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

Additional optimizations to ZGEMM SUP and Tiny codepaths(ZEN4 and ZEN5)

Browse Source 
- Added a set of AVX512 fringe kernels(using masked loads and
  stores) in order to avoid rerouting to the GEMV typed API
  interface(when m = 1). This ensures uniformity in performance
  across the main and fringe cases, when the calls are multithreaded.

- Further tuned the thresholds to decide between ZGEMM Tiny, Small
  SUP and Native paths for ZEN4 and ZEN5 architectures(in case
  of parallel execution). This would account for additional
  combinations of the input dimensions.

- Moved the call to Tiny-ZGEMM before the BLIS object creation,
  since this code-path operates on raw buffers.

- Added the necessary test-cases for functional and memory testing
  of the newly added kernels.

AMD-Internal: [CPUPL-6378][CPUPL-6661]
Change-Id: I9af73d1b6ef82b26503d4fc373111132aee3afd6
This commit is contained in:
Vignesh Balasubramanian
2025-04-07 14:26:01 +05:30
committed by Vignesh Balasubramanian
parent 87c9230cac
commit b4b0887ca4
9 changed files with 1910 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1731
kernels/zen4/3/sup/bli_gemmsup_cv_zen4_z12x4m.c
View File

File diff suppressed because it is too large Load Diff

6
kernels/zen4/aocl_smart/bli_aocl_smart_zen4.c
View File

@@ -86,8 +86,10 @@ bool bli_cntx_gemmsup_thresh_is_met_zen4( obj_t* a, obj_t* b, obj_t* c, cntx_t*
// For skinny sizes where m and/or n is small
// The threshold for m is a single value, but for n, it is
// also based on the packing size of A, since the kernels are
// column preferential
if( ( m <= 84 ) || ( ( n <= 84 ) && ( m < 4000 ) ) ) return TRUE;
// column preferential
if( ( ( m <= 120 ) && ( n <= 7515 ) && ( k <= 128 ) ) ||
// ( ( m <= 96 ) && ( n <= 7515 ) && ( k <= 128 ) ) ||
( ( m <= 1200 ) && ( n <= 1200 ) && ( k <= 64 ) ) ) return TRUE;
// For all combinations in small sizes
if( ( m <= 216 ) && ( n <= 216 ) && ( k <= 216 ) ) return TRUE;

5
kernels/zen4/bli_kernels_zen4.h
View File

@@ -361,6 +361,11 @@ GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_8x3 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_8x2 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_8x1 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_fx4 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_fx3 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_fx2 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_fx1 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_4x4 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_4x3 )
GEMMSUP_KER_PROT( dcomplex, z, gemmsup_cv_zen4_asm_4x2 )

2
kernels/zen5/aocl_smart/bli_aocl_smart_zen5.c
View File

@@ -87,7 +87,7 @@ bool bli_cntx_gemmsup_thresh_is_met_zen5( obj_t* a, obj_t* b, obj_t* c, cntx_t*
// The threshold for m is a single value, but for n, it is
// also based on the packing size of A, since the kernels are
// column preferential
if( ( m <= 84 ) || ( ( n <= 84 ) && ( ( m * k ) <= 983040 ) ) ) return TRUE;
if( ( m <= 60 ) || ( ( n <= 60 ) && ( m <= 960 ) && ( k <= 16384 ) ) || ( k <= 8 ) ) return TRUE;
// For all combinations in small sizes
if( ( m <= 216 ) && ( n <= 216 ) && ( k <= 216 ) ) return TRUE;