This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[PATCH 07/36] PowerPC: memrchr multilib for PowerPC32
- From: Adhemerval Zanella <azanella at linux dot vnet dot ibm dot com>
- To: "GNU C. Library" <libc-alpha at sourceware dot org>
- Date: Mon, 19 Aug 2013 18:28:02 -0300
- Subject: [PATCH 07/36] PowerPC: memrchr multilib for PowerPC32
- References: <52127ABC dot 40008 at linux dot vnet dot ibm dot com>
2013-08-19 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
* sysdeps/powerpc/powerpc32/power7/memrchr.S: Move to ...
* sysdeps/powerpc/powerpc32/multiarch/memrchr-power7.S : ... here.
(__memrchr): Rename symbol name to __memrchr_power7 and remove the
libc_hidden_builtin_def and weak_alias.
* sysdeps/powerpc/powerpc32/multiarch/memrchr-ppc32.c: New file:
default memrchr PPC32 implementation.
* sysdeps/powerpc/powerpc32/multiarch/memrchr.c: New file: multiarch
memrchr for PPC32.
* sysdeps/powerpc/powerpc32/multiarch/Makefile: Added memrchr
multiarch objects.
* sysdeps/powerpc/powerpc32/multiarch/ifunc-impl-list
(__libc_ifunc_impl_list): Likewise.
--
diff --git a/sysdeps/powerpc/powerpc32/multiarch/Makefile b/sysdeps/powerpc/powerpc32/multiarch/Makefile
index 9feb3c6..fde4fbf 100644
--- a/sysdeps/powerpc/powerpc32/multiarch/Makefile
+++ b/sysdeps/powerpc/powerpc32/multiarch/Makefile
@@ -3,5 +3,5 @@ sysdep_routines += memcpy-power7 memcpy-a2 memcpy-power6 memcpy-cell \
memcpy-power4 memcpy-ppc32 memcmp-power7 memcmp-power4 \
memcmp-ppc32 memset-power7 memset-power6 memset-power4 \
memset-ppc32 mempcpy-power7 mempcpy-ppc32 memchr-power7 \
- memchr-ppc32
+ memchr-ppc32 memrchr-power7 memrchr-ppc32
endif
diff --git a/sysdeps/powerpc/powerpc32/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc32/multiarch/ifunc-impl-list.c
index 279cc0b..4b3dd75 100644
--- a/sysdeps/powerpc/powerpc32/multiarch/ifunc-impl-list.c
+++ b/sysdeps/powerpc/powerpc32/multiarch/ifunc-impl-list.c
@@ -63,6 +63,13 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__memchr_power7)
IFUNC_IMPL_ADD (array, i, memchr, 1,
__memchr_ppc32))
+ /* Support sysdeps/powerpc/powerpc32/multiarch/memrchr.c. */
+ IFUNC_IMPL (i, name, memrchr,
+ IFUNC_IMPL_ADD (array, i, memrchr,
+ hwcap & PPC_FEATURE_HAS_VSX,
+ __memrchr_power7)
+ IFUNC_IMPL_ADD (array, i, memrchr, 1,
+ __memrchr_ppc32))
#ifdef SHARED
/* Support sysdeps/powerpc/powerpc32/multiarch/memset.c. */
diff --git a/sysdeps/powerpc/powerpc32/multiarch/memrchr-power7.S b/sysdeps/powerpc/powerpc32/multiarch/memrchr-power7.S
new file mode 100644
index 0000000..0aaf017
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/multiarch/memrchr-power7.S
@@ -0,0 +1,173 @@
+/* Optimized memrchr implementation for PowerPC32/POWER7 using cmpb insn.
+ Copyright (C) 2010-2013 Free Software Foundation, Inc.
+ Contributed by Luis Machado <luisgpm@br.ibm.com>.
+ 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/>. */
+
+#include <sysdep.h>
+
+/* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */
+ .machine power7
+ENTRY (__memrchr_power7)
+ CALL_MCOUNT
+ dcbt 0,r3
+ mr r7,r3
+ add r3,r7,r5 /* Calculate the last acceptable address. */
+ cmplw cr7,r3,r7 /* Is the address equal or less than r3? */
+
+ /* Replicate BYTE to word. */
+ rlwimi r4,r4,8,16,23
+ rlwimi r4,r4,16,0,15
+ bge cr7,L(proceed)
+
+ li r3,-1 /* Make r11 the biggest if r4 <= 0. */
+L(proceed):
+ li r6,-4
+ addi r9,r3,-1
+ clrrwi r8,r9,2
+ addi r8,r8,4
+ neg r0,r3
+ rlwinm r0,r0,3,27,28 /* Calculate padding. */
+
+ cmplwi r5,16
+ ble L(small_range)
+
+ lwbrx r12,r8,r6 /* Load reversed word from memory. */
+ cmpb r10,r12,r4 /* Check for BYTE in WORD1. */
+ slw r10,r10,r0
+ srw r10,r10,r0
+ cmplwi cr7,r10,0 /* If r10 == 0, no BYTEs have been found. */
+ bne cr7,L(done)
+
+ /* Are we done already? */
+ addi r9,r8,-4
+ cmplw cr6,r9,r7
+ ble cr6,L(null)
+
+ mtcrf 0x01,r8
+ /* Are we now aligned to a doubleword boundary? If so, skip to
+ the main loop. Otherwise, go through the alignment code. */
+ mr r8,r9
+ bt 29,L(loop_setup)
+
+ /* Handle WORD2 of pair. */
+ lwbrx r12,r8,r6
+ cmpb r10,r12,r4
+ cmplwi cr7,r10,0
+ bne cr7,L(done)
+
+ /* Are we done already? */
+ addi r8,r8,-4
+ cmplw cr6,r8,r7
+ ble cr6,L(null)
+
+L(loop_setup):
+ li r0,-8
+ sub r5,r8,r7
+ srwi r9,r5,3 /* Number of loop iterations. */
+ mtctr r9 /* Setup the counter. */
+ b L(loop)
+ /* Main loop to look for BYTE backwards in the string. Since it's a
+ small loop (< 8 instructions), align it to 32-bytes. */
+ .p2align 5
+L(loop):
+ /* Load two words, compare and merge in a
+ single register for speed. This is an attempt
+ to speed up the byte-checking process for bigger strings. */
+
+ lwbrx r12,r8,r6
+ lwbrx r11,r8,r0
+ addi r8,r8,-4
+ cmpb r10,r12,r4
+ cmpb r9,r11,r4
+ or r5,r9,r10 /* Merge everything in one word. */
+ cmplwi cr7,r5,0
+ bne cr7,L(found)
+ addi r8,r8,-4
+ bdnz L(loop)
+ /* We're here because the counter reached 0, and that means we
+ didn't have any matches for BYTE in the whole range. Just return
+ the original range. */
+ addi r9,r8,4
+ cmplw cr6,r9,r7
+ bgt cr6,L(loop_small)
+ b L(null)
+
+ /* OK, one (or both) of the words contains BYTE. Check
+ the first word and decrement the address in case the first
+ word really contains BYTE. */
+ .align 4
+L(found):
+ cmplwi cr6,r10,0
+ addi r8,r8,4
+ bne cr6,L(done)
+
+ /* BYTE must be in the second word. Adjust the address
+ again and move the result of cmpb to r10 so we can calculate the
+ pointer. */
+
+ mr r10,r9
+ addi r8,r8,-4
+
+ /* r10 has the output of the cmpb instruction, that is, it contains
+ 0xff in the same position as BYTE in the original
+ word from the string. Use that to calculate the pointer.
+ We need to make sure BYTE is *before* the end of the
+ range. */
+L(done):
+ cntlzw r0,r10 /* Count leading zeroes before the match. */
+ srwi r6,r0,3 /* Convert leading zeroes to bytes. */
+ addi r0,r6,1
+ sub r3,r8,r0
+ cmplw r3,r7
+ blt L(null)
+ blr
+
+ .align 4
+L(null):
+ li r3,0
+ blr
+
+/* Deals with size <= 16. */
+ .align 4
+L(small_range):
+ cmplwi r5,0
+ beq L(null)
+
+ lwbrx r12,r8,r6 /* Load reversed word from memory. */
+ cmpb r10,r12,r4 /* Check for null bytes in WORD1. */
+ slw r10,r10,r0
+ srw r10,r10,r0
+ cmplwi cr7,r10,0
+ bne cr7,L(done)
+
+ addi r8,r8,-4
+ cmplw r8,r7
+ ble L(null)
+ b L(loop_small)
+
+ .p2align 5
+L(loop_small):
+ lwbrx r12,r8,r6
+ cmpb r10,r12,r4
+ cmplwi cr6,r10,0
+ bne cr6,L(done)
+ addi r8,r8,-4
+ cmplw r8,r7
+ ble L(null)
+ b L(loop_small)
+
+END (__memrchr_power7)
diff --git a/sysdeps/powerpc/powerpc32/multiarch/memrchr-ppc32.c b/sysdeps/powerpc/powerpc32/multiarch/memrchr-ppc32.c
new file mode 100644
index 0000000..19ae015
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/multiarch/memrchr-ppc32.c
@@ -0,0 +1,25 @@
+/* PowerPC32 default implementation of memrchr.
+ Copyright (C) 2013 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/>. */
+
+#ifndef NOT_IN_libc
+# define MEMRCHR __memrchr_ppc32
+# include <string.h>
+extern void *__memrchr_ppc32 (const void *, int, size_t);
+#endif
+
+#include <string/memrchr.c>
diff --git a/sysdeps/powerpc/powerpc32/multiarch/memrchr.c b/sysdeps/powerpc/powerpc32/multiarch/memrchr.c
new file mode 100644
index 0000000..c3dde9c
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/multiarch/memrchr.c
@@ -0,0 +1,37 @@
+/* Multiple versions of memrchr.
+ Copyright (C) 2013 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/>. */
+
+#ifndef NOT_IN_libc
+# include <string.h>
+# include <shlib-compat.h>
+# include "init-arch.h"
+
+extern __typeof (__memrchr) __memrchr_ppc32 attribute_hidden;
+extern __typeof (__memrchr) __memrchr_power7 attribute_hidden;
+
+/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
+ ifunc symbol properly. */
+libc_ifunc (__memrchr,
+ (hwcap & PPC_FEATURE_HAS_VSX)
+ ? __memrchr_power7
+ : __memrchr_ppc32);
+
+weak_alias (__memrchr, memrchr)
+#else
+#include <string/memrchr.c>
+#endif
diff --git a/sysdeps/powerpc/powerpc32/power7/memrchr.S b/sysdeps/powerpc/powerpc32/power7/memrchr.S
deleted file mode 100644
index d1e3fda..0000000
--- a/sysdeps/powerpc/powerpc32/power7/memrchr.S
+++ /dev/null
@@ -1,175 +0,0 @@
-/* Optimized memrchr implementation for PowerPC32/POWER7 using cmpb insn.
- Copyright (C) 2010-2013 Free Software Foundation, Inc.
- Contributed by Luis Machado <luisgpm@br.ibm.com>.
- 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/>. */
-
-#include <sysdep.h>
-
-/* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */
- .machine power7
-ENTRY (__memrchr)
- CALL_MCOUNT
- dcbt 0,r3
- mr r7,r3
- add r3,r7,r5 /* Calculate the last acceptable address. */
- cmplw cr7,r3,r7 /* Is the address equal or less than r3? */
-
- /* Replicate BYTE to word. */
- rlwimi r4,r4,8,16,23
- rlwimi r4,r4,16,0,15
- bge cr7,L(proceed)
-
- li r3,-1 /* Make r11 the biggest if r4 <= 0. */
-L(proceed):
- li r6,-4
- addi r9,r3,-1
- clrrwi r8,r9,2
- addi r8,r8,4
- neg r0,r3
- rlwinm r0,r0,3,27,28 /* Calculate padding. */
-
- cmplwi r5,16
- ble L(small_range)
-
- lwbrx r12,r8,r6 /* Load reversed word from memory. */
- cmpb r10,r12,r4 /* Check for BYTE in WORD1. */
- slw r10,r10,r0
- srw r10,r10,r0
- cmplwi cr7,r10,0 /* If r10 == 0, no BYTEs have been found. */
- bne cr7,L(done)
-
- /* Are we done already? */
- addi r9,r8,-4
- cmplw cr6,r9,r7
- ble cr6,L(null)
-
- mtcrf 0x01,r8
- /* Are we now aligned to a doubleword boundary? If so, skip to
- the main loop. Otherwise, go through the alignment code. */
- mr r8,r9
- bt 29,L(loop_setup)
-
- /* Handle WORD2 of pair. */
- lwbrx r12,r8,r6
- cmpb r10,r12,r4
- cmplwi cr7,r10,0
- bne cr7,L(done)
-
- /* Are we done already? */
- addi r8,r8,-4
- cmplw cr6,r8,r7
- ble cr6,L(null)
-
-L(loop_setup):
- li r0,-8
- sub r5,r8,r7
- srwi r9,r5,3 /* Number of loop iterations. */
- mtctr r9 /* Setup the counter. */
- b L(loop)
- /* Main loop to look for BYTE backwards in the string. Since it's a
- small loop (< 8 instructions), align it to 32-bytes. */
- .p2align 5
-L(loop):
- /* Load two words, compare and merge in a
- single register for speed. This is an attempt
- to speed up the byte-checking process for bigger strings. */
-
- lwbrx r12,r8,r6
- lwbrx r11,r8,r0
- addi r8,r8,-4
- cmpb r10,r12,r4
- cmpb r9,r11,r4
- or r5,r9,r10 /* Merge everything in one word. */
- cmplwi cr7,r5,0
- bne cr7,L(found)
- addi r8,r8,-4
- bdnz L(loop)
- /* We're here because the counter reached 0, and that means we
- didn't have any matches for BYTE in the whole range. Just return
- the original range. */
- addi r9,r8,4
- cmplw cr6,r9,r7
- bgt cr6,L(loop_small)
- b L(null)
-
- /* OK, one (or both) of the words contains BYTE. Check
- the first word and decrement the address in case the first
- word really contains BYTE. */
- .align 4
-L(found):
- cmplwi cr6,r10,0
- addi r8,r8,4
- bne cr6,L(done)
-
- /* BYTE must be in the second word. Adjust the address
- again and move the result of cmpb to r10 so we can calculate the
- pointer. */
-
- mr r10,r9
- addi r8,r8,-4
-
- /* r10 has the output of the cmpb instruction, that is, it contains
- 0xff in the same position as BYTE in the original
- word from the string. Use that to calculate the pointer.
- We need to make sure BYTE is *before* the end of the
- range. */
-L(done):
- cntlzw r0,r10 /* Count leading zeroes before the match. */
- srwi r6,r0,3 /* Convert leading zeroes to bytes. */
- addi r0,r6,1
- sub r3,r8,r0
- cmplw r3,r7
- blt L(null)
- blr
-
- .align 4
-L(null):
- li r3,0
- blr
-
-/* Deals with size <= 16. */
- .align 4
-L(small_range):
- cmplwi r5,0
- beq L(null)
-
- lwbrx r12,r8,r6 /* Load reversed word from memory. */
- cmpb r10,r12,r4 /* Check for null bytes in WORD1. */
- slw r10,r10,r0
- srw r10,r10,r0
- cmplwi cr7,r10,0
- bne cr7,L(done)
-
- addi r8,r8,-4
- cmplw r8,r7
- ble L(null)
- b L(loop_small)
-
- .p2align 5
-L(loop_small):
- lwbrx r12,r8,r6
- cmpb r10,r12,r4
- cmplwi cr6,r10,0
- bne cr6,L(done)
- addi r8,r8,-4
- cmplw r8,r7
- ble L(null)
- b L(loop_small)
-
-END (__memrchr)
-weak_alias (__memrchr, memrchr)
-libc_hidden_builtin_def (memrchr)