weaselab/conflict-set
2
0
Fork 0
Code Issues 4 Packages Releases 13 Activity

Fixes for symbol multi-versioning with avx512f
Some checks failed
weaselab/conflict-set/pipeline/head There was a failure building this commit
Details

Browse Source
This commit is contained in:
Andrew Noyes
2024-06-29 22:20:50 -07:00
parent 11a022dcf7
commit 0b839b9d7e
2 changed files with 25 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
ConflictSet.cpp
View File

@@ -38,6 +38,12 @@ limitations under the License.
#include <arm_neon.h> #include <arm_neon.h>
#endif #endif
#if defined(__has_feature)
# if __has_feature(thread_sanitizer)
#define __SANITIZE_THREAD__
# endif
#endif
#include <memcheck.h> #include <memcheck.h>
using namespace weaselab; using namespace weaselab;
@@ -1767,11 +1773,17 @@ downLeftSpine:
} }
#ifdef HAS_AVX #ifdef HAS_AVX
__attribute__((target("default"))) uint32_t compare16_32bit(const uint32_t *vs, #ifndef __SANITIZE_THREAD__
uint32_t rv) { __attribute__((target("default")))
#endif
uint32_t compare16_32bit(const InternalVersionT *vs,
InternalVersionT rv) {
uint32_t compared = 0;
__m128i w[4]; __m128i w[4];
memcpy(w, vs, sizeof(w)); memcpy(w, vs, sizeof(w));
const auto rvVec = _mm_set1_epi32(rv); uint32_t r;
memcpy(&r, &rv, sizeof(r));
const auto rvVec = _mm_set1_epi32(r);
const auto zero = _mm_setzero_si128(); const auto zero = _mm_setzero_si128();
for (int i = 0; i < 4; ++i) { for (int i = 0; i < 4; ++i) {
compared |= compared |=
@@ -1781,14 +1793,18 @@ __attribute__((target("default"))) uint32_t compare16_32bit(const uint32_t *vs,
return compared; return compared;
} }
__attribute__((target("avx512f"))) uint32_t compare16_32bit(const uint32_t *vs, #ifndef __SANITIZE_THREAD__
uint32_t rv) { __attribute__((target("avx512f"))) uint32_t compare16_32bit(const InternalVersionT *vs,
__mm512i w; InternalVersionT rv) {
__m512i w;
memcpy(&w, vs, sizeof(w)); memcpy(&w, vs, sizeof(w));
return _mm512_cmpgt_epi32_mask(_mm512_sub_epi32(w, _mm512_set1_epi32(rv)), uint32_t r;
memcpy(&r, &rv, sizeof(r));
return _mm512_cmpgt_epi32_mask(_mm512_sub_epi32(w, _mm512_set1_epi32(r)),
_mm512_setzero()); _mm512_setzero());
} }
#endif #endif
#endif
// Returns true if v[i] <= readVersion for all i such that begin <= is[i] < end // Returns true if v[i] <= readVersion for all i such that begin <= is[i] < end
// Preconditions: begin <= end, end - begin < 256 // Preconditions: begin <= end, end - begin < 256

2
symbol-imports.txt
View File

@@ -1,4 +1,6 @@
_GLOBAL_OFFSET_TABLE_ _GLOBAL_OFFSET_TABLE_
__cpu_indicator_init
__cpu_model
__stack_chk_fail@GLIBC_2.4 __stack_chk_fail@GLIBC_2.4
__tls_get_addr@GLIBC_2.3 __tls_get_addr@GLIBC_2.3
abort@GLIBC_2.2.5 abort@GLIBC_2.2.5