frozenleaves/oneDNN
1
0
Fork 0
mirror of https://github.com/uxlfoundation/oneDNN.git synced 2025-10-20 18:43:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

fixup: cpu: x64: brgemm: fix handling big offsets

Browse Source
This commit is contained in:
xuxinzen
2025-10-13 17:48:00 -04:00
committed by Dmitrii Zarukin
parent c9affbff3c
commit d1a336fe78
1 changed files with 11 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
src/cpu/x64/brgemm/jit_brgemm_kernel.cpp
View File

@ -2380,12 +2380,18 @@ void jit_brgemm_kernel_t<Wmm>::gemm_microkernel(dim_t bd_block2,
= B_offset(prefetch_count_B++, rd)
+ static_cast<dim_t>(brg.LDB) * brg.rd_block
* brg.typesize_B;
if (is_superset(brg.isa_impl, avx512_core)) {
prefetcht0(EVEX_compress_addr_safe(reg_aux_B,
prefetch_offset, reg_tmp_microkernel));
// Only use EVEX_compress_addr_safe/make_safe_addr
// when prefetch_offset > INT_MAX forr perf purpose
if (prefetch_offset <= INT_MAX) {
prefetcht0(ptr[reg_aux_B + prefetch_offset]);
} else {
prefetcht0(make_safe_addr(reg_aux_B, prefetch_offset,
reg_tmp_microkernel));
if (is_superset(brg.isa_impl, avx512_core)) {
prefetcht0(EVEX_compress_addr_safe(reg_aux_B,
prefetch_offset, reg_tmp_microkernel));
} else {
prefetcht0(make_safe_addr(reg_aux_B,
prefetch_offset, reg_tmp_microkernel));
}
}
}
for (dim_t ld = 0; ld < ld_block2; ld++) {