LoongArch64: Fixed LASX version of cscal and zscal

OpenMathLib · Jan 16, 2025 · cb8cc3f · cb8cc3f
1 parent b5ed48b
commit cb8cc3f
Showing 1 changed file with 13 additions and 3 deletions.
diff --git a/kernel/loongarch64/cscal_lasx.S b/kernel/loongarch64/cscal_lasx.S
@@ -87,9 +87,9 @@ 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
@@ -182,10 +182,10 @@ 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
@@ -477,6 +477,16 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
     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