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

Enable thread affinity on Loongson 3B. Fixed the bug of reading cycle counter. 

In Loongson 3A and 3B, the CPU core increases the counter in every 2 cycles by default.
This commit is contained in:
Xianyi Zhang 2011-11-11 17:49:41 +00:00
parent 0884f6b78d
commit d1baf14a64
2 changed files with 23 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Makefile.system
View File

@ -591,9 +591,11 @@ endif
ifneq ($(ARCH), x86_64)
ifneq ($(ARCH), x86)
ifneq ($(CORE), LOONGSON3B)
NO_AFFINITY = 1
endif
endif
endif
ifdef NO_AFFINITY
CCOMMON_OPT += -DNO_AFFINITY

28
common_mips64.h
View File

@ -101,13 +101,15 @@ static void INLINE blas_lock(volatile unsigned long *address){
static inline unsigned int rpcc(void){
unsigned long ret;
#if defined(LOONGSON3A)
unsigned long long tmp;
__asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
ret=tmp;
#elif defined(LOONGSON3B)
//Temp Implementation.
return 1;
#if defined(LOONGSON3A) || defined(LOONGSON3B)
// unsigned long long tmp;
//__asm__ __volatile__("dmfc0 %0, $25, 1": "=r"(tmp):: "memory");
//ret=tmp;
__asm__ __volatile__(".set push \n"
".set mips32r2\n"
"rdhwr %0, $2\n"
".set pop": "=r"(ret):: "memory");
#else
__asm__ __volatile__(".set push \n"
".set mips32r2\n"
@ -117,6 +119,18 @@ static inline unsigned int rpcc(void){
return ret;
}
//#if defined(LOONGSON3A) || defined(LOONGSON3B)
static inline int WhereAmI(void){
int ret=0;
__asm__ __volatile__(".set push \n"
".set mips32r2\n"
"rdhwr %0, $0\n"
".set pop": "=r"(ret):: "memory");
return ret;
}
//#endif
static inline int blas_quickdivide(blasint x, blasint y){
return x / y;
}