alpha/lib/ev67-strlen.S

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 /*
0003  * arch/alpha/lib/ev67-strlen.S
0004  * 21264 version by Rick Gorton <rick.gorton@alpha-processor.com>
0005  *
0006  * Finds length of a 0-terminated string.  Optimized for the
0007  * Alpha architecture:
0008  *
0009  *  - memory accessed as aligned quadwords only
0010  *  - uses bcmpge to compare 8 bytes in parallel
0011  *
0012  * Much of the information about 21264 scheduling/coding comes from:
0013  *  Compiler Writer's Guide for the Alpha 21264
0014  *  abbreviated as 'CWG' in other comments here
0015  *  ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html
0016  * Scheduling notation:
0017  *  E   - either cluster
0018  *  U   - upper subcluster; U0 - subcluster U0; U1 - subcluster U1
0019  *  L   - lower subcluster; L0 - subcluster L0; L1 - subcluster L1
0020  */
0021 #include <asm/export.h>
0022     .set noreorder
0023     .set noat
0024
0025     .globl  strlen
0026     .ent    strlen
0027     .align 4
0028 strlen:
0029     ldq_u   $1, 0($16)  # L : load first quadword ($16  may be misaligned)
0030     lda $2, -1($31) # E :
0031     insqh   $2, $16, $2 # U :
0032     andnot  $16, 7, $0  # E :
0033
0034     or  $2, $1, $1  # E :
0035     cmpbge  $31, $1, $2 # E : $2  <- bitmask: bit i == 1 <==> i-th byte == 0
0036     nop         # E :
0037     bne $2, $found  # U :
0038
0039 $loop:  ldq $1, 8($0)   # L :
0040     addq    $0, 8, $0   # E : addr += 8
0041     cmpbge  $31, $1, $2 # E :
0042     beq $2, $loop   # U :
0043
0044 $found:
0045     cttz    $2, $3      # U0 :
0046     addq    $0, $3, $0  # E :
0047     subq    $0, $16, $0 # E :
0048     ret $31, ($26)  # L0 :
0049
0050     .end    strlen
0051     EXPORT_SYMBOL(strlen)