1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
|
#!/usr/bin/env perl
##
## Copyright (c) 2016, Alliance for Open Media. All rights reserved
##
## This source code is subject to the terms of the BSD 2 Clause License and
## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
## was not distributed with this source code in the LICENSE file, you can
## obtain it at www.aomedia.org/license/software. If the Alliance for Open
## Media Patent License 1.0 was not distributed with this source code in the
## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
##
package thumb;
sub FixThumbInstructions($$)
{
my $short_branches = $_[1];
my $branch_shift_offset = $short_branches ? 1 : 0;
# Write additions with shifts, such as "add r10, r11, lsl #8",
# in three operand form, "add r10, r10, r11, lsl #8".
s/(add\s+)(r\d+),\s*(r\d+),\s*(lsl #\d+)/$1$2, $2, $3, $4/g;
# Convert additions with a non-constant shift into a sequence
# with left shift, addition and a right shift (to restore the
# register to the original value). Currently the right shift
# isn't necessary in the code base since the values in these
# registers aren't used, but doing the shift for consistency.
# This converts instructions such as "add r12, r12, r5, lsl r4"
# into the sequence "lsl r5, r4", "add r12, r12, r5", "lsr r5, r4".
s/^(\s*)(add)(\s+)(r\d+),\s*(r\d+),\s*(r\d+),\s*lsl (r\d+)/$1lsl$3$6, $7\n$1$2$3$4, $5, $6\n$1lsr$3$6, $7/g;
# Convert loads with right shifts in the indexing into a
# sequence of an add, load and sub. This converts
# "ldrb r4, [r9, lr, asr #1]" into "add r9, r9, lr, asr #1",
# "ldrb r9, [r9]", "sub r9, r9, lr, asr #1".
s/^(\s*)(ldrb)(\s+)(r\d+),\s*\[(\w+),\s*(\w+),\s*(asr #\d+)\]/$1add $3$5, $5, $6, $7\n$1$2$3$4, [$5]\n$1sub $3$5, $5, $6, $7/g;
# Convert register indexing with writeback into a separate add
# instruction. This converts "ldrb r12, [r1, r2]!" into
# "ldrb r12, [r1, r2]", "add r1, r1, r2".
s/^(\s*)(ldrb)(\s+)(r\d+),\s*\[(\w+),\s*(\w+)\]!/$1$2$3$4, [$5, $6]\n$1add $3$5, $6/g;
# Convert negative register indexing into separate sub/add instructions.
# This converts "ldrne r4, [src, -pstep, lsl #1]" into
# "subne src, src, pstep, lsl #1", "ldrne r4, [src]",
# "addne src, src, pstep, lsl #1". In a couple of cases where
# this is used, it's used for two subsequent load instructions,
# where a hand-written version of it could merge two subsequent
# add and sub instructions.
s/^(\s*)((ldr|str|pld)(ne)?)(\s+)(r\d+,\s*)?\[(\w+), -([^\]]+)\]/$1sub$4$5$7, $7, $8\n$1$2$5$6\[$7\]\n$1add$4$5$7, $7, $8/g;
# Convert register post indexing to a separate add instruction.
# This converts "ldrneb r9, [r0], r2" into "ldrneb r9, [r0]",
# "addne r0, r0, r2".
s/^(\s*)((ldr|str)(ne)?[bhd]?)(\s+)(\w+),(\s*\w+,)?\s*\[(\w+)\],\s*(\w+)/$1$2$5$6,$7 [$8]\n$1add$4$5$8, $8, $9/g;
# Convert "mov pc, lr" into "bx lr", since the former only works
# for switching from arm to thumb (and only in armv7), but not
# from thumb to arm.
s/mov(\s*)pc\s*,\s*lr/bx$1lr/g;
}
1;
|
