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

added new sgemm kernel for BULLDOZER

This commit is contained in:
wernsaar 2013-06-09 15:57:42 +02:00
parent e4c39c7c26
commit d65bbec99b
3 changed files with 4549 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
kernel/x86_64/KERNEL.BULLDOZER
View File

@ -1,11 +1,11 @@
ZGEMVNKERNEL = zgemv_n_dup.S
ZGEMVTKERNEL = zgemv_t_dup.S
SGEMMKERNEL = gemm_kernel_8x4_barcelona.S
SGEMMINCOPY = ../generic/gemm_ncopy_8.c
SGEMMITCOPY = ../generic/gemm_tcopy_8.c
SGEMMONCOPY = gemm_ncopy_4_opteron.S
SGEMMOTCOPY = gemm_tcopy_4_opteron.S
SGEMMKERNEL = sgemm_kernel_16x2_bulldozer.S
SGEMMINCOPY = ../generic/gemm_ncopy_16.c
SGEMMITCOPY = ../generic/gemm_tcopy_16.c
SGEMMONCOPY = ../generic/gemm_ncopy_2.c
SGEMMOTCOPY = ../generic/gemm_tcopy_2.c
SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)

4532
kernel/x86_64/sgemm_kernel_16x2_bulldozer.S Normal file
View File

File diff suppressed because it is too large Load Diff

16
param.h
View File

@ -152,10 +152,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define GEMM_DEFAULT_OFFSET_B 832
#define GEMM_DEFAULT_ALIGN 0x0fffUL
#define SGEMM_DEFAULT_UNROLL_N 4
#if defined(BULLDOZER) && defined(ARCH_X86_64) && !defined(COMPLEX)
#define SGEMM_DEFAULT_UNROLL_N 2
#define DGEMM_DEFAULT_UNROLL_N 2
#else
#define SGEMM_DEFAULT_UNROLL_N 4
#define DGEMM_DEFAULT_UNROLL_N 4
#endif
#define QGEMM_DEFAULT_UNROLL_N 2
@ -171,16 +172,21 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define ZGEMM_DEFAULT_UNROLL_M 1
#define XGEMM_DEFAULT_UNROLL_M 1
#else
#define SGEMM_DEFAULT_UNROLL_M 8
#if defined(BULLDOZER) && !defined(COMPLEX)
#define SGEMM_DEFAULT_UNROLL_M 16
#define DGEMM_DEFAULT_UNROLL_M 8
#else
#define SGEMM_DEFAULT_UNROLL_M 8
#define DGEMM_DEFAULT_UNROLL_M 4
#endif
#define QGEMM_DEFAULT_UNROLL_M 2
#define CGEMM_DEFAULT_UNROLL_M 4
#define ZGEMM_DEFAULT_UNROLL_M 2
#define XGEMM_DEFAULT_UNROLL_M 1
#define CGEMM3M_DEFAULT_UNROLL_N 4
#define CGEMM3M_DEFAULT_UNROLL_M 8
#define ZGEMM3M_DEFAULT_UNROLL_N 4
#define ZGEMM3M_DEFAULT_UNROLL_M 4
#endif
#if 0
@ -200,10 +206,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#else
#define SGEMM_DEFAULT_P 448
#if defined(BULLDOZER) && defined(ARCH_X86_64) && !defined(COMPLEX)
#define SGEMM_DEFAULT_P 768
#define DGEMM_DEFAULT_P 384
#else
#define SGEMM_DEFAULT_P 448
#define DGEMM_DEFAULT_P 224
#endif
#define QGEMM_DEFAULT_P 112
@ -211,10 +218,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define ZGEMM_DEFAULT_P 112
#define XGEMM_DEFAULT_P 56
#define SGEMM_DEFAULT_Q 224
#if defined(BULLDOZER) && defined(ARCH_X86_64) && !defined(COMPLEX)
#define SGEMM_DEFAULT_Q 168
#define DGEMM_DEFAULT_Q 168
#else
#define SGEMM_DEFAULT_Q 224
#define DGEMM_DEFAULT_Q 224
#endif
#define QGEMM_DEFAULT_Q 224