162306a36Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0 */ 262306a36Sopenharmony_ci/* 362306a36Sopenharmony_ci * arch/alpha/lib/ev67-strcat.S 462306a36Sopenharmony_ci * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com> 562306a36Sopenharmony_ci * 662306a36Sopenharmony_ci * Append a null-terminated string from SRC to DST. 762306a36Sopenharmony_ci * 862306a36Sopenharmony_ci * Much of the information about 21264 scheduling/coding comes from: 962306a36Sopenharmony_ci * Compiler Writer's Guide for the Alpha 21264 1062306a36Sopenharmony_ci * abbreviated as 'CWG' in other comments here 1162306a36Sopenharmony_ci * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html 1262306a36Sopenharmony_ci * Scheduling notation: 1362306a36Sopenharmony_ci * E - either cluster 1462306a36Sopenharmony_ci * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 1562306a36Sopenharmony_ci * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 1662306a36Sopenharmony_ci * Try not to change the actual algorithm if possible for consistency. 1762306a36Sopenharmony_ci * Commentary: It seems bogus to walk the input string twice - once 1862306a36Sopenharmony_ci * to determine the length, and then again while doing the copy. 1962306a36Sopenharmony_ci * A significant (future) enhancement would be to only read the input 2062306a36Sopenharmony_ci * string once. 2162306a36Sopenharmony_ci */ 2262306a36Sopenharmony_ci 2362306a36Sopenharmony_ci#include <linux/export.h> 2462306a36Sopenharmony_ci .text 2562306a36Sopenharmony_ci 2662306a36Sopenharmony_ci .align 4 2762306a36Sopenharmony_ci .globl strcat 2862306a36Sopenharmony_ci .ent strcat 2962306a36Sopenharmony_cistrcat: 3062306a36Sopenharmony_ci .frame $30, 0, $26 3162306a36Sopenharmony_ci .prologue 0 3262306a36Sopenharmony_ci 3362306a36Sopenharmony_ci mov $16, $0 # E : set up return value 3462306a36Sopenharmony_ci /* Find the end of the string. */ 3562306a36Sopenharmony_ci ldq_u $1, 0($16) # L : load first quadword (a0 may be misaligned) 3662306a36Sopenharmony_ci lda $2, -1 # E : 3762306a36Sopenharmony_ci insqh $2, $16, $2 # U : 3862306a36Sopenharmony_ci 3962306a36Sopenharmony_ci andnot $16, 7, $16 # E : 4062306a36Sopenharmony_ci or $2, $1, $1 # E : 4162306a36Sopenharmony_ci cmpbge $31, $1, $2 # E : bits set iff byte == 0 4262306a36Sopenharmony_ci bne $2, $found # U : 4362306a36Sopenharmony_ci 4462306a36Sopenharmony_ci$loop: ldq $1, 8($16) # L : 4562306a36Sopenharmony_ci addq $16, 8, $16 # E : 4662306a36Sopenharmony_ci cmpbge $31, $1, $2 # E : 4762306a36Sopenharmony_ci beq $2, $loop # U : 4862306a36Sopenharmony_ci 4962306a36Sopenharmony_ci$found: cttz $2, $3 # U0 : 5062306a36Sopenharmony_ci addq $16, $3, $16 # E : 5162306a36Sopenharmony_ci /* Now do the append. */ 5262306a36Sopenharmony_ci mov $26, $23 # E : 5362306a36Sopenharmony_ci br __stxcpy # L0 : 5462306a36Sopenharmony_ci 5562306a36Sopenharmony_ci .end strcat 5662306a36Sopenharmony_ci EXPORT_SYMBOL(strcat) 57