From ea25ba255a640a6f96bd349c59ed724d7a131ba3 Mon Sep 17 00:00:00 2001 From: Kiran Varaganti Date: Fri, 31 May 2019 15:14:22 +0530 Subject: [PATCH] Added back BLIS_ENABLE_ZEN_BLOCK_SIZES macro to zen configuration, this is same as release 1.3. This was added before to improve DGEMM Multithreaded scalability on Naples for when number of threads is greater than 16. By mistake this got deleted in many changes done for 2.0 release, now we are adding this change back., in bli_gemm_front.c - code clean Change-Id: I6827b58d2dab1041fe182fef5a007b679ac4bb1f --- config/zen/bli_cntx_init_zen.c | 3 ++- config/zen/bli_family_zen.h | 2 +- frame/3/gemm/bli_gemm_front.c | 6 ------ 3 files changed, 3 insertions(+), 8 deletions(-) diff --git a/config/zen/bli_cntx_init_zen.c b/config/zen/bli_cntx_init_zen.c index bcf793cb8..09ca2dee0 100644 --- a/config/zen/bli_cntx_init_zen.c +++ b/config/zen/bli_cntx_init_zen.c @@ -136,6 +136,7 @@ void bli_cntx_init_zen( cntx_t* cntx ) */ + // Zen optmized level 3 cache block sizes #ifdef BLIS_ENABLE_ZEN_BLOCK_SIZES #if BLIS_ENABLE_SINGLE_INSTANCE_BLOCK_SIZES @@ -155,7 +156,7 @@ void bli_cntx_init_zen( cntx_t* cntx ) bli_blksz_init_easy( &blkszs[ BLIS_NC ], 4080, 4080, 4080, 4080 ); #endif - bli_blksz_init_easy( &blkszs[ BLIS_NC ], 4080, 2040, 4080, 4080 ); + //bli_blksz_init_easy( &blkszs[ BLIS_NC ], 4080, 2040, 4080, 4080 ); bli_blksz_init_easy( &blkszs[ BLIS_AF ], 8, 8, -1, -1 ); bli_blksz_init_easy( &blkszs[ BLIS_DF ], 8, 8, -1, -1 ); diff --git a/config/zen/bli_family_zen.h b/config/zen/bli_family_zen.h index 01544e8b5..526e3a8b0 100644 --- a/config/zen/bli_family_zen.h +++ b/config/zen/bli_family_zen.h @@ -42,7 +42,7 @@ #define BLIS_DEFAULT_MR_THREAD_MAX 1 #define BLIS_DEFAULT_NR_THREAD_MAX 1 - +#define BLIS_ENABLE_ZEN_BLOCK_SIZES #define BLIS_ENABLE_SMALL_MATRIX #define BLIS_ENABLE_SMALL_MATRIX_TRSM diff --git a/frame/3/gemm/bli_gemm_front.c b/frame/3/gemm/bli_gemm_front.c index 610c86f24..bb6f2ee3b 100644 --- a/frame/3/gemm/bli_gemm_front.c +++ b/frame/3/gemm/bli_gemm_front.c @@ -52,13 +52,7 @@ void bli_gemm_front obj_t a_local; obj_t b_local; obj_t c_local; -#if 0 - gint_t M = bli_obj_length( c ); - gint_t N = bli_obj_width( c ); - gint_t K = bli_obj_width( a ); - if( !(M && N && K)) return; -#endif #ifdef BLIS_ENABLE_SMALL_MATRIX