|
 |
00db10 |
From 250962053fadd0911accad515802d9b957c9d019 Mon Sep 17 00:00:00 2001
|
|
|
00db10 |
From: Stefan Liebler <stli@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Thu, 8 Oct 2015 13:24:58 +0200
|
|
|
00db10 |
Subject: [PATCH 30/30] S390: Optimize memrchr.
|
|
|
00db10 |
|
|
|
00db10 |
upstream-commit-id: 798f5b4b5d24e5687dd55acc535954ed1d59c10c
|
|
|
00db10 |
https://www.sourceware.org/ml/libc-alpha/2015-07/msg00092.html
|
|
|
00db10 |
|
|
|
00db10 |
This patch provides optimized version of memrchr with the z13 vector
|
|
|
00db10 |
instructions.
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/s390/multiarch/memrchr-c.c: New File.
|
|
|
00db10 |
* sysdeps/s390/multiarch/memrchr-vx.S: Likewise.
|
|
|
00db10 |
* sysdeps/s390/multiarch/memrchr.c: Likewise.
|
|
|
00db10 |
* sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
(sysdep_routines): Add memrchr functions.
|
|
|
00db10 |
* sysdeps/s390/multiarch/ifunc-impl-list-common.c
|
|
|
00db10 |
(__libc_ifunc_impl_list_common): Add ifunc test for memrchr.
|
|
|
00db10 |
---
|
|
|
00db10 |
sysdeps/s390/multiarch/Makefile | 3 +-
|
|
|
00db10 |
sysdeps/s390/multiarch/ifunc-impl-list.c | 2 +
|
|
|
00db10 |
sysdeps/s390/multiarch/memrchr-c.c | 25 +++++
|
|
|
00db10 |
sysdeps/s390/multiarch/memrchr-vx.S | 160 +++++++++++++++++++++++++++++++
|
|
|
00db10 |
sysdeps/s390/multiarch/memrchr.c | 28 ++++++
|
|
|
00db10 |
5 files changed, 217 insertions(+), 1 deletion(-)
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/memrchr-c.c
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/memrchr-vx.S
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/memrchr.c
|
|
|
00db10 |
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/Makefile b/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
index 929a545..0805b07 100644
|
|
|
00db10 |
--- a/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
@@ -17,7 +17,8 @@ sysdep_routines += strlen strlen-vx strlen-c \
|
|
|
00db10 |
strcspn strcspn-vx strcspn-c \
|
|
|
00db10 |
memchr memchr-vx \
|
|
|
00db10 |
rawmemchr rawmemchr-vx rawmemchr-c \
|
|
|
00db10 |
- memccpy memccpy-vx memccpy-c
|
|
|
00db10 |
+ memccpy memccpy-vx memccpy-c \
|
|
|
00db10 |
+ memrchr memrchr-vx memrchr-c
|
|
|
00db10 |
endif
|
|
|
00db10 |
|
|
|
00db10 |
ifeq ($(subdir),wcsmbs)
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/ifunc-impl-list.c b/sysdeps/s390/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
index 5ea258b..c235bdc 100644
|
|
|
00db10 |
--- a/sysdeps/s390/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
@@ -137,6 +137,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
00db10 |
|
|
|
00db10 |
IFUNC_VX_IMPL (wmemcmp);
|
|
|
00db10 |
|
|
|
00db10 |
+ IFUNC_VX_IMPL (memrchr);
|
|
|
00db10 |
+
|
|
|
00db10 |
#endif /* HAVE_S390_VX_ASM_SUPPORT */
|
|
|
00db10 |
|
|
|
00db10 |
return i;
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/memrchr-c.c b/sysdeps/s390/multiarch/memrchr-c.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..6a32861
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/memrchr-c.c
|
|
|
00db10 |
@@ -0,0 +1,25 @@
|
|
|
00db10 |
+/* Default memrchr implementation for S/390.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
|
|
|
00db10 |
+# define MEMRCHR __memrchr_c
|
|
|
00db10 |
+
|
|
|
00db10 |
+# include <string.h>
|
|
|
00db10 |
+extern __typeof (__memrchr) __memrchr_c;
|
|
|
00db10 |
+# include <string/memrchr.c>
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/memrchr-vx.S b/sysdeps/s390/multiarch/memrchr-vx.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..da6dbdf
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/memrchr-vx.S
|
|
|
00db10 |
@@ -0,0 +1,160 @@
|
|
|
00db10 |
+/* Vector optimized 32/64 bit S/390 version of memrchr.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
|
|
|
00db10 |
+
|
|
|
00db10 |
+# include "sysdep.h"
|
|
|
00db10 |
+# include "asm-syntax.h"
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .text
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* void *memrchr (const void *s, int c, size_t n)
|
|
|
00db10 |
+ Scans memory for character c backwards
|
|
|
00db10 |
+ and returns pointer to first c.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ Register usage:
|
|
|
00db10 |
+ -r0=tmp
|
|
|
00db10 |
+ -r1=tmp
|
|
|
00db10 |
+ -r2=s
|
|
|
00db10 |
+ -r3=c
|
|
|
00db10 |
+ -r4=n
|
|
|
00db10 |
+ -r5=s in loop
|
|
|
00db10 |
+
|
|
|
00db10 |
+ -v16=part of s
|
|
|
00db10 |
+ -v17=index of found c
|
|
|
00db10 |
+ -v18=c replicated
|
|
|
00db10 |
+ -v20=permute pattern
|
|
|
00db10 |
+*/
|
|
|
00db10 |
+ENTRY(__memrchr_vx)
|
|
|
00db10 |
+ .machine "z13"
|
|
|
00db10 |
+ .machinemode "zarch_nohighgprs"
|
|
|
00db10 |
+
|
|
|
00db10 |
+# if !defined __s390x__
|
|
|
00db10 |
+ llgfr %r4,%r4
|
|
|
00db10 |
+# endif /* !defined __s390x__ */
|
|
|
00db10 |
+ clgije %r4,0,.Lnot_found
|
|
|
00db10 |
+
|
|
|
00db10 |
+ vlvgb %v18,%r3,0 /* Generate vector which elements are all c.
|
|
|
00db10 |
+ If c > 255, c will be truncated. */
|
|
|
00db10 |
+ vrepb %v18,%v18,0
|
|
|
00db10 |
+
|
|
|
00db10 |
+ llcr %r3,%r3 /* char c_char = (char) c. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* check byte n - 1. */
|
|
|
00db10 |
+ llc %r0,-1(%r4,%r2)
|
|
|
00db10 |
+ slgfi %r4,1
|
|
|
00db10 |
+ clrje %r0,%r3,.Lfound_end
|
|
|
00db10 |
+ jh .Lnot_found /* Return NULL if n is now 0. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ larl %r1,.Lpermute_mask /* Load permute mask. */
|
|
|
00db10 |
+ vl %v20,0(%r1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* check byte n - 2. */
|
|
|
00db10 |
+ llc %r0,-1(%r4,%r2)
|
|
|
00db10 |
+ slgfi %r4,1
|
|
|
00db10 |
+ clrje %r0,%r3,.Lfound_end
|
|
|
00db10 |
+ jh .Lnot_found /* Return NULL if n is now 0. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ clgijhe %r4,64,.Lloop64 /* If n >= 64 -> loop64. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+.Llt64:
|
|
|
00db10 |
+ /* Process n < 64 bytes. */
|
|
|
00db10 |
+ clgijl %r4,16,.Llt16 /* Jump away if n < 16. */
|
|
|
00db10 |
+ aghi %r4,-16
|
|
|
00db10 |
+ vl %v16,0(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound0
|
|
|
00db10 |
+ clgijl %r4,16,.Llt16
|
|
|
00db10 |
+ aghi %r4,-16
|
|
|
00db10 |
+ vl %v16,0(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound0
|
|
|
00db10 |
+ clgijl %r4,16,.Llt16
|
|
|
00db10 |
+ aghi %r4,-16
|
|
|
00db10 |
+ vl %v16,0(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound0
|
|
|
00db10 |
+.Llt16:
|
|
|
00db10 |
+ clgfi %r4,0 /* if remaining bytes == 0, return NULL. */
|
|
|
00db10 |
+ locghie %r2,0
|
|
|
00db10 |
+ ber %r14
|
|
|
00db10 |
+
|
|
|
00db10 |
+ aghi %r4,-1 /* vll needs highest index. */
|
|
|
00db10 |
+ vll %v16,%r4,0(%r2) /* Load remaining bytes. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Right-shift of v16 to mask bytes after highest index. */
|
|
|
00db10 |
+ lhi %r0,15
|
|
|
00db10 |
+ slr %r0,%r4 /* Compute byte count for vector shift right. */
|
|
|
00db10 |
+ sll %r0,3 /* Convert to bit count. */
|
|
|
00db10 |
+ vlvgb %v17,%r0,7
|
|
|
00db10 |
+ vsrlb %v16,%v16,%v17 /* Vector shift right by byte by number of bytes
|
|
|
00db10 |
+ specified in bits 1-4 of byte 7 in v17. */
|
|
|
00db10 |
+ j .Lfound_permute
|
|
|
00db10 |
+
|
|
|
00db10 |
+.Lfound48:
|
|
|
00db10 |
+ aghi %r4,16
|
|
|
00db10 |
+.Lfound32:
|
|
|
00db10 |
+ aghi %r4,16
|
|
|
00db10 |
+.Lfound16:
|
|
|
00db10 |
+ aghi %r4,16
|
|
|
00db10 |
+.Lfound0:
|
|
|
00db10 |
+ la %r2,0(%r4,%r2) /* Set pointer to start of v16. */
|
|
|
00db10 |
+ lghi %r4,15 /* Set highest index in v16 to last index. */
|
|
|
00db10 |
+.Lfound_permute:
|
|
|
00db10 |
+ /* Search for a c in v16 in reversed byte order. v16 contains %r4 + 1
|
|
|
00db10 |
+ bytes. If v16 was not fully loaded, the bytes are already
|
|
|
00db10 |
+ right shifted, so that the bytes in v16 can simply be reversed. */
|
|
|
00db10 |
+ vperm %v16,%v16,%v16,%v20 /* Permute v16 to reversed order. */
|
|
|
00db10 |
+ vfeeb %v16,%v16,%v18 /* Find c in reversed v16. */
|
|
|
00db10 |
+ vlgvb %r1,%v16,7 /* Index of c or 16 if not found. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Return NULL if there is no c in loaded bytes. */
|
|
|
00db10 |
+ clrjh %r1,%r4,.Lnot_found
|
|
|
00db10 |
+
|
|
|
00db10 |
+ slgr %r4,%r1
|
|
|
00db10 |
+.Lfound_end:
|
|
|
00db10 |
+ la %r2,0(%r4,%r2) /* Return pointer to c. */
|
|
|
00db10 |
+ br %r14
|
|
|
00db10 |
+
|
|
|
00db10 |
+.Lnot_found:
|
|
|
00db10 |
+ lghi %r2,0
|
|
|
00db10 |
+ br %r14
|
|
|
00db10 |
+
|
|
|
00db10 |
+.Lpermute_mask:
|
|
|
00db10 |
+ .byte 0x0F,0x0E,0x0D,0x0C,0x0B,0x0A,0x09,0x08
|
|
|
00db10 |
+ .byte 0x07,0x06,0x05,0x04,0x03,0x02,0x01,0x00
|
|
|
00db10 |
+
|
|
|
00db10 |
+.Lloop64:
|
|
|
00db10 |
+ aghi %r4,-64
|
|
|
00db10 |
+ vl %v16,48(%r4,%r2) /* Load 16bytes of memory area. */
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18 /* Find c. */
|
|
|
00db10 |
+ jno .Lfound48 /* Jump away if c was found. */
|
|
|
00db10 |
+ vl %v16,32(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound32
|
|
|
00db10 |
+ vl %v16,16(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound16
|
|
|
00db10 |
+ vl %v16,0(%r4,%r2)
|
|
|
00db10 |
+ vfeebs %v17,%v16,%v18
|
|
|
00db10 |
+ jno .Lfound0
|
|
|
00db10 |
+
|
|
|
00db10 |
+ clgijhe %r4,64,.Lloop64 /* If n >= 64 -> loop64. */
|
|
|
00db10 |
+ j .Llt64
|
|
|
00db10 |
+END(__memrchr_vx)
|
|
|
00db10 |
+#endif /* HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc */
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/memrchr.c b/sysdeps/s390/multiarch/memrchr.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..6596067
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/memrchr.c
|
|
|
00db10 |
@@ -0,0 +1,28 @@
|
|
|
00db10 |
+/* Multiple versions of memrchr.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
|
|
|
00db10 |
+# include <string.h>
|
|
|
00db10 |
+# include <ifunc-resolve.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+s390_vx_libc_ifunc (__memrchr)
|
|
|
00db10 |
+weak_alias (__memrchr, memrchr)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+# include <string/memrchr.c>
|
|
|
00db10 |
+#endif /* !(defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc) */
|
|
|
00db10 |
--
|
|
|
00db10 |
2.3.0
|
|
|
00db10 |
|