LoongArch: Add ifunc support for strncmp{aligned, lsx}

Based on the glibc microbenchmark, only a few short inputs with this
strncmp-aligned and strncmp-lsx implementation experience performance
degradation, overall, strncmp-aligned could reduce the runtime 0%-10%
for aligned comparision, 10%-25% for unaligend comparision, strncmp-lsx
could reduce the runtime about 0%-60%.

6 files changed, 508 insertions, 0 deletions

diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile
index d5a500d..5d7ae7a 100644
--- a/sysdeps/loongarch/lp64/multiarch/Makefile
+++ b/sysdeps/loongarch/lp64/multiarch/Makefile
@@ -14,6 +14,8 @@ sysdep_routines += \
   strchrnul-lasx \
   strcmp-aligned \
   strcmp-lsx \
+  strncmp-aligned \
+  strncmp-lsx \
   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 9183b7d..c8ba87b 100644
--- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c
@@ -69,6 +69,13 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 	      IFUNC_IMPL_ADD (array, i, strcmp, 1, __strcmp_aligned)
 	      )
 
+  IFUNC_IMPL (i, name, strncmp,
+#if !defined __loongarch_soft_float
+	      IFUNC_IMPL_ADD (array, i, strncmp, SUPPORT_LSX, __strncmp_lsx)
+#endif
+	      IFUNC_IMPL_ADD (array, i, strncmp, 1, __strncmp_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-strncmp.h b/sysdeps/loongarch/lp64/multiarch/ifunc-strncmp.h
new file mode 100644
index 0000000..1a7dc36
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/ifunc-strncmp.h
@@ -0,0 +1,38 @@
+/* Common definition for strncmp ifunc selection.
+   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 (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_LSX)
+    return OPTIMIZE (lsx);
+  else
+#endif
+    return OPTIMIZE (aligned);
+}
diff --git a/sysdeps/loongarch/lp64/multiarch/strncmp-aligned.S b/sysdeps/loongarch/lp64/multiarch/strncmp-aligned.S
new file mode 100644
index 0000000..e2687fa
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/strncmp-aligned.S
@@ -0,0 +1,218 @@
+/* Optimized strncmp 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 STRNCMP __strncmp_aligned
+#else
+# define STRNCMP strncmp
+#endif
+
+LEAF(STRNCMP, 6)
+    beqz        a2, L(ret0)
+    lu12i.w     a5, 0x01010
+    andi        a3, a0, 0x7
+    ori         a5, a5, 0x101
+
+    andi        a4, a1, 0x7
+    bstrins.d   a5, a5, 63, 32
+    li.d        t7, -1
+    li.d        t8, 8
+
+    addi.d      a2, a2, -1
+    slli.d      a6, a5, 7
+    bne         a3, a4, L(unaligned)
+    bstrins.d   a0, zero, 2, 0
+
+    bstrins.d   a1, zero, 2, 0
+    ld.d        t0, a0, 0
+    ld.d        t1, a1, 0
+    slli.d      t2, a3, 3
+
+
+    sub.d       t5, t8, a3
+    srl.d       t3, t7, t2
+    srl.d       t0, t0, t2
+    srl.d       t1, t1, t2
+
+    orn         t0, t0, t3
+    orn         t1, t1, t3
+    sub.d       t2, t0, a5
+    andn        t3, a6, t0
+
+    and         t2, t2, t3
+    bne         t0, t1, L(al_end)
+    sltu        t4, a2, t5
+    sub.d       a2, a2, t5
+
+L(al_loop):
+    or          t4, t2, t4
+    bnez        t4, L(ret0)
+    ldx.d       t0, a0, t8
+    ldx.d       t1, a1, t8
+
+
+    addi.d      t8, t8, 8
+    sltui       t4, a2, 8
+    addi.d      a2, a2, -8
+    sub.d       t2, t0, a5
+
+    andn        t3, a6, t0
+    and         t2, t2, t3
+    beq         t0, t1, L(al_loop)
+    addi.d      a2, a2, 8
+
+L(al_end):
+    xor         t3, t0, t1
+    or          t2, t2, t3
+    ctz.d       t2, t2
+    srli.d      t4, t2, 3
+
+    bstrins.d   t2, zero, 2, 0
+    srl.d       t0, t0, t2
+    srl.d       t1, t1, t2
+    andi        t0, t0, 0xff
+
+
+    andi        t1, t1, 0xff
+    sltu        t2, a2, t4
+    sub.d       a0, t0, t1
+    masknez     a0, a0, t2
+
+    jr          ra
+L(ret0):
+    move        a0, zero
+    jr          ra
+    nop
+
+L(unaligned):
+    slt         a7, a4, a3
+    xor         t0, a0, a1
+    maskeqz     t0, t0, a7
+    xor         a0, a0, t0
+
+    xor         a1, a1, t0
+    andi        a3, a0, 0x7
+    andi        a4, a1, 0x7
+    bstrins.d   a0, zero, 2, 0
+
+
+    bstrins.d   a1, zero, 2, 0
+    ld.d        t4, a0, 0
+    ld.d        t1, a1, 0
+    slli.d      t2, a3, 3
+
+    slli.d      t3, a4, 3
+    srl.d       t5, t7, t3
+    srl.d       t0, t4, t2
+    srl.d       t1, t1, t3
+
+    orn         t0, t0, t5
+    orn         t1, t1, t5
+    bne         t0, t1, L(not_equal)
+    sub.d       t6, t8, a4
+
+    sub.d       a4, t2, t3
+    sll.d       t2, t7, t2
+    sub.d       t5, t8, a3
+    orn         t4, t4, t2
+
+
+    sub.d       t2, t4, a5
+    andn        t3, a6, t4
+    sltu        t7, a2, t5
+    and         t2, t2, t3
+
+    sub.d       a3, zero, a4
+    or          t2, t2, t7
+    bnez        t2, L(un_end)
+    sub.d       t7, t5, t6
+
+    sub.d       a2, a2, t5
+    sub.d       t6, t8, t7
+L(un_loop):
+    srl.d       t5, t4, a4
+    ldx.d       t4, a0, t8
+
+    ldx.d       t1, a1, t8
+    addi.d      t8, t8, 8
+    sll.d       t0, t4, a3
+    or          t0, t0, t5
+
+
+    bne         t0, t1, L(loop_not_equal)
+    sub.d       t2, t4, a5
+    andn        t3, a6, t4
+    sltui       t5, a2, 8
+
+    and         t2, t2, t3
+    addi.d      a2, a2, -8
+    or          t3, t2, t5
+    beqz        t3, L(un_loop)
+
+    addi.d      a2, a2, 8
+L(un_end):
+    sub.d       t2, t0, a5
+    andn        t3, a6, t0
+    sltu        t5, a2, t6
+
+    and         t2, t2, t3
+    or          t2, t2, t5
+    bnez        t2, L(ret0)
+    ldx.d       t1, a1, t8
+
+
+    srl.d       t0, t4, a4
+    sub.d       a2, a2, t6
+L(not_equal):
+    sub.d       t2, t0, a5
+    andn        t3, a6, t0
+
+    xor         t4, t0, t1
+    and         t2, t2, t3
+    or          t2, t2, t4
+    ctz.d       t2, t2
+
+    bstrins.d   t2, zero, 2, 0
+    srli.d      t4, t2, 3
+    srl.d       t0, t0, t2
+    srl.d       t1, t1, t2
+
+    andi        t0, t0, 0xff
+    andi        t1, t1, 0xff
+    sub.d       t2, t0, t1
+    sub.d       t3, t1, t0
+
+
+    masknez     t0, t2, a7
+    maskeqz     t1, t3, a7
+    sltu        t2, a2, t4
+    or          a0, t0, t1
+
+    masknez     a0, a0, t2
+    jr          ra
+L(loop_not_equal):
+    add.d       a2, a2, t7
+    b           L(not_equal)
+END(STRNCMP)
+
+libc_hidden_builtin_def (STRNCMP)
diff --git a/sysdeps/loongarch/lp64/multiarch/strncmp-lsx.S b/sysdeps/loongarch/lp64/multiarch/strncmp-lsx.S
new file mode 100644
index 0000000..0b4eee2
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/strncmp-lsx.S
@@ -0,0 +1,208 @@
+/* Optimized strncmp 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 STRNCMP __strncmp_lsx
+
+LEAF(STRNCMP, 6)
+    beqz            a2, L(ret0)
+    pcalau12i       t0, %pc_hi20(L(INDEX))
+    andi            a3, a0, 0xf
+    vld             vr2, t0, %pc_lo12(L(INDEX))
+
+    andi            a4, a1, 0xf
+    li.d            t2, 16
+    bne             a3, a4, L(unaligned)
+    xor             t0, a0, a3
+
+    xor             t1, a1, a4
+    vld             vr0, t0, 0
+    vld             vr1, t1, 0
+    vreplgr2vr.b    vr3, a3
+
+
+    sub.d           t2, t2, a3
+    vadd.b          vr3, vr3, vr2
+    vshuf.b         vr0, vr3, vr0, vr3
+    vshuf.b         vr1, vr3, vr1, vr3
+
+    vseq.b          vr3, vr0, vr1
+    vmin.bu         vr3, vr0, vr3
+    bgeu            t2, a2, L(al_early_end)
+    vsetanyeqz.b    fcc0, vr3
+
+    bcnez           fcc0, L(al_end)
+    add.d           a3, a0, a2
+    addi.d          a4, a3, -1
+    bstrins.d       a4, zero, 3, 0
+
+    sub.d           a2, a3, a4
+L(al_loop):
+    vld             vr0, t0, 16
+    vld             vr1, t1, 16
+    addi.d          t0, t0, 16
+
+
+    addi.d          t1, t1, 16
+    vseq.b          vr3, vr0, vr1
+    vmin.bu         vr3, vr0, vr3
+    beq             t0, a4, L(al_early_end)
+
+    vsetanyeqz.b    fcc0, vr3
+    bceqz           fcc0, L(al_loop)
+L(al_end):
+    vseqi.b         vr3, vr3, 0
+    vfrstpi.b       vr3, vr3, 0
+
+    vshuf.b         vr0, vr0, vr0, vr3
+    vshuf.b         vr1, vr1, vr1, vr3
+    vpickve2gr.bu   t0, vr0, 0
+    vpickve2gr.bu   t1, vr1, 0
+
+    sub.d           a0, t0, t1
+    jr              ra
+L(al_early_end):
+    vreplgr2vr.b    vr4, a2
+    vslt.b          vr4, vr2, vr4
+
+
+    vorn.v          vr3, vr3, vr4
+    b               L(al_end)
+L(unaligned):
+    slt             a5, a3, a4
+    xor             t0, a0, a1
+
+    maskeqz         t0, t0, a5
+    xor             a0, a0, t0
+    xor             a1, a1, t0
+    andi            a3, a0, 0xf
+
+    andi            a4, a1, 0xf
+    xor             t0, a0, a3
+    xor             t1, a1, a4
+    vld             vr0, t0, 0
+
+    vld             vr3, t1, 0
+    sub.d           t2, t2, a3
+    vreplgr2vr.b    vr4, a3
+    vreplgr2vr.b    vr5, a4
+
+
+    vaddi.bu        vr6, vr2, 16
+    vsub.b          vr7, vr4, vr5
+    vsub.b          vr6, vr6, vr7
+    vadd.b          vr4, vr2, vr4
+
+    vshuf.b         vr1, vr3, vr3, vr6
+    vshuf.b         vr0, vr7, vr0, vr4
+    vshuf.b         vr1, vr7, vr1, vr4
+    vseq.b          vr4, vr0, vr1
+
+    vmin.bu         vr4, vr0, vr4
+    bgeu            t2, a2, L(un_early_end)
+    vsetanyeqz.b    fcc0, vr4
+    bcnez           fcc0, L(un_end)
+
+    add.d           a6, a0, a2
+    vslt.b          vr5, vr2, vr5
+    addi.d          a7, a6, -1
+    vor.v           vr3, vr3, vr5
+
+
+    bstrins.d       a7, zero, 3, 0
+    sub.d           a2, a6, a7
+L(un_loop):
+    vld             vr0, t0, 16
+    addi.d          t0, t0, 16
+
+    vsetanyeqz.b    fcc0, vr3
+    bcnez           fcc0, L(has_zero)
+    beq             t0, a7, L(end_with_len)
+    vor.v           vr1, vr3, vr3
+
+    vld             vr3, t1, 16
+    addi.d          t1, t1, 16
+    vshuf.b         vr1, vr3, vr1, vr6
+    vseq.b          vr4, vr0, vr1
+
+    vmin.bu         vr4, vr0, vr4
+    vsetanyeqz.b    fcc0, vr4
+    bceqz           fcc0, L(un_loop)
+L(un_end):
+    vseqi.b         vr4, vr4, 0
+
+
+    vfrstpi.b       vr4, vr4, 0
+    vshuf.b         vr0, vr0, vr0, vr4
+    vshuf.b         vr1, vr1, vr1, vr4
+    vpickve2gr.bu   t0, vr0, 0
+
+    vpickve2gr.bu   t1, vr1, 0
+    sub.d           t2, t0, t1
+    sub.d           t3, t1, t0
+    masknez         t0, t2, a5
+
+    maskeqz         t1, t3, a5
+    or              a0, t0, t1
+    jr              ra
+L(has_zero):
+    vshuf.b         vr1, vr3, vr3, vr6
+
+    vseq.b          vr4, vr0, vr1
+    vmin.bu         vr4, vr0, vr4
+    bne             t0, a7, L(un_end)
+L(un_early_end):
+    vreplgr2vr.b    vr5, a2
+
+    vslt.b          vr5, vr2, vr5
+    vorn.v          vr4, vr4, vr5
+    b               L(un_end)
+L(end_with_len):
+    sub.d           a6, a3, a4
+
+    bgeu            a6, a2, 1f
+    vld             vr4, t1, 16
+1:
+    vshuf.b         vr1, vr4, vr3, vr6
+    vseq.b          vr4, vr0, vr1
+
+    vmin.bu         vr4, vr0, vr4
+    vreplgr2vr.b    vr5, a2
+    vslt.b          vr5, vr2, vr5
+    vorn.v          vr4, vr4, vr5
+
+    b               L(un_end)
+L(ret0):
+    move            a0, zero
+    jr              ra
+END(STRNCMP)
+
+    .section         .rodata.cst16,"M",@progbits,16
+    .align           4
+L(INDEX):
+    .dword           0x0706050403020100
+    .dword           0x0f0e0d0c0b0a0908
+
+libc_hidden_builtin_def (STRNCMP)
+#endif
diff --git a/sysdeps/loongarch/lp64/multiarch/strncmp.c b/sysdeps/loongarch/lp64/multiarch/strncmp.c
new file mode 100644
index 0000000..af6d0bc
--- /dev/null
+++ b/sysdeps/loongarch/lp64/multiarch/strncmp.c
@@ -0,0 +1,35 @@
+/* Multiple versions of strncmp.
+   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
+   <http://www.gnu.org/licenses/>.  */
+
+/* Define multiple versions only for the definition in libc.  */
+#if IS_IN (libc)
+# define strncmp __redirect_strncmp
+# include <string.h>
+# undef strncmp
+
+# define SYMBOL_NAME strncmp
+# include "ifunc-strncmp.h"
+
+libc_ifunc_redirected (__redirect_strncmp, strncmp, IFUNC_SELECTOR ());
+
+# ifdef SHARED
+__hidden_ver1 (strncmp, __GI_strncmp, __redirect_strncmp)
+  __attribute__ ((visibility ("hidden"))) __attribute_copy__ (strncmp);
+# endif
+#endif