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

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

This commit is contained in:
Zhang Xianyi 2013-03-08 22:28:34 +08:00
parent 2c9a203bd1
commit 724ae159ce
2 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
kernel/x86_64/axpy_sse.S
View File

@ -69,7 +69,7 @@
#endif #endif
movaps %xmm0, ALPHA movaps %xmm0, ALPHA
#else #else
movaps %xmm3, ALPHA
movq 40(%rsp), X movq 40(%rsp), X
movq 48(%rsp), INCX movq 48(%rsp), INCX
@ -79,6 +79,10 @@
SAVEREGISTERS SAVEREGISTERS
#ifdef WINDOWS_ABI
movaps %xmm3, ALPHA
#endif
shufps $0, ALPHA, ALPHA shufps $0, ALPHA, ALPHA
leaq (, INCX, SIZE), INCX leaq (, INCX, SIZE), INCX

5
kernel/x86_64/axpy_sse2.S
View File

@ -69,7 +69,6 @@
#endif #endif
movaps %xmm0, ALPHA movaps %xmm0, ALPHA
#else #else
movaps %xmm3, ALPHA
movq 40(%rsp), X movq 40(%rsp), X
movq 48(%rsp), INCX movq 48(%rsp), INCX
@ -79,6 +78,10 @@
SAVEREGISTERS SAVEREGISTERS
#ifdef WINDOWS_ABI
movaps %xmm3, ALPHA
#endif
unpcklpd ALPHA, ALPHA unpcklpd ALPHA, ALPHA
leaq (, INCX, SIZE), INCX leaq (, INCX, SIZE), INCX