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
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
|
/*
* Copyright (c) 2017 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include <tmmintrin.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
void vpx_quantize_b_ssse3(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan_ptr, const int16_t *iscan_ptr) {
const __m128i zero = _mm_setzero_si128();
intptr_t index = 16;
__m128i zbin, round, quant, dequant, shift;
__m128i coeff0, coeff1;
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
__m128i qtmp0, qtmp1;
__m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
__m128i eob, eob0, eob1;
(void)scan_ptr;
(void)skip_block;
assert(!skip_block);
// Setup global values.
zbin = _mm_load_si128((const __m128i *)zbin_ptr);
// x86 has no "greater *or equal*" comparison. Subtract 1 from zbin so
// it is a strict "greater" comparison.
zbin = _mm_sub_epi16(zbin, _mm_set1_epi16(1));
round = _mm_load_si128((const __m128i *)round_ptr);
quant = _mm_load_si128((const __m128i *)quant_ptr);
dequant = _mm_load_si128((const __m128i *)dequant_ptr);
shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
// Do DC and first 15 AC.
coeff0 = load_tran_low(coeff_ptr);
coeff1 = load_tran_low(coeff_ptr + 8);
qcoeff0 = _mm_abs_epi16(coeff0);
qcoeff1 = _mm_abs_epi16(coeff1);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
round = _mm_unpackhi_epi64(round, round);
qcoeff1 = _mm_adds_epi16(qcoeff1, round);
qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
// Reinsert signs
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
// Mask out zbin threshold coeffs
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
store_tran_low(qcoeff0, qcoeff_ptr);
store_tran_low(qcoeff1, qcoeff_ptr + 8);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
// Scan for eob.
zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
eob = _mm_andnot_si128(zero_coeff0, iscan0);
eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
eob = _mm_max_epi16(eob, eob1);
// AC only loop.
while (index < n_coeffs) {
coeff0 = load_tran_low(coeff_ptr + index);
coeff1 = load_tran_low(coeff_ptr + index + 8);
qcoeff0 = _mm_abs_epi16(coeff0);
qcoeff1 = _mm_abs_epi16(coeff1);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
qcoeff1 = _mm_adds_epi16(qcoeff1, round);
qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
store_tran_low(qcoeff0, qcoeff_ptr + index);
store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
eob0 = _mm_max_epi16(eob0, eob1);
eob = _mm_max_epi16(eob, eob0);
index += 16;
}
// Accumulate eob.
{
__m128i eob_shuffled;
eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
eob = _mm_max_epi16(eob, eob_shuffled);
eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
eob = _mm_max_epi16(eob, eob_shuffled);
eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
eob = _mm_max_epi16(eob, eob_shuffled);
*eob_ptr = _mm_extract_epi16(eob, 1);
}
}
void vpx_quantize_b_32x32_ssse3(
const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan_ptr, const int16_t *iscan_ptr) {
const __m128i zero = _mm_setzero_si128();
const __m128i one = _mm_set1_epi16(1);
intptr_t index = 16;
__m128i zbin, round, quant, dequant, shift;
__m128i coeff0, coeff1;
__m128i qcoeff0, qcoeff1;
__m128i cmp_mask0, cmp_mask1;
__m128i all_zero;
__m128i qtmp0, qtmp1;
__m128i zero_coeff0, zero_coeff1, iscan0, iscan1;
__m128i eob = zero, eob0, eob1;
(void)scan_ptr;
(void)n_coeffs;
(void)skip_block;
assert(!skip_block);
// Setup global values.
// The 32x32 halves zbin and round.
zbin = _mm_load_si128((const __m128i *)zbin_ptr);
// Shift with rounding.
zbin = _mm_add_epi16(zbin, one);
zbin = _mm_srli_epi16(zbin, 1);
// x86 has no "greater *or equal*" comparison. Subtract 1 from zbin so
// it is a strict "greater" comparison.
zbin = _mm_sub_epi16(zbin, one);
round = _mm_load_si128((const __m128i *)round_ptr);
round = _mm_add_epi16(round, one);
round = _mm_srli_epi16(round, 1);
quant = _mm_load_si128((const __m128i *)quant_ptr);
dequant = _mm_load_si128((const __m128i *)dequant_ptr);
shift = _mm_load_si128((const __m128i *)quant_shift_ptr);
// I suspect this is not technically OK because quant_shift can be up
// to 1 << 16 and shifting up again will outrange that, but the test is not
// comprehensive enough to catch that and "it's been that way forever"
shift = _mm_slli_epi16(shift, 1);
// Do DC and first 15 AC.
coeff0 = load_tran_low(coeff_ptr);
coeff1 = load_tran_low(coeff_ptr + 8);
qcoeff0 = _mm_abs_epi16(coeff0);
qcoeff1 = _mm_abs_epi16(coeff1);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC.
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
all_zero = _mm_or_si128(cmp_mask0, cmp_mask1);
if (_mm_movemask_epi8(all_zero) == 0) {
_mm_store_si128((__m128i *)(qcoeff_ptr), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + 8), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + 8), zero);
#if CONFIG_VP9_HIGHBITDEPTH
_mm_store_si128((__m128i *)(qcoeff_ptr + 4), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + 12), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + 4), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + 12), zero);
#endif
round = _mm_unpackhi_epi64(round, round);
quant = _mm_unpackhi_epi64(quant, quant);
shift = _mm_unpackhi_epi64(shift, shift);
dequant = _mm_unpackhi_epi64(dequant, dequant);
} else {
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
round = _mm_unpackhi_epi64(round, round);
qcoeff1 = _mm_adds_epi16(qcoeff1, round);
qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
quant = _mm_unpackhi_epi64(quant, quant);
qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
shift = _mm_unpackhi_epi64(shift, shift);
qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
// Reinsert signs.
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
// Mask out zbin threshold coeffs.
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
store_tran_low(qcoeff0, qcoeff_ptr);
store_tran_low(qcoeff1, qcoeff_ptr + 8);
// Un-sign to bias rounding like C.
// dequant is almost always negative, so this is probably the backwards way
// to handle the sign. However, it matches the previous assembly.
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
coeff0 = _mm_mullo_epi16(coeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
coeff1 = _mm_mullo_epi16(coeff1, dequant);
// "Divide" by 2.
coeff0 = _mm_srli_epi16(coeff0, 1);
coeff1 = _mm_srli_epi16(coeff1, 1);
coeff0 = _mm_sign_epi16(coeff0, qcoeff0);
coeff1 = _mm_sign_epi16(coeff1, qcoeff1);
store_tran_low(coeff0, dqcoeff_ptr);
store_tran_low(coeff1, dqcoeff_ptr + 8);
// Scan for eob.
zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + 8));
// Add one to convert from indices to counts.
iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
eob = _mm_andnot_si128(zero_coeff0, iscan0);
eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
eob = _mm_max_epi16(eob, eob1);
}
// AC only loop.
for (index = 16; index < 32 * 32; index += 16) {
coeff0 = load_tran_low(coeff_ptr + index);
coeff1 = load_tran_low(coeff_ptr + index + 8);
qcoeff0 = _mm_abs_epi16(coeff0);
qcoeff1 = _mm_abs_epi16(coeff1);
cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
all_zero = _mm_or_si128(cmp_mask0, cmp_mask1);
if (_mm_movemask_epi8(all_zero) == 0) {
_mm_store_si128((__m128i *)(qcoeff_ptr + index), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + index + 8), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index + 8), zero);
#if CONFIG_VP9_HIGHBITDEPTH
_mm_store_si128((__m128i *)(qcoeff_ptr + index + 4), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + index + 12), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index + 4), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + index + 12), zero);
#endif
continue;
}
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
qcoeff1 = _mm_adds_epi16(qcoeff1, round);
qtmp0 = _mm_mulhi_epi16(qcoeff0, quant);
qtmp1 = _mm_mulhi_epi16(qcoeff1, quant);
qtmp0 = _mm_add_epi16(qtmp0, qcoeff0);
qtmp1 = _mm_add_epi16(qtmp1, qcoeff1);
qcoeff0 = _mm_mulhi_epi16(qtmp0, shift);
qcoeff1 = _mm_mulhi_epi16(qtmp1, shift);
qcoeff0 = _mm_sign_epi16(qcoeff0, coeff0);
qcoeff1 = _mm_sign_epi16(qcoeff1, coeff1);
qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
store_tran_low(qcoeff0, qcoeff_ptr + index);
store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
coeff0 = _mm_abs_epi16(qcoeff0);
coeff1 = _mm_abs_epi16(qcoeff1);
coeff0 = _mm_mullo_epi16(coeff0, dequant);
coeff1 = _mm_mullo_epi16(coeff1, dequant);
coeff0 = _mm_srli_epi16(coeff0, 1);
coeff1 = _mm_srli_epi16(coeff1, 1);
coeff0 = _mm_sign_epi16(coeff0, qcoeff0);
coeff1 = _mm_sign_epi16(coeff1, qcoeff1);
store_tran_low(coeff0, dqcoeff_ptr + index);
store_tran_low(coeff1, dqcoeff_ptr + index + 8);
zero_coeff0 = _mm_cmpeq_epi16(coeff0, zero);
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + index));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + index + 8));
iscan0 = _mm_sub_epi16(iscan0, cmp_mask0);
iscan1 = _mm_sub_epi16(iscan1, cmp_mask1);
eob0 = _mm_andnot_si128(zero_coeff0, iscan0);
eob1 = _mm_andnot_si128(zero_coeff1, iscan1);
eob0 = _mm_max_epi16(eob0, eob1);
eob = _mm_max_epi16(eob, eob0);
}
{
__m128i eob_shuffled;
eob_shuffled = _mm_shuffle_epi32(eob, 0xe);
eob = _mm_max_epi16(eob, eob_shuffled);
eob_shuffled = _mm_shufflelo_epi16(eob, 0xe);
eob = _mm_max_epi16(eob, eob_shuffled);
eob_shuffled = _mm_shufflelo_epi16(eob, 0x1);
eob = _mm_max_epi16(eob, eob_shuffled);
*eob_ptr = _mm_extract_epi16(eob, 1);
}
}
|
