amd/blis
1
0
Fork 0
mirror of https://github.com/amd/blis.git synced 2026-05-24 18:34:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

AOCL Dynamic for zen3 dcopy

Browse Source 
-  Create seperate AOCL Dynamic values for
   multithreading dcopy API for zen1, zen2 and zen3

AMD-Internal: [CPUPL-5238]
Change-Id: I42f56393716edeeace8bfe71d7adab0ba7325b47
This commit is contained in:
Hari Govind S
2024-08-01 13:47:12 +05:30
parent 9843bd0317
commit f2acd4fd49
1 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
frame/base/bli_rntm.c
View File

@@ -2120,9 +2120,6 @@ BLIS_INLINE void aocl_dcopyv_dynamic
break;
case BLIS_ARCH_ZEN4:
case BLIS_ARCH_ZEN:
case BLIS_ARCH_ZEN2:
case BLIS_ARCH_ZEN3:
if ( n_elem <= 17000 )
*nt_ideal = 1;
@@ -2134,6 +2131,18 @@ BLIS_INLINE void aocl_dcopyv_dynamic
*nt_ideal = 8;
// dcopy does not scale with more than 8 threads
break;
case BLIS_ARCH_ZEN:
case BLIS_ARCH_ZEN2:
case BLIS_ARCH_ZEN3:
if ( n_elem <= 17000 )
*nt_ideal = 1;
else if (n_elem <= 52200)
*nt_ideal = 4;
else
*nt_ideal = 8;
// dcopy does not scale with more than 8 threads
break;
default:
// Without this default condition, compiler will throw