Block parameters tuning to improve sgemm performance on Rome

Details:
    - Tuned block sizes to get better performance for sgemm default path.

Change-Id: I892e8642fa2d03a07a6d53537131536e6b1b091e
Signed-off-by: Kiran N D <kiran.Devrajegowda@amd.com>
AMD-Internal: [CPUPL-832]
This commit is contained in:
Devrajegowda, Kiran
2020-04-16 12:21:47 +05:30
committed by Kiran Devrajegowda
parent 139fbbb77f
commit 1c76723320

View File

@@ -115,9 +115,9 @@ void bli_cntx_init_zen2( cntx_t* cntx )
bli_blksz_init_easy( &blkszs[ BLIS_KC ], 256, 512, 256, 256 );
bli_blksz_init_easy( &blkszs[ BLIS_NC ], 4080, 2040, 4080, 4080 );
#else
bli_blksz_init_easy( &blkszs[ BLIS_MC ], 144, 72, 144, 72 );
bli_blksz_init_easy( &blkszs[ BLIS_KC ], 256, 256, 256, 256 );
bli_blksz_init_easy( &blkszs[ BLIS_NC ], 4080, 4080, 4080, 4080 );
bli_blksz_init_easy( &blkszs[ BLIS_MC ], 768, 72, 144, 72 );
bli_blksz_init_easy( &blkszs[ BLIS_KC ], 1024, 256, 256, 256 );
bli_blksz_init_easy( &blkszs[ BLIS_NC ], 8160, 4080, 4080, 4080 );
#endif
bli_blksz_init_easy( &blkszs[ BLIS_AF ], 8, 8, -1, -1 );