162306a36Sopenharmony_ci/* SPDX-License-Identifier: GPL-2.0-only */ 262306a36Sopenharmony_ci/* 362306a36Sopenharmony_ci * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com) 462306a36Sopenharmony_ci */ 562306a36Sopenharmony_ci 662306a36Sopenharmony_ci/* If dst and src are 4 byte aligned, copy 8 bytes at a time. 762306a36Sopenharmony_ci If the src is 4, but not 8 byte aligned, we first read 4 bytes to get 862306a36Sopenharmony_ci it 8 byte aligned. Thus, we can do a little read-ahead, without 962306a36Sopenharmony_ci dereferencing a cache line that we should not touch. 1062306a36Sopenharmony_ci Note that short and long instructions have been scheduled to avoid 1162306a36Sopenharmony_ci branch stalls. 1262306a36Sopenharmony_ci The beq_s to r3z could be made unaligned & long to avoid a stall 1362306a36Sopenharmony_ci there, but the it is not likely to be taken often, and it 1462306a36Sopenharmony_ci would also be likey to cost an unaligned mispredict at the next call. */ 1562306a36Sopenharmony_ci 1662306a36Sopenharmony_ci#include <linux/linkage.h> 1762306a36Sopenharmony_ci 1862306a36Sopenharmony_ciENTRY_CFI(strcpy) 1962306a36Sopenharmony_ci or r2,r0,r1 2062306a36Sopenharmony_ci bmsk_s r2,r2,1 2162306a36Sopenharmony_ci brne.d r2,0,charloop 2262306a36Sopenharmony_ci mov_s r10,r0 2362306a36Sopenharmony_ci ld_s r3,[r1,0] 2462306a36Sopenharmony_ci mov r8,0x01010101 2562306a36Sopenharmony_ci bbit0.d r1,2,loop_start 2662306a36Sopenharmony_ci ror r12,r8 2762306a36Sopenharmony_ci sub r2,r3,r8 2862306a36Sopenharmony_ci bic_s r2,r2,r3 2962306a36Sopenharmony_ci tst_s r2,r12 3062306a36Sopenharmony_ci bne r3z 3162306a36Sopenharmony_ci mov_s r4,r3 3262306a36Sopenharmony_ci .balign 4 3362306a36Sopenharmony_ciloop: 3462306a36Sopenharmony_ci ld.a r3,[r1,4] 3562306a36Sopenharmony_ci st.ab r4,[r10,4] 3662306a36Sopenharmony_ciloop_start: 3762306a36Sopenharmony_ci ld.a r4,[r1,4] 3862306a36Sopenharmony_ci sub r2,r3,r8 3962306a36Sopenharmony_ci bic_s r2,r2,r3 4062306a36Sopenharmony_ci tst_s r2,r12 4162306a36Sopenharmony_ci bne_s r3z 4262306a36Sopenharmony_ci st.ab r3,[r10,4] 4362306a36Sopenharmony_ci sub r2,r4,r8 4462306a36Sopenharmony_ci bic r2,r2,r4 4562306a36Sopenharmony_ci tst r2,r12 4662306a36Sopenharmony_ci beq loop 4762306a36Sopenharmony_ci mov_s r3,r4 4862306a36Sopenharmony_ci#ifdef __LITTLE_ENDIAN__ 4962306a36Sopenharmony_cir3z: bmsk.f r1,r3,7 5062306a36Sopenharmony_ci lsr_s r3,r3,8 5162306a36Sopenharmony_ci#else 5262306a36Sopenharmony_cir3z: lsr.f r1,r3,24 5362306a36Sopenharmony_ci asl_s r3,r3,8 5462306a36Sopenharmony_ci#endif 5562306a36Sopenharmony_ci bne.d r3z 5662306a36Sopenharmony_ci stb.ab r1,[r10,1] 5762306a36Sopenharmony_ci j_s [blink] 5862306a36Sopenharmony_ci 5962306a36Sopenharmony_ci .balign 4 6062306a36Sopenharmony_cicharloop: 6162306a36Sopenharmony_ci ldb.ab r3,[r1,1] 6262306a36Sopenharmony_ci 6362306a36Sopenharmony_ci 6462306a36Sopenharmony_ci brne.d r3,0,charloop 6562306a36Sopenharmony_ci stb.ab r3,[r10,1] 6662306a36Sopenharmony_ci j [blink] 6762306a36Sopenharmony_ciEND_CFI(strcpy) 68