Browse Source

enabled gemv assembly on power8

tags/v0.2.16^2
Werner Saar 9 years ago
parent
commit
d5130ce7e3
5 changed files with 32 additions and 9 deletions
  1. +10
    -9
      kernel/power/KERNEL.POWER8
  2. +6
    -0
      kernel/power/gemv_n.S
  3. +5
    -0
      kernel/power/gemv_t.S
  4. +5
    -0
      kernel/power/zgemv_n.S
  5. +6
    -0
      kernel/power/zgemv_t.S

+ 10
- 9
kernel/power/KERNEL.POWER8 View File

@@ -137,15 +137,16 @@ ZGEMM3MKERNEL = zgemm3m_kernel_4x4_sse3.S
#CSWAPKERNEL = ../arm/zswap.c
#ZSWAPKERNEL = ../arm/zswap.c
#
SGEMVNKERNEL = ../arm/gemv_n.c
DGEMVNKERNEL = ../arm/gemv_n.c
CGEMVNKERNEL = ../arm/zgemv_n.c
ZGEMVNKERNEL = ../arm/zgemv_n.c

SGEMVTKERNEL = ../arm/gemv_t.c
DGEMVTKERNEL = ../arm/gemv_t.c
CGEMVTKERNEL = ../arm/zgemv_t.c
ZGEMVTKERNEL = ../arm/zgemv_t.c

#SGEMVNKERNEL = ../arm/gemv_n.c
#DGEMVNKERNEL = ../arm/gemv_n.c
#CGEMVNKERNEL = ../arm/zgemv_n.c
#ZGEMVNKERNEL = ../arm/zgemv_n.c
#
#SGEMVTKERNEL = ../arm/gemv_t.c
#DGEMVTKERNEL = ../arm/gemv_t.c
#CGEMVTKERNEL = ../arm/zgemv_t.c
#ZGEMVTKERNEL = ../arm/zgemv_t.c

SSYMV_U_KERNEL = ../generic/symv_k.c
SSYMV_L_KERNEL = ../generic/symv_k.c


+ 6
- 0
kernel/power/gemv_n.S View File

@@ -174,6 +174,12 @@
#define PREFETCHSIZE_C 40
#endif

#ifdef POWER8
#define PREFETCHSIZE_A 96
#define PREFETCHSIZE_C 40
#endif


#ifndef NEEDPARAM

#ifndef __64BIT__


+ 5
- 0
kernel/power/gemv_t.S View File

@@ -139,6 +139,11 @@
#define PREFETCHSIZE_C 8
#endif

#ifdef POWER8
#define PREFETCHSIZE_A 96
#define PREFETCHSIZE_C 8
#endif

#define y01 f0
#define y02 f1
#define y03 f2


+ 5
- 0
kernel/power/zgemv_n.S View File

@@ -170,6 +170,11 @@
#define PREFETCHSIZE_C 24
#endif

#ifdef POWER8
#define PREFETCHSIZE_A 24
#define PREFETCHSIZE_C 24
#endif

#ifndef XCONJ
#define FMADDR FMADD
#define FMSUBR FNMSUB


+ 6
- 0
kernel/power/zgemv_t.S View File

@@ -144,6 +144,12 @@
#define PREFETCHSIZE_C 8
#endif

#ifdef POWER8
#define PREFETCHSIZE_A 24
#define PREFETCHSIZE_C 8
#endif


#if !(defined(CONJ) && defined(XCONJ))
#define FMADDR FMADD
#define FMSUBR FNMSUB


Loading…
Cancel
Save