LoongArch64: Fixed LSX version of cscal and zscal

OpenMathLib · Jan 16, 2025 · b5ed48b · b5ed48b
1 parent 9b98103
commit b5ed48b
Showing 1 changed file with 12 additions and 4 deletions.
diff --git a/kernel/loongarch64/cscal_lsx.S b/kernel/loongarch64/cscal_lsx.S
@@ -85,9 +85,9 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     bne INCX, TEMP, .L22
 
 .L11:
-    bge $r0, I, .L997
     CMPEQ  $fcc0, ALPHAR, a1
     CMPEQ  $fcc1, ALPHAI, a1
+    bge $r0, I, .L997
     bceqz $fcc0, .L13
     b .L14
     .align 3
@@ -201,10 +201,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     .align 3
 
 .L22:
-    bge $r0, I, .L997
-    move XX, X
     CMPEQ  $fcc0, ALPHAR, a1
     CMPEQ  $fcc1, ALPHAI, a1
+    bge $r0, I, .L997
+    move XX, X
     bceqz $fcc0, .L23
     b .L24
     .align 3
@@ -422,7 +422,15 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     andi I, N, 3
     bge $r0, I, .L999
     .align 3
-
+    bceqz $fcc0, .L998
+    bceqz $fcc1, .L998
+.L996:
+    addi.d I, I, -1
+    ST   a1, X, 0 * SIZE
+    ST   a1, X, 1 * SIZE
+    add.d X, X, INCX
+    blt $r0, I, .L996
+    b .L999
 .L998:
     LD   a1, X, 0 * SIZE
     LD   a2, X, 1 * SIZE