From 4158e7fffef676882ca4401712de8237679fb356 Mon Sep 17 00:00:00 2001 From: "Devrajegowda, Kiran" Date: Wed, 23 Oct 2019 10:33:43 +0530 Subject: [PATCH] missed changes while rebasing field's SUP code Change-Id: I560b93c42901ca2bbd4c22e833f55ba884a38a50 --- config/zen2/make_defs.mk | 41 ++++++++++++++++++++++++----------- frame/3/trsm/bli_trsm_front.c | 8 +++---- frame/base/bli_clock.c | 3 --- 3 files changed, 32 insertions(+), 20 deletions(-) diff --git a/config/zen2/make_defs.mk b/config/zen2/make_defs.mk index 404df2a19..832681ca6 100644 --- a/config/zen2/make_defs.mk +++ b/config/zen2/make_defs.mk @@ -33,32 +33,41 @@ # # -# FLAGS that are specific to 'zen2' architecture are added here. -# FLAGS that are common for all the AMD architectures are present in config/zen/amd_config.mk -# # Declare the name of the current configuration and add it to the # running list of configurations included by common.mk. THIS_CONFIG := zen2 #CONFIGS_INCL += $(THIS_CONFIG) -# Include file containing common flags for all AMD architectures -AMD_CONFIG_FILE := amd_config.mk -AMD_CONFIG_PATH := $(BASE_SHARE_PATH)/config/zen --include $(AMD_CONFIG_PATH)/$(AMD_CONFIG_FILE) # # --- Determine the C compiler and related flags --- # + +# NOTE: The build system will append these variables with various +# general-purpose/configuration-agnostic flags in common.mk. You +# may specify additional flags here as needed. +CPPROCFLAGS := +CMISCFLAGS := +CPICFLAGS := +CWARNFLAGS := + +ifneq ($(DEBUG_TYPE),off) +CDBGFLAGS := -g +endif + +ifeq ($(DEBUG_TYPE),noopt) +COPTFLAGS := -O0 +else +COPTFLAGS := -O3 -fomit-frame-pointer +endif + # Flags specific to optimized kernels. +CKOPTFLAGS := $(COPTFLAGS) ifeq ($(CC_VENDOR),gcc) # gcc 9.0 (clang ?) or later: -GCC_VERSION := $(strip $(shell gcc -dumpversion)) -ifeq ($(shell test $(GCC_VERSION) -ge 9; echo $$?),0) -CKVECFLAGS += -march=znver2 +#CKVECFLAGS := -mavx2 -mfpmath=sse -mfma -march=znver2 # gcc 6.0 (clang 4.0) or later: -else -CKVECFLAGS += -march=znver1 -mno-avx256-split-unaligned-store -endif +CKVECFLAGS := -mavx2 -mfpmath=sse -mfma -march=znver1 -mno-avx256-split-unaligned-store # gcc 4.9 (clang 3.5) or later: # possibly add zen-specific instructions: -mclzero -madx -mrdseed -mmwaitx -msha -mxsavec -mxsaves -mclflushopt -mpopcnt #CKVECFLAGS := -mavx2 -mfpmath=sse -mfma -march=bdver4 -mno-fma4 -mno-tbm -mno-xop -mno-lwp @@ -68,6 +77,12 @@ CKVECFLAGS := -mavx2 -mfpmath=sse -mfma -march=znver1 -mno-fma4 -mno-tbm -mn else $(error gcc or clang are required for this configuration.) endif +endif + +# Flags specific to reference kernels. +CROPTFLAGS := $(CKOPTFLAGS) +CRVECFLAGS := $(CKVECFLAGS) + # Store all of the variables here to new variables containing the # configuration name. $(eval $(call store-make-defs,$(THIS_CONFIG))) diff --git a/frame/3/trsm/bli_trsm_front.c b/frame/3/trsm/bli_trsm_front.c index 38af5388b..eb88913ea 100644 --- a/frame/3/trsm/bli_trsm_front.c +++ b/frame/3/trsm/bli_trsm_front.c @@ -53,10 +53,10 @@ void bli_trsm_front obj_t b_local; obj_t c_local; -int m = bli_obj_length(*b); -int n = bli_obj_width(*b); -float *L = a->buffer; - float *B = b->buffer; +//int m = bli_obj_length(*b); +//int n = bli_obj_width(*b); +//float *L = a->buffer; + // float *B = b->buffer; #ifdef PRINT_SMALL_TRSM_INFO printf("Side:: %c\n", side ? 'R' : 'L'); diff --git a/frame/base/bli_clock.c b/frame/base/bli_clock.c index 35f3361ef..f5bf2c49f 100644 --- a/frame/base/bli_clock.c +++ b/frame/base/bli_clock.c @@ -59,12 +59,9 @@ double bli_clock_min_diff( double time_min, double time_start ) // - under a nanosecond // is actually garbled due to the clocks being taken too closely together. if ( time_min <= 0.0 ) time_min = time_min_prev; -<<<<<<< HEAD -======= // To genuinely measure time for an application taking more than an hour, the below // line is commented. If wrongly measuring higher time we could always use previous_min. /* else if ( time_min > 3600.0 ) time_min = time_min_prev; */ ->>>>>>> This is a fix to floating-point exception error for BLIS SGEMM with larger matrix sizes. else if ( time_min < 1.0e-9 ) time_min = time_min_prev; return time_min;