floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refs #139. Added NO_AVX flag to use old Nehalem kernels on Sandy Bridge. 

For example, make NO_AVX=1 or make DYNAMIC_ARCH=1 NO_AVX=1
This commit is contained in:
Zhang Xianyi 2012-09-17 23:24:04 +08:00
parent 9419a43a7f
commit f76a384841
4 changed files with 34 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Makefile.rule
View File

@ -71,6 +71,10 @@ VERSION = 0.2.3
# If you want to disable CPU/Memory affinity on Linux. # If you want to disable CPU/Memory affinity on Linux.
# NO_AFFINITY = 1 # NO_AFFINITY = 1
# Don't use AVX kernel on Sandy Bridge. It is compatible with old compilers
# and OS. However, the performance is low.
# NO_AVX = 1
# If you would like to know minute performance report of GotoBLAS. # If you would like to know minute performance report of GotoBLAS.
# FUNCTION_PROFILE = 1 # FUNCTION_PROFILE = 1

18
Makefile.system
View File

@ -57,6 +57,10 @@ GEMM_MULTITHREAD_THRESHOLD=50
endif endif
GETARCH_FLAGS += -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD) GETARCH_FLAGS += -DGEMM_MULTITHREAD_THRESHOLD=$(GEMM_MULTITHREAD_THRESHOLD)
ifeq ($(NO_AVX), 1)
GETARCH_FLAGS += -DNO_AVX
endif
# This operation is expensive, so execution should be once. # This operation is expensive, so execution should be once.
ifndef GOTOBLAS_MAKEFILE ifndef GOTOBLAS_MAKEFILE
export GOTOBLAS_MAKEFILE = 1 export GOTOBLAS_MAKEFILE = 1
@ -247,11 +251,17 @@ endif
ifeq ($(DYNAMIC_ARCH), 1) ifeq ($(DYNAMIC_ARCH), 1)
ifeq ($(ARCH), x86) ifeq ($(ARCH), x86)
DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \ DYNAMIC_CORE = KATMAI COPPERMINE NORTHWOOD PRESCOTT BANIAS \
CORE2 PENRYN DUNNINGTON NEHALEM SANDYBRIDGE ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO CORE2 PENRYN DUNNINGTON NEHALEM ATHLON OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
ifneq ($(NO_AVX), 1)
DYNAMIC_CORE += SANDYBRIDGE
endif
endif endif
ifeq ($(ARCH), x86_64) ifeq ($(ARCH), x86_64)
DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM SANDYBRIDGE OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO DYNAMIC_CORE = PRESCOTT CORE2 PENRYN DUNNINGTON NEHALEM OPTERON OPTERON_SSE3 BARCELONA BOBCAT ATOM NANO
ifneq ($(NO_AVX), 1)
DYNAMIC_CORE += SANDYBRIDGE
endif
endif endif
ifndef DYNAMIC_CORE ifndef DYNAMIC_CORE
@ -562,6 +572,10 @@ ifeq ($(NO_LAPACKE), 1)
CCOMMON_OPT += -DNO_LAPACKE CCOMMON_OPT += -DNO_LAPACKE
endif endif
ifeq ($(NO_AVX), 1)
CCOMMON_OPT += -DNO_AVX
endif
ifdef SMP ifdef SMP
CCOMMON_OPT += -DSMP_SERVER CCOMMON_OPT += -DSMP_SERVER

7
cpuid_x86.c
View File

@ -40,6 +40,11 @@
#include <string.h> #include <string.h>
#include "cpuid.h" #include "cpuid.h"
#ifdef NO_AVX
#define CPUTYPE_SANDYBRIDGE CPUTYPE_NEHALEM
#define CORE_SANDYBRIDGE CORE_NEHALEM
#endif
#ifndef CPUIDEMU #ifndef CPUIDEMU
#if defined(__APPLE__) && defined(__i386__) #if defined(__APPLE__) && defined(__i386__)
@ -189,7 +194,9 @@ int get_cputype(int gettype){
if ((ecx & (1 << 9)) != 0) feature |= HAVE_SSSE3; if ((ecx & (1 << 9)) != 0) feature |= HAVE_SSSE3;
if ((ecx & (1 << 19)) != 0) feature |= HAVE_SSE4_1; if ((ecx & (1 << 19)) != 0) feature |= HAVE_SSE4_1;
if ((ecx & (1 << 20)) != 0) feature |= HAVE_SSE4_2; if ((ecx & (1 << 20)) != 0) feature |= HAVE_SSE4_2;
#ifndef NO_AVX
if ((ecx & (1 << 28)) != 0) feature |= HAVE_AVX; if ((ecx & (1 << 28)) != 0) feature |= HAVE_AVX;
#endif
if (have_excpuid() >= 0x01) { if (have_excpuid() >= 0x01) {
cpuid(0x80000001, &eax, &ebx, &ecx, &edx); cpuid(0x80000001, &eax, &ebx, &ecx, &edx);

8
driver/others/dynamic.c
View File

@ -60,8 +60,14 @@ extern gotoblas_t gotoblas_NEHALEM;
extern gotoblas_t gotoblas_OPTERON; extern gotoblas_t gotoblas_OPTERON;
extern gotoblas_t gotoblas_OPTERON_SSE3; extern gotoblas_t gotoblas_OPTERON_SSE3;
extern gotoblas_t gotoblas_BARCELONA; extern gotoblas_t gotoblas_BARCELONA;
extern gotoblas_t gotoblas_SANDYBRIDGE;
extern gotoblas_t gotoblas_BOBCAT; extern gotoblas_t gotoblas_BOBCAT;
#ifndef NO_AVX
extern gotoblas_t gotoblas_SANDYBRIDGE;
#else
//Use NEHALEM kernels for sandy bridge
#define gotoblas_SANDYBRIDGE gotoblas_NEHALEM
#endif
#define VENDOR_INTEL 1 #define VENDOR_INTEL 1
#define VENDOR_AMD 2 #define VENDOR_AMD 2