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

fix NAN handling

Browse Source
This commit is contained in:
Martin Kroeker
2024-07-20 18:27:15 +02:00
committed by GitHub
parent c064319ecb
commit dfbc2348a8
2 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
kernel/x86_64/scal_atom.S
View File

@@ -60,8 +60,10 @@
#ifdef WINDOWS_ABI #ifdef WINDOWS_ABI
movq 40(%rsp), X movq 40(%rsp), X
movq 48(%rsp), INCX movq 48(%rsp), INCX
movq 64(%rsp), %r9
movaps %xmm3, %xmm0 movaps %xmm3, %xmm0
#else
movq 24(%rsp), %r9
#endif #endif
SAVEREGISTERS SAVEREGISTERS
@@ -73,6 +75,10 @@
lea (, INCX, SIZE), INCX lea (, INCX, SIZE), INCX
comisd %xmm0, %xmm1 comisd %xmm0, %xmm1
jne .L100 jne .L100
jp .L100
cmpq $1, %r9
je .L100
/* Alpha == ZERO */ /* Alpha == ZERO */
cmpq $SIZE, INCX cmpq $SIZE, INCX

2
kernel/x86_64/zscal_atom.S
View File

@@ -74,7 +74,7 @@
pxor %xmm15, %xmm15 pxor %xmm15, %xmm15
comisd %xmm0, %xmm15 comisd %xmm0, %xmm15
jne .L30 # Alpha_r != ZERO jne .L30 # Alpha_r != ZERO
jp .L30
comisd %xmm1, %xmm15 comisd %xmm1, %xmm15
jne .L30 # Alpha_i != ZERO jne .L30 # Alpha_i != ZERO