18c2ecf20Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0 */ 28c2ecf20Sopenharmony_ci/* 38c2ecf20Sopenharmony_ci * arch/alpha/lib/ev67-strcat.S 48c2ecf20Sopenharmony_ci * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com> 58c2ecf20Sopenharmony_ci * 68c2ecf20Sopenharmony_ci * Append a null-terminated string from SRC to DST. 78c2ecf20Sopenharmony_ci * 88c2ecf20Sopenharmony_ci * Much of the information about 21264 scheduling/coding comes from: 98c2ecf20Sopenharmony_ci * Compiler Writer's Guide for the Alpha 21264 108c2ecf20Sopenharmony_ci * abbreviated as 'CWG' in other comments here 118c2ecf20Sopenharmony_ci * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html 128c2ecf20Sopenharmony_ci * Scheduling notation: 138c2ecf20Sopenharmony_ci * E - either cluster 148c2ecf20Sopenharmony_ci * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 158c2ecf20Sopenharmony_ci * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 168c2ecf20Sopenharmony_ci * Try not to change the actual algorithm if possible for consistency. 178c2ecf20Sopenharmony_ci * Commentary: It seems bogus to walk the input string twice - once 188c2ecf20Sopenharmony_ci * to determine the length, and then again while doing the copy. 198c2ecf20Sopenharmony_ci * A significant (future) enhancement would be to only read the input 208c2ecf20Sopenharmony_ci * string once. 218c2ecf20Sopenharmony_ci */ 228c2ecf20Sopenharmony_ci 238c2ecf20Sopenharmony_ci#include <asm/export.h> 248c2ecf20Sopenharmony_ci .text 258c2ecf20Sopenharmony_ci 268c2ecf20Sopenharmony_ci .align 4 278c2ecf20Sopenharmony_ci .globl strcat 288c2ecf20Sopenharmony_ci .ent strcat 298c2ecf20Sopenharmony_cistrcat: 308c2ecf20Sopenharmony_ci .frame $30, 0, $26 318c2ecf20Sopenharmony_ci .prologue 0 328c2ecf20Sopenharmony_ci 338c2ecf20Sopenharmony_ci mov $16, $0 # E : set up return value 348c2ecf20Sopenharmony_ci /* Find the end of the string. */ 358c2ecf20Sopenharmony_ci ldq_u $1, 0($16) # L : load first quadword (a0 may be misaligned) 368c2ecf20Sopenharmony_ci lda $2, -1 # E : 378c2ecf20Sopenharmony_ci insqh $2, $16, $2 # U : 388c2ecf20Sopenharmony_ci 398c2ecf20Sopenharmony_ci andnot $16, 7, $16 # E : 408c2ecf20Sopenharmony_ci or $2, $1, $1 # E : 418c2ecf20Sopenharmony_ci cmpbge $31, $1, $2 # E : bits set iff byte == 0 428c2ecf20Sopenharmony_ci bne $2, $found # U : 438c2ecf20Sopenharmony_ci 448c2ecf20Sopenharmony_ci$loop: ldq $1, 8($16) # L : 458c2ecf20Sopenharmony_ci addq $16, 8, $16 # E : 468c2ecf20Sopenharmony_ci cmpbge $31, $1, $2 # E : 478c2ecf20Sopenharmony_ci beq $2, $loop # U : 488c2ecf20Sopenharmony_ci 498c2ecf20Sopenharmony_ci$found: cttz $2, $3 # U0 : 508c2ecf20Sopenharmony_ci addq $16, $3, $16 # E : 518c2ecf20Sopenharmony_ci /* Now do the append. */ 528c2ecf20Sopenharmony_ci mov $26, $23 # E : 538c2ecf20Sopenharmony_ci br __stxcpy # L0 : 548c2ecf20Sopenharmony_ci 558c2ecf20Sopenharmony_ci .end strcat 568c2ecf20Sopenharmony_ci EXPORT_SYMBOL(strcat) 57