diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64/power7/strnlen.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/power7/strnlen.S | 121 |
1 files changed, 63 insertions, 58 deletions
diff --git a/sysdeps/powerpc/powerpc64/power7/strnlen.S b/sysdeps/powerpc/powerpc64/power7/strnlen.S index 043cc159b2..7993dae69e 100644 --- a/sysdeps/powerpc/powerpc64/power7/strnlen.S +++ b/sysdeps/powerpc/powerpc64/power7/strnlen.S @@ -1,5 +1,5 @@ /* Optimized strnlen implementation for PowerPC64/POWER7 using cmpb insn. - Copyright (C) 2010 Free Software Foundation, Inc. + Copyright (C) 2010-2014 Free Software Foundation, Inc. Contributed by Luis Machado <luisgpm@br.ibm.com>. This file is part of the GNU C Library. @@ -18,41 +18,35 @@ <http://www.gnu.org/licenses/>. */ #include <sysdep.h> -#include <bp-sym.h> -#include <bp-asm.h> /* int [r3] strnlen (char *s [r3], int size [r4]) */ .machine power7 -ENTRY (BP_SYM (__strnlen)) +ENTRY (__strnlen) CALL_MCOUNT 2 dcbt 0,r3 - clrrdi r8,r3,3 + clrrdi r8,r3,3 add r7,r3,r4 /* Calculate the last acceptable address. */ cmpldi r4,32 li r0,0 /* Doubleword with null chars. */ + addi r7,r7,-1 + /* If we have less than 33 bytes to search, skip to a faster code. */ ble L(small_range) - cmpld cr7,r3,r7 /* Is the address equal or less than r3? If - it's equal or less, it means size is either 0 - or a negative number. */ - ble cr7,L(proceed) - - li r7,-1 /* Make r11 the biggest if r4 <= 0. */ -L(proceed): rlwinm r6,r3,3,26,28 /* Calculate padding. */ ld r12,0(r8) /* Load doubleword from memory. */ cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + sld r10,r10,r6 +#else sld r10,r10,r6 srd r10,r10,r6 +#endif cmpldi cr7,r10,0 /* If r10 == 0, no null's have been found. */ bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(end_max) - + clrrdi r7,r7,3 /* Address of last doubleword. */ mtcrf 0x01,r8 /* Are we now aligned to a quadword boundary? If so, skip to the main loop. Otherwise, go through the alignment code. */ @@ -65,17 +59,18 @@ L(proceed): cmpldi cr7,r10,0 bne cr7,L(done) - /* Are we done already? */ - addi r9,r8,8 - cmpld cr6,r9,r7 - bge cr6,L(end_max) - L(loop_setup): - sub r5,r7,r9 + /* The last dword we want to read in the loop below is the one + containing the last byte of the string, ie. the dword at + (s + size - 1) & ~7, or r7. The first dword read is at + r8 + 8, we read 2 * cnt dwords, so the last dword read will + be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives + cnt = (r7 - r8) / 16 */ + sub r5,r7,r8 srdi r6,r5,4 /* Number of loop iterations. */ mtctr r6 /* Setup the counter. */ - b L(loop) - /* Main loop to look for the null byte backwards in the string. Since + + /* Main loop to look for the null byte in the string. Since it's a small loop (< 8 instructions), align it to 32-bytes. */ .p2align 5 L(loop): @@ -91,15 +86,18 @@ L(loop): cmpldi cr7,r5,0 bne cr7,L(found) bdnz L(loop) - /* We're here because the counter reached 0, and that means we - didn't have any matches for null in the whole range. Just return - the original size. */ - addi r9,r8,8 - cmpld cr6,r9,r7 - blt cr6,L(loop_small) + + /* We may have one more dword to read. */ + cmpld cr6,r8,r7 + beq cr6,L(end_max) + + ldu r12,8(r8) + cmpb r10,r12,r0 + cmpldi cr6,r10,0 + bne cr6,L(done) L(end_max): - sub r3,r7,r3 + mr r3,r4 blr /* OK, one (or both) of the doublewords contains a null byte. Check @@ -121,52 +119,59 @@ L(found): /* r10 has the output of the cmpb instruction, that is, it contains 0xff in the same position as the null byte in the original doubleword from the string. Use that to calculate the length. - We need to make sure the null char is *before* the start of the - range (since we're going backwards). */ + We need to make sure the null char is *before* the end of the + range. */ L(done): - cntlzd r0,r10 /* Count leading zeroes before the match. */ - srdi r0,r0,3 /* Convert leading zeroes to bytes. */ - add r9,r8,r0 - sub r6,r9,r3 /* Length until the match. */ - cmpld r9,r7 - bgt L(end_max) - mr r3,r6 - blr - - .align 4 -L(zero): - li r3,0 +#ifdef __LITTLE_ENDIAN__ + addi r0,r10,-1 + andc r0,r0,r10 + popcntd r0,r0 +#else + cntlzd r0,r10 /* Count leading zeros before the match. */ +#endif + sub r3,r8,r3 + srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */ + add r3,r3,r0 /* Length until the match. */ + cmpld r3,r4 + blelr + mr r3,r4 blr /* Deals with size <= 32. */ .align 4 L(small_range): cmpldi r4,0 - beq L(zero) + beq L(end_max) + + clrrdi r7,r7,3 /* Address of last doubleword. */ rlwinm r6,r3,3,26,28 /* Calculate padding. */ - ld r12,0(r8) /* Load word from memory. */ + ld r12,0(r8) /* Load doubleword from memory. */ cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */ +#ifdef __LITTLE_ENDIAN__ + srd r10,r10,r6 + sld r10,r10,r6 +#else sld r10,r10,r6 srd r10,r10,r6 +#endif cmpldi cr7,r10,0 bne cr7,L(done) - addi r9,r8,8 - cmpld r9,r7 - bge L(end_max) - b L(loop_small) + cmpld r8,r7 + beq L(end_max) .p2align 5 L(loop_small): ldu r12,8(r8) cmpb r10,r12,r0 - addi r9,r8,8 cmpldi cr6,r10,0 bne cr6,L(done) - cmpld r9,r7 - bge L(end_max) - b L(loop_small) -END (BP_SYM (__strnlen)) -weak_alias (BP_SYM (__strnlen), BP_SYM(strnlen)) + cmpld r8,r7 + bne L(loop_small) + mr r3,r4 + blr + +END (__strnlen) +weak_alias (__strnlen, strnlen) libc_hidden_builtin_def (strnlen) |