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

Merge pull request #101 from xianyi/develop 

rebase
This commit is contained in:
Martin Kroeker 2020-10-16 20:48:58 +02:00 committed by GitHub
parent a85ac71633 9789375389
commit 25907e672b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 44 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
Makefile.x86
View File

@ -54,3 +54,19 @@ LIBATLAS = -L$(ATLASPATH)/32 -lcblas -lf77blas -latlas -lm
else else
LIBATLAS = -L$(ATLASPATH)/32 -lptf77blas -lptatlas -lpthread -lm LIBATLAS = -L$(ATLASPATH)/32 -lptf77blas -lptatlas -lpthread -lm
endif endif
ifdef HAVE_SSE3
ifndef DYNAMIC_ARCH
CCOMMON_OPT += -msse3
FCOMMON_OPT += -msse3
ifdef HAVE_SSSE3
CCOMMON_OPT += -mssse3
FCOMMON_OPT += -mssse3
endif
ifdef HAVE_SSE4_1
CCOMMON_OPT += -msse4.1
FCOMMON_OPT += -msse4.1
endif
endif
endif

9
cmake/cc.cmake
View File

@ -124,10 +124,19 @@ if (NOT DYNAMIC_ARCH)
if (HAVE_AVX) if (HAVE_AVX)
set (CCOMMON_OPT "${CCOMMON_OPT} -mavx") set (CCOMMON_OPT "${CCOMMON_OPT} -mavx")
endif () endif ()
if (HAVE_SSE)
set (CCOMMON_OPT "${CCOMMON_OPT} -msse")
endif ()
if (HAVE_SSE2)
set (CCOMMON_OPT "${CCOMMON_OPT} -msse2")
endif ()
if (HAVE_SSE3) if (HAVE_SSE3)
set (CCOMMON_OPT "${CCOMMON_OPT} -msse3") set (CCOMMON_OPT "${CCOMMON_OPT} -msse3")
endif () endif ()
if (HAVE_SSSE3) if (HAVE_SSSE3)
set (CCOMMON_OPT "${CCOMMON_OPT} -mssse3") set (CCOMMON_OPT "${CCOMMON_OPT} -mssse3")
endif () endif ()
if (HAVE_SSE4_1)
set (CCOMMON_OPT "${CCOMMON_OPT} -msse4.1")
endif ()
endif() endif()

12
cmake/system.cmake
View File

@ -70,9 +70,21 @@ if (DEFINED TARGET)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
endif() endif()
endif() endif()
if (DEFINED HAVE_SSE)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
endif()
if (DEFINED HAVE_SSE2)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse2")
endif()
if (DEFINED HAVE_SSE3) if (DEFINED HAVE_SSE3)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3") set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse3")
endif() endif()
if (DEFINED HAVE_SSSE3)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mssse3")
endif()
if (DEFINED HAVE_SSE4_1)
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse4.1")
endif()
endif() endif()
if (DEFINED TARGET) if (DEFINED TARGET)

5
kernel/Makefile
View File

@ -45,7 +45,10 @@ endif
ifdef TARGET_CORE ifdef TARGET_CORE
ifeq ($(TARGET_CORE), $(filter $(TARGET_CORE),PRESCOTT CORE2 PENRYN DUNNINGTON ATOM NANO SANDYBRIDGE HASWELL NEHALEM ZEN BARCELONA BOBCAT BULLDOZER PILEDRIVER EXCAVATOR STEAMROLLER OPTERON_SSE3)) ifeq ($(TARGET_CORE), $(filter $(TARGET_CORE),PRESCOTT CORE2 PENRYN DUNNINGTON ATOM NANO SANDYBRIDGE HASWELL NEHALEM ZEN BARCELONA BOBCAT BULLDOZER PILEDRIVER EXCAVATOR STEAMROLLER OPTERON_SSE3))
override CFLAGS += -msse3 -mssse3 -msse4.1 override CFLAGS += -msse -msse2 -msse3 -mssse3 -msse4.1
endif
ifeq ($(TARGET_CORE), $(filter $(TARGET_CORE),KATMAI COPPERMINE BANIAS NORTHWOOD ATHLON OPTERON))
override CFLAGS += -msse -msse2
endif endif
ifeq ($(TARGET_CORE), COOPERLAKE) ifeq ($(TARGET_CORE), COOPERLAKE)
override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)

2
kernel/setparam-ref.c
View File

@ -1164,7 +1164,7 @@ static void init_parameter(void) {
TABLE_NAME.xgemm3m_q = QGEMM_DEFAULT_Q; TABLE_NAME.xgemm3m_q = QGEMM_DEFAULT_Q;
#endif #endif
#if (CORE_KATMAI) || (CORE_COPPERMINE) || (CORE_BANIAS) || (CORE_YONAH) || (CORE_ATHLON) #if defined(CORE_KATMAI) || defined(CORE_COPPERMINE) || defined(CORE_BANIAS) || defined(CORE_YONAH) || defined(CORE_ATHLON)
#ifdef DEBUG #ifdef DEBUG
fprintf(stderr, "Katmai, Coppermine, Banias, Athlon\n"); fprintf(stderr, "Katmai, Coppermine, Banias, Athlon\n");

2
kernel/x86_64/iamax_sse.S
View File

@ -51,6 +51,8 @@
#define MAXPS maxps #define MAXPS maxps
#define MAXSS maxss #define MAXSS maxss
#ifdef USE_MIN #ifdef USE_MIN
#undef MAXPS
#undef MAXSS
#define MAXPS minps #define MAXPS minps
#define MAXSS minss #define MAXSS minss
#endif #endif