glibc/6_6-LoongArch-Optimize-string-functions-strlen-strnlen.patch
Xue Liu 4adfab66ba LoongArch: Optimize string functions including memcpy, memmove,
memset, strchr, strchrnul, strcmp, strncmp, ctrcpy, ctrlen, strnlen.

(cherry picked from commit 4ba365320a633ecd4cb47d8f171aa81fcd1dd6ef)
2023-01-29 14:28:25 +08:00

256 lines
6.5 KiB
Diff

From 86290dbec63a9688ab0e0085ab8ab686fa256f18 Mon Sep 17 00:00:00 2001
From: Xue Liu <liuxue@loongson.cn>
Date: Sun, 29 Jan 2023 10:25:47 +0800
Subject: [PATCH 6/6] LoongArch: Optimize string functions strlen, strnlen.
Change-Id: I5df3398f9dbd9ea72c3de14e1e5f7793f6dbd794
---
sysdeps/loongarch/lp64/strlen.S | 102 +++++++++++++++++++++++++
sysdeps/loongarch/lp64/strnlen.S | 125 +++++++++++++++++++++++++++++++
2 files changed, 227 insertions(+)
create mode 100644 sysdeps/loongarch/lp64/strlen.S
create mode 100644 sysdeps/loongarch/lp64/strnlen.S
diff --git a/sysdeps/loongarch/lp64/strlen.S b/sysdeps/loongarch/lp64/strlen.S
new file mode 100644
index 00000000..3569598c
--- /dev/null
+++ b/sysdeps/loongarch/lp64/strlen.S
@@ -0,0 +1,102 @@
+/* Optimized strlen implementation for LoongArch.
+ Copyright (C) 2021 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/asm.h>
+
+/* size_t strlen (const char *s1); */
+LEAF(strlen)
+ .align 5
+
+ nor t4, zero, zero
+ lu12i.w a2, 0x01010
+ andi t5, a0, 0x7
+
+ li.w t7, 0x7
+ slli.d t6, t5, 0x3
+ andn t7, a0, t7
+ ld.d a1, t7, 0
+ sub.d t7, zero, t6
+ sll.d t4, t4, t7
+ maskeqz t4, t4, t6
+ srl.d a1, a1, t6
+ or a1, a1, t4
+
+
+ ori a2, a2, 0x101
+ nor t1, a1, zero
+ li.w a4, 8
+
+ bstrins.d a2, a2, 63, 32
+ sub.d a5, a4, t5
+ move t5, a0
+
+ sub.d t0, a1, a2
+ slli.d t4, a2, 7
+ nor a3, zero, t4
+ nor t1, a1, a3
+
+ and t0, t0, t1
+ bnez t0, strlen_count1
+ add.d a0, a0, a5
+strlen_loop:
+ ld.d a1, a0, 0
+ sub.d t0, a1, a2
+ and t1, t0, t4
+ bnez t1, strlen_count_pre
+ ld.d a1, a0, 8
+ sub.d t0, a1, a2
+ and t1, t0, t4
+ addi.d a0, a0, 16
+ beqz t1, strlen_loop
+strlen_count:
+ addi.d a0, a0, -8
+strlen_count_pre:
+ nor t1, a1, a3
+ and t0, t0, t1
+ beqz t0, strlen_noascii_start
+strlen_count1:
+ ctz.d t1, t0
+ sub.d v0, a0, t5
+ srli.w t1, t1, 3
+ add.d v0, v0, t1
+ jr ra
+strlen_noascii_start:
+ addi.d a0, a0, 8
+strlen_loop_noascii:
+ ld.d a1, a0, 0
+ sub.d t0, a1, a2
+ nor t1, a1, a3
+ and t0, t0, t1
+ bnez t0, strlen_count1
+ ld.d a1, a0, 8
+ sub.d t0, a1, a2
+ nor t1, a1, a3
+ and t0, t0, t1
+ addi.d a0, a0, 16
+ beqz t0, strlen_loop_noascii
+ addi.d a0, a0, -8
+ ctz.d t1, t0
+ sub.d v0, a0, t5
+ srli.w t1, t1, 3
+ add.d v0, v0, t1
+ jr ra
+END(strlen)
+
+libc_hidden_builtin_def (strlen)
+
diff --git a/sysdeps/loongarch/lp64/strnlen.S b/sysdeps/loongarch/lp64/strnlen.S
new file mode 100644
index 00000000..8eaa60e2
--- /dev/null
+++ b/sysdeps/loongarch/lp64/strnlen.S
@@ -0,0 +1,125 @@
+/* Optimized strlen implementation for LoongArch.
+ Copyright (C) 2021 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 <sys/asm.h>
+
+/* rd <- if rc then ra else rb
+ a5 will be destroyed. */
+#define CONDITIONSEL(rd, ra, rb, rc)\
+ masknez a5, rb, rc;\
+ maskeqz rd, ra, rc;\
+ or rd, rd, a5
+
+/* Parameters and Results */
+#define srcin a0
+#define limit a1
+#define len v0
+
+/* Internal variable */
+#define data1 t0
+#define data2 t1
+#define has_nul1 t2
+#define has_nul2 t3
+#define src t4
+#define zeroones t5
+#define sevenf t6
+#define data2a t7
+#define tmp6 t7
+#define pos t8
+#define tmp1 a2
+#define tmp2 a3
+#define tmp3 a4
+#define tmp4 a5
+#define tmp5 a6
+#define limit_wd a7
+
+/* size_t strnlen (const char *s1,size_t maxlen); */
+LEAF(__strnlen)
+ .align 4
+ beqz limit, _hit_limit
+ lu12i.w zeroones, 0x01010
+ lu12i.w sevenf, 0x7f7f7
+ ori zeroones, zeroones, 0x101
+ ori sevenf, sevenf, 0xf7f
+ bstrins.d zeroones, zeroones, 63, 32
+ bstrins.d sevenf, sevenf, 63, 32
+ andi tmp1, srcin, 15
+ sub.d src, srcin, tmp1
+ bnez tmp1, misaligned
+ addi.d limit_wd, limit, -1
+ srli.d limit_wd, limit_wd, 4
+_loop:
+ ld.d data1, src, 0
+ ld.d data2, src, 8
+ addi.d src, src, 16
+_realigned:
+ sub.d tmp1, data1, zeroones
+ or tmp2, data1, sevenf
+ sub.d tmp3, data2, zeroones
+ or tmp4, data2, sevenf
+ andn has_nul1, tmp1, tmp2
+ andn has_nul2, tmp3, tmp4
+ addi.d limit_wd, limit_wd, -1
+ srli.d tmp1, limit_wd, 63
+ or tmp2, has_nul1, has_nul2
+ or tmp3, tmp1, tmp2
+ beqz tmp3, _loop
+ beqz tmp2, _hit_limit
+ sub.d len, src, srcin
+ beqz has_nul1, _nul_in_data2
+ move has_nul2, has_nul1
+ addi.d len, len, -8
+_nul_in_data2:
+ ctz.d pos, has_nul2
+ srli.d pos, pos, 3
+ addi.d len, len, -8
+ add.d len, len, pos
+ sltu tmp1, len, limit
+ CONDITIONSEL(len, len, limit, tmp1)
+ jr ra
+
+misaligned:
+ addi.d limit_wd, limit, -1
+ sub.d tmp4, zero, tmp1
+ andi tmp3, limit_wd, 15
+ srli.d limit_wd, limit_wd, 4
+ li.d tmp5, -1
+ ld.d data1, src, 0
+ ld.d data2, src, 8
+ addi.d src, src, 16
+ slli.d tmp4, tmp4, 3
+ add.d tmp3, tmp3, tmp1
+ srl.d tmp2, tmp5, tmp4
+ srli.d tmp3, tmp3, 4
+ add.d limit_wd, limit_wd, tmp3
+ or data1, data1, tmp2
+ or data2a, data2, tmp2
+ li.w tmp3, 9
+ sltu tmp1, tmp1, tmp3
+ CONDITIONSEL(data1, data1, tmp5, tmp1)
+ CONDITIONSEL(data2, data2, data2a, tmp1)
+ b _realigned
+
+_hit_limit:
+ move len, limit
+ jr ra
+END(__strnlen)
+
+weak_alias (__strnlen, strnlen)
+libc_hidden_def (strnlen)
+libc_hidden_def (__strnlen)
--
2.33.0