xref: /aosp_15_r20/external/libxaac/decoder/armv7/ixheaacd_radix4_bfly.s (revision 15dc779a375ca8b5125643b829a8aa4b70d7f451)
1@/******************************************************************************
2@ *
3@ * Copyright (C) 2018 The Android Open Source Project
4@ *
5@ * Licensed under the Apache License, Version 2.0 (the "License");
6@ * you may not use this file except in compliance with the License.
7@ * You may obtain a copy of the License at:
8@ *
9@ * http://www.apache.org/licenses/LICENSE-2.0
10@ *
11@ * Unless required by applicable law or agreed to in writing, software
12@ * distributed under the License is distributed on an "AS IS" BASIS,
13@ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14@ * See the License for the specific language governing permissions and
15@ * limitations under the License.
16@ *
17@ *****************************************************************************
18@ * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
19@*/
20
21
22.text
23.p2align 2
24        .global ixheaacd_radix4bfly
25	.type ixheaacd_radix4bfly, %function
26
27ixheaacd_radix4bfly:
28
29    STMFD           sp!, {r4-r12, r14}
30
31    SUB             sp, sp, #16
32
33    MOV             r6, #6
34    MUL             r7, r6, r3
35    MOV             r4, r3
36    STR             r7, [sp]
37
38
39
40    MOV             r3, r3, lsl #1
41
42    STR             r2, [sp, #8]
43    STR             r4, [sp, #12]
44
45
46    ADD             r2, r1, r3, lsl #2
47    ADD             r0, r0, #8
48
49
50RADIX4_OUTLOOP:
51RADIX4_INLOOP:
52
53
54    LDR             r6, [r1]
55    LDR             r7, [r2]
56    LDR             r8, [r2, r3, lsl #2]
57    LDR             r9, [r2, r3, lsl #3]
58
59    ADD             r10, r6, r8
60    SUB             r11, r6, r8
61    ADD             r12, r7, r9
62    SUB             r14, r7, r9
63
64    ADD             r6, r10, r12
65    SUB             r7, r10, r12
66    STR             r6, [r1], #4
67
68    LDR             r8, [r1]
69    LDR             r6, [r2, #4]!
70    LDR             r9, [r2, r3, lsl #2]!
71    LDR             r10, [r2, r3, lsl #2]!
72
73    ADD             r12, r8, r9
74    SUB             r8, r8, r9
75    ADD             r9, r6, r10
76    SUB             r6, r6, r10
77
78    ADD             r10, r12, r9
79    STR             r10, [r1], #4
80    SUB             r12, r12, r9
81
82    ADD             r9, r11, r6
83    SUB             r10, r11, r6
84    ADD             r11, r8, r14
85    LDR             r5, [r0], #-4
86    SUB             r6, r8, r14
87
88    SMULWB          r14, r10, r5
89    SMULWT          r8, r11, r5
90
91    SUBS            r4, r4, #1
92    SUB             r8, r8, r14
93    MOV             r8, r8, lsl #1
94    STR             r8, [r2], #-4
95
96    SMULWT          r14, r10, r5
97    SMLAWB          r8, r11, r5, r14
98    LDR             r11, [r0], #-4
99    MOV             r8, r8, lsl #1
100    STR             r8, [r2], -r3, lsl #2
101
102    SMULWT          r10, r7, r11
103    SMLAWB          r8, r12, r11, r10
104
105    LDR             r14, [r0], #20
106    MOV             r5, r8, lsl #1
107
108    SMULWB          r10, r7, r11
109    SMULWT          r8, r12, r11
110
111    STR             r5, [r2], #4
112    SUB             r7, r8, r10
113    MOV             r7, r7, lsl #1
114
115    SMULWB          r11, r9, r14
116    SMULWT          r12, r6, r14
117
118    STR             r7, [r2], -r3, lsl #2
119    SUB             r12, r12, r11
120    MOV             r12, r12, lsl #1
121
122    SMULWT          r10, r9, r14
123    SMLAWB          r7, r6, r14, r10
124
125    STR             r12, [r2], #-4
126    MOV             r7, r7, lsl #1
127    STR             r7, [r2], #8
128
129
130    BNE             RADIX4_INLOOP
131
132    LDR             r8, [sp]
133    LDR             r4, [sp, #12]
134    LDR             r6, [sp, #8]
135
136
137    SUB             r0, r0, r8, lsl #1
138    ADD             r1, r1, r8, lsl #2
139    ADD             r2, r2, r8, lsl #2
140
141    SUBS            r6, r6, #1
142    STR             r6, [sp, #8]
143    BNE             RADIX4_OUTLOOP
144
145
146
147    ADD             sp, sp, #16
148    LDMFD           sp!, {r4-r12, r15}
149
150
151