floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #4150 from martin-frbg/armsve 

Fix runtime detection in ARMV8 DYNAMIC_ARCH to check SVE capability
This commit is contained in:
Martin Kroeker 2023-07-19 22:25:55 +02:00 committed by GitHub
parent 9e81a3a0a2 5c58994eb2
commit 8a27a274a1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
driver/others/dynamic_arm64.c
View File

@ -137,6 +137,8 @@ extern gotoblas_t gotoblas_CORTEXA55;
#endif
extern void openblas_warning(int verbose, const char * msg);
#define FALLBACK_VERBOSE 1
#define NEOVERSEN1_FALLBACK "OpenBLAS : Your OS does not support SVE instructions. OpenBLAS is using Neoverse N1 kernels as a fallback, which may give poorer performance.\n"
#define NUM_CORETYPES 13
@ -147,6 +149,9 @@ extern void openblas_warning(int verbose, const char * msg);
#ifndef HWCAP_CPUID
#define HWCAP_CPUID (1 << 11)
#endif
#ifndef HWCAP_SVE
#define HWCAP_SVE (1 << 22)
#endif
#define get_cpu_ftr(id, var) ({ \
__asm__ __volatile__ ("mrs %0, "#id : "=r" (var)); \
@ -281,8 +286,16 @@ static gotoblas_t *get_coretype(void) {
return &gotoblas_NEOVERSEN1;
#ifndef NO_SVE
case 0xd49:
if (!(getauxval(AT_HWCAP) & HWCAP_SVE)) {
openblas_warning(FALLBACK_VERBOSE, NEOVERSEN1_FALLBACK);
return &gotoblas_NEOVERSEN1;
} else
return &gotoblas_NEOVERSEN2;
case 0xd40:
if (!(getauxval(AT_HWCAP) & HWCAP_SVE)) {
openblas_warning(FALLBACK_VERBOSE, NEOVERSEN1_FALLBACK);
return &gotoblas_NEOVERSEN1;
}else
return &gotoblas_NEOVERSEV1;
#endif
case 0xd05: // Cortex A55