[2/4] LoongArch: Add ifunc support for stpcpy{aligned, lsx, lasx}

Message ID 20230908093357.3119822-3-dengjianbo@loongson.cn
State New
Headers
Series LoongArch: Add ifunc support for str{cpy, rchr}, |

Checks

Context Check Description
redhat-pt-bot/TryBot-apply_patch success Patch applied to master at the time it was sent
linaro-tcwg-bot/tcwg_glibc_build--master-arm success Testing passed
linaro-tcwg-bot/tcwg_glibc_build--master-aarch64 success Testing passed
linaro-tcwg-bot/tcwg_glibc_check--master-arm success Testing passed
linaro-tcwg-bot/tcwg_glibc_check--master-aarch64 success Testing passed
redhat-pt-bot/TryBot-still_applies warning Patch no longer applies to master

Commit Message

dengjianbo Sept. 8, 2023, 9:33 a.m. UTC
  According to glibc stpcpy microbenchmark test results(changed to use
generic_stpcpy instead of strlen + memcpy), This implementation could
reduce the runtime as following:

Name                Percent of rutime reduced
stpcpy-lasx         10%-87%
stpcpy-lsx          10%-80%
stpcpy-aligned      5%-45%
---
 sysdeps/loongarch/lp64/multiarch/Makefile     |   3 +
 .../lp64/multiarch/ifunc-impl-list.c          |   8 +
 .../loongarch/lp64/multiarch/ifunc-stpcpy.h   |  40 ++++
 .../loongarch/lp64/multiarch/stpcpy-aligned.S | 191 ++++++++++++++++
 .../loongarch/lp64/multiarch/stpcpy-lasx.S    | 208 ++++++++++++++++++
 sysdeps/loongarch/lp64/multiarch/stpcpy-lsx.S | 206 +++++++++++++++++
 sysdeps/loongarch/lp64/multiarch/stpcpy.c     |  42 ++++
 7 files changed, 698 insertions(+)
 create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-stpcpy.h
 create mode 100644 sysdeps/loongarch/lp64/multiarch/stpcpy-aligned.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/stpcpy-lasx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/stpcpy-lsx.S
 create mode 100644 sysdeps/loongarch/lp64/multiarch/stpcpy.c
  

Patch

diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile
index f05685ceec..f95eb5c4fe 100644
--- a/sysdeps/loongarch/lp64/multiarch/Makefile
+++ b/sysdeps/loongarch/lp64/multiarch/Makefile
@@ -20,6 +20,9 @@  sysdep_routines += \
   strcpy-unaligned \
   strcpy-lsx \
   strcpy-lasx \
+  stpcpy-aligned \
+  stpcpy-lsx \
+  stpcpy-lasx \
   memcpy-aligned \
   memcpy-unaligned \
   memmove-unaligned \
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
index b556bacbd1..539aa681f9 100644
--- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
@@ -85,6 +85,14 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 	      IFUNC_IMPL_ADD (array, i, strcpy, 1, __strcpy_aligned)
 	      )
 
+  IFUNC_IMPL (i, name, stpcpy,
+#if !defined __loongarch_soft_float
+	      IFUNC_IMPL_ADD (array, i, stpcpy, SUPPORT_LASX, __stpcpy_lasx)
+	      IFUNC_IMPL_ADD (array, i, stpcpy, SUPPORT_LSX, __stpcpy_lsx)
+#endif
+	      IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_aligned)
+	      )
+
   IFUNC_IMPL (i, name, memcpy,
 #if !defined __loongarch_soft_float
               IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_LASX, __memcpy_lasx)
diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-stpcpy.h b/sysdeps/loongarch/lp64/multiarch/ifunc-stpcpy.h
new file mode 100644
index 0000000000..3827ec5a7e
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-stpcpy.h
@@ -0,0 +1,40 @@ 
+/* Common definition for stpcpy ifunc selections.
+   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;
+
+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
+    return OPTIMIZE (aligned);
+}
diff --git a/sysdeps/loongarch/lp64/multiarch/stpcpy-aligned.S b/sysdeps/loongarch/lp64/multiarch/stpcpy-aligned.S
new file mode 100644
index 0000000000..1520597b91
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/stpcpy-aligned.S
@@ -0,0 +1,191 @@ 
+/* Optimized stpcpy 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 STPCPY_NAME __stpcpy_aligned
+#else
+# define STPCPY_NAME __stpcpy
+#endif
+
+LEAF(STPCPY_NAME, 6)
+    andi        a3, a0, 0x7
+    beqz        a3, L(dest_align)
+    sub.d       a5, a1, a3
+    addi.d      a5, a5, 8
+
+L(make_dest_align):
+    ld.b        t0, a1, 0
+    addi.d      a1, a1, 1
+    st.b        t0, a0, 0
+    addi.d      a0, a0, 1
+
+    beqz        t0, L(al_out)
+    bne         a1, a5, L(make_dest_align)
+
+L(dest_align):
+    andi        a4, a1, 7
+    bstrins.d   a1, zero, 2, 0
+
+    lu12i.w     t5, 0x1010
+    ld.d        t0, a1, 0
+    ori         t5, t5, 0x101
+    bstrins.d   t5, t5, 63, 32
+
+    slli.d      t6, t5, 0x7
+    bnez        a4, L(unalign)
+    sub.d       t1, t0, t5
+    andn        t2, t6, t0
+
+    and         t3, t1, t2
+    bnez        t3, L(al_end)
+
+L(al_loop):
+    st.d        t0, a0, 0
+    ld.d        t0, a1, 8
+
+    addi.d      a1, a1, 8
+    addi.d      a0, a0, 8
+    sub.d       t1, t0, t5
+    andn        t2, t6, t0
+
+    and         t3, t1, t2
+    beqz        t3, L(al_loop)
+
+L(al_end):
+    ctz.d       t1, t3
+    srli.d      t1, t1, 3
+    addi.d      t1, t1, 1
+
+    andi        a3, t1, 8
+    andi        a4, t1, 4
+    andi        a5, t1, 2
+    andi        a6, t1, 1
+
+L(al_end_8):
+    beqz        a3, L(al_end_4)
+    st.d        t0, a0, 0
+    addi.d      a0, a0, 7
+    jr          ra
+L(al_end_4):
+    beqz        a4, L(al_end_2)
+    st.w        t0, a0, 0
+    addi.d      a0, a0, 4
+    srli.d      t0, t0, 32
+L(al_end_2):
+    beqz        a5, L(al_end_1)
+    st.h        t0, a0, 0
+    addi.d      a0, a0, 2
+    srli.d      t0, t0, 16
+L(al_end_1):
+    beqz        a6, L(al_out)
+    st.b        t0, a0, 0
+    addi.d      a0, a0, 1
+L(al_out):
+    addi.d      a0, a0, -1
+    jr          ra
+
+L(unalign):
+    slli.d      a5, a4, 3
+    li.d        t1, -1
+    sub.d       a6, zero, a5
+
+    srl.d       a7, t0, a5
+    sll.d       t7, t1, a6
+
+    or          t0, a7, t7
+    sub.d       t1, t0, t5
+    andn        t2, t6, t0
+    and         t3, t1, t2
+
+    bnez        t3, L(un_end)
+
+    ld.d        t4, a1, 8
+    addi.d      a1, a1, 8
+
+    sub.d       t1, t4, t5
+    andn        t2, t6, t4
+    sll.d       t0, t4, a6
+    and         t3, t1, t2
+
+    or          t0, t0, a7
+    bnez        t3, L(un_end_with_remaining)
+
+L(un_loop):
+    srl.d       a7, t4, a5
+
+    ld.d        t4, a1, 8
+    addi.d      a1, a1, 8
+
+    st.d        t0, a0, 0
+    addi.d      a0, a0, 8
+
+    sub.d       t1, t4, t5
+    andn        t2, t6, t4
+    sll.d       t0, t4, a6
+    and         t3, t1, t2
+
+    or          t0, t0, a7
+    beqz        t3, L(un_loop)
+
+L(un_end_with_remaining):
+    ctz.d       t1, t3
+    srli.d      t1, t1, 3
+    addi.d      t1, t1, 1
+    sub.d       t1, t1, a4
+
+    blt         t1, zero, L(un_end_less_8)
+    st.d        t0, a0, 0
+    addi.d      a0, a0, 8
+    beqz        t1, L(un_out)
+    srl.d       t0, t4, a5
+    b           L(un_end_less_8)
+
+L(un_end):
+    ctz.d       t1, t3
+    srli.d      t1, t1, 3
+    addi.d      t1, t1, 1
+
+L(un_end_less_8):
+    andi        a4, t1, 4
+    andi        a5, t1, 2
+    andi        a6, t1, 1
+L(un_end_4):
+    beqz        a4, L(un_end_2)
+    st.w        t0, a0, 0
+    addi.d      a0, a0, 4
+    srli.d      t0, t0, 32
+L(un_end_2):
+    beqz        a5, L(un_end_1)
+    st.h        t0, a0, 0
+    addi.d      a0, a0, 2
+    srli.d      t0, t0, 16
+L(un_end_1):
+    beqz        a6, L(un_out)
+    st.b        t0, a0, 0
+    addi.d      a0, a0, 1
+L(un_out):
+    addi.d      a0, a0, -1
+    jr          ra
+
+END(STPCPY_NAME)
+
+libc_hidden_builtin_def (STPCPY_NAME)
diff --git a/sysdeps/loongarch/lp64/multiarch/stpcpy-lasx.S b/sysdeps/loongarch/lp64/multiarch/stpcpy-lasx.S
new file mode 100644
index 0000000000..c21b132239
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/stpcpy-lasx.S
@@ -0,0 +1,208 @@ 
+/* Optimized stpcpy 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
+
+# define STPCPY __stpcpy_lasx
+
+LEAF(STPCPY, 6)
+    ori             t8, zero, 0xfe0
+    andi            t0, a1, 0xfff
+    li.d            t7, -1
+    move            a2, a0
+
+    bltu            t8, t0, L(page_cross_start)
+L(start_entry):
+    xvld            xr0, a1, 0
+    li.d            t0, 32
+    andi            t1, a2, 0x1f
+
+    xvsetanyeqz.b   fcc0, xr0
+    sub.d           t0, t0, t1
+    bcnez           fcc0, L(end)
+    add.d           a1, a1, t0
+
+    xvst            xr0, a2, 0
+    andi            a3, a1, 0x1f
+    add.d           a2, a2, t0
+    bnez            a3, L(unaligned)
+
+
+    xvld            xr0, a1, 0
+    xvsetanyeqz.b   fcc0, xr0
+    bcnez           fcc0, L(al_end)
+L(al_loop):
+    xvst            xr0, a2, 0
+
+    xvld            xr0, a1, 32
+    addi.d          a2, a2, 32
+    addi.d          a1, a1, 32
+    xvsetanyeqz.b   fcc0, xr0
+
+    bceqz           fcc0, L(al_loop)
+L(al_end):
+    xvmsknz.b       xr0, xr0
+    xvpickve.w      xr1, xr0, 4
+    vilvl.h         vr0, vr1, vr0
+
+    movfr2gr.s      t0, fa0
+    cto.w           t0, t0
+    add.d           a1, a1, t0
+    xvld            xr0, a1, -31
+
+
+    add.d           a0, a2, t0
+    xvst            xr0, a0, -31
+    jr              ra
+    nop
+
+L(page_cross_start):
+    move            a4, a1
+    bstrins.d       a4, zero, 4, 0
+    xvld            xr0, a4, 0
+    xvmsknz.b       xr0, xr0
+
+    xvpickve.w      xr1, xr0, 4
+    vilvl.h         vr0, vr1, vr0
+    movfr2gr.s      t0, fa0
+    sra.w           t0, t0, a1
+
+    beq             t0, t7, L(start_entry)
+    b               L(tail)
+L(unaligned):
+    andi            t0, a1, 0xfff
+    bltu            t8, t0, L(un_page_cross)
+
+
+L(un_start_entry):
+    xvld            xr0, a1, 0
+    xvsetanyeqz.b   fcc0, xr0
+    bcnez           fcc0, L(un_end)
+    addi.d          a1, a1, 32
+
+L(un_loop):
+    xvst            xr0, a2, 0
+    andi            t0, a1, 0xfff
+    addi.d          a2, a2, 32
+    bltu            t8, t0, L(page_cross_loop)
+
+L(un_loop_entry):
+    xvld            xr0, a1, 0
+    addi.d          a1, a1, 32
+    xvsetanyeqz.b   fcc0, xr0
+    bceqz           fcc0, L(un_loop)
+
+    addi.d          a1, a1, -32
+L(un_end):
+    xvmsknz.b       xr0, xr0
+    xvpickve.w      xr1, xr0, 4
+    vilvl.h         vr0, vr1, vr0
+
+
+    movfr2gr.s      t0, fa0
+L(un_tail):
+    cto.w           t0, t0
+    add.d           a1, a1, t0
+    xvld            xr0, a1, -31
+
+    add.d           a0, a2, t0
+    xvst            xr0, a0, -31
+    jr              ra
+L(un_page_cross):
+    sub.d           a4, a1, a3
+
+    xvld            xr0, a4, 0
+    xvmsknz.b       xr0, xr0
+    xvpickve.w      xr1, xr0, 4
+    vilvl.h         vr0, vr1, vr0
+
+    movfr2gr.s      t0, fa0
+    sra.w           t0, t0, a1
+    beq             t0, t7, L(un_start_entry)
+    b               L(un_tail)
+
+
+L(page_cross_loop):
+    sub.d           a4, a1, a3
+    xvld            xr0, a4, 0
+    xvmsknz.b       xr0, xr0
+    xvpickve.w      xr1, xr0, 4
+
+    vilvl.h         vr0, vr1, vr0
+    movfr2gr.s      t0, fa0
+    sra.w           t0, t0, a1
+    beq             t0, t7, L(un_loop_entry)
+
+    b               L(un_tail)
+L(end):
+    xvmsknz.b       xr0, xr0
+    xvpickve.w      xr1, xr0, 4
+    vilvl.h         vr0, vr1, vr0
+
+    movfr2gr.s      t0, fa0
+L(tail):
+    cto.w           t0, t0
+    add.d           a0, a2, t0
+    add.d           a5, a1, t0
+
+L(less_32):
+    srli.d          t1, t0, 4
+    beqz            t1, L(less_16)
+    vld             vr0, a1, 0
+    vld             vr1, a5, -15
+
+    vst             vr0, a2, 0
+    vst             vr1, a0, -15
+    jr              ra
+L(less_16):
+    srli.d          t1, t0, 3
+
+    beqz            t1, L(less_8)
+    ld.d            t2, a1, 0
+    ld.d            t3, a5, -7
+    st.d            t2, a2, 0
+
+    st.d            t3, a0, -7
+    jr              ra
+L(less_8):
+    li.d            t1, 3
+    bltu            t0, t1, L(less_4)
+
+    ld.w            t2, a1, 0
+    ld.w            t3, a5, -3
+    st.w            t2, a2, 0
+    st.w            t3, a0, -3
+
+    jr              ra
+L(less_4):
+    srli.d          t1, t0, 2
+    bgeu            t1, t0, L(zero_byte)
+    ld.h            t2, a1, 0
+
+    st.h            t2, a2, 0
+L(zero_byte):
+    st.b            zero, a0, 0
+    jr              ra
+END(STPCPY)
+
+libc_hidden_builtin_def (STPCPY)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/stpcpy-lsx.S b/sysdeps/loongarch/lp64/multiarch/stpcpy-lsx.S
new file mode 100644
index 0000000000..34ceadee66
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/stpcpy-lsx.S
@@ -0,0 +1,206 @@ 
+/* Optimized stpcpy 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 STPCPY __stpcpy_lsx
+
+LEAF(STPCPY, 6)
+    pcalau12i       t0, %pc_hi20(L(INDEX))
+    andi            a4, a1, 0xf
+    vld             vr1, t0, %pc_lo12(L(INDEX))
+    beqz            a4, L(load_start)
+
+    xor             t0, a1, a4
+    vld             vr0, t0, 0
+    vreplgr2vr.b    vr2, a4
+    vadd.b          vr2, vr2, vr1
+
+    vshuf.b         vr0, vr2, vr0, vr2
+    vsetanyeqz.b    fcc0, vr0
+    bcnez           fcc0, L(end)
+L(load_start):
+    vld             vr0, a1, 0
+
+    li.d            t1, 16
+    andi            a3, a0, 0xf
+    vsetanyeqz.b    fcc0, vr0
+    sub.d           t0, t1, a3
+
+
+    bcnez           fcc0, L(end)
+    add.d           a1, a1, t0
+    vst             vr0, a0, 0
+    add.d           a0, a0, t0
+
+    bne             a3, a4, L(unaligned)
+    vld             vr0, a1, 0
+    vsetanyeqz.b    fcc0, vr0
+    bcnez           fcc0, L(al_end)
+
+L(al_loop):
+    vst             vr0, a0, 0
+    vld             vr0, a1, 16
+    addi.d          a0, a0, 16
+    addi.d          a1, a1, 16
+
+    vsetanyeqz.b    fcc0, vr0
+    bceqz           fcc0, L(al_loop)
+L(al_end):
+    vmsknz.b        vr1, vr0
+    movfr2gr.s      t0, fa1
+
+
+    cto.w           t0, t0
+    add.d           a1, a1, t0
+    vld             vr0, a1, -15
+    add.d           a0, a0, t0
+
+    vst             vr0, a0, -15
+    jr              ra
+    nop
+    nop
+
+L(end):
+    vseqi.b         vr1, vr0, 0
+    vfrstpi.b       vr1, vr1, 0
+    vpickve2gr.bu   t0, vr1, 0
+    addi.d          t0, t0, 1
+
+L(end_16):
+    andi            t1, t0, 16
+    beqz            t1, L(end_8)
+    vst             vr0, a0, 0
+    addi.d          a0, a0, 15
+
+
+    jr              ra
+L(end_8):
+    andi            t2, t0, 8
+    andi            t3, t0, 4
+    andi            t4, t0, 2
+
+    andi            t5, t0, 1
+    beqz            t2, L(end_4)
+    vstelm.d        vr0, a0, 0, 0
+    addi.d          a0, a0, 8
+
+    vbsrl.v         vr0, vr0, 8
+L(end_4):
+    beqz            t3, L(end_2)
+    vstelm.w        vr0, a0, 0, 0
+    addi.d          a0, a0, 4
+
+    vbsrl.v         vr0, vr0, 4
+L(end_2):
+    beqz            t4, L(end_1)
+    vstelm.h        vr0, a0, 0, 0
+    addi.d          a0, a0, 2
+
+
+    vbsrl.v         vr0, vr0, 2
+L(end_1):
+    beqz            t5, L(out)
+    vstelm.b        vr0, a0, 0, 0
+    addi.d          a0, a0, 1
+
+L(out):
+    addi.d          a0, a0, -1
+    jr              ra
+    nop
+    nop
+
+L(unaligned):
+    andi            a3, a1, 0xf
+    bstrins.d       a1, zero, 3, 0
+    vld             vr2, a1, 0
+    vreplgr2vr.b    vr3, a3
+
+    vslt.b          vr4, vr1, vr3
+    vor.v           vr0, vr2, vr4
+    vsetanyeqz.b    fcc0, vr0
+    bcnez           fcc0, L(un_first_end)
+
+
+    vld             vr0, a1, 16
+    vadd.b          vr3, vr3, vr1
+    vshuf.b         vr4, vr0, vr2, vr3
+    vsetanyeqz.b    fcc0, vr0
+
+    bcnez           fcc0, L(un_end)
+    vor.v           vr2, vr0, vr0
+    addi.d          a1, a1, 16
+L(un_loop):
+    vld             vr0, a1, 16
+
+    vst             vr4, a0, 0
+    addi.d          a0, a0, 16
+    vshuf.b         vr4, vr0, vr2, vr3
+    vsetanyeqz.b    fcc0, vr0
+
+    bcnez           fcc0, L(un_end)
+    vld             vr2, a1, 32
+    vst             vr4, a0, 0
+    addi.d          a1, a1, 32
+
+
+    addi.d          a0, a0, 16
+    vshuf.b         vr4, vr2, vr0, vr3
+    vsetanyeqz.b    fcc0, vr2
+    bceqz           fcc0, L(un_loop)
+
+    vor.v           vr0, vr2, vr2
+    addi.d          a1, a1, -16
+L(un_end):
+    vsetanyeqz.b    fcc0, vr4
+    bcnez           fcc0, 1f
+
+    vst             vr4, a0, 0
+1:
+    vmsknz.b        vr1, vr0
+    movfr2gr.s      t0, fa1
+    cto.w           t0, t0
+
+    add.d           a1, a1, t0
+    vld             vr0, a1, 1
+    add.d           a0, a0, t0
+    sub.d           a0, a0, a3
+
+
+    vst             vr0, a0, 1
+    addi.d          a0, a0, 16
+    jr              ra
+L(un_first_end):
+    addi.d          a0, a0, -16
+
+    addi.d          a1, a1, -16
+    b               1b
+END(STPCPY)
+
+    .section        .rodata.cst16,"M",@progbits,16
+    .align          4
+L(INDEX):
+    .dword          0x0706050403020100
+    .dword          0x0f0e0d0c0b0a0908
+
+libc_hidden_builtin_def (STPCPY)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/stpcpy.c b/sysdeps/loongarch/lp64/multiarch/stpcpy.c
new file mode 100644
index 0000000000..62115e4055
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/stpcpy.c
@@ -0,0 +1,42 @@ 
+/* Multiple versions of stpcpy.
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2017-2023 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 stpcpy __redirect_stpcpy
+# define __stpcpy __redirect___stpcpy
+# define NO_MEMPCPY_STPCPY_REDIRECT
+# define __NO_STRING_INLINES
+# include <string.h>
+# undef stpcpy
+# undef __stpcpy
+
+# define SYMBOL_NAME stpcpy
+# include "ifunc-stpcpy.h"
+
+libc_ifunc_redirected (__redirect_stpcpy, __stpcpy, IFUNC_SELECTOR ());
+
+weak_alias (__stpcpy, stpcpy)
+# ifdef SHARED
+__hidden_ver1 (__stpcpy, __GI___stpcpy, __redirect___stpcpy)
+  __attribute__ ((visibility ("hidden"))) __attribute_copy__ (stpcpy);
+__hidden_ver1 (stpcpy, __GI_stpcpy, __redirect_stpcpy)
+  __attribute__ ((visibility ("hidden"))) __attribute_copy__ (stpcpy);
+# endif
+#endif