Browse Source

bugfix in sgemv_t_microk_sandy.c

tags/v0.2.11^2
wernsaar 11 years ago
parent
commit
02eb72ac42
2 changed files with 10 additions and 1 deletions
  1. +8
    -0
      kernel/x86_64/KERNEL.SANDYBRIDGE
  2. +2
    -1
      kernel/x86_64/sgemv_t_microk_sandy.c

+ 8
- 0
kernel/x86_64/KERNEL.SANDYBRIDGE View File

@@ -1,3 +1,11 @@
ifdef OS_WINDOWS
#SGEMVNKERNEL = ../arm/gemv_n.c
SGEMVTKERNEL = ../arm/gemv_t.c
else
#SGEMVNKERNEL = sgemv_n_avx.c
SGEMVTKERNEL = sgemv_t_avx.c
endif


SGEMMKERNEL = sgemm_kernel_16x4_sandy.S
SGEMMINCOPY = ../generic/gemm_ncopy_16.c


+ 2
- 1
kernel/x86_64/sgemv_t_microk_sandy.c View File

@@ -80,7 +80,8 @@ static void sgemv_kernel_16( long n, float alpha, float *a, long lda, float *x,
"vhaddps %%xmm12, %%xmm12, %%xmm12\n\t"
"vhaddps %%xmm12, %%xmm12, %%xmm12\n\t"

"vfmaddss (%%rdx), %%xmm12, %%xmm1, %%xmm12\n\t"
"vmulss %%xmm12, %%xmm1, %%xmm12 \n\t"
"vaddss (%%rdx), %%xmm12, %%xmm12\n\t"
"vmovss %%xmm12, (%%rdx) \n\t" // store temp -> y

:


Loading…
Cancel
Save