Skip to content

Fix instabilities in CGEMM/CTRMM/DNRM2 on Apple M1/M2 under OSX #4003

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Apr 18, 2023
Merged

Fix instabilities in CGEMM/CTRMM/DNRM2 on Apple M1/M2 under OSX #4003

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
cda2963
move ALPHA_I out of register 18 (reserved on OSX)
martin-frbg Apr 13, 2023
c7bbad0
Move ALPHA_I out of register 18 (reserved on OSX)
martin-frbg Apr 13, 2023
0b1acb0
Move ALPHA_I out of register 18 (reserved on OSX)
martin-frbg Apr 13, 2023
108a21e
Move ALPHA out of register 18 (reserved on OSX)
martin-frbg Apr 13, 2023
3727672
Improve workaround and keep compilers from optimizing it out
martin-frbg Apr 13, 2023
f096a33
Use long value fields for cpu ident on OSX
martin-frbg Apr 13, 2023
8be68fa
move declaration of sca to really keep the compiler from throwing it …
martin-frbg Apr 15, 2023
44164e3
revert "move alpha out of register 18" (out of PR scope, no SVE on Ap…
martin-frbg Apr 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions cpuid_arm64.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -267,9 +267,9 @@ int detect(void)
}
#else
#ifdef __APPLE__
sysctlbyname("hw.cpufamily",&value,&length,NULL,0);
if (value ==131287967|| value == 458787763 ) return CPU_VORTEX; //A12/M1
if (value == 3660830781) return CPU_VORTEX; //A15/M2
sysctlbyname("hw.cpufamily",&value64,&length64,NULL,0);
if (value64 ==131287967|| value64 == 458787763 ) return CPU_VORTEX; //A12/M1
if (value64 == 3660830781) return CPU_VORTEX; //A15/M2
#endif
return CPU_ARMV8;
#endif
Expand Down
2 changes: 1 addition & 1 deletion kernel/arm64/cgemm_kernel_8x4.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define pCRow3 x15
#define pA x16
#define alphaR w17
#define alphaI w18
#define alphaI w19

#define alpha0_R s10
#define alphaV0_R v10.s[0]
Expand Down
2 changes: 1 addition & 1 deletion kernel/arm64/cgemm_kernel_8x4_thunderx2t99.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define pCRow3 x15
#define pA x16
#define alphaR w17
#define alphaI w18
#define alphaI w19

#define alpha0_R s10
#define alphaV0_R v10.s[0]
Expand Down
8 changes: 4 additions & 4 deletions kernel/arm64/ctrmm_kernel_8x4.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,10 +49,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define pCRow3 x15
#define pA x16
#define alphaR w17
#define alphaI w18
#define temp x19
#define tempOffset x20
#define tempK x21
#define alphaI w19
#define temp x20
#define tempOffset x21
#define tempK x22

#define alpha0_R s10
#define alphaV0_R v10.s[0]
Expand Down
5 changes: 3 additions & 2 deletions kernel/arm64/dznrm2_thunderx2t99.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.


#include "common.h"

#include <float.h>
#include <arm_neon.h>

#if defined(SMP)
Expand Down Expand Up @@ -404,7 +404,8 @@ FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
#else
nrm2_compute(n, x, inc_x, &ssq, &scale);
#endif
if (fabs(scale) <1.e-300) return 0.;
volatile FLOAT sca = fabs(scale);
if (sca < DBL_MIN) return 0.;
ssq = sqrt(ssq) * scale;

return ssq;
Expand Down