) {
return {_mm256_alignr_epi8(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(2, 0, 0, 1)), u.v256[0], N)}; return {_mm256_alignr_epi8(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(2, 0, 0, 1)), u.v256[0], N)};
@ -1050,18 +1052,15 @@ really_inline SuperVector<32> SuperVector<32>::operator>>(uint8_t const N) const
return {_mm256_srli_si256(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(2, 0, 0, 1)), N - 16)}; return {_mm256_srli_si256(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(2, 0, 0, 1)), N - 16)};
} }
} }
#else #endif
template <>
really_inline SuperVector<32> SuperVector<32>::operator>>(uint8_t const N) const
{
return vshr_256(N); return vshr_256(N);
} }
#endif
#ifdef HS_OPTIMIZE
template <> template <>
really_inline SuperVector<32> SuperVector<32>::operator<<(uint8_t const N) const really_inline SuperVector<32> SuperVector<32>::operator<<(uint8_t const N) const
{ {
#if defined(HAVE__BUILTIN_CONSTANT_P)
if (__builtin_constant_p(N)) {
// As found here: https://stackoverflow.com/questions/25248766/emulating-shifts-on-32-bytes-with-avx // As found here: https://stackoverflow.com/questions/25248766/emulating-shifts-on-32-bytes-with-avx
if (N < 16) { if (N < 16) {
return {_mm256_alignr_epi8(u.v256[0], _mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(0, 0, 2, 0)), 16 - N)}; return {_mm256_alignr_epi8(u.v256[0], _mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(0, 0, 2, 0)), 16 - N)};
@ -1071,13 +1070,9 @@ really_inline SuperVector<32> SuperVector<32>::operator<<(uint8_t const N) const
return {_mm256_slli_si256(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(0, 0, 2, 0)), N - 16)}; return {_mm256_slli_si256(_mm256_permute2x128_si256(u.v256[0], u.v256[0], _MM_SHUFFLE(0, 0, 2, 0)), N - 16)};
} }
} }
#else #endif
template <>
really_inline SuperVector<32> SuperVector<32>::operator<<(uint8_t const N) const
{
return vshl_256(N); return vshl_256(N);
} }
#endif
template<> template<>
really_inline SuperVector<32> SuperVector<32>::Ones_vshr(uint8_t const N) really_inline SuperVector<32> SuperVector<32>::Ones_vshr(uint8_t const N)
@ -1132,16 +1127,18 @@ really_inline SuperVector<32> SuperVector<32>::loadu_maskz(void const *ptr, uint
#endif #endif
} }
#ifdef HS_OPTIMIZE
template<> template<>
really_inline SuperVector<32> SuperVector<32>::alignr(SuperVector<32> &other, int8_t offset) really_inline SuperVector<32> SuperVector<32>::alignr(SuperVector<32> &other, int8_t offset)
{ {
#if defined(HAVE__BUILTIN_CONSTANT_P)
if (__builtin_constant_p(offset)) {
if (offset == 16) {
return *this;
} else {
return {_mm256_alignr_epi8(u.v256[0], other.u.v256[0], offset)}; return {_mm256_alignr_epi8(u.v256[0], other.u.v256[0], offset)};
} }
#else }
template<> #endif
really_inline SuperVector<32> SuperVector<32>::alignr(SuperVector<32> &other, int8_t offset)
{
// As found here: https://stackoverflow.com/questions/8517970/mm-alignr-epi8-palignr-equivalent-in-avx2#8637458 // As found here: https://stackoverflow.com/questions/8517970/mm-alignr-epi8-palignr-equivalent-in-avx2#8637458
switch (offset){ switch (offset){
case 0 : return _mm256_set_m128i(_mm_alignr_epi8(u.v128[0], other.u.v128[1], 0), _mm_alignr_epi8(other.u.v128[1], other.u.v128[0], 0)); break; case 0 : return _mm256_set_m128i(_mm_alignr_epi8(u.v128[0], other.u.v128[1], 0), _mm_alignr_epi8(other.u.v128[1], other.u.v128[0], 0)); break;
@ -1180,7 +1177,6 @@ really_inline SuperVector<32> SuperVector<32>::alignr(SuperVector<32> &other, in
} }
return *this; return *this;
} }
#endif
template<> template<>
template<> template<>
@ -1772,16 +1768,18 @@ really_inline SuperVector<64> SuperVector<64>::pshufb_maskz(SuperVector<64> b, u
return {_mm512_maskz_shuffle_epi8(mask, u.v512[0], b.u.v512[0])}; return {_mm512_maskz_shuffle_epi8(mask, u.v512[0], b.u.v512[0])};
} }
#ifdef HS_OPTIMIZE
template<> template<>
really_inline SuperVector<64> SuperVector<64>::alignr(SuperVector<64> &l, int8_t offset) really_inline SuperVector<64> SuperVector<64>::alignr(SuperVector<64> &l, int8_t offset)
{ {
#if defined(HAVE__BUILTIN_CONSTANT_P)
if (__builtin_constant_p(offset)) {
if (offset == 16) {
return *this;
} else {
return {_mm512_alignr_epi8(u.v512[0], l.u.v512[0], offset)}; return {_mm512_alignr_epi8(u.v512[0], l.u.v512[0], offset)};
} }
#else }
template<> #endif
really_inline SuperVector<64> SuperVector<64>::alignr(SuperVector<64> &l, int8_t offset)
{
if(offset == 0) { if(offset == 0) {
return *this; return *this;
} else if (offset < 32){ } else if (offset < 32){
@ -1802,7 +1800,6 @@ really_inline SuperVector<64> SuperVector<64>::alignr(SuperVector<64> &l, int8_t
return *this; return *this;
} }
} }
#endif
#endif // HAVE_AVX512 #endif // HAVE_AVX512