Warning: That file was not part of the compilation database. It may have many parsing errors.

1/* Copyright (C) 1996-2019 Free Software Foundation, Inc.
2 Contributed by Richard Henderson (rth@tamu.edu)
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library. If not, see
17 <http://www.gnu.org/licenses/>. */
18
19/* Fill a block of memory with a character. Optimized for the Alpha
20 architecture:
21
22 - memory accessed as aligned quadwords only
23 - destination memory not read unless needed for good cache behaviour
24 - basic blocks arranged to optimize branch prediction for full-quadword
25 aligned memory blocks.
26 - partial head and tail quadwords constructed with byte-mask instructions
27
28 This is generally scheduled for the EV5 (got to look out for my own
29 interests :-), but with EV4 needs in mind. There *should* be no more
30 stalls for the EV4 than there are for the EV5.
31*/
32
33
34#include <sysdep.h>
35
36 .set noat
37 .set noreorder
38
39 .text
40 .type memset, @function
41 .globl memset
42 .usepv memset, USEPV_PROF
43
44 cfi_startproc
45
46 /* On entry to this basic block:
47 t3 == loop counter
48 t4 == bytes in partial final word
49 a0 == possibly misaligned destination pointer
50 a1 == replicated source character */
51
52 .align 3
53memset_loop:
54 beq t3, $tail
55 blbc t3, 0f # skip single store if count even
56
57 stq_u a1, 0(a0) # e0 : store one word
58 subq t3, 1, t3 # .. e1 :
59 addq a0, 8, a0 # e0 :
60 beq t3, $tail # .. e1 :
61
620: stq_u a1, 0(a0) # e0 : store two words
63 subq t3, 2, t3 # .. e1 :
64 stq_u a1, 8(a0) # e0 :
65 addq a0, 16, a0 # .. e1 :
66 bne t3, 0b # e1 :
67
68$tail: bne t4, 1f # is there a tail to do?
69 ret # no
70
71 .align 3
721: ldq_u t0, 0(a0) # e1 : yes, load original data
73 mskql a1, t4, t1 # .. e0 :
74 mskqh t0, t4, t0 # e0 :
75 or t0, t1, t0 # e1 (stall)
76 stq_u t0, 0(a0) # e0 :
77 ret # .. e1 :
78
79memset:
80#ifdef PROF
81 ldgp gp, 0(pv)
82 lda AT, _mcount
83 jsr AT, (AT), _mcount
84#endif
85
86 and a1, 0xff, a1 # e0 : zero extend input character
87 mov a0, v0 # .. e1 : move return value in place
88 sll a1, 8, t0 # e0 : begin replicating the char
89 beq a2, $done # .. e1 : early exit for zero-length store
90 or t0, a1, a1 # e0 :
91 and a0, 7, t1 # .. e1 : dest misalignment
92 sll a1, 16, t0 # e0 :
93 addq a2, t1, a2 # .. e1 : add dest misalignment to count
94 or t0, a1, a1 # e0 :
95 srl a2, 3, t3 # .. e1 : loop = count >> 3
96 sll a1, 32, t0 # e0 :
97 and a2, 7, t4 # .. e1 : find number of bytes in tail
98 or t0, a1, a1 # e0 : character replication done
99
100 beq t1, memset_loop # .. e1 : aligned head, jump right in
101
102 ldq_u t0, 0(a0) # e1 : load original data to mask into
103 mskqh a1, a0, t1 # .. e0 :
104
105 cmpult a2, 8, t2 # e0 : is this a sub-word set?
106 bne t2, $oneq # .. e1 (zdb)
107
108 mskql t0, a0, t0 # e0 : we span words. finish this partial
109 subq t3, 1, t3 # .. e1 :
110 addq a0, 8, a0 # e0 :
111 or t0, t1, t0 # .. e1 :
112 stq_u t0, -8(a0) # e0 :
113 br memset_loop # .. e1 :
114
115 .align 3
116$oneq:
117 mskql t1, a2, t1 # e0 : entire operation within one word
118 mskql t0, a0, t2 # e0 :
119 mskqh t0, a2, t3 # e0 :
120 or t1, t2, t0 # .. e1 :
121 or t0, t3, t0 # e1 :
122 stq_u t0, 0(a0) # e0 (stall)
123
124$done: ret
125
126 cfi_endproc
127libc_hidden_builtin_def (memset)
128

Warning: That file was not part of the compilation database. It may have many parsing errors.