18c2ecf20Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0 */ 28c2ecf20Sopenharmony_ci/* 38c2ecf20Sopenharmony_ci * arch/alpha/lib/ev67-strchr.S 48c2ecf20Sopenharmony_ci * 21264 version contributed by Rick Gorton <rick.gorton@alpha-processor.com> 58c2ecf20Sopenharmony_ci * 68c2ecf20Sopenharmony_ci * Return the address of a given character within a null-terminated 78c2ecf20Sopenharmony_ci * string, or null if it is not found. 88c2ecf20Sopenharmony_ci * 98c2ecf20Sopenharmony_ci * Much of the information about 21264 scheduling/coding comes from: 108c2ecf20Sopenharmony_ci * Compiler Writer's Guide for the Alpha 21264 118c2ecf20Sopenharmony_ci * abbreviated as 'CWG' in other comments here 128c2ecf20Sopenharmony_ci * ftp.digital.com/pub/Digital/info/semiconductor/literature/dsc-library.html 138c2ecf20Sopenharmony_ci * Scheduling notation: 148c2ecf20Sopenharmony_ci * E - either cluster 158c2ecf20Sopenharmony_ci * U - upper subcluster; U0 - subcluster U0; U1 - subcluster U1 168c2ecf20Sopenharmony_ci * L - lower subcluster; L0 - subcluster L0; L1 - subcluster L1 178c2ecf20Sopenharmony_ci * Try not to change the actual algorithm if possible for consistency. 188c2ecf20Sopenharmony_ci */ 198c2ecf20Sopenharmony_ci#include <asm/export.h> 208c2ecf20Sopenharmony_ci#include <asm/regdef.h> 218c2ecf20Sopenharmony_ci 228c2ecf20Sopenharmony_ci .set noreorder 238c2ecf20Sopenharmony_ci .set noat 248c2ecf20Sopenharmony_ci 258c2ecf20Sopenharmony_ci .align 4 268c2ecf20Sopenharmony_ci .globl strchr 278c2ecf20Sopenharmony_ci .ent strchr 288c2ecf20Sopenharmony_cistrchr: 298c2ecf20Sopenharmony_ci .frame sp, 0, ra 308c2ecf20Sopenharmony_ci .prologue 0 318c2ecf20Sopenharmony_ci 328c2ecf20Sopenharmony_ci ldq_u t0, 0(a0) # L : load first quadword Latency=3 338c2ecf20Sopenharmony_ci and a1, 0xff, t3 # E : 00000000000000ch 348c2ecf20Sopenharmony_ci insbl a1, 1, t5 # U : 000000000000ch00 358c2ecf20Sopenharmony_ci insbl a1, 7, a2 # U : ch00000000000000 368c2ecf20Sopenharmony_ci 378c2ecf20Sopenharmony_ci insbl t3, 6, a3 # U : 00ch000000000000 388c2ecf20Sopenharmony_ci or t5, t3, a1 # E : 000000000000chch 398c2ecf20Sopenharmony_ci andnot a0, 7, v0 # E : align our loop pointer 408c2ecf20Sopenharmony_ci lda t4, -1 # E : build garbage mask 418c2ecf20Sopenharmony_ci 428c2ecf20Sopenharmony_ci mskqh t4, a0, t4 # U : only want relevant part of first quad 438c2ecf20Sopenharmony_ci or a2, a3, a2 # E : chch000000000000 448c2ecf20Sopenharmony_ci inswl a1, 2, t5 # E : 00000000chch0000 458c2ecf20Sopenharmony_ci inswl a1, 4, a3 # E : 0000chch00000000 468c2ecf20Sopenharmony_ci 478c2ecf20Sopenharmony_ci or a1, a2, a1 # E : chch00000000chch 488c2ecf20Sopenharmony_ci or a3, t5, t5 # E : 0000chchchch0000 498c2ecf20Sopenharmony_ci cmpbge zero, t0, t2 # E : bits set iff byte == zero 508c2ecf20Sopenharmony_ci cmpbge zero, t4, t4 # E : bits set iff byte is garbage 518c2ecf20Sopenharmony_ci 528c2ecf20Sopenharmony_ci /* This quad is _very_ serialized. Lots of stalling happens */ 538c2ecf20Sopenharmony_ci or t5, a1, a1 # E : chchchchchchchch 548c2ecf20Sopenharmony_ci xor t0, a1, t1 # E : make bytes == c zero 558c2ecf20Sopenharmony_ci cmpbge zero, t1, t3 # E : bits set iff byte == c 568c2ecf20Sopenharmony_ci or t2, t3, t0 # E : bits set iff char match or zero match 578c2ecf20Sopenharmony_ci 588c2ecf20Sopenharmony_ci andnot t0, t4, t0 # E : clear garbage bits 598c2ecf20Sopenharmony_ci cttz t0, a2 # U0 : speculative (in case we get a match) 608c2ecf20Sopenharmony_ci nop # E : 618c2ecf20Sopenharmony_ci bne t0, $found # U : 628c2ecf20Sopenharmony_ci 638c2ecf20Sopenharmony_ci /* 648c2ecf20Sopenharmony_ci * Yuk. This loop is going to stall like crazy waiting for the 658c2ecf20Sopenharmony_ci * data to be loaded. Not much can be done about it unless it's 668c2ecf20Sopenharmony_ci * unrolled multiple times - is that safe to do in kernel space? 678c2ecf20Sopenharmony_ci * Or would exception handling recovery code do the trick here? 688c2ecf20Sopenharmony_ci */ 698c2ecf20Sopenharmony_ci$loop: ldq t0, 8(v0) # L : Latency=3 708c2ecf20Sopenharmony_ci addq v0, 8, v0 # E : 718c2ecf20Sopenharmony_ci xor t0, a1, t1 # E : 728c2ecf20Sopenharmony_ci cmpbge zero, t0, t2 # E : bits set iff byte == 0 738c2ecf20Sopenharmony_ci 748c2ecf20Sopenharmony_ci cmpbge zero, t1, t3 # E : bits set iff byte == c 758c2ecf20Sopenharmony_ci or t2, t3, t0 # E : 768c2ecf20Sopenharmony_ci cttz t3, a2 # U0 : speculative (in case we get a match) 778c2ecf20Sopenharmony_ci beq t0, $loop # U : 788c2ecf20Sopenharmony_ci 798c2ecf20Sopenharmony_ci$found: negq t0, t1 # E : clear all but least set bit 808c2ecf20Sopenharmony_ci and t0, t1, t0 # E : 818c2ecf20Sopenharmony_ci and t0, t3, t1 # E : bit set iff byte was the char 828c2ecf20Sopenharmony_ci addq v0, a2, v0 # E : Add in the bit number from above 838c2ecf20Sopenharmony_ci 848c2ecf20Sopenharmony_ci cmoveq t1, $31, v0 # E : Two mapping slots, latency = 2 858c2ecf20Sopenharmony_ci nop 868c2ecf20Sopenharmony_ci nop 878c2ecf20Sopenharmony_ci ret # L0 : 888c2ecf20Sopenharmony_ci 898c2ecf20Sopenharmony_ci .end strchr 908c2ecf20Sopenharmony_ci EXPORT_SYMBOL(strchr) 91