From 9c522272146423c1ef9fb9e071737a8ad26e844e Mon Sep 17 00:00:00 2001
From: dengjianbo <dengjianbo@loongson.cn>
Date: Tue, 15 Aug 2023 09:11:53 +0800
Subject: [PATCH 07/29] Loongarch: Add ifunc support for memcpy{aligned,
 unaligned, lsx, lasx} and memmove{aligned, unaligned, lsx, lasx}

These implementations improve the time to copy data in the glibc
microbenchmark as below:
memcpy-lasx       reduces the runtime about 8%-76%
memcpy-lsx        reduces the runtime about 8%-72%
memcpy-unaligned  reduces the runtime of unaligned data copying up to 40%
memcpy-aligned    reduece the runtime of unaligned data copying up to 25%
memmove-lasx      reduces the runtime about 20%-73%
memmove-lsx       reduces the runtime about 50%
memmove-unaligned reduces the runtime of unaligned data moving up to 40%
memmove-aligned   reduces the runtime of unaligned data moving up to 25%

Signed-off-by: Peng Fan <fanpeng@loongson.cn>
Signed-off-by: ticat_fp <fanpeng@loongson.cn>
---
 sysdeps/loongarch/lp64/multiarch/Makefile     |   5 +
 .../lp64/multiarch/ifunc-impl-list.c          |  19 +
 sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h |  45 +
 .../loongarch/lp64/multiarch/memcpy-aligned.S | 783 ++++++++++++++++++
 .../loongarch/lp64/multiarch/memcpy-lasx.S    |  20 +
 sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S |  20 +
 .../lp64/multiarch/memcpy-unaligned.S         | 247 ++++++
 sysdeps/loongarch/lp64/multiarch/memcpy.c     |  37 +
 .../lp64/multiarch/memmove-aligned.S          |  20 +
 .../loongarch/lp64/multiarch/memmove-lasx.S   | 287 +++++++
 .../loongarch/lp64/multiarch/memmove-lsx.S    | 534 ++++++++++++
 .../lp64/multiarch/memmove-unaligned.S        | 380 +++++++++
 sysdeps/loongarch/lp64/multiarch/memmove.c    |  38 +
 13 files changed, 2435 insertions(+)
 create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy.c
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-aligned.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-lasx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-lsx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove.c

diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile
index 110a8c5c..afa51041 100644
--- a/sysdeps/loongarch/lp64/multiarch/Makefile
+++ b/sysdeps/loongarch/lp64/multiarch/Makefile
@@ -9,5 +9,10 @@ sysdep_routines += \
   strchrnul-aligned \
   strchrnul-lsx \
   strchrnul-lasx \
+  memcpy-aligned \
+  memcpy-unaligned \
+  memmove-unaligned \
+  memmove-lsx \
+  memmove-lasx \
 # sysdep_routines
 endif
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
index c7164b45..25eb96b0 100644
--- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
@@ -53,5 +53,24 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 #endif
 	      IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_aligned)
 	      )
+
+  IFUNC_IMPL (i, name, memcpy,
+#if !defined __loongarch_soft_float
+              IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_LASX, __memcpy_lasx)
+              IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_LSX, __memcpy_lsx)
+#endif
+              IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_UAL, __memcpy_unaligned)
+              IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_aligned)
+              )
+
+  IFUNC_IMPL (i, name, memmove,
+#if !defined __loongarch_soft_float
+              IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_LASX, __memmove_lasx)
+              IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_LSX, __memmove_lsx)
+#endif
+              IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_UAL, __memmove_unaligned)
+              IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_aligned)
+              )
+
   return i;
 }
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h b/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h
new file mode 100644
index 00000000..3be67da6
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h
@@ -0,0 +1,45 @@
+/* Common definition for ifunc selection implementation.
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <ldsodefs.h>
+#include <ifunc-init.h>
+
+#if !defined __loongarch_soft_float
+extern __typeof (REDIRECT_NAME) OPTIMIZE (lasx) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (lsx) attribute_hidden;
+#endif
+
+extern __typeof (REDIRECT_NAME) OPTIMIZE (aligned) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (unaligned) attribute_hidden;
+
+static inline void *
+IFUNC_SELECTOR (void)
+{
+#if !defined __loongarch_soft_float
+  if (SUPPORT_LASX)
+    return OPTIMIZE (lasx);
+  else if (SUPPORT_LSX)
+    return OPTIMIZE (lsx);
+  else
+#endif
+  if (SUPPORT_UAL)
+    return OPTIMIZE (unaligned);
+  else
+    return OPTIMIZE (aligned);
+}
diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S b/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S
new file mode 100644
index 00000000..299dd49c
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S
@@ -0,0 +1,783 @@
+/* Optimized memcpy_aligned implementation using basic Loongarch instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc)
+# define MEMCPY_NAME __memcpy_aligned
+# define MEMMOVE_NAME __memmove_aligned
+#else
+# define MEMCPY_NAME memcpy
+# define MEMMOVE_NAME memmove
+#endif
+
+#define LD_64(reg, n)            \
+    ld.d        t0, reg, n;      \
+    ld.d        t1, reg, n + 8;  \
+    ld.d        t2, reg, n + 16; \
+    ld.d        t3, reg, n + 24; \
+    ld.d        t4, reg, n + 32; \
+    ld.d        t5, reg, n + 40; \
+    ld.d        t6, reg, n + 48; \
+    ld.d        t7, reg, n + 56;
+
+#define ST_64(reg, n)            \
+    st.d        t0, reg, n;      \
+    st.d        t1, reg, n + 8;  \
+    st.d        t2, reg, n + 16; \
+    st.d        t3, reg, n + 24; \
+    st.d        t4, reg, n + 32; \
+    st.d        t5, reg, n + 40; \
+    st.d        t6, reg, n + 48; \
+    st.d        t7, reg, n + 56;
+
+LEAF(MEMMOVE_NAME, 6)
+    sub.d       t0, a0, a1
+    bltu        t0, a2, L(copy_back)
+END(MEMMOVE_NAME)
+
+LEAF_NO_ALIGN(MEMCPY_NAME)
+    srai.d      a3, a2, 4
+    beqz        a3, L(short_data)
+
+    move        a4, a0
+    andi        a5, a0, 0x7
+    andi        a6, a1, 0x7
+    li.d        t8, 8
+    beqz        a5, L(check_align)
+
+    sub.d       t2, t8, a5
+    sub.d       a2, a2, t2
+    pcaddi      t1, 20
+    slli.d      t3, t2, 3
+
+    add.d       a1, a1, t2
+    sub.d       t1, t1, t3
+    add.d       a4, a4, t2
+    jr          t1
+
+L(al7):
+    ld.b        t0, a1, -7
+    st.b        t0, a4, -7
+L(al6):
+    ld.b        t0, a1, -6
+    st.b        t0, a4, -6
+L(al5):
+    ld.b        t0, a1, -5
+    st.b        t0, a4, -5
+L(al4):
+    ld.b        t0, a1, -4
+    st.b        t0, a4, -4
+L(al3):
+    ld.b        t0, a1, -3
+    st.b        t0, a4, -3
+L(al2):
+    ld.b        t0, a1, -2
+    st.b        t0, a4, -2
+L(al1):
+    ld.b        t0, a1, -1
+    st.b        t0, a4, -1
+
+L(check_align):
+    bne         a5, a6, L(unalign)
+    srai.d      a3, a2, 4
+    beqz        a3, L(al_less_16bytes)
+    andi        a3, a2, 0x3f
+
+    beq         a3, a2, L(al_less_64bytes)
+    sub.d       t0, a2, a3
+    move        a2, a3
+    add.d       a5, a1, t0
+
+L(loop_64bytes):
+    LD_64(a1, 0)
+    addi.d      a1, a1, 64
+    ST_64(a4, 0)
+
+    addi.d      a4, a4, 64
+    bne         a1, a5, L(loop_64bytes)
+
+L(al_less_64bytes):
+    srai.d     a3, a2, 5
+    beqz       a3, L(al_less_32bytes)
+
+    ld.d       t0, a1, 0
+    ld.d       t1, a1, 8
+    ld.d       t2, a1, 16
+    ld.d       t3, a1, 24
+
+    addi.d     a1, a1, 32
+    addi.d     a2, a2, -32
+
+    st.d       t0, a4, 0
+    st.d       t1, a4, 8
+    st.d       t2, a4, 16
+    st.d       t3, a4, 24
+
+    addi.d     a4, a4, 32
+
+L(al_less_32bytes):
+    srai.d     a3, a2, 4
+    beqz       a3, L(al_less_16bytes)
+
+    ld.d       t0, a1, 0
+    ld.d       t1, a1, 8
+    addi.d     a1, a1, 16
+    addi.d     a2, a2, -16
+
+    st.d       t0, a4, 0
+    st.d       t1, a4, 8
+    addi.d     a4, a4, 16
+
+L(al_less_16bytes):
+    srai.d     a3, a2, 3
+    beqz       a3, L(al_less_8bytes)
+
+    ld.d       t0, a1, 0
+    addi.d     a1, a1, 8
+    addi.d     a2, a2, -8
+    st.d       t0, a4, 0
+    addi.d     a4, a4, 8
+
+L(al_less_8bytes):
+    srai.d      a3, a2, 2
+    beqz        a3, L(al_less_4bytes)
+
+    ld.w        t0, a1, 0
+    addi.d      a1, a1, 4
+    addi.d      a2, a2, -4
+    st.w        t0, a4, 0
+    addi.d      a4, a4, 4
+
+L(al_less_4bytes):
+    srai.d      a3, a2, 1
+    beqz        a3, L(al_less_2bytes)
+
+    ld.h        t0, a1, 0
+    addi.d      a1, a1, 2
+    addi.d      a2, a2, -2
+    st.h        t0, a4, 0
+    addi.d      a4, a4, 2
+
+L(al_less_2bytes):
+    beqz        a2, L(al_less_1byte)
+
+    ld.b        t0, a1, 0
+    st.b        t0, a4, 0
+
+L(al_less_1byte):
+    jr          ra
+
+L(unalign):
+    andi        a5, a1, 0x7
+    bstrins.d   a1, zero, 2, 0
+    sub.d       t8, t8, a5
+    slli.d      a5, a5, 3
+
+    ld.d        t0, a1, 0
+    addi.d      a1, a1, 8
+    slli.d      a6, t8, 3
+    srl.d       a7, t0, a5
+
+    srai.d      a3, a2, 4
+    beqz        a3, L(un_less_16bytes)
+    andi        a3, a2, 0x3f
+    beq         a3, a2, L(un_less_64bytes)
+
+    sub.d       t0, a2, a3
+    move        a2, a3
+    add.d       a3, a1, t0
+
+L(un_long_bytes):
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a1, 16
+    ld.d        t3, a1, 24
+
+    srl.d       t4, t0, a5
+    sll.d       t0, t0, a6
+    srl.d       t5, t1, a5
+    sll.d       t1, t1, a6
+
+    srl.d       t6, t2, a5
+    sll.d       t2, t2, a6
+    srl.d       t7, t3, a5
+    sll.d       t3, t3, a6
+
+    or          t0, a7, t0
+    or          t1, t4, t1
+    or          t2, t5, t2
+    or          t3, t6, t3
+
+    ld.d        t4, a1, 32
+    ld.d        t5, a1, 40
+    ld.d        t6, a1, 48
+    ld.d        a7, a1, 56
+
+    st.d        t0, a4, 0
+    st.d        t1, a4, 8
+    st.d        t2, a4, 16
+    st.d        t3, a4, 24
+
+    addi.d      a1, a1, 64
+
+    srl.d       t0, t4, a5
+    sll.d       t4, t4, a6
+    srl.d       t1, t5, a5
+    sll.d       t5, t5, a6
+
+    srl.d       t2, t6, a5
+    sll.d       t6, t6, a6
+    sll.d       t3, a7, a6
+    srl.d       a7, a7, a5
+
+    or          t4, t7, t4
+    or          t5, t0, t5
+    or          t6, t1, t6
+    or          t3, t2, t3
+
+    st.d        t4, a4, 32
+    st.d        t5, a4, 40
+    st.d        t6, a4, 48
+    st.d        t3, a4, 56
+
+    addi.d      a4, a4, 64
+    bne         a3, a1, L(un_long_bytes)
+
+L(un_less_64bytes):
+    srai.d	a3, a2, 5
+    beqz	a3, L(un_less_32bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a1, 16
+    ld.d        t3, a1, 24
+
+    addi.d      a1, a1, 32
+    addi.d      a2, a2, -32
+
+    srl.d       t4, t0, a5
+    sll.d       t0, t0, a6
+    srl.d       t5, t1, a5
+    sll.d       t1, t1, a6
+
+    srl.d       t6, t2, a5
+    sll.d       t2, t2, a6
+    or          t0, a7, t0
+    srl.d       a7, t3, a5
+    sll.d       t3, t3, a6
+
+    or          t1, t4, t1
+    or          t2, t5, t2
+    or          t3, t6, t3
+
+    st.d        t0, a4, 0
+    st.d        t1, a4, 8
+    st.d        t2, a4, 16
+    st.d        t3, a4, 24
+
+    addi.d      a4, a4, 32
+
+L(un_less_32bytes):
+    srai.d      a3, a2, 4
+    beqz        a3, L(un_less_16bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    addi.d      a1, a1, 16
+    addi.d      a2, a2, -16
+
+    srl.d       t2, t0, a5
+    sll.d       t3, t0, a6
+    sll.d       t4, t1, a6
+    or          t3, a7, t3
+    or          t4, t2, t4
+
+    srl.d       a7, t1, a5
+    st.d        t3, a4, 0
+    st.d        t4, a4, 8
+    addi.d      a4, a4, 16
+
+L(un_less_16bytes):
+    srai.d      a3, a2, 3
+    beqz        a3, L(un_less_8bytes)
+
+    ld.d        t0, a1, 0
+    addi.d      a1, a1, 8
+    addi.d      a2, a2, -8
+    sll.d       t1, t0, a6
+
+    or          t2, a7, t1
+    srl.d       a7, t0, a5
+    st.d        t2, a4, 0
+    addi.d      a4, a4, 8
+
+L(un_less_8bytes):
+    beqz        a2, L(un_less_1byte)
+    bge         t8, a2, 1f
+
+    ld.d        t0, a1, 0
+    sll.d       t0, t0, a6
+    or          a7, a7, t0
+
+1:
+    srai.d      a3, a2, 2
+    beqz        a3, L(un_less_4bytes)
+
+    addi.d      a2, a2, -4
+    st.w        a7, a4, 0
+    addi.d      a4, a4, 4
+    srai.d      a7, a7, 32
+
+L(un_less_4bytes):
+    srai.d      a3, a2, 1
+    beqz        a3, L(un_less_2bytes)
+
+    addi.d      a2, a2, -2
+    st.h        a7, a4, 0
+    addi.d      a4, a4, 2
+    srai.d      a7, a7, 16
+
+L(un_less_2bytes):
+    beqz        a2, L(un_less_1byte)
+    st.b        a7, a4, 0
+
+L(un_less_1byte):
+    jr          ra
+
+L(short_data):
+    pcaddi      t1, 36
+    slli.d      t2, a2, 3
+    add.d       a4, a0, a2
+    sub.d       t1, t1, t2
+    add.d       a1, a1, a2
+    jr          t1
+
+L(short_15_bytes):
+    ld.b       t0, a1, -15
+    st.b       t0, a4, -15
+L(short_14_bytes):
+    ld.b       t0, a1, -14
+    st.b       t0, a4, -14
+L(short_13_bytes):
+    ld.b       t0, a1, -13
+    st.b       t0, a4, -13
+L(short_12_bytes):
+    ld.b       t0, a1, -12
+    st.b       t0, a4, -12
+L(short_11_bytes):
+    ld.b       t0, a1, -11
+    st.b       t0, a4, -11
+L(short_10_bytes):
+    ld.b       t0, a1, -10
+    st.b       t0, a4, -10
+L(short_9_bytes):
+    ld.b       t0, a1, -9
+    st.b       t0, a4, -9
+L(short_8_bytes):
+    ld.b       t0, a1, -8
+    st.b       t0, a4, -8
+L(short_7_bytes):
+    ld.b       t0, a1, -7
+    st.b       t0, a4, -7
+L(short_6_bytes):
+    ld.b       t0, a1, -6
+    st.b       t0, a4, -6
+L(short_5_bytes):
+    ld.b       t0, a1, -5
+    st.b       t0, a4, -5
+L(short_4_bytes):
+    ld.b       t0, a1, -4
+    st.b       t0, a4, -4
+L(short_3_bytes):
+    ld.b       t0, a1, -3
+    st.b       t0, a4, -3
+L(short_2_bytes):
+    ld.b       t0, a1, -2
+    st.b       t0, a4, -2
+L(short_1_bytes):
+    ld.b       t0, a1, -1
+    st.b       t0, a4, -1
+    jr         ra
+
+L(copy_back):
+    srai.d      a3, a2, 4
+    beqz        a3, L(back_short_data)
+
+    add.d       a4, a0, a2
+    add.d       a1, a1, a2
+
+    andi        a5, a4, 0x7
+    andi        a6, a1, 0x7
+    beqz        a5, L(back_check_align)
+
+    sub.d       a2, a2, a5
+    sub.d       a1, a1, a5
+    sub.d       a4, a4, a5
+
+    pcaddi      t1, 18
+    slli.d      t3, a5, 3
+    sub.d       t1, t1, t3
+    jr          t1
+
+    ld.b        t0, a1, 6
+    st.b        t0, a4, 6
+    ld.b        t0, a1, 5
+    st.b        t0, a4, 5
+    ld.b        t0, a1, 4
+    st.b        t0, a4, 4
+    ld.b        t0, a1, 3
+    st.b        t0, a4, 3
+    ld.b        t0, a1, 2
+    st.b        t0, a4, 2
+    ld.b        t0, a1, 1
+    st.b        t0, a4, 1
+    ld.b        t0, a1, 0
+    st.b        t0, a4, 0
+
+L(back_check_align):
+    bne         a5, a6, L(back_unalign)
+
+    srai.d      a3, a2, 4
+    beqz        a3, L(back_less_16bytes)
+
+    andi        a3, a2, 0x3f
+    beq         a3, a2, L(back_less_64bytes)
+
+    sub.d       t0, a2, a3
+    move        a2, a3
+    sub.d       a5, a1, t0
+
+L(back_loop_64bytes):
+    LD_64(a1, -64)
+    addi.d      a1, a1, -64
+    ST_64(a4, -64)
+
+    addi.d      a4, a4, -64
+    bne         a1, a5, L(back_loop_64bytes)
+
+L(back_less_64bytes):
+    srai.d     a3, a2, 5
+    beqz       a3, L(back_less_32bytes)
+
+    ld.d       t0, a1, -32
+    ld.d       t1, a1, -24
+    ld.d       t2, a1, -16
+    ld.d       t3, a1, -8
+
+    addi.d     a1, a1, -32
+    addi.d     a2, a2, -32
+
+    st.d       t0, a4, -32
+    st.d       t1, a4, -24
+    st.d       t2, a4, -16
+    st.d       t3, a4, -8
+
+    addi.d     a4, a4, -32
+
+L(back_less_32bytes):
+    srai.d     a3, a2, 4
+    beqz       a3, L(back_less_16bytes)
+
+    ld.d       t0, a1, -16
+    ld.d       t1, a1, -8
+
+    addi.d     a2, a2, -16
+    addi.d     a1, a1, -16
+
+    st.d       t0, a4, -16
+    st.d       t1, a4, -8
+    addi.d     a4, a4, -16
+
+L(back_less_16bytes):
+    srai.d      a3, a2, 3
+    beqz        a3, L(back_less_8bytes)
+
+    ld.d        t0, a1, -8
+    addi.d      a2, a2, -8
+    addi.d      a1, a1, -8
+
+    st.d        t0, a4, -8
+    addi.d      a4, a4, -8
+
+L(back_less_8bytes):
+    srai.d      a3, a2, 2
+    beqz        a3, L(back_less_4bytes)
+
+    ld.w        t0, a1, -4
+    addi.d      a2, a2, -4
+    addi.d      a1, a1, -4
+
+    st.w        t0, a4, -4
+    addi.d      a4, a4, -4
+
+L(back_less_4bytes):
+    srai.d      a3, a2, 1
+    beqz        a3, L(back_less_2bytes)
+
+    ld.h        t0, a1, -2
+    addi.d      a2, a2, -2
+    addi.d      a1, a1, -2
+
+    st.h        t0, a4, -2
+    addi.d      a4, a4, -2
+
+L(back_less_2bytes):
+    beqz        a2, L(back_less_1byte)
+
+    ld.b        t0, a1, -1
+    st.b        t0, a4, -1
+
+L(back_less_1byte):
+    jr          ra
+
+L(back_unalign):
+    andi        t8, a1, 0x7
+    bstrins.d   a1, zero, 2, 0
+
+    sub.d       a6, zero, t8
+
+    ld.d        t0, a1, 0
+    slli.d      a6, a6, 3
+    slli.d      a5, t8, 3
+    sll.d       a7, t0, a6
+
+    srai.d      a3, a2, 4
+    beqz        a3, L(back_un_less_16bytes)
+
+    andi        a3, a2, 0x3f
+    beq         a3, a2, L(back_un_less_64bytes)
+
+    sub.d       t0, a2, a3
+    move        a2, a3
+    sub.d       a3, a1, t0
+
+L(back_un_long_bytes):
+    ld.d        t0, a1, -8
+    ld.d        t1, a1, -16
+    ld.d        t2, a1, -24
+    ld.d        t3, a1, -32
+
+    sll.d       t4, t0, a6
+    srl.d       t0, t0, a5
+
+    sll.d       t5, t1, a6
+    srl.d       t1, t1, a5
+
+    sll.d       t6, t2, a6
+    srl.d       t2, t2, a5
+
+    sll.d       t7, t3, a6
+    srl.d       t3, t3, a5
+
+    or          t0, t0, a7
+    or          t1, t1, t4
+    or          t2, t2, t5
+    or          t3, t3, t6
+
+    ld.d        t4, a1, -40
+    ld.d        t5, a1, -48
+    ld.d        t6, a1, -56
+    ld.d        a7, a1, -64
+    st.d        t0, a4, -8
+    st.d        t1, a4, -16
+    st.d        t2, a4, -24
+    st.d        t3, a4, -32
+
+    addi.d      a1, a1, -64
+
+    sll.d       t0, t4, a6
+    srl.d       t4, t4, a5
+
+    sll.d       t1, t5, a6
+    srl.d       t5, t5, a5
+
+    sll.d       t2, t6, a6
+    srl.d       t6, t6, a5
+
+    srl.d       t3, a7, a5
+    sll.d       a7, a7, a6
+
+    or          t4, t7, t4
+    or          t5, t0, t5
+    or          t6, t1, t6
+    or          t3, t2, t3
+
+    st.d        t4, a4, -40
+    st.d        t5, a4, -48
+    st.d        t6, a4, -56
+    st.d        t3, a4, -64
+
+    addi.d      a4, a4, -64
+    bne         a3, a1, L(back_un_long_bytes)
+
+L(back_un_less_64bytes):
+    srai.d	a3, a2, 5
+    beqz	a3, L(back_un_less_32bytes)
+
+    ld.d        t0, a1, -8
+    ld.d        t1, a1, -16
+    ld.d        t2, a1, -24
+    ld.d        t3, a1, -32
+
+    addi.d      a1, a1, -32
+    addi.d      a2, a2, -32
+
+    sll.d       t4, t0, a6
+    srl.d       t0, t0, a5
+
+    sll.d       t5, t1, a6
+    srl.d       t1, t1, a5
+
+    sll.d       t6, t2, a6
+    srl.d       t2, t2, a5
+
+    or          t0, a7, t0
+
+    sll.d       a7, t3, a6
+    srl.d       t3, t3, a5
+
+    or          t1, t4, t1
+    or          t2, t5, t2
+    or          t3, t6, t3
+
+    st.d        t0, a4, -8
+    st.d        t1, a4, -16
+    st.d        t2, a4, -24
+    st.d        t3, a4, -32
+
+    addi.d      a4, a4, -32
+
+L(back_un_less_32bytes):
+    srai.d      a3, a2, 4
+    beqz        a3, L(back_un_less_16bytes)
+
+    ld.d        t0, a1, -8
+    ld.d        t1, a1, -16
+
+    addi.d      a1, a1, -16
+    addi.d      a2, a2, -16
+
+    sll.d       t2, t0, a6
+    srl.d       t3, t0, a5
+
+    srl.d       t4, t1, a5
+    or          t3, a7, t3
+    or          t4, t2, t4
+    sll.d       a7, t1, a6
+
+    st.d        t3, a4, -8
+    st.d        t4, a4, -16
+
+    addi.d      a4, a4, -16
+
+L(back_un_less_16bytes):
+    srai.d      a3, a2, 3
+    beqz        a3, L(back_un_less_8bytes)
+
+    ld.d        t0, a1, -8
+
+    addi.d      a1, a1, -8
+    addi.d      a2, a2, -8
+
+    srl.d       t1, t0, a5
+    or          t2, a7, t1
+    sll.d       a7, t0, a6
+
+    st.d        t2, a4, -8
+    addi.d      a4, a4, -8
+
+L(back_un_less_8bytes):
+    beqz        a2, L(back_end)
+    bge         t8, a2, 1f
+
+    ld.d        t0, a1, -8
+    srl.d       t0, t0, a5
+    or          a7, a7, t0
+
+1:
+    srai.d      a3, a2, 2
+    beqz        a3, L(back_un_less_4bytes)
+
+    srai.d      t0, a7, 32
+    addi.d      a2, a2, -4
+    st.w        t0, a4, -4
+    addi.d      a4, a4, -4
+    slli.d      a7, a7, 32
+
+L(back_un_less_4bytes):
+    srai.d      a3, a2, 1
+    beqz        a3, L(back_un_less_2bytes)
+    srai.d      t0, a7, 48
+    addi.d      a2, a2, -2
+    st.h        t0, a4, -2
+    addi.d      a4, a4, -2
+    slli.d      a7, a7, 16
+L(back_un_less_2bytes):
+    beqz        a2, L(back_un_less_1byte)
+    srai.d      t0, a7, 56
+    st.b        t0, a4, -1
+L(back_un_less_1byte):
+    jr          ra
+
+L(back_short_data):
+    pcaddi     t1, 34
+    slli.d     t2, a2, 3
+    sub.d      t1, t1, t2
+    jr         t1
+
+    ld.b       t0, a1, 14
+    st.b       t0, a0, 14
+    ld.b       t0, a1, 13
+    st.b       t0, a0, 13
+    ld.b       t0, a1, 12
+    st.b       t0, a0, 12
+    ld.b       t0, a1, 11
+    st.b       t0, a0, 11
+    ld.b       t0, a1, 10
+    st.b       t0, a0, 10
+    ld.b       t0, a1, 9
+    st.b       t0, a0, 9
+    ld.b       t0, a1, 8
+    st.b       t0, a0, 8
+    ld.b       t0, a1, 7
+    st.b       t0, a0, 7
+    ld.b       t0, a1, 6
+    st.b       t0, a0, 6
+    ld.b       t0, a1, 5
+    st.b       t0, a0, 5
+    ld.b       t0, a1, 4
+    st.b       t0, a0, 4
+    ld.b       t0, a1, 3
+    st.b       t0, a0, 3
+    ld.b       t0, a1, 2
+    st.b       t0, a0, 2
+    ld.b       t0, a1, 1
+    st.b       t0, a0, 1
+    ld.b       t0, a1, 0
+    st.b       t0, a0, 0
+L(back_end):
+    jr         ra
+
+END(MEMCPY_NAME)
+
+libc_hidden_builtin_def (MEMMOVE_NAME)
+libc_hidden_builtin_def (MEMCPY_NAME)
diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S b/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S
new file mode 100644
index 00000000..4aae5bf8
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S
@@ -0,0 +1,20 @@
+/* Optimized memcpy implementation using Loongarch LASX instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* memcpy is part of memmove.S */
diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S b/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S
new file mode 100644
index 00000000..6ebbe7a2
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S
@@ -0,0 +1,20 @@
+/* Optimized memcpy implementation using Loongarch LSX instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* memcpy is part of memmove.S */
diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S b/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S
new file mode 100644
index 00000000..8e60a22d
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S
@@ -0,0 +1,247 @@
+/* Optimized unaligned memcpy implementation using basic Loongarch instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc)
+
+# define MEMCPY_NAME __memcpy_unaligned
+
+# define LD_64(reg, n)           \
+    ld.d        t0, reg, n;      \
+    ld.d        t1, reg, n + 8;  \
+    ld.d        t2, reg, n + 16; \
+    ld.d        t3, reg, n + 24; \
+    ld.d        t4, reg, n + 32; \
+    ld.d        t5, reg, n + 40; \
+    ld.d        t6, reg, n + 48; \
+    ld.d        t7, reg, n + 56;
+
+# define ST_64(reg, n)           \
+    st.d        t0, reg, n;      \
+    st.d        t1, reg, n + 8;  \
+    st.d        t2, reg, n + 16; \
+    st.d        t3, reg, n + 24; \
+    st.d        t4, reg, n + 32; \
+    st.d        t5, reg, n + 40; \
+    st.d        t6, reg, n + 48; \
+    st.d        t7, reg, n + 56;
+
+LEAF(MEMCPY_NAME, 3)
+    add.d       a4, a1, a2
+    add.d       a3, a0, a2
+    li.w        a6, 16
+    bge         a6, a2, L(less_16bytes)
+
+    li.w        a6, 128
+    blt         a6, a2, L(long_bytes)
+    li.w        a6, 64
+    blt         a6, a2, L(more_64bytes)
+
+    li.w        a6, 32
+    blt         a6, a2, L(more_32bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a4, -16
+    ld.d        t3, a4, -8
+
+    st.d        t0, a0, 0
+    st.d        t1, a0, 8
+    st.d        t2, a3, -16
+    st.d        t3, a3, -8
+    jr          ra
+
+L(more_64bytes):
+    srli.d      t8, a0, 3
+    slli.d      t8, t8, 3
+    addi.d      t8, t8,  0x8
+    sub.d       a7, a0, t8
+
+    ld.d        t0, a1, 0
+    sub.d       a1, a1, a7
+    st.d        t0, a0, 0
+    add.d       a7, a7, a2
+    addi.d      a7, a7, -0x20
+
+L(loop_32):
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a1, 16
+    ld.d        t3, a1, 24
+
+    st.d        t0, t8, 0
+    st.d        t1, t8, 8
+    st.d        t2, t8, 16
+    st.d        t3, t8, 24
+
+    addi.d      t8, t8, 0x20
+    addi.d	a1, a1, 0x20
+    addi.d	a7, a7, -0x20
+    blt         zero, a7, L(loop_32)
+
+    ld.d        t4, a4, -32
+    ld.d        t5, a4, -24
+    ld.d        t6, a4, -16
+    ld.d        t7, a4, -8
+
+    st.d        t4, a3, -32
+    st.d        t5, a3, -24
+    st.d        t6, a3, -16
+    st.d        t7, a3, -8
+
+    jr          ra
+
+L(more_32bytes):
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a1, 16
+    ld.d        t3, a1, 24
+
+    ld.d        t4, a4, -32
+    ld.d        t5, a4, -24
+    ld.d        t6, a4, -16
+    ld.d        t7, a4, -8
+
+    st.d        t0, a0, 0
+    st.d        t1, a0, 8
+    st.d        t2, a0, 16
+    st.d        t3, a0, 24
+
+    st.d        t4, a3, -32
+    st.d        t5, a3, -24
+    st.d        t6, a3, -16
+    st.d        t7, a3, -8
+
+    jr          ra
+
+L(less_16bytes):
+    srai.d      a6, a2, 3
+    beqz        a6, L(less_8bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a4, -8
+    st.d        t0, a0, 0
+    st.d        t1, a3, -8
+
+    jr          ra
+
+L(less_8bytes):
+    srai.d      a6, a2, 2
+    beqz        a6, L(less_4bytes)
+
+    ld.w        t0, a1, 0
+    ld.w        t1, a4, -4
+    st.w        t0, a0, 0
+    st.w        t1, a3, -4
+
+    jr          ra
+
+L(less_4bytes):
+    srai.d      a6, a2, 1
+    beqz        a6, L(less_2bytes)
+
+    ld.h        t0, a1, 0
+    ld.h        t1, a4, -2
+    st.h        t0, a0, 0
+    st.h        t1, a3, -2
+
+    jr          ra
+
+L(less_2bytes):
+    beqz        a2, L(less_1bytes)
+
+    ld.b        t0, a1, 0
+    st.b        t0, a0, 0
+    jr          ra
+
+L(less_1bytes):
+    jr          ra
+
+L(long_bytes):
+    srli.d      t8, a0, 3
+    slli.d      t8, t8, 3
+    beq         a0, t8, L(start)
+    ld.d        t0, a1, 0
+
+    addi.d      t8, t8, 0x8
+    st.d        t0, a0, 0
+    sub.d       a7, a0, t8
+    sub.d       a1, a1, a7
+
+L(start):
+    addi.d     a5, a3, -0x80
+    blt        a5, t8, L(align_end_proc)
+
+L(loop_128):
+    LD_64(a1, 0)
+    ST_64(t8, 0)
+    LD_64(a1, 64)
+    addi.d     a1, a1, 0x80
+    ST_64(t8, 64)
+    addi.d     t8, t8, 0x80
+    bge        a5, t8, L(loop_128)
+
+L(align_end_proc):
+    sub.d      a2, a3, t8
+    pcaddi     t1, 34
+    andi       t2, a2, 0x78
+    sub.d      t1, t1, t2
+    jr         t1
+
+    ld.d       t0, a1, 112
+    st.d       t0, t8, 112
+    ld.d       t0, a1, 104
+    st.d       t0, t8, 104
+    ld.d       t0, a1, 96
+    st.d       t0, t8, 96
+    ld.d       t0, a1, 88
+    st.d       t0, t8, 88
+    ld.d       t0, a1, 80
+    st.d       t0, t8, 80
+    ld.d       t0, a1, 72
+    st.d       t0, t8, 72
+    ld.d       t0, a1, 64
+    st.d       t0, t8, 64
+    ld.d       t0, a1, 56
+    st.d       t0, t8, 56
+    ld.d       t0, a1, 48
+    st.d       t0, t8, 48
+    ld.d       t0, a1, 40
+    st.d       t0, t8, 40
+    ld.d       t0, a1, 32
+    st.d       t0, t8, 32
+    ld.d       t0, a1, 24
+    st.d       t0, t8, 24
+    ld.d       t0, a1, 16
+    st.d       t0, t8, 16
+    ld.d       t0, a1, 8
+    st.d       t0, t8, 8
+    ld.d       t0, a1, 0
+    st.d       t0, t8, 0
+    ld.d       t0, a4, -8
+    st.d       t0, a3, -8
+
+    jr         ra
+END(MEMCPY_NAME)
+
+libc_hidden_builtin_def (MEMCPY_NAME)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy.c b/sysdeps/loongarch/lp64/multiarch/memcpy.c
new file mode 100644
index 00000000..93b238ce
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memcpy.c
@@ -0,0 +1,37 @@
+/* Multiple versions of memcpy.
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* Define multiple versions only for the definition in libc.  */
+#if IS_IN (libc)
+# define memcpy __redirect_memcpy
+# include <string.h>
+# undef memcpy
+
+# define SYMBOL_NAME memcpy
+# include "ifunc-lasx.h"
+
+libc_ifunc_redirected (__redirect_memcpy, memcpy,
+		       IFUNC_SELECTOR ());
+
+# ifdef SHARED
+__hidden_ver1 (memcpy, __GI_memcpy, __redirect_memcpy)
+  __attribute__ ((visibility ("hidden"))) __attribute_copy__ (memcmp);
+# endif
+
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S b/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S
new file mode 100644
index 00000000..5354f383
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S
@@ -0,0 +1,20 @@
+/* Optimized memmove_aligned implementation using basic Loongarch instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* memmove_aligned is part of memcpy_aligned, see memcpy-aligned.S.  */
diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S b/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S
new file mode 100644
index 00000000..ff68e7a2
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S
@@ -0,0 +1,287 @@
+/* Optimized memmove implementation using Loongarch LASX instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc) && !defined __loongarch_soft_float
+
+#ifndef MEMCPY_NAME
+# define MEMCPY_NAME __memcpy_lasx
+#endif
+
+#ifndef MEMMOVE_NAME
+# define MEMMOVE_NAME __memmove_lasx
+#endif
+
+LEAF(MEMCPY_NAME, 6)
+    li.d            t0, 32
+    add.d           a3, a0, a2
+    add.d           a4, a1, a2
+    bgeu            t0, a2, L(less_32bytes)
+
+    li.d            t1, 64
+    bltu            t1, a2, L(copy_long)
+    xvld            xr0, a1, 0
+    xvld            xr1, a4, -32
+
+    xvst            xr0, a0, 0
+    xvst            xr1, a3, -32
+    jr              ra
+L(less_32bytes):
+    srli.d          t0, a2, 4
+
+    beqz            t0, L(less_16bytes)
+    vld             vr0, a1, 0
+    vld             vr1, a4, -16
+    vst             vr0, a0, 0
+
+
+    vst             vr1, a3, -16
+    jr              ra
+L(less_16bytes):
+    srli.d          t0, a2, 3
+    beqz            t0, L(less_8bytes)
+
+    ld.d            t0, a1, 0
+    ld.d            t1, a4, -8
+    st.d            t0, a0, 0
+    st.d            t1, a3, -8
+
+    jr              ra
+L(less_8bytes):
+    srli.d          t0, a2, 2
+    beqz            t0, L(less_4bytes)
+    ld.w            t0, a1, 0
+
+    ld.w            t1, a4, -4
+    st.w            t0, a0, 0
+    st.w            t1, a3, -4
+    jr              ra
+
+
+L(less_4bytes):
+    srli.d          t0, a2, 1
+    beqz            t0, L(less_2bytes)
+    ld.h            t0, a1, 0
+    ld.h            t1, a4, -2
+
+    st.h            t0, a0, 0
+    st.h            t1, a3, -2
+    jr              ra
+L(less_2bytes):
+    beqz            a2, L(less_1bytes)
+
+    ld.b            t0, a1, 0
+    st.b            t0, a0, 0
+L(less_1bytes):
+    jr              ra
+END(MEMCPY_NAME)
+
+LEAF(MEMMOVE_NAME, 6)
+
+    li.d            t0, 32
+    add.d           a3, a0, a2
+    add.d           a4, a1, a2
+    bgeu            t0, a2, L(less_32bytes)
+
+    li.d            t1, 64
+    bltu            t1, a2, L(move_long)
+    xvld            xr0, a1, 0
+    xvld            xr1, a4, -32
+
+    xvst            xr0, a0, 0
+    xvst            xr1, a3, -32
+    jr              ra
+L(move_long):
+    sub.d           t2, a0, a1
+
+    bltu            t2, a2, L(copy_back)
+L(copy_long):
+    andi            t2, a0, 0x1f
+    addi.d          a2, a2, -1
+    sub.d           t2, t0, t2
+
+
+    xvld            xr8, a1, 0
+    xvld            xr9, a4, -32
+    sub.d           t3, a2, t2
+    add.d           a5, a0, t2
+
+    andi            a2, t3, 0xff
+    add.d           a1, a1, t2
+    beq             a2, t3, L(lt256)
+    sub.d           a6, a4, a2
+
+    addi.d          a6, a6, -1
+L(loop_256):
+    xvld            xr0, a1, 0
+    xvld            xr1, a1, 32
+    xvld            xr2, a1, 64
+
+    xvld            xr3, a1, 96
+    xvld            xr4, a1, 128
+    xvld            xr5, a1, 160
+    xvld            xr6, a1, 192
+
+
+    xvld            xr7, a1, 224
+    addi.d          a1, a1, 256
+    xvst            xr0, a5, 0
+    xvst            xr1, a5, 32
+
+    xvst            xr2, a5, 64
+    xvst            xr3, a5, 96
+    xvst            xr4, a5, 128
+    xvst            xr5, a5, 160
+
+    xvst            xr6, a5, 192
+    xvst            xr7, a5, 224
+    addi.d          a5, a5, 256
+    bne             a1, a6, L(loop_256)
+
+L(lt256):
+    srli.d          t2, a2, 7
+    beqz            t2, L(lt128)
+    xvld            xr0, a1, 0
+    xvld            xr1, a1, 32
+
+
+    xvld            xr2, a1, 64
+    xvld            xr3, a1, 96
+    addi.d          a1, a1, 128
+    addi.d          a2, a2, -128
+
+    xvst            xr0, a5, 0
+    xvst            xr1, a5, 32
+    xvst            xr2, a5, 64
+    xvst            xr3, a5, 96
+
+    addi.d          a5, a5, 128
+L(lt128):
+    bltu            a2, t1, L(lt64)
+    xvld            xr0, a1, 0
+    xvld            xr1, a1, 32
+
+    addi.d          a1, a1, 64
+    addi.d          a2, a2, -64
+    xvst            xr0, a5, 0
+    xvst            xr1, a5, 32
+
+
+    addi.d          a5, a5, 64
+L(lt64):
+    bltu            a2, t0, L(lt32)
+    xvld            xr0, a1, 0
+    xvst            xr0, a5, 0
+
+L(lt32):
+    xvst            xr8, a0, 0
+    xvst            xr9, a3, -32
+    jr              ra
+    nop
+
+L(copy_back):
+    addi.d          a3, a3, -1
+    addi.d          a2, a2, -2
+    andi            t2, a3, 0x1f
+    xvld            xr8, a1, 0
+
+    xvld            xr9, a4, -32
+    sub.d           t3, a2, t2
+    sub.d           a5, a3, t2
+    sub.d           a4, a4, t2
+
+
+    andi            a2, t3, 0xff
+    beq             a2, t3, L(back_lt256)
+    add.d           a6, a1, a2
+    addi.d          a6, a6, 2
+
+L(back_loop_256):
+    xvld            xr0, a4, -33
+    xvld            xr1, a4, -65
+    xvld            xr2, a4, -97
+    xvld            xr3, a4, -129
+
+    xvld            xr4, a4, -161
+    xvld            xr5, a4, -193
+    xvld            xr6, a4, -225
+    xvld            xr7, a4, -257
+
+    addi.d          a4, a4, -256
+    xvst            xr0, a5, -32
+    xvst            xr1, a5, -64
+    xvst            xr2, a5, -96
+
+
+    xvst            xr3, a5, -128
+    xvst            xr4, a5, -160
+    xvst            xr5, a5, -192
+    xvst            xr6, a5, -224
+
+    xvst            xr7, a5, -256
+    addi.d          a5, a5, -256
+    bne             a4, a6, L(back_loop_256)
+L(back_lt256):
+    srli.d          t2, a2, 7
+
+    beqz            t2, L(back_lt128)
+    xvld            xr0, a4, -33
+    xvld            xr1, a4, -65
+    xvld            xr2, a4, -97
+
+    xvld            xr3, a4, -129
+    addi.d          a2, a2, -128
+    addi.d          a4, a4, -128
+    xvst            xr0, a5, -32
+
+
+    xvst            xr1, a5, -64
+    xvst            xr2, a5, -96
+    xvst            xr3, a5, -128
+    addi.d          a5, a5, -128
+
+L(back_lt128):
+    blt             a2, t1, L(back_lt64)
+    xvld            xr0, a4, -33
+    xvld            xr1, a4, -65
+    addi.d          a2, a2, -64
+
+    addi.d          a4, a4, -64
+    xvst            xr0, a5, -32
+    xvst            xr1, a5, -64
+    addi.d          a5, a5, -64
+
+L(back_lt64):
+    bltu            a2, t0, L(back_lt32)
+    xvld            xr0, a4, -33
+    xvst            xr0, a5, -32
+L(back_lt32):
+    xvst            xr8, a0, 0
+
+
+    xvst            xr9, a3, -31
+    jr              ra
+END(MEMMOVE_NAME)
+
+libc_hidden_builtin_def (MEMCPY_NAME)
+libc_hidden_builtin_def (MEMMOVE_NAME)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S b/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S
new file mode 100644
index 00000000..9e1502a7
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S
@@ -0,0 +1,534 @@
+/* Optimized memmove implementation using Loongarch LSX instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc) && !defined __loongarch_soft_float
+
+# define MEMCPY_NAME __memcpy_lsx
+# define MEMMOVE_NAME __memmove_lsx
+
+LEAF(MEMCPY_NAME, 6)
+    li.d            t6, 16
+    add.d           a3, a0, a2
+    add.d           a4, a1, a2
+    bgeu            t6, a2, L(less_16bytes)
+
+    li.d            t8, 64
+    li.d            t7, 32
+    bltu            t8, a2, L(copy_long)
+    bltu            t7, a2, L(more_32bytes)
+
+    vld             vr0, a1, 0
+    vld             vr1, a4, -16
+    vst             vr0, a0, 0
+    vst             vr1, a3, -16
+
+    jr              ra
+L(more_32bytes):
+    vld             vr0, a1, 0
+    vld             vr1, a1, 16
+    vld             vr2, a4, -32
+
+
+    vld             vr3, a4, -16
+    vst             vr0, a0, 0
+    vst             vr1, a0, 16
+    vst             vr2, a3, -32
+
+    vst             vr3, a3, -16
+    jr              ra
+L(less_16bytes):
+    srli.d          t0, a2, 3
+    beqz            t0, L(less_8bytes)
+
+    vldrepl.d       vr0, a1, 0
+    vldrepl.d       vr1, a4, -8
+    vstelm.d        vr0, a0, 0, 0
+    vstelm.d        vr1, a3, -8, 0
+
+    jr              ra
+L(less_8bytes):
+    srli.d          t0, a2, 2
+    beqz            t0, L(less_4bytes)
+    vldrepl.w       vr0, a1, 0
+
+
+    vldrepl.w       vr1, a4, -4
+    vstelm.w        vr0, a0, 0, 0
+    vstelm.w        vr1, a3, -4, 0
+    jr              ra
+
+L(less_4bytes):
+    srli.d          t0, a2, 1
+    beqz            t0, L(less_2bytes)
+    vldrepl.h       vr0, a1, 0
+    vldrepl.h       vr1, a4, -2
+
+    vstelm.h        vr0, a0, 0, 0
+    vstelm.h        vr1, a3, -2, 0
+    jr              ra
+L(less_2bytes):
+    beqz            a2, L(less_1bytes)
+
+    ld.b            t0, a1, 0
+    st.b            t0, a0, 0
+L(less_1bytes):
+    jr              ra
+    nop
+END(MEMCPY_NAME)
+
+LEAF(MEMMOVE_NAME, 6)
+    li.d            t6, 16
+    add.d           a3, a0, a2
+    add.d           a4, a1, a2
+    bgeu            t6, a2, L(less_16bytes)
+
+    li.d            t8, 64
+    li.d            t7, 32
+    bltu            t8, a2, L(move_long)
+    bltu            t7, a2, L(more_32bytes)
+
+    vld             vr0, a1, 0
+    vld             vr1, a4, -16
+    vst             vr0, a0, 0
+    vst             vr1, a3, -16
+
+    jr              ra
+    nop
+L(move_long):
+    sub.d           t0, a0, a1
+    bltu            t0, a2, L(copy_back)
+
+
+L(copy_long):
+    vld             vr2, a1, 0
+    andi            t0, a0, 0xf
+    sub.d           t0, t6, t0
+    add.d           a1, a1, t0
+
+    sub.d           a2, a2, t0
+    andi            t1, a1, 0xf
+    bnez            t1, L(unaligned)
+    vld             vr0, a1, 0
+
+    addi.d          a2, a2, -16
+    vst             vr2, a0, 0
+    andi            t2, a2, 0x7f
+    add.d           a5, a0, t0
+
+    beq             a2, t2, L(al_less_128)
+    sub.d           t3, a2, t2
+    move            a2, t2
+    add.d           a6, a1, t3
+
+
+L(al_loop):
+    vld             vr1, a1, 16
+    vld             vr2, a1, 32
+    vld             vr3, a1, 48
+    vld             vr4, a1, 64
+
+    vld             vr5, a1, 80
+    vld             vr6, a1, 96
+    vld             vr7, a1, 112
+    vst             vr0, a5, 0
+
+    vld             vr0, a1, 128
+    addi.d          a1, a1, 128
+    vst             vr1, a5, 16
+    vst             vr2, a5, 32
+
+    vst             vr3, a5, 48
+    vst             vr4, a5, 64
+    vst             vr5, a5, 80
+    vst             vr6, a5, 96
+
+
+    vst             vr7, a5, 112
+    addi.d          a5, a5, 128
+    bne             a1, a6, L(al_loop)
+L(al_less_128):
+    blt             a2, t8, L(al_less_64)
+
+    vld             vr1, a1, 16
+    vld             vr2, a1, 32
+    vld             vr3, a1, 48
+    addi.d          a2, a2, -64
+
+    vst             vr0, a5, 0
+    vld             vr0, a1, 64
+    addi.d          a1, a1, 64
+    vst             vr1, a5, 16
+
+    vst             vr2, a5, 32
+    vst             vr3, a5, 48
+    addi.d          a5, a5, 64
+L(al_less_64):
+    blt             a2, t7, L(al_less_32)
+
+
+    vld             vr1, a1, 16
+    addi.d          a2, a2, -32
+    vst             vr0, a5, 0
+    vld             vr0, a1, 32
+
+    addi.d          a1, a1, 32
+    vst             vr1, a5, 16
+    addi.d          a5, a5, 32
+L(al_less_32):
+    blt             a2, t6, L(al_less_16)
+
+    vst             vr0, a5, 0
+    vld             vr0, a1, 16
+    addi.d          a5, a5, 16
+L(al_less_16):
+    vld             vr1, a4, -16
+
+    vst             vr0, a5, 0
+    vst             vr1, a3, -16
+    jr              ra
+    nop
+
+
+L(magic_num):
+    .dword          0x0706050403020100
+    .dword          0x0f0e0d0c0b0a0908
+L(unaligned):
+    pcaddi          t2, -4
+    bstrins.d       a1, zero, 3, 0
+    vld             vr8, t2, 0
+    vld             vr0, a1, 0
+
+    vld             vr1, a1, 16
+    addi.d          a2, a2, -16
+    vst             vr2, a0, 0
+    add.d           a5, a0, t0
+
+    vreplgr2vr.b    vr9, t1
+    andi            t2, a2, 0x7f
+    vadd.b          vr9, vr9, vr8
+    addi.d          a1, a1, 32
+
+
+    beq             t2, a2, L(un_less_128)
+    sub.d           t3, a2, t2
+    move            a2, t2
+    add.d           a6, a1, t3
+
+L(un_loop):
+    vld             vr2, a1, 0
+    vld             vr3, a1, 16
+    vld             vr4, a1, 32
+    vld             vr5, a1, 48
+
+    vld             vr6, a1, 64
+    vld             vr7, a1, 80
+    vshuf.b         vr8, vr1, vr0, vr9
+    vld             vr0, a1, 96
+
+    vst             vr8, a5, 0
+    vshuf.b         vr8, vr2, vr1, vr9
+    vld             vr1, a1, 112
+    vst             vr8, a5, 16
+
+
+    addi.d          a1, a1, 128
+    vshuf.b         vr2, vr3, vr2, vr9
+    vshuf.b         vr3, vr4, vr3, vr9
+    vst             vr2, a5, 32
+
+    vshuf.b         vr4, vr5, vr4, vr9
+    vst             vr3, a5, 48
+    vshuf.b         vr5, vr6, vr5, vr9
+    vst             vr4, a5, 64
+
+    vshuf.b         vr6, vr7, vr6, vr9
+    vst             vr5, a5, 80
+    vshuf.b         vr7, vr0, vr7, vr9
+    vst             vr6, a5, 96
+
+    vst             vr7, a5, 112
+    addi.d          a5, a5, 128
+    bne             a1, a6, L(un_loop)
+L(un_less_128):
+    blt             a2, t8, L(un_less_64)
+
+
+    vld             vr2, a1, 0
+    vld             vr3, a1, 16
+    vshuf.b         vr4, vr1, vr0, vr9
+    vld             vr0, a1, 32
+
+    vst             vr4, a5, 0
+    addi.d          a2, a2, -64
+    vshuf.b         vr4, vr2, vr1, vr9
+    vld             vr1, a1, 48
+
+    addi.d          a1, a1, 64
+    vst             vr4, a5, 16
+    vshuf.b         vr2, vr3, vr2, vr9
+    vshuf.b         vr3, vr0, vr3, vr9
+
+    vst             vr2, a5, 32
+    vst             vr3, a5, 48
+    addi.d          a5, a5, 64
+L(un_less_64):
+    blt             a2, t7, L(un_less_32)
+
+
+    vshuf.b         vr3, vr1, vr0, vr9
+    vld             vr0, a1, 0
+    vst             vr3, a5, 0
+    addi.d          a2, a2, -32
+
+    vshuf.b         vr3, vr0, vr1, vr9
+    vld             vr1, a1, 16
+    addi.d          a1, a1, 32
+    vst             vr3, a5, 16
+
+    addi.d          a5, a5, 32
+L(un_less_32):
+    blt             a2, t6, L(un_less_16)
+    vshuf.b         vr2, vr1, vr0, vr9
+    vor.v           vr0, vr1, vr1
+
+    vld             vr1, a1, 0
+    vst             vr2, a5, 0
+    addi.d          a5, a5, 16
+L(un_less_16):
+    vld             vr2, a4, -16
+
+
+    vshuf.b         vr0, vr1, vr0, vr9
+    vst             vr0, a5, 0
+    vst             vr2, a3, -16
+    jr              ra
+
+L(copy_back):
+    addi.d          t0, a3, -1
+    vld             vr2, a4, -16
+    andi            t0, t0, 0xf
+    addi.d          t0, t0, 1
+
+    sub.d           a4, a4, t0
+    sub.d           a2, a2, t0
+    andi            t1, a4, 0xf
+    bnez            t1, L(back_unaligned)
+
+    vld             vr0, a4, -16
+    addi.d          a2, a2, -16
+    vst             vr2, a3, -16
+    andi            t2, a2, 0x7f
+
+
+    sub.d           a3, a3, t0
+    beq             t2, a2, L(back_al_less_128)
+    sub.d           t3, a2, t2
+    move            a2, t2
+
+    sub.d           a6, a4, t3
+L(back_al_loop):
+    vld             vr1, a4, -32
+    vld             vr2, a4, -48
+    vld             vr3, a4, -64
+
+    vld             vr4, a4, -80
+    vld             vr5, a4, -96
+    vld             vr6, a4, -112
+    vld             vr7, a4, -128
+
+    vst             vr0, a3, -16
+    vld             vr0, a4, -144
+    addi.d          a4, a4, -128
+    vst             vr1, a3, -32
+
+
+    vst             vr2, a3, -48
+    vst             vr3, a3, -64
+    vst             vr4, a3, -80
+    vst             vr5, a3, -96
+
+    vst             vr6, a3, -112
+    vst             vr7, a3, -128
+    addi.d          a3, a3, -128
+    bne             a4, a6, L(back_al_loop)
+
+L(back_al_less_128):
+    blt             a2, t8, L(back_al_less_64)
+    vld             vr1, a4, -32
+    vld             vr2, a4, -48
+    vld             vr3, a4, -64
+
+    addi.d          a2, a2, -64
+    vst             vr0, a3, -16
+    vld             vr0, a4, -80
+    addi.d          a4, a4, -64
+
+
+    vst             vr1, a3, -32
+    vst             vr2, a3, -48
+    vst             vr3, a3, -64
+    addi.d          a3, a3, -64
+
+L(back_al_less_64):
+    blt             a2, t7, L(back_al_less_32)
+    vld             vr1, a4, -32
+    addi.d          a2, a2, -32
+    vst             vr0, a3, -16
+
+    vld             vr0, a4, -48
+    vst             vr1, a3, -32
+    addi.d          a3, a3, -32
+    addi.d          a4, a4, -32
+
+L(back_al_less_32):
+    blt             a2, t6, L(back_al_less_16)
+    vst             vr0, a3, -16
+    vld             vr0, a4, -32
+    addi.d          a3, a3, -16
+
+
+L(back_al_less_16):
+    vld             vr1, a1, 0
+    vst             vr0, a3, -16
+    vst             vr1, a0, 0
+    jr              ra
+
+L(magic_num_2):
+    .dword          0x0706050403020100
+    .dword          0x0f0e0d0c0b0a0908
+L(back_unaligned):
+    pcaddi          t2, -4
+    bstrins.d       a4, zero, 3, 0
+    vld             vr8, t2, 0
+    vld             vr0, a4, 0
+
+    vld             vr1, a4, -16
+    addi.d          a2, a2, -16
+    vst             vr2, a3, -16
+    sub.d           a3, a3, t0
+
+
+    vreplgr2vr.b    vr9, t1
+    andi            t2, a2, 0x7f
+    vadd.b          vr9, vr9, vr8
+    addi.d          a4, a4, -16
+
+    beq             t2, a2, L(back_un_less_128)
+    sub.d           t3, a2, t2
+    move            a2, t2
+    sub.d           a6, a4, t3
+
+L(back_un_loop):
+    vld             vr2, a4, -16
+    vld             vr3, a4, -32
+    vld             vr4, a4, -48
+
+    vld             vr5, a4, -64
+    vld             vr6, a4, -80
+    vld             vr7, a4, -96
+    vshuf.b         vr8, vr0, vr1, vr9
+
+
+    vld             vr0, a4, -112
+    vst             vr8, a3, -16
+    vshuf.b         vr8, vr1, vr2, vr9
+    vld             vr1, a4, -128
+
+    vst             vr8, a3, -32
+    addi.d          a4, a4, -128
+    vshuf.b         vr2, vr2, vr3, vr9
+    vshuf.b         vr3, vr3, vr4, vr9
+
+    vst             vr2, a3, -48
+    vshuf.b         vr4, vr4, vr5, vr9
+    vst             vr3, a3, -64
+    vshuf.b         vr5, vr5, vr6, vr9
+
+    vst             vr4, a3, -80
+    vshuf.b         vr6, vr6, vr7, vr9
+    vst             vr5, a3, -96
+    vshuf.b         vr7, vr7, vr0, vr9
+
+
+    vst             vr6, a3, -112
+    vst             vr7, a3, -128
+    addi.d          a3, a3, -128
+    bne             a4, a6, L(back_un_loop)
+
+L(back_un_less_128):
+    blt             a2, t8, L(back_un_less_64)
+    vld             vr2, a4, -16
+    vld             vr3, a4, -32
+    vshuf.b         vr4, vr0, vr1, vr9
+
+    vld             vr0, a4, -48
+    vst             vr4, a3, -16
+    addi.d          a2, a2, -64
+    vshuf.b         vr4, vr1, vr2, vr9
+
+    vld             vr1, a4, -64
+    addi.d          a4, a4, -64
+    vst             vr4, a3, -32
+    vshuf.b         vr2, vr2, vr3, vr9
+
+
+    vshuf.b         vr3, vr3, vr0, vr9
+    vst             vr2, a3, -48
+    vst             vr3, a3, -64
+    addi.d          a3, a3, -64
+
+L(back_un_less_64):
+    blt             a2, t7, L(back_un_less_32)
+    vshuf.b         vr3, vr0, vr1, vr9
+    vld             vr0, a4, -16
+    vst             vr3, a3, -16
+
+    addi.d          a2, a2, -32
+    vshuf.b         vr3, vr1, vr0, vr9
+    vld             vr1, a4, -32
+    addi.d          a4, a4, -32
+
+    vst             vr3, a3, -32
+    addi.d          a3, a3, -32
+L(back_un_less_32):
+    blt             a2, t6, L(back_un_less_16)
+    vshuf.b         vr2, vr0, vr1, vr9
+
+
+    vor.v           vr0, vr1, vr1
+    vld             vr1, a4, -16
+    vst             vr2, a3, -16
+    addi.d          a3, a3, -16
+
+L(back_un_less_16):
+    vld             vr2, a1, 0
+    vshuf.b         vr0, vr0, vr1, vr9
+    vst             vr0, a3, -16
+    vst             vr2, a0, 0
+
+    jr              ra
+END(MEMMOVE_NAME)
+
+libc_hidden_builtin_def (MEMCPY_NAME)
+libc_hidden_builtin_def (MEMMOVE_NAME)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S b/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S
new file mode 100644
index 00000000..90a64b6b
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S
@@ -0,0 +1,380 @@
+/* Optimized memmove_unaligned implementation using basic Loongarch instructions.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <sys/regdef.h>
+#include <sys/asm.h>
+
+#if IS_IN (libc)
+
+# define MEMMOVE_NAME __memmove_unaligned
+
+# define LD_64(reg, n)            \
+    ld.d        t0, reg, n;      \
+    ld.d        t1, reg, n + 8;  \
+    ld.d        t2, reg, n + 16; \
+    ld.d        t3, reg, n + 24; \
+    ld.d        t4, reg, n + 32; \
+    ld.d        t5, reg, n + 40; \
+    ld.d        t6, reg, n + 48; \
+    ld.d        t7, reg, n + 56;
+
+# define ST_64(reg, n)            \
+    st.d        t0, reg, n;      \
+    st.d        t1, reg, n + 8;  \
+    st.d        t2, reg, n + 16; \
+    st.d        t3, reg, n + 24; \
+    st.d        t4, reg, n + 32; \
+    st.d        t5, reg, n + 40; \
+    st.d        t6, reg, n + 48; \
+    st.d        t7, reg, n + 56;
+
+LEAF(MEMMOVE_NAME, 3)
+    add.d       a4, a1, a2
+    add.d       a3, a0, a2
+    beq         a1, a0, L(less_1bytes)
+    move        t8, a0
+
+    srai.d      a6, a2, 4
+    beqz        a6, L(less_16bytes)
+    srai.d      a6, a2, 6
+    bnez        a6, L(more_64bytes)
+    srai.d      a6, a2, 5
+    beqz        a6, L(less_32bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a1, 16
+    ld.d        t3, a1, 24
+
+    ld.d        t4, a4, -32
+    ld.d        t5, a4, -24
+    ld.d        t6, a4, -16
+    ld.d        t7, a4, -8
+
+    st.d        t0, a0, 0
+    st.d        t1, a0, 8
+    st.d        t2, a0, 16
+    st.d        t3, a0, 24
+
+    st.d        t4, a3, -32
+    st.d        t5, a3, -24
+    st.d        t6, a3, -16
+    st.d        t7, a3, -8
+
+    jr          ra
+
+L(less_32bytes):
+    ld.d        t0, a1, 0
+    ld.d        t1, a1, 8
+    ld.d        t2, a4, -16
+    ld.d        t3, a4, -8
+
+    st.d        t0, a0, 0
+    st.d        t1, a0, 8
+    st.d        t2, a3, -16
+    st.d        t3, a3, -8
+
+    jr          ra
+
+L(less_16bytes):
+    srai.d      a6, a2, 3
+    beqz        a6, L(less_8bytes)
+
+    ld.d        t0, a1, 0
+    ld.d        t1, a4, -8
+    st.d        t0, a0, 0
+    st.d        t1, a3, -8
+
+    jr          ra
+
+L(less_8bytes):
+    srai.d      a6, a2, 2
+    beqz        a6, L(less_4bytes)
+
+    ld.w        t0, a1, 0
+    ld.w        t1, a4, -4
+    st.w        t0, a0, 0
+    st.w        t1, a3, -4
+
+    jr          ra
+
+L(less_4bytes):
+    srai.d      a6, a2, 1
+    beqz        a6, L(less_2bytes)
+
+    ld.h        t0, a1, 0
+    ld.h        t1, a4, -2
+    st.h        t0, a0, 0
+    st.h        t1, a3, -2
+
+    jr          ra
+
+L(less_2bytes):
+    beqz        a2, L(less_1bytes)
+
+    ld.b        t0, a1, 0
+    st.b        t0, a0, 0
+
+    jr          ra
+
+L(less_1bytes):
+    jr          ra
+
+L(more_64bytes):
+    sub.d       a7, a0, a1
+    bltu        a7, a2, L(copy_backward)
+
+L(copy_forward):
+    srli.d      a0, a0, 3
+    slli.d      a0, a0, 3
+    beq         a0, t8, L(all_align)
+    addi.d      a0, a0, 0x8
+    sub.d       a7, t8, a0
+    sub.d       a1, a1, a7
+    add.d       a2, a7, a2
+
+L(start_unalign_proc):
+    pcaddi      t1, 18
+    slli.d      a6, a7, 3
+    add.d       t1, t1, a6
+    jr          t1
+
+    ld.b        t0, a1, -7
+    st.b        t0, a0, -7
+    ld.b        t0, a1, -6
+    st.b        t0, a0, -6
+    ld.b        t0, a1, -5
+    st.b        t0, a0, -5
+    ld.b        t0, a1, -4
+    st.b        t0, a0, -4
+    ld.b        t0, a1, -3
+    st.b        t0, a0, -3
+    ld.b        t0, a1, -2
+    st.b        t0, a0, -2
+    ld.b        t0, a1, -1
+    st.b        t0, a0, -1
+L(start_over):
+
+    addi.d      a2, a2, -0x80
+    blt         a2, zero, L(end_unalign_proc)
+
+L(loop_less):
+    LD_64(a1, 0)
+    ST_64(a0, 0)
+    LD_64(a1, 64)
+    ST_64(a0, 64)
+
+    addi.d      a0, a0, 0x80
+    addi.d      a1, a1, 0x80
+    addi.d      a2, a2, -0x80
+    bge         a2, zero, L(loop_less)
+
+L(end_unalign_proc):
+    addi.d      a2, a2, 0x80
+
+    pcaddi      t1, 36
+    andi        t2, a2, 0x78
+    add.d       a1, a1, t2
+    add.d       a0, a0, t2
+    sub.d       t1, t1, t2
+    jr          t1
+
+    ld.d        t0, a1, -120
+    st.d        t0, a0, -120
+    ld.d        t0, a1, -112
+    st.d        t0, a0, -112
+    ld.d        t0, a1, -104
+    st.d        t0, a0, -104
+    ld.d        t0, a1, -96
+    st.d        t0, a0, -96
+    ld.d        t0, a1, -88
+    st.d        t0, a0, -88
+    ld.d        t0, a1, -80
+    st.d        t0, a0, -80
+    ld.d        t0, a1, -72
+    st.d        t0, a0, -72
+    ld.d        t0, a1, -64
+    st.d        t0, a0, -64
+    ld.d        t0, a1, -56
+    st.d        t0, a0, -56
+    ld.d        t0, a1, -48
+    st.d        t0, a0, -48
+    ld.d        t0, a1, -40
+    st.d        t0, a0, -40
+    ld.d        t0, a1, -32
+    st.d        t0, a0, -32
+    ld.d        t0, a1, -24
+    st.d        t0, a0, -24
+    ld.d        t0, a1, -16
+    st.d        t0, a0, -16
+    ld.d        t0, a1, -8
+    st.d        t0, a0, -8
+
+    andi        a2, a2, 0x7
+    pcaddi      t1, 18
+    slli.d      a2, a2, 3
+    sub.d       t1, t1, a2
+    jr          t1
+
+    ld.b        t0, a4, -7
+    st.b        t0, a3, -7
+    ld.b        t0, a4, -6
+    st.b        t0, a3, -6
+    ld.b        t0, a4, -5
+    st.b        t0, a3, -5
+    ld.b        t0, a4, -4
+    st.b        t0, a3, -4
+    ld.b        t0, a4, -3
+    st.b        t0, a3, -3
+    ld.b        t0, a4, -2
+    st.b        t0, a3, -2
+    ld.b        t0, a4, -1
+    st.b        t0, a3, -1
+L(end):
+    move        a0, t8
+    jr          ra
+
+L(all_align):
+    addi.d      a1, a1, 0x8
+    addi.d      a0, a0, 0x8
+    ld.d        t0, a1, -8
+    st.d        t0, a0, -8
+    addi.d      a2, a2, -8
+    b           L(start_over)
+
+L(all_align_back):
+    addi.d      a4, a4, -0x8
+    addi.d      a3, a3, -0x8
+    ld.d        t0, a4, 0
+    st.d        t0, a3, 0
+    addi.d      a2, a2, -8
+    b           L(start_over_back)
+
+L(copy_backward):
+    move        a5, a3
+    srli.d      a3, a3, 3
+    slli.d      a3, a3, 3
+    beq         a3, a5, L(all_align_back)
+    sub.d       a7, a3, a5
+    add.d       a4, a4, a7
+    add.d       a2, a7, a2
+
+    pcaddi      t1, 18
+    slli.d      a6, a7, 3
+    add.d       t1, t1, a6
+    jr          t1
+
+    ld.b        t0, a4, 6
+    st.b        t0, a3, 6
+    ld.b        t0, a4, 5
+    st.b        t0, a3, 5
+    ld.b        t0, a4, 4
+    st.b        t0, a3, 4
+    ld.b        t0, a4, 3
+    st.b        t0, a3, 3
+    ld.b        t0, a4, 2
+    st.b        t0, a3, 2
+    ld.b        t0, a4, 1
+    st.b        t0, a3, 1
+    ld.b        t0, a4, 0
+    st.b        t0, a3, 0
+L(start_over_back):
+    addi.d      a2, a2, -0x80
+    blt         a2, zero, L(end_unalign_proc_back)
+
+L(loop_less_back):
+    LD_64(a4, -64)
+    ST_64(a3, -64)
+    LD_64(a4, -128)
+    ST_64(a3, -128)
+
+    addi.d      a4, a4, -0x80
+    addi.d      a3, a3, -0x80
+    addi.d      a2, a2, -0x80
+    bge         a2, zero, L(loop_less_back)
+
+L(end_unalign_proc_back):
+    addi.d      a2, a2, 0x80
+
+    pcaddi      t1, 36
+    andi        t2, a2, 0x78
+    sub.d       a4, a4, t2
+    sub.d       a3, a3, t2
+    sub.d       t1, t1, t2
+    jr          t1
+
+    ld.d        t0, a4, 112
+    st.d        t0, a3, 112
+    ld.d        t0, a4, 104
+    st.d        t0, a3, 104
+    ld.d        t0, a4, 96
+    st.d        t0, a3, 96
+    ld.d        t0, a4, 88
+    st.d        t0, a3, 88
+    ld.d        t0, a4, 80
+    st.d        t0, a3, 80
+    ld.d        t0, a4, 72
+    st.d        t0, a3, 72
+    ld.d        t0, a4, 64
+    st.d        t0, a3, 64
+    ld.d        t0, a4, 56
+    st.d        t0, a3, 56
+    ld.d        t0, a4, 48
+    st.d        t0, a3, 48
+    ld.d        t0, a4, 40
+    st.d        t0, a3, 40
+    ld.d        t0, a4, 32
+    st.d        t0, a3, 32
+    ld.d        t0, a4, 24
+    st.d        t0, a3, 24
+    ld.d        t0, a4, 16
+    st.d        t0, a3, 16
+    ld.d        t0, a4, 8
+    st.d        t0, a3, 8
+    ld.d        t0, a4, 0
+    st.d        t0, a3, 0
+
+    andi        a2, a2, 0x7
+    pcaddi      t1, 18
+    slli.d      a2, a2, 3
+    sub.d       t1, t1, a2
+    jr          t1
+
+    ld.b        t0, a1, 6
+    st.b        t0, a0, 6
+    ld.b        t0, a1, 5
+    st.b        t0, a0, 5
+    ld.b        t0, a1, 4
+    st.b        t0, a0, 4
+    ld.b        t0, a1, 3
+    st.b        t0, a0, 3
+    ld.b        t0, a1, 2
+    st.b        t0, a0, 2
+    ld.b        t0, a1, 1
+    st.b        t0, a0, 1
+    ld.b        t0, a1, 0
+    st.b        t0, a0, 0
+
+    move        a0, t8
+    jr          ra
+END(MEMMOVE_NAME)
+
+libc_hidden_builtin_def (MEMMOVE_NAME)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/memmove.c b/sysdeps/loongarch/lp64/multiarch/memmove.c
new file mode 100644
index 00000000..7e3ca4c4
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/memmove.c
@@ -0,0 +1,38 @@
+/* Multiple versions of memmove.
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2023 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+/* Define multiple versions only for the definition in libc.  */
+#if IS_IN (libc)
+# define memmove __redirect_memmove
+# include <string.h>
+# undef memmove
+
+# define SYMBOL_NAME memmove
+# include "ifunc-lasx.h"
+
+libc_ifunc_redirected (__redirect_memmove, __libc_memmove,
+		       IFUNC_SELECTOR ());
+strong_alias (__libc_memmove, memmove);
+
+# ifdef SHARED
+__hidden_ver1 (__libc_memmove, __GI_memmove, __redirect_memmove)
+  __attribute__ ((visibility ("hidden")));
+# endif
+
+#endif
-- 
2.33.0