From 546f13558caad680fdd82778c106ebaa3bd7331c Mon Sep 17 00:00:00 2001
From: gxw <guxiwei-hf@loongson.cn>
Date: Fri, 29 Dec 2023 11:03:53 +0800
Subject: [PATCH] loongarch64: Add {c/z}swap and {c/z}sum optimization

---
 kernel/loongarch64/KERNEL.LOONGSON2K1000 |   6 +
 kernel/loongarch64/KERNEL.LOONGSON3R5    |   6 +
 kernel/loongarch64/csum_lasx.S           | 274 +++++++++++++++
 kernel/loongarch64/csum_lsx.S            | 266 ++++++++++++++
 kernel/loongarch64/cswap_lasx.S          | 394 +++++++++++++++++++++
 kernel/loongarch64/cswap_lsx.S           | 421 +++++++++++++++++++++++
 6 files changed, 1367 insertions(+)
 create mode 100644 kernel/loongarch64/csum_lasx.S
 create mode 100644 kernel/loongarch64/csum_lsx.S
 create mode 100644 kernel/loongarch64/cswap_lasx.S
 create mode 100644 kernel/loongarch64/cswap_lsx.S

diff --git a/kernel/loongarch64/KERNEL.LOONGSON2K1000 b/kernel/loongarch64/KERNEL.LOONGSON2K1000
index 0fb0bb68f..01f8e4782 100644
--- a/kernel/loongarch64/KERNEL.LOONGSON2K1000
+++ b/kernel/loongarch64/KERNEL.LOONGSON2K1000
@@ -75,6 +75,12 @@ DNRM2KERNEL =  dnrm2_lsx.S
 CNRM2KERNEL =  cnrm2_lsx.S
 ZNRM2KERNEL =  znrm2_lsx.S
 
+CSWAPKERNEL = cswap_lsx.S
+ZSWAPKERNEL = cswap_lsx.S
+
+CSUMKERNEL = csum_lsx.S
+ZSUMKERNEL = csum_lsx.S
+
 DGEMMKERNEL    = dgemm_kernel_8x4.S
 DGEMMINCOPY    = dgemm_ncopy_8_lsx.S
 DGEMMITCOPY    = dgemm_tcopy_8_lsx.S
diff --git a/kernel/loongarch64/KERNEL.LOONGSON3R5 b/kernel/loongarch64/KERNEL.LOONGSON3R5
index 1a6a04532..a9e8abaf0 100644
--- a/kernel/loongarch64/KERNEL.LOONGSON3R5
+++ b/kernel/loongarch64/KERNEL.LOONGSON3R5
@@ -75,6 +75,12 @@ DNRM2KERNEL =  dnrm2_lasx.S
 CNRM2KERNEL =  cnrm2_lasx.S
 ZNRM2KERNEL =  znrm2_lasx.S
 
+CSWAPKERNEL = cswap_lasx.S
+ZSWAPKERNEL = cswap_lasx.S
+
+CSUMKERNEL = csum_lasx.S
+ZSUMKERNEL = csum_lasx.S
+
 DGEMMKERNEL    = dgemm_kernel_16x4.S
 DGEMMINCOPY    = dgemm_ncopy_16.S
 DGEMMITCOPY    = dgemm_tcopy_16.S
diff --git a/kernel/loongarch64/csum_lasx.S b/kernel/loongarch64/csum_lasx.S
new file mode 100644
index 000000000..3e65f2c15
--- /dev/null
+++ b/kernel/loongarch64/csum_lasx.S
@@ -0,0 +1,274 @@
+/*******************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define t1     $r15
+#define t2     $r12
+#define t3     $r13
+#define t4     $r14
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define s1     $f16
+#define VX0    $xr12
+#define VX1    $xr13
+#define VX2    $xr14
+#define VX3    $xr15
+#define res1   $xr16
+#define res2   $xr17
+    PROLOGUE
+    xvxor.v res1, res1, res1
+    xvxor.v res2, res2, res2
+    bge $r0, N, .L999
+    bge $r0, INCX, .L999
+    li.d  TEMP, 1
+    slli.d  TEMP, TEMP, ZBASE_SHIFT
+    slli.d  INCX, INCX, ZBASE_SHIFT
+    srai.d I, N, 3
+    bne INCX, TEMP, .L20
+    bge $r0, I, .L13
+    .align 3
+
+.L11:
+#ifdef DOUBLE
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 4 * SIZE
+    xvfadd.d res2, VX0, VX1
+    xvfadd.d res1, res1, res2
+    xvld VX2, X, 8 * SIZE
+    xvld VX3, X, 12 * SIZE
+    xvfadd.d res2, VX2, VX3
+    xvfadd.d res1, res1, res2
+#else
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 8 * SIZE
+    xvfadd.s res2, VX0, VX1
+    xvfadd.s res1, res2, res1
+#endif
+    addi.d X, X, 16 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L11
+    .align 3
+
+.L12:
+#ifdef DOUBLE
+    xvpickve.d VX1, res1, 1
+    xvpickve.d VX2, res1, 2
+    xvpickve.d VX3, res1, 3
+    xvfadd.d res1, VX1, res1
+    xvfadd.d res1, VX2, res1
+    xvfadd.d res1, VX3, res1
+#else
+    xvfadd.s res2, res1, res2
+    xvpickve.w VX1, res1, 1
+    xvpickve.w VX2, res1, 2
+    xvpickve.w VX3, res1, 3
+    xvfadd.s res1, VX1, res1
+    xvfadd.s res1, VX2, res1
+    xvfadd.s res1, VX3, res1
+    xvpickve.w VX0, res2, 4
+    xvpickve.w VX1, res2, 5
+    xvpickve.w VX2, res2, 6
+    xvpickve.w VX3, res2, 7
+    xvfadd.s res1, VX0, res1
+    xvfadd.s res1, VX1, res1
+    xvfadd.s res1, VX2, res1
+    xvfadd.s res1, VX2, res1
+#endif
+    .align 3
+
+.L13:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L14:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    ADD a1, a1, a2
+    ADD s1, a1, s1
+
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    blt $r0, I, .L14
+    b .L999
+    .align 3
+
+.L20:
+    bge $r0, I, .L23
+    .align 3
+
+.L21:
+#ifdef DOUBLE
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX0, t3, 2
+    xvinsgr2vr.d VX0, t4, 3
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t1, 0
+    xvinsgr2vr.d VX1, t2, 1
+    xvinsgr2vr.d VX1, t3, 2
+    xvinsgr2vr.d VX1, t4, 3
+    xvfadd.d res2, VX0, VX1
+    xvfadd.d res1, res1, res2
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX0, t3, 2
+    xvinsgr2vr.d VX0, t4, 3
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.d VX1, t1, 0
+    xvinsgr2vr.d VX1, t2, 1
+    xvinsgr2vr.d VX1, t3, 2
+    xvinsgr2vr.d VX1, t4, 3
+    xvfadd.d res2, VX0, VX1
+    xvfadd.d res1, res1, res2
+#else
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.w VX0, t1, 0
+    xvinsgr2vr.w VX0, t2, 1
+    xvinsgr2vr.w VX0, t3, 2
+    xvinsgr2vr.w VX0, t4, 3
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.w VX0, t1, 4
+    xvinsgr2vr.w VX0, t2, 5
+    xvinsgr2vr.w VX0, t3, 6
+    xvinsgr2vr.w VX0, t4, 7
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.w VX1, t1, 0
+    xvinsgr2vr.w VX1, t2, 1
+    xvinsgr2vr.w VX1, t3, 2
+    xvinsgr2vr.w VX1, t4, 3
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    xvinsgr2vr.w VX1, t1, 4
+    xvinsgr2vr.w VX1, t2, 5
+    xvinsgr2vr.w VX1, t3, 6
+    xvinsgr2vr.w VX1, t4, 7
+    xvfadd.s res2, VX0, VX1
+    xvfadd.s res1, res2, res1
+#endif
+    addi.d  I, I, -1
+    blt $r0, I, .L21
+    .align 3
+
+.L22:
+#ifdef DOUBLE
+    xvpickve.d VX1, res1, 1
+    xvpickve.d VX2, res1, 2
+    xvpickve.d VX3, res1, 3
+    xvfadd.d res1, VX1, res1
+    xvfadd.d res1, VX2, res1
+    xvfadd.d res1, VX3, res1
+#else
+    xvfadd.s res2, res1, res2
+    xvpickve.w VX1, res1, 1
+    xvpickve.w VX2, res1, 2
+    xvpickve.w VX3, res1, 3
+    xvfadd.s res1, VX1, res1
+    xvfadd.s res1, VX2, res1
+    xvfadd.s res1, VX3, res1
+    xvpickve.w VX0, res2, 4
+    xvpickve.w VX1, res2, 5
+    xvpickve.w VX2, res2, 6
+    xvpickve.w VX3, res2, 7
+    xvfadd.s res1, VX0, res1
+    xvfadd.s res1, VX1, res1
+    xvfadd.s res1, VX2, res1
+    xvfadd.s res1, VX2, res1
+#endif
+    .align 3
+
+.L23:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L24:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    ADD a1, a1, a2
+    ADD s1, a1, s1
+
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    blt $r0, I, .L24
+    .align 3
+
+.L999:
+    fmov.s $f0, $f16
+    jirl $r0, $r1, 0x0
+    .align 3
+
+    EPILOGUE
diff --git a/kernel/loongarch64/csum_lsx.S b/kernel/loongarch64/csum_lsx.S
new file mode 100644
index 000000000..8de8e27ca
--- /dev/null
+++ b/kernel/loongarch64/csum_lsx.S
@@ -0,0 +1,266 @@
+/*******************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#define N      $r4
+#define X      $r5
+#define INCX   $r6
+#define I      $r17
+#define TEMP   $r18
+#define t1     $r15
+#define t2     $r12
+#define t3     $r13
+#define t4     $r14
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define s1     $f16
+#define VX0    $vr12
+#define VX1    $vr13
+#define VX2    $vr14
+#define VX3    $vr15
+#define res1   $vr16
+#define res2   $vr17
+#define res3   $vr18
+    PROLOGUE
+    vxor.v res1, res1, res1
+    vxor.v res2, res2, res2
+    bge $r0, N, .L999
+    bge $r0, INCX, .L999
+    li.d  TEMP, 1
+    slli.d  TEMP, TEMP, ZBASE_SHIFT
+    slli.d  INCX, INCX, ZBASE_SHIFT
+    srai.d I, N, 3
+    bne INCX, TEMP, .L20
+    bge $r0, I, .L13
+    .align 3
+
+.L11:
+#ifdef DOUBLE
+    vld VX0, X, 0 * SIZE
+    vld VX1, X, 2 * SIZE
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+    vld VX2, X, 4 * SIZE
+    vld VX3, X, 6 * SIZE
+    vfadd.d res2, VX2, VX3
+    vfadd.d res1, res1, res2
+    vld VX0, X, 8 * SIZE
+    vld VX1, X, 10 * SIZE
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+    vld VX2, X, 12 * SIZE
+    vld VX3, X, 14 * SIZE
+    vfadd.d res2, VX2, VX3
+    vfadd.d res1, res1, res2
+#else
+    vld VX0, X, 0 * SIZE
+    vld VX1, X, 4 * SIZE
+    vfadd.s res2, VX0, VX1
+    vld VX2, X, 8 * SIZE
+    vld VX3, X, 12 * SIZE
+    vfadd.s res3, VX2, VX3
+    vfadd.s res2, res3, res2
+    vfadd.s res1, res1, res2
+#endif
+
+    addi.d  I, I, -1
+    addi.d X, X, 16 * SIZE
+    blt $r0, I, .L11
+    .align 3
+
+.L12:
+#ifdef DOUBLE
+    vreplvei.d VX1, res1, 1
+    vfadd.d res1, VX1, res1
+#else
+    vreplvei.w VX1, res1, 1
+    vreplvei.w VX2, res1, 2
+    vreplvei.w VX3, res1, 3
+    vfadd.s res1, VX1, res1
+    vfadd.s res1, VX2, res1
+    vfadd.s res1, VX3, res1
+#endif
+    .align 3
+
+.L13:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L14:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    ADD a1, a1, a2
+    ADD s1, a1, s1
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    blt $r0, I, .L14
+    b .L999
+    .align 3
+
+.L20:
+    bge $r0, I, .L23
+    .align 3
+
+.L21:
+#ifdef DOUBLE
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.d VX0, t1, 0
+    vinsgr2vr.d VX0, t2, 1
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    vinsgr2vr.d VX1, t1, 0
+    vinsgr2vr.d VX1, t2, 1
+    add.d X, X, INCX
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.d VX0, t3, 0
+    vinsgr2vr.d VX0, t4, 1
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    vinsgr2vr.d VX1, t3, 0
+    vinsgr2vr.d VX1, t4, 1
+    add.d X, X, INCX
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.d VX0, t1, 0
+    vinsgr2vr.d VX0, t2, 1
+    ld.d t1, X, 0 * SIZE
+    ld.d t2, X, 1 * SIZE
+    vinsgr2vr.d VX1, t1, 0
+    vinsgr2vr.d VX1, t2, 1
+    add.d X, X, INCX
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.d VX0, t3, 0
+    vinsgr2vr.d VX0, t4, 1
+    ld.d t3, X, 0 * SIZE
+    ld.d t4, X, 1 * SIZE
+    vinsgr2vr.d VX1, t3, 0
+    vinsgr2vr.d VX1, t4, 1
+    add.d X, X, INCX
+    vfadd.d res2, VX0, VX1
+    vfadd.d res1, res1, res2
+#else
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.w VX0, t1, 0
+    vinsgr2vr.w VX0, t2, 1
+    vinsgr2vr.w VX0, t3, 2
+    vinsgr2vr.w VX0, t4, 3
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.w VX1, t1, 0
+    vinsgr2vr.w VX1, t2, 1
+    vinsgr2vr.w VX1, t3, 2
+    vinsgr2vr.w VX1, t4, 3
+    vfadd.s res2, VX0, VX1
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.w VX2, t1, 0
+    vinsgr2vr.w VX2, t2, 1
+    vinsgr2vr.w VX2, t3, 2
+    vinsgr2vr.w VX2, t4, 3
+    ld.w t1, X, 0 * SIZE
+    ld.w t2, X, 1 * SIZE
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    ld.w t4, X, 1 * SIZE
+    add.d X, X, INCX
+    vinsgr2vr.w VX3, t1, 0
+    vinsgr2vr.w VX3, t2, 1
+    vinsgr2vr.w VX3, t3, 2
+    vinsgr2vr.w VX3, t4, 3
+    vfadd.s res3, VX2, VX3
+    vfadd.s res2, res3, res2
+    vfadd.s res1, res1, res2
+#endif
+    addi.d  I, I, -1
+    blt $r0, I, .L21
+    .align 3
+
+.L22:
+#ifdef DOUBLE
+    vreplvei.d VX1, res1, 1
+    vfadd.d res1, VX1, res1
+#else
+    vreplvei.w VX1, res1, 1
+    vreplvei.w VX2, res1, 2
+    vreplvei.w VX3, res1, 3
+    vfadd.s res1, VX1, res1
+    vfadd.s res1, VX2, res1
+    vfadd.s res1, VX3, res1
+#endif
+    .align 3
+
+.L23:
+    andi I, N, 7
+    bge $r0, I, .L999
+    .align 3
+
+.L24:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    ADD a1, a1, a2
+    ADD s1, a1, s1
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    blt $r0, I, .L24
+    .align 3
+
+.L999:
+    fmov.s $f0, $f16
+    jirl $r0, $r1, 0x0
+    .align 3
+
+    EPILOGUE
diff --git a/kernel/loongarch64/cswap_lasx.S b/kernel/loongarch64/cswap_lasx.S
new file mode 100644
index 000000000..d53773d5a
--- /dev/null
+++ b/kernel/loongarch64/cswap_lasx.S
@@ -0,0 +1,394 @@
+/*******************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+#define Y      $r9
+#define INCY   $r10
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define YY     $r6
+#define t1     $r14
+#define t2     $r15
+#define t3     $r16
+#define t4     $r19
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define b1     $f16
+#define b2     $f17
+#define b3     $f18
+#define b4     $f19
+#define VX0    $xr12
+#define VX1    $xr13
+#define VX2    $xr14
+#define VX3    $xr15
+
+
+    PROLOGUE
+    bge $r0, N, .L999
+    li.d TEMP, 1
+    slli.d  TEMP, TEMP, ZBASE_SHIFT
+    slli.d  INCX, INCX, ZBASE_SHIFT
+    slli.d  INCY, INCY, ZBASE_SHIFT
+    srai.d I, N, 2
+    bne INCX, TEMP, .L20
+    bne INCY, TEMP, .L12 // INCX==1 and INCY!=1
+    b .L11  // INCX==1 and INCY==1
+.L20:
+    bne INCY, TEMP, .L22 // INCX!=1 and INCY!=1
+    b .L21 // INCX!=1 and INCY==1
+
+.L11:
+    bge $r0, I, .L112
+    .align 3
+
+.L111:
+#ifdef DOUBLE
+    xvld VX0, X, 0 * SIZE
+    xvld VX1, X, 4 * SIZE
+    xvld VX2, Y, 0 * SIZE
+    xvld VX3, Y, 4 * SIZE
+    xvst VX2, X, 0 * SIZE
+    xvst VX3, X, 4 * SIZE
+    xvst VX0, Y, 0 * SIZE
+    xvst VX1, Y, 4 * SIZE
+#else
+    xvld VX0, X, 0 * SIZE
+    xvld VX2, Y, 0 * SIZE
+    xvst VX2, X, 0 * SIZE
+    xvst VX0, Y, 0 * SIZE
+#endif
+    addi.d X, X, 8 * SIZE
+    addi.d Y, Y, 8 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L111
+    .align 3
+
+.L112:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+.L113:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    addi.d  Y, Y, 2 * SIZE
+    blt $r0, I, .L113
+    b .L999
+    .align 3
+
+.L12: // INCX==1 and INCY!=1
+    bge $r0, I, .L122
+    .align 3
+
+.L121:
+#ifdef DOUBLE
+    xvld VX0, X, 0 * SIZE
+    ld.d t1, Y, 0 * SIZE
+    xvstelm.d VX0, Y, 0 * SIZE, 0
+    ld.d t2, Y, 1 * SIZE
+    xvstelm.d VX0, Y, 1 * SIZE, 1
+    add.d Y, Y, INCY
+    ld.d t3, Y, 0 * SIZE
+    xvstelm.d VX0, Y, 0 * SIZE, 2
+    ld.d t4, Y, 1 * SIZE
+    xvstelm.d VX0, Y, 1 * SIZE, 3
+    xvinsgr2vr.d VX2, t1, 0
+    xvinsgr2vr.d VX2, t2, 1
+    xvinsgr2vr.d VX2, t3, 2
+    xvinsgr2vr.d VX2, t4, 3
+    add.d Y, Y, INCY
+    xvst VX2, X, 0 * SIZE
+    xvld VX1, X, 4 * SIZE
+    ld.d t1, Y, 0 * SIZE
+    xvstelm.d VX1, Y, 0 * SIZE, 0
+    ld.d t2, Y, 1 * SIZE
+    xvstelm.d VX1, Y, 1 * SIZE, 1
+    add.d Y, Y, INCY
+    ld.d t3, Y, 0 * SIZE
+    xvstelm.d VX1, Y, 0 * SIZE, 2
+    ld.d t4, Y, 1 * SIZE
+    xvstelm.d VX1, Y, 1 * SIZE, 3
+    xvinsgr2vr.d VX3, t1, 0
+    xvinsgr2vr.d VX3, t2, 1
+    xvinsgr2vr.d VX3, t3, 2
+    xvinsgr2vr.d VX3, t4, 3
+    add.d Y, Y, INCY
+    xvst VX3, X, 4 * SIZE
+#else
+    xvld VX0, X, 0 * SIZE
+    ld.w t1, Y, 0 * SIZE
+    xvstelm.w VX0, Y, 0 * SIZE, 0
+    ld.w t2, Y, 1 * SIZE
+    xvstelm.w VX0, Y, 1 * SIZE, 1
+    add.d Y, Y, INCY
+    ld.w t3, Y, 0 * SIZE
+    xvstelm.w VX0, Y, 0 * SIZE, 2
+    ld.w t4, Y, 1 * SIZE
+    xvstelm.w VX0, Y, 1 * SIZE, 3
+    xvinsgr2vr.w VX2, t1, 0
+    xvinsgr2vr.w VX2, t2, 1
+    xvinsgr2vr.w VX2, t3, 2
+    xvinsgr2vr.w VX2, t4, 3
+    add.d Y, Y, INCY
+    ld.w t1, Y, 0 * SIZE
+    xvstelm.w VX0, Y, 0 * SIZE, 4
+    ld.w t2, Y, 1 * SIZE
+    xvstelm.w VX0, Y, 1 * SIZE, 5
+    add.d Y, Y, INCY
+    ld.w t3, Y, 0 * SIZE
+    xvstelm.w VX0, Y, 0 * SIZE, 6
+    ld.w t4, Y, 1 * SIZE
+    xvstelm.w VX0, Y, 1 * SIZE, 7
+    xvinsgr2vr.w VX2, t1, 4
+    xvinsgr2vr.w VX2, t2, 5
+    xvinsgr2vr.w VX2, t3, 6
+    xvinsgr2vr.w VX2, t4, 7
+    add.d Y, Y, INCY
+    xvst VX2, X, 0 * SIZE
+#endif
+    addi.d X, X, 8 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L121
+    .align 3
+
+.L122:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L123:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    add.d  Y, Y, INCY
+    blt $r0, I, .L123
+    b .L999
+    .align 3
+
+.L21:
+    bge $r0, I, .L212
+    .align 3
+
+.L211:
+#ifdef DOUBLE
+    xvld VX2, Y, 0 * SIZE
+    ld.d t1, X, 0 * SIZE
+    xvstelm.d VX2, X, 0 * SIZE, 0
+    ld.d t2, X, 1 * SIZE
+    xvstelm.d VX2, X, 1 * SIZE, 1
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    xvstelm.d VX2, X, 0 * SIZE, 2
+    ld.d t4, X, 1 * SIZE
+    xvstelm.d VX2, X, 1 * SIZE, 3
+    xvinsgr2vr.d VX0, t1, 0
+    xvinsgr2vr.d VX0, t2, 1
+    xvinsgr2vr.d VX0, t3, 2
+    xvinsgr2vr.d VX0, t4, 3
+    add.d X, X, INCX
+    xvst VX0, Y, 0 * SIZE
+    xvld VX3, Y, 4 * SIZE
+    ld.d t1, X, 0 * SIZE
+    xvstelm.d VX3, X, 0 * SIZE, 0
+    ld.d t2, X, 1 * SIZE
+    xvstelm.d VX3, X, 1 * SIZE, 1
+    add.d X, X, INCX
+    ld.d t3, X, 0 * SIZE
+    xvstelm.d VX3, X, 0 * SIZE, 2
+    ld.d t4, X, 1 * SIZE
+    xvstelm.d VX3, X, 1 * SIZE, 3
+    xvinsgr2vr.d VX1, t1, 0
+    xvinsgr2vr.d VX1, t2, 1
+    xvinsgr2vr.d VX1, t3, 2
+    xvinsgr2vr.d VX1, t4, 3
+    add.d X, X, INCX
+    xvst VX1, Y, 4 * SIZE
+#else
+    xvld VX2, Y, 0 * SIZE
+    ld.w t1, X, 0 * SIZE
+    xvstelm.w VX2, X, 0 * SIZE, 0
+    ld.w t2, X, 1 * SIZE
+    xvstelm.w VX2, X, 1 * SIZE, 1
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    xvstelm.w VX2, X, 0 * SIZE, 2
+    ld.w t4, X, 1 * SIZE
+    xvstelm.w VX2, X, 1 * SIZE, 3
+    xvinsgr2vr.w VX0, t1, 0
+    xvinsgr2vr.w VX0, t2, 1
+    xvinsgr2vr.w VX0, t3, 2
+    xvinsgr2vr.w VX0, t4, 3
+    add.d X, X, INCX
+    ld.w t1, X, 0 * SIZE
+    xvstelm.w VX2, X, 0 * SIZE, 4
+    ld.w t2, X, 1 * SIZE
+    xvstelm.w VX2, X, 1 * SIZE, 5
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    xvstelm.w VX2, X, 0 * SIZE, 6
+    ld.w t4, X, 1 * SIZE
+    xvstelm.w VX2, X, 1 * SIZE, 7
+    xvinsgr2vr.w VX0, t1, 4
+    xvinsgr2vr.w VX0, t2, 5
+    xvinsgr2vr.w VX0, t3, 6
+    xvinsgr2vr.w VX0, t4, 7
+    add.d X, X, INCX
+    xvst VX0, Y, 0 * SIZE
+#endif
+    addi.d Y, Y, 8 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L211
+    .align 3
+
+.L212:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L213:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    addi.d  Y, Y, 2 * SIZE
+    blt $r0, I, .L213
+    b .L999
+    .align 3
+
+.L22:
+    bge $r0, I, .L223
+    .align 3
+    move XX, X
+
+.L222:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    add.d X, X, INCX
+    LD a3, X, 0 * SIZE
+    LD a4, X, 1 * SIZE
+    add.d X, X, INCX
+    LD b1, Y, 0 * SIZE
+    ST a1, Y, 0 * SIZE
+    LD b2, Y, 1 * SIZE
+    ST a2, Y, 1 * SIZE
+    add.d Y, Y, INCY
+    LD b3, Y, 0 * SIZE
+    ST a3, Y, 0 * SIZE
+    LD b4, Y, 1 * SIZE
+    ST a4, Y, 1 * SIZE
+    add.d Y, Y, INCY
+
+    LD a1, X, 0 * SIZE
+    ST b1, XX, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    add.d X, X, INCX
+    ST b2, XX, 1 * SIZE
+    add.d XX, XX, INCX
+    LD a3, X, 0 * SIZE
+    ST b3, XX, 0 * SIZE
+    LD a4, X, 1 * SIZE
+    add.d X, X, INCX
+    ST b4, XX, 1 * SIZE
+    add.d XX, XX, INCX
+
+    LD b1, Y, 0 * SIZE
+    ST a1, Y, 0 * SIZE
+    LD b2, Y, 1 * SIZE
+    ST a2, Y, 1 * SIZE
+    add.d Y, Y, INCY
+    LD b3, Y, 0 * SIZE
+    ST a3, Y, 0 * SIZE
+    LD b4, Y, 1 * SIZE
+    ST a4, Y, 1 * SIZE
+    add.d Y, Y, INCY
+
+    ST b1, XX, 0 * SIZE
+    ST b2, XX, 1 * SIZE
+    add.d XX, XX, INCX
+    ST b3, XX, 0 * SIZE
+    ST b4, XX, 1 * SIZE
+
+    add.d XX, XX, INCX
+    addi.d  I, I, -1
+    blt $r0, I, .L222
+    .align 3
+
+.L223:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L224:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    add.d  Y, Y, INCY
+    blt $r0, I, .L224
+    .align 3
+
+.L999:
+    move $r4, $r12
+    jirl $r0, $r1, 0x0
+    .align 3
+
+    EPILOGUE
diff --git a/kernel/loongarch64/cswap_lsx.S b/kernel/loongarch64/cswap_lsx.S
new file mode 100644
index 000000000..62a869066
--- /dev/null
+++ b/kernel/loongarch64/cswap_lsx.S
@@ -0,0 +1,421 @@
+/*******************************************************************************
+Copyright (c) 2023, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+
+#include "common.h"
+#define N      $r4
+#define X      $r7
+#define INCX   $r8
+#define Y      $r9
+#define INCY   $r10
+
+#define I      $r17
+#define TEMP   $r18
+#define XX     $r5
+#define YY     $r6
+#define t1     $r14
+#define t2     $r15
+#define t3     $r16
+#define t4     $r19
+#define a1     $f12
+#define a2     $f13
+#define a3     $f14
+#define a4     $f15
+#define b1     $f16
+#define b2     $f17
+#define b3     $f18
+#define b4     $f19
+#define VX0    $vr12
+#define VX1    $vr13
+#define VX2    $vr14
+#define VX3    $vr15
+
+
+    PROLOGUE
+    bge $r0, N, .L999
+    li.d TEMP, 1
+    slli.d  TEMP, TEMP, ZBASE_SHIFT
+    slli.d  INCX, INCX, ZBASE_SHIFT
+    slli.d  INCY, INCY, ZBASE_SHIFT
+    srai.d I, N, 2
+    bne INCX, TEMP, .L20
+    bne INCY, TEMP, .L12 // INCX==1 and INCY!=1
+    b .L11  // INCX==1 and INCY==1
+.L20:
+    bne INCY, TEMP, .L22 // INCX!=1 and INCY!=1
+    b .L21 // INCX!=1 and INCY==1
+
+.L11:
+    bge $r0, I, .L112
+    .align 3
+
+.L111:
+#ifdef DOUBLE
+    vld VX0, X, 0 * SIZE
+    vld VX1, X, 2 * SIZE
+    vld VX2, Y, 0 * SIZE
+    vld VX3, Y, 2 * SIZE
+    vst VX2, X, 0 * SIZE
+    vst VX3, X, 2 * SIZE
+    vst VX0, Y, 0 * SIZE
+    vst VX1, Y, 2 * SIZE
+    vld VX0, X, 4 * SIZE
+    vld VX1, X, 6 * SIZE
+    vld VX2, Y, 4 * SIZE
+    vld VX3, Y, 6 * SIZE
+    vst VX2, X, 4 * SIZE
+    vst VX3, X, 6 * SIZE
+    vst VX0, Y, 4 * SIZE
+    vst VX1, Y, 6 * SIZE
+#else
+    vld VX0, X, 0 * SIZE
+    vld VX1, X, 4 * SIZE
+    vld VX2, Y, 0 * SIZE
+    vld VX3, Y, 4 * SIZE
+    vst VX2, X, 0 * SIZE
+    vst VX3, X, 4 * SIZE
+    vst VX0, Y, 0 * SIZE
+    vst VX1, Y, 4 * SIZE
+#endif
+    addi.d  I, I, -1
+    addi.d X, X, 8 * SIZE
+    addi.d Y, Y, 8 * SIZE
+    blt $r0, I, .L111
+    .align 3
+
+.L112:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L113:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    addi.d  Y, Y, 2 * SIZE
+    blt $r0, I, .L113
+    b .L999
+    .align 3
+
+.L12: // INCX==1 and INCY!=1
+    bge $r0, I, .L122
+    .align 3
+
+.L121:
+#ifdef DOUBLE
+    vld VX0, X, 0 * SIZE
+    ld.d t1, Y, 0 * SIZE
+    vstelm.d VX0, Y, 0 * SIZE, 0
+    ld.d t2, Y, 1 * SIZE
+    vstelm.d VX0, Y, 1 * SIZE, 1
+    vinsgr2vr.d VX2, t1, 0
+    vinsgr2vr.d VX2, t2, 1
+    add.d Y, Y, INCY
+    vst VX2, X, 0 * SIZE
+    vld VX1, X, 2 * SIZE
+    ld.d t3, Y, 0 * SIZE
+    vstelm.d VX1, Y, 0 * SIZE, 0
+    ld.d t4, Y, 1 * SIZE
+    vstelm.d VX1, Y, 1 * SIZE, 1
+    vinsgr2vr.d VX3, t3, 0
+    vinsgr2vr.d VX3, t4, 1
+    add.d Y, Y, INCY
+    vst VX3, X, 2 * SIZE
+    vld VX0, X, 4 * SIZE
+    ld.d t1, Y, 0 * SIZE
+    vstelm.d VX0, Y, 0 * SIZE, 0
+    ld.d t2, Y, 1 * SIZE
+    vstelm.d VX0, Y, 1 * SIZE, 1
+    vinsgr2vr.d VX2, t1, 0
+    vinsgr2vr.d VX2, t2, 1
+    add.d Y, Y, INCY
+    vst VX2, X, 4 * SIZE
+    vld VX1, X, 6 * SIZE
+    ld.d t3, Y, 0 * SIZE
+    vstelm.d VX1, Y, 0 * SIZE, 0
+    ld.d t4, Y, 1 * SIZE
+    vstelm.d VX1, Y, 1 * SIZE, 1
+    vinsgr2vr.d VX3, t3, 0
+    vinsgr2vr.d VX3, t4, 1
+    add.d Y, Y, INCY
+    vst VX3, X, 6 * SIZE
+#else
+    vld VX0, X, 0 * SIZE
+    ld.w t1, Y, 0 * SIZE
+    vstelm.w VX0, Y, 0 * SIZE, 0
+    ld.w t2, Y, 1 * SIZE
+    vstelm.w VX0, Y, 1 * SIZE, 1
+    add.d Y, Y, INCY
+    ld.w t3, Y, 0 * SIZE
+    vstelm.w VX0, Y, 0 * SIZE, 2
+    ld.w t4, Y, 1 * SIZE
+    vstelm.w VX0, Y, 1 * SIZE, 3
+    vinsgr2vr.w VX2, t1, 0
+    vinsgr2vr.w VX2, t2, 1
+    vinsgr2vr.w VX2, t3, 2
+    vinsgr2vr.w VX2, t4, 3
+    add.d Y, Y, INCY
+    vst VX2, X, 0 * SIZE
+
+    vld VX1, X, 4 * SIZE
+    ld.w t1, Y, 0 * SIZE
+    vstelm.w VX1, Y, 0 * SIZE, 0
+    ld.w t2, Y, 1 * SIZE
+    vstelm.w VX1, Y, 1 * SIZE, 1
+    add.d Y, Y, INCY
+    ld.w t3, Y, 0 * SIZE
+    vstelm.w VX1, Y, 0 * SIZE, 2
+    ld.w t4, Y, 1 * SIZE
+    vstelm.w VX1, Y, 1 * SIZE, 3
+    vinsgr2vr.w VX3, t1, 0
+    vinsgr2vr.w VX3, t2, 1
+    vinsgr2vr.w VX3, t3, 2
+    vinsgr2vr.w VX3, t4, 3
+    add.d Y, Y, INCY
+    vst VX3, X, 4 * SIZE
+#endif
+    addi.d X, X, 8 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L121
+    .align 3
+
+.L122:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L123:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    addi.d  X, X, 2 * SIZE
+    add.d  Y, Y, INCY
+    blt $r0, I, .L123
+    b .L999
+    .align 3
+
+.L21:// INCX!=1 and INCY==1
+    bge $r0, I, .L212
+    .align 3
+
+.L211:
+#ifdef DOUBLE
+    vld VX2, Y, 0 * SIZE
+    ld.d t1, X, 0 * SIZE
+    vstelm.d VX2, X, 0 * SIZE, 0
+    ld.d t2, X, 1 * SIZE
+    vstelm.d VX2, X, 1 * SIZE, 1
+    vinsgr2vr.d VX0, t1, 0
+    vinsgr2vr.d VX0, t2, 1
+    add.d X, X, INCX
+    vst VX0, Y, 0 * SIZE
+    vld VX3, Y, 2 * SIZE
+    ld.d t3, X, 0 * SIZE
+    vstelm.d VX3, X, 0 * SIZE, 0
+    ld.d t4, X, 1 * SIZE
+    vstelm.d VX3, X, 1 * SIZE, 1
+    vinsgr2vr.d VX1, t3, 0
+    vinsgr2vr.d VX1, t4, 1
+    add.d X, X, INCX
+    vst VX1, Y, 2 * SIZE
+    vld VX2, Y, 4 * SIZE
+    ld.d t1, X, 0 * SIZE
+    vstelm.d VX2, X, 0 * SIZE, 0
+    ld.d t2, X, 1 * SIZE
+    vstelm.d VX2, X, 1 * SIZE, 1
+    vinsgr2vr.d VX0, t1, 0
+    vinsgr2vr.d VX0, t2, 1
+    add.d X, X, INCX
+    vst VX0, Y, 4 * SIZE
+    vld VX3, Y, 6 * SIZE
+    ld.d t3, X, 0 * SIZE
+    vstelm.d VX3, X, 0 * SIZE, 0
+    ld.d t4, X, 1 * SIZE
+    vstelm.d VX3, X, 1 * SIZE, 1
+    vinsgr2vr.d VX1, t3, 0
+    vinsgr2vr.d VX1, t4, 1
+    add.d X, X, INCX
+    vst VX1, Y, 6 * SIZE
+#else
+    vld VX2, Y, 0 * SIZE
+    ld.w t1, X, 0 * SIZE
+    vstelm.w VX2, X, 0 * SIZE, 0
+    ld.w t2, X, 1 * SIZE
+    vstelm.w VX2, X, 1 * SIZE, 1
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    vstelm.w VX2, X, 0 * SIZE, 2
+    ld.w t4, X, 1 * SIZE
+    vstelm.w VX2, X, 1 * SIZE, 3
+    vinsgr2vr.w VX0, t1, 0
+    vinsgr2vr.w VX0, t2, 1
+    vinsgr2vr.w VX0, t3, 2
+    vinsgr2vr.w VX0, t4, 3
+    add.d X, X, INCX
+    vst VX0, Y, 0 * SIZE
+
+    vld VX3, Y, 4 * SIZE
+    ld.w t1, X, 0 * SIZE
+    vstelm.w VX3, X, 0 * SIZE, 0
+    ld.w t2, X, 1 * SIZE
+    vstelm.w VX3, X, 1 * SIZE, 1
+    add.d X, X, INCX
+    ld.w t3, X, 0 * SIZE
+    vstelm.w VX3, X, 0 * SIZE, 2
+    ld.w t4, X, 1 * SIZE
+    vstelm.w VX3, X, 1 * SIZE, 3
+    vinsgr2vr.w VX1, t1, 0
+    vinsgr2vr.w VX1, t2, 1
+    vinsgr2vr.w VX1, t3, 2
+    vinsgr2vr.w VX1, t4, 3
+    add.d X, X, INCX
+    vst VX1, Y, 4 * SIZE
+#endif
+    addi.d Y, Y, 8 * SIZE
+    addi.d  I, I, -1
+    blt $r0, I, .L211
+    .align 3
+
+.L212:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L213:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    addi.d  Y, Y, 2 * SIZE
+    blt $r0, I, .L213
+    b .L999
+    .align 3
+
+.L22:
+    bge $r0, I, .L223
+    .align 3
+    move XX, X
+
+.L222:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    add.d X, X, INCX
+    LD a3, X, 0 * SIZE
+    LD a4, X, 1 * SIZE
+    add.d X, X, INCX
+    LD b1, Y, 0 * SIZE
+    ST a1, Y, 0 * SIZE
+    LD b2, Y, 1 * SIZE
+    ST a2, Y, 1 * SIZE
+    add.d Y, Y, INCY
+    LD b3, Y, 0 * SIZE
+    ST a3, Y, 0 * SIZE
+    LD b4, Y, 1 * SIZE
+    ST a4, Y, 1 * SIZE
+    add.d Y, Y, INCY
+
+    LD a1, X, 0 * SIZE
+    ST b1, XX, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    add.d X, X, INCX
+    ST b2, XX, 1 * SIZE
+    add.d XX, XX, INCX
+    LD a3, X, 0 * SIZE
+    ST b3, XX, 0 * SIZE
+    LD a4, X, 1 * SIZE
+    add.d X, X, INCX
+    ST b4, XX, 1 * SIZE
+    add.d XX, XX, INCX
+
+    LD b1, Y, 0 * SIZE
+    ST a1, Y, 0 * SIZE
+    LD b2, Y, 1 * SIZE
+    ST a2, Y, 1 * SIZE
+    add.d Y, Y, INCY
+    LD b3, Y, 0 * SIZE
+    ST a3, Y, 0 * SIZE
+    LD b4, Y, 1 * SIZE
+    ST a4, Y, 1 * SIZE
+    add.d Y, Y, INCY
+
+    ST b1, XX, 0 * SIZE
+    ST b2, XX, 1 * SIZE
+    add.d XX, XX, INCX
+    ST b3, XX, 0 * SIZE
+    ST b4, XX, 1 * SIZE
+    add.d XX, XX, INCX
+    addi.d  I, I, -1
+    blt $r0, I, .L222
+    .align 3
+
+.L223:
+    andi I, N, 3
+    bge $r0, I, .L999
+    .align 3
+
+.L224:
+    LD a1, X, 0 * SIZE
+    LD a2, X, 1 * SIZE
+    LD a3, Y, 0 * SIZE
+    LD a4, Y, 1 * SIZE
+    ST a1, Y, 0 * SIZE
+    ST a2, Y, 1 * SIZE
+    ST a3, X, 0 * SIZE
+    ST a4, X, 1 * SIZE
+
+    addi.d I, I, -1
+    add.d  X, X, INCX
+    add.d  Y, Y, INCY
+    blt $r0, I, .L224
+    .align 3
+
+.L999:
+    move $r4, $r12
+    jirl $r0, $r1, 0x0
+    .align 3
+
+    EPILOGUE