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

Ref #65. Fixed 64-bit Windows calling convention bug in cdot and zdot. 

According to 64-bit Windows calling convention, the return value is in %rax instead of %xmm0 in cdot kernel.
In zdot, the caller allocates a memory space for return value and sets this memory address to the first hidden parameter. Thus, the callee (zdot) should assign the result to this memory space and return the memory address in %rax.
This commit is contained in:
traits 2011-10-18 10:23:17 +08:00
parent ba31b19c00
commit c852ce3981
2 changed files with 19 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
kernel/x86_64/zdot_sse.S
View File

@ -3483,6 +3483,10 @@
subss %xmm3, %xmm1 subss %xmm3, %xmm1
#endif #endif
unpcklps %xmm1, %xmm0 unpcklps %xmm1, %xmm0
#ifdef WINDOWS_ABI
movq %xmm0, %rax
#endif
RESTOREREGISTERS RESTOREREGISTERS

18
kernel/x86_64/zdot_sse2.S
View File

@ -39,14 +39,19 @@
#define ASSEMBLER #define ASSEMBLER
#include "common.h" #include "common.h"
#ifndef WINDOWS_ABI
#define N ARG1 /* rdi */ #define N ARG1 /* rdi */
#define X ARG2 /* rsi */ #define X ARG2 /* rsi */
#define INCX ARG3 /* rdx */ #define INCX ARG3 /* rdx */
#define Y ARG4 /* rcx */ #define Y ARG4 /* rcx */
#ifndef WINDOWS_ABI
#define INCY ARG5 /* r8 */ #define INCY ARG5 /* r8 */
#else #else
#define INCY %r10 #define RESULT_ADDRESS ARG1 /*rcx*/
#define N ARG2 /* rdx */
#define X ARG3 /* r8 */
#define INCX ARG4 /* r9*/
#define Y %r10
#define INCY %r11
#endif #endif
#include "l1param.h" #include "l1param.h"
@ -64,7 +69,8 @@
PROFCODE PROFCODE
#ifdef WINDOWS_ABI #ifdef WINDOWS_ABI
movq 40(%rsp), INCY movq 40(%rsp), Y
movq 48(%rsp), INCY
#endif #endif
SAVEREGISTERS SAVEREGISTERS
@ -1544,6 +1550,12 @@
subsd %xmm3, %xmm1 subsd %xmm3, %xmm1
#endif #endif
#ifdef WINDOWS_ABI
movq RESULT_ADDRESS, %rax
movsd %xmm0, (%rax)
movsd %xmm1, 8(%rax)
#endif
RESTOREREGISTERS RESTOREREGISTERS
ret ret