diff --git a/kernel/loongarch64/dgemv_t_lsx.S b/kernel/loongarch64/dgemv_t_lsx.S
index 76f0d9bdcc..ee16c798b9 100644
--- a/kernel/loongarch64/dgemv_t_lsx.S
+++ b/kernel/loongarch64/dgemv_t_lsx.S
@@ -26,10 +26,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *******************************************************************************/
 
 #define ASSEMBLER
-
 #include "common.h"
 
-/* Param */
 #define M      $r4
 #define N      $r5
 #define A      $r7
@@ -41,44 +39,26 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define BUFFER $r16
 #define ALPHA  $f0
 
-#define YORIG  $r18
-#define T0     $r19
-#define T1     $r20
-#define AO3    $r12
-#define AO4    $r13
 #define I      $r14
 #define J      $r15
-#define AO1    $r23
-#define AO2    $r24
+#define AO     $r23
+#define XORG   $r24
 #define IX     $r25
 #define IY     $r26
 #define II     $r27
+#define T0     $r19
+#define T1     $r20
 #define T2     $r28
 #define T3     $r29
 #define T4     $r30
 
-/* LSX vectors */
-#define U0     $vr11
-#define U1     $vr12
-#define U2     $vr2
-#define U3     $vr3
-#define U4     $vr4
-#define U5     $vr5
-#define U6     $vr6
-#define U7     $vr7
-#define U8     $vr8
-#define U9     $vr9
-#define VALPHA $vr10
-
-#define a1     $f3
-#define a2     $f4
-#define a3     $f5
-#define a4     $f6
-#define a5     $f7
-#define a6     $f8
-#define a7     $f9
-#define a8     $f10
-
+#define ACC    $vr11
+#define VX     $vr12
+#define VA     $vr13
+#define VT     $vr14
+#define FAC    $f17
+#define FAX    $f18
+#define FAY    $f19
 
     PROLOGUE
 
@@ -86,7 +66,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     LDARG  BUFFER,  $sp,  8
 
     addi.d  $sp, $sp, -80
-
     SDARG  $r23,  $sp,    0
     SDARG  $r24,  $sp,    8
     SDARG  $r25,  $sp,    16
@@ -97,8 +76,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     SDARG  $r30,  $sp,    64
     ST     ALPHA, $sp,    72
 
-    vldrepl.d  VALPHA, $sp, 72
-
     slli.d     LDA,     LDA,  BASE_SHIFT
     slli.d     INCX,    INCX, BASE_SHIFT
     slli.d     INCY,    INCY, BASE_SHIFT
@@ -108,160 +85,82 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
     move   J,      $r0
     move   IY,     $r0
-
-    move   AO1,    A  //a_ptr1
-
-    srai.d     T0,     N,     2  //n/4
-    beq        J,      T0,    .L04
-
-.L01: /* j<n/4 */
-    vxor.v    U0,     U0,   U0
-    vxor.v    U7,     U7,   U7
-    add.d    AO2,    AO1,  LDA
-    add.d    AO3,    AO2,  LDA
-    add.d    AO4,    AO3,  LDA
-
-    move      IX,     $r0
-    move      I,      $r0
+    move   XORG,   X
+    move   AO,     A
+    li.d   T1,     8
+
+.L01:  /* for each column J */
+    vxor.v    ACC,    ACC,   ACC
+    MTC       FAC,    $r0
+    move      X,      XORG
     move      II,     $r0
-
-    beq       $r0,    M,    .L03
-
-.L02:  /* i<m */
-    vldx        U1,    X,    IX
-    fldx.d      $f2,   AO1,   II
-    fldx.d      $f3,   AO2,   II
-    fldx.d      $f4,   AO3,   II
-    fldx.d      $f5,   AO4,   II
-
-    vshuf4i.d   U1,    U1,    0x00
-    vextrins.d  U2,    U3,    0x10
-    vextrins.d  U4,    U5,    0x10
-
-    vfmadd.d    U0,    U2,    U1,    U0  //temp1,2
-    vfmadd.d    U7,    U4,    U1,    U7  //temp3,4
-
-    add.d       IX,    IX,    INCX
-
-    addi.d     II,     II,    8
-    addi.d     I,      I,     1
-    blt        I,      M,    .L02
+    move      I,      $r0
+    srli.d    T2,     M,     1
+    slli.d    T2,     T2,    1       /* T2 = M & ~1 */
+
+    bne       INCX,   T1,    .L_SCALAR
+
+.L02:
+    beq       I,      T2,    .L03
+    vld       VX,     X,     0
+    vld       VA,     AO,    0
+    vfmadd.d  ACC,    VA,    VX,   ACC
+    addi.d    II,     II,    16
+    addi.d    I,      I,     2
+    addi.d    X,      X,     16
+    addi.d    AO,     AO,    16
+    b         .L02
 
 .L03:
-    move        T1,    IY
-    add.d       T2,    T1,    INCY
-    add.d       T3,    T2,    INCY
-    add.d       T4,    T3,    INCY
-
-    fldx.d      $f3,    Y,    T1
-    fldx.d      $f4,    Y,    T2
-    fldx.d      $f5,    Y,    T3
-    fldx.d      $f6,    Y,    T4
-
-    vextrins.d  U3,    U4,    0x10
-    vextrins.d  U5,    U6,    0x10
-
-    vfmadd.d    U3,    VALPHA,    U0,    U3
-    vfmadd.d    U5,    VALPHA,    U7,    U5
-
-    vextrins.d  U4,    U3,    0x01
-    vextrins.d  U6,    U5,    0x01
-
-    fstx.d     $f3,     Y,    T1
-    fstx.d     $f4,     Y,    T2
-    fstx.d     $f5,     Y,    T3
-    fstx.d     $f6,     Y,    T4
-
-    slli.d     T1,     LDA,     2
-    add.d      AO1,    AO1,     T1
-    add.d      IY,     T4,      INCY
-
-    addi.d     J,      J,     1
-    blt        J,      T0,    .L01
-
-.L04: /* if(n&2) */
-    andi       T0,     N,     2
-    beq        $r0,    T0,    .L07
-
-    vxor.v    U0,     U0,   U0
-
-    add.d    AO2,    AO1,  LDA
-
+    andi      T3,     M,     1
+    beq       $r0,    T3,    .L05
+    fld.d     FAY,    AO,    0
+    fld.d     FAX,    X,     0
+    fmadd.d   FAC,    FAY,   FAX,  FAC
+.L05:
+    /* dot = (lane0+lane1) + tail */
+    vextrins.d VT,    ACC,   0x01
+    vfadd.d   VT,     ACC,    VT
+    vpickve2gr.d T0,  VT,    0
+    movgr2fr.d  FAX,  T0
+    fadd.d      FAC,  FAX,   FAC
+    /* y += alpha*dot */
+    fldx.d    FAY,    Y,     IY
+    fmadd.d   FAY,    ALPHA, FAC,  FAY
+    fstx.d    FAY,    Y,     IY
+    /* next column: AO = A + (J+1)*LDA */
+    addi.d    T3,     J,     1
+    mul.d     T3,     T3,    LDA
+    add.d     AO,     A,     T3
+    add.d     IY,     IY,    INCY
+    addi.d    J,      J,     1
+    blt       J,      N,     .L01
+    b         .L999
+
+.L_SCALAR:
     move      IX,     $r0
     move      I,      $r0
     move      II,     $r0
-
-    beq       $r0,    M,    .L06
-
-.L05:  /* i<m */
-    vldx        U1,    X,    IX
-    fldx.d      $f2,   AO1,   II
-    fldx.d      $f3,   AO2,   II
-
-    vshuf4i.d   U1,    U1,    0x00
-    vextrins.d  U2,    U3,    0x10
-
-    vfmadd.d    U0,    U2,    U1,    U0  //temp1,2
-
-    add.d       IX,    IX,    INCX
-
-    addi.d     II,     II,    8
-    addi.d     I,      I,     1
-    blt        I,      M,    .L05
-
 .L06:
-    move        T1,    IY
-    add.d       T2,    T1,    INCY
-
-    fldx.d      a1,    Y,    T1
-    fldx.d      a2,    Y,    T2
-
-    vextrins.d  U3,    U4,    0x10
-
-    vfmadd.d    U3,    VALPHA,    U0,    U3
-
-    vextrins.d  U4,    U3,    0x01
-
-    fstx.d     a1,     Y,    T1
-    fstx.d     a2,     Y,    T2
-
-    slli.d     T0,     LDA,     1
-    add.d      AO1,    AO1,     T0
-    add.d      IY,     T2,      INCY
-
-.L07: /* if(n&1) */
-    andi       T0,     N,     1
-    beq        $r0,    T0,    .L999
-
-    MTC       a1,     $r0
-
-    move      IX,     $r0
-    move      I,      $r0
-    move      II,      $r0
-
-    beq       $r0,    M,    .L09
-
-.L08:  /* i<m */
-    fldx.d      a3,    X,     IX
-    fldx.d      a4,    AO1,   II
-
-    fmadd.d     a1,    a4,    a3,    a1  //temp1
-
-    add.d       IX,    IX,    INCX
-
-    addi.d     II,     II,    8
-    addi.d     I,      I,     1
-    blt        I,      M,    .L08
-
-.L09:
-    fldx.d      a3,    Y,    IY
-
-    fmadd.d    a3,    ALPHA,    a1,    a3
-
-    fstx.d     a3,     Y,    IY
-
-    add.d      AO1,    AO1,     LDA
-    add.d      IY,     IY,      INCY
+    beq       I,      M,     .L07
+    fldx.d    FAX,    X,     IX
+    fldx.d    FAY,    AO,    II
+    fmadd.d   FAC,    FAY,   FAX,  FAC
+    add.d     IX,     IX,    INCX
+    addi.d    II,     II,    8
+    addi.d    I,      I,     1
+    b         .L06
+.L07:
+    fldx.d    FAY,    Y,     IY
+    fmadd.d   FAY,    ALPHA, FAC,  FAY
+    fstx.d    FAY,    Y,     IY
+    addi.d    T3,     J,     1
+    mul.d     T3,     T3,    LDA
+    add.d     AO,     A,     T3
+    add.d     IY,     IY,    INCY
+    addi.d    J,      J,     1
+    blt       J,      N,     .L01
+    b         .L999
 
 .L999:
     LDARG  $r23,  $sp,    0