Browse Source

Fixed the Windows x86_64 ABI bug in s/daxpy kernels.

tags/v0.2.7
Zhang Xianyi 13 years ago
parent
commit
724ae159ce
2 changed files with 9 additions and 2 deletions
  1. +5
    -1
      kernel/x86_64/axpy_sse.S
  2. +4
    -1
      kernel/x86_64/axpy_sse2.S

+ 5
- 1
kernel/x86_64/axpy_sse.S View File

@@ -69,7 +69,7 @@
#endif
movaps %xmm0, ALPHA
#else
movaps %xmm3, ALPHA

movq 40(%rsp), X
movq 48(%rsp), INCX
@@ -79,6 +79,10 @@

SAVEREGISTERS

#ifdef WINDOWS_ABI
movaps %xmm3, ALPHA
#endif
shufps $0, ALPHA, ALPHA

leaq (, INCX, SIZE), INCX


+ 4
- 1
kernel/x86_64/axpy_sse2.S View File

@@ -69,7 +69,6 @@
#endif
movaps %xmm0, ALPHA
#else
movaps %xmm3, ALPHA

movq 40(%rsp), X
movq 48(%rsp), INCX
@@ -79,6 +78,10 @@

SAVEREGISTERS

#ifdef WINDOWS_ABI
movaps %xmm3, ALPHA
#endif

unpcklpd ALPHA, ALPHA

leaq (, INCX, SIZE), INCX


Loading…
Cancel
Save