Skip to content
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

Jump to bottom

Tune generic SVE parameters closer to other SVE cores and add tunings to baseline AArch64 #4833

Draft
wants to merge 2 commits into
base: develop
Choose a base branch
from
Draft

Tune generic SVE parameters closer to other SVE cores and add tunings to baseline AArch64 #4833

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
22d8803
Tune generic SVE constants closer to other SVE cores
Mousius Aug 1, 2024
5f8744d
Add tunings for baseline AArch64
Mousius Aug 1, 2024
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
26 changes: 18 additions & 8 deletions param.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3671,8 +3671,10 @@ Until then, just keep it different than DGEMM_DEFAULT_UNROLL_N to keep copy rout

#if defined(XDOUBLE) || defined(DOUBLE)
#define SWITCH_RATIO 8
#define GEMM_PREFERED_SIZE 4
#else
#define SWITCH_RATIO 16
#define GEMM_PREFERED_SIZE 8
#endif

#define SGEMM_DEFAULT_UNROLL_M 4 // Actually 1VL (8) but kept seperate to keep copies seperate
Expand All @@ -3689,13 +3691,13 @@ Until then, just keep it different than DGEMM_DEFAULT_UNROLL_N to keep copy rout
#define ZGEMM_DEFAULT_UNROLL_N 4
#define ZGEMM_DEFAULT_UNROLL_MN 16

#define SGEMM_DEFAULT_P 128
#define DGEMM_DEFAULT_P 160
#define SGEMM_DEFAULT_P 240
#define DGEMM_DEFAULT_P 240
#define CGEMM_DEFAULT_P 128
#define ZGEMM_DEFAULT_P 128

#define SGEMM_DEFAULT_Q 352
#define DGEMM_DEFAULT_Q 128
#define SGEMM_DEFAULT_Q 640
#define DGEMM_DEFAULT_Q 320
#define CGEMM_DEFAULT_Q 224
#define ZGEMM_DEFAULT_Q 112

Expand All @@ -3706,6 +3708,14 @@ Until then, just keep it different than DGEMM_DEFAULT_UNROLL_N to keep copy rout

#else /* Other/undetected ARMv8 cores */

#if defined(XDOUBLE) || defined(DOUBLE)
#define SWITCH_RATIO 8
#define GEMM_PREFERED_SIZE 4
#else
#define SWITCH_RATIO 16
#define GEMM_PREFERED_SIZE 8
#endif

#define SGEMM_DEFAULT_UNROLL_M 16
#define SGEMM_DEFAULT_UNROLL_N 4

Expand All @@ -3718,13 +3728,13 @@ Until then, just keep it different than DGEMM_DEFAULT_UNROLL_N to keep copy rout
#define ZGEMM_DEFAULT_UNROLL_M 4
#define ZGEMM_DEFAULT_UNROLL_N 4

#define SGEMM_DEFAULT_P 128
#define DGEMM_DEFAULT_P 160
#define SGEMM_DEFAULT_P 240
#define DGEMM_DEFAULT_P 240
#define CGEMM_DEFAULT_P 128
#define ZGEMM_DEFAULT_P 128

#define SGEMM_DEFAULT_Q 352
#define DGEMM_DEFAULT_Q 128
#define SGEMM_DEFAULT_Q 640
#define DGEMM_DEFAULT_Q 320
#define CGEMM_DEFAULT_Q 224
#define ZGEMM_DEFAULT_Q 112

Expand Down
Loading