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

loongarch: Fixed {sc/dz}amax LSX opt

This commit is contained in:
gxw 2024-03-19 09:56:11 +08:00
parent ad13e04669
commit b5eb9d6bac
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
kernel/loongarch64/camax_lsx.S
View File

@ -177,7 +177,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
FABS t4, t4 FABS t4, t4
ADD t1, t1, t2 ADD t1, t1, t2
ADD t3, t3, t4 ADD t3, t3, t4
FMAX s1, t1, t3 FMAX s2, t1, t3
LD t1, X, 0 * SIZE LD t1, X, 0 * SIZE
LD t2, X, 1 * SIZE LD t2, X, 1 * SIZE
add.d X, X, INCX add.d X, X, INCX
@ -205,13 +205,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
ADD t1, t1, t2 ADD t1, t1, t2
ADD t3, t3, t4 ADD t3, t3, t4
FMAX s4, t1, t3 FMAX s4, t1, t3
FMAX s1, s1, s2
FMAX s3, s3, s4
FMAX a0, a0, s3
FMAX a0, a0, s1
blt $r0, I, .L21 blt $r0, I, .L21
.align 3 .align 3
.L22: .L22:
FMAX s1, s1, s2 MOV s1, a0
FMAX s3, s3, s4
FMAX s1, s1, s3
.align 3 .align 3
.L23: //N<8 .L23: //N<8