Debian Codegraph @

.cargo/vendor/av-scenechange/src/asm/x86/ipred_sse.asm

raw
1 ; Copyright © 2018-2021, VideoLAN and dav1d authors
2 ; Copyright © 2018, Two Orioles, LLC
3 ; All rights reserved.
4 ;
5 ; Redistribution and use in source and binary forms, with or without
6 ; modification, are permitted provided that the following conditions are met:
7 ;
8 ; 1. Redistributions of source code must retain the above copyright notice, this
9 ; list of conditions and the following disclaimer.
10 ;
11 ; 2. Redistributions in binary form must reproduce the above copyright notice,
12 ; this list of conditions and the following disclaimer in the documentation
13 ; and/or other materials provided with the distribution.
14 ;
15 ; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
16 ; ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
17 ; WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
18 ; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
19 ; ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
20 ; (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
21 ; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
22 ; ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 ; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24 ; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25
26 %include "config.asm"
27 %include "src/asm/x86/x86inc.asm"
28
29 SECTION_RODATA 16
30
31 %macro SMOOTH_WEIGHT_TABLE 1-*
32 %rep %0
33 db %1-128, 127-%1
34 %rotate 1
35 %endrep
36 %endmacro
37
38 ; sm_weights[], but modified to precalculate x and 256-x with offsets to
39 ; enable efficient use of pmaddubsw (which requires signed values)
40 smooth_weights: SMOOTH_WEIGHT_TABLE \
41 0, 0, 255, 128, 255, 149, 85, 64, \
42 255, 197, 146, 105, 73, 50, 37, 32, \
43 255, 225, 196, 170, 145, 123, 102, 84, \
44 68, 54, 43, 33, 26, 20, 17, 16, \
45 255, 240, 225, 210, 196, 182, 169, 157, \
46 145, 133, 122, 111, 101, 92, 83, 74, \
47 66, 59, 52, 45, 39, 34, 29, 25, \
48 21, 17, 14, 12, 10, 9, 8, 8, \
49 255, 248, 240, 233, 225, 218, 210, 203, \
50 196, 189, 182, 176, 169, 163, 156, 150, \
51 144, 138, 133, 127, 121, 116, 111, 106, \
52 101, 96, 91, 86, 82, 77, 73, 69, \
53 65, 61, 57, 54, 50, 47, 44, 41, \
54 38, 35, 32, 29, 27, 25, 22, 20, \
55 18, 16, 15, 13, 12, 10, 9, 8, \
56 7, 6, 6, 5, 5, 4, 4, 4
57
58 ipred_v_shuf: db 0, 1, 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7
59 ipred_h_shuf: db 3, 3, 3, 3, 2, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0
60 ipred_paeth_shuf: db 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0
61 z_upsample1: db 1, 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7
62 z_upsample2: db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 8, 8, 8
63 z_transpose4: db 8, 12, 0, 4, 9, 13, 1, 5, 10, 14, 2, 6, 11, 15, 3, 7
64 z3_shuf: db 1, 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7
65 z3_shuf_h4: db 4, 3, 3, 2, 2, 1, 1, 0, 12, 11, 11, 10, 10, 9, 9, 8
66 filter_shuf1: db 3, 4, 3, 4, 5, 6, 5, 6, 7, 2, 7, 2, 1, -1, 1, -1
67 filter_shuf2: db 3, 4, 3, 4, 5, 6, 5, 6, 7, 11, 7, 11, 15, -1, 15, -1
68 z_filter_wh4: db 7, 7, 19, 7,
69 z_filter_wh8: db 19, 19, 11, 19, 11, 15, 15, 15, 23, 23, 23, 23, 39, 39, 39, 39
70 pd_32768: dd 32768
71 z3_filter_k_tail: db 64, 0, 64, 0, 64, 0, 56, 8
72 z1_shuf_w4: db 0, 1, 1, 2, 2, 3, 3, 4, 8, 9, 9, 10, 10, 11, 11, 12
73 pb_0to15: db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
74 pb_15to0: db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
75 z_base_inc: dw 0*64, 1*64, 2*64, 3*64, 4*64, 5*64, 6*64, 7*64
76 z3_base_inc: dw 7*64, 6*64, 5*64, 4*64, 3*64, 2*64, 1*64, 0*64
77 z_filter_wh16: db 19, 19, 19, 23, 23, 23, 31, 31, 31, 47, 47, 47, 79, 79, 79, -1
78 z_filter_t_w48: db 55,127, 7,127, 15, 31, 39, 31,127, 39,127, 39, 7, 15, 31, 15
79 db 39, 63, 3, 63, 3, 3, 19, 3, 47, 19, 47, 19, 3, 3, 3, 3
80 z_filter_t_w16: db 15, 31, 7, 15, 31, 7, 3, 31, 3, 3, 3, 3, 3, 3, 0, 0
81 z_filter_s: db 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7
82 db 7, 8, 8, 9, 9, 10, 10, 11
83 z_filter_k_tail: db 0, 64, 0, 64, 8, 56, 0, 64
84 z2_h_shuf: db 7, 6, 15, 14, 6, 5, 14, 13, 5, 4, 13, 12, 4, 3, 12, 11
85 z2_upsample: db 7, 6, 15, 14, 5, 4, 13, 12, 3, 2, 11, 10, 1, 0, 9, 8
86 z2_dy_offset: dw 88*64, 88*64, 87*64, 87*64
87 pw_m1to4: dw -1, -2, -3, -4
88 z_filter_k: times 4 db 0, 16
89 times 4 db 0, 20
90 times 4 db 8, 16
91 times 4 db 32, 16
92 times 4 db 24, 20
93 times 4 db 16, 16
94 times 4 db 0, 0
95 times 4 db 0, 0
96 pw_8: times 8 db 8, 0
97 pb_3: times 16 db 3
98 pb_16: times 16 db 16
99 pw_62: times 8 dw 62
100 pw_64: times 8 dw 64
101 pw_256: times 8 dw 256
102 pw_512: times 8 dw 512
103 pw_m256: times 8 dw -256
104 pb_2: times 8 db 2
105 pb_4: times 8 db 4
106 pb_8: times 8 db 8
107 pb_128: times 8 db 128
108 pb_m16: times 8 db -16
109 pw_128: times 4 dw 128
110 pw_255: times 4 dw 255
111 pb_36_m4: times 4 db 36, -4
112 pb_127_m127: times 4 db 127, -127
113
114 %macro JMP_TABLE 3-*
115 %xdefine %1_%2_table (%%table - 2*4)
116 %xdefine %%base mangle(private_prefix %+ _%1_8bpc_%2)
117 %%table:
118 %rep %0 - 2
119 dd %%base %+ .%3 - (%%table - 2*4)
120 %rotate 1
121 %endrep
122 %endmacro
123
124 %define ipred_dc_splat_ssse3_table (ipred_dc_ssse3_table + 10*4)
125 %define ipred_cfl_splat_ssse3_table (ipred_cfl_ssse3_table + 8*4)
126
127 JMP_TABLE ipred_h, ssse3, w4, w8, w16, w32, w64
128 JMP_TABLE ipred_dc, ssse3, h4, h8, h16, h32, h64, w4, w8, w16, w32, w64, \
129 s4-10*4, s8-10*4, s16-10*4, s32-10*4, s64-10*4
130 JMP_TABLE ipred_dc_left, ssse3, h4, h8, h16, h32, h64
131 JMP_TABLE ipred_smooth, ssse3, w4, w8, w16, w32, w64
132 JMP_TABLE ipred_smooth_v, ssse3, w4, w8, w16, w32, w64
133 JMP_TABLE ipred_smooth_h, ssse3, w4, w8, w16, w32, w64
134 JMP_TABLE ipred_paeth, ssse3, w4, w8, w16, w32, w64
135 JMP_TABLE ipred_z1, ssse3, w4, w8, w16, w32, w64
136 JMP_TABLE ipred_z2, ssse3, w4, w8, w16, w32, w64
137 JMP_TABLE ipred_z3, ssse3, h4, h8, h16, h32, h64
138 JMP_TABLE pal_pred, ssse3, w4, w8, w16, w32, w64
139 JMP_TABLE ipred_cfl, ssse3, h4, h8, h16, h32, w4, w8, w16, w32, \
140 s4-8*4, s8-8*4, s16-8*4, s32-8*4
141 JMP_TABLE ipred_cfl_left, ssse3, h4, h8, h16, h32
142 JMP_TABLE ipred_filter, ssse3, w4, w8, w16, w32
143
144 cextern dr_intra_derivative
145 cextern filter_intra_taps
146
147 SECTION .text
148
149 ;---------------------------------------------------------------------------------------
150 ;int dav1d_ipred_h_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
151 ; const int width, const int height, const int a);
152 ;---------------------------------------------------------------------------------------
153 %macro IPRED_SET 3 ; width, stride, stride size pshuflw_imm8
154 pshuflw m1, m0, %3 ; extend 8 byte for 2 pos
155 punpcklqdq m1, m1
156 mova [dstq + %2], m1
157 %if %1 > 16
158 mova [dstq + 16 + %2], m1
159 %endif
160 %if %1 > 32
161 mova [dstq + 32 + %2], m1
162 mova [dstq + 48 + %2], m1
163 %endif
164 %endmacro
165
166 %macro IPRED_H 1 ; width
167 sub tlq, 4
168 movd m0, [tlq] ; get 4 bytes of topleft data
169 punpcklbw m0, m0 ; extend 2 byte
170 %if %1 == 4
171 pshuflw m1, m0, q2233
172 movd [dstq+strideq*0], m1
173 psrlq m1, 32
174 movd [dstq+strideq*1], m1
175 pshuflw m0, m0, q0011
176 movd [dstq+strideq*2], m0
177 psrlq m0, 32
178 movd [dstq+stride3q ], m0
179
180 %elif %1 == 8
181 punpcklwd m0, m0
182 punpckhdq m1, m0, m0
183 punpckldq m0, m0
184 movq [dstq+strideq*1], m1
185 movhps [dstq+strideq*0], m1
186 movq [dstq+stride3q ], m0
187 movhps [dstq+strideq*2], m0
188 %else
189 IPRED_SET %1, 0, q3333
190 IPRED_SET %1, strideq, q2222
191 IPRED_SET %1, strideq*2, q1111
192 IPRED_SET %1, stride3q, q0000
193 %endif
194 lea dstq, [dstq+strideq*4]
195 sub hd, 4
196 jg .w%1
197 RET
198 %endmacro
199
200 INIT_XMM ssse3
201 cglobal ipred_h_8bpc, 3, 6, 2, dst, stride, tl, w, h, stride3
202 LEA r5, ipred_h_ssse3_table
203 tzcnt wd, wm
204 movifnidn hd, hm
205 movsxd wq, [r5+wq*4]
206 add wq, r5
207 lea stride3q, [strideq*3]
208 jmp wq
209 .w4:
210 IPRED_H 4
211 .w8:
212 IPRED_H 8
213 .w16:
214 IPRED_H 16
215 .w32:
216 IPRED_H 32
217 .w64:
218 IPRED_H 64
219
220 ;---------------------------------------------------------------------------------------
221 ;int dav1d_ipred_v_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
222 ; const int width, const int height, const int a);
223 ;---------------------------------------------------------------------------------------
224 cglobal ipred_v_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
225 LEA r5, ipred_dc_splat_ssse3_table
226 tzcnt wd, wm
227 movu m0, [tlq+ 1]
228 movu m1, [tlq+17]
229 movu m2, [tlq+33]
230 movu m3, [tlq+49]
231 movifnidn hd, hm
232 movsxd wq, [r5+wq*4]
233 add wq, r5
234 lea stride3q, [strideq*3]
235 jmp wq
236
237 ;---------------------------------------------------------------------------------------
238 ;int dav1d_ipred_dc_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
239 ; const int width, const int height, const int a);
240 ;---------------------------------------------------------------------------------------
241 cglobal ipred_dc_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
242 movifnidn hd, hm
243 movifnidn wd, wm
244 tzcnt r6d, hd
245 lea r5d, [wq+hq]
246 movd m4, r5d
247 tzcnt r5d, r5d
248 movd m5, r5d
249 LEA r5, ipred_dc_ssse3_table
250 tzcnt wd, wd
251 movsxd r6, [r5+r6*4]
252 movsxd wq, [r5+wq*4+20]
253 pcmpeqd m3, m3
254 psrlw m4, 1 ; dc = (width + height) >> 1;
255 add r6, r5
256 add wq, r5
257 lea stride3q, [strideq*3]
258 jmp r6
259 .h4:
260 movd m0, [tlq-4]
261 pmaddubsw m0, m3
262 jmp wq
263 .w4:
264 movd m1, [tlq+1]
265 pmaddubsw m1, m3
266 psubw m0, m4
267 paddw m0, m1
268 pmaddwd m0, m3
269 cmp hd, 4
270 jg .w4_mul
271 psrlw m0, 3 ; dc >>= ctz(width + height);
272 jmp .w4_end
273 .w4_mul:
274 punpckhqdq m1, m0, m0
275 paddw m0, m1
276 psrlq m1, m0, 32
277 paddw m0, m1
278 psrlw m0, 2
279 mov r6d, 0x5556
280 mov r2d, 0x3334
281 test hd, 8
282 cmovz r6d, r2d
283 movd m5, r6d
284 pmulhuw m0, m5
285 .w4_end:
286 pxor m1, m1
287 pshufb m0, m1
288 .s4:
289 movd [dstq+strideq*0], m0
290 movd [dstq+strideq*1], m0
291 movd [dstq+strideq*2], m0
292 movd [dstq+stride3q ], m0
293 lea dstq, [dstq+strideq*4]
294 sub hd, 4
295 jg .s4
296 RET
297 ALIGN function_align
298 .h8:
299 movq m0, [tlq-8]
300 pmaddubsw m0, m3
301 jmp wq
302 .w8:
303 movq m1, [tlq+1]
304 pmaddubsw m1, m3
305 psubw m4, m0
306 punpckhqdq m0, m0
307 psubw m0, m4
308 paddw m0, m1
309 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
310 paddw m0, m1
311 pmaddwd m0, m3
312 psrlw m0, m5
313 cmp hd, 8
314 je .w8_end
315 mov r6d, 0x5556
316 mov r2d, 0x3334
317 cmp hd, 32
318 cmovz r6d, r2d
319 movd m1, r6d
320 pmulhuw m0, m1
321 .w8_end:
322 pxor m1, m1
323 pshufb m0, m1
324 .s8:
325 movq [dstq+strideq*0], m0
326 movq [dstq+strideq*1], m0
327 movq [dstq+strideq*2], m0
328 movq [dstq+stride3q ], m0
329 lea dstq, [dstq+strideq*4]
330 sub hd, 4
331 jg .s8
332 RET
333 ALIGN function_align
334 .h16:
335 mova m0, [tlq-16]
336 pmaddubsw m0, m3
337 jmp wq
338 .w16:
339 movu m1, [tlq+1]
340 pmaddubsw m1, m3
341 paddw m0, m1
342 psubw m4, m0
343 punpckhqdq m0, m0
344 psubw m0, m4
345 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
346 paddw m0, m1
347 pmaddwd m0, m3
348 psrlw m0, m5
349 cmp hd, 16
350 je .w16_end
351 mov r6d, 0x5556
352 mov r2d, 0x3334
353 test hd, 8|32
354 cmovz r6d, r2d
355 movd m1, r6d
356 pmulhuw m0, m1
357 .w16_end:
358 pxor m1, m1
359 pshufb m0, m1
360 .s16:
361 mova [dstq+strideq*0], m0
362 mova [dstq+strideq*1], m0
363 mova [dstq+strideq*2], m0
364 mova [dstq+stride3q ], m0
365 lea dstq, [dstq+strideq*4]
366 sub hd, 4
367 jg .s16
368 RET
369 ALIGN function_align
370 .h32:
371 mova m0, [tlq-32]
372 pmaddubsw m0, m3
373 mova m2, [tlq-16]
374 pmaddubsw m2, m3
375 paddw m0, m2
376 jmp wq
377 .w32:
378 movu m1, [tlq+1]
379 pmaddubsw m1, m3
380 movu m2, [tlq+17]
381 pmaddubsw m2, m3
382 paddw m1, m2
383 paddw m0, m1
384 psubw m4, m0
385 punpckhqdq m0, m0
386 psubw m0, m4
387 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
388 paddw m0, m1
389 pmaddwd m0, m3
390 psrlw m0, m5
391 cmp hd, 32
392 je .w32_end
393 lea r2d, [hq*2]
394 mov r6d, 0x5556
395 mov r2d, 0x3334
396 test hd, 64|16
397 cmovz r6d, r2d
398 movd m1, r6d
399 pmulhuw m0, m1
400 .w32_end:
401 pxor m1, m1
402 pshufb m0, m1
403 mova m1, m0
404 .s32:
405 mova [dstq], m0
406 mova [dstq+16], m1
407 mova [dstq+strideq], m0
408 mova [dstq+strideq+16], m1
409 mova [dstq+strideq*2], m0
410 mova [dstq+strideq*2+16], m1
411 mova [dstq+stride3q], m0
412 mova [dstq+stride3q+16], m1
413 lea dstq, [dstq+strideq*4]
414 sub hd, 4
415 jg .s32
416 RET
417 ALIGN function_align
418 .h64:
419 mova m0, [tlq-64]
420 mova m1, [tlq-48]
421 pmaddubsw m0, m3
422 pmaddubsw m1, m3
423 paddw m0, m1
424 mova m1, [tlq-32]
425 pmaddubsw m1, m3
426 paddw m0, m1
427 mova m1, [tlq-16]
428 pmaddubsw m1, m3
429 paddw m0, m1
430 jmp wq
431 .w64:
432 movu m1, [tlq+ 1]
433 movu m2, [tlq+17]
434 pmaddubsw m1, m3
435 pmaddubsw m2, m3
436 paddw m1, m2
437 movu m2, [tlq+33]
438 pmaddubsw m2, m3
439 paddw m1, m2
440 movu m2, [tlq+49]
441 pmaddubsw m2, m3
442 paddw m1, m2
443 paddw m0, m1
444 psubw m4, m0
445 punpckhqdq m0, m0
446 psubw m0, m4
447 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
448 paddw m0, m1
449 pmaddwd m0, m3
450 psrlw m0, m5
451 cmp hd, 64
452 je .w64_end
453 mov r6d, 0x5556
454 mov r2d, 0x3334
455 test hd, 32
456 cmovz r6d, r2d
457 movd m1, r6d
458 pmulhuw m0, m1
459 .w64_end:
460 pxor m1, m1
461 pshufb m0, m1
462 mova m1, m0
463 mova m2, m0
464 mova m3, m0
465 .s64:
466 mova [dstq], m0
467 mova [dstq+16], m1
468 mova [dstq+32], m2
469 mova [dstq+48], m3
470 mova [dstq+strideq], m0
471 mova [dstq+strideq+16], m1
472 mova [dstq+strideq+32], m2
473 mova [dstq+strideq+48], m3
474 lea dstq, [dstq+strideq*2]
475 sub hd, 2
476 jg .s64
477 RET
478
479 ;---------------------------------------------------------------------------------------
480 ;int dav1d_ipred_dc_left_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
481 ; const int width, const int height, const int a);
482 ;---------------------------------------------------------------------------------------
483 cglobal ipred_dc_left_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
484 LEA r5, ipred_dc_left_ssse3_table
485 mov hd, hm ; zero upper half
486 tzcnt r6d, hd
487 sub tlq, hq
488 tzcnt wd, wm
489 movu m0, [tlq]
490 movd m3, [r5-ipred_dc_left_ssse3_table+pd_32768]
491 movd m2, r6d
492 psrld m3, m2
493 movsxd r6, [r5+r6*4]
494 pcmpeqd m2, m2
495 pmaddubsw m0, m2
496 add r6, r5
497 add r5, ipred_dc_splat_ssse3_table-ipred_dc_left_ssse3_table
498 movsxd wq, [r5+wq*4]
499 add wq, r5
500 jmp r6
501 .h64:
502 movu m1, [tlq+48] ; unaligned when jumping here from dc_top
503 pmaddubsw m1, m2
504 paddw m0, m1
505 movu m1, [tlq+32] ; unaligned when jumping here from dc_top
506 pmaddubsw m1, m2
507 paddw m0, m1
508 .h32:
509 movu m1, [tlq+16] ; unaligned when jumping here from dc_top
510 pmaddubsw m1, m2
511 paddw m0, m1
512 .h16:
513 pshufd m1, m0, q3232 ; psrlq m1, m0, 16
514 paddw m0, m1
515 .h8:
516 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
517 paddw m0, m1
518 .h4:
519 pmaddwd m0, m2
520 pmulhrsw m0, m3
521 lea stride3q, [strideq*3]
522 pxor m1, m1
523 pshufb m0, m1
524 mova m1, m0
525 mova m2, m0
526 mova m3, m0
527 jmp wq
528
529 ;---------------------------------------------------------------------------------------
530 ;int dav1d_ipred_dc_128_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
531 ; const int width, const int height, const int a);
532 ;---------------------------------------------------------------------------------------
533 cglobal ipred_dc_128_8bpc, 2, 7, 6, dst, stride, tl, w, h, stride3
534 LEA r5, ipred_dc_splat_ssse3_table
535 tzcnt wd, wm
536 movifnidn hd, hm
537 movsxd wq, [r5+wq*4]
538 movddup m0, [r5-ipred_dc_splat_ssse3_table+pb_128]
539 mova m1, m0
540 mova m2, m0
541 mova m3, m0
542 add wq, r5
543 lea stride3q, [strideq*3]
544 jmp wq
545
546 ;---------------------------------------------------------------------------------------
547 ;int dav1d_ipred_dc_top_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
548 ; const int width, const int height, const int a);
549 ;---------------------------------------------------------------------------------------
550 cglobal ipred_dc_top_8bpc, 3, 7, 6, dst, stride, tl, w, h
551 LEA r5, ipred_dc_left_ssse3_table
552 tzcnt wd, wm
553 inc tlq
554 movu m0, [tlq]
555 movifnidn hd, hm
556 movd m3, [r5-ipred_dc_left_ssse3_table+pd_32768]
557 movd m2, wd
558 psrld m3, m2
559 movsxd r6, [r5+wq*4]
560 pcmpeqd m2, m2
561 pmaddubsw m0, m2
562 add r6, r5
563 add r5, ipred_dc_splat_ssse3_table-ipred_dc_left_ssse3_table
564 movsxd wq, [r5+wq*4]
565 add wq, r5
566 jmp r6
567
568 ;---------------------------------------------------------------------------------------
569 ;int dav1d_ipred_smooth_v_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
570 ; const int width, const int height, const int a);
571 ;---------------------------------------------------------------------------------------
572 %macro SMOOTH 6 ; src[1-2], mul[1-2], add[1-2]
573 ; w * a = (w - 128) * a + 128 * a
574 ; (256 - w) * b = (127 - w) * b + 129 * b
575 ; => w * a + (256 - w) * b = [(w - 128) * a + (127 - w) * b] + [128 * a + 129 * b]
576 pmaddubsw m6, m%3, m%1
577 pmaddubsw m0, m%4, m%2 ; (w - 128) * a + (127 - w) * b
578 paddw m6, m%5
579 paddw m0, m%6 ; [(w - 128) * a + (127 - w) * b] + [128 * a + 129 * b + 128]
580 psrlw m6, 8
581 psrlw m0, 8
582 packuswb m6, m0
583 %endmacro
584
585 cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl, w, h, weights
586 %define base r6-ipred_smooth_v_ssse3_table
587 LEA r6, ipred_smooth_v_ssse3_table
588 tzcnt wd, wm
589 mov hd, hm
590 movsxd wq, [r6+wq*4]
591 movddup m0, [base+pb_127_m127]
592 movddup m1, [base+pw_128]
593 lea weightsq, [base+smooth_weights+hq*4]
594 neg hq
595 movd m5, [tlq+hq]
596 pxor m2, m2
597 pshufb m5, m2
598 add wq, r6
599 jmp wq
600 .w4:
601 movd m2, [tlq+1]
602 punpckldq m2, m2
603 punpcklbw m2, m5 ; top, bottom
604 lea r3, [strideq*3]
605 mova m4, [base+ipred_v_shuf]
606 mova m5, m4
607 punpckldq m4, m4
608 punpckhdq m5, m5
609 pmaddubsw m3, m2, m0 ; m3: 127 * top - 127 * bottom
610 paddw m1, m2 ; m1: 1 * top + 256 * bottom + 128, overflow is ok
611 paddw m3, m1 ; m3: 128 * top + 129 * bottom + 128
612 .w4_loop:
613 movu m1, [weightsq+hq*2]
614 pshufb m0, m1, m4 ;m2, m3, m4 and m5 should be stable in loop
615 pshufb m1, m5
616 SMOOTH 0, 1, 2, 2, 3, 3
617 movd [dstq+strideq*0], m6
618 pshuflw m1, m6, q1032
619 movd [dstq+strideq*1], m1
620 punpckhqdq m6, m6
621 movd [dstq+strideq*2], m6
622 psrlq m6, 32
623 movd [dstq+r3 ], m6
624 lea dstq, [dstq+strideq*4]
625 add hq, 4
626 jl .w4_loop
627 RET
628 ALIGN function_align
629 .w8:
630 movq m2, [tlq+1]
631 punpcklbw m2, m5
632 mova m5, [base+ipred_v_shuf]
633 lea r3, [strideq*3]
634 pshufd m4, m5, q0000
635 pshufd m5, m5, q1111
636 pmaddubsw m3, m2, m0
637 paddw m1, m2
638 paddw m3, m1 ; m3 is output for loop
639 .w8_loop:
640 movq m1, [weightsq+hq*2]
641 pshufb m0, m1, m4
642 pshufb m1, m5
643 SMOOTH 0, 1, 2, 2, 3, 3
644 movq [dstq+strideq*0], m6
645 movhps [dstq+strideq*1], m6
646 lea dstq, [dstq+strideq*2]
647 add hq, 2
648 jl .w8_loop
649 RET
650 ALIGN function_align
651 .w16:
652 movu m3, [tlq+1]
653 punpcklbw m2, m3, m5
654 punpckhbw m3, m5
655 pmaddubsw m4, m2, m0
656 pmaddubsw m5, m3, m0
657 paddw m0, m1, m2
658 paddw m1, m3
659 paddw m4, m0
660 paddw m5, m1 ; m4 and m5 is output for loop
661 .w16_loop:
662 movd m1, [weightsq+hq*2]
663 pshuflw m1, m1, q0000
664 punpcklqdq m1, m1
665 SMOOTH 1, 1, 2, 3, 4, 5
666 mova [dstq], m6
667 add dstq, strideq
668 add hq, 1
669 jl .w16_loop
670 RET
671 ALIGN function_align
672 .w32:
673 %if WIN64
674 movaps [rsp+24], xmm7
675 %define xmm_regs_used 8
676 %endif
677 mova m7, m5
678 .w32_loop_init:
679 mov r3d, 2
680 .w32_loop:
681 movddup m0, [base+pb_127_m127]
682 movddup m1, [base+pw_128]
683 movu m3, [tlq+1]
684 punpcklbw m2, m3, m7
685 punpckhbw m3, m7
686 pmaddubsw m4, m2, m0
687 pmaddubsw m5, m3, m0
688 paddw m0, m1, m2
689 paddw m1, m3
690 paddw m4, m0
691 paddw m5, m1
692 movd m1, [weightsq+hq*2]
693 pshuflw m1, m1, q0000
694 punpcklqdq m1, m1
695 SMOOTH 1, 1, 2, 3, 4, 5
696 mova [dstq], m6
697 add tlq, 16
698 add dstq, 16
699 dec r3d
700 jg .w32_loop
701 lea dstq, [dstq-32+strideq]
702 sub tlq, 32
703 add hq, 1
704 jl .w32_loop_init
705 RET
706 ALIGN function_align
707 .w64:
708 %if WIN64
709 movaps [rsp+24], xmm7
710 %define xmm_regs_used 8
711 %endif
712 mova m7, m5
713 .w64_loop_init:
714 mov r3d, 4
715 .w64_loop:
716 movddup m0, [base+pb_127_m127]
717 movddup m1, [base+pw_128]
718 movu m3, [tlq+1]
719 punpcklbw m2, m3, m7
720 punpckhbw m3, m7
721 pmaddubsw m4, m2, m0
722 pmaddubsw m5, m3, m0
723 paddw m0, m1, m2
724 paddw m1, m3
725 paddw m4, m0
726 paddw m5, m1
727 movd m1, [weightsq+hq*2]
728 pshuflw m1, m1, q0000
729 punpcklqdq m1, m1
730 SMOOTH 1, 1, 2, 3, 4, 5
731 mova [dstq], m6
732 add tlq, 16
733 add dstq, 16
734 dec r3d
735 jg .w64_loop
736 lea dstq, [dstq-64+strideq]
737 sub tlq, 64
738 add hq, 1
739 jl .w64_loop_init
740 RET
741
742 ;---------------------------------------------------------------------------------------
743 ;int dav1d_ipred_smooth_h_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
744 ; const int width, const int height, const int a);
745 ;---------------------------------------------------------------------------------------
746 cglobal ipred_smooth_h_8bpc, 3, 7, 8, dst, stride, tl, w, h
747 %define base r6-ipred_smooth_h_ssse3_table
748 LEA r6, ipred_smooth_h_ssse3_table
749 mov wd, wm
750 movd m3, [tlq+wq]
751 pxor m1, m1
752 pshufb m3, m1 ; right
753 tzcnt wd, wd
754 mov hd, hm
755 movsxd wq, [r6+wq*4]
756 movddup m4, [base+pb_127_m127]
757 movddup m5, [base+pw_128]
758 add wq, r6
759 jmp wq
760 .w4:
761 movddup m6, [base+smooth_weights+4*2]
762 mova m7, [base+ipred_h_shuf]
763 sub tlq, 4
764 sub tlq, hq
765 lea r3, [strideq*3]
766 .w4_loop:
767 movd m2, [tlq+hq] ; left
768 pshufb m2, m7
769 punpcklbw m1, m2, m3 ; left, right
770 punpckhbw m2, m3
771 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
772 paddw m0, m1 ; 128 * left + 129 * right
773 pmaddubsw m1, m6
774 paddw m1, m5
775 paddw m0, m1
776 pmaddubsw m1, m2, m4
777 paddw m1, m2
778 pmaddubsw m2, m6
779 paddw m2, m5
780 paddw m1, m2
781 psrlw m0, 8
782 psrlw m1, 8
783 packuswb m0, m1
784 movd [dstq+strideq*0], m0
785 pshuflw m1, m0, q1032
786 movd [dstq+strideq*1], m1
787 punpckhqdq m0, m0
788 movd [dstq+strideq*2], m0
789 psrlq m0, 32
790 movd [dstq+r3 ], m0
791 lea dstq, [dstq+strideq*4]
792 sub hd, 4
793 jg .w4_loop
794 RET
795 ALIGN function_align
796 .w8:
797 mova m6, [base+smooth_weights+8*2]
798 mova m7, [base+ipred_h_shuf]
799 sub tlq, 4
800 sub tlq, hq
801 punpckldq m7, m7
802 .w8_loop:
803 movd m2, [tlq+hq] ; left
804 pshufb m2, m7
805 punpcklbw m1, m2, m3 ; left, right
806 punpckhbw m2, m3
807 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
808 paddw m0, m1 ; 128 * left + 129 * right
809 pmaddubsw m1, m6
810 paddw m1, m5
811 paddw m0, m1
812 pmaddubsw m1, m2, m4
813 paddw m1, m2
814 pmaddubsw m2, m6
815 paddw m2, m5
816 paddw m1, m2
817 psrlw m0, 8
818 psrlw m1, 8
819 packuswb m0, m1
820 movq [dstq+strideq*0], m0
821 movhps [dstq+strideq*1], m0
822 lea dstq, [dstq+strideq*2]
823 sub hd, 2
824 jg .w8_loop
825 RET
826 ALIGN function_align
827 .w16:
828 mova m6, [base+smooth_weights+16*2]
829 mova m7, [base+smooth_weights+16*3]
830 sub tlq, 1
831 sub tlq, hq
832 .w16_loop:
833 pxor m1, m1
834 movd m2, [tlq+hq] ; left
835 pshufb m2, m1
836 punpcklbw m1, m2, m3 ; left, right
837 punpckhbw m2, m3
838 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
839 paddw m0, m1 ; 128 * left + 129 * right
840 pmaddubsw m1, m6
841 paddw m1, m5
842 paddw m0, m1
843 pmaddubsw m1, m2, m4
844 paddw m1, m2
845 pmaddubsw m2, m7
846 paddw m2, m5
847 paddw m1, m2
848 psrlw m0, 8
849 psrlw m1, 8
850 packuswb m0, m1
851 mova [dstq], m0
852 lea dstq, [dstq+strideq]
853 sub hd, 1
854 jg .w16_loop
855 RET
856 ALIGN function_align
857 .w32:
858 sub tlq, 1
859 sub tlq, hq
860 pxor m6, m6
861 .w32_loop_init:
862 mov r5, 2
863 lea r3, [base+smooth_weights+16*4]
864 .w32_loop:
865 mova m7, [r3]
866 add r3, 16
867 movd m2, [tlq+hq] ; left
868 pshufb m2, m6
869 punpcklbw m1, m2, m3 ; left, right
870 punpckhbw m2, m3
871 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
872 paddw m0, m1 ; 128 * left + 129 * right
873 pmaddubsw m1, m7
874 paddw m1, m5
875 paddw m0, m1
876 pmaddubsw m1, m2, m4
877 paddw m1, m2
878 mova m7, [r3]
879 add r3, 16
880 pmaddubsw m2, m7
881 paddw m2, m5
882 paddw m1, m2
883 psrlw m0, 8
884 psrlw m1, 8
885 packuswb m0, m1
886 mova [dstq], m0
887 add dstq, 16
888 dec r5
889 jg .w32_loop
890 lea dstq, [dstq-32+strideq]
891 sub hd, 1
892 jg .w32_loop_init
893 RET
894 ALIGN function_align
895 .w64:
896 sub tlq, 1
897 sub tlq, hq
898 pxor m6, m6
899 .w64_loop_init:
900 mov r5, 4
901 lea r3, [base+smooth_weights+16*8]
902 .w64_loop:
903 mova m7, [r3]
904 add r3, 16
905 movd m2, [tlq+hq] ; left
906 pshufb m2, m6
907 punpcklbw m1, m2, m3 ; left, right
908 punpckhbw m2, m3
909 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
910 paddw m0, m1 ; 128 * left + 129 * right
911 pmaddubsw m1, m7
912 paddw m1, m5
913 paddw m0, m1
914 pmaddubsw m1, m2, m4
915 paddw m1, m2
916 mova m7, [r3]
917 add r3, 16
918 pmaddubsw m2, m7
919 paddw m2, m5
920 paddw m1, m2
921 psrlw m0, 8
922 psrlw m1, 8
923 packuswb m0, m1
924 mova [dstq], m0
925 add dstq, 16
926 dec r5
927 jg .w64_loop
928 lea dstq, [dstq-64+strideq]
929 sub hd, 1
930 jg .w64_loop_init
931 RET
932
933 ;---------------------------------------------------------------------------------------
934 ;int dav1d_ipred_smooth_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
935 ; const int width, const int height, const int a);
936 ;---------------------------------------------------------------------------------------
937 %macro SMOOTH_2D_END 7 ; src[1-2], mul[1-2], add[1-2], m3
938 pmaddubsw m6, m%3, m%1
939 mova m0, m6
940 pmaddubsw m6, m%4, m%2
941 mova m1, m6
942 %ifnum %5
943 paddw m0, m%5
944 %else
945 paddw m0, %5
946 %endif
947 %ifnum %6
948 paddw m1, m%6
949 %else
950 paddw m1, %6
951 %endif
952 %ifnum %7
953 %else
954 mova m3, %7
955 %endif
956 pavgw m0, m2
957 pavgw m1, m3
958 psrlw m0, 8
959 psrlw m1, 8
960 packuswb m0, m1
961 %endmacro
962
963 %macro SMOOTH_OUTPUT_16B 12 ; m1, [buffer1, buffer2, buffer3, buffer4,] [w1, w2,] m3, m7, [m0, m4, m5]
964 mova m1, [rsp+16*%1] ; top
965 punpckhbw m6, m1, m0 ; top, bottom
966 punpcklbw m1, m0 ; top, bottom
967 pmaddubsw m2, m1, m5
968 mova [rsp+16*%2], m1
969 paddw m1, m3 ; 1 * top + 255 * bottom + 255
970 paddw m2, m1 ; 128 * top + 129 * bottom + 255
971 mova [rsp+16*%3], m2
972 pmaddubsw m2, m6, m5
973 mova [rsp+16*%4], m6
974 paddw m6, m3 ; 1 * top + 255 * bottom + 255
975 paddw m2, m6 ; 128 * top + 129 * bottom + 255
976 mova [rsp+16*%5], m2
977 movd m1, [tlq+hq] ; left
978 pshufb m1, [base+pb_3] ; topleft[-(1 + y)]
979 punpcklbw m1, m4 ; left, right
980 pmaddubsw m2, m1, m5 ; 127 * left - 127 * right
981 paddw m2, m1 ; 128 * left + 129 * right
982 mova m3, m2
983 pmaddubsw m0, m1, %6 ; weights_hor = &dav1d_sm_weights[width];
984 pmaddubsw m1, %7
985 paddw m2, m3, m0
986 paddw m3, m1
987 movd m1, [v_weightsq] ; weights_ver = &dav1d_sm_weights[height];
988 mova m7, [rsp+16*%9]
989 pshufb m1, m7
990 mova [rsp+16*%8], m3
991 mova m4, [rsp+16*%2]
992 mova m5, [rsp+16*%3]
993 mova m3, [rsp+16*%4]
994 mova m7, [rsp+16*%5]
995 SMOOTH_2D_END 1, 1, 4, 3, 5, 7, [rsp+16*%8]
996 mova [dstq], m0
997 movddup m3, [base+pw_255] ; recovery
998 mova m0, [rsp+16*%10] ; recovery
999 mova m4, [rsp+16*%11] ; recovery
1000 mova m5, [rsp+16*%12] ; recovery
1001 %endmacro
1002
1003 cglobal ipred_smooth_8bpc, 3, 7, 8, -13*16, dst, stride, tl, w, h, v_weights
1004 %define base r6-ipred_smooth_ssse3_table
1005 mov wd, wm
1006 mov hd, hm
1007 LEA r6, ipred_smooth_ssse3_table
1008 movd m4, [tlq+wq] ; right
1009 pxor m2, m2
1010 pshufb m4, m2
1011 tzcnt wd, wd
1012 mov r5, tlq
1013 sub r5, hq
1014 movsxd wq, [r6+wq*4]
1015 movddup m5, [base+pb_127_m127]
1016 movd m0, [r5]
1017 pshufb m0, m2 ; bottom
1018 movddup m3, [base+pw_255]
1019 add wq, r6
1020 lea v_weightsq, [base+smooth_weights+hq*2] ; weights_ver = &dav1d_sm_weights[height]
1021 jmp wq
1022 .w4:
1023 mova m7, [base+ipred_v_shuf]
1024 movd m1, [tlq+1] ; left
1025 pshufd m1, m1, q0000
1026 sub tlq, 4
1027 lea r3, [strideq*3]
1028 sub tlq, hq
1029 punpcklbw m1, m0 ; top, bottom
1030 pshufd m6, m7, q1100
1031 pshufd m7, m7, q3322
1032 pmaddubsw m2, m1, m5
1033 paddw m3, m1 ; 1 * top + 255 * bottom + 255
1034 paddw m2, m3 ; 128 * top + 129 * bottom + 255
1035 mova [rsp+16*0], m1
1036 mova [rsp+16*1], m2
1037 movq m1, [base+smooth_weights+4*2] ; weights_hor = &dav1d_sm_weights[width];
1038 punpcklqdq m1, m1
1039 mova [rsp+16*2], m1
1040 mova [rsp+16*3], m4
1041 mova [rsp+16*4], m6
1042 mova [rsp+16*5], m5
1043 .w4_loop:
1044 movd m1, [tlq+hq] ; left
1045 pshufb m1, [base+ipred_h_shuf]
1046 punpcklbw m0, m1, m4 ; left, right
1047 punpckhbw m1, m4
1048 pmaddubsw m2, m0, m5 ; 127 * left - 127 * right
1049 pmaddubsw m3, m1, m5
1050 paddw m2, m0 ; 128 * left + 129 * right
1051 paddw m3, m1
1052 mova m4, [rsp+16*2]
1053 pmaddubsw m0, m4
1054 pmaddubsw m1, m4
1055 paddw m2, m0
1056 paddw m3, m1
1057 movq m1, [v_weightsq] ; weights_ver = &dav1d_sm_weights[height];
1058 add v_weightsq, 8
1059 pshufb m0, m1, m6
1060 pshufb m1, m7
1061 mova m4, [rsp+16*0]
1062 mova m5, [rsp+16*1]
1063 SMOOTH_2D_END 0, 1, 4, 4, 5, 5, 3
1064 mova m4, [rsp+16*3]
1065 mova m6, [rsp+16*4]
1066 mova m5, [rsp+16*5]
1067 movd [dstq+strideq*0], m0
1068 pshuflw m1, m0, q1032
1069 movd [dstq+strideq*1], m1
1070 punpckhqdq m0, m0
1071 movd [dstq+strideq*2], m0
1072 psrlq m0, 32
1073 movd [dstq+r3 ], m0
1074 lea dstq, [dstq+strideq*4]
1075 sub hd, 4
1076 jg .w4_loop
1077 RET
1078 ALIGN function_align
1079 .w8:
1080 mova m7, [base+ipred_v_shuf]
1081 movq m1, [tlq+1] ; left
1082 punpcklqdq m1, m1
1083 sub tlq, 4
1084 sub tlq, hq
1085 punpcklbw m1, m0
1086 pshufd m6, m7, q0000
1087 pshufd m7, m7, q1111
1088 pmaddubsw m2, m1, m5
1089 paddw m3, m1
1090 paddw m2, m3
1091 mova [rsp+16*0], m1
1092 mova [rsp+16*1], m2
1093 mova m1, [base+smooth_weights+8*2] ; weights_hor = &dav1d_sm_weights[width];
1094 mova [rsp+16*2], m1
1095 mova [rsp+16*3], m4
1096 mova [rsp+16*4], m6
1097 mova [rsp+16*5], m5
1098 .w8_loop:
1099 movd m1, [tlq+hq] ; left
1100 pshufb m1, [base+ipred_h_shuf]
1101 pshufd m1, m1, q1100
1102 punpcklbw m0, m1, m4
1103 punpckhbw m1, m4
1104 pmaddubsw m2, m0, m5
1105 pmaddubsw m3, m1, m5
1106 paddw m2, m0
1107 paddw m3, m1
1108 mova m4, [rsp+16*2]
1109 pmaddubsw m0, m4
1110 pmaddubsw m1, m4
1111 paddw m2, m0
1112 paddw m3, m1
1113 movd m1, [v_weightsq] ; weights_ver = &dav1d_sm_weights[height];
1114 add v_weightsq, 4
1115 pshufb m0, m1, m6
1116 pshufb m1, m7
1117 mova m4, [rsp+16*0]
1118 mova m5, [rsp+16*1]
1119 SMOOTH_2D_END 0, 1, 4, 4, 5, 5, 3
1120 mova m4, [rsp+16*3]
1121 mova m6, [rsp+16*4]
1122 mova m5, [rsp+16*5]
1123 movq [dstq+strideq*0], m0
1124 movhps [dstq+strideq*1], m0
1125 lea dstq, [dstq+strideq*2]
1126 sub hd, 2
1127 jg .w8_loop
1128 RET
1129 ALIGN function_align
1130 .w16:
1131 mova m7, [base+ipred_v_shuf]
1132 movu m1, [tlq+1] ; left
1133 sub tlq, 4
1134 sub tlq, hq
1135 punpckhbw m6, m1, m0 ; top, bottom
1136 punpcklbw m1, m0 ; top, bottom
1137 pshufd m7, m7, q0000
1138 mova [rsp+16*2], m7
1139 pmaddubsw m2, m6, m5
1140 mova [rsp+16*5], m6
1141 paddw m6, m3 ; 1 * top + 255 * bottom + 255
1142 paddw m2, m6 ; 128 * top + 129 * bottom + 255
1143 mova [rsp+16*6], m2
1144 pmaddubsw m2, m1, m5
1145 paddw m3, m1 ; 1 * top + 255 * bottom + 255
1146 mova [rsp+16*0], m1
1147 paddw m2, m3 ; 128 * top + 129 * bottom + 255
1148 mova [rsp+16*1], m2
1149 mova [rsp+16*3], m4
1150 mova [rsp+16*4], m5
1151 .w16_loop:
1152 movd m1, [tlq+hq] ; left
1153 pshufb m1, [base+pb_3] ; topleft[-(1 + y)]
1154 punpcklbw m1, m4 ; left, right
1155 pmaddubsw m2, m1, m5 ; 127 * left - 127 * right
1156 paddw m2, m1 ; 128 * left + 129 * right
1157 mova m0, m1
1158 mova m3, m2
1159 pmaddubsw m0, [base+smooth_weights+16*2] ; weights_hor = &dav1d_sm_weights[width];
1160 pmaddubsw m1, [base+smooth_weights+16*3]
1161 paddw m2, m0
1162 paddw m3, m1
1163 movd m1, [v_weightsq] ; weights_ver = &dav1d_sm_weights[height];
1164 add v_weightsq, 2
1165 mova m7, [rsp+16*2]
1166 pshufb m1, m7
1167 mova [rsp+16*7], m3
1168 mova m4, [rsp+16*0]
1169 mova m5, [rsp+16*1]
1170 mova m3, [rsp+16*5]
1171 mova m7, [rsp+16*6]
1172 SMOOTH_2D_END 1, 1, 4, 3, 5, 7, [rsp+16*7]
1173 mova m4, [rsp+16*3]
1174 mova m5, [rsp+16*4]
1175 mova [dstq], m0
1176 lea dstq, [dstq+strideq]
1177 sub hd, 1
1178 jg .w16_loop
1179 RET
1180 ALIGN function_align
1181 .w32:
1182 movu m1, [tlq+1] ; top topleft[1 + x]
1183 movu m2, [tlq+17] ; top
1184 mova [rsp+16*0], m1
1185 mova [rsp+16*1], m2
1186 sub tlq, 4
1187 sub tlq, hq
1188 mova m7, [base+ipred_v_shuf]
1189 pshufd m7, m7, q0000
1190 mova [rsp+16*2], m7
1191 mova [rsp+16*3], m0
1192 mova [rsp+16*4], m4
1193 mova [rsp+16*5], m5
1194 .w32_loop:
1195 SMOOTH_OUTPUT_16B 0, 6, 7, 8, 9, [base+smooth_weights+16*4], [base+smooth_weights+16*5], 10, 2, 3, 4, 5
1196 add dstq, 16
1197 SMOOTH_OUTPUT_16B 1, 6, 7, 8, 9, [base+smooth_weights+16*6], [base+smooth_weights+16*7], 10, 2, 3, 4, 5
1198 lea dstq, [dstq-16+strideq]
1199 add v_weightsq, 2
1200 sub hd, 1
1201 jg .w32_loop
1202 RET
1203 ALIGN function_align
1204 .w64:
1205 movu m1, [tlq+1] ; top topleft[1 + x]
1206 movu m2, [tlq+17] ; top
1207 mova [rsp+16*0], m1
1208 mova [rsp+16*1], m2
1209 movu m1, [tlq+33] ; top
1210 movu m2, [tlq+49] ; top
1211 mova [rsp+16*11], m1
1212 mova [rsp+16*12], m2
1213 sub tlq, 4
1214 sub tlq, hq
1215 mova m7, [base+ipred_v_shuf]
1216 pshufd m7, m7, q0000
1217 mova [rsp+16*2], m7
1218 mova [rsp+16*3], m0
1219 mova [rsp+16*4], m4
1220 mova [rsp+16*5], m5
1221 .w64_loop:
1222 SMOOTH_OUTPUT_16B 0, 6, 7, 8, 9, [base+smooth_weights+16*8], [base+smooth_weights+16*9], 10, 2, 3, 4, 5
1223 add dstq, 16
1224 SMOOTH_OUTPUT_16B 1, 6, 7, 8, 9, [base+smooth_weights+16*10], [base+smooth_weights+16*11], 10, 2, 3, 4, 5
1225 add dstq, 16
1226 SMOOTH_OUTPUT_16B 11, 6, 7, 8, 9, [base+smooth_weights+16*12], [base+smooth_weights+16*13], 10, 2, 3, 4, 5
1227 add dstq, 16
1228 SMOOTH_OUTPUT_16B 12, 6, 7, 8, 9, [base+smooth_weights+16*14], [base+smooth_weights+16*15], 10, 2, 3, 4, 5
1229 lea dstq, [dstq-48+strideq]
1230 add v_weightsq, 2
1231 sub hd, 1
1232 jg .w64_loop
1233 RET
1234
1235 %if ARCH_X86_64
1236 cglobal ipred_z1_8bpc, 3, 8, 11, 16*12, dst, stride, tl, w, h, angle, dx
1237 %define base r7-$$
1238 lea r7, [$$]
1239 mova m8, [base+pw_62]
1240 mova m9, [base+pw_64]
1241 mova m10, [base+pw_512]
1242 %else
1243 cglobal ipred_z1_8bpc, 3, 7, 8, -16*13, dst, _, tl, w, h, angle, dx
1244 %define base r1-$$
1245 %define m8 [base+pw_62]
1246 %define m9 [base+pw_64]
1247 %define m10 [base+pw_512]
1248 %define strideq r3
1249 %define stridemp dword [rsp+16*12]
1250 mov stridemp, r1
1251 LEA r1, $$
1252 %endif
1253 tzcnt wd, wm
1254 movifnidn angled, anglem
1255 movifnidn hd, hm
1256 inc tlq
1257 movsxd wq, [base+ipred_z1_ssse3_table+wq*4]
1258 mov dxd, angled
1259 and dxd, 0x7e
1260 add angled, 165 ; ~90
1261 lea wq, [base+wq+ipred_z1_ssse3_table]
1262 movzx dxd, word [base+dr_intra_derivative+dxq]
1263 xor angled, 0x4ff ; d = 90 - angle
1264 jmp wq
1265 .w4:
1266 lea r3d, [angleq+88]
1267 test r3d, 0x480
1268 jnz .w4_no_upsample ; !enable_intra_edge_filter || angle >= 40
1269 sar r3d, 9
1270 add r3d, hd
1271 cmp r3d, 8
1272 jg .w4_no_upsample ; h > 8 || (w == h && is_sm)
1273 mova m1, [tlq-1]
1274 pshufb m0, m1, [base+z_upsample1]
1275 pshufb m1, [base+z_upsample2]
1276 movddup m2, [base+pb_36_m4]
1277 add dxd, dxd
1278 pmaddubsw m0, m2
1279 pshufd m7, m1, q3333
1280 movd [rsp+16], m7 ; top[max_base_x]
1281 pmaddubsw m1, m2
1282 movd m6, dxd
1283 mov r5d, dxd ; xpos
1284 pshufb m6, [base+pw_256]
1285 paddw m1, m0
1286 movq m0, [tlq]
1287 pmulhrsw m1, m10
1288 paddw m7, m6, m6
1289 punpcklqdq m6, m7 ; xpos0 xpos1
1290 packuswb m1, m1
1291 punpcklbw m0, m1
1292 movifnidn strideq, stridemp
1293 mova [rsp], m0
1294 .w4_upsample_loop:
1295 lea r2d, [r5+dxq]
1296 shr r5d, 6 ; base0
1297 movq m0, [rsp+r5]
1298 lea r5d, [r2+dxq]
1299 shr r2d, 6 ; base1
1300 movhps m0, [rsp+r2]
1301 pand m2, m8, m6 ; frac
1302 psubw m1, m9, m2 ; 64-frac
1303 psllw m2, 8
1304 por m1, m2 ; 64-frac, frac
1305 pmaddubsw m0, m1
1306 paddw m6, m7 ; xpos += dx
1307 pmulhrsw m0, m10
1308 packuswb m0, m0
1309 movd [dstq+strideq*0], m0
1310 pshuflw m0, m0, q1032
1311 movd [dstq+strideq*1], m0
1312 lea dstq, [dstq+strideq*2]
1313 sub hd, 2
1314 jg .w4_upsample_loop
1315 RET
1316 .w4_no_upsample:
1317 mov r3d, 7 ; max_base
1318 test angled, 0x400 ; !enable_intra_edge_filter
1319 jnz .w4_main
1320 lea r3d, [hq+3]
1321 movd m0, r3d
1322 movd m2, angled
1323 shr angled, 8 ; is_sm << 1
1324 pxor m1, m1
1325 pshufb m0, m1
1326 pshufb m2, m1
1327 pcmpeqb m1, m0, [base+z_filter_wh4]
1328 pand m1, m2
1329 pcmpgtb m1, [base+z_filter_t_w48+angleq*8]
1330 pmovmskb r5d, m1
1331 mov r3d, 7
1332 test r5d, r5d
1333 jz .w4_main ; filter_strength == 0
1334 mova m3, [tlq-1]
1335 imul r5d, 0x55555555
1336 movu m7, [base+z_filter_s+8]
1337 shr r5d, 30 ; filter_strength
1338 movddup m0, [base+pb_8]
1339 pminub m7, m0
1340 pshufb m0, m3, [base+z_filter_s]
1341 movddup m4, [base+z_filter_k-8+r5*8+24*0]
1342 pshufb m3, m7
1343 movddup m5, [base+z_filter_k-8+r5*8+24*1]
1344 shufps m2, m0, m3, q2121
1345 movddup m6, [base+z_filter_k-8+r5*8+24*2]
1346 pmaddubsw m0, m4
1347 pmaddubsw m1, m2, m4
1348 pmaddubsw m2, m5
1349 paddd m5, m6
1350 pmaddubsw m4, m3, m5
1351 pmaddubsw m3, m6
1352 paddw m0, m2
1353 paddw m1, m4
1354 paddw m0, m3
1355 pshufd m1, m1, q3333
1356 pmulhrsw m0, m10
1357 pmulhrsw m1, m10
1358 mov r5d, 9
1359 mov tlq, rsp
1360 cmp hd, 4
1361 cmovne r3d, r5d
1362 packuswb m0, m1
1363 mova [tlq], m0
1364 .w4_main:
1365 add tlq, r3
1366 movd m5, dxd
1367 movddup m0, [base+z_base_inc] ; base_inc << 6
1368 movd m7, [tlq] ; top[max_base_x]
1369 shl r3d, 6
1370 movd m4, r3d
1371 pshufb m5, [base+pw_256]
1372 mov r5d, dxd ; xpos
1373 pshufb m7, [base+pw_m256]
1374 sub r5, r3
1375 pshufb m4, [base+pw_256]
1376 mova m3, [base+z1_shuf_w4]
1377 paddw m6, m5, m5
1378 psubw m4, m0 ; max_base_x
1379 punpcklqdq m5, m6 ; xpos0 xpos1
1380 .w4_loop:
1381 lea r3, [r5+dxq]
1382 sar r5, 6 ; base0
1383 movq m0, [tlq+r5]
1384 lea r5, [r3+dxq]
1385 sar r3, 6 ; base1
1386 movhps m0, [tlq+r3]
1387 pand m2, m8, m5 ; frac
1388 psubw m1, m9, m2 ; 64-frac
1389 psllw m2, 8
1390 pshufb m0, m3
1391 por m1, m2 ; 64-frac, frac
1392 pmaddubsw m0, m1
1393 movifnidn strideq, stridemp
1394 pcmpgtw m1, m4, m5 ; base < max_base_x
1395 pmulhrsw m0, m10
1396 paddw m5, m6 ; xpos += dx
1397 pand m0, m1
1398 pandn m1, m7
1399 por m0, m1
1400 packuswb m0, m0
1401 movd [dstq+strideq*0], m0
1402 pshuflw m0, m0, q1032
1403 movd [dstq+strideq*1], m0
1404 sub hd, 2
1405 jz .w4_end
1406 lea dstq, [dstq+strideq*2]
1407 test r5d, r5d
1408 jl .w4_loop
1409 packuswb m7, m7
1410 .w4_end_loop:
1411 movd [dstq+strideq*0], m7
1412 movd [dstq+strideq*1], m7
1413 lea dstq, [dstq+strideq*2]
1414 sub hd, 2
1415 jg .w4_end_loop
1416 .w4_end:
1417 RET
1418 .w8:
1419 lea r3d, [angleq+88]
1420 and r3d, ~0x7f
1421 or r3d, hd
1422 cmp r3d, 8
1423 ja .w8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || h > 8
1424 mova m5, [base+z_upsample1]
1425 movu m3, [base+z_filter_s+6]
1426 movd m4, hd
1427 mova m0, [tlq-1]
1428 movu m1, [tlq+7]
1429 pxor m7, m7
1430 pshufb m4, m7
1431 movddup m7, [base+pb_36_m4]
1432 pminub m4, m3
1433 add dxd, dxd
1434 pshufb m2, m0, m5
1435 pmaddubsw m2, m7
1436 pshufb m0, m3
1437 pmaddubsw m0, m7
1438 movd m6, dxd
1439 pshufb m3, m1, m5
1440 pmaddubsw m3, m7
1441 pshufb m1, m4
1442 pmaddubsw m1, m7
1443 pshufb m6, [base+pw_256]
1444 mov r5d, dxd
1445 paddw m2, m0
1446 paddw m7, m6, m6
1447 paddw m3, m1
1448 punpcklqdq m6, m7 ; xpos0 xpos1
1449 movu m1, [tlq]
1450 pmulhrsw m2, m10
1451 pmulhrsw m3, m10
1452 packuswb m2, m3
1453 punpcklbw m0, m1, m2
1454 punpckhbw m1, m2
1455 movifnidn strideq, stridemp
1456 mova [rsp+16*0], m0
1457 mova [rsp+16*1], m1
1458 .w8_upsample_loop:
1459 lea r2d, [r5+dxq]
1460 shr r5d, 6 ; base0
1461 movu m0, [rsp+r5]
1462 lea r5d, [r2+dxq]
1463 shr r2d, 6 ; base1
1464 movu m1, [rsp+r2]
1465 pand m2, m8, m6
1466 psubw m3, m9, m2
1467 psllw m2, 8
1468 por m3, m2
1469 punpcklqdq m2, m3, m3 ; frac0
1470 pmaddubsw m0, m2
1471 punpckhqdq m3, m3 ; frac1
1472 pmaddubsw m1, m3
1473 paddw m6, m7
1474 pmulhrsw m0, m10
1475 pmulhrsw m1, m10
1476 packuswb m0, m1
1477 movq [dstq+strideq*0], m0
1478 movhps [dstq+strideq*1], m0
1479 lea dstq, [dstq+strideq*2]
1480 sub hd, 2
1481 jg .w8_upsample_loop
1482 RET
1483 .w8_no_upsample:
1484 lea r3d, [hq+7]
1485 movd m0, r3d
1486 and r3d, 7
1487 or r3d, 8 ; imin(h+7, 15)
1488 test angled, 0x400
1489 jnz .w8_main
1490 movd m2, angled
1491 shr angled, 8 ; is_sm << 1
1492 pxor m1, m1
1493 pshufb m0, m1
1494 pshufb m2, m1
1495 movu m1, [base+z_filter_wh8]
1496 psrldq m3, [base+z_filter_t_w48+angleq*8], 4
1497 pcmpeqb m1, m0
1498 pand m1, m2
1499 pcmpgtb m1, m3
1500 pmovmskb r5d, m1
1501 test r5d, r5d
1502 jz .w8_main ; filter_strength == 0
1503 movd m3, [tlq-1]
1504 movu m0, [tlq+16*0]
1505 imul r5d, 0x55555555
1506 movu m1, [tlq+16*1]
1507 shr r5d, 30 ; filter_strength
1508 movd m2, [tlq+r3]
1509 lea tlq, [rsp+16*4]
1510 sub r5, 3
1511 mova [tlq-16*1], m0
1512 pxor m7, m7
1513 mova [tlq+16*0], m1
1514 pshufb m3, m7
1515 pshufb m2, m7
1516 mova [tlq-16*2], m3
1517 movq [tlq+r3-15], m2
1518 call .filter_edge
1519 sar r5d, 1
1520 add r5d, 17
1521 cmp hd, 8
1522 cmova r3d, r5d
1523 .w8_main:
1524 add tlq, r3
1525 movd m5, dxd
1526 movd m7, [tlq]
1527 shl r3d, 6
1528 movu m3, [base+z_filter_s+2]
1529 movd m4, r3d
1530 pshufb m5, [base+pw_256]
1531 mov r5d, dxd
1532 pshufb m7, [base+pw_m256]
1533 sub r5, r3
1534 pshufb m4, [base+pw_256]
1535 psubw m4, [base+z_base_inc]
1536 mova m6, m5
1537 .w8_loop:
1538 mov r3, r5
1539 sar r3, 6
1540 movu m0, [tlq+r3]
1541 pand m1, m8, m5
1542 psubw m2, m9, m1
1543 psllw m1, 8
1544 pshufb m0, m3
1545 por m1, m2
1546 pmaddubsw m0, m1
1547 pcmpgtw m1, m4, m5
1548 paddw m5, m6
1549 pmulhrsw m0, m10
1550 pand m0, m1
1551 pandn m1, m7
1552 por m0, m1
1553 packuswb m0, m0
1554 movq [dstq], m0
1555 dec hd
1556 jz .w8_end
1557 movifnidn strideq, stridemp
1558 add dstq, strideq
1559 add r5, dxq
1560 jl .w8_loop
1561 packuswb m7, m7
1562 .w8_end_loop:
1563 movq [dstq], m7
1564 add dstq, strideq
1565 dec hd
1566 jg .w8_end_loop
1567 .w8_end:
1568 RET
1569 .w16:
1570 lea r3d, [hq+15]
1571 movd m0, r3d
1572 and r3d, 15
1573 or r3d, 16 ; imin(h+15, 31)
1574 test angled, 0x400
1575 jnz .w16_main
1576 movd m2, angled
1577 shr angled, 8 ; is_sm << 1
1578 pxor m1, m1
1579 pshufb m0, m1
1580 pshufb m2, m1
1581 movq m3, [base+z_filter_t_w16+angleq*4]
1582 pcmpeqb m0, [base+z_filter_wh16]
1583 pand m0, m2
1584 pcmpgtb m0, m3
1585 pmovmskb r5d, m0
1586 test r5d, r5d
1587 jz .w16_main ; filter_strength == 0
1588 movd m4, [tlq-1]
1589 movu m0, [tlq+16*0]
1590 imul r5d, 0x24924924
1591 movu m1, [tlq+16*1]
1592 shr r5d, 30
1593 movd m2, [tlq+30]
1594 adc r5, -4 ; filter_strength-3
1595 movd m3, [tlq+r3]
1596 lea tlq, [rsp+16*4]
1597 mova [tlq-16*1], m0
1598 pxor m7, m7
1599 mova [tlq+16*0], m1
1600 pshufb m4, m7
1601 movd [rsp], m2
1602 pshufb m3, m7
1603 mova [tlq-16*2], m4
1604 movd [tlq+r3-16], m3
1605 call .filter_edge
1606 cmp hd, 16
1607 jle .w16_main
1608 pshuflw m0, [rsp], q0000
1609 sar r5, 1
1610 movd m1, [base+z_filter_k_tail+4+r5*4]
1611 lea r3d, [r5+33]
1612 pmaddubsw m0, m1
1613 %if ARCH_X86_64
1614 pmulhrsw m0, m10
1615 %else
1616 pmulhrsw m0, m4
1617 %endif
1618 packuswb m0, m0
1619 movd [tlq+32], m0
1620 .w16_main:
1621 add tlq, r3
1622 movd m5, dxd
1623 movd m7, [tlq]
1624 movd m4, r3d
1625 shl r3d, 6
1626 pshufb m5, [base+pw_256]
1627 pxor m6, m6
1628 pshufb m7, m6
1629 mov r5d, dxd
1630 pshufb m4, m6
1631 sub r5, r3
1632 psubb m4, [base+pb_0to15]
1633 mova m6, m5
1634 .w16_loop:
1635 mov r3, r5
1636 sar r3, 6
1637 movu m1, [tlq+r3+0]
1638 pand m0, m8, m5
1639 movu m2, [tlq+r3+1]
1640 psubw m3, m9, m0
1641 psllw m0, 8
1642 por m3, m0
1643 punpcklbw m0, m1, m2
1644 pmaddubsw m0, m3
1645 punpckhbw m1, m2
1646 pmaddubsw m1, m3
1647 psrlw m3, m5, 6
1648 packsswb m3, m3
1649 pmulhrsw m0, m10
1650 pmulhrsw m1, m10
1651 paddw m5, m6
1652 pcmpgtb m2, m4, m3
1653 packuswb m0, m1
1654 pand m0, m2
1655 pandn m2, m7
1656 por m0, m2
1657 mova [dstq], m0
1658 dec hd
1659 jz .w16_end
1660 movifnidn strideq, stridemp
1661 add dstq, strideq
1662 add r5, dxq
1663 jl .w16_loop
1664 .w16_end_loop:
1665 mova [dstq], m7
1666 add dstq, strideq
1667 dec hd
1668 jg .w16_end_loop
1669 .w16_end:
1670 RET
1671 .w32:
1672 lea r3d, [hq+31]
1673 and r3d, 31
1674 or r3d, 32 ; imin(h+31, 63)
1675 test angled, 0x400 ; !enable_intra_edge_filter
1676 jnz .w32_main
1677 movd m6, [tlq-1]
1678 movu m0, [tlq+16*0]
1679 movu m1, [tlq+16*1]
1680 movu m2, [tlq+16*2]
1681 movu m3, [tlq+16*3]
1682 movd m4, [tlq+62]
1683 movd m5, [tlq+r3]
1684 lea tlq, [rsp+16*6]
1685 mova [tlq-16*3], m0
1686 pxor m7, m7
1687 mova [tlq-16*2], m1
1688 pshufb m6, m7
1689 mova [tlq-16*1], m2
1690 xor r5d, r5d ; filter_strength = 3
1691 mova [tlq+16*0], m3
1692 movd [rsp], m4
1693 pshufb m5, m7
1694 mova [tlq-16*4], m6
1695 movd [tlq+r3-48], m5
1696 call .filter_edge
1697 sub tlq, 16*2
1698 call .filter_edge
1699 cmp hd, 32
1700 jle .w32_main
1701 pshuflw m0, [rsp], q0000
1702 movd m1, [base+z_filter_k_tail+4]
1703 add r3d, 2
1704 pmaddubsw m0, m1
1705 %if ARCH_X86_64
1706 pmulhrsw m0, m10
1707 %else
1708 pmulhrsw m0, m4
1709 %endif
1710 packuswb m0, m0
1711 movd [tlq+64], m0
1712 .w32_main:
1713 add tlq, r3
1714 movd m0, r3d
1715 movd m7, [tlq]
1716 shl r3d, 6
1717 movd m5, dxd
1718 pxor m6, m6
1719 mov r5d, dxd
1720 pshufb m0, m6
1721 pshufb m5, [base+pw_256]
1722 sub r5, r3
1723 pshufb m7, m6
1724 psubb m0, [base+pb_0to15]
1725 movddup m1, [base+pb_m16]
1726 mova [rsp+16*0], m0
1727 paddb m0, m1
1728 mova [rsp+16*1], m0
1729 mova m6, m5
1730 .w32_loop:
1731 mov r3, r5
1732 sar r3, 6
1733 movu m1, [tlq+r3+16*0+0]
1734 pand m0, m8, m5
1735 movu m2, [tlq+r3+16*0+1]
1736 psubw m3, m9, m0
1737 psllw m0, 8
1738 por m3, m0
1739 punpcklbw m0, m1, m2
1740 pmaddubsw m0, m3
1741 punpckhbw m1, m2
1742 pmaddubsw m1, m3
1743 psrlw m4, m5, 6
1744 pmulhrsw m0, m10
1745 pmulhrsw m1, m10
1746 packsswb m4, m4
1747 pcmpgtb m2, [rsp+16*0], m4
1748 packuswb m0, m1
1749 pand m0, m2
1750 pandn m2, m7
1751 por m0, m2
1752 movu m1, [tlq+r3+16*1+0]
1753 movu m2, [tlq+r3+16*1+1]
1754 mova [dstq+16*0], m0
1755 punpcklbw m0, m1, m2
1756 pmaddubsw m0, m3
1757 punpckhbw m1, m2
1758 pmaddubsw m1, m3
1759 paddw m5, m6
1760 pmulhrsw m0, m10
1761 pmulhrsw m1, m10
1762 pcmpgtb m2, [rsp+16*1], m4
1763 packuswb m0, m1
1764 pand m0, m2
1765 pandn m2, m7
1766 por m0, m2
1767 mova [dstq+16*1], m0
1768 dec hd
1769 jz .w32_end
1770 movifnidn strideq, stridemp
1771 add dstq, strideq
1772 add r5, dxq
1773 jl .w32_loop
1774 .w32_end_loop:
1775 mova [dstq+16*0], m7
1776 mova [dstq+16*1], m7
1777 add dstq, strideq
1778 dec hd
1779 jg .w32_end_loop
1780 .w32_end:
1781 RET
1782 .w64:
1783 lea r3d, [hq+63]
1784 test angled, 0x400 ; !enable_intra_edge_filter
1785 jnz .w64_main
1786 movd m4, [tlq-1]
1787 movu m0, [tlq+16*0]
1788 movu m1, [tlq+16*1]
1789 movu m2, [tlq+16*2]
1790 movu m3, [tlq+16*3]
1791 mova [rsp+16*3], m0
1792 pxor m7, m7
1793 mova [rsp+16*4], m1
1794 pshufb m4, m7
1795 mova [rsp+16*5], m2
1796 mova [rsp+16*6], m3
1797 mova [rsp+16*2], m4
1798 movu m0, [tlq+16*4]
1799 movu m1, [tlq+16*5]
1800 movu m2, [tlq+16*6]
1801 movu m3, [tlq+16*7]
1802 movd m4, [tlq+r3]
1803 lea tlq, [rsp+16*10]
1804 mova [tlq-16*3], m0
1805 xor r5d, r5d ; filter_strength = 3
1806 mova [tlq-16*2], m1
1807 pshufb m4, m7
1808 mova [tlq-16*1], m2
1809 mova [tlq+16*0], m3
1810 movd [tlq+r3-16*7], m4
1811 cmp hd, 64
1812 jl .w64_filter96 ; skip one call if the last 32 bytes aren't used
1813 call .filter_edge
1814 .w64_filter96:
1815 sub tlq, 16*2
1816 call .filter_edge
1817 sub tlq, 16*2
1818 call .filter_edge
1819 sub tlq, 16*2
1820 call .filter_edge
1821 .w64_main:
1822 add tlq, r3
1823 movd m0, r3d
1824 movd m7, [tlq]
1825 shl r3d, 6
1826 movd m5, dxd
1827 pxor m6, m6
1828 mov r5d, dxd
1829 pshufb m0, m6
1830 sub r5, r3
1831 pshufb m5, [base+pw_256]
1832 pshufb m7, m6
1833 psubb m0, [base+pb_0to15]
1834 movddup m1, [base+pb_m16]
1835 mova [rsp+16*0], m0
1836 paddb m0, m1
1837 mova [rsp+16*1], m0
1838 paddb m0, m1
1839 mova [rsp+16*2], m0
1840 paddb m0, m1
1841 mova [rsp+16*3], m0
1842 mova m6, m5
1843 .w64_loop:
1844 mov r3, r5
1845 sar r3, 6
1846 movu m1, [tlq+r3+16*0+0]
1847 pand m0, m8, m5
1848 movu m2, [tlq+r3+16*0+1]
1849 psubw m3, m9, m0
1850 psllw m0, 8
1851 por m3, m0
1852 punpcklbw m0, m1, m2
1853 pmaddubsw m0, m3
1854 punpckhbw m1, m2
1855 pmaddubsw m1, m3
1856 psrlw m4, m5, 6
1857 pmulhrsw m0, m10
1858 pmulhrsw m1, m10
1859 packsswb m4, m4
1860 pcmpgtb m2, [rsp+16*0], m4
1861 packuswb m0, m1
1862 pand m0, m2
1863 pandn m2, m7
1864 por m0, m2
1865 movu m1, [tlq+r3+16*1+0]
1866 movu m2, [tlq+r3+16*1+1]
1867 mova [dstq+16*0], m0
1868 punpcklbw m0, m1, m2
1869 pmaddubsw m0, m3
1870 punpckhbw m1, m2
1871 pmaddubsw m1, m3
1872 pmulhrsw m0, m10
1873 pmulhrsw m1, m10
1874 pcmpgtb m2, [rsp+16*1], m4
1875 packuswb m0, m1
1876 pand m0, m2
1877 pandn m2, m7
1878 por m0, m2
1879 movu m1, [tlq+r3+16*2+0]
1880 movu m2, [tlq+r3+16*2+1]
1881 mova [dstq+16*1], m0
1882 punpcklbw m0, m1, m2
1883 pmaddubsw m0, m3
1884 punpckhbw m1, m2
1885 pmaddubsw m1, m3
1886 pmulhrsw m0, m10
1887 pmulhrsw m1, m10
1888 pcmpgtb m2, [rsp+16*2], m4
1889 packuswb m0, m1
1890 pand m0, m2
1891 pandn m2, m7
1892 por m0, m2
1893 movu m1, [tlq+r3+16*3+0]
1894 movu m2, [tlq+r3+16*3+1]
1895 mova [dstq+16*2], m0
1896 punpcklbw m0, m1, m2
1897 pmaddubsw m0, m3
1898 punpckhbw m1, m2
1899 pmaddubsw m1, m3
1900 paddw m5, m6
1901 pmulhrsw m0, m10
1902 pmulhrsw m1, m10
1903 pcmpgtb m2, [rsp+16*3], m4
1904 packuswb m0, m1
1905 pand m0, m2
1906 pandn m2, m7
1907 por m0, m2
1908 mova [dstq+16*3], m0
1909 dec hd
1910 jz .w64_end
1911 movifnidn strideq, stridemp
1912 add dstq, strideq
1913 add r5, dxq
1914 jl .w64_loop
1915 .w64_end_loop:
1916 mova [dstq+16*0], m7
1917 mova [dstq+16*1], m7
1918 mova [dstq+16*2], m7
1919 mova [dstq+16*3], m7
1920 add dstq, strideq
1921 dec hd
1922 jg .w64_end_loop
1923 .w64_end:
1924 RET
1925 ALIGN function_align
1926 .filter_edge: ; 32 pixels/iteration
1927 movddup m7, [base+z_filter_k+8*2+r5*8+24*0]
1928 movu m2, [tlq-18]
1929 movu m1, [tlq-17]
1930 movu m3, [tlq- 2]
1931 movu m4, [tlq- 1]
1932 punpcklbw m0, m2, m1
1933 pmaddubsw m0, m7
1934 punpckhbw m2, m1
1935 pmaddubsw m2, m7
1936 punpcklbw m1, m3, m4
1937 pmaddubsw m1, m7
1938 punpckhbw m3, m4
1939 pmaddubsw m3, m7
1940 movddup m7, [base+z_filter_k+8*2+r5*8+24*1]
1941 mova m5, [tlq-16]
1942 movu m6, [tlq-15]
1943 punpcklbw m4, m5, m6
1944 pmaddubsw m4, m7
1945 punpckhbw m5, m6
1946 pmaddubsw m5, m7
1947 paddw m0, m4
1948 paddw m2, m5
1949 mova m5, [tlq+ 0]
1950 movu m6, [tlq+ 1]
1951 punpcklbw m4, m5, m6
1952 pmaddubsw m4, m7
1953 punpckhbw m5, m6
1954 pmaddubsw m5, m7
1955 paddw m1, m4
1956 paddw m3, m5
1957 test r5d, r5d
1958 jnz .filter_end ; 3-tap
1959 movddup m7, [base+z_filter_k+8*8]
1960 movu m5, [tlq-14]
1961 movu m6, [tlq+ 2]
1962 punpcklbw m4, m5, m5
1963 pmaddubsw m4, m7
1964 punpckhbw m5, m5
1965 pmaddubsw m5, m7
1966 paddw m0, m4
1967 paddw m2, m5
1968 punpcklbw m5, m6, m6
1969 pmaddubsw m5, m7
1970 punpckhbw m6, m6
1971 pmaddubsw m6, m7
1972 paddw m1, m5
1973 paddw m3, m6
1974 .filter_end:
1975 %if ARCH_X86_64
1976 REPX {pmulhrsw x, m10}, m0, m2, m1, m3
1977 %else
1978 mova m4, m10
1979 REPX {pmulhrsw x, m4 }, m0, m2, m1, m3
1980 %endif
1981 packuswb m0, m2
1982 packuswb m1, m3
1983 mova [tlq+16*0], m0
1984 mova [tlq+16*1], m1
1985 ret
1986
1987 %if ARCH_X86_64
1988 cglobal ipred_z2_8bpc, 4, 12, 13, 16*16, dst, stride, tl, w, h, angle, dx, _, dy
1989 %define base r7-$$
1990 %define maxwm r6m
1991 %define maxhm r7m
1992 lea r7, [$$]
1993 mov hd, hm
1994 mova m8, [base+pw_62]
1995 mova m9, [base+pw_64]
1996 lea r9d, [wq-4]
1997 mova m10, [base+pw_512]
1998 shl r9d, 6
1999 mova m11, [base+z1_shuf_w4]
2000 or r9d, hd
2001 mova m12, [base+z2_h_shuf]
2002 %else
2003 cglobal ipred_z2_8bpc, 4, 7, 8, -16*20, dst, _, tl, w, h, angle, dx
2004 %define base r1-$$
2005 %define m8 [base+pw_62]
2006 %define m9 [base+pw_64]
2007 %define m10 [base+pw_512]
2008 %define m11 [rsp+16*16]
2009 %define m12 [rsp+16*17]
2010 %define r9b byte [rsp+16*18+4*0]
2011 %define r9d dword [rsp+16*18+4*0]
2012 %define r10d dword [rsp+16*18+4*1]
2013 %define r11d dword [rsp+16*18+4*2]
2014 %define maxwm [rsp+16*18+4*3]
2015 %define maxhm [rsp+16*19+4*0]
2016 %define stridemp [rsp+16*19+4*1]
2017 %define strideq r3
2018 %define dyd r4
2019 %define dyq r4
2020 mov stridemp, r1
2021 mov r1d, r6m
2022 mov r4d, r7m
2023 mov maxwm, r1d
2024 mov maxhm, r4d
2025 LEA r1, $$
2026 lea hd, [wq-4]
2027 mova m0, [base+z1_shuf_w4]
2028 shl hd, 6
2029 mova m1, [base+z2_h_shuf]
2030 or hd, hm
2031 mova m11, m0
2032 mov r9d, hd
2033 mova m12, m1
2034 %endif
2035 tzcnt wd, wd
2036 movifnidn angled, anglem
2037 movsxd wq, [base+ipred_z2_ssse3_table+wq*4]
2038 %if ARCH_X86_64
2039 movzx dxd, angleb
2040 %else
2041 movzx dxd, byte anglem
2042 %endif
2043 xor angled, 0x400
2044 mova m0, [tlq-16*4]
2045 mov dyd, dxd
2046 mova m1, [tlq-16*3]
2047 neg dxq
2048 mova m2, [tlq-16*2]
2049 and dyd, ~1
2050 mova m3, [tlq-16*1]
2051 and dxq, ~1
2052 movd m4, [tlq]
2053 movu m5, [tlq+16*0+1]
2054 movu m6, [tlq+16*1+1]
2055 movzx dyd, word [base+dr_intra_derivative+dyq-90] ; angle - 90
2056 movzx dxd, word [base+dr_intra_derivative+dxq+180] ; 180 - angle
2057 mova [rsp+16*2], m0
2058 pxor m7, m7
2059 mova [rsp+16*3], m1
2060 pshufb m4, m7
2061 mova [rsp+16*4], m2
2062 lea wq, [base+ipred_z2_ssse3_table+wq]
2063 mova [rsp+16*5], m3
2064 neg dxd
2065 mova [rsp+16*6], m4
2066 or dyd, 4<<16
2067 mova [rsp+16*7], m4
2068 mova [rsp+16*8], m5
2069 mova [rsp+16*9], m6
2070 movq m0, [base+z_base_inc+2]
2071 movsldup m1, [base+z2_dy_offset]
2072 movq m2, [base+pw_256] ; 4<<6
2073 movq [rsp+16*14+8*0], m0
2074 movq [rsp+16*15+8*0], m1
2075 movq [rsp+16*15+8*1], m2
2076 %if ARCH_X86_64
2077 lea r10d, [dxq+(128<<6)] ; xpos
2078 %else
2079 mov [rsp+16*7+4*1], dyd
2080 lea r4d, [dxq+(128<<6)]
2081 mov r10d, r4d
2082 movzx hd, r9b
2083 %endif
2084 mov r11d, (128-4)<<6
2085 jmp wq
2086 .w4:
2087 test angled, 0x400
2088 jnz .w4_main
2089 movd m5, [tlq+4]
2090 lea r3d, [hq+2]
2091 add angled, 1022
2092 pshufb m5, m7
2093 shl r3d, 6
2094 movd [rsp+16*8+4], m5
2095 test r3d, angled
2096 jnz .w4_no_upsample_above ; angle >= 130 || h > 8 || (is_sm && h == 8)
2097 call .upsample_above
2098 sub angled, 1075 ; angle - 53
2099 lea r3d, [hq+3]
2100 xor angled, 0x7f ; 180 - angle
2101 movd m0, r3d
2102 movd m6, angled
2103 shr angled, 8 ; is_sm << 1
2104 pshufb m0, m7
2105 pshufb m6, m7
2106 pcmpeqb m0, [base+z_filter_wh4]
2107 pand m6, m0
2108 pcmpgtb m6, [base+z_filter_t_w48+angleq*8]
2109 jmp .w8_filter_left
2110 .upsample_above: ; w4/w8
2111 movq m3, [rsp+gprsize+16*8-2]
2112 movq m1, [rsp+gprsize+16*8-1]
2113 movq m0, [rsp+gprsize+16*8+0]
2114 movq m4, [rsp+gprsize+16*8+1]
2115 movddup m5, [base+pb_36_m4]
2116 punpcklbw m1, m3
2117 punpcklbw m2, m0, m4
2118 pmaddubsw m1, m5
2119 pmaddubsw m2, m5
2120 %if ARCH_X86_64
2121 mova m11, [base+pb_0to15]
2122 lea r10d, [r10+dxq+(1<<6)]
2123 mov r11d, (128-7)<<6
2124 %else
2125 mova m3, [base+pb_0to15]
2126 mov r3d, [rsp+gprsize+16*18+4*1]
2127 mov dword [rsp+gprsize+16*18+4*2], (128-7)<<6
2128 lea r3d, [r3+dxq+(1<<6)]
2129 mov [rsp+gprsize+16*18+4*1], r3d
2130 mova [rsp+gprsize+16*16], m3
2131 %endif
2132 add dxd, dxd
2133 paddw m1, m2
2134 pmulhrsw m1, m10
2135 movq m2, [rsp+gprsize+16*14]
2136 paddw m2, m2
2137 movq [rsp+gprsize+16*14], m2
2138 packuswb m1, m1
2139 punpcklbw m1, m0
2140 mova [rsp+gprsize+16*8], m1
2141 ret
2142 .w4_no_upsample_above:
2143 lea r3d, [hq+3]
2144 mov [rsp], angled
2145 sub angled, 1112 ; angle - 90
2146 movd m0, r3d
2147 mov r3d, 90
2148 movd m1, angled
2149 sub r3d, angled ; 180 - angle
2150 shr angled, 8 ; is_sm << 1
2151 movu m3, [base+z_filter_wh4]
2152 mova m4, [base+z_filter_t_w48+angleq*8]
2153 call .w8_filter_top
2154 mov angled, [rsp]
2155 lea r3d, [hq+2]
2156 sub angled, 139
2157 shl r3d, 6
2158 test r3d, angled
2159 jnz .w8_filter_left ; angle <= 140 || h > 8 || (is_sm && h == 8)
2160 .upsample_left: ; w4/w8
2161 neg hq
2162 movd m0, [tlq+hq]
2163 pshufb m0, m7
2164 movd [rsp+16*6+hq-4], m0
2165 movq m3, [rsp+16*5+7]
2166 movq m0, [rsp+16*5+8]
2167 movq m2, [rsp+16*5+9]
2168 movq m4, [rsp+16*5+10]
2169 movddup m5, [base+pb_36_m4]
2170 punpcklbw m1, m0, m3
2171 punpcklbw m2, m4
2172 pmaddubsw m1, m5
2173 pmaddubsw m2, m5
2174 movshdup m3, [base+z2_dy_offset]
2175 %if ARCH_X86_64
2176 mova m12, [base+z2_upsample]
2177 add dyd, dyd
2178 %else
2179 mova m4, [base+z2_upsample]
2180 shl dword [rsp+16*7+4*1], 1
2181 mova m12, m4
2182 %endif
2183 paddw m1, m2
2184 pmulhrsw m1, m10
2185 movq [rsp+16*15], m3
2186 packuswb m1, m1
2187 punpcklbw m0, m1
2188 mova [rsp+16*5], m0
2189 .w4_main:
2190 movd m6, dxd
2191 %if ARCH_X86_64
2192 movd m3, dyd
2193 %else
2194 movd m3, [rsp+16*7+4*1]
2195 %endif
2196 movddup m0, [rsp+16*14+8*0]
2197 pshufb m6, [base+pw_256]
2198 paddw m7, m6, m6
2199 movq m5, [base+pw_m1to4]
2200 pshuflw m4, m3, q0000
2201 punpcklqdq m6, m7
2202 pmullw m4, m5
2203 pshuflw m3, m3, q1111
2204 paddw m6, m0
2205 mov r2d, r10d
2206 pshuflw m0, m4, q3333
2207 psubw m4, [rsp+16*15]
2208 movq [rsp+16*6+8*1], m3
2209 movq [rsp+8*1], m0 ; dy*4
2210 mov r5, dstq
2211 .w4_loop0:
2212 mova [rsp+16*12], m6
2213 movq [rsp+8*0], m4
2214 pand m0, m4, m8
2215 psraw m4, 6
2216 psubw m1, m9, m0
2217 psllw m0, 8
2218 por m0, m1 ; 64-frac_y, frac_y
2219 movq [rsp+8*3], m0
2220 pabsw m4, m4
2221 movq [rsp+8*2], m4
2222 movzx hd, r9b
2223 .w4_loop:
2224 lea r3d, [r2+dxq]
2225 shr r2d, 6 ; base_x0
2226 movq m0, [rsp+r2]
2227 lea r2d, [r3+dxq]
2228 shr r3d, 6 ; base_x1
2229 movhps m0, [rsp+r3]
2230 lea r3d, [r2+dxq]
2231 shr r2d, 6 ; base_x2
2232 movq m1, [rsp+r2]
2233 lea r2d, [r3+dxq]
2234 shr r3d, 6 ; base_x3
2235 movhps m1, [rsp+r3]
2236 pand m2, m8, m6
2237 paddsw m5, m6, m7
2238 psubw m3, m9, m2
2239 psllw m2, 8
2240 pshufb m0, m11
2241 por m2, m3
2242 pmaddubsw m0, m2
2243 pand m2, m8, m5
2244 psubw m3, m9, m2
2245 psllw m2, 8
2246 pshufb m1, m11
2247 por m2, m3
2248 pmaddubsw m1, m2
2249 cmp r3d, 127 ; topleft
2250 jge .w4_toponly
2251 movzx r3d, byte [rsp+8*2+0] ; base_y0
2252 movq m3, [rsp+r3]
2253 movzx r3d, byte [rsp+8*2+2] ; base_y1
2254 movhps m3, [rsp+r3]
2255 movzx r3d, byte [rsp+8*2+4] ; base_y2
2256 movq m4, [rsp+r3]
2257 movzx r3d, byte [rsp+8*2+6] ; base_y3
2258 movhps m4, [rsp+r3]
2259 pshufb m3, m12
2260 pshufb m4, m12
2261 punpckldq m2, m3, m4
2262 punpckhdq m3, m4
2263 movddup m4, [rsp+8*3]
2264 pmaddubsw m2, m4
2265 pmaddubsw m3, m4
2266 psraw m6, 15 ; base_x < topleft
2267 pand m2, m6
2268 pandn m6, m0
2269 por m0, m2, m6
2270 psraw m6, m5, 15
2271 pand m3, m6
2272 pandn m6, m1
2273 por m1, m3, m6
2274 .w4_toponly:
2275 pmulhrsw m0, m10
2276 pmulhrsw m1, m10
2277 movifnidn strideq, stridemp
2278 packuswb m0, m1
2279 movd [dstq+strideq*0], m0
2280 pshuflw m1, m0, q1032
2281 movd [dstq+strideq*1], m1
2282 lea dstq, [dstq+strideq*2]
2283 punpckhqdq m0, m0
2284 movd [dstq+strideq*0], m0
2285 psrlq m0, 32
2286 movd [dstq+strideq*1], m0
2287 sub hd, 4
2288 jz .w4_end
2289 movq m4, [rsp+8*2]
2290 movq m3, [rsp+16*6+8*1]
2291 paddw m6, m5, m7 ; xpos += dx
2292 psubw m4, m3
2293 movq [rsp+8*2], m4
2294 lea dstq, [dstq+strideq*2]
2295 cmp r2d, r11d
2296 jge .w4_loop
2297 movddup m5, [rsp+8*3]
2298 .w4_leftonly_loop:
2299 movzx r2d, byte [rsp+8*2+0] ; base_y0
2300 movq m1, [rsp+r2]
2301 movzx r2d, byte [rsp+8*2+2] ; base_y1
2302 movhps m1, [rsp+r2]
2303 movzx r2d, byte [rsp+8*2+4] ; base_y2
2304 movq m2, [rsp+r2]
2305 movzx r2d, byte [rsp+8*2+6] ; base_y3
2306 movhps m2, [rsp+r2]
2307 psubw m4, m3
2308 pshufb m1, m12
2309 pshufb m2, m12
2310 movq [rsp+8*2], m4
2311 punpckldq m0, m1, m2
2312 punpckhdq m1, m2
2313 pmaddubsw m0, m5
2314 pmaddubsw m1, m5
2315 pmulhrsw m0, m10
2316 pmulhrsw m1, m10
2317 packuswb m0, m1
2318 movd [dstq+strideq*0], m0
2319 pshuflw m1, m0, q1032
2320 movd [dstq+strideq*1], m1
2321 lea dstq, [dstq+strideq*2]
2322 punpckhqdq m0, m0
2323 movd [dstq+strideq*0], m0
2324 psrlq m0, 32
2325 movd [dstq+strideq*1], m0
2326 lea dstq, [dstq+strideq*2]
2327 sub hd, 4
2328 jg .w4_leftonly_loop
2329 .w4_end:
2330 sub r9d, 1<<8
2331 jl .w4_ret
2332 movq m4, [rsp+8*1]
2333 add r5, 4
2334 mov dstq, r5
2335 paddw m4, [rsp+8*0] ; base_y += 4*dy
2336 movzx r2d, word [rsp+16*15+8*1]
2337 movddup m6, [rsp+16*15+8*1]
2338 paddw m6, [rsp+16*12] ; base_x += (4 << upsample_above)
2339 add r2d, r10d
2340 mov r10d, r2d
2341 jmp .w4_loop0
2342 .w4_ret:
2343 RET
2344 .w8:
2345 test angled, 0x400
2346 jnz .w4_main
2347 movd m5, [tlq+8]
2348 lea r3d, [angleq+126]
2349 pshufb m5, m7
2350 %if ARCH_X86_64
2351 mov r3b, hb
2352 %else
2353 xor r3b, r3b
2354 or r3d, hd
2355 %endif
2356 movd [rsp+16*8+8], m5
2357 cmp r3d, 8
2358 ja .w8_no_upsample_above ; angle >= 130 || h > 8 || is_sm
2359 call .upsample_above
2360 sub angled, 53
2361 lea r3d, [hq+7]
2362 xor angled, 0x7f ; 180 - angle
2363 movu m1, [base+z_filter_wh8]
2364 movd m0, r3d
2365 movd m6, angled
2366 shr angled, 8 ; is_sm << 1
2367 psrldq m2, [base+z_filter_t_w48+angleq*8], 4
2368 pshufb m0, m7
2369 pshufb m6, m7
2370 pcmpeqb m0, m1
2371 pand m6, m0
2372 pcmpgtb m6, m2
2373 %if ARCH_X86_64
2374 movq [rsp+16*15+8*1], m10 ; 8<<6
2375 %else
2376 movq m0, m10
2377 movq [rsp+16*15+8*1], m0
2378 %endif
2379 jmp .w8_filter_left
2380 .w8_no_upsample_above:
2381 lea r3d, [hq+7]
2382 mov [rsp], angled
2383 sub angled, 90
2384 movd m0, r3d
2385 mov r3d, 90
2386 movd m1, angled
2387 sub r3d, angled ; 180 - angle
2388 shr angled, 8 ; is_sm << 1
2389 movu m3, [base+z_filter_wh8]
2390 psrldq m4, [base+z_filter_t_w48+angleq*8], 4
2391 call .w8_filter_top
2392 mov r3d, [rsp]
2393 sub r3d, 141
2394 %if ARCH_X86_64
2395 mov r3b, hb
2396 %else
2397 xor r3b, r3b
2398 or r3d, hd
2399 %endif
2400 cmp r3d, 8
2401 jbe .upsample_left ; angle > 140 && h <= 8 && !is_sm
2402 .w8_filter_left:
2403 pmovmskb r5d, m6
2404 test r5d, r5d
2405 jz .w4_main
2406 imul r5d, 0x55555555
2407 mov r3, tlq
2408 shr r5d, 30
2409 sub r5, 3 ; filter_strength-3
2410 jmp .filter_left
2411 .w8_filter_top:
2412 movd m6, r3d
2413 REPX {pshufb x, m7}, m0, m1, m6
2414 pcmpeqb m0, m3
2415 pand m1, m0
2416 pand m6, m0
2417 pcmpgtb m1, m4
2418 pcmpgtb m6, m4
2419 pmovmskb r5d, m1
2420 test r5d, r5d
2421 jz .w8_filter_top_end ; filter_strength == 0
2422 imul r5d, 0x55555555
2423 movq m0, [rsp+gprsize+16*8-2]
2424 shr r5d, 30
2425 movq m1, [rsp+gprsize+16*8-1]
2426 sub r5, 3 ; filter_strength-3
2427 movddup m7, [base+z_filter_k+8*2+r5*8+24*0]
2428 punpcklbw m0, m1
2429 pmaddubsw m0, m7
2430 movq m1, [rsp+gprsize+16*8+0]
2431 movq m2, [rsp+gprsize+16*8+1]
2432 movddup m7, [base+z_filter_k+8*2+r5*8+24*1]
2433 punpcklbw m1, m2
2434 pmaddubsw m1, m7
2435 movq m2, [rsp+gprsize+16*8+2]
2436 movddup m7, [base+z_filter_k+8*2+r5*8+24*2]
2437 punpcklbw m2, m2
2438 pmaddubsw m2, m7
2439 paddw m0, m1
2440 paddw m0, m2
2441 %if ARCH_X86_64
2442 mov r3d, r7m ; maxw, offset due to call
2443 %else
2444 mov r3d, [rsp+gprsize+16*18+4*3]
2445 %endif
2446 pmulhrsw m0, m10
2447 pmulhrsw m1, m10
2448 packuswb m0, m1
2449 movq [rsp+gprsize+16*8], m0
2450 cmp r3d, 8
2451 jge .w8_filter_top_end
2452 movq m0, [tlq+r3+1]
2453 movq [rsp+gprsize+r3+16*8], m0
2454 .w8_filter_top_end:
2455 ret
2456 .w16:
2457 test angled, 0x400
2458 jnz .w4_main
2459 lea r3d, [hq+15]
2460 sub angled, 90
2461 movd m0, r3d
2462 mov r3d, 90
2463 movd m1, angled
2464 sub r3d, angled ; 180 - angle
2465 shr angled, 8 ; is_sm << 1
2466 movd m6, r3d
2467 REPX {pshufb x, m7}, m0, m1, m6
2468 movq m3, [base+z_filter_t_w16+angleq*4]
2469 pcmpeqb m0, [base+z_filter_wh16]
2470 pand m1, m0
2471 pand m6, m0
2472 pcmpgtb m1, m3
2473 pcmpgtb m6, m3
2474 pmovmskb r5d, m1
2475 mov r3, tlq
2476 test r5d, r5d
2477 jz .w16_filter_left ; filter_strength == 0
2478 imul r5d, 0x24924924
2479 pshufb m5, [base+z_filter_t_w16] ; tlq[16]
2480 shr r5d, 30
2481 adc r5, -4 ; filter_strength-3
2482 movd [rsp+16*9], m5
2483 movddup m7, [base+z_filter_k+8*2+r5*8+24*0]
2484 movu m1, [rsp+16*8-2]
2485 movu m2, [rsp+16*8-1]
2486 punpcklbw m0, m1, m2
2487 pmaddubsw m0, m7
2488 punpckhbw m1, m2
2489 pmaddubsw m1, m7
2490 movddup m7, [base+z_filter_k+8*2+r5*8+24*1]
2491 mova m3, [rsp+16*8+0]
2492 movu m4, [rsp+16*8+1]
2493 punpcklbw m2, m3, m4
2494 pmaddubsw m2, m7
2495 punpckhbw m3, m4
2496 pmaddubsw m3, m7
2497 paddw m0, m2
2498 paddw m1, m3
2499 test r5d, r5d
2500 jnz .w16_filter_end ; 3-tap
2501 movddup m7, [base+z_filter_k+8*8]
2502 movu m3, [rsp+16*8+2]
2503 punpcklbw m2, m3, m3
2504 pmaddubsw m2, m7
2505 punpckhbw m3, m3
2506 pmaddubsw m3, m7
2507 paddw m0, m2
2508 paddw m1, m3
2509 .w16_filter_end:
2510 mov r2d, maxwm
2511 pmulhrsw m0, m10
2512 pmulhrsw m1, m10
2513 packuswb m0, m1
2514 mova [rsp+16*8], m0
2515 cmp r2d, 16
2516 jge .w16_filter_left
2517 movu m0, [r3+r2+1]
2518 movu [rsp+r2+16*8], m0
2519 .w16_filter_left:
2520 pmovmskb r5d, m6
2521 test r5d, r5d
2522 jz .w4_main
2523 imul r5d, 0x24924924
2524 shr r5d, 30
2525 adc r5, -4 ; filter_strength-3
2526 jmp .filter_left
2527 .w32:
2528 test angled, 0x400
2529 jnz .w4_main
2530 pshufb m6, [base+z_filter_t_w16] ; tlq[32]
2531 mov r3, tlq
2532 lea tlq, [rsp+16*9]
2533 movd [tlq+16*1], m6
2534 xor r5d, r5d ; filter_strength = 3
2535 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2536 mova m0, [tlq+16*0]
2537 mova m1, [tlq+16*1]
2538 mov r2d, maxwm
2539 mova [rsp+16*8], m0
2540 mova [rsp+16*9], m1
2541 cmp r2d, 32
2542 jge .filter_left
2543 movu m0, [r3+r2+16*0+1]
2544 movu m1, [r3+r2+16*1+1]
2545 movu [rsp+r2+16*8], m0
2546 movu [rsp+r2+16*9], m1
2547 jmp .filter_left
2548 .w64:
2549 movu m0, [tlq+16*2+1]
2550 movu m1, [tlq+16*3+1]
2551 mova [rsp+16*10], m0
2552 mova [rsp+16*11], m1
2553 test angled, 0x400
2554 jnz .w4_main
2555 pshufb m1, [base+z_filter_t_w16] ; tlq[64]
2556 mov r3, tlq
2557 lea tlq, [rsp+16*11]
2558 movd [tlq+16*1], m1
2559 xor r5d, r5d ; filter_strength = 3
2560 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2561 sub tlq, 16*2
2562 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2563 mova m0, [tlq+16*0]
2564 mova m1, [tlq+16*1]
2565 mova m2, [tlq+16*2]
2566 mova m3, [tlq+16*3]
2567 mov r2d, maxwm
2568 mova [rsp+16* 8], m0
2569 mova [rsp+16* 9], m1
2570 mova [rsp+16*10], m2
2571 mova [rsp+16*11], m3
2572 cmp r2d, 64
2573 jge .filter_left
2574 movu m0, [r3+r2+16*0+1]
2575 movu m1, [r3+r2+16*1+1]
2576 movu [rsp+r2+16* 8], m0
2577 movu [rsp+r2+16* 9], m1
2578 cmp r2d, 32
2579 jge .filter_left
2580 movu m0, [r3+r2+16*2+1]
2581 movu m1, [r3+r2+16*3+1]
2582 movu [rsp+r2+16*10], m0
2583 movu [rsp+r2+16*11], m1
2584 .filter_left:
2585 neg hq
2586 movd m0, [r3+hq]
2587 pxor m1, m1
2588 pshufb m0, m1
2589 movd [rsp+16*6+hq-4], m0
2590 lea tlq, [rsp+16*5]
2591 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2592 cmp hd, -32
2593 jge .filter_left_end
2594 sub tlq, 16*2
2595 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2596 mova m0, [tlq+16*0]
2597 mova m1, [tlq+16*1]
2598 mova [rsp+16*2], m0
2599 mova [rsp+16*3], m1
2600 .filter_left_end:
2601 mov r2d, maxhm
2602 mova m0, [rsp+16*5]
2603 mova m1, [rsp+16*6]
2604 mova m2, [rsp+16*7]
2605 neg r2
2606 mova [rsp+16*4], m0
2607 mova [rsp+16*5], m1
2608 mova [rsp+16*6], m2
2609 cmp r2d, hd
2610 jle .w4_main
2611 movu m0, [r3+r2-16*2]
2612 movu m1, [r3+r2-16*1]
2613 movu [rsp+r2+16*4], m0
2614 movu [rsp+r2+16*5], m1
2615 cmp r2d, -32
2616 jle .w4_main
2617 movu m0, [r3+r2-16*4]
2618 movu m1, [r3+r2-16*3]
2619 movu [rsp+r2+16*2], m0
2620 movu [rsp+r2+16*3], m1
2621 jmp .w4_main
2622
2623 %if ARCH_X86_64
2624 cglobal ipred_z3_8bpc, 4, 9, 11, 16*10, dst, stride, tl, w, h, angle, dy, _, org_w
2625 %define base r7-$$
2626 lea r7, [$$]
2627 mova m8, [base+pw_62]
2628 mova m9, [base+pw_64]
2629 mova m10, [base+pw_512]
2630 mov org_wd, wd
2631 %else
2632 cglobal ipred_z3_8bpc, 4, 7, 8, -16*10, dst, stride, tl, w, h, angle, dy
2633 %define base r1-$$
2634 %define m8 [base+pw_62]
2635 %define m9 [base+pw_64]
2636 %define m10 [base+pw_512]
2637 %define org_wd r5
2638 %define org_wq r5
2639 mov [dstq+strideq*0], strideq
2640 mov [dstq+strideq*1], wd
2641 LEA r1, $$
2642 %endif
2643 tzcnt hd, hm
2644 movifnidn angled, anglem
2645 dec tlq
2646 movsxd hq, [base+ipred_z3_ssse3_table+hq*4]
2647 sub angled, 180
2648 mov dyd, angled
2649 neg dyd
2650 xor angled, 0x400
2651 or dyq, ~0x7e
2652 lea hq, [base+ipred_z3_ssse3_table+hq]
2653 movzx dyd, word [base+dr_intra_derivative+45*2-1+dyq]
2654 jmp hq
2655 .h4:
2656 lea r4d, [angleq+88]
2657 test r4d, 0x480
2658 jnz .h4_no_upsample ; !enable_intra_edge_filter || angle >= 40
2659 sar r4d, 9
2660 add r4d, wd
2661 cmp r4d, 8
2662 jg .h4_no_upsample ; w > 8 || (w == 8 && is_sm)
2663 movu m3, [tlq-7]
2664 movu m1, [base+z_upsample1-4]
2665 movu m4, [base+z_filter_s+2]
2666 pshufb m0, m3, m1
2667 pxor m1, m1
2668 pshufb m2, m3, m1
2669 pshufb m1, m3, m4
2670 mova [rsp+16], m2 ; top[max_base_y]
2671 movddup m2, [base+pb_36_m4]
2672 add dyd, dyd
2673 pmaddubsw m0, m2
2674 pmaddubsw m1, m2
2675 movd m5, dyd
2676 mov r5d, dyd
2677 pshufb m5, [base+pw_256]
2678 paddw m0, m1
2679 pmulhrsw m0, m10
2680 shl wd, 2
2681 mov tlq, rsp
2682 sub rsp, wq
2683 packuswb m0, m0
2684 punpcklbw m0, m3
2685 paddw m6, m5, m5
2686 punpcklqdq m5, m6
2687 pshufb m0, [base+pb_15to0]
2688 mova [tlq], m0
2689 .h4_upsample_loop:
2690 lea r4d, [r5+dyq]
2691 shr r5d, 6
2692 movq m0, [tlq+r5]
2693 lea r5d, [r4+dyq]
2694 shr r4d, 6
2695 movhps m0, [tlq+r4]
2696 pand m2, m8, m5
2697 psubw m1, m9, m2
2698 psllw m2, 8
2699 por m1, m2
2700 pmaddubsw m0, m1
2701 paddw m5, m6
2702 pmulhrsw m0, m10
2703 packuswb m0, m0
2704 movq [rsp+wq-8], m0
2705 sub wd, 8
2706 jg .h4_upsample_loop
2707 jmp .h4_transpose
2708 .h4_no_upsample:
2709 mov r4d, 7
2710 test angled, 0x400 ; !enable_intra_edge_filter
2711 jnz .h4_main
2712 lea r4d, [wq+3]
2713 movd m0, r4d
2714 movd m2, angled
2715 shr angled, 8 ; is_sm << 1
2716 pxor m1, m1
2717 pshufb m0, m1
2718 pshufb m2, m1
2719 pcmpeqb m1, m0, [base+z_filter_wh4]
2720 pand m1, m2
2721 pcmpgtb m1, [base+z_filter_t_w48+angleq*8]
2722 pmovmskb r5d, m1
2723 mov r4d, 7
2724 test r5d, r5d
2725 jz .h4_main ; filter_strength == 0
2726 movu m2, [tlq-7]
2727 imul r5d, 0x55555555
2728 movu m3, [base+z_filter_s-2]
2729 shr r5d, 30 ; filter_strength
2730 mova m4, [base+z_upsample2]
2731 movddup m5, [base+z_filter_k-8+r5*8+24*0]
2732 movddup m6, [base+z_filter_k-8+r5*8+24*1]
2733 movddup m7, [base+z_filter_k-8+r5*8+24*2]
2734 pshufb m0, m2, m3
2735 shufps m3, m4, q2121
2736 pmaddubsw m1, m0, m5
2737 pmaddubsw m0, m6
2738 pshufb m5, m2, m3
2739 pmaddubsw m3, m5, m6
2740 pmaddubsw m5, m7
2741 pshufb m2, m4
2742 pmaddubsw m2, m7
2743 paddw m0, m1
2744 paddw m1, m3
2745 paddw m0, m5
2746 paddw m1, m2
2747 pmulhrsw m0, m10
2748 pmulhrsw m1, m10
2749 lea r2d, [r4+2]
2750 cmp wd, 4
2751 cmovne r4d, r2d
2752 pshufd m0, m0, q0000
2753 lea tlq, [rsp+15]
2754 packuswb m0, m1
2755 mova [rsp], m0
2756 .h4_main:
2757 movd m5, dyd
2758 movddup m0, [base+z_base_inc] ; base_inc << 6
2759 sub tlq, r4
2760 shl r4d, 6
2761 movd m7, [tlq]
2762 movd m4, r4d
2763 pshufb m5, [base+pw_256]
2764 neg dyq
2765 pshufb m7, [base+pw_m256]
2766 mova m3, [base+z3_shuf_h4]
2767 lea r5, [dyq+r4+63] ; ypos
2768 pshufb m4, [base+pw_256]
2769 psubw m4, m0 ; max_base_y
2770 shl wd, 2
2771 paddw m6, m5, m5
2772 sub rsp, wq
2773 punpcklqdq m5, m6
2774 .h4_loop:
2775 lea r4, [r5+dyq]
2776 sar r5, 6
2777 movq m0, [tlq+r5-4]
2778 lea r5, [r4+dyq]
2779 sar r4, 6
2780 movhps m0, [tlq+r4-4]
2781 pand m2, m8, m5
2782 psubw m1, m9, m2
2783 psllw m2, 8
2784 pshufb m0, m3
2785 por m1, m2
2786 pmaddubsw m0, m1
2787 pcmpgtw m1, m4, m5
2788 paddw m5, m6
2789 pmulhrsw m0, m10
2790 pand m0, m1
2791 pandn m1, m7
2792 por m0, m1
2793 packuswb m0, m0
2794 movq [rsp+wq-8], m0
2795 sub wd, 8
2796 jz .h4_transpose
2797 test r5d, r5d
2798 jg .h4_loop
2799 packuswb m7, m7
2800 .h4_end_loop:
2801 movq [rsp+wq-8], m7
2802 sub wd, 8
2803 jg .h4_end_loop
2804 .h4_transpose:
2805 mova m1, [base+z_transpose4]
2806 %if ARCH_X86_32
2807 mov strideq, [dstq]
2808 mov org_wd, [dstq+strideq]
2809 %endif
2810 lea r2, [strideq*3]
2811 lea dstq, [dstq+org_wq-4]
2812 .h4_transpose_loop:
2813 mova m0, [rsp]
2814 add rsp, 16
2815 pshufb m0, m1
2816 movd [dstq+strideq*0], m0
2817 pshuflw m2, m0, q1032
2818 movd [dstq+strideq*1], m2
2819 punpckhqdq m0, m0
2820 movd [dstq+strideq*2], m0
2821 psrlq m0, 32
2822 movd [dstq+r2 ], m0
2823 sub dstq, 4
2824 sub org_wd, 4
2825 jg .h4_transpose_loop
2826 RET
2827 .h8:
2828 lea r4d, [angleq+88]
2829 and r4d, ~0x7f
2830 or r4d, wd
2831 cmp r4d, 8
2832 ja .h8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || w > 8
2833 mova m4, [tlq-15]
2834 and r4d, 4
2835 movu m3, [tlq- 9]
2836 movd m1, r4d
2837 movu m2, [base+z_filter_s+2]
2838 pxor m0, m0
2839 movu m5, [base+z_filter_s+6]
2840 movddup m7, [base+pb_36_m4]
2841 pshufb m1, m0 ; w & 4
2842 movu m0, [base+z_upsample1-4]
2843 pmaxub m1, m0 ; clip 4x8
2844 add dyd, dyd
2845 pshufb m0, m4, m1
2846 pmaddubsw m0, m7
2847 pshufb m1, m4, m2
2848 pmaddubsw m1, m7
2849 pshufb m2, m3, [base+z_upsample1]
2850 pmaddubsw m2, m7
2851 pshufb m3, m5
2852 pmaddubsw m3, m7
2853 movd m5, dyd
2854 neg dyq
2855 paddw m1, m0
2856 paddw m2, m3
2857 pmulhrsw m1, m10
2858 pmulhrsw m2, m10
2859 shl wd, 3
2860 lea tlq, [rsp+16]
2861 pshufb m5, [base+pw_256]
2862 sub rsp, wq
2863 packuswb m1, m2
2864 lea r5, [dyq+63]
2865 punpcklbw m0, m1, m4
2866 punpckhbw m1, m4
2867 mova [tlq-16*1], m0
2868 mova [tlq-16*0], m1
2869 paddw m6, m5, m5
2870 punpcklqdq m5, m6
2871 .h8_upsample_loop:
2872 lea r4, [r5+dyq]
2873 sar r5, 6
2874 movu m0, [tlq+r5]
2875 lea r5, [r4+dyq]
2876 sar r4, 6
2877 movu m1, [tlq+r4]
2878 pand m3, m8, m5
2879 psubw m2, m9, m3
2880 psllw m2, 8
2881 por m3, m2
2882 pshufd m2, m3, q1010
2883 pmaddubsw m0, m2
2884 punpckhqdq m3, m3
2885 pmaddubsw m1, m3
2886 paddw m5, m6
2887 pmulhrsw m0, m10
2888 pmulhrsw m1, m10
2889 packuswb m1, m0
2890 mova [rsp+wq-16], m1
2891 sub wd, 16
2892 jg .h8_upsample_loop
2893 jmp .h8_transpose
2894 .h8_no_upsample:
2895 lea r4d, [wq+7]
2896 movd m0, r4d
2897 and r4d, 7
2898 or r4d, 8 ; imin(w+7, 15)
2899 test angled, 0x400
2900 jnz .h8_main
2901 movd m2, angled
2902 shr angled, 8 ; is_sm << 1
2903 pxor m1, m1
2904 pshufb m0, m1
2905 pshufb m2, m1
2906 movu m1, [base+z_filter_wh8]
2907 psrldq m3, [base+z_filter_t_w48+angleq*8], 4
2908 pcmpeqb m1, m0
2909 pand m1, m2
2910 pcmpgtb m1, m3
2911 pmovmskb r5d, m1
2912 test r5d, r5d
2913 jz .h8_main ; filter_strength == 0
2914 mova m0, [tlq-15]
2915 imul r5d, 0x55555555
2916 movd m1, [tlq+1]
2917 neg r4
2918 movd m2, [tlq+r4]
2919 shr r5d, 30
2920 pxor m7, m7
2921 lea tlq, [rsp+16*2]
2922 sub r5, 3 ; filter_strength-3
2923 mova [tlq+16*0], m0
2924 pshufb m1, m7
2925 mova [tlq+16*1], m1
2926 pshufb m2, m7
2927 movq [tlq+r4+8], m2
2928 neg r4d
2929 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
2930 sar r5d, 1
2931 add tlq, 31
2932 add r5d, 17
2933 cmp wd, 8
2934 cmova r4d, r5d
2935 .h8_main:
2936 movd m5, dyd
2937 sub tlq, r4
2938 shl r4d, 6
2939 movd m7, [tlq]
2940 movd m4, r4d
2941 pshufb m5, [base+pw_256]
2942 neg dyq
2943 pshufb m7, [base+pw_m256]
2944 mova m3, [base+z3_shuf]
2945 lea r5, [dyq+r4+63]
2946 pshufb m4, [base+pw_256]
2947 psubw m4, [base+z3_base_inc]
2948 shl wd, 3
2949 mova m6, m5
2950 sub rsp, wq
2951 .h8_loop:
2952 mov r4, r5
2953 sar r4, 6
2954 movu m0, [tlq+r4-8]
2955 pand m2, m8, m5
2956 psubw m1, m9, m2
2957 psllw m2, 8
2958 pshufb m0, m3
2959 por m1, m2
2960 pmaddubsw m0, m1
2961 pcmpgtw m1, m4, m5
2962 paddw m5, m6
2963 pmulhrsw m0, m10
2964 pand m0, m1
2965 pandn m1, m7
2966 por m0, m1
2967 packuswb m0, m0
2968 movq [rsp+wq-8], m0
2969 sub wd, 8
2970 jz .h8_transpose
2971 add r5, dyq
2972 jg .h8_loop
2973 packuswb m7, m7
2974 .h8_end_loop:
2975 movq [rsp+wq-8], m7
2976 sub wd, 8
2977 jg .h8_end_loop
2978 .h8_transpose:
2979 %if ARCH_X86_32
2980 mov strideq, [dstq]
2981 mov org_wd, [dstq+strideq]
2982 %endif
2983 or r3d, 8
2984 cmp org_wd, 4
2985 %if ARCH_X86_64
2986 jne .end_transpose_main
2987 %else
2988 jne .end_transpose_loop
2989 %endif
2990 mova m1, [rsp+16*1]
2991 mova m0, [rsp+16*0]
2992 lea r2, [strideq*3]
2993 add rsp, 16*2
2994 punpcklbw m2, m1, m0
2995 punpckhbw m1, m0
2996 punpckhbw m0, m1, m2
2997 punpcklbw m1, m2
2998 .write_4x8_end:
2999 call .write_4x8
3000 RET
3001 .write_4x8:
3002 movd [dstq+r2 ], m0
3003 pshuflw m4, m0, q1032
3004 movd [dstq+strideq*2], m4
3005 punpckhqdq m0, m0
3006 movd [dstq+strideq*1], m0
3007 psrlq m0, 32
3008 movd [dstq+strideq*0], m0
3009 lea dstq, [dstq+strideq*4]
3010 movd [dstq+r2 ], m1
3011 pshuflw m4, m1, q1032
3012 movd [dstq+strideq*2], m4
3013 punpckhqdq m1, m1
3014 movd [dstq+strideq*1], m1
3015 psrlq m1, 32
3016 movd [dstq+strideq*0], m1
3017 ret
3018 .h16:
3019 lea r4d, [wq+15]
3020 movd m0, r4d
3021 and r4d, 15
3022 or r4d, 16 ; imin(w+15, 31)
3023 test angled, 0x400
3024 jnz .h16_main
3025 movd m2, angled
3026 shr angled, 8 ; is_sm << 1
3027 pxor m1, m1
3028 pshufb m0, m1
3029 pshufb m2, m1
3030 movq m3, [base+z_filter_t_w16+angleq*4]
3031 pcmpeqb m1, m0, [base+z_filter_wh16]
3032 pand m1, m2
3033 pcmpgtb m1, m3
3034 pmovmskb r5d, m1
3035 test r5d, r5d
3036 jz .h16_main ; filter_strength == 0
3037 mova m0, [tlq-16*2+1]
3038 imul r5d, 0x24924924
3039 mova m1, [tlq-16*1+1]
3040 neg r4
3041 movd m2, [tlq-16*0+1]
3042 shr r5d, 30
3043 movd m3, [tlq+r4]
3044 adc r5, -4 ; filter_strength-3
3045 pxor m7, m7
3046 lea tlq, [rsp+16*2]
3047 mova [tlq-16*1], m0
3048 pshufb m2, m7
3049 mova [tlq+16*0], m1
3050 pshufb m3, m7
3051 mova [tlq+16*1], m2
3052 movq [tlq+r4+8], m3
3053 neg r4d
3054 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3055 add tlq, 31
3056 cmp wd, 16
3057 jle .h16_main
3058 pshuflw m0, [tlq-47], q0000
3059 sar r5, 1
3060 movq m1, [base+z3_filter_k_tail+r5*4]
3061 lea r4d, [r5+33]
3062 pmaddubsw m0, m1
3063 %if ARCH_X86_64
3064 pmulhrsw m0, m10
3065 %else
3066 pmulhrsw m0, m4
3067 %endif
3068 packuswb m0, m0
3069 movd [tlq-35], m0
3070 .h16_main:
3071 movd m5, dyd
3072 sub tlq, r4
3073 movd m4, r4d
3074 shl r4d, 6
3075 movd m7, [tlq]
3076 pxor m6, m6
3077 pshufb m5, [base+pw_256]
3078 neg dyq
3079 pshufb m7, m6
3080 mova m3, [base+z3_shuf]
3081 lea r5, [dyq+r4+63]
3082 pshufb m4, m6
3083 psubb m4, [base+pb_15to0]
3084 shl wd, 4
3085 mova m6, m5
3086 sub rsp, wq
3087 .h16_loop:
3088 mov r4, r5
3089 pand m2, m8, m5
3090 sar r4, 6
3091 psubw m1, m9, m2
3092 psllw m2, 8
3093 movu m0, [tlq+r4-8*2]
3094 por m2, m1
3095 movu m1, [tlq+r4-8*1]
3096 pshufb m0, m3
3097 pmaddubsw m0, m2
3098 pshufb m1, m3
3099 pmaddubsw m1, m2
3100 psrlw m2, m5, 6
3101 paddw m5, m6
3102 pmulhrsw m0, m10
3103 pmulhrsw m1, m10
3104 packsswb m2, m2
3105 packuswb m0, m1
3106 pcmpgtb m1, m4, m2
3107 pand m0, m1
3108 pandn m1, m7
3109 por m0, m1
3110 mova [rsp+wq-16], m0
3111 sub wd, 16
3112 jz .h16_transpose
3113 add r5, dyq
3114 jg .h16_loop
3115 .h16_end_loop:
3116 mova [rsp+wq-16], m7
3117 sub wd, 16
3118 jg .h16_end_loop
3119 .h16_transpose:
3120 %if ARCH_X86_32
3121 mov strideq, [dstq]
3122 mov org_wd, [dstq+strideq]
3123 %endif
3124 or r3d, 16
3125 cmp org_wd, 4
3126 %if ARCH_X86_64
3127 jne .end_transpose_main
3128 %else
3129 jne .end_transpose_loop
3130 %endif
3131 .h16_transpose_w4:
3132 mova m2, [rsp+16*3]
3133 mova m4, [rsp+16*2]
3134 mova m3, [rsp+16*1]
3135 mova m0, [rsp+16*0]
3136 lea r2, [strideq*3]
3137 add rsp, 16*4
3138 punpckhbw m1, m2, m4
3139 punpcklbw m2, m4
3140 punpckhbw m4, m3, m0
3141 punpcklbw m3, m0
3142 punpckhwd m0, m1, m4
3143 punpcklwd m1, m4
3144 call .write_4x8
3145 lea dstq, [dstq+strideq*4]
3146 punpckhwd m0, m2, m3
3147 punpcklwd m1, m2, m3
3148 jmp .write_4x8_end
3149 .h32:
3150 lea r4d, [wq+31]
3151 and r4d, 31
3152 or r4d, 32 ; imin(w+31, 63)
3153 test angled, 0x400 ; !enable_intra_edge_filter
3154 jnz .h32_main
3155 mova m0, [tlq-16*4+1]
3156 mova m1, [tlq-16*3+1]
3157 mova m2, [tlq-16*2+1]
3158 mova m3, [tlq-16*1+1]
3159 movd m4, [tlq-16*0+1]
3160 neg r4
3161 movd m5, [tlq+r4]
3162 pxor m7, m7
3163 lea tlq, [rsp+16*4]
3164 mova [tlq-16*3], m0
3165 mova [tlq-16*2], m1
3166 xor r5d, r5d ; filter_strength = 3
3167 mova [tlq-16*1], m2
3168 pshufb m4, m7
3169 mova [tlq+16*0], m3
3170 pshufb m5, m7
3171 mova [tlq+16*1], m4
3172 movq [tlq+r4+8], m5
3173 neg r4d
3174 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3175 sub tlq, 16*2
3176 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3177 add tlq, 63
3178 cmp wd, 32
3179 jle .h32_main
3180 pshuflw m0, [tlq-79], q0000
3181 movq m1, [base+z3_filter_k_tail]
3182 add r4d, 2
3183 pmaddubsw m0, m1
3184 %if ARCH_X86_64
3185 pmulhrsw m0, m10
3186 %else
3187 pmulhrsw m0, m4
3188 %endif
3189 packuswb m0, m0
3190 movd [tlq-67], m0
3191 .h32_main:
3192 movd m5, dyd
3193 sub tlq, r4
3194 movd m4, r4d
3195 shl r4d, 6
3196 movd m7, [tlq]
3197 pxor m6, m6
3198 pshufb m5, [base+pw_256]
3199 neg dyq
3200 pshufb m7, m6
3201 mova m3, [base+z3_shuf]
3202 lea r5, [dyq+r4+63]
3203 pshufb m4, m6
3204 psubb m4, [base+pb_15to0]
3205 mova m6, m5
3206 .h32_loop:
3207 mov r4, r5
3208 pand m2, m8, m5
3209 sar r4, 6
3210 psubw m1, m9, m2
3211 psllw m2, 8
3212 movu m0, [tlq+r4-8*4]
3213 por m2, m1
3214 movu m1, [tlq+r4-8*3]
3215 pshufb m0, m3
3216 pmaddubsw m0, m2
3217 pshufb m1, m3
3218 pmaddubsw m1, m2
3219 pmulhrsw m0, m10
3220 pmulhrsw m1, m10
3221 sub rsp, 32
3222 packuswb m0, m1
3223 mova [rsp+16*0], m0
3224 movu m0, [tlq+r4-8*2]
3225 movu m1, [tlq+r4-8*1]
3226 pshufb m0, m3
3227 pshufb m1, m3
3228 pmaddubsw m0, m2
3229 pmaddubsw m1, m2
3230 pmulhrsw m0, m10
3231 pmulhrsw m1, m10
3232 psrlw m2, m5, 6
3233 paddw m5, m6
3234 packsswb m2, m2
3235 packuswb m0, m1
3236 pcmpgtb m1, m4, m2
3237 paddsb m2, [base+pb_16]
3238 pand m0, m1
3239 pandn m1, m7
3240 por m0, m1
3241 pcmpgtb m1, m4, m2
3242 mova [rsp+16*1], m0
3243 pand m0, m1, [rsp+16*0]
3244 pandn m1, m7
3245 por m0, m1
3246 mova [rsp+16*0], m0
3247 dec wd
3248 jz .h32_transpose
3249 add r5, dyq
3250 jg .h32_loop
3251 .h32_end_loop:
3252 sub rsp, 32
3253 mova [rsp+16*1], m7
3254 mova [rsp+16*0], m7
3255 dec wd
3256 jg .h32_end_loop
3257 .h32_transpose:
3258 or r3d, 32
3259 jmp .end_transpose_main
3260 .h64:
3261 lea r4d, [wq+63]
3262 test angled, 0x400 ; !enable_intra_edge_filter
3263 jnz .h64_main
3264 mova m0, [tlq-16*8+1]
3265 mova m1, [tlq-16*7+1]
3266 mova m2, [tlq-16*6+1]
3267 mova m3, [tlq-16*5+1]
3268 mova [rsp+16*1], m0
3269 mova [rsp+16*2], m1
3270 mova [rsp+16*3], m2
3271 mova [rsp+16*4], m3
3272 mova m0, [tlq-16*4+1]
3273 mova m1, [tlq-16*3+1]
3274 mova m2, [tlq-16*2+1]
3275 mova m3, [tlq-16*1+1]
3276 movd m4, [tlq-16*0+1]
3277 neg r4
3278 movd m5, [tlq+r4]
3279 pxor m7, m7
3280 lea tlq, [rsp+16*8]
3281 mova [tlq-16*3], m0
3282 mova [tlq-16*2], m1
3283 xor r5d, r5d ; filter_strength = 3
3284 mova [tlq-16*1], m2
3285 pshufb m4, m7
3286 mova [tlq+16*0], m3
3287 pshufb m5, m7
3288 mova [tlq+16*1], m4
3289 movq [tlq+r4+8], m5
3290 neg r4d
3291 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3292 sub tlq, 16*2
3293 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3294 sub tlq, 16*2
3295 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3296 sub tlq, 16*2
3297 cmp wd, 64
3298 jl .h64_filter96 ; skip one call if the last 32 bytes aren't used
3299 call mangle(private_prefix %+ _ipred_z1_8bpc_ssse3).filter_edge
3300 .h64_filter96:
3301 add tlq, 127
3302 .h64_main:
3303 movd m5, dyd
3304 sub tlq, r4
3305 movd m4, r4d
3306 shl r4d, 6
3307 movd m7, [tlq]
3308 pxor m6, m6
3309 pshufb m5, [base+pw_256]
3310 neg dyq
3311 pshufb m7, m6
3312 mova m3, [base+z3_shuf]
3313 lea r5, [dyq+r4+63]
3314 pshufb m4, m6
3315 psubb m4, [base+pb_15to0]
3316 mova m6, m5
3317 .h64_loop:
3318 mov r4, r5
3319 pand m2, m8, m5
3320 sar r4, 6
3321 psubw m1, m9, m2
3322 psllw m2, 8
3323 movu m0, [tlq+r4-8*8]
3324 por m2, m1
3325 movu m1, [tlq+r4-8*7]
3326 pshufb m0, m3
3327 pmaddubsw m0, m2
3328 pshufb m1, m3
3329 pmaddubsw m1, m2
3330 pmulhrsw m0, m10
3331 pmulhrsw m1, m10
3332 sub rsp, 64
3333 packuswb m0, m1
3334 mova [rsp+16*0], m0
3335 movu m0, [tlq+r4-8*6]
3336 movu m1, [tlq+r4-8*5]
3337 pshufb m0, m3
3338 pshufb m1, m3
3339 pmaddubsw m0, m2
3340 pmaddubsw m1, m2
3341 pmulhrsw m0, m10
3342 pmulhrsw m1, m10
3343 packuswb m0, m1
3344 mova [rsp+16*1], m0
3345 movu m0, [tlq+r4-8*4]
3346 movu m1, [tlq+r4-8*3]
3347 pshufb m0, m3
3348 pshufb m1, m3
3349 pmaddubsw m0, m2
3350 pmaddubsw m1, m2
3351 pmulhrsw m0, m10
3352 pmulhrsw m1, m10
3353 packuswb m0, m1
3354 mova [rsp+16*2], m0
3355 movu m0, [tlq+r4-8*2]
3356 movu m1, [tlq+r4-8*1]
3357 pshufb m0, m3
3358 pshufb m1, m3
3359 pmaddubsw m0, m2
3360 pmaddubsw m1, m2
3361 pmulhrsw m0, m10
3362 pmulhrsw m1, m10
3363 psrlw m2, m5, 6
3364 paddw m5, m6
3365 packsswb m2, m2
3366 packuswb m0, m1
3367 pcmpgtb m1, m4, m2
3368 paddsb m2, [base+pb_16]
3369 pand m0, m1
3370 pandn m1, m7
3371 por m0, m1
3372 pcmpgtb m1, m4, m2
3373 paddsb m2, [base+pb_16]
3374 mova [rsp+16*3], m0
3375 pand m0, m1, [rsp+16*2]
3376 pandn m1, m7
3377 por m0, m1
3378 pcmpgtb m1, m4, m2
3379 paddsb m2, [base+pb_16]
3380 mova [rsp+16*2], m0
3381 pand m0, m1, [rsp+16*1]
3382 pandn m1, m7
3383 por m0, m1
3384 pcmpgtb m1, m4, m2
3385 mova [rsp+16*1], m0
3386 pand m0, m1, [rsp+16*0]
3387 pandn m1, m7
3388 por m0, m1
3389 mova [rsp+16*0], m0
3390 dec wd
3391 jz .h64_transpose
3392 add r5, dyq
3393 jg .h64_loop
3394 .h64_end_loop:
3395 sub rsp, 64
3396 mova [rsp+16*3], m7
3397 mova [rsp+16*2], m7
3398 mova [rsp+16*1], m7
3399 mova [rsp+16*0], m7
3400 dec wd
3401 jg .h64_end_loop
3402 .h64_transpose:
3403 or r3d, 64
3404 .end_transpose_main:
3405 %if ARCH_X86_64
3406 lea r5, [r3*3]
3407 lea r7, [strideq*3]
3408 %else
3409 mov strideq, [dstq]
3410 mov org_wd, [dstq+strideq]
3411 %endif
3412 .end_transpose_loop:
3413 lea r4, [rsp+r3-8]
3414 lea r6, [dstq+org_wq-8]
3415 .end_transpose_loop_y:
3416 movq m0, [r4+r3*1]
3417 movq m4, [r4+r3*0]
3418 %if ARCH_X86_64
3419 movq m1, [r4+r5 ]
3420 movq m5, [r4+r3*2]
3421 lea r2, [r4+r3*4]
3422 %else
3423 lea r2, [r4+r3*2]
3424 movq m1, [r2+r3*1]
3425 movq m5, [r2+r3*0]
3426 lea r2, [r2+r3*2]
3427 %endif
3428 movq m2, [r2+r3*1]
3429 movq m6, [r2+r3*0]
3430 %if ARCH_X86_64
3431 movq m3, [r2+r5 ]
3432 movq m7, [r2+r3*2]
3433 %else
3434 lea r2, [r2+r3*2]
3435 movq m3, [r2+r3*1]
3436 movq m7, [r2+r3*0]
3437 %endif
3438 sub r4, 8
3439 punpcklbw m0, m4
3440 punpcklbw m1, m5
3441 punpcklbw m2, m6
3442 punpcklbw m3, m7
3443 punpckhwd m4, m1, m0
3444 punpcklwd m1, m0
3445 punpckhwd m0, m3, m2
3446 punpcklwd m3, m2
3447 punpckhdq m2, m3, m1
3448 punpckldq m3, m1
3449 punpckldq m1, m0, m4
3450 punpckhdq m0, m4
3451 movhps [r6+strideq*0], m0
3452 movq [r6+strideq*1], m0
3453 %if ARCH_X86_64
3454 movhps [r6+strideq*2], m1
3455 movq [r6+r7 ], m1
3456 lea r6, [r6+strideq*4]
3457 %else
3458 lea r6, [r6+strideq*2]
3459 movhps [r6+strideq*0], m1
3460 movq [r6+strideq*1], m1
3461 lea r6, [r6+strideq*2]
3462 %endif
3463 movhps [r6+strideq*0], m2
3464 movq [r6+strideq*1], m2
3465 %if ARCH_X86_64
3466 movhps [r6+strideq*2], m3
3467 movq [r6+r7 ], m3
3468 lea r6, [r6+strideq*4]
3469 %else
3470 lea r6, [r6+strideq*2]
3471 movhps [r6+strideq*0], m3
3472 movq [r6+strideq*1], m3
3473 lea r6, [r6+strideq*2]
3474 %endif
3475 cmp r4, rsp
3476 jae .end_transpose_loop_y
3477 lea rsp, [rsp+r3*8]
3478 sub org_wd, 8
3479 jg .end_transpose_loop
3480 RET
3481
3482 ;---------------------------------------------------------------------------------------
3483 ;int dav1d_pal_pred_ssse3(pixel *dst, const ptrdiff_t stride, const uint16_t *const pal,
3484 ; const uint8_t *idx, const int w, const int h);
3485 ;---------------------------------------------------------------------------------------
3486 cglobal pal_pred_8bpc, 4, 6, 5, dst, stride, pal, idx, w, h
3487 mova m4, [palq]
3488 LEA r2, pal_pred_ssse3_table
3489 tzcnt wd, wm
3490 movifnidn hd, hm
3491 movsxd wq, [r2+wq*4]
3492 packuswb m4, m4
3493 add wq, r2
3494 lea r2, [strideq*3]
3495 jmp wq
3496 .w4:
3497 pshufb m0, m4, [idxq]
3498 add idxq, 16
3499 movd [dstq ], m0
3500 pshuflw m1, m0, q1032
3501 movd [dstq+strideq ], m1
3502 punpckhqdq m0, m0
3503 movd [dstq+strideq*2], m0
3504 psrlq m0, 32
3505 movd [dstq+r2 ], m0
3506 lea dstq, [dstq+strideq*4]
3507 sub hd, 4
3508 jg .w4
3509 RET
3510 ALIGN function_align
3511 .w8:
3512 pshufb m0, m4, [idxq]
3513 pshufb m1, m4, [idxq+16]
3514 add idxq, 32
3515 movq [dstq ], m0
3516 movhps [dstq+strideq ], m0
3517 movq [dstq+strideq*2], m1
3518 movhps [dstq+r2 ], m1
3519 lea dstq, [dstq+strideq*4]
3520 sub hd, 4
3521 jg .w8
3522 RET
3523 ALIGN function_align
3524 .w16:
3525 pshufb m0, m4, [idxq]
3526 pshufb m1, m4, [idxq+16]
3527 pshufb m2, m4, [idxq+32]
3528 pshufb m3, m4, [idxq+48]
3529 add idxq, 64
3530 mova [dstq ], m0
3531 mova [dstq+strideq ], m1
3532 mova [dstq+strideq*2], m2
3533 mova [dstq+r2 ], m3
3534 lea dstq, [dstq+strideq*4]
3535 sub hd, 4
3536 jg .w16
3537 RET
3538 ALIGN function_align
3539 .w32:
3540 pshufb m0, m4, [idxq]
3541 pshufb m1, m4, [idxq+16]
3542 pshufb m2, m4, [idxq+32]
3543 pshufb m3, m4, [idxq+48]
3544 add idxq, 64
3545 mova [dstq ], m0
3546 mova [dstq+16 ], m1
3547 mova [dstq+strideq ], m2
3548 mova [dstq+strideq+16], m3
3549 lea dstq, [dstq+strideq*2]
3550 sub hd, 2
3551 jg .w32
3552 RET
3553 ALIGN function_align
3554 .w64:
3555 pshufb m0, m4, [idxq]
3556 pshufb m1, m4, [idxq+16]
3557 pshufb m2, m4, [idxq+32]
3558 pshufb m3, m4, [idxq+48]
3559 add idxq, 64
3560 mova [dstq ], m0
3561 mova [dstq+16], m1
3562 mova [dstq+32], m2
3563 mova [dstq+48], m3
3564 add dstq, strideq
3565 sub hd, 1
3566 jg .w64
3567 RET
3568
3569 ;---------------------------------------------------------------------------------------
3570 ;void dav1d_ipred_cfl_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
3571 ; const int width, const int height, const int16_t *ac, const int alpha);
3572 ;---------------------------------------------------------------------------------------
3573 %macro IPRED_CFL 1 ; ac in, unpacked pixels out
3574 psignw m3, m%1, m1
3575 pabsw m%1, m%1
3576 pmulhrsw m%1, m2
3577 psignw m%1, m3
3578 paddw m%1, m0
3579 %endmacro
3580
3581 %if UNIX64
3582 DECLARE_REG_TMP 7
3583 %else
3584 DECLARE_REG_TMP 5
3585 %endif
3586
3587 cglobal ipred_cfl_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
3588 movifnidn wd, wm
3589 movifnidn hd, hm
3590 tzcnt r6d, hd
3591 lea t0d, [wq+hq]
3592 movd m4, t0d
3593 tzcnt t0d, t0d
3594 movd m5, t0d
3595 LEA t0, ipred_cfl_ssse3_table
3596 tzcnt wd, wd
3597 movsxd r6, [t0+r6*4]
3598 movsxd wq, [t0+wq*4+16]
3599 pcmpeqd m3, m3
3600 psrlw m4, 1
3601 add r6, t0
3602 add wq, t0
3603 movifnidn acq, acmp
3604 jmp r6
3605 .h4:
3606 movd m0, [tlq-4]
3607 pmaddubsw m0, m3
3608 jmp wq
3609 .w4:
3610 movd m1, [tlq+1]
3611 pmaddubsw m1, m3
3612 psubw m0, m4
3613 paddw m0, m1
3614 pmaddwd m0, m3
3615 cmp hd, 4
3616 jg .w4_mul
3617 psrlw m0, 3 ; dc >>= ctz(width + height);
3618 jmp .w4_end
3619 .w4_mul:
3620 punpckhqdq m1, m0, m0
3621 paddw m0, m1
3622 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
3623 paddw m0, m1
3624 psrlw m0, 2
3625 mov r6d, 0x5556
3626 mov r2d, 0x3334
3627 test hd, 8
3628 cmovz r6d, r2d
3629 movd m5, r6d
3630 pmulhuw m0, m5
3631 .w4_end:
3632 pshuflw m0, m0, q0000
3633 punpcklqdq m0, m0
3634 .s4:
3635 movd m1, alpham
3636 pshuflw m1, m1, q0000
3637 punpcklqdq m1, m1
3638 lea r6, [strideq*3]
3639 pabsw m2, m1
3640 psllw m2, 9
3641 .s4_loop:
3642 mova m4, [acq]
3643 mova m5, [acq+16]
3644 IPRED_CFL 4
3645 IPRED_CFL 5
3646 packuswb m4, m5
3647 movd [dstq+strideq*0], m4
3648 pshuflw m4, m4, q1032
3649 movd [dstq+strideq*1], m4
3650 punpckhqdq m4, m4
3651 movd [dstq+strideq*2], m4
3652 psrlq m4, 32
3653 movd [dstq+r6 ], m4
3654 lea dstq, [dstq+strideq*4]
3655 add acq, 32
3656 sub hd, 4
3657 jg .s4_loop
3658 RET
3659 ALIGN function_align
3660 .h8:
3661 movq m0, [tlq-8]
3662 pmaddubsw m0, m3
3663 jmp wq
3664 .w8:
3665 movq m1, [tlq+1]
3666 pmaddubsw m1, m3
3667 psubw m4, m0
3668 punpckhqdq m0, m0
3669 psubw m0, m4
3670 paddw m0, m1
3671 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
3672 paddw m0, m1
3673 pmaddwd m0, m3
3674 psrlw m0, m5
3675 cmp hd, 8
3676 je .w8_end
3677 mov r6d, 0x5556
3678 mov r2d, 0x3334
3679 cmp hd, 32
3680 cmovz r6d, r2d
3681 movd m1, r6d
3682 pmulhuw m0, m1
3683 .w8_end:
3684 pshuflw m0, m0, q0000
3685 punpcklqdq m0, m0
3686 .s8:
3687 movd m1, alpham
3688 pshuflw m1, m1, q0000
3689 punpcklqdq m1, m1
3690 lea r6, [strideq*3]
3691 pabsw m2, m1
3692 psllw m2, 9
3693 .s8_loop:
3694 mova m4, [acq]
3695 mova m5, [acq+16]
3696 IPRED_CFL 4
3697 IPRED_CFL 5
3698 packuswb m4, m5
3699 movq [dstq ], m4
3700 movhps [dstq+strideq ], m4
3701 mova m4, [acq+32]
3702 mova m5, [acq+48]
3703 IPRED_CFL 4
3704 IPRED_CFL 5
3705 packuswb m4, m5
3706 movq [dstq+strideq*2], m4
3707 movhps [dstq+r6 ], m4
3708 lea dstq, [dstq+strideq*4]
3709 add acq, 64
3710 sub hd, 4
3711 jg .s8_loop
3712 RET
3713 ALIGN function_align
3714 .h16:
3715 mova m0, [tlq-16]
3716 pmaddubsw m0, m3
3717 jmp wq
3718 .w16:
3719 movu m1, [tlq+1]
3720 pmaddubsw m1, m3
3721 paddw m0, m1
3722 psubw m4, m0
3723 punpckhqdq m0, m0
3724 psubw m0, m4
3725 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
3726 paddw m0, m1
3727 pmaddwd m0, m3
3728 psrlw m0, m5
3729 cmp hd, 16
3730 je .w16_end
3731 mov r6d, 0x5556
3732 mov r2d, 0x3334
3733 test hd, 8|32
3734 cmovz r6d, r2d
3735 movd m1, r6d
3736 pmulhuw m0, m1
3737 .w16_end:
3738 pshuflw m0, m0, q0000
3739 punpcklqdq m0, m0
3740 .s16:
3741 movd m1, alpham
3742 pshuflw m1, m1, q0000
3743 punpcklqdq m1, m1
3744 pabsw m2, m1
3745 psllw m2, 9
3746 .s16_loop:
3747 mova m4, [acq]
3748 mova m5, [acq+16]
3749 IPRED_CFL 4
3750 IPRED_CFL 5
3751 packuswb m4, m5
3752 mova [dstq], m4
3753 mova m4, [acq+32]
3754 mova m5, [acq+48]
3755 IPRED_CFL 4
3756 IPRED_CFL 5
3757 packuswb m4, m5
3758 mova [dstq+strideq], m4
3759 lea dstq, [dstq+strideq*2]
3760 add acq, 64
3761 sub hd, 2
3762 jg .s16_loop
3763 RET
3764 ALIGN function_align
3765 .h32:
3766 mova m0, [tlq-32]
3767 pmaddubsw m0, m3
3768 mova m2, [tlq-16]
3769 pmaddubsw m2, m3
3770 paddw m0, m2
3771 jmp wq
3772 .w32:
3773 movu m1, [tlq+1]
3774 pmaddubsw m1, m3
3775 movu m2, [tlq+17]
3776 pmaddubsw m2, m3
3777 paddw m1, m2
3778 paddw m0, m1
3779 psubw m4, m0
3780 punpckhqdq m0, m0
3781 psubw m0, m4
3782 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
3783 paddw m0, m1
3784 pmaddwd m0, m3
3785 psrlw m0, m5
3786 cmp hd, 32
3787 je .w32_end
3788 lea r2d, [hq*2]
3789 mov r6d, 0x5556
3790 mov r2d, 0x3334
3791 test hd, 64|16
3792 cmovz r6d, r2d
3793 movd m1, r6d
3794 pmulhuw m0, m1
3795 .w32_end:
3796 pshuflw m0, m0, q0000
3797 punpcklqdq m0, m0
3798 .s32:
3799 movd m1, alpham
3800 pshuflw m1, m1, q0000
3801 punpcklqdq m1, m1
3802 pabsw m2, m1
3803 psllw m2, 9
3804 .s32_loop:
3805 mova m4, [acq]
3806 mova m5, [acq+16]
3807 IPRED_CFL 4
3808 IPRED_CFL 5
3809 packuswb m4, m5
3810 mova [dstq], m4
3811 mova m4, [acq+32]
3812 mova m5, [acq+48]
3813 IPRED_CFL 4
3814 IPRED_CFL 5
3815 packuswb m4, m5
3816 mova [dstq+16], m4
3817 add dstq, strideq
3818 add acq, 64
3819 dec hd
3820 jg .s32_loop
3821 RET
3822
3823 ;---------------------------------------------------------------------------------------
3824 ;void dav1d_ipred_cfl_left_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
3825 ; const int width, const int height, const int16_t *ac, const int alpha);
3826 ;---------------------------------------------------------------------------------------
3827 cglobal ipred_cfl_left_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
3828 mov hd, hm ; zero upper half
3829 tzcnt r6d, hd
3830 sub tlq, hq
3831 tzcnt wd, wm
3832 movu m0, [tlq]
3833 mov t0d, 0x8000
3834 movd m3, t0d
3835 movd m2, r6d
3836 psrld m3, m2
3837 LEA t0, ipred_cfl_left_ssse3_table
3838 movsxd r6, [t0+r6*4]
3839 pcmpeqd m2, m2
3840 pmaddubsw m0, m2
3841 add r6, t0
3842 add t0, ipred_cfl_splat_ssse3_table-ipred_cfl_left_ssse3_table
3843 movsxd wq, [t0+wq*4]
3844 add wq, t0
3845 movifnidn acq, acmp
3846 jmp r6
3847 .h32:
3848 movu m1, [tlq+16] ; unaligned when jumping here from dc_top
3849 pmaddubsw m1, m2
3850 paddw m0, m1
3851 .h16:
3852 pshufd m1, m0, q3232 ; psrlq m1, m0, 16
3853 paddw m0, m1
3854 .h8:
3855 pshuflw m1, m0, q1032 ; psrlq m1, m0, 32
3856 paddw m0, m1
3857 .h4:
3858 pmaddwd m0, m2
3859 pmulhrsw m0, m3
3860 pshuflw m0, m0, q0000
3861 punpcklqdq m0, m0
3862 jmp wq
3863
3864 ;---------------------------------------------------------------------------------------
3865 ;void dav1d_ipred_cfl_top_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
3866 ; const int width, const int height, const int16_t *ac, const int alpha);
3867 ;---------------------------------------------------------------------------------------
3868 cglobal ipred_cfl_top_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
3869 LEA t0, ipred_cfl_left_ssse3_table
3870 tzcnt wd, wm
3871 inc tlq
3872 movu m0, [tlq]
3873 movifnidn hd, hm
3874 mov r6d, 0x8000
3875 movd m3, r6d
3876 movd m2, wd
3877 psrld m3, m2
3878 movsxd r6, [t0+wq*4]
3879 pcmpeqd m2, m2
3880 pmaddubsw m0, m2
3881 add r6, t0
3882 add t0, ipred_cfl_splat_ssse3_table-ipred_cfl_left_ssse3_table
3883 movsxd wq, [t0+wq*4]
3884 add wq, t0
3885 movifnidn acq, acmp
3886 jmp r6
3887
3888 ;---------------------------------------------------------------------------------------
3889 ;void dav1d_ipred_cfl_128_ssse3(pixel *dst, const ptrdiff_t stride, const pixel *const topleft,
3890 ; const int width, const int height, const int16_t *ac, const int alpha);
3891 ;---------------------------------------------------------------------------------------
3892 cglobal ipred_cfl_128_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
3893 tzcnt wd, wm
3894 movifnidn hd, hm
3895 LEA r6, ipred_cfl_splat_ssse3_table
3896 movsxd wq, [r6+wq*4]
3897 movddup m0, [r6-ipred_cfl_splat_ssse3_table+pw_128]
3898 add wq, r6
3899 movifnidn acq, acmp
3900 jmp wq
3901
3902 %macro RELOAD_ACQ_32 1
3903 mov acq, ac_bakq ; restore acq
3904 %endmacro
3905
3906 %if ARCH_X86_64
3907 cglobal ipred_cfl_ac_420_8bpc, 4, 8, 7, ac, y, stride, wpad, hpad, w, h, ac_bak
3908 DECLARE_REG_TMP 7
3909 movddup m2, [pb_2]
3910 %else
3911 cglobal ipred_cfl_ac_420_8bpc, 4, 7, 7, ac, y, stride, wpad, hpad, w, h
3912 DECLARE_REG_TMP 4
3913 %define ac_bakq acmp
3914 mov t0d, 0x02020202
3915 movd m2, t0d
3916 pshufd m2, m2, q0000
3917 %endif
3918 movifnidn wd, wm
3919 mov t0d, hm
3920 mov hd, t0d
3921 imul t0d, wd
3922 movd m5, t0d
3923 movifnidn hpadd, hpadm
3924 %if ARCH_X86_64
3925 mov ac_bakq, acq
3926 %endif
3927 shl hpadd, 2
3928 sub hd, hpadd
3929 pxor m4, m4
3930 cmp wd, 8
3931 jg .w16
3932 je .w8
3933 ; fall-through
3934 %if ARCH_X86_64
3935 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, ac_bak
3936 %else
3937 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h
3938 %endif
3939 .w4:
3940 lea stride3q, [strideq*3]
3941 .w4_loop:
3942 movq m0, [yq]
3943 movq m1, [yq+strideq]
3944 movhps m0, [yq+strideq*2]
3945 movhps m1, [yq+stride3q]
3946 pmaddubsw m0, m2
3947 pmaddubsw m1, m2
3948 paddw m0, m1
3949 mova [acq], m0
3950 paddw m4, m0
3951 lea yq, [yq+strideq*4]
3952 add acq, 16
3953 sub hd, 2
3954 jg .w4_loop
3955 test hpadd, hpadd
3956 jz .calc_avg_4_8
3957 punpckhqdq m0, m0
3958 .w4_hpad_loop:
3959 mova [acq], m0
3960 paddw m4, m0
3961 add acq, 16
3962 sub hpadd, 2
3963 jg .w4_hpad_loop
3964 jmp .calc_avg_4_8
3965 .w8:
3966 lea stride3q, [strideq*3]
3967 test wpadd, wpadd
3968 jnz .w8_wpad
3969 .w8_loop:
3970 mova m0, [yq]
3971 mova m1, [yq+strideq]
3972 pmaddubsw m0, m2
3973 pmaddubsw m1, m2
3974 paddw m0, m1
3975 mova [acq], m0
3976 paddw m4, m0
3977 mova m0, [yq+strideq*2]
3978 mova m1, [yq+stride3q]
3979 pmaddubsw m0, m2
3980 pmaddubsw m1, m2
3981 paddw m0, m1
3982 mova [acq+16], m0
3983 paddw m4, m0
3984 lea yq, [yq+strideq*4]
3985 add acq, 32
3986 sub hd, 2
3987 jg .w8_loop
3988 test hpadd, hpadd
3989 jz .calc_avg_4_8
3990 jmp .w8_hpad
3991 .w8_wpad: ; wpadd=1
3992 movddup m0, [yq]
3993 movddup m1, [yq+strideq]
3994 pmaddubsw m0, m2
3995 pmaddubsw m1, m2
3996 paddw m0, m1
3997 pshufhw m0, m0, q3333
3998 mova [acq], m0
3999 paddw m4, m0
4000 lea yq, [yq+strideq*2]
4001 add acq, 16
4002 sub hd, 1
4003 jg .w8_wpad
4004 test hpadd, hpadd
4005 jz .calc_avg_4_8
4006 .w8_hpad:
4007 mova [acq], m0
4008 paddw m4, m0
4009 add acq, 16
4010 sub hpadd, 1
4011 jg .w8_hpad
4012 jmp .calc_avg_4_8
4013 .w16:
4014 test wpadd, wpadd
4015 jnz .w16_wpad
4016 .w16_loop:
4017 mova m0, [yq]
4018 mova m1, [yq+strideq]
4019 pmaddubsw m0, m2
4020 pmaddubsw m1, m2
4021 paddw m0, m1
4022 mova [acq], m0
4023 paddw m4, m0
4024 mova m6, [yq+16]
4025 mova m1, [yq+strideq+16]
4026 pmaddubsw m6, m2
4027 pmaddubsw m1, m2
4028 paddw m6, m1
4029 mova [acq+16], m6
4030 paddw m4, m6
4031 lea yq, [yq+strideq*2]
4032 add acq, 32
4033 dec hd
4034 jg .w16_loop
4035 test hpadd, hpadd
4036 jz .calc_avg16
4037 jmp .w16_hpad_loop
4038 .w16_wpad:
4039 cmp wpadd, 2
4040 jl .w16_pad1
4041 je .w16_pad2
4042 .w16_pad3:
4043 movddup m0, [yq]
4044 movddup m1, [yq+strideq]
4045 pmaddubsw m0, m2
4046 pmaddubsw m1, m2
4047 paddw m0, m1
4048 pshufhw m0, m0, q3333
4049 mova [acq], m0
4050 paddw m4, m0
4051 mova m6, m0
4052 punpckhqdq m6, m0, m0
4053 mova [acq+16], m6
4054 paddw m4, m6
4055 lea yq, [yq+strideq*2]
4056 add acq, 32
4057 dec hd
4058 jg .w16_pad3
4059 jmp .w16_wpad_done
4060 .w16_pad2:
4061 mova m0, [yq]
4062 mova m1, [yq+strideq]
4063 pmaddubsw m0, m2
4064 pmaddubsw m1, m2
4065 paddw m0, m1
4066 mova [acq], m0
4067 paddw m4, m0
4068 pshufhw m6, m0, q3333
4069 punpckhqdq m6, m6
4070 mova [acq+16], m6
4071 paddw m4, m6
4072 lea yq, [yq+strideq*2]
4073 add acq, 32
4074 dec hd
4075 jg .w16_pad2
4076 jmp .w16_wpad_done
4077 .w16_pad1:
4078 mova m0, [yq]
4079 mova m1, [yq+strideq]
4080 pmaddubsw m0, m2
4081 pmaddubsw m1, m2
4082 paddw m0, m1
4083 mova [acq], m0
4084 paddw m4, m0
4085 movddup m6, [yq+16]
4086 movddup m1, [yq+strideq+16]
4087 pmaddubsw m6, m2
4088 pmaddubsw m1, m2
4089 paddw m6, m1
4090 pshufhw m6, m6, q3333
4091 mova [acq+16], m6
4092 paddw m4, m6
4093 lea yq, [yq+strideq*2]
4094 add acq, 32
4095 dec hd
4096 jg .w16_pad1
4097 .w16_wpad_done:
4098 test hpadd, hpadd
4099 jz .calc_avg16
4100 .w16_hpad_loop:
4101 mova [acq], m0
4102 paddw m4, m0
4103 mova [acq+16], m6
4104 paddw m4, m6
4105 add acq, 32
4106 dec hpadd
4107 jg .w16_hpad_loop
4108 jmp .calc_avg16
4109
4110 %if ARCH_X86_64
4111 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h, ac_bak
4112 %else
4113 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h
4114 %endif
4115 .calc_avg_4_8:
4116 psrlw m2, 9
4117 pmaddwd m4, m2
4118 jmp .calc_avg
4119 .calc_avg16:
4120 psrld m0, m4, 16
4121 pslld m4, 16
4122 psrld m4, 16
4123 paddd m4, m0
4124 .calc_avg:
4125 movd szd, m5
4126 psrad m5, 1
4127 tzcnt r1d, szd
4128 paddd m4, m5
4129 movd m1, r1d
4130 pshufd m0, m4, q2301
4131 paddd m0, m4
4132 pshufd m4, m0, q1032
4133 paddd m0, m4
4134 psrad m0, m1 ; sum >>= log2sz;
4135 packssdw m0, m0
4136 RELOAD_ACQ_32 acq
4137 .sub_loop:
4138 mova m1, [acq]
4139 psubw m1, m0 ; ac[x] -= sum;
4140 mova [acq], m1
4141 add acq, 16
4142 sub szd, 8
4143 jg .sub_loop
4144 RET
4145
4146 %if ARCH_X86_64
4147 cglobal ipred_cfl_ac_422_8bpc, 4, 8, 7, ac, y, stride, wpad, hpad, w, h, ac_bak
4148 movddup m2, [pb_4]
4149 %else
4150 cglobal ipred_cfl_ac_422_8bpc, 4, 7, 7, ac, y, stride, wpad, hpad, w, h
4151 mov t0d, 0x04040404
4152 movd m2, t0d
4153 pshufd m2, m2, q0000
4154 %endif
4155 movifnidn wd, wm
4156 mov t0d, hm
4157 mov hd, t0d
4158 imul t0d, wd
4159 movd m6, t0d
4160 movifnidn hpadd, hpadm
4161 %if ARCH_X86_64
4162 mov ac_bakq, acq
4163 %endif
4164 shl hpadd, 2
4165 sub hd, hpadd
4166 pxor m4, m4
4167 pxor m5, m5
4168 cmp wd, 8
4169 jg .w16
4170 je .w8
4171 ; fall-through
4172
4173 %if ARCH_X86_64
4174 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, ac_bak
4175 %else
4176 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h
4177 %endif
4178 .w4:
4179 lea stride3q, [strideq*3]
4180 .w4_loop:
4181 movq m1, [yq]
4182 movhps m1, [yq+strideq]
4183 movq m0, [yq+strideq*2]
4184 movhps m0, [yq+stride3q]
4185 pmaddubsw m0, m2
4186 pmaddubsw m1, m2
4187 mova [acq], m1
4188 mova [acq+16], m0
4189 paddw m4, m0
4190 paddw m5, m1
4191 lea yq, [yq+strideq*4]
4192 add acq, 32
4193 sub hd, 4
4194 jg .w4_loop
4195 test hpadd, hpadd
4196 jz .calc_avg_4
4197 punpckhqdq m0, m0
4198 .w4_hpad_loop:
4199 mova [acq], m0
4200 paddw m4, m0
4201 add acq, 16
4202 sub hpadd, 2
4203 jg .w4_hpad_loop
4204 jmp .calc_avg_4
4205 .w8:
4206 lea stride3q, [strideq*3]
4207 test wpadd, wpadd
4208 jnz .w8_wpad
4209 .w8_loop:
4210 mova m1, [yq]
4211 mova m0, [yq+strideq]
4212 pmaddubsw m0, m2
4213 pmaddubsw m1, m2
4214 mova [acq], m1
4215 mova [acq+16], m0
4216 paddw m4, m0
4217 paddw m5, m1
4218 mova m1, [yq+strideq*2]
4219 mova m0, [yq+stride3q]
4220 pmaddubsw m0, m2
4221 pmaddubsw m1, m2
4222 mova [acq+32], m1
4223 mova [acq+48], m0
4224 paddw m4, m0
4225 paddw m5, m1
4226 lea yq, [yq+strideq*4]
4227 add acq, 64
4228 sub hd, 4
4229 jg .w8_loop
4230 test hpadd, hpadd
4231 jz .calc_avg_8_16
4232 jmp .w8_hpad
4233 .w8_wpad:
4234 movddup m1, [yq]
4235 pmaddubsw m1, m2
4236 pshufhw m1, m1, q3333
4237 mova [acq], m1
4238 paddw m5, m1
4239 movddup m0, [yq+strideq]
4240 pmaddubsw m0, m2
4241 pshufhw m0, m0, q3333
4242 mova [acq+16], m0
4243 paddw m4, m0
4244 lea yq, [yq+strideq*2]
4245 add acq, 32
4246 sub hd, 2
4247 jg .w8_wpad
4248 test hpadd, hpadd
4249 jz .calc_avg_8_16
4250 .w8_hpad:
4251 mova [acq], m0
4252 paddw m4, m0
4253 mova [acq+16], m0
4254 paddw m4, m0
4255 add acq, 32
4256 sub hpadd, 2
4257 jg .w8_hpad
4258 jmp .calc_avg_8_16
4259 .w16:
4260 test wpadd, wpadd
4261 jnz .w16_wpad
4262 .w16_loop:
4263 mova m1, [yq]
4264 mova m0, [yq+16]
4265 pmaddubsw m0, m2
4266 pmaddubsw m1, m2
4267 mova [acq], m1
4268 mova [acq+16], m0
4269 paddw m5, m0
4270 paddw m5, m1
4271 mova m1, [yq+strideq]
4272 mova m0, [yq+strideq+16]
4273 pmaddubsw m0, m2
4274 pmaddubsw m1, m2
4275 mova [acq+32], m1
4276 mova [acq+48], m0
4277 paddw m4, m0
4278 paddw m4, m1
4279 lea yq, [yq+strideq*2]
4280 add acq, 64
4281 sub hd, 2
4282 jg .w16_loop
4283 test hpadd, hpadd
4284 jz .calc_avg_8_16
4285 jmp .w16_hpad_loop
4286 .w16_wpad:
4287 cmp wpadd, 2
4288 jl .w16_pad1
4289 je .w16_pad2
4290 .w16_pad3:
4291 movddup m1, [yq]
4292 pmaddubsw m1, m2
4293 pshufhw m1, m1, q3333
4294 mova [acq], m1
4295 paddw m5, m1
4296 punpckhqdq m1, m1
4297 mova [acq+16], m1
4298 paddw m5, m1
4299 movddup m1, [yq+strideq]
4300 pmaddubsw m1, m2
4301 pshufhw m1, m1, q3333
4302 mova [acq+32], m1
4303 paddw m4, m1
4304 punpckhqdq m0, m1, m1
4305 mova [acq+48], m0
4306 paddw m4, m0
4307 lea yq, [yq+strideq*2]
4308 add acq, 64
4309 sub hd, 2
4310 jg .w16_pad3
4311 jmp .w16_wpad_done
4312 .w16_pad2:
4313 mova m1, [yq]
4314 pmaddubsw m1, m2
4315 mova [acq], m1
4316 paddw m5, m1
4317 pshufhw m1, m1, q3333
4318 punpckhqdq m1, m1
4319 mova [acq+16], m1
4320 paddw m5, m1
4321 mova m1, [yq+strideq]
4322 pmaddubsw m1, m2
4323 mova [acq+32], m1
4324 paddw m4, m1
4325 mova m0, m1
4326 pshufhw m0, m0, q3333
4327 punpckhqdq m0, m0
4328 mova [acq+48], m0
4329 paddw m4, m0
4330 lea yq, [yq+strideq*2]
4331 add acq, 64
4332 sub hd, 2
4333 jg .w16_pad2
4334 jmp .w16_wpad_done
4335 .w16_pad1:
4336 mova m1, [yq]
4337 pmaddubsw m1, m2
4338 mova [acq], m1
4339 paddw m5, m1
4340 movddup m0, [yq+16]
4341 pmaddubsw m0, m2
4342 pshufhw m0, m0, q3333
4343 mova [acq+16], m0
4344 paddw m5, m0
4345 mova m1, [yq+strideq]
4346 pmaddubsw m1, m2
4347 mova [acq+32], m1
4348 paddw m4, m1
4349 movddup m0, [yq+strideq+16]
4350 pmaddubsw m0, m2
4351 pshufhw m0, m0, q3333
4352 mova [acq+48], m0
4353 paddw m4, m0
4354 lea yq, [yq+strideq*2]
4355 add acq, 64
4356 sub hd, 2
4357 jg .w16_pad1
4358 .w16_wpad_done:
4359 test hpadd, hpadd
4360 jz .calc_avg_8_16
4361 .w16_hpad_loop:
4362 mova [acq], m1
4363 mova [acq+16], m0
4364 paddw m4, m1
4365 paddw m5, m0
4366 mova [acq+32], m1
4367 mova [acq+48], m0
4368 paddw m4, m1
4369 paddw m5, m0
4370 add acq, 64
4371 sub hpadd, 2
4372 jg .w16_hpad_loop
4373 jmp .calc_avg_8_16
4374
4375 %if ARCH_X86_64
4376 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h, ac_bak
4377 %else
4378 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h
4379 %endif
4380 .calc_avg_4:
4381 psrlw m2, 10
4382 pmaddwd m5, m2
4383 pmaddwd m0, m4, m2
4384 jmp .calc_avg
4385 .calc_avg_8_16:
4386 mova m0, m5
4387 psrld m5, 16
4388 pslld m0, 16
4389 psrld m0, 16
4390 paddd m5, m0
4391 mova m0, m4
4392 psrld m0, 16
4393 pslld m4, 16
4394 psrld m4, 16
4395 paddd m0, m4
4396 .calc_avg:
4397 paddd m5, m0
4398 movd szd, m6
4399 psrad m6, 1
4400 tzcnt r1d, szd ; const int log2sz = ctz(width) + ctz(height);
4401 paddd m5, m6
4402 movd m1, r1d
4403 pshufd m0, m5, q2301
4404 paddd m0, m5
4405 pshufd m5, m0, q1032
4406 paddd m0, m5
4407 psrad m0, m1 ; sum >>= log2sz;
4408 packssdw m0, m0
4409 RELOAD_ACQ_32 acq ; ac = ac_orig
4410 .sub_loop:
4411 mova m1, [acq]
4412 psubw m1, m0
4413 mova [acq], m1
4414 add acq, 16
4415 sub szd, 8
4416 jg .sub_loop
4417 RET
4418
4419 %if ARCH_X86_64
4420 cglobal ipred_cfl_ac_444_8bpc, 4, 8, 7, -4*16, ac, y, stride, wpad, hpad, w, h, ac_bak
4421 movddup m2, [pb_4]
4422 %else
4423 cglobal ipred_cfl_ac_444_8bpc, 4, 7, 7, -5*16, ac, y, stride, wpad, hpad, w, h
4424 %define ac_bakq [rsp+16*4]
4425 mov t0d, 0x04040404
4426 movd m2, t0d
4427 pshufd m2, m2, q0000
4428 %endif
4429 movifnidn wd, wm
4430 movifnidn hpadd, hpadm
4431 movd m0, hpadd
4432 mov t0d, hm
4433 mov hd, t0d
4434 imul t0d, wd
4435 movd m6, t0d
4436 movd hpadd, m0
4437 mov ac_bakq, acq
4438 shl hpadd, 2
4439 sub hd, hpadd
4440 pxor m5, m5
4441 pxor m4, m4
4442 cmp wd, 16
4443 jg .w32
4444 cmp wd, 8
4445 jg .w16
4446 je .w8
4447 ; fall-through
4448
4449 %if ARCH_X86_64
4450 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, ac_bak
4451 %else
4452 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h
4453 %endif
4454 .w4:
4455 lea stride3q, [strideq*3]
4456 .w4_loop:
4457 movd m1, [yq]
4458 movd m3, [yq+strideq]
4459 punpckldq m1, m3
4460 punpcklbw m1, m1
4461 movd m0, [yq+strideq*2]
4462 movd m3, [yq+stride3q]
4463 punpckldq m0, m3
4464 punpcklbw m0, m0
4465 pmaddubsw m1, m2
4466 pmaddubsw m0, m2
4467 mova [acq], m1
4468 mova [acq+16], m0
4469 paddw m5, m0
4470 paddw m5, m1
4471 lea yq, [yq+strideq*4]
4472 add acq, 32
4473 sub hd, 4
4474 jg .w4_loop
4475 test hpadd, hpadd
4476 jz .calc_avg_4
4477 punpckhqdq m0, m0
4478 .w4_hpad_loop:
4479 mova [acq], m0
4480 paddw m5, m0
4481 add acq, 16
4482 sub hpadd, 2
4483 jg .w4_hpad_loop
4484 .calc_avg_4:
4485 psrlw m2, 10
4486 pmaddwd m5, m2
4487 jmp .calc_avg
4488
4489 .w8:
4490 lea stride3q, [strideq*3]
4491 test wpadd, wpadd
4492 jnz .w8_wpad
4493 .w8_loop:
4494 movq m1, [yq]
4495 punpcklbw m1, m1
4496 pmaddubsw m1, m2
4497 mova [acq], m1
4498 paddw m5, m1
4499 movq m0, [yq+strideq]
4500 punpcklbw m0, m0
4501 pmaddubsw m0, m2
4502 mova [acq+16], m0
4503 paddw m5, m0
4504 movq m1, [yq+strideq*2]
4505 punpcklbw m1, m1
4506 pmaddubsw m1, m2
4507 mova [acq+32], m1
4508 paddw m4, m1
4509 movq m0, [yq+stride3q]
4510 punpcklbw m0, m0
4511 pmaddubsw m0, m2
4512 mova [acq+48], m0
4513 paddw m4, m0
4514 lea yq, [yq+strideq*4]
4515 add acq, 64
4516 sub hd, 4
4517 jg .w8_loop
4518 test hpadd, hpadd
4519 jz .calc_avg_8_16
4520 jmp .w8_hpad
4521 .w8_wpad:
4522 movd m1, [yq]
4523 punpcklbw m1, m1
4524 punpcklqdq m1, m1
4525 pmaddubsw m1, m2
4526 pshufhw m1, m1, q3333
4527 mova [acq], m1
4528 paddw m5, m1
4529 movd m0, [yq+strideq]
4530 punpcklbw m0, m0
4531 punpcklqdq m0, m0
4532 pmaddubsw m0, m2
4533 pshufhw m0, m0, q3333
4534 mova [acq+16], m0
4535 paddw m4, m0
4536 lea yq, [yq+strideq*2]
4537 add acq, 32
4538 sub hd, 2
4539 jg .w8_wpad
4540 test hpadd, hpadd
4541 jz .calc_avg_8_16
4542 .w8_hpad:
4543 mova [acq], m0
4544 paddw m5, m0
4545 mova [acq+16], m0
4546 paddw m4, m0
4547 add acq, 32
4548 sub hpadd, 2
4549 jg .w8_hpad
4550 jmp .calc_avg_8_16
4551
4552 .w16:
4553 test wpadd, wpadd
4554 jnz .w16_wpad
4555 .w16_loop:
4556 mova m0, [yq]
4557 mova m1, m0
4558 punpcklbw m1, m1
4559 pmaddubsw m1, m2
4560 mova [acq], m1
4561 paddw m5, m1
4562 punpckhbw m0, m0
4563 pmaddubsw m0, m2
4564 mova [acq+16], m0
4565 paddw m5, m0
4566 mova m0, [yq+strideq]
4567 mova m1, m0
4568 punpcklbw m1, m1
4569 pmaddubsw m1, m2
4570 mova [acq+32], m1
4571 paddw m4, m1
4572 punpckhbw m0, m0
4573 pmaddubsw m0, m2
4574 mova [acq+48], m0
4575 paddw m4, m0
4576 lea yq, [yq+strideq*2]
4577 add acq, 64
4578 sub hd, 2
4579 jg .w16_loop
4580 test hpadd, hpadd
4581 jz .calc_avg_8_16
4582 jmp .w16_hpad_loop
4583 .w16_wpad:
4584 cmp wpadd, 2
4585 jl .w16_pad1
4586 je .w16_pad2
4587 .w16_pad3:
4588 movd m1, [yq]
4589 punpcklbw m1, m1
4590 punpcklqdq m1, m1
4591 pshufhw m1, m1, q3333
4592 pmaddubsw m1, m2
4593 mova [acq], m1
4594 paddw m5, m1
4595 punpckhqdq m1, m1
4596 mova [acq+16], m1
4597 paddw m5, m1
4598 movd m1, [yq+strideq]
4599 punpcklbw m1, m1
4600 punpcklqdq m1, m1
4601 pshufhw m1, m1, q3333
4602 pmaddubsw m1, m2
4603 mova [acq+32], m1
4604 paddw m4, m1
4605 punpckhqdq m0, m1, m1
4606 mova [acq+48], m0
4607 paddw m4, m0
4608 lea yq, [yq+strideq*2]
4609 add acq, 64
4610 sub hd, 2
4611 jg .w16_pad3
4612 jmp .w16_wpad_done
4613 .w16_pad2:
4614 movq m1, [yq]
4615 punpcklbw m1, m1
4616 pmaddubsw m1, m2
4617 mova [acq], m1
4618 paddw m5, m1
4619 pshufhw m1, m1, q3333
4620 punpckhqdq m1, m1
4621 mova [acq+16], m1
4622 paddw m5, m1
4623 movq m1, [yq+strideq]
4624 punpcklbw m1, m1
4625 pmaddubsw m1, m2
4626 mova [acq+32], m1
4627 paddw m4, m1
4628 mova m0, m1
4629 pshufhw m0, m0, q3333
4630 punpckhqdq m0, m0
4631 mova [acq+48], m0
4632 paddw m4, m0
4633 lea yq, [yq+strideq*2]
4634 add acq, 64
4635 sub hd, 2
4636 jg .w16_pad2
4637 jmp .w16_wpad_done
4638 .w16_pad1:
4639 mova m0, [yq]
4640 mova m1, m0
4641 punpcklbw m1, m1
4642 pmaddubsw m1, m2
4643 mova [acq], m1
4644 paddw m5, m1
4645 punpckhbw m0, m0
4646 punpcklqdq m0, m0
4647 pshufhw m0, m0, q3333
4648 pmaddubsw m0, m2
4649 mova [acq+16], m0
4650 paddw m5, m0
4651 mova m0, [yq+strideq]
4652 mova m1, m0
4653 punpcklbw m1, m1
4654 pmaddubsw m1, m2
4655 mova [acq+32], m1
4656 paddw m4, m1
4657 punpckhbw m0, m0
4658 punpcklqdq m0, m0
4659 pshufhw m0, m0, q3333
4660 pmaddubsw m0, m2
4661 mova [acq+48], m0
4662 paddw m4, m0
4663 lea yq, [yq+strideq*2]
4664 add acq, 64
4665 sub hd, 2
4666 jg .w16_pad1
4667 .w16_wpad_done:
4668 test hpadd, hpadd
4669 jz .calc_avg_8_16
4670 .w16_hpad_loop:
4671 mova [acq], m1
4672 mova [acq+16], m0
4673 paddw m4, m1
4674 paddw m5, m0
4675 mova [acq+32], m1
4676 mova [acq+48], m0
4677 paddw m4, m1
4678 paddw m5, m0
4679 add acq, 64
4680 sub hpadd, 2
4681 jg .w16_hpad_loop
4682 .calc_avg_8_16:
4683 mova m0, m5
4684 psrld m5, 16
4685 pslld m0, 16
4686 psrld m0, 16
4687 paddd m5, m0
4688 mova m0, m4
4689 psrld m0, 16
4690 pslld m4, 16
4691 psrld m4, 16
4692 paddd m0, m4
4693 paddd m5, m0
4694 jmp .calc_avg
4695
4696 .w32:
4697 pxor m0, m0
4698 mova [rsp ], m0
4699 mova [rsp+16], m0
4700 mova [rsp+32], m0
4701 mova [rsp+48], m0
4702 test wpadd, wpadd
4703 jnz .w32_wpad
4704 .w32_loop:
4705 mova m0, [yq]
4706 mova m1, m0
4707 punpcklbw m1, m1
4708 pmaddubsw m1, m2
4709 mova [acq], m1
4710 paddw m5, m1, [rsp]
4711 mova [rsp ], m5
4712 punpckhbw m0, m0
4713 pmaddubsw m0, m2
4714 mova [acq+16], m0
4715 paddw m5, m0, [rsp+16]
4716 mova [rsp+16], m5
4717 mova m4, [yq+16]
4718 mova m3, m4
4719 punpcklbw m3, m3
4720 pmaddubsw m3, m2
4721 mova [acq+32], m3
4722 paddw m5, m3, [rsp+32]
4723 mova [rsp+32], m5
4724 punpckhbw m4, m4
4725 pmaddubsw m4, m2
4726 mova [acq+48], m4
4727 paddw m5, m4, [rsp+48]
4728 mova [rsp+48], m5
4729 lea yq, [yq+strideq]
4730 add acq, 64
4731 sub hd, 1
4732 jg .w32_loop
4733 test hpadd, hpadd
4734 jz .calc_avg_32
4735 jmp .w32_hpad_loop
4736 .w32_wpad:
4737 cmp wpadd, 2
4738 jl .w32_pad1
4739 je .w32_pad2
4740 cmp wpadd, 4
4741 jl .w32_pad3
4742 je .w32_pad4
4743 cmp wpadd, 6
4744 jl .w32_pad5
4745 je .w32_pad6
4746 .w32_pad7:
4747 movd m1, [yq]
4748 punpcklbw m1, m1
4749 punpcklqdq m1, m1
4750 pshufhw m1, m1, q3333
4751 pmaddubsw m1, m2
4752 mova [acq], m1
4753 paddw m5, m1, [rsp]
4754 mova [rsp ], m5
4755 mova m0, m1
4756 punpckhqdq m0, m0
4757 mova [acq+16], m0
4758 paddw m5, m0, [rsp+16]
4759 mova [rsp+16], m5
4760 mova m3, m0
4761 mova [acq+32], m3
4762 paddw m5, m3, [rsp+32]
4763 mova [rsp+32], m5
4764 mova m4, m3
4765 mova [acq+48], m4
4766 paddw m5, m4, [rsp+48]
4767 mova [rsp+48], m5
4768 lea yq, [yq+strideq]
4769 add acq, 64
4770 sub hd, 1
4771 jg .w32_pad7
4772 jmp .w32_wpad_done
4773 .w32_pad6:
4774 mova m0, [yq]
4775 mova m1, m0
4776 punpcklbw m1, m1
4777 pmaddubsw m1, m2
4778 mova [acq], m1
4779 paddw m5, m1, [rsp]
4780 mova [rsp ], m5
4781 pshufhw m0, m1, q3333
4782 punpckhqdq m0, m0
4783 mova [acq+16], m0
4784 paddw m5, m0, [rsp+16]
4785 mova [rsp+16], m5
4786 mova m3, m0
4787 mova [acq+32], m3
4788 paddw m5, m3, [rsp+32]
4789 mova [rsp+32], m5
4790 mova m4, m3
4791 mova [acq+48], m4
4792 paddw m5, m4, [rsp+48]
4793 mova [rsp+48], m5
4794 lea yq, [yq+strideq]
4795 add acq, 64
4796 sub hd, 1
4797 jg .w32_pad6
4798 jmp .w32_wpad_done
4799 .w32_pad5:
4800 mova m0, [yq]
4801 mova m1, m0
4802 punpcklbw m1, m1
4803 pmaddubsw m1, m2
4804 mova [acq], m1
4805 mova m5, [rsp]
4806 paddw m5, m1
4807 mova [rsp ], m5
4808 punpckhbw m0, m0
4809 punpcklqdq m0, m0
4810 pshufhw m0, m0, q3333
4811 pmaddubsw m0, m2
4812 mova [acq+16], m0
4813 paddw m5, m0, [rsp+16]
4814 mova [rsp+16], m5
4815 mova m3, m0
4816 punpckhqdq m3, m3
4817 mova [acq+32], m3
4818 paddw m5, m3, [rsp+32]
4819 mova [rsp+32], m5
4820 mova m4, m3
4821 mova [acq+48], m4
4822 paddw m5, m4, [rsp+48]
4823 mova [rsp+48], m5
4824 lea yq, [yq+strideq]
4825 add acq, 64
4826 sub hd, 1
4827 jg .w32_pad5
4828 jmp .w32_wpad_done
4829 .w32_pad4:
4830 mova m0, [yq]
4831 mova m1, m0
4832 punpcklbw m1, m1
4833 pmaddubsw m1, m2
4834 mova [acq], m1
4835 paddw m5, m1, [rsp]
4836 mova [rsp ], m5
4837 punpckhbw m0, m0
4838 pmaddubsw m0, m2
4839 mova [acq+16], m0
4840 paddw m5, m0, [rsp+16]
4841 mova [rsp+16], m5
4842 mova m3, m0
4843 pshufhw m3, m3, q3333
4844 punpckhqdq m3, m3
4845 mova [acq+32], m3
4846 paddw m5, m3, [rsp+32]
4847 mova [rsp+32], m5
4848 mova m4, m3
4849 mova [acq+48], m4
4850 paddw m5, m4, [rsp+48]
4851 mova [rsp+48], m5
4852 lea yq, [yq+strideq]
4853 add acq, 64
4854 sub hd, 1
4855 jg .w32_pad4
4856 jmp .w32_wpad_done
4857 .w32_pad3:
4858 mova m0, [yq]
4859 mova m1, m0
4860 punpcklbw m1, m1
4861 pmaddubsw m1, m2
4862 mova [acq], m1
4863 paddw m5, m1, [rsp]
4864 mova [rsp ], m5
4865 punpckhbw m0, m0
4866 pmaddubsw m0, m2
4867 mova [acq+16], m0
4868 paddw m5, m0, [rsp+16]
4869 mova [rsp+16], m5
4870 movd m3, [yq+16]
4871 punpcklbw m3, m3
4872 punpcklqdq m3, m3
4873 pshufhw m3, m3, q3333
4874 pmaddubsw m3, m2
4875 mova [acq+32], m3
4876 paddw m5, m3, [rsp+32]
4877 mova [rsp+32], m5
4878 mova m4, m3
4879 punpckhqdq m4, m4
4880 mova [acq+48], m4
4881 paddw m5, m4, [rsp+48]
4882 mova [rsp+48], m5
4883 lea yq, [yq+strideq]
4884 add acq, 64
4885 sub hd, 1
4886 jg .w32_pad3
4887 jmp .w32_wpad_done
4888 .w32_pad2:
4889 mova m0, [yq]
4890 mova m1, m0
4891 punpcklbw m1, m1
4892 pmaddubsw m1, m2
4893 mova [acq], m1
4894 paddw m5, m1, [rsp]
4895 mova [rsp ], m5
4896 punpckhbw m0, m0
4897 pmaddubsw m0, m2
4898 mova [acq+16], m0
4899 paddw m5, m0, [rsp+16]
4900 mova [rsp+16], m5
4901 mova m3, [yq+16]
4902 punpcklbw m3, m3
4903 pmaddubsw m3, m2
4904 mova [acq+32], m3
4905 paddw m5, m3, [rsp+32]
4906 mova [rsp+32], m5
4907 pshufhw m4, m3, q3333
4908 punpckhqdq m4, m4
4909 mova [acq+48], m4
4910 paddw m5, m4, [rsp+48]
4911 mova [rsp+48], m5
4912 lea yq, [yq+strideq]
4913 add acq, 64
4914 sub hd, 1
4915 jg .w32_pad2
4916 jmp .w32_wpad_done
4917 .w32_pad1:
4918 mova m0, [yq]
4919 mova m1, m0
4920 punpcklbw m1, m1
4921 pmaddubsw m1, m2
4922 mova [acq], m1
4923 paddw m5, m1, [rsp]
4924 mova [rsp ], m5
4925 punpckhbw m0, m0
4926 pmaddubsw m0, m2
4927 mova [acq+16], m0
4928 paddw m5, m0, [rsp+16]
4929 mova [rsp+16], m5
4930 mova m4, [yq+16]
4931 mova m3, m4
4932 punpcklbw m3, m3
4933 pmaddubsw m3, m2
4934 mova [acq+32], m3
4935 paddw m5, m3, [rsp+32]
4936 mova [rsp+32], m5
4937 punpckhbw m4, m4
4938 punpcklqdq m4, m4
4939 pshufhw m4, m4, q3333
4940 pmaddubsw m4, m2
4941 mova [acq+48], m4
4942 paddw m5, m4, [rsp+48]
4943 mova [rsp+48], m5
4944 lea yq, [yq+strideq]
4945 add acq, 64
4946 sub hd, 1
4947 jg .w32_pad1
4948 .w32_wpad_done:
4949 test hpadd, hpadd
4950 jz .calc_avg_32
4951 .w32_hpad_loop:
4952 mova [acq], m1
4953 mova [acq+16], m0
4954 paddw m5, m1, [rsp]
4955 mova [rsp ], m5
4956 paddw m5, m0, [rsp+16]
4957 mova [rsp+16], m5
4958 mova [acq+32], m3
4959 mova [acq+48], m4
4960 paddw m5, m3, [rsp+32]
4961 mova [rsp+32], m5
4962 paddw m5, m4, [rsp+48]
4963 mova [rsp+48], m5
4964 add acq, 64
4965 sub hpadd, 1
4966 jg .w32_hpad_loop
4967
4968 %if ARCH_X86_64
4969 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h, ac_bak
4970 %else
4971 DEFINE_ARGS ac, y, iptr, wpad, hpad, sz, h
4972 %endif
4973
4974 .calc_avg_32:
4975 mova m5, [rsp]
4976 mova m0, m5
4977 psrld m5, 16
4978 pslld m0, 16
4979 psrld m0, 16
4980 paddd m5, m0
4981 mova m0, [rsp+16]
4982 mova m3, m0
4983 psrld m0, 16
4984 pslld m3, 16
4985 psrld m3, 16
4986 paddd m0, m3
4987 paddd m5, m0
4988 mova m0, [rsp+32]
4989 mova m3, m0
4990 psrld m0, 16
4991 pslld m3, 16
4992 psrld m3, 16
4993 paddd m0, m3
4994 mova m1, [rsp+48]
4995 mova m3, m1
4996 psrld m1, 16
4997 pslld m3, 16
4998 psrld m3, 16
4999 paddd m1, m3
5000 paddd m1, m0
5001 paddd m5, m1
5002 .calc_avg:
5003 movd szd, m6
5004 psrad m6, 1
5005 tzcnt r1d, szd ; const int log2sz = ctz(width) + ctz(height);
5006 paddd m5, m6
5007 movd m1, r1d
5008 pshufd m0, m5, q2301
5009 paddd m0, m5
5010 pshufd m5, m0, q1032
5011 paddd m0, m5
5012 psrad m0, m1 ; sum >>= log2sz;
5013 packssdw m0, m0
5014 RELOAD_ACQ_32 acq ; ac = ac_orig
5015 .sub_loop:
5016 mova m1, [acq]
5017 psubw m1, m0
5018 mova [acq], m1
5019 add acq, 16
5020 sub szd, 8
5021 jg .sub_loop
5022 RET
5023
5024 ; %1 simd register that hold the mask and will hold the result
5025 ; %2 simd register that holds the "true" values
5026 ; %3 location of the "false" values (simd register/memory)
5027 %macro BLEND 3 ; mask, true, false
5028 pand %2, %1
5029 pandn %1, %3
5030 por %1, %2
5031 %endmacro
5032
5033 %macro PAETH 2 ; top, ldiff
5034 pavgb m1, m%1, m3
5035 pxor m0, m%1, m3
5036 pand m0, m4
5037 psubusb m2, m5, m1
5038 psubb m1, m0
5039 psubusb m1, m5
5040 por m1, m2
5041 paddusb m1, m1
5042 por m1, m0 ; min(tldiff, 255)
5043 psubusb m2, m5, m3
5044 psubusb m0, m3, m5
5045 por m2, m0 ; tdiff
5046 %ifnum %2
5047 pminub m2, m%2
5048 pcmpeqb m0, m%2, m2 ; ldiff <= tdiff
5049 %else
5050 mova m0, %2
5051 pminub m2, m0
5052 pcmpeqb m0, m2
5053 %endif
5054 pminub m1, m2
5055 pcmpeqb m1, m2 ; ldiff <= tldiff && tdiff <= tldiff
5056 mova m2, m3
5057 BLEND m0, m2, m%1
5058 BLEND m1, m0, m5
5059 %endmacro
5060
5061 cglobal ipred_paeth_8bpc, 3, 6, 8, -7*16, dst, stride, tl, w, h
5062 %define base r5-ipred_paeth_ssse3_table
5063 tzcnt wd, wm
5064 movifnidn hd, hm
5065 pxor m0, m0
5066 movd m5, [tlq]
5067 pshufb m5, m0
5068 LEA r5, ipred_paeth_ssse3_table
5069 movsxd wq, [r5+wq*4]
5070 movddup m4, [base+ipred_paeth_shuf]
5071 add wq, r5
5072 jmp wq
5073 .w4:
5074 movd m6, [tlq+1] ; top
5075 pshufd m6, m6, q0000
5076 lea r3, [strideq*3]
5077 psubusb m7, m5, m6
5078 psubusb m0, m6, m5
5079 por m7, m0 ; ldiff
5080 .w4_loop:
5081 sub tlq, 4
5082 movd m3, [tlq]
5083 mova m1, [base+ipred_h_shuf]
5084 pshufb m3, m1 ; left
5085 PAETH 6, 7
5086 movd [dstq ], m1
5087 pshuflw m0, m1, q1032
5088 movd [dstq+strideq ], m0
5089 punpckhqdq m1, m1
5090 movd [dstq+strideq*2], m1
5091 psrlq m1, 32
5092 movd [dstq+r3 ], m1
5093 lea dstq, [dstq+strideq*4]
5094 sub hd, 4
5095 jg .w4_loop
5096 RET
5097 ALIGN function_align
5098 .w8:
5099 movddup m6, [tlq+1]
5100 psubusb m7, m5, m6
5101 psubusb m0, m6, m5
5102 por m7, m0
5103 .w8_loop:
5104 sub tlq, 2
5105 movd m3, [tlq]
5106 pshufb m3, [base+ipred_paeth_shuf]
5107 PAETH 6, 7
5108 movq [dstq ], m1
5109 movhps [dstq+strideq], m1
5110 lea dstq, [dstq+strideq*2]
5111 sub hd, 2
5112 jg .w8_loop
5113 RET
5114 ALIGN function_align
5115 .w16:
5116 movu m6, [tlq+1]
5117 psubusb m7, m5, m6
5118 psubusb m0, m6, m5
5119 por m7, m0
5120 .w16_loop:
5121 sub tlq, 1
5122 movd m3, [tlq]
5123 pxor m1, m1
5124 pshufb m3, m1
5125 PAETH 6, 7
5126 mova [dstq], m1
5127 add dstq, strideq
5128 sub hd, 1
5129 jg .w16_loop
5130 RET
5131 ALIGN function_align
5132 .w32:
5133 movu m6, [tlq+1]
5134 psubusb m7, m5, m6
5135 psubusb m0, m6, m5
5136 por m7, m0
5137 mova [rsp ], m6
5138 mova [rsp+16], m7
5139 movu m6, [tlq+17]
5140 psubusb m7, m5, m6
5141 psubusb m0, m6, m5
5142 por m7, m0
5143 mova [rsp+32], m6
5144 .w32_loop:
5145 dec tlq
5146 movd m3, [tlq]
5147 pxor m1, m1
5148 pshufb m3, m1
5149 mova m6, [rsp]
5150 PAETH 6, [rsp+16]
5151 mova [dstq ], m1
5152 mova m6, [rsp+32]
5153 PAETH 6, 7
5154 mova [dstq+16], m1
5155 add dstq, strideq
5156 dec hd
5157 jg .w32_loop
5158 RET
5159 ALIGN function_align
5160 .w64:
5161 movu m6, [tlq+1]
5162 psubusb m7, m5, m6
5163 psubusb m0, m6, m5
5164 por m7, m0
5165 mova [rsp ], m6
5166 mova [rsp+16], m7
5167 movu m6, [tlq+17]
5168 psubusb m7, m5, m6
5169 psubusb m0, m6, m5
5170 por m7, m0
5171 mova [rsp+32], m6
5172 mova [rsp+48], m7
5173 movu m6, [tlq+33]
5174 psubusb m7, m5, m6
5175 psubusb m0, m6, m5
5176 por m7, m0
5177 mova [rsp+64], m6
5178 mova [rsp+80], m7
5179 movu m6, [tlq+49]
5180 psubusb m7, m5, m6
5181 psubusb m0, m6, m5
5182 por m7, m0
5183 mova [rsp+96], m6
5184 .w64_loop:
5185 dec tlq
5186 movd m3, [tlq]
5187 pxor m1, m1
5188 pshufb m3, m1
5189 mova m6, [rsp]
5190 PAETH 6, [rsp+16]
5191 mova [dstq ], m1
5192 mova m6, [rsp+32]
5193 PAETH 6, [rsp+48]
5194 mova [dstq+16], m1
5195 mova m6, [rsp+64]
5196 PAETH 6, [rsp+80]
5197 mova [dstq+32], m1
5198 mova m6, [rsp+96]
5199 PAETH 6, 7
5200 mova [dstq+48], m1
5201 add dstq, strideq
5202 dec hd
5203 jg .w64_loop
5204 RET
5205
5206
5207 %macro FILTER 4 ;dst, src, tmp, shuf
5208 %ifnum %4
5209 pshufb m%2, m%4
5210 %else
5211 pshufb m%2, %4
5212 %endif
5213 pshufd m%1, m%2, q0000 ;p0 p1
5214 pmaddubsw m%1, m2
5215 pshufd m%3, m%2, q1111 ;p2 p3
5216 pmaddubsw m%3, m3
5217 paddw m%1, [base+pw_8]
5218 paddw m%1, m%3
5219 pshufd m%3, m%2, q2222 ;p4 p5
5220 pmaddubsw m%3, m4
5221 paddw m%1, m%3
5222 pshufd m%3, m%2, q3333 ;p6 __
5223 pmaddubsw m%3, m5
5224 paddw m%1, m%3
5225 psraw m%1, 4
5226 packuswb m%1, m%1
5227 %endmacro
5228
5229 cglobal ipred_filter_8bpc, 3, 7, 8, dst, stride, tl, w, h, filter
5230 %define base r6-$$
5231 LEA r6, $$
5232 tzcnt wd, wm
5233 %ifidn filterd, filterm
5234 movzx filterd, filterb
5235 %else
5236 movzx filterd, byte filterm
5237 %endif
5238 shl filterd, 6
5239 lea filterq, [base+filter_intra_taps+filterq]
5240 movq m0, [tlq-3] ;_ 6 5 0 1 2 3 4
5241 movsxd wq, [base+ipred_filter_ssse3_table+wq*4]
5242 mova m2, [filterq+16*0]
5243 mova m3, [filterq+16*1]
5244 mova m4, [filterq+16*2]
5245 mova m5, [filterq+16*3]
5246 lea wq, [base+ipred_filter_ssse3_table+wq]
5247 mov hd, hm
5248 jmp wq
5249 .w4:
5250 mova m1, [base+filter_shuf1]
5251 sub tlq, 3
5252 sub tlq, hq
5253 jmp .w4_loop_start
5254 .w4_loop:
5255 movd m0, [tlq+hq]
5256 punpckldq m0, m6
5257 lea dstq, [dstq+strideq*2]
5258 .w4_loop_start:
5259 FILTER 6, 0, 7, 1
5260 movd [dstq+strideq*0], m6
5261 pshuflw m6, m6, q1032
5262 movd [dstq+strideq*1], m6
5263 sub hd, 2
5264 jg .w4_loop
5265 RET
5266
5267 ALIGN function_align
5268 .w8:
5269 movq m6, [tlq+1] ;_ _ _ 0 1 2 3 4
5270 sub tlq, 5
5271 sub tlq, hq
5272
5273 .w8_loop:
5274 FILTER 7, 0, 1, [base+filter_shuf1]
5275 punpcklqdq m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5276 FILTER 0, 6, 1, [base+filter_shuf2]
5277
5278 punpckldq m6, m7, m0
5279 movq [dstq+strideq*0], m6
5280 punpckhqdq m6, m6
5281 movq [dstq+strideq*1], m6
5282
5283 movd m0, [tlq+hq] ;_ 6 5 0
5284 punpckldq m0, m6 ;_ 6 5 0 1 2 3 4
5285
5286 lea dstq, [dstq+strideq*2]
5287 sub hd, 2
5288 jg .w8_loop
5289 RET
5290
5291 ALIGN function_align
5292 .w16:
5293 movu m6, [tlq+1] ;top row
5294 sub tlq, 5
5295 sub tlq, hq
5296
5297 .w16_loop:
5298 FILTER 7, 0, 1, [base+filter_shuf1]
5299 punpcklqdq m0, m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5300 movd [dstq+strideq*0], m7
5301 psrlq m7, 32
5302 palignr m7, m6, 4
5303
5304 FILTER 6, 0, 1, [base+filter_shuf2]
5305 punpcklqdq m0, m7, m6 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5306 movd [dstq+4+strideq*0], m6
5307 psrlq m6, 32
5308 palignr m6, m7, 4
5309
5310 FILTER 7, 0, 1, [base+filter_shuf2]
5311 punpcklqdq m0, m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5312 movd [dstq+8+strideq*0], m7
5313 psrlq m7, 32
5314 palignr m7, m6, 4
5315
5316 FILTER 6, 0, 1, [base+filter_shuf2]
5317 movd [dstq+12+strideq*0], m6
5318 psrlq m6, 32
5319 palignr m6, m7, 4
5320 mova [dstq+strideq*1], m6
5321
5322 movd m0, [tlq+hq] ;_ 6 5 0
5323 punpckldq m0, m6 ;_ 6 5 0 1 2 3 4
5324
5325 lea dstq, [dstq+strideq*2]
5326 sub hd, 2
5327 jg .w16_loop
5328 RET
5329
5330 ALIGN function_align
5331 .w32:
5332 movu m6, [tlq+1] ;top row
5333 lea filterq, [tlq+17]
5334 sub tlq, 5
5335 sub tlq, hq
5336
5337 .w32_loop:
5338 FILTER 7, 0, 1, [base+filter_shuf1]
5339 punpcklqdq m0, m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5340 movd [dstq+strideq*0], m7
5341 psrlq m7, 32
5342 palignr m7, m6, 4
5343
5344 FILTER 6, 0, 1, [base+filter_shuf2]
5345 punpcklqdq m0, m7, m6 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5346 movd [dstq+4+strideq*0], m6
5347 psrlq m6, 32
5348 palignr m6, m7, 4
5349
5350 FILTER 7, 0, 1, [base+filter_shuf2]
5351 punpcklqdq m0, m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5352 movd [dstq+8+strideq*0], m7
5353 psrlq m7, 32
5354 palignr m7, m6, 4
5355
5356 FILTER 6, 0, 1, [base+filter_shuf2]
5357 movu m1, [filterq]
5358 punpckldq m0, m7, m1 ;_ _ _ 0 1 2 3 4 _ _ _ _ _ _ _ _
5359 punpcklqdq m0, m6 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5360 movd [dstq+12+strideq*0], m6
5361 psrlq m6, 32
5362 palignr m6, m7, 4
5363 mova [dstq+strideq*1], m6
5364
5365 mova m6, m1
5366
5367 FILTER 7, 0, 6, [base+filter_shuf2]
5368 punpcklqdq m0, m1, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5369 movd [dstq+16+strideq*0], m7
5370 psrlq m7, 32
5371 palignr m7, m1, 4
5372
5373 FILTER 6, 0, 1, [base+filter_shuf2]
5374 punpcklqdq m0, m7, m6 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5375 movd [dstq+20+strideq*0], m6
5376 psrlq m6, 32
5377 palignr m6, m7, 4
5378
5379 FILTER 7, 0, 1, [base+filter_shuf2]
5380 punpcklqdq m0, m6, m7 ;_ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
5381 movd [dstq+24+strideq*0], m7
5382 psrlq m7, 32
5383 palignr m7, m6, 4
5384
5385 FILTER 6, 0, 1, [base+filter_shuf2]
5386 movd [dstq+28+strideq*0], m6
5387 psrlq m6, 32
5388 palignr m6, m7, 4
5389 mova [dstq+16+strideq*1], m6
5390
5391 mova m6, [dstq+strideq*1]
5392 movd m0, [tlq+hq] ;_ 6 5 0
5393 punpckldq m0, m6 ;_ 6 5 0 1 2 3 4
5394 lea filterq, [dstq+16+strideq*1]
5395 lea dstq, [dstq+strideq*2]
5396 sub hd, 2
5397 jg .w32_loop
5398 RET