happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Disable SVE (except SVE2_128) for MatMul due to compiler crash 

PiperOrigin-RevId: 886190686
This commit is contained in:
Jan Wassenberg 2026-03-19 08:20:17 -07:00 committed by Copybara-Service
parent 90f3de7f15
commit 79f2bf7a07
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
compression/types.h
View File

@ -42,8 +42,9 @@ namespace gcpp {
#define HWY_DISABLED_TARGETS (HWY_SCALAR | HWY_ALL_NEON)
#elif HWY_ARCH_ARM_A64
// We do not yet use AES (e.g. for random generation), hence NEON is the same
// as NEON_WITHOUT_AES. Also skip SVE because SVE2_128 and SVE_256 cover most.
#define GEMMA_DISABLED_TARGETS (HWY_SCALAR | HWY_NEON | HWY_SVE)
// as NEON_WITHOUT_AES. Also skip SVE except SVE2_128 due to compiler bugs.
#define GEMMA_DISABLED_TARGETS \
(HWY_SCALAR | HWY_NEON | HWY_SVE | HWY_SVE2 | HWY_SVE_256)
#elif HWY_ARCH_X86
// Skip anything older than Haswell (2013); use Zen4/SPR for recent CPUs.
// Although we do not use SPR's F16, Zen4 is only enabled for AMD. We do not