Browse Source

Changed kernels for dgemm and dtrmm

tags/v0.2.9.rc1
wernsaar 12 years ago
parent
commit
e0b968c3a7
4 changed files with 3554 additions and 12 deletions
  1. +8
    -8
      kernel/arm/KERNEL.ARMV7
  2. +1589
    -0
      kernel/arm/dgemm_kernel_4x4_vfpv3.S
  3. +1953
    -0
      kernel/arm/dtrmm_kernel_4x4_vfpv3.S
  4. +4
    -4
      param.h

+ 8
- 8
kernel/arm/KERNEL.ARMV7 View File

@@ -81,7 +81,7 @@ CGEMVTKERNEL = zgemv_t.c
ZGEMVTKERNEL = zgemv_t.c

STRMMKERNEL = ../generic/trmmkernel_2x2.c
DTRMMKERNEL = dtrmm_kernel_8x2_vfpv3.S
DTRMMKERNEL = dtrmm_kernel_4x4_vfpv3.S
CTRMMKERNEL = ../generic/ztrmmkernel_2x2.c
ZTRMMKERNEL = ../generic/ztrmmkernel_2x2.c

@@ -93,13 +93,13 @@ SGEMMOTCOPYOBJ = sgemm_otcopy.o

#DGEMMKERNEL = ../generic/gemmkernel_2x2.c
#DGEMMKERNEL = dgemm_kernel_4x2_vfpv2.S
DGEMMKERNEL = dgemm_kernel_8x2_vfpv3.S
DGEMMINCOPY = ../generic/gemm_ncopy_8.c
DGEMMITCOPY = ../generic/gemm_tcopy_8.c
DGEMMONCOPY = ../generic/gemm_ncopy_2.c
DGEMMOTCOPY = ../generic/gemm_tcopy_2.c
DGEMMINCOPYOBJ = dgemm_incopy.o
DGEMMITCOPYOBJ = dgemm_itcopy.o
DGEMMKERNEL = dgemm_kernel_4x4_vfpv3.S
DGEMMINCOPY =
DGEMMITCOPY =
DGEMMONCOPY = ../generic/gemm_ncopy_4.c
DGEMMOTCOPY = ../generic/gemm_tcopy_4.c
DGEMMINCOPYOBJ =
DGEMMITCOPYOBJ =
DGEMMONCOPYOBJ = dgemm_oncopy.o
DGEMMOTCOPYOBJ = dgemm_otcopy.o



+ 1589
- 0
kernel/arm/dgemm_kernel_4x4_vfpv3.S
File diff suppressed because it is too large
View File


+ 1953
- 0
kernel/arm/dtrmm_kernel_4x4_vfpv3.S
File diff suppressed because it is too large
View File


+ 4
- 4
param.h View File

@@ -1805,8 +1805,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define SGEMM_DEFAULT_UNROLL_M 2
#define SGEMM_DEFAULT_UNROLL_N 2

#define DGEMM_DEFAULT_UNROLL_M 8
#define DGEMM_DEFAULT_UNROLL_N 2
#define DGEMM_DEFAULT_UNROLL_M 4
#define DGEMM_DEFAULT_UNROLL_N 4

#define CGEMM_DEFAULT_UNROLL_M 2
#define CGEMM_DEFAULT_UNROLL_N 2
@@ -1815,12 +1815,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define ZGEMM_DEFAULT_UNROLL_N 2

#define SGEMM_DEFAULT_P 64
#define DGEMM_DEFAULT_P 64
#define DGEMM_DEFAULT_P 128
#define CGEMM_DEFAULT_P 24
#define ZGEMM_DEFAULT_P 20

#define SGEMM_DEFAULT_Q 192
#define DGEMM_DEFAULT_Q 64
#define DGEMM_DEFAULT_Q 96
#define CGEMM_DEFAULT_Q 128
#define ZGEMM_DEFAULT_Q 64



Loading…
Cancel
Save