amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-04-20 15:48:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Bug Fix in INT8 reference un-reorder API

Browse Source 
- For int8/uint8 reorder function, the k dimension is made multiple of 4 to
 meet the alignment requirements.
 - Modified the logic to update the k_updated to use multiples of 4.

[AMD - Internal : SWLCSG - 3686 ]
This commit is contained in:
V, Varsha
2025-07-24 11:26:49 +05:30
committed by GitHub
parent 4bc5287f72
commit 8a86620753
1 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
addon/aocl_gemm/frame/s8s8s32/lpgemm_reorder_s8.c
View File

@@ -58,7 +58,7 @@ void unreorderb_nr64_s8s8s32os32_reference
dim_t k = b->length;
dim_t k_updated = k;
k_updated += (k_updated & 0x3);
k_updated = make_multiple_of_n( k, 4 );
dim_t n_threads = bli_rntm_num_threads( rntm );
n_threads = ( n_threads > 0 ) ? n_threads : 1;
@@ -101,12 +101,11 @@ void unreorderb_nr64_s8s8s32os32_reference
{
dim_t kc0 = bli_min( ( k - pc ), KC );
// k needs to be a multiple of 2 so that it can be used with dpbf
// k needs to be a multiple of 4 so that it can be used with dpbf
// instruction. Padding is added in cases this condition is not
// satisfied, and therefore the k offset used for packed/reordered
// buffer needs to be updated.
dim_t kc0_updated = kc0;
kc0_updated += (kc0_updated & 0x3);
dim_t kc0_updated = make_multiple_of_n( kc0, 4 );
unpackb_nr64_s8_reference
(