-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcdef_rav1e.asm
470 lines (430 loc) · 15.8 KB
/
cdef_rav1e.asm
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
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
; Copyright © 2018, VideoLAN and dav1d authors
; Copyright © 2018, Two Orioles, LLC
; All rights reserved.
;
; Redistribution and use in source and binary forms, with or without
; modification, are permitted provided that the following conditions are met:
;
; 1. Redistributions of source code must retain the above copyright notice, this
; list of conditions and the following disclaimer.
;
; 2. 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.
;
; 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 COPYRIGHT OWNER 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.
%include "config.asm"
%include "ext/x86/x86inc.asm"
%if ARCH_X86_64
SECTION_RODATA 32
pd_47130256: dd 4, 7, 1, 3, 0, 2, 5, 6
div_table: dd 840, 420, 280, 210, 168, 140, 120, 105
dd 420, 210, 140, 105
shufw_6543210x: db 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1, 14, 15
shufb_lohi: db 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15
tap_table: ; masks for 8 bit shifts
db 0xFF, 0x7F, 0x3F, 0x1F, 0x0F, 0x07, 0x03, 0x01
; weights
db 4, 2, 3, 3, 2, 1
db 0, 0 ; padding
db -1, -2, 0, 0, 1, 2, 0, 0
db 0, -1, 0, 1, 1, 2, 0, 0
db 0, 0, 1, 2, -1, -2, 0, 0
db 0, 1, 1, 2, 0, -1, 0, 0
db 1, 2, 1, 2, 0, 0, 0, 0
db 1, 2, 1, 2, 0, 1, 0, 0
db 1, 2, -1, -2, 1, 2, 0, 0
db 1, 2, 0, -1, 1, 2, 0, 0
db 1, 2, 1, 2, 0, 0, 0, 0
db 1, 2, 1, 2, 0, -1, 0, 0
db 1, 2, 1, 2, 1, 2, 0, 0
db 1, 2, 0, 1, 1, 2, 0, 0
db 1, 2, 0, 0, 1, 2, 0, 0
db 0, 1, 0, -1, 1, 2, 0, 0
db 0, 0, 1, 2, 1, 2, 0, 0
db 0, -1, 1, 2, 0, 1, 0, 0
pw_128: times 2 dw 128
pw_2048: times 2 dw 2048
SECTION .text
; stride unused
%macro ACCUMULATE_TAP 7 ; tap_offset, shift, mask, strength, mul_tap, w, stride
; load p0/p1
movsx offq, dword [offsets+kq*4+%1] ; off1
%if %6 == 4
movq xm5, [tmp0q+offq*2] ; p0
movq xm6, [tmp2q+offq*2]
movhps xm5, [tmp1q+offq*2]
movhps xm6, [tmp3q+offq*2]
vinserti128 m5, xm6, 1
%else
movu xm5, [tmp0q+offq*2] ; p0
vinserti128 m5, [tmp1q+offq*2], 1
%endif
neg offq ; -off1
%if %6 == 4
movq xm6, [tmp0q+offq*2] ; p1
movq xm9, [tmp2q+offq*2]
movhps xm6, [tmp1q+offq*2]
movhps xm9, [tmp3q+offq*2]
vinserti128 m6, xm9, 1
%else
movu xm6, [tmp0q+offq*2] ; p1
vinserti128 m6, [tmp1q+offq*2], 1
%endif
; out of bounds values are set to a value that is a both a large unsigned
; value and a negative signed value.
; use signed max and unsigned min to remove them
pmaxsw m7, m5 ; max after p0
pminuw m8, m5 ; min after p0
pmaxsw m7, m6 ; max after p1
pminuw m8, m6 ; min after p1
; accumulate sum[m15] over p0/p1
; calculate difference before converting
psubw m5, m4 ; diff_p0(p0 - px)
psubw m6, m4 ; diff_p1(p1 - px)
; convert to 8-bits with signed saturation
; saturating to large diffs has no impact on the results
packsswb m5, m6
; group into pairs so we can accumulate using maddubsw
pshufb m5, m12
pabsb m9, m5
psignb m10, %5, m5
psrlw m5, m9, %2 ; emulate 8-bit shift
pand m5, %3
psubusb m5, %4, m5
; use unsigned min since abs diff can equal 0x80
pminub m5, m9
pmaddubsw m5, m10
paddw m15, m5
%endmacro
%macro CDEF_FILTER 2 ; w, h
INIT_YMM avx2
%if %1*%2*2/mmsize > 1
%if %1 == 4
cglobal cdef_filter_%1x%2, 4, 13, 16, 64
%else
cglobal cdef_filter_%1x%2, 4, 11, 16, 64
%endif
%else
cglobal cdef_filter_%1x%2, 4, 12, 16, 64
%endif
%define offsets rsp+32
DEFINE_ARGS dst, dst_stride, tmp, tmp_stride, pri, sec, pridmp, table, \
secdmp, damping, dir
lea tableq, [tap_table]
; off1/2/3[k] [6 total]
movd xm2, tmp_strided
vpbroadcastd m2, xm2
mov dird, r6m
pmovsxbd m3, [tableq+dirq*8+16]
pmulld m2, m3
pmovsxbd m4, [tableq+dirq*8+16+64]
paddd m2, m4
mova [offsets], m2
; register to shuffle values into after packing
vbroadcasti128 m12, [shufb_lohi]
movifnidn prid, prim
mov dampingd, r7m
lzcnt pridmpd, prid
%if UNIX64
movd xm0, prid
movd xm1, secd
%endif
lzcnt secdmpd, secm
sub dampingd, 31
DEFINE_ARGS dst, dst_stride, tmp, tmp_stride, pri, sec, pridmp, table, \
secdmp, damping, zero
xor zerod, zerod
add pridmpd, dampingd
cmovl pridmpd, zerod
add secdmpd, dampingd
cmovl secdmpd, zerod
mov [rsp+0], pridmpq ; pri_shift
mov [rsp+8], secdmpq ; sec_shift
DEFINE_ARGS dst, dst_stride, tmp, tmp_stride, pri, sec, pridmp, table, \
secdmp
vpbroadcastb m13, [tableq+pridmpq] ; pri_shift_mask
vpbroadcastb m14, [tableq+secdmpq] ; sec_shift_mask
; pri/sec_taps[k] [4 total]
DEFINE_ARGS dst, dst_stride, tmp, tmp_stride, pri, sec, dummy, table, \
secdmp
%if UNIX64
vpbroadcastb m0, xm0 ; pri_strength
vpbroadcastb m1, xm1 ; sec_strength
%else
vpbroadcastb m0, prim
vpbroadcastb m1, secm
%endif
and prid, 1
lea priq, [tableq+priq*2+8] ; pri_taps
lea secq, [tableq+12] ; sec_taps
%if %1*%2*2/mmsize > 1
%if %1 == 4
DEFINE_ARGS dst, dst_stride, tmp0, tmp_stride, pri, sec, dst_stride3, h, off, k, tmp1, tmp2, tmp3
lea dst_stride3q, [dst_strideq*3]
%else
DEFINE_ARGS dst, dst_stride, tmp0, tmp_stride, pri, sec, h, off, k, tmp1
%endif
mov hd, %1*%2*2/mmsize
%else
DEFINE_ARGS dst, dst_stride, tmp0, tmp_stride, pri, sec, dst_stride3, off, k, tmp1, tmp2, tmp3
lea dst_stride3q, [dst_strideq*3]
%endif
pxor m11, m11
%if %1*%2*2/mmsize > 1
.v_loop:
%endif
lea tmp1q, [tmp0q+tmp_strideq*2]
%if %1 == 4
lea tmp2q, [tmp0q+tmp_strideq*4]
lea tmp3q, [tmp1q+tmp_strideq*4]
%endif
mov kd, 1
%if %1 == 4
movq xm4, [tmp0q]
movhps xm4, [tmp1q]
movq xm5, [tmp2q]
movhps xm5, [tmp3q]
vinserti128 m4, xm5, 1
%else
mova xm4, [tmp0q] ; px
vinserti128 m4, [tmp1q], 1
%endif
pxor m15, m15 ; sum
mova m7, m4 ; max
mova m8, m4 ; min
.k_loop:
vpbroadcastb m2, [priq+kq] ; pri_taps
vpbroadcastb m3, [secq+kq] ; sec_taps
ACCUMULATE_TAP 0*8, [rsp+0], m13, m0, m2, %1, %3
ACCUMULATE_TAP 1*8, [rsp+8], m14, m1, m3, %1, %3
ACCUMULATE_TAP 2*8, [rsp+8], m14, m1, m3, %1, %3
dec kq
jge .k_loop
vpbroadcastd m10, [pw_2048]
pcmpgtw m9, m11, m15
paddw m15, m9
pmulhrsw m15, m10
paddw m4, m15
pminsw m4, m7
pmaxsw m4, m8
packuswb m4, m4
vextracti128 xm5, m4, 1
%if %1 == 4
movd [dstq+dst_strideq*0], xm4
pextrd [dstq+dst_strideq*1], xm4, 1
movd [dstq+dst_strideq*2], xm5
pextrd [dstq+dst_stride3q], xm5, 1
%else
movq [dstq+dst_strideq*0], xm4
movq [dstq+dst_strideq*1], xm5
%endif
%if %1*%2*2/mmsize > 1
%define vloop_lines (mmsize/(%1*2))
lea dstq, [dstq+dst_strideq*vloop_lines]
lea tmp0q, [tmp0q+tmp_strideq*2*vloop_lines]
dec hd
jg .v_loop
%endif
RET
%endmacro
CDEF_FILTER 8, 8
CDEF_FILTER 4, 8
CDEF_FILTER 4, 4
INIT_YMM avx2
cglobal cdef_dir_8bpc, 3, 4, 15, src, stride, var, stride3
lea stride3q, [strideq*3]
movq xm0, [srcq+strideq*0]
movq xm1, [srcq+strideq*1]
movq xm2, [srcq+strideq*2]
movq xm3, [srcq+stride3q]
lea srcq, [srcq+strideq*4]
vpbroadcastq m4, [srcq+strideq*0]
vpbroadcastq m5, [srcq+strideq*1]
vpbroadcastq m6, [srcq+strideq*2]
vpbroadcastq m7, [srcq+stride3q]
vpbroadcastd m8, [pw_128]
pxor m9, m9
vpblendd m0, m0, m7, 0xf0
vpblendd m1, m1, m6, 0xf0
vpblendd m2, m2, m5, 0xf0
vpblendd m3, m3, m4, 0xf0
punpcklbw m0, m9
punpcklbw m1, m9
punpcklbw m2, m9
punpcklbw m3, m9
psubw m0, m8
psubw m1, m8
psubw m2, m8
psubw m3, m8
; shuffle registers to generate partial_sum_diag[0-1] together
vpermq m7, m0, q1032
vpermq m6, m1, q1032
vpermq m5, m2, q1032
vpermq m4, m3, q1032
; start with partial_sum_hv[0-1]
paddw m8, m0, m1
paddw m9, m2, m3
phaddw m10, m0, m1
phaddw m11, m2, m3
paddw m8, m9
phaddw m10, m11
vextracti128 xm9, m8, 1
vextracti128 xm11, m10, 1
paddw xm8, xm9 ; partial_sum_hv[1]
phaddw xm10, xm11 ; partial_sum_hv[0]
vinserti128 m8, xm10, 1
vpbroadcastd m9, [div_table+44]
pmaddwd m8, m8
pmulld m8, m9 ; cost6[2a-d] | cost2[a-d]
; create aggregates [lower half]:
; m9 = m0:01234567+m1:x0123456+m2:xx012345+m3:xxx01234+
; m4:xxxx0123+m5:xxxxx012+m6:xxxxxx01+m7:xxxxxxx0
; m10= m1:7xxxxxxx+m2:67xxxxxx+m3:567xxxxx+
; m4:4567xxxx+m5:34567xxx+m6:234567xx+m7:1234567x
; and [upper half]:
; m9 = m0:xxxxxxx0+m1:xxxxxx01+m2:xxxxx012+m3:xxxx0123+
; m4:xxx01234+m5:xx012345+m6:x0123456+m7:01234567
; m10= m0:1234567x+m1:234567xx+m2:34567xxx+m3:4567xxxx+
; m4:567xxxxx+m5:67xxxxxx+m6:7xxxxxxx
; and then shuffle m11 [shufw_6543210x], unpcklwd, pmaddwd, pmulld, paddd
pslldq m9, m1, 2
psrldq m10, m1, 14
pslldq m11, m2, 4
psrldq m12, m2, 12
pslldq m13, m3, 6
psrldq m14, m3, 10
paddw m9, m11
paddw m10, m12
paddw m9, m13
paddw m10, m14
pslldq m11, m4, 8
psrldq m12, m4, 8
pslldq m13, m5, 10
psrldq m14, m5, 6
paddw m9, m11
paddw m10, m12
paddw m9, m13
paddw m10, m14
pslldq m11, m6, 12
psrldq m12, m6, 4
pslldq m13, m7, 14
psrldq m14, m7, 2
paddw m9, m11
paddw m10, m12
paddw m9, m13
paddw m10, m14 ; partial_sum_diag[0/1][8-14,zero]
vbroadcasti128 m14, [shufw_6543210x]
vbroadcasti128 m13, [div_table+16]
vbroadcasti128 m12, [div_table+0]
paddw m9, m0 ; partial_sum_diag[0/1][0-7]
pshufb m10, m14
punpckhwd m11, m9, m10
punpcklwd m9, m10
pmaddwd m11, m11
pmaddwd m9, m9
pmulld m11, m13
pmulld m9, m12
paddd m9, m11 ; cost0[a-d] | cost4[a-d]
; merge horizontally and vertically for partial_sum_alt[0-3]
paddw m10, m0, m1
paddw m11, m2, m3
paddw m12, m4, m5
paddw m13, m6, m7
phaddw m0, m4
phaddw m1, m5
phaddw m2, m6
phaddw m3, m7
; create aggregates [lower half]:
; m4 = m10:01234567+m11:x0123456+m12:xx012345+m13:xxx01234
; m11= m11:7xxxxxxx+m12:67xxxxxx+m13:567xxxxx
; and [upper half]:
; m4 = m10:xxx01234+m11:xx012345+m12:x0123456+m13:01234567
; m11= m10:567xxxxx+m11:67xxxxxx+m12:7xxxxxxx
; and then pshuflw m11 3012, unpcklwd, pmaddwd, pmulld, paddd
pslldq m4, m11, 2
psrldq m11, 14
pslldq m5, m12, 4
psrldq m12, 12
pslldq m6, m13, 6
psrldq m13, 10
paddw m4, m10
paddw m11, m12
vpbroadcastd m12, [div_table+44]
paddw m5, m6
paddw m11, m13 ; partial_sum_alt[3/2] right
vbroadcasti128 m13, [div_table+32]
paddw m4, m5 ; partial_sum_alt[3/2] left
pshuflw m5, m11, q3012
punpckhwd m6, m11, m4
punpcklwd m4, m5
pmaddwd m6, m6
pmaddwd m4, m4
pmulld m6, m12
pmulld m4, m13
paddd m4, m6 ; cost7[a-d] | cost5[a-d]
; create aggregates [lower half]:
; m5 = m0:01234567+m1:x0123456+m2:xx012345+m3:xxx01234
; m1 = m1:7xxxxxxx+m2:67xxxxxx+m3:567xxxxx
; and [upper half]:
; m5 = m0:xxx01234+m1:xx012345+m2:x0123456+m3:01234567
; m1 = m0:567xxxxx+m1:67xxxxxx+m2:7xxxxxxx
; and then pshuflw m1 3012, unpcklwd, pmaddwd, pmulld, paddd
pslldq m5, m1, 2
psrldq m1, 14
pslldq m6, m2, 4
psrldq m2, 12
pslldq m7, m3, 6
psrldq m3, 10
paddw m5, m0
paddw m1, m2
paddw m6, m7
paddw m1, m3 ; partial_sum_alt[0/1] right
paddw m5, m6 ; partial_sum_alt[0/1] left
pshuflw m0, m1, q3012
punpckhwd m1, m5
punpcklwd m5, m0
pmaddwd m1, m1
pmaddwd m5, m5
pmulld m1, m12
pmulld m5, m13
paddd m5, m1 ; cost1[a-d] | cost3[a-d]
mova xm0, [pd_47130256+ 16]
mova m1, [pd_47130256]
phaddd m9, m8
phaddd m5, m4
phaddd m9, m5
vpermd m0, m9 ; cost[0-3]
vpermd m1, m9 ; cost[4-7] | cost[0-3]
; now find the best cost
pmaxsd xm2, xm0, xm1
pshufd xm3, xm2, q1032
pmaxsd xm2, xm3
pshufd xm3, xm2, q2301
pmaxsd xm2, xm3 ; best cost
; find the idx using minpos
; make everything other than the best cost negative via subtraction
; find the min of unsigned 16-bit ints to sort out the negative values
psubd xm4, xm1, xm2
psubd xm3, xm0, xm2
packssdw xm3, xm4
phminposuw xm3, xm3
; convert idx to 32-bits
psrld xm3, 16
movd eax, xm3
; get idx^4 complement
vpermd m3, m1
psubd xm2, xm3
psrld xm2, 10
movd [varq], xm2
RET
%endif ; ARCH_X86_64