Skip to content

Fix CORTEXA72 Level-3 correctness (SYMM/TRMM/SYRK) #6002

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

Open
hugomeiland wants to merge 6 commits into
base: develop
Choose a base branch
from
Open

Fix CORTEXA72 Level-3 correctness (SYMM/TRMM/SYRK) #6002

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
c1594a0
Fix CORTEXA72 Level-3 correctness (SYMM/TRMM/SYRK)
hugomeiland Aug 28, 2026
528e873
syr2k: use division for UNROLL_MN stepping when MN is non-PoT
hugomeiland Aug 28, 2026
1effe16
CORTEXA72: set DGEMM_UNROLL_MN=8 for SYRK/SYR2K B packs
hugomeiland Aug 28, 2026
4201c78
CORTEXA72: set DGEMM_UNROLL_MN=LCM(6,8)=24 for SYRK panels
hugomeiland Aug 28, 2026
bad3679
CORTEXA72: add ASM DTRMM 6x8 microkernel
hugomeiland Aug 29, 2026
c8a6e79
CORTEXA72: fix DTRSM LN rem indexing for non-PoT UNROLL_M
hugomeiland Aug 29, 2026
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
3 changes: 2 additions & 1 deletion driver/level3/syr2k_kernel.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,7 +144,8 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT alpha_r,

int mm, nn;

mm = (loop & ~(GEMM_UNROLL_MN - 1));
/* Division form works for non-PoT UNROLL_MN (e.g. CORTEXA72 MN=24). */
mm = (loop / GEMM_UNROLL_MN) * GEMM_UNROLL_MN;
nn = MIN(GEMM_UNROLL_MN, n - loop);

#ifndef LOWER
Expand Down
3 changes: 1 addition & 2 deletions kernel/arm64/KERNEL.CORTEXA72
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,11 +6,10 @@ include $(KERNELDIR)/KERNEL.CORTEXA57
#
# Stock generic/gemm_{t,n}copy_6.c packs as 4+2, not contiguous MR=6 —
# use the panel packers. NR=8 reuses the asm dgemm_{n,t}copy_8.S.
# No dtrmm_kernel_6x8.S yet; fall back to generic 2x2 for DTRMM.
# Stock generic/trsm_kernel_*.c assumes power-of-2 UNROLL_M; use the
# UNROLLM6-aware kernels (same idea as loongarch64 UNROLLN6).
DGEMMKERNEL = dgemm_kernel_6x8_cortexa72.S
DTRMMKERNEL = ../generic/trmmkernel_2x2.c
DTRMMKERNEL = dtrmm_kernel_6x8_cortexa72.S

DTRSMKERNEL_LN = trsm_kernel_LN_UNROLLM6.c
DTRSMKERNEL_LT = trsm_kernel_LT_UNROLLM6.c
Expand Down
Loading
Loading