summaryrefslogtreecommitdiffstats
path: root/sbc/sbc_primitives_neon.c
blob: d9c12f9e3f04428410aae7d41e640042e15dc369 (plain)
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
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
/*
 *
 *  Bluetooth low-complexity, subband codec (SBC) library
 *
 *  Copyright (C) 2004-2009  Marcel Holtmann <marcel@holtmann.org>
 *  Copyright (C) 2004-2005  Henryk Ploetz <henryk@ploetzli.ch>
 *  Copyright (C) 2005-2006  Brad Midgley <bmidgley@xmission.com>
 *
 *
 *  This library is free software; you can redistribute it and/or
 *  modify it under the terms of the GNU Lesser General Public
 *  License as published by the Free Software Foundation; either
 *  version 2.1 of the License, or (at your option) any later version.
 *
 *  This library is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 *  Lesser General Public License for more details.
 *
 *  You should have received a copy of the GNU Lesser General Public
 *  License along with this library; if not, write to the Free Software
 *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 *
 */

#include <stdint.h>
#include <limits.h>
#include "sbc.h"
#include "sbc_math.h"
#include "sbc_tables.h"

#include "sbc_primitives_neon.h"

/*
 * ARM NEON optimizations
 */

#ifdef SBC_BUILD_WITH_NEON_SUPPORT

static inline void _sbc_analyze_four_neon(const int16_t *in, int32_t *out,
							const FIXED_T *consts)
{
	/* TODO: merge even and odd cases (or even merge all four calls to this
	 * function) in order to have only aligned reads from 'in' array
	 * and reduce number of load instructions */
	asm volatile (
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmull.s16  q0, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmull.s16  q1, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"

		"vmlal.s16  q0, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmlal.s16  q1, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q0, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmlal.s16  q1, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"

		"vmlal.s16  q0, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmlal.s16  q1, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q0, d4, d8\n"
		"vmlal.s16  q1, d5, d9\n"

		"vpadd.s32  d0, d0, d1\n"
		"vpadd.s32  d1, d2, d3\n"

		"vrshrn.s32 d0, q0, %3\n"

		"vld1.16    {d2, d3, d4, d5}, [%1, :128]!\n"

		"vdup.i32   d1, d0[1]\n"  /* TODO: can be eliminated */
		"vdup.i32   d0, d0[0]\n"  /* TODO: can be eliminated */

		"vmull.s16  q3, d2, d0\n"
		"vmull.s16  q4, d3, d0\n"
		"vmlal.s16  q3, d4, d1\n"
		"vmlal.s16  q4, d5, d1\n"

		"vpadd.s32  d0, d6, d7\n" /* TODO: can be eliminated */
		"vpadd.s32  d1, d8, d9\n" /* TODO: can be eliminated */

		"vst1.32    {d0, d1}, [%2, :128]\n"
		: "+r" (in), "+r" (consts)
		: "r" (out),
			"i" (SBC_PROTO_FIXED4_SCALE)
		: "memory",
			"d0", "d1", "d2", "d3", "d4", "d5",
			"d6", "d7", "d8", "d9", "d10", "d11");
}

static inline void _sbc_analyze_eight_neon(const int16_t *in, int32_t *out,
							const FIXED_T *consts)
{
	/* TODO: merge even and odd cases (or even merge all four calls to this
	 * function) in order to have only aligned reads from 'in' array
	 * and reduce number of load instructions */
	asm volatile (
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmull.s16  q6, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmull.s16  q7, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"
		"vmull.s16  q8, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmull.s16  q9, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q6, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmlal.s16  q7, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"
		"vmlal.s16  q8, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmlal.s16  q9, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q6, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmlal.s16  q7, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"
		"vmlal.s16  q8, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmlal.s16  q9, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q6, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmlal.s16  q7, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"
		"vmlal.s16  q8, d6, d10\n"
		"vld1.16    {d4, d5}, [%0, :64]!\n"
		"vmlal.s16  q9, d7, d11\n"
		"vld1.16    {d8, d9}, [%1, :128]!\n"

		"vmlal.s16  q6, d4, d8\n"
		"vld1.16    {d6,  d7}, [%0, :64]!\n"
		"vmlal.s16  q7, d5, d9\n"
		"vld1.16    {d10, d11}, [%1, :128]!\n"

		"vmlal.s16  q8, d6, d10\n"
		"vmlal.s16  q9, d7, d11\n"

		"vpadd.s32  d0, d12, d13\n"
		"vpadd.s32  d1, d14, d15\n"
		"vpadd.s32  d2, d16, d17\n"
		"vpadd.s32  d3, d18, d19\n"

		"vrshr.s32 q0, q0, %3\n"
		"vrshr.s32 q1, q1, %3\n"
		"vmovn.s32 d0, q0\n"
		"vmovn.s32 d1, q1\n"

		"vdup.i32   d3, d1[1]\n"  /* TODO: can be eliminated */
		"vdup.i32   d2, d1[0]\n"  /* TODO: can be eliminated */
		"vdup.i32   d1, d0[1]\n"  /* TODO: can be eliminated */
		"vdup.i32   d0, d0[0]\n"  /* TODO: can be eliminated */

		"vld1.16    {d4, d5}, [%1, :128]!\n"
		"vmull.s16  q6, d4, d0\n"
		"vld1.16    {d6, d7}, [%1, :128]!\n"
		"vmull.s16  q7, d5, d0\n"
		"vmull.s16  q8, d6, d0\n"
		"vmull.s16  q9, d7, d0\n"

		"vld1.16    {d4, d5}, [%1, :128]!\n"
		"vmlal.s16  q6, d4, d1\n"
		"vld1.16    {d6, d7}, [%1, :128]!\n"
		"vmlal.s16  q7, d5, d1\n"
		"vmlal.s16  q8, d6, d1\n"
		"vmlal.s16  q9, d7, d1\n"

		"vld1.16    {d4, d5}, [%1, :128]!\n"
		"vmlal.s16  q6, d4, d2\n"
		"vld1.16    {d6, d7}, [%1, :128]!\n"
		"vmlal.s16  q7, d5, d2\n"
		"vmlal.s16  q8, d6, d2\n"
		"vmlal.s16  q9, d7, d2\n"

		"vld1.16    {d4, d5}, [%1, :128]!\n"
		"vmlal.s16  q6, d4, d3\n"
		"vld1.16    {d6, d7}, [%1, :128]!\n"
		"vmlal.s16  q7, d5, d3\n"
		"vmlal.s16  q8, d6, d3\n"
		"vmlal.s16  q9, d7, d3\n"

		"vpadd.s32  d0, d12, d13\n" /* TODO: can be eliminated */
		"vpadd.s32  d1, d14, d15\n" /* TODO: can be eliminated */
		"vpadd.s32  d2, d16, d17\n" /* TODO: can be eliminated */
		"vpadd.s32  d3, d18, d19\n" /* TODO: can be eliminated */

		"vst1.32    {d0, d1, d2, d3}, [%2, :128]\n"
		: "+r" (in), "+r" (consts)
		: "r" (out),
			"i" (SBC_PROTO_FIXED8_SCALE)
		: "memory",
			"d0", "d1", "d2", "d3", "d4", "d5",
			"d6", "d7", "d8", "d9", "d10", "d11",
			"d12", "d13", "d14", "d15", "d16", "d17",
			"d18", "d19");
}

static inline void sbc_analyze_4b_4s_neon(int16_t *x,
						int32_t *out, int out_stride)
{
	/* Analyze blocks */
	_sbc_analyze_four_neon(x + 12, out, analysis_consts_fixed4_simd_odd);
	out += out_stride;
	_sbc_analyze_four_neon(x + 8, out, analysis_consts_fixed4_simd_even);
	out += out_stride;
	_sbc_analyze_four_neon(x + 4, out, analysis_consts_fixed4_simd_odd);
	out += out_stride;
	_sbc_analyze_four_neon(x + 0, out, analysis_consts_fixed4_simd_even);
}

static inline void sbc_analyze_4b_8s_neon(int16_t *x,
						int32_t *out, int out_stride)
{
	/* Analyze blocks */
	_sbc_analyze_eight_neon(x + 24, out, analysis_consts_fixed8_simd_odd);
	out += out_stride;
	_sbc_analyze_eight_neon(x + 16, out, analysis_consts_fixed8_simd_even);
	out += out_stride;
	_sbc_analyze_eight_neon(x + 8, out, analysis_consts_fixed8_simd_odd);
	out += out_stride;
	_sbc_analyze_eight_neon(x + 0, out, analysis_consts_fixed8_simd_even);
}

void sbc_init_primitives_neon(struct sbc_encoder_state *state)
{
	state->sbc_analyze_4b_4s = sbc_analyze_4b_4s_neon;
	state->sbc_analyze_4b_8s = sbc_analyze_4b_8s_neon;
}

#endif