Skip to content

Commit ad13e04

Browse files
committed
loongarch: Fixed {s/d/sc/dz}amin LSX opt
1 parent bbf82cb commit ad13e04

File tree

2 files changed

+13
-10
lines changed

2 files changed

+13
-10
lines changed

kernel/loongarch64/amin_lsx.S

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -146,7 +146,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
146146
add.d X, X, INCX
147147
vinsgr2vr.d VX1, t3, 0
148148
vinsgr2vr.d VX1, t4, 1
149-
vfmaxa.d VM1, VX0, VX1
149+
vfmina.d VM1, VX0, VX1
150150
ld.d t1, X, 0 * SIZE
151151
add.d X, X, INCX
152152
ld.d t2, X, 0 * SIZE
@@ -159,9 +159,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
159159
add.d X, X, INCX
160160
vinsgr2vr.d VX1, t3, 0
161161
vinsgr2vr.d VX1, t4, 1
162-
vfmaxa.d VM2, VX0, VX1
163-
vfmaxa.d VM1, VM1, VM2
164-
vfmaxa.d VM0, VM0, VM1
162+
vfmina.d VM2, VX0, VX1
163+
vfmina.d VM1, VM1, VM2
164+
vfmina.d VM0, VM0, VM1
165165
#else
166166
ld.w t1, X, 0
167167
add.d X, X, INCX
@@ -187,8 +187,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
187187
vinsgr2vr.w VX1, t2, 1
188188
vinsgr2vr.w VX1, t3, 2
189189
vinsgr2vr.w VX1, t4, 3
190-
vfmaxa.s VM1, VX0, VX1
191-
vfmaxa.s VM0, VM0, VM1
190+
vfmina.s VM1, VX0, VX1
191+
vfmina.s VM0, VM0, VM1
192192
#endif
193193
addi.d I, I, -1
194194
blt $r0, I, .L21

kernel/loongarch64/camin_lsx.S

Lines changed: 7 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -186,7 +186,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
186186
FABS t4, t4
187187
ADD t1, t1, t2
188188
ADD t3, t3, t4
189-
FMIN s1, t1, t3
189+
FMIN s2, t1, t3
190190
LD t1, X, 0 * SIZE
191191
LD t2, X, 1 * SIZE
192192
add.d X, X, INCX
@@ -214,13 +214,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
214214
ADD t1, t1, t2
215215
ADD t3, t3, t4
216216
FMIN s4, t1, t3
217+
218+
FMIN s1, s1, s2
219+
FMIN s3, s3, s4
220+
FMIN a0, a0, s3
221+
FMIN a0, a0, s1
217222
blt $r0, I, .L21
218223
.align 3
219224

220225
.L22:
221-
FMIN s1, s1, s2
222-
FMIN s3, s3, s4
223-
FMIN s1, s1, s3
226+
MOV s1, a0
224227
.align 3
225228

226229
.L23: //N<8

0 commit comments

Comments
 (0)