2 * arch/xtensa/lib/memset.S
4 * ANSI C standard library function memset
5 * (Well, almost. .fixup code might return zero.)
7 * This file is subject to the terms and conditions of the GNU General
8 * Public License. See the file "COPYING" in the main directory of
9 * this archive for more details.
11 * Copyright (C) 2002 Tensilica Inc.
14 #include <xtensa/coreasm.h>
17 * void *memset(void *dst, int c, size_t length)
19 * The algorithm is as follows:
20 * Create a word with c in all byte positions
21 * If the destination is aligned,
22 * do 16B chucks with a loop, and then finish up with
23 * 8B, 4B, 2B, and 1B stores conditional on the length.
24 * If destination is unaligned, align it by conditionally
25 * setting 1B and 2B and then go to aligned case.
26 * This code tries to use fall-through branches for the common
27 * case of an aligned destination (except for the branches to
28 * the alignment labels).
31 /* Load or store instructions that may cause exceptions use the EX macro. */
33 #define EX(insn,reg1,reg2,offset,handler) \
34 9: insn reg1, reg2, offset; \
35 .section __ex_table, "a"; \
43 .type memset,@function
45 entry sp, 16 # minimal stack frame
46 # a2/ dst, a3/ c, a4/ length
47 extui a3, a3, 0, 8 # mask to just 8 bits
48 slli a7, a3, 8 # duplicate character in all bytes of word
52 mov a5, a2 # copy dst so that a2 is return value
53 movi a6, 3 # for alignment tests
54 bany a2, a6, .Ldstunaligned # if dst is unaligned
55 .L0: # return here from .Ldstunaligned when dst is aligned
56 srli a7, a4, 4 # number of loop iterations with 16B
62 * Destination is word-aligned.
64 # set 16 bytes per iteration for word-aligned dst
65 .align 4 # 1 mod 4 alignment for LOOPNEZ
66 .byte 0 # (0 mod 4 alignment for LBEG)
69 loopnez a7, .Loop1done
70 #else /* !XCHAL_HAVE_LOOPS */
73 add a6, a6, a5 # a6 = end of last 16B chunk
74 #endif /* !XCHAL_HAVE_LOOPS */
76 EX(s32i, a3, a5, 0, memset_fixup)
77 EX(s32i, a3, a5, 4, memset_fixup)
78 EX(s32i, a3, a5, 8, memset_fixup)
79 EX(s32i, a3, a5, 12, memset_fixup)
83 #endif /* !XCHAL_HAVE_LOOPS */
87 EX(s32i, a3, a5, 0, memset_fixup)
88 EX(s32i, a3, a5, 4, memset_fixup)
93 EX(s32i, a3, a5, 0, memset_fixup)
98 EX(s16i, a3, a5, 0, memset_fixup)
103 EX(s8i, a3, a5, 0, memset_fixup)
109 * Destination is unaligned
113 bltui a4, 8, .Lbyteset # do short copies byte by byte
114 bbci.l a5, 0, .L20 # branch if dst alignment half-aligned
115 # dst is only byte aligned
117 EX(s8i, a3, a5, 0, memset_fixup)
120 # now retest if dst aligned
121 bbci.l a5, 1, .L0 # if now aligned, return to main algorithm
125 EX(s16i, a3, a5, 0, memset_fixup)
128 j .L0 # dst is now aligned, return to main algorithm
134 .byte 0 # 1 mod 4 alignment for LOOPNEZ
135 # (0 mod 4 alignment for LBEG)
138 loopnez a4, .Lbytesetdone
139 #else /* !XCHAL_HAVE_LOOPS */
140 beqz a4, .Lbytesetdone
141 add a6, a5, a4 # a6 = ending address
142 #endif /* !XCHAL_HAVE_LOOPS */
144 EX(s8i, a3, a5, 0, memset_fixup)
146 #if !XCHAL_HAVE_LOOPS
147 blt a5, a6, .Lbyteloop
148 #endif /* !XCHAL_HAVE_LOOPS */
153 .section .fixup, "ax"
156 /* We return zero if a failure occurred. */