aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/powerpc/powerpc32/power7/memrchr.S
diff options
context:
space:
mode:
Diffstat (limited to 'sysdeps/powerpc/powerpc32/power7/memrchr.S')
-rw-r--r--sysdeps/powerpc/powerpc32/power7/memrchr.S178
1 files changed, 178 insertions, 0 deletions
diff --git a/sysdeps/powerpc/powerpc32/power7/memrchr.S b/sysdeps/powerpc/powerpc32/power7/memrchr.S
new file mode 100644
index 0000000000..ea31166f4a
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/power7/memrchr.S
@@ -0,0 +1,178 @@
+/* Optimized memrchr implementation for PowerPC32/POWER7 using cmpb insn.
+ Copyright (C) 2010 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, write to the Free
+ Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA
+ 02110-1301 USA. */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+/* int [r3] memrchr (char *s [r3], int byte [r4], int size [r5]) */
+ .machine power7
+ENTRY (BP_SYM (__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 BYTE's 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 (BP_SYM (__memrchr))
+weak_alias (BP_SYM (__memrchr), BP_SYM(memrchr))
+libc_hidden_builtin_def (memrchr)