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

bugfixes for sgemm- and cgemm-kernel

This commit is contained in:
Werner Saar 2016-04-06 11:15:21 +02:00
parent 7282419525
commit a51102e9b7
3 changed files with 11 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
kernel/power/cgemm_kernel_8x4_power8.S
View File

@ -82,7 +82,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#ifdef __64BIT__
#define STACKSIZE 400
#define STACKSIZE 512
#define ALPHA_R_SP 296(SP)
#define ALPHA_I_SP 304(SP)
#define FZERO 312(SP)
@ -290,9 +290,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
li o32 , 32
li o48 , 48
li T1, 256
slwi T1, T1, 9 // 131072
sub BBUFFER, A, T1 // temp buffer for B unrolled
li T1, 512
slwi T1, T1, 16
add BBUFFER, A, T1
#ifdef __64BIT__

8
kernel/power/sgemm_kernel_16x8_power8.S
View File

@ -82,7 +82,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#ifdef __64BIT__
#define STACKSIZE 340
#define STACKSIZE 512
#define ALPHA_SP 296(SP)
#define FZERO 304(SP)
#else
@ -271,9 +271,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
li o32, 32
li o48, 48
li T1, 256
slwi T1, T1, 9 // 131072
sub BBUFFER, A, T1 // temp buffer for B unrolled
li T1, 512
slwi T1, T1, 16
add BBUFFER, A, T1
addi T1, SP, 300
stxsspx f1, o0 , T1

6
param.h
View File

@ -1965,7 +1965,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define DNUMOPT 8
#define GEMM_DEFAULT_OFFSET_A 131072
#define GEMM_DEFAULT_OFFSET_B 1024
#define GEMM_DEFAULT_OFFSET_B 131072
#define GEMM_DEFAULT_ALIGN 0x03fffUL
#define SGEMM_DEFAULT_UNROLL_M 16
@ -1985,12 +1985,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define SGEMM_DEFAULT_Q 720
#define DGEMM_DEFAULT_Q 720
#define CGEMM_DEFAULT_Q 720
#define ZGEMM_DEFAULT_Q 360
#define ZGEMM_DEFAULT_Q 720
#define SGEMM_DEFAULT_R 14400
#define DGEMM_DEFAULT_R 14400
#define CGEMM_DEFAULT_R 14400
#define ZGEMM_DEFAULT_R 7200
#define ZGEMM_DEFAULT_R 14400
#define SYMV_P 8