diff options
Diffstat (limited to 'string/arm/memchr.S')
| -rw-r--r-- | string/arm/memchr.S | 46 |
1 files changed, 41 insertions, 5 deletions
diff --git a/string/arm/memchr.S b/string/arm/memchr.S index 3f1ac4d..823d601 100644 --- a/string/arm/memchr.S +++ b/string/arm/memchr.S @@ -1,8 +1,8 @@ /* * memchr - scan memory for a character * - * Copyright (c) 2010-2021, Arm Limited. - * SPDX-License-Identifier: MIT + * Copyright (c) 2010-2022, Arm Limited. + * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception */ /* @@ -23,7 +23,11 @@ @ Removed unneeded cbz from align loop .syntax unified +#if __ARM_ARCH >= 8 && __ARM_ARCH_PROFILE == 'M' + /* keep config inherited from -march= */ +#else .arch armv7-a +#endif @ this lets us check a flag in a 00/ff byte easily in either endianness #ifdef __ARMEB__ @@ -32,6 +36,8 @@ #define CHARTSTMASK(c) 1<<(c*8) #endif .thumb +#include "asmdefs.h" + @ --------------------------------------------------------------------------- .thumb_func @@ -39,11 +45,14 @@ .p2align 4,,15 .global __memchr_arm .type __memchr_arm,%function + .fnstart + .cfi_startproc __memchr_arm: @ r0 = start of memory to scan @ r1 = character to look for @ r2 = length @ returns r0 = pointer to character or NULL if not found + prologue and r1,r1,#0xff @ Don't think we can trust the caller to actually pass a char cmp r2,#16 @ If it's short don't bother with anything clever @@ -64,6 +73,11 @@ __memchr_arm: 10: @ At this point, we are aligned, we know we have at least 8 bytes to work with push {r4,r5,r6,r7} + .cfi_adjust_cfa_offset 16 + .cfi_rel_offset 4, 0 + .cfi_rel_offset 5, 4 + .cfi_rel_offset 6, 8 + .cfi_rel_offset 7, 12 orr r1, r1, r1, lsl #8 @ expand the match word across to all bytes orr r1, r1, r1, lsl #16 bic r4, r2, #7 @ Number of double words to work with @@ -83,6 +97,11 @@ __memchr_arm: bne 15b @ (Flags from the subs above) If not run out of bytes then go around again pop {r4,r5,r6,r7} + .cfi_restore 7 + .cfi_restore 6 + .cfi_restore 5 + .cfi_restore 4 + .cfi_adjust_cfa_offset -16 and r1,r1,#0xff @ Get r1 back to a single character from the expansion above and r2,r2,#7 @ Leave the count remaining as the number after the double words have been done @@ -97,16 +116,25 @@ __memchr_arm: bne 21b @ on r2 flags 40: + .cfi_remember_state movs r0,#0 @ not found - bx lr + epilogue 50: + .cfi_restore_state + .cfi_remember_state subs r0,r0,#1 @ found - bx lr + epilogue 60: @ We're here because the fast path found a hit - now we have to track down exactly which word it was @ r0 points to the start of the double word after the one that was tested @ r5 has the 00/ff pattern for the first word, r6 has the chained value + .cfi_restore_state @ Standard post-prologue state + .cfi_adjust_cfa_offset 16 + .cfi_rel_offset 4, 0 + .cfi_rel_offset 5, 4 + .cfi_rel_offset 6, 8 + .cfi_rel_offset 7, 12 cmp r5, #0 itte eq moveq r5, r6 @ the end is in the 2nd word @@ -126,7 +154,15 @@ __memchr_arm: 61: pop {r4,r5,r6,r7} + .cfi_restore 7 + .cfi_restore 6 + .cfi_restore 5 + .cfi_restore 4 + .cfi_adjust_cfa_offset -16 subs r0,r0,#1 - bx lr + epilogue + .cfi_endproc + .cantunwind + .fnend .size __memchr_arm, . - __memchr_arm |