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
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
|
;
; 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.
;
;
%include "third_party/x86inc/x86inc.asm"
SECTION .text
%macro QUANTIZE_FN 2
cglobal quantize_%1, 0, %2, 15, coeff, ncoeff, zbin, round, quant, \
shift, qcoeff, dqcoeff, dequant, \
eob, scan, iscan
vzeroupper
%ifnidn %1, b_32x32
; Special case for ncoeff == 16, as it is frequent and we can save on
; not setting up a loop.
cmp ncoeffmp, 16
jne .generic
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Special case of ncoeff == 16
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
.single:
movifnidn coeffq, coeffmp
movifnidn zbinq, zbinmp
mova m0, [zbinq] ; m0 = zbin
; Get DC and first 15 AC coeffs - in this special case, that is all.
; coeff stored as 32bit numbers but we process them as 16 bit numbers
mova m9, [coeffq]
packssdw m9, [coeffq+16] ; m9 = c[i]
mova m10, [coeffq+32]
packssdw m10, [coeffq+48] ; m10 = c[i]
mov r0, eobmp ; Output pointer
mov r1, qcoeffmp ; Output pointer
mov r2, dqcoeffmp ; Output pointer
pxor m5, m5 ; m5 = dedicated zero
pcmpeqw m4, m4 ; All word lanes -1
paddw m0, m4 ; m0 = zbin - 1
pabsw m6, m9 ; m6 = abs(m9)
pabsw m11, m10 ; m11 = abs(m10)
pcmpgtw m7, m6, m0 ; m7 = c[i] >= zbin
punpckhqdq m0, m0
pcmpgtw m12, m11, m0 ; m12 = c[i] >= zbin
; Check if all coeffs are less than zbin. If yes, we just write zeros
; to the outputs and we are done.
por m14, m7, m12
ptest m14, m14
jnz .single_nonzero
mova [r1 ], ymm5
mova [r1+32], ymm5
mova [r2 ], ymm5
mova [r2+32], ymm5
mov [r0], word 0
vzeroupper
RET
.single_nonzero:
; Actual quantization of size 16 block - setup pointers, rounders, etc.
movifnidn r3, roundmp
movifnidn r4, quantmp
mov r6, dequantmp
mov r5, shiftmp
mova m1, [r3] ; m1 = round
mova m2, [r4] ; m2 = quant
mova m3, [r6] ; m3 = dequant
mova m4, [r5] ; m4 = shift
mov r3, iscanmp
DEFINE_ARGS eob, qcoeff, dqcoeff, iscan
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
paddsw m6, m1 ; m6 += round
punpckhqdq m1, m1
paddsw m11, m1 ; m11 += round
pmulhw m8, m6, m2 ; m8 = m6*q>>16
punpckhqdq m2, m2
pmulhw m13, m11, m2 ; m13 = m11*q>>16
paddw m8, m6 ; m8 += m6
paddw m13, m11 ; m13 += m11
pmulhw m8, m4 ; m8 = m8*qsh>>16
punpckhqdq m4, m4
pmulhw m13, m4 ; m13 = m13*qsh>>16
psignw m8, m9 ; m8 = reinsert sign
psignw m13, m10 ; m13 = reinsert sign
pand m8, m7
pand m13, m12
; Store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m8
punpckhwd m6, m8, m6
pmovsxwd m11, m8
mova [qcoeffq ], m11
mova [qcoeffq+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [qcoeffq+32], m11
mova [qcoeffq+48], m6
pmullw m8, m3 ; dqc[i] = qc[i] * q
punpckhqdq m3, m3
pmullw m13, m3 ; dqc[i] = qc[i] * q
; Store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m8
punpckhwd m6, m8, m6
pmovsxwd m11, m8
mova [dqcoeffq ], m11
mova [dqcoeffq+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [dqcoeffq+32], m11
mova [dqcoeffq+48], m6
mova m6, [iscanq] ; m6 = scan[i]
mova m11, [iscanq+16] ; m11 = scan[i]
pcmpeqw m8, m8, m5 ; m8 = c[i] == 0
pcmpeqw m13, m13, m5 ; m13 = c[i] == 0
psubw m6, m6, m7 ; m6 = scan[i] + 1
psubw m11, m11, m12 ; m11 = scan[i] + 1
pandn m8, m8, m6 ; m8 = max(eob)
pandn m13, m13, m11 ; m13 = max(eob)
pmaxsw m8, m8, m13
; Horizontally accumulate/max eobs and write into [eob] memory pointer
pshufd m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0x1
pmaxsw m8, m7
movq rax, m8
mov [eobq], ax
vzeroupper
RET
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Generic case of ncoeff != 16
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
.generic:
%endif ; %ifnidn %1, b_32x32
DEFINE_ARGS coeff, ncoeff, zbin, round, quant, shift, \
qcoeff, dqcoeff, dequant, eob, scan, iscan
; Actual quantization loop - setup pointers, rounders, etc.
movifnidn coeffq, coeffmp
movifnidn ncoeffq, ncoeffmp
movifnidn zbinq, zbinmp
movifnidn roundq, roundmp
movifnidn quantq, quantmp
movifnidn dequantq, dequantmp
mova m0, [zbinq] ; m0 = zbin
mova m1, [roundq] ; m1 = round
mova m2, [quantq] ; m2 = quant
mova m3, [dequantq] ; m3 = dequant
pcmpeqw m4, m4 ; All lanes -1
%ifidn %1, b_32x32
psubw m0, m4
psubw m1, m4
psrlw m0, 1 ; m0 = (m0 + 1) / 2
psrlw m1, 1 ; m1 = (m1 + 1) / 2
%endif
paddw m0, m4 ; m0 = m0 + 1
mov r2, shiftmp
mov r3, qcoeffmp
mova m4, [r2] ; m4 = shift
mov r4, dqcoeffmp
mov r5, iscanmp
%ifidn %1, b_32x32
psllw m4, 1
%endif
pxor m5, m5 ; m5 = dedicated zero
DEFINE_ARGS coeff, ncoeff, d1, qcoeff, dqcoeff, iscan, d2, d3, d4, eob
lea coeffq, [ coeffq+ncoeffq*4]
lea qcoeffq, [ qcoeffq+ncoeffq*4]
lea dqcoeffq, [dqcoeffq+ncoeffq*4]
lea iscanq, [ iscanq+ncoeffq*2]
neg ncoeffq
; get DC and first 15 AC coeffs
; coeff stored as 32bit numbers & require 16bit numbers
mova m9, [coeffq+ncoeffq*4+ 0]
packssdw m9, [coeffq+ncoeffq*4+16]
mova m10, [coeffq+ncoeffq*4+32]
packssdw m10, [coeffq+ncoeffq*4+48]
pabsw m6, m9 ; m6 = abs(m9)
pabsw m11, m10 ; m11 = abs(m10)
pcmpgtw m7, m6, m0 ; m7 = c[i] >= zbin
punpckhqdq m0, m0
pcmpgtw m12, m11, m0 ; m12 = c[i] >= zbin
; Check if all coeffs are less than zbin. If yes, skip forward quickly.
por m14, m7, m12
ptest m14, m14
jnz .first_nonzero
mova [qcoeffq+ncoeffq*4 ], ymm5
mova [qcoeffq+ncoeffq*4+32], ymm5
mova [dqcoeffq+ncoeffq*4 ], ymm5
mova [dqcoeffq+ncoeffq*4+32], ymm5
add ncoeffq, mmsize
punpckhqdq m1, m1
punpckhqdq m2, m2
punpckhqdq m3, m3
punpckhqdq m4, m4
pxor m8, m8
jmp .ac_only_loop
.first_nonzero:
paddsw m6, m1 ; m6 += round
punpckhqdq m1, m1
paddsw m11, m1 ; m11 += round
pmulhw m8, m6, m2 ; m8 = m6*q>>16
punpckhqdq m2, m2
pmulhw m13, m11, m2 ; m13 = m11*q>>16
paddw m8, m6 ; m8 += m6
paddw m13, m11 ; m13 += m11
pmulhw m8, m4 ; m8 = m8*qsh>>16
punpckhqdq m4, m4
pmulhw m13, m4 ; m13 = m13*qsh>>16
psignw m8, m9 ; m8 = reinsert sign
psignw m13, m10 ; m13 = reinsert sign
pand m8, m7
pand m13, m12
; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m8
punpckhwd m6, m8, m6
pmovsxwd m11, m8
mova [qcoeffq+ncoeffq*4+ 0], m11
mova [qcoeffq+ncoeffq*4+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [qcoeffq+ncoeffq*4+32], m11
mova [qcoeffq+ncoeffq*4+48], m6
%ifidn %1, b_32x32
pabsw m8, m8
pabsw m13, m13
%endif
pmullw m8, m3 ; dqc[i] = qc[i] * q
punpckhqdq m3, m3
pmullw m13, m3 ; dqc[i] = qc[i] * q
%ifidn %1, b_32x32
psrlw m8, 1
psrlw m13, 1
psignw m8, m9
psignw m13, m10
%endif
; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m8
punpckhwd m6, m8, m6
pmovsxwd m11, m8
mova [dqcoeffq+ncoeffq*4+ 0], m11
mova [dqcoeffq+ncoeffq*4+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [dqcoeffq+ncoeffq*4+32], m11
mova [dqcoeffq+ncoeffq*4+48], m6
pcmpeqw m8, m5 ; m8 = c[i] == 0
pcmpeqw m13, m5 ; m13 = c[i] == 0
mova m6, [iscanq+ncoeffq*2] ; m6 = scan[i]
mova m11, [iscanq+ncoeffq*2+16] ; m11 = scan[i]
psubw m6, m7 ; m6 = scan[i] + 1
psubw m11, m12 ; m11 = scan[i] + 1
pandn m8, m6 ; m8 = max(eob)
pandn m13, m11 ; m13 = max(eob)
pmaxsw m8, m13
add ncoeffq, mmsize
.ac_only_loop:
; pack coeff from 32bit to 16bit array
mova m9, [coeffq+ncoeffq*4+ 0]
packssdw m9, [coeffq+ncoeffq*4+16]
mova m10, [coeffq+ncoeffq*4+32]
packssdw m10, [coeffq+ncoeffq*4+48]
pabsw m6, m9 ; m6 = abs(m9)
pabsw m11, m10 ; m11 = abs(m10)
pcmpgtw m7, m6, m0 ; m7 = c[i] >= zbin
pcmpgtw m12, m11, m0 ; m12 = c[i] >= zbin
; Check if all coeffs are less than zbin. If yes, skip this itertion.
; And just write zeros as the result would be.
por m14, m7, m12
ptest m14, m14
jnz .rest_nonzero
mova [qcoeffq+ncoeffq*4+ 0], ymm5
mova [qcoeffq+ncoeffq*4+32], ymm5
mova [dqcoeffq+ncoeffq*4+ 0], ymm5
mova [dqcoeffq+ncoeffq*4+32], ymm5
add ncoeffq, mmsize
jnz .ac_only_loop
; Horizontally accumulate/max eobs and write into [eob] memory pointer
mov r2, eobmp
pshufd m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0x1
pmaxsw m8, m7
movq rax, m8
mov [r2], ax
vzeroupper
RET
.rest_nonzero:
paddsw m6, m1 ; m6 += round
paddsw m11, m1 ; m11 += round
pmulhw m14, m6, m2 ; m14 = m6*q>>16
pmulhw m13, m11, m2 ; m13 = m11*q>>16
paddw m14, m6 ; m14 += m6
paddw m13, m11 ; m13 += m11
pmulhw m14, m4 ; m14 = m14*qsh>>16
pmulhw m13, m4 ; m13 = m13*qsh>>16
psignw m14, m9 ; m14 = reinsert sign
psignw m13, m10 ; m13 = reinsert sign
pand m14, m7
pand m13, m12
; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m14
punpckhwd m6, m14, m6
pmovsxwd m11, m14
mova [qcoeffq+ncoeffq*4+ 0], m11
mova [qcoeffq+ncoeffq*4+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [qcoeffq+ncoeffq*4+32], m11
mova [qcoeffq+ncoeffq*4+48], m6
%ifidn %1, b_32x32
pabsw m14, m14
pabsw m13, m13
%endif
pmullw m14, m3 ; dqc[i] = qc[i] * q
pmullw m13, m3 ; dqc[i] = qc[i] * q
%ifidn %1, b_32x32
psrlw m14, 1
psrlw m13, 1
psignw m14, m9
psignw m13, m10
%endif
; store 16bit numbers as 32bit numbers in array pointed to by qcoeff
pcmpgtw m6, m5, m14
punpckhwd m6, m14, m6
pmovsxwd m11, m14
mova [dqcoeffq+ncoeffq*4+ 0], m11
mova [dqcoeffq+ncoeffq*4+16], m6
pcmpgtw m6, m5, m13
punpckhwd m6, m13, m6
pmovsxwd m11, m13
mova [dqcoeffq+ncoeffq*4+32], m11
mova [dqcoeffq+ncoeffq*4+48], m6
pcmpeqw m14, m5 ; m14 = c[i] == 0
pcmpeqw m13, m5 ; m13 = c[i] == 0
mova m6, [iscanq+ncoeffq*2+ 0] ; m6 = scan[i]
mova m11, [iscanq+ncoeffq*2+16] ; m11 = scan[i]
psubw m6, m7 ; m6 = scan[i] + 1
psubw m11, m12 ; m11 = scan[i] + 1
pandn m14, m6 ; m14 = max(eob)
pandn m13, m11 ; m13 = max(eob)
pmaxsw m8, m14
pmaxsw m8, m13
add ncoeffq, mmsize
jnz .ac_only_loop
; Horizontally accumulate/max eobs and write into [eob] memory pointer
mov r2, eobmp
pshufd m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0xe
pmaxsw m8, m7
pshuflw m7, m8, 0x1
pmaxsw m8, m7
movq rax, m8
mov [r2], ax
vzeroupper
RET
%endmacro
INIT_XMM avx
QUANTIZE_FN b, 9
QUANTIZE_FN b_32x32, 9
|
