fixed_intrin_sse2.c 11.4 KB
Newer Older
1
2
/* libFLAC - Free Lossless Audio Codec library
 * Copyright (C) 2000-2009  Josh Coalson
3
 * Copyright (C) 2011-2016  Xiph.Org Foundation
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
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer.
 *
 * - Redistributions in binary form must reproduce the above copyright
 * notice, this list of conditions and the following disclaimer in the
 * documentation and/or other materials provided with the distribution.
 *
 * - Neither the name of the Xiph.org Foundation nor the names of its
 * contributors may be used to endorse or promote products derived from
 * this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#ifdef HAVE_CONFIG_H
#  include <config.h>
#endif

37
38
#include "private/cpu.h"

39
40
#ifndef FLAC__INTEGER_ONLY_LIBRARY
#ifndef FLAC__NO_ASM
41
#if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN
42
43
44
45
46
47
48
49
50
51
#include "private/fixed.h"
#ifdef FLAC__SSE2_SUPPORTED

#include <emmintrin.h> /* SSE2 */
#include <math.h>
#include "private/macros.h"
#include "share/compat.h"
#include "FLAC/assert.h"

#ifdef FLAC__CPU_IA32
52
53
54
#define m128i_to_i64(dest, src) _mm_storel_epi64((__m128i*)&dest, src)
#else
#define m128i_to_i64(dest, src) dest = _mm_cvtsi128_si64(src)
55
56
57
#endif

FLAC__SSE_TARGET("sse2")
58
unsigned FLAC__fixed_compute_best_predictor_intrin_sse2(const FLAC__int32 data[], unsigned data_len, float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER + 1])
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
{
	FLAC__uint32 total_error_0, total_error_1, total_error_2, total_error_3, total_error_4;
	unsigned i, order;

	__m128i total_err0, total_err1, total_err2;

	{
		FLAC__int32 itmp;
		__m128i last_error;

		last_error = _mm_cvtsi32_si128(data[-1]);							// 0   0   0   le0
		itmp = data[-2];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// 0   0   le0 le1
		itmp -= data[-3];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// 0   le0 le1 le2
		itmp -= data[-3] - data[-4];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// le0 le1 le2 le3

		total_err0 = total_err1 = _mm_setzero_si128();
		for(i = 0; i < data_len; i++) {
			__m128i err0, err1, tmp;
			err0 = _mm_cvtsi32_si128(data[i]);								// 0   0   0   e0
			err1 = _mm_shuffle_epi32(err0, _MM_SHUFFLE(0,0,0,0));			// e0  e0  e0  e0
85
#if 1 /* OPT_SSE */
86
87
88
89
90
91
92
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   le0 le1 le2
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   0   le0 le1
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   0   0   le0
			err1 = _mm_sub_epi32(err1, last_error);							// e1  e2  e3  e4
93
94
95
96
97
#else
			last_error = _mm_add_epi32(last_error, _mm_srli_si128(last_error, 8));	// le0  le1  le2+le0  le3+le1
			last_error = _mm_add_epi32(last_error, _mm_srli_si128(last_error, 4));	// le0  le1+le0  le2+le0+le1  le3+le1+le2+le0
			err1 = _mm_sub_epi32(err1, last_error);							// e1  e2  e3  e4
#endif
98
			tmp = _mm_slli_si128(err0, 12);									// e0   0   0   0
99
			last_error = _mm_srli_si128(err1, 4);							//  0  e1  e2  e3
100
101
102
103
104
105
106
107
108
109
110
111
112
			last_error = _mm_or_si128(last_error, tmp);						// e0  e1  e2  e3

			tmp = _mm_srai_epi32(err0, 31);
			err0 = _mm_xor_si128(err0, tmp);
			err0 = _mm_sub_epi32(err0, tmp);
			tmp = _mm_srai_epi32(err1, 31);
			err1 = _mm_xor_si128(err1, tmp);
			err1 = _mm_sub_epi32(err1, tmp);

			total_err0 = _mm_add_epi32(total_err0, err0);					// 0   0   0   te0
			total_err1 = _mm_add_epi32(total_err1, err1);					// te1 te2 te3 te4
		}
	}
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
	total_error_0 = _mm_cvtsi128_si32(total_err0);
	total_err2 = total_err1;											// te1  te2  te3  te4
	total_err1 = _mm_srli_si128(total_err1, 8);							//  0    0   te1  te2
	total_error_4 = _mm_cvtsi128_si32(total_err2);
	total_error_2 = _mm_cvtsi128_si32(total_err1);
	total_err2 = _mm_srli_si128(total_err2,	4);							//  0   te1  te2  te3
	total_err1 = _mm_srli_si128(total_err1, 4);							//  0    0    0   te1
	total_error_3 = _mm_cvtsi128_si32(total_err2);
	total_error_1 = _mm_cvtsi128_si32(total_err1);

	/* prefer higher order */
	if(total_error_0 < flac_min(flac_min(flac_min(total_error_1, total_error_2), total_error_3), total_error_4))
		order = 0;
	else if(total_error_1 < flac_min(flac_min(total_error_2, total_error_3), total_error_4))
		order = 1;
	else if(total_error_2 < flac_min(total_error_3, total_error_4))
		order = 2;
	else if(total_error_3 < total_error_4)
		order = 3;
	else
		order = 4;

	/* Estimate the expected number of bits per residual signal sample. */
	/* 'total_error*' is linearly related to the variance of the residual */
	/* signal, so we use it directly to compute E(|x|) */
	FLAC__ASSERT(data_len > 0 || total_error_0 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_1 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_2 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_3 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_4 == 0);

145
146
147
148
149
	residual_bits_per_sample[0] = (float)((total_error_0 > 0) ? log(M_LN2 * (double)total_error_0 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[1] = (float)((total_error_1 > 0) ? log(M_LN2 * (double)total_error_1 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[2] = (float)((total_error_2 > 0) ? log(M_LN2 * (double)total_error_2 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[3] = (float)((total_error_3 > 0) ? log(M_LN2 * (double)total_error_3 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[4] = (float)((total_error_4 > 0) ? log(M_LN2 * (double)total_error_4 / (double)data_len) / M_LN2 : 0.0);
150
151
152
153
154

	return order;
}

FLAC__SSE_TARGET("sse2")
155
unsigned FLAC__fixed_compute_best_predictor_wide_intrin_sse2(const FLAC__int32 data[], unsigned data_len, float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER + 1])
156
157
158
159
160
161
162
163
164
{
	FLAC__uint64 total_error_0, total_error_1, total_error_2, total_error_3, total_error_4;
	unsigned i, order;

	__m128i total_err0, total_err1, total_err3;

	{
		FLAC__int32 itmp;
		__m128i last_error, zero = _mm_setzero_si128();
165

166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
		last_error = _mm_cvtsi32_si128(data[-1]);							// 0   0   0   le0
		itmp = data[-2];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// 0   0   le0 le1
		itmp -= data[-3];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// 0   le0 le1 le2
		itmp -= data[-3] - data[-4];
		last_error = _mm_shuffle_epi32(last_error, _MM_SHUFFLE(2,1,0,0));
		last_error = _mm_sub_epi32(last_error, _mm_cvtsi32_si128(itmp));	// le0 le1 le2 le3

		total_err0 = total_err1 = total_err3 = _mm_setzero_si128();
		for(i = 0; i < data_len; i++) {
			__m128i err0, err1, tmp;
			err0 = _mm_cvtsi32_si128(data[i]);								// 0   0   0   e0
			err1 = _mm_shuffle_epi32(err0, _MM_SHUFFLE(0,0,0,0));			// e0  e0  e0  e0
182
#if 1 /* OPT_SSE */
183
184
185
186
187
188
189
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   le0 le1 le2
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   0   le0 le1
			err1 = _mm_sub_epi32(err1, last_error);
			last_error = _mm_srli_si128(last_error, 4);						// 0   0   0   le0
			err1 = _mm_sub_epi32(err1, last_error);							// e1  e2  e3  e4
190
191
192
193
194
#else
			last_error = _mm_add_epi32(last_error, _mm_srli_si128(last_error, 8));	// le0  le1  le2+le0  le3+le1
			last_error = _mm_add_epi32(last_error, _mm_srli_si128(last_error, 4));	// le0  le1+le0  le2+le0+le1  le3+le1+le2+le0
			err1 = _mm_sub_epi32(err1, last_error);							// e1  e2  e3  e4
#endif
195
			tmp = _mm_slli_si128(err0, 12);									// e0   0   0   0
196
			last_error = _mm_srli_si128(err1, 4);							//  0  e1  e2  e3
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
			last_error = _mm_or_si128(last_error, tmp);						// e0  e1  e2  e3

			tmp = _mm_srai_epi32(err0, 31);
			err0 = _mm_xor_si128(err0, tmp);
			err0 = _mm_sub_epi32(err0, tmp);
			tmp = _mm_srai_epi32(err1, 31);
			err1 = _mm_xor_si128(err1, tmp);
			err1 = _mm_sub_epi32(err1, tmp);

			total_err0 = _mm_add_epi64(total_err0, err0);					//        0       te0
			err0 = _mm_unpacklo_epi32(err1, zero);							//   0  |e3|   0  |e4|
			err1 = _mm_unpackhi_epi32(err1, zero);							//   0  |e1|   0  |e2|
			total_err3 = _mm_add_epi64(total_err3, err0);					//       te3      te4
			total_err1 = _mm_add_epi64(total_err1, err1);					//       te1      te2
		}
	}
213

214
215
216
	m128i_to_i64(total_error_0, total_err0);
	m128i_to_i64(total_error_4, total_err3);
	m128i_to_i64(total_error_2, total_err1);
217
218
	total_err3 = _mm_srli_si128(total_err3,	8);							//         0      te3
	total_err1 = _mm_srli_si128(total_err1, 8);							//         0      te1
219
220
	m128i_to_i64(total_error_3, total_err3);
	m128i_to_i64(total_error_1, total_err1);
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242

	/* prefer higher order */
	if(total_error_0 < flac_min(flac_min(flac_min(total_error_1, total_error_2), total_error_3), total_error_4))
		order = 0;
	else if(total_error_1 < flac_min(flac_min(total_error_2, total_error_3), total_error_4))
		order = 1;
	else if(total_error_2 < flac_min(total_error_3, total_error_4))
		order = 2;
	else if(total_error_3 < total_error_4)
		order = 3;
	else
		order = 4;

	/* Estimate the expected number of bits per residual signal sample. */
	/* 'total_error*' is linearly related to the variance of the residual */
	/* signal, so we use it directly to compute E(|x|) */
	FLAC__ASSERT(data_len > 0 || total_error_0 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_1 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_2 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_3 == 0);
	FLAC__ASSERT(data_len > 0 || total_error_4 == 0);

243
244
245
246
247
	residual_bits_per_sample[0] = (float)((total_error_0 > 0) ? log(M_LN2 * (double)total_error_0 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[1] = (float)((total_error_1 > 0) ? log(M_LN2 * (double)total_error_1 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[2] = (float)((total_error_2 > 0) ? log(M_LN2 * (double)total_error_2 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[3] = (float)((total_error_3 > 0) ? log(M_LN2 * (double)total_error_3 / (double)data_len) / M_LN2 : 0.0);
	residual_bits_per_sample[4] = (float)((total_error_4 > 0) ? log(M_LN2 * (double)total_error_4 / (double)data_len) / M_LN2 : 0.0);
248
249
250
251
252

	return order;
}

#endif /* FLAC__SSE2_SUPPORTED */
253
#endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */
254
255
#endif /* FLAC__NO_ASM */
#endif /* FLAC__INTEGER_ONLY_LIBRARY */