Debian Codegraph @

.cargo/vendor/rav1e/src/x86/ipred_avx2.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 "ext/x86/x86inc.asm"
28
29 %if ARCH_X86_64
30
31 SECTION_RODATA 64
32
33 %macro SMOOTH_WEIGHT_TABLE 1-*
34 %rep %0
35 db %1-128, 127-%1
36 %rotate 1
37 %endrep
38 %endmacro
39
40 ; sm_weights[], but modified to precalculate x and 256-x with offsets to
41 ; enable efficient use of pmaddubsw (which requires signed values)
42 smooth_weights: SMOOTH_WEIGHT_TABLE \
43 0, 0, 255, 128, 255, 149, 85, 64, \
44 255, 197, 146, 105, 73, 50, 37, 32, \
45 255, 225, 196, 170, 145, 123, 102, 84, \
46 68, 54, 43, 33, 26, 20, 17, 16, \
47 255, 240, 225, 210, 196, 182, 169, 157, \
48 145, 133, 122, 111, 101, 92, 83, 74, \
49 66, 59, 52, 45, 39, 34, 29, 25, \
50 21, 17, 14, 12, 10, 9, 8, 8, \
51 255, 248, 240, 233, 225, 218, 210, 203, \
52 196, 189, 182, 176, 169, 163, 156, 150, \
53 144, 138, 133, 127, 121, 116, 111, 106, \
54 101, 96, 91, 86, 82, 77, 73, 69, \
55 65, 61, 57, 54, 50, 47, 44, 41, \
56 38, 35, 32, 29, 27, 25, 22, 20, \
57 18, 16, 15, 13, 12, 10, 9, 8, \
58 7, 6, 6, 5, 5, 4, 4, 4
59
60 pb_1to32: db 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
61 db 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32
62 pb_32to1: db 32, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 18, 17
63 pb_16to1: db 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
64 z_filter_wh: db 7, 7, 11, 11, 15, 15, 19, 19, 19, 23, 23, 23, 31, 31, 31, 39
65 db 39, 39, 47, 47, 47, 63, 63, 63, 79, 79, 79, -1
66 z_filter_k: db 0, 16, 0, 16, 0, 20, 0, 20, 8, 16, 8, 16
67 db 32, 16, 32, 16, 24, 20, 24, 20, 16, 16, 16, 16
68 db 0, 0, 0, 0, 0, 0, 0, 0, 8, 0, 8, 0
69 z_filter_s: db 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7
70 db 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14, 15
71 db 15, 15, 15, 15, 15, 15, 15, 15 ; should be in one cache line
72 pb_128: times 4 db 128 ; those are just placed here for alignment.
73 pb_36_m4: times 2 db 36, -4
74 z3_shuf: db 8, 7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0
75 z_filter_t0: db 55,127, 39,127, 39,127, 7, 15, 31, 7, 15, 31, 0, 3, 31, 0
76 z_filter_t1: db 39, 63, 19, 47, 19, 47, 3, 3, 3, 3, 3, 3, 0, 0, 0, 0
77 z_upsample1: db 1, 0, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7, 6, 8, 7
78 z_upsample2: db 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 8, 8, 8
79 z2_upsample: db 7, 6, 15, 14, 5, 4, 13, 12, 3, 2, 11, 10, 1, 0, 9, 8
80 z1_shuf_w4: db 0, 1, 1, 2, 2, 3, 3, 4, 8, 9, 9, 10, 10, 11, 11, 12
81 z2_shuf_h2: db 3, 2, 7, 6, 11, 10, 15, 14, 2, 1, 6, 5, 10, 9, 14, 13
82 z2_shuf_h4: db 7, 6, 15, 14, 6, 5, 14, 13, 5, 4, 13, 12, 4, 3, 12, 11
83 z3_shuf_w4: db 4, 3, 3, 2, 2, 1, 1, 0, 12, 11, 11, 10, 10, 9, 9, 8
84 z_transpose4: db 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15
85 z_base_inc: dw 0*64, 1*64, 2*64, 3*64, 4*64, 5*64, 6*64, 7*64
86 dw 16*64, 17*64, 18*64, 19*64, 20*64, 21*64, 22*64, 23*64
87 z2_base_inc: dw 1*64, 2*64, 3*64, 4*64, 5*64, 6*64, 7*64, 8*64
88 dw 9*64, 10*64, 11*64, 12*64, 13*64, 14*64, 15*64, 16*64
89 z2_ymul: dw 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
90 z2_y_shuf_h4: db 90, 90, 90, 90, 14, 14, 14, 14, 27, 27, 27, 27, 31, 31, 31, 31 ; 2, 6, 3, 7
91 db 32, 32, 32, 32, 12, 12, 12, 12, 1, 0, 1, 0, 5, -1, -1, -1 ; 0, 4, 1, 5
92 ; vpermd indices in bits 4..6 of filter_shuf1: 0, 2, 6, 4, 1, 3, 7, 5
93 filter_shuf1: db 10, 4, 10, 4, 37, 6, 5, 6,103, 9, 7, 9, 72, -1, 8, -1
94 db 16, 4, 0, 4, 53, 6, 5, 6,119, 11, 7, 11, 95, -1, 15, -1
95 filter_shuf2: db 3, 4, 3, 4, 5, 6, 5, 6, 7, 2, 7, 2, 1, -1, 1, -1
96 filter_shuf3: db 3, 4, 3, 4, 5, 6, 5, 6, 7, 11, 7, 11; 15, -1, 15, -1
97 pb_127_m127: times 2 db 127, -127
98 ipred_v_shuf: db 0, 1, 0, 1, 4, 5, 4, 5, 8, 9, 8, 9, 12, 13, 12, 13
99 db 2, 3, 2, 3, 6, 7, 6, 7, 10, 11, 10, 11, 14, 15, 14, 15
100 ipred_h_shuf: db 7, 7, 7, 7, 3, 3, 3, 3, 5, 5, 5, 5, 1, 1, 1, 1
101 db 6, 6, 6, 6, 2, 2, 2, 2, 4, 4, 4, 4; 0, 0, 0, 0
102 pw_64: times 2 dw 64
103
104 cfl_ac_444_w16_pad1_shuffle: db 0, -1, 1, -1, 2, -1, 3, -1, 4, -1, 5, -1, 6, -1
105 times 9 db 7, -1
106 cfl_ac_w16_pad_shuffle: ; w=16, w_pad=1
107 db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
108 ; w=8, w_pad=1 as well as second half of previous one
109 cfl_ac_w8_pad1_shuffle: db 0, 1, 2, 3, 4, 5
110 times 5 db 6, 7
111 ; w=16,w_pad=2
112 db 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
113 times 8 db 14, 15
114 ; w=16,w_pad=3
115 db 0, 1, 2, 3, 4, 5
116 times 13 db 6, 7
117 pb_15to0: db 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
118
119 %define pb_0to15 cfl_ac_w16_pad_shuffle
120 %define pb_1 (ipred_h_shuf+12)
121 %define pb_2 (ipred_h_shuf+20)
122 %define pb_3 (ipred_h_shuf+ 4)
123 %define pb_4 (ipred_h_shuf+24)
124 %define pb_5 (ipred_h_shuf+ 8)
125 %define pb_7 (ipred_h_shuf+ 0)
126 %define pb_8 (z_upsample2 +12)
127 %define pb_12 (z2_y_shuf_h4+20)
128 %define pb_14 (z2_y_shuf_h4+ 4)
129 %define pb_15 (z_filter_s +32)
130 %define pb_27 (z2_y_shuf_h4+ 8)
131 %define pb_31 (z2_y_shuf_h4+12)
132 %define pb_32 (z2_y_shuf_h4+16)
133 %define pb_90 (z2_y_shuf_h4+ 0)
134 %define pw_1 (z2_y_shuf_h4+24)
135 %define pw_8 (z_filter_k +32)
136
137 pw_62: times 2 dw 62
138 pw_128: times 2 dw 128
139 pw_255: times 2 dw 255
140 pw_512: times 2 dw 512
141
142 %macro JMP_TABLE 3-*
143 %xdefine %1_%2_table (%%table - 2*4)
144 %xdefine %%base mangle(private_prefix %+ _%1_8bpc_%2)
145 %%table:
146 %rep %0 - 2
147 dd %%base %+ .%3 - (%%table - 2*4)
148 %rotate 1
149 %endrep
150 %endmacro
151
152 %define ipred_dc_splat_avx2_table (ipred_dc_avx2_table + 10*4)
153 %define ipred_cfl_splat_avx2_table (ipred_cfl_avx2_table + 8*4)
154
155 JMP_TABLE ipred_smooth, avx2, w4, w8, w16, w32, w64
156 JMP_TABLE ipred_smooth_v, avx2, w4, w8, w16, w32, w64
157 JMP_TABLE ipred_smooth_h, avx2, w4, w8, w16, w32, w64
158 JMP_TABLE ipred_paeth, avx2, w4, w8, w16, w32, w64
159 JMP_TABLE ipred_filter, avx2, w4, w8, w16, w32
160 JMP_TABLE ipred_dc, avx2, h4, h8, h16, h32, h64, w4, w8, w16, w32, w64, \
161 s4-10*4, s8-10*4, s16-10*4, s32-10*4, s64-10*4
162 JMP_TABLE ipred_dc_left, avx2, h4, h8, h16, h32, h64
163 JMP_TABLE ipred_h, avx2, w4, w8, w16, w32, w64
164 JMP_TABLE ipred_z1, avx2, w4, w8, w16, w32, w64
165 JMP_TABLE ipred_z2, avx2, w4, w8, w16, w32, w64
166 JMP_TABLE ipred_z3, avx2, h4, h8, h16, h32, h64
167 JMP_TABLE ipred_cfl, avx2, h4, h8, h16, h32, w4, w8, w16, w32, \
168 s4-8*4, s8-8*4, s16-8*4, s32-8*4
169 JMP_TABLE ipred_cfl_left, avx2, h4, h8, h16, h32
170 JMP_TABLE ipred_cfl_ac_420, avx2, w16_pad1, w16_pad2, w16_pad3
171 JMP_TABLE ipred_cfl_ac_422, avx2, w16_pad1, w16_pad2, w16_pad3
172 JMP_TABLE ipred_cfl_ac_444, avx2, w32_pad1, w32_pad2, w32_pad3, w4, w8, w16, w32
173 JMP_TABLE pal_pred, avx2, w4, w8, w16, w32, w64
174
175 cextern dr_intra_derivative
176 cextern filter_intra_taps
177
178 SECTION .text
179
180 INIT_YMM avx2
181 cglobal ipred_dc_top_8bpc, 3, 7, 6, dst, stride, tl, w, h
182 lea r5, [ipred_dc_left_avx2_table]
183 tzcnt wd, wm
184 inc tlq
185 movu m0, [tlq]
186 movifnidn hd, hm
187 mov r6d, 0x8000
188 shrx r6d, r6d, wd
189 movd xm3, r6d
190 movsxd r6, [r5+wq*4]
191 pcmpeqd m2, m2
192 pmaddubsw m0, m2
193 add r6, r5
194 add r5, ipred_dc_splat_avx2_table-ipred_dc_left_avx2_table
195 movsxd wq, [r5+wq*4]
196 add wq, r5
197 jmp r6
198
199 cglobal ipred_dc_left_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
200 mov hd, hm ; zero upper half
201 tzcnt r6d, hd
202 sub tlq, hq
203 tzcnt wd, wm
204 movu m0, [tlq]
205 mov r5d, 0x8000
206 shrx r5d, r5d, r6d
207 movd xm3, r5d
208 lea r5, [ipred_dc_left_avx2_table]
209 movsxd r6, [r5+r6*4]
210 pcmpeqd m2, m2
211 pmaddubsw m0, m2
212 add r6, r5
213 add r5, ipred_dc_splat_avx2_table-ipred_dc_left_avx2_table
214 movsxd wq, [r5+wq*4]
215 add wq, r5
216 jmp r6
217 .h64:
218 movu m1, [tlq+32] ; unaligned when jumping here from dc_top
219 pmaddubsw m1, m2
220 paddw m0, m1
221 .h32:
222 vextracti128 xm1, m0, 1
223 paddw xm0, xm1
224 .h16:
225 punpckhqdq xm1, xm0, xm0
226 paddw xm0, xm1
227 .h8:
228 psrlq xm1, xm0, 32
229 paddw xm0, xm1
230 .h4:
231 pmaddwd xm0, xm2
232 pmulhrsw xm0, xm3
233 lea stride3q, [strideq*3]
234 vpbroadcastb m0, xm0
235 mova m1, m0
236 jmp wq
237
238 cglobal ipred_dc_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
239 movifnidn hd, hm
240 movifnidn wd, wm
241 tzcnt r6d, hd
242 lea r5d, [wq+hq]
243 movd xm4, r5d
244 tzcnt r5d, r5d
245 movd xm5, r5d
246 lea r5, [ipred_dc_avx2_table]
247 tzcnt wd, wd
248 movsxd r6, [r5+r6*4]
249 movsxd wq, [r5+wq*4+5*4]
250 pcmpeqd m3, m3
251 psrlw xm4, 1
252 add r6, r5
253 add wq, r5
254 lea stride3q, [strideq*3]
255 jmp r6
256 .h4:
257 movd xm0, [tlq-4]
258 pmaddubsw xm0, xm3
259 jmp wq
260 .w4:
261 movd xm1, [tlq+1]
262 pmaddubsw xm1, xm3
263 psubw xm0, xm4
264 paddw xm0, xm1
265 pmaddwd xm0, xm3
266 cmp hd, 4
267 jg .w4_mul
268 psrlw xm0, 3
269 jmp .w4_end
270 .w4_mul:
271 punpckhqdq xm1, xm0, xm0
272 lea r2d, [hq*2]
273 mov r6d, 0x55563334
274 paddw xm0, xm1
275 shrx r6d, r6d, r2d
276 psrlq xm1, xm0, 32
277 paddw xm0, xm1
278 movd xm1, r6d
279 psrlw xm0, 2
280 pmulhuw xm0, xm1
281 .w4_end:
282 vpbroadcastb xm0, xm0
283 .s4:
284 movd [dstq+strideq*0], xm0
285 movd [dstq+strideq*1], xm0
286 movd [dstq+strideq*2], xm0
287 movd [dstq+stride3q ], xm0
288 lea dstq, [dstq+strideq*4]
289 sub hd, 4
290 jg .s4
291 RET
292 ALIGN function_align
293 .h8:
294 movq xm0, [tlq-8]
295 pmaddubsw xm0, xm3
296 jmp wq
297 .w8:
298 movq xm1, [tlq+1]
299 vextracti128 xm2, m0, 1
300 pmaddubsw xm1, xm3
301 psubw xm0, xm4
302 paddw xm0, xm2
303 punpckhqdq xm2, xm0, xm0
304 paddw xm0, xm2
305 paddw xm0, xm1
306 psrlq xm1, xm0, 32
307 paddw xm0, xm1
308 pmaddwd xm0, xm3
309 psrlw xm0, xm5
310 cmp hd, 8
311 je .w8_end
312 mov r6d, 0x5556
313 mov r2d, 0x3334
314 cmp hd, 32
315 cmove r6d, r2d
316 movd xm1, r6d
317 pmulhuw xm0, xm1
318 .w8_end:
319 vpbroadcastb xm0, xm0
320 .s8:
321 movq [dstq+strideq*0], xm0
322 movq [dstq+strideq*1], xm0
323 movq [dstq+strideq*2], xm0
324 movq [dstq+stride3q ], xm0
325 lea dstq, [dstq+strideq*4]
326 sub hd, 4
327 jg .s8
328 RET
329 ALIGN function_align
330 .h16:
331 mova xm0, [tlq-16]
332 pmaddubsw xm0, xm3
333 jmp wq
334 .w16:
335 movu xm1, [tlq+1]
336 vextracti128 xm2, m0, 1
337 pmaddubsw xm1, xm3
338 psubw xm0, xm4
339 paddw xm0, xm2
340 paddw xm0, xm1
341 punpckhqdq xm1, xm0, xm0
342 paddw xm0, xm1
343 psrlq xm1, xm0, 32
344 paddw xm0, xm1
345 pmaddwd xm0, xm3
346 psrlw xm0, xm5
347 cmp hd, 16
348 je .w16_end
349 mov r6d, 0x5556
350 mov r2d, 0x3334
351 test hb, 8|32
352 cmovz r6d, r2d
353 movd xm1, r6d
354 pmulhuw xm0, xm1
355 .w16_end:
356 vpbroadcastb xm0, xm0
357 .s16:
358 mova [dstq+strideq*0], xm0
359 mova [dstq+strideq*1], xm0
360 mova [dstq+strideq*2], xm0
361 mova [dstq+stride3q ], xm0
362 lea dstq, [dstq+strideq*4]
363 sub hd, 4
364 jg .s16
365 RET
366 ALIGN function_align
367 .h32:
368 mova m0, [tlq-32]
369 pmaddubsw m0, m3
370 jmp wq
371 .w32:
372 movu m1, [tlq+1]
373 pmaddubsw m1, m3
374 paddw m0, m1
375 vextracti128 xm1, m0, 1
376 psubw xm0, xm4
377 paddw xm0, xm1
378 punpckhqdq xm1, xm0, xm0
379 paddw xm0, xm1
380 psrlq xm1, xm0, 32
381 paddw xm0, xm1
382 pmaddwd xm0, xm3
383 psrlw xm0, xm5
384 cmp hd, 32
385 je .w32_end
386 lea r2d, [hq*2]
387 mov r6d, 0x33345556
388 shrx r6d, r6d, r2d
389 movd xm1, r6d
390 pmulhuw xm0, xm1
391 .w32_end:
392 vpbroadcastb m0, xm0
393 .s32:
394 mova [dstq+strideq*0], m0
395 mova [dstq+strideq*1], m0
396 mova [dstq+strideq*2], m0
397 mova [dstq+stride3q ], m0
398 lea dstq, [dstq+strideq*4]
399 sub hd, 4
400 jg .s32
401 RET
402 ALIGN function_align
403 .h64:
404 mova m0, [tlq-64]
405 mova m1, [tlq-32]
406 pmaddubsw m0, m3
407 pmaddubsw m1, m3
408 paddw m0, m1
409 jmp wq
410 .w64:
411 movu m1, [tlq+ 1]
412 movu m2, [tlq+33]
413 pmaddubsw m1, m3
414 pmaddubsw m2, m3
415 paddw m0, m1
416 paddw m0, m2
417 vextracti128 xm1, m0, 1
418 psubw xm0, xm4
419 paddw xm0, xm1
420 punpckhqdq xm1, xm0, xm0
421 paddw xm0, xm1
422 psrlq xm1, xm0, 32
423 paddw xm0, xm1
424 pmaddwd xm0, xm3
425 psrlw xm0, xm5
426 cmp hd, 64
427 je .w64_end
428 mov r6d, 0x33345556
429 shrx r6d, r6d, hd
430 movd xm1, r6d
431 pmulhuw xm0, xm1
432 .w64_end:
433 vpbroadcastb m0, xm0
434 mova m1, m0
435 .s64:
436 mova [dstq+strideq*0+32*0], m0
437 mova [dstq+strideq*0+32*1], m1
438 mova [dstq+strideq*1+32*0], m0
439 mova [dstq+strideq*1+32*1], m1
440 mova [dstq+strideq*2+32*0], m0
441 mova [dstq+strideq*2+32*1], m1
442 mova [dstq+stride3q +32*0], m0
443 mova [dstq+stride3q +32*1], m1
444 lea dstq, [dstq+strideq*4]
445 sub hd, 4
446 jg .s64
447 RET
448
449 cglobal ipred_dc_128_8bpc, 2, 7, 6, dst, stride, tl, w, h, stride3
450 lea r5, [ipred_dc_splat_avx2_table]
451 tzcnt wd, wm
452 movifnidn hd, hm
453 movsxd wq, [r5+wq*4]
454 vpbroadcastd m0, [r5-ipred_dc_splat_avx2_table+pb_128]
455 mova m1, m0
456 add wq, r5
457 lea stride3q, [strideq*3]
458 jmp wq
459
460 cglobal ipred_v_8bpc, 3, 7, 6, dst, stride, tl, w, h, stride3
461 lea r5, [ipred_dc_splat_avx2_table]
462 tzcnt wd, wm
463 movu m0, [tlq+ 1]
464 movu m1, [tlq+33]
465 movifnidn hd, hm
466 movsxd wq, [r5+wq*4]
467 add wq, r5
468 lea stride3q, [strideq*3]
469 jmp wq
470
471 %macro IPRED_H 2 ; w, store_type
472 vpbroadcastb m0, [tlq-1]
473 vpbroadcastb m1, [tlq-2]
474 vpbroadcastb m2, [tlq-3]
475 sub tlq, 4
476 vpbroadcastb m3, [tlq+0]
477 mov%2 [dstq+strideq*0], m0
478 mov%2 [dstq+strideq*1], m1
479 mov%2 [dstq+strideq*2], m2
480 mov%2 [dstq+stride3q ], m3
481 lea dstq, [dstq+strideq*4]
482 sub hd, 4
483 jg .w%1
484 RET
485 ALIGN function_align
486 %endmacro
487
488 INIT_XMM avx2
489 cglobal ipred_h_8bpc, 3, 6, 4, dst, stride, tl, w, h, stride3
490 lea r5, [ipred_h_avx2_table]
491 tzcnt wd, wm
492 movifnidn hd, hm
493 movsxd wq, [r5+wq*4]
494 add wq, r5
495 lea stride3q, [strideq*3]
496 jmp wq
497 .w4:
498 IPRED_H 4, d
499 .w8:
500 IPRED_H 8, q
501 .w16:
502 IPRED_H 16, a
503 INIT_YMM avx2
504 .w32:
505 IPRED_H 32, a
506 .w64:
507 vpbroadcastb m0, [tlq-1]
508 vpbroadcastb m1, [tlq-2]
509 vpbroadcastb m2, [tlq-3]
510 sub tlq, 4
511 vpbroadcastb m3, [tlq+0]
512 mova [dstq+strideq*0+32*0], m0
513 mova [dstq+strideq*0+32*1], m0
514 mova [dstq+strideq*1+32*0], m1
515 mova [dstq+strideq*1+32*1], m1
516 mova [dstq+strideq*2+32*0], m2
517 mova [dstq+strideq*2+32*1], m2
518 mova [dstq+stride3q +32*0], m3
519 mova [dstq+stride3q +32*1], m3
520 lea dstq, [dstq+strideq*4]
521 sub hd, 4
522 jg .w64
523 RET
524
525 %macro PAETH 2 ; top, ldiff
526 pavgb m1, m%1, m3 ; Calculating tldiff normally requires
527 pxor m0, m%1, m3 ; 10-bit intermediates, but we can do it
528 pand m0, m4 ; in 8-bit with some tricks which avoids
529 psubusb m2, m5, m1 ; having to unpack everything to 16-bit.
530 psubb m1, m0
531 psubusb m1, m5
532 por m1, m2
533 paddusb m1, m1
534 por m1, m0 ; min(tldiff, 255)
535 psubusb m2, m5, m3
536 psubusb m0, m3, m5
537 por m2, m0 ; tdiff
538 pminub m2, m%2
539 pcmpeqb m0, m%2, m2 ; ldiff <= tdiff
540 vpblendvb m0, m%1, m3, m0
541 pminub m1, m2
542 pcmpeqb m1, m2 ; ldiff <= tldiff || tdiff <= tldiff
543 vpblendvb m0, m5, m0, m1
544 %endmacro
545
546 cglobal ipred_paeth_8bpc, 3, 6, 9, dst, stride, tl, w, h
547 %define base r5-ipred_paeth_avx2_table
548 lea r5, [ipred_paeth_avx2_table]
549 tzcnt wd, wm
550 vpbroadcastb m5, [tlq] ; topleft
551 movifnidn hd, hm
552 movsxd wq, [r5+wq*4]
553 vpbroadcastd m4, [base+pb_1]
554 add wq, r5
555 jmp wq
556 .w4:
557 vpbroadcastd m6, [tlq+1] ; top
558 mova m8, [base+ipred_h_shuf]
559 lea r3, [strideq*3]
560 psubusb m7, m5, m6
561 psubusb m0, m6, m5
562 por m7, m0 ; ldiff
563 .w4_loop:
564 sub tlq, 8
565 vpbroadcastq m3, [tlq]
566 pshufb m3, m8 ; left
567 PAETH 6, 7
568 vextracti128 xm1, m0, 1
569 movd [dstq+strideq*0], xm0
570 movd [dstq+strideq*1], xm1
571 pextrd [dstq+strideq*2], xm0, 2
572 pextrd [dstq+r3 ], xm1, 2
573 cmp hd, 4
574 je .ret
575 lea dstq, [dstq+strideq*4]
576 pextrd [dstq+strideq*0], xm0, 1
577 pextrd [dstq+strideq*1], xm1, 1
578 pextrd [dstq+strideq*2], xm0, 3
579 pextrd [dstq+r3 ], xm1, 3
580 lea dstq, [dstq+strideq*4]
581 sub hd, 8
582 jg .w4_loop
583 .ret:
584 RET
585 ALIGN function_align
586 .w8:
587 vpbroadcastq m6, [tlq+1]
588 mova m8, [base+ipred_h_shuf]
589 lea r3, [strideq*3]
590 psubusb m7, m5, m6
591 psubusb m0, m6, m5
592 por m7, m0
593 .w8_loop:
594 sub tlq, 4
595 vpbroadcastd m3, [tlq]
596 pshufb m3, m8
597 PAETH 6, 7
598 vextracti128 xm1, m0, 1
599 movq [dstq+strideq*0], xm0
600 movq [dstq+strideq*1], xm1
601 movhps [dstq+strideq*2], xm0
602 movhps [dstq+r3 ], xm1
603 lea dstq, [dstq+strideq*4]
604 sub hd, 4
605 jg .w8_loop
606 RET
607 ALIGN function_align
608 .w16:
609 vbroadcasti128 m6, [tlq+1]
610 mova xm8, xm4 ; lower half = 1, upper half = 0
611 psubusb m7, m5, m6
612 psubusb m0, m6, m5
613 por m7, m0
614 .w16_loop:
615 sub tlq, 2
616 vpbroadcastd m3, [tlq]
617 pshufb m3, m8
618 PAETH 6, 7
619 mova [dstq+strideq*0], xm0
620 vextracti128 [dstq+strideq*1], m0, 1
621 lea dstq, [dstq+strideq*2]
622 sub hd, 2
623 jg .w16_loop
624 RET
625 ALIGN function_align
626 .w32:
627 movu m6, [tlq+1]
628 psubusb m7, m5, m6
629 psubusb m0, m6, m5
630 por m7, m0
631 .w32_loop:
632 dec tlq
633 vpbroadcastb m3, [tlq]
634 PAETH 6, 7
635 mova [dstq], m0
636 add dstq, strideq
637 dec hd
638 jg .w32_loop
639 RET
640 ALIGN function_align
641 .w64:
642 movu m6, [tlq+ 1]
643 movu m7, [tlq+33]
644 %if WIN64
645 movaps r4m, xmm9
646 %endif
647 psubusb m8, m5, m6
648 psubusb m0, m6, m5
649 psubusb m9, m5, m7
650 psubusb m1, m7, m5
651 por m8, m0
652 por m9, m1
653 .w64_loop:
654 dec tlq
655 vpbroadcastb m3, [tlq]
656 PAETH 6, 8
657 mova [dstq+32*0], m0
658 PAETH 7, 9
659 mova [dstq+32*1], m0
660 add dstq, strideq
661 dec hd
662 jg .w64_loop
663 %if WIN64
664 movaps xmm9, r4m
665 %endif
666 RET
667
668 %macro SMOOTH 6 ; src[1-2], mul[1-2], add[1-2]
669 ; w * a = (w - 128) * a + 128 * a
670 ; (256 - w) * b = (127 - w) * b + 129 * b
671 pmaddubsw m0, m%3, m%1
672 pmaddubsw m1, m%4, m%2
673 paddw m0, m%5
674 paddw m1, m%6
675 psrlw m0, 8
676 psrlw m1, 8
677 packuswb m0, m1
678 %endmacro
679
680 cglobal ipred_smooth_v_8bpc, 3, 7, 0, dst, stride, tl, w, h, weights
681 %define base r6-ipred_smooth_v_avx2_table
682 lea r6, [ipred_smooth_v_avx2_table]
683 tzcnt wd, wm
684 mov hd, hm
685 movsxd wq, [r6+wq*4]
686 vpbroadcastd m0, [base+pb_127_m127]
687 vpbroadcastd m1, [base+pw_128]
688 lea weightsq, [base+smooth_weights+hq*4]
689 neg hq
690 vpbroadcastb m5, [tlq+hq] ; bottom
691 add wq, r6
692 jmp wq
693 .w4:
694 vpbroadcastd m2, [tlq+1]
695 punpcklbw m2, m5 ; top, bottom
696 mova m5, [base+ipred_v_shuf]
697 lea r3, [strideq*3]
698 punpckldq m4, m5, m5
699 punpckhdq m5, m5
700 pmaddubsw m3, m2, m0
701 paddw m1, m2 ; 1 * top + 256 * bottom + 128, overflow is ok
702 paddw m3, m1 ; 128 * top + 129 * bottom + 128
703 .w4_loop:
704 vbroadcasti128 m1, [weightsq+hq*2]
705 pshufb m0, m1, m4
706 pshufb m1, m5
707 SMOOTH 0, 1, 2, 2, 3, 3
708 vextracti128 xm1, m0, 1
709 movd [dstq+strideq*0], xm0
710 movd [dstq+strideq*1], xm1
711 pextrd [dstq+strideq*2], xm0, 1
712 pextrd [dstq+r3 ], xm1, 1
713 cmp hd, -4
714 je .ret
715 lea dstq, [dstq+strideq*4]
716 pextrd [dstq+strideq*0], xm0, 2
717 pextrd [dstq+strideq*1], xm1, 2
718 pextrd [dstq+strideq*2], xm0, 3
719 pextrd [dstq+r3 ], xm1, 3
720 lea dstq, [dstq+strideq*4]
721 add hq, 8
722 jl .w4_loop
723 .ret:
724 RET
725 ALIGN function_align
726 .w8:
727 vpbroadcastq m2, [tlq+1]
728 punpcklbw m2, m5
729 mova m5, [base+ipred_v_shuf]
730 lea r3, [strideq*3]
731 pshufd m4, m5, q0000
732 pshufd m5, m5, q1111
733 pmaddubsw m3, m2, m0
734 paddw m1, m2
735 paddw m3, m1
736 .w8_loop:
737 vpbroadcastq m1, [weightsq+hq*2]
738 pshufb m0, m1, m4
739 pshufb m1, m5
740 SMOOTH 0, 1, 2, 2, 3, 3
741 vextracti128 xm1, m0, 1
742 movq [dstq+strideq*0], xm0
743 movq [dstq+strideq*1], xm1
744 movhps [dstq+strideq*2], xm0
745 movhps [dstq+r3 ], xm1
746 lea dstq, [dstq+strideq*4]
747 add hq, 4
748 jl .w8_loop
749 RET
750 ALIGN function_align
751 .w16:
752 WIN64_SPILL_XMM 7
753 vbroadcasti128 m3, [tlq+1]
754 mova m6, [base+ipred_v_shuf]
755 punpcklbw m2, m3, m5
756 punpckhbw m3, m5
757 pmaddubsw m4, m2, m0
758 pmaddubsw m5, m3, m0
759 paddw m0, m1, m2
760 paddw m1, m3
761 paddw m4, m0
762 paddw m5, m1
763 .w16_loop:
764 vpbroadcastd m1, [weightsq+hq*2]
765 pshufb m1, m6
766 SMOOTH 1, 1, 2, 3, 4, 5
767 mova [dstq+strideq*0], xm0
768 vextracti128 [dstq+strideq*1], m0, 1
769 lea dstq, [dstq+strideq*2]
770 add hq, 2
771 jl .w16_loop
772 RET
773 ALIGN function_align
774 .w32:
775 %assign stack_offset stack_offset - stack_size_padded
776 WIN64_SPILL_XMM 6
777 movu m3, [tlq+1]
778 punpcklbw m2, m3, m5
779 punpckhbw m3, m5
780 pmaddubsw m4, m2, m0
781 pmaddubsw m5, m3, m0
782 paddw m0, m1, m2
783 paddw m1, m3
784 paddw m4, m0
785 paddw m5, m1
786 .w32_loop:
787 vpbroadcastw m1, [weightsq+hq*2]
788 SMOOTH 1, 1, 2, 3, 4, 5
789 mova [dstq], m0
790 add dstq, strideq
791 inc hq
792 jl .w32_loop
793 RET
794 ALIGN function_align
795 .w64:
796 WIN64_SPILL_XMM 11
797 movu m4, [tlq+ 1]
798 movu m8, [tlq+33]
799 punpcklbw m3, m4, m5
800 punpckhbw m4, m5
801 punpcklbw m7, m8, m5
802 punpckhbw m8, m5
803 pmaddubsw m5, m3, m0
804 pmaddubsw m6, m4, m0
805 pmaddubsw m9, m7, m0
806 pmaddubsw m10, m8, m0
807 paddw m2, m1, m3
808 paddw m5, m2
809 paddw m2, m1, m4
810 paddw m6, m2
811 paddw m0, m1, m7
812 paddw m9, m0
813 paddw m1, m8
814 paddw m10, m1
815 .w64_loop:
816 vpbroadcastw m2, [weightsq+hq*2]
817 SMOOTH 2, 2, 3, 4, 5, 6
818 mova [dstq+32*0], m0
819 SMOOTH 2, 2, 7, 8, 9, 10
820 mova [dstq+32*1], m0
821 add dstq, strideq
822 inc hq
823 jl .w64_loop
824 RET
825
826 %macro SETUP_STACK_FRAME 3 ; stack_size, regs_used, xmm_regs_used
827 %assign stack_offset 0
828 %assign stack_size_padded 0
829 %assign regs_used %2
830 %xdefine rstk rsp
831 SETUP_STACK_POINTER %1
832 %if regs_used != %2 && WIN64
833 PUSH r%2
834 %endif
835 ALLOC_STACK %1, %3
836 %endmacro
837
838 cglobal ipred_smooth_h_8bpc, 3, 7, 0, dst, stride, tl, w, h
839 %define base r6-ipred_smooth_h_avx2_table
840 lea r6, [ipred_smooth_h_avx2_table]
841 mov wd, wm
842 vpbroadcastb m3, [tlq+wq] ; right
843 tzcnt wd, wd
844 mov hd, hm
845 movsxd wq, [r6+wq*4]
846 vpbroadcastd m4, [base+pb_127_m127]
847 vpbroadcastd m5, [base+pw_128]
848 add wq, r6
849 jmp wq
850 .w4:
851 WIN64_SPILL_XMM 8
852 vpbroadcastq m6, [base+smooth_weights+4*2]
853 mova m7, [base+ipred_h_shuf]
854 sub tlq, 8
855 sub tlq, hq
856 lea r3, [strideq*3]
857 .w4_loop:
858 vpbroadcastq m2, [tlq+hq]
859 pshufb m2, m7
860 punpcklbw m1, m2, m3 ; left, right
861 punpckhbw m2, m3
862 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
863 paddw m0, m1 ; 128 * left + 129 * right
864 pmaddubsw m1, m6
865 paddw m1, m5
866 paddw m0, m1
867 pmaddubsw m1, m2, m4
868 paddw m1, m2
869 pmaddubsw m2, m6
870 paddw m2, m5
871 paddw m1, m2
872 psrlw m0, 8
873 psrlw m1, 8
874 packuswb m0, m1
875 vextracti128 xm1, m0, 1
876 movd [dstq+strideq*0], xm0
877 movd [dstq+strideq*1], xm1
878 pextrd [dstq+strideq*2], xm0, 2
879 pextrd [dstq+r3 ], xm1, 2
880 cmp hd, 4
881 je .ret
882 lea dstq, [dstq+strideq*4]
883 pextrd [dstq+strideq*0], xm0, 1
884 pextrd [dstq+strideq*1], xm1, 1
885 pextrd [dstq+strideq*2], xm0, 3
886 pextrd [dstq+r3 ], xm1, 3
887 lea dstq, [dstq+strideq*4]
888 sub hd, 8
889 jg .w4_loop
890 .ret:
891 RET
892 ALIGN function_align
893 .w8:
894 %assign stack_offset stack_offset - stack_size_padded
895 WIN64_SPILL_XMM 8
896 vbroadcasti128 m6, [base+smooth_weights+8*2]
897 mova m7, [base+ipred_h_shuf]
898 sub tlq, 4
899 lea r3, [strideq*3]
900 sub tlq, hq
901 .w8_loop:
902 vpbroadcastd m2, [tlq+hq]
903 pshufb m2, m7
904 punpcklbw m1, m2, m3
905 punpckhbw m2, m3
906 pmaddubsw m0, m1, m4
907 paddw m0, m1
908 pmaddubsw m1, m6
909 paddw m1, m5
910 paddw m0, m1
911 pmaddubsw m1, m2, m4
912 paddw m1, m2
913 pmaddubsw m2, m6
914 paddw m2, m5
915 paddw m1, m2
916 psrlw m0, 8
917 psrlw m1, 8
918 packuswb m0, m1
919 vextracti128 xm1, m0, 1
920 movq [dstq+strideq*0], xm0
921 movq [dstq+strideq*1], xm1
922 movhps [dstq+strideq*2], xm0
923 movhps [dstq+r3 ], xm1
924 lea dstq, [dstq+strideq*4]
925 sub hd, 4
926 jg .w8_loop
927 RET
928 ALIGN function_align
929 .w16:
930 SETUP_STACK_FRAME 32*4, 7, 8
931 lea r3, [rsp+64*2-4]
932 call .prep ; only worthwhile for for w16 and above
933 sub tlq, 2
934 vpbroadcastd xm6, [base+pb_1]
935 mova xm7, [base+ipred_v_shuf+16]
936 vinserti128 m7, [base+ipred_v_shuf+ 0], 1
937 vbroadcasti128 m4, [base+smooth_weights+16*2]
938 vbroadcasti128 m5, [base+smooth_weights+16*3]
939 .w16_loop:
940 vpbroadcastd m1, [tlq+hq]
941 vpbroadcastd m2, [r3+hq*2]
942 pshufb m1, m6
943 punpcklbw m1, m3
944 pshufb m2, m7
945 SMOOTH 4, 5, 1, 1, 2, 2
946 mova [dstq+strideq*0], xm0
947 vextracti128 [dstq+strideq*1], m0, 1
948 lea dstq, [dstq+strideq*2]
949 sub hd, 2
950 jg .w16_loop
951 RET
952 ALIGN function_align
953 .w32:
954 SETUP_STACK_FRAME 32*4, 7, 6
955 lea r3, [rsp+64*2-2]
956 call .prep
957 dec tlq
958 mova xm4, [base+smooth_weights+16*4]
959 vinserti128 m4, [base+smooth_weights+16*6], 1
960 mova xm5, [base+smooth_weights+16*5]
961 vinserti128 m5, [base+smooth_weights+16*7], 1
962 .w32_loop:
963 vpbroadcastb m1, [tlq+hq]
964 punpcklbw m1, m3
965 vpbroadcastw m2, [r3+hq*2]
966 SMOOTH 4, 5, 1, 1, 2, 2
967 mova [dstq], m0
968 add dstq, strideq
969 dec hd
970 jg .w32_loop
971 RET
972 ALIGN function_align
973 .w64:
974 SETUP_STACK_FRAME 32*4, 7, 9
975 lea r3, [rsp+64*2-2]
976 call .prep
977 add r6, smooth_weights+16*15-ipred_smooth_h_avx2_table
978 dec tlq
979 mova xm5, [r6-16*7]
980 vinserti128 m5, [r6-16*5], 1
981 mova xm6, [r6-16*6]
982 vinserti128 m6, [r6-16*4], 1
983 mova xm7, [r6-16*3]
984 vinserti128 m7, [r6-16*1], 1
985 mova xm8, [r6-16*2]
986 vinserti128 m8, [r6-16*0], 1
987 .w64_loop:
988 vpbroadcastb m2, [tlq+hq]
989 punpcklbw m2, m3
990 vpbroadcastw m4, [r3+hq*2]
991 SMOOTH 5, 6, 2, 2, 4, 4
992 mova [dstq+32*0], m0
993 SMOOTH 7, 8, 2, 2, 4, 4
994 mova [dstq+32*1], m0
995 add dstq, strideq
996 dec hd
997 jg .w64_loop
998 RET
999 ALIGN function_align
1000 .prep:
1001 vpermq m2, [tlq-32*1], q3120
1002 punpckhbw m1, m2, m3
1003 punpcklbw m2, m3
1004 pmaddubsw m0, m1, m4 ; 127 * left - 127 * right
1005 paddw m1, m5 ; 1 * left + 256 * right + 128
1006 paddw m0, m1 ; 128 * left + 129 * right + 128
1007 pmaddubsw m1, m2, m4
1008 paddw m2, m5
1009 paddw m1, m2
1010 vpermq m2, [tlq-32*2], q3120
1011 mova [rsp+gprsize+32*3], m0
1012 mova [rsp+gprsize+32*2], m1
1013 punpckhbw m1, m2, m3
1014 punpcklbw m2, m3
1015 pmaddubsw m0, m1, m4
1016 paddw m1, m5
1017 paddw m0, m1
1018 pmaddubsw m1, m2, m4
1019 paddw m2, m5
1020 paddw m1, m2
1021 mova [rsp+gprsize+32*1], m0
1022 mova [rsp+gprsize+32*0], m1
1023 sub r3, hq
1024 sub tlq, hq
1025 sub r3, hq
1026 ret
1027
1028 %macro SMOOTH_2D_END 6 ; src[1-2], mul[1-2], add[1-2]
1029 pmaddubsw m0, m%3, m%1
1030 pmaddubsw m1, m%4, m%2
1031 %ifnum %5
1032 paddw m0, m%5
1033 %else
1034 paddw m0, %5
1035 %endif
1036 %ifnum %6
1037 paddw m1, m%6
1038 %else
1039 paddw m1, %6
1040 %endif
1041 pavgw m0, m2
1042 pavgw m1, m3
1043 psrlw m0, 8
1044 psrlw m1, 8
1045 packuswb m0, m1
1046 %endmacro
1047
1048 cglobal ipred_smooth_8bpc, 3, 7, 0, dst, stride, tl, w, h, v_weights
1049 %define base r6-ipred_smooth_avx2_table
1050 lea r6, [ipred_smooth_avx2_table]
1051 mov wd, wm
1052 vpbroadcastb m4, [tlq+wq] ; right
1053 tzcnt wd, wd
1054 mov hd, hm
1055 mov r5, tlq
1056 sub r5, hq
1057 movsxd wq, [r6+wq*4]
1058 vpbroadcastd m5, [base+pb_127_m127]
1059 vpbroadcastb m0, [r5] ; bottom
1060 vpbroadcastd m3, [base+pw_255]
1061 add wq, r6
1062 lea v_weightsq, [base+smooth_weights+hq*2]
1063 jmp wq
1064 .w4:
1065 WIN64_SPILL_XMM 12
1066 mova m10, [base+ipred_h_shuf]
1067 vpbroadcastq m11, [base+smooth_weights+4*2]
1068 mova m7, [base+ipred_v_shuf]
1069 vpbroadcastd m8, [tlq+1]
1070 sub tlq, 8
1071 lea r3, [strideq*3]
1072 sub tlq, hq
1073 punpcklbw m8, m0 ; top, bottom
1074 pshufd m6, m7, q2200
1075 pshufd m7, m7, q3311
1076 pmaddubsw m9, m8, m5
1077 paddw m3, m8 ; 1 * top + 255 * bottom + 255
1078 paddw m9, m3 ; 128 * top + 129 * bottom + 255
1079 .w4_loop:
1080 vpbroadcastq m1, [tlq+hq]
1081 pshufb m1, m10
1082 punpcklbw m0, m1, m4 ; left, right
1083 punpckhbw m1, m4
1084 pmaddubsw m2, m0, m5 ; 127 * left - 127 * right
1085 pmaddubsw m3, m1, m5
1086 paddw m2, m0 ; 128 * left + 129 * right
1087 paddw m3, m1
1088 pmaddubsw m0, m11
1089 pmaddubsw m1, m11
1090 paddw m2, m0
1091 paddw m3, m1
1092 vbroadcasti128 m1, [v_weightsq]
1093 add v_weightsq, 16
1094 pshufb m0, m1, m6
1095 pshufb m1, m7
1096 SMOOTH_2D_END 0, 1, 8, 8, 9, 9
1097 vextracti128 xm1, m0, 1
1098 movd [dstq+strideq*0], xm0
1099 movd [dstq+strideq*1], xm1
1100 pextrd [dstq+strideq*2], xm0, 2
1101 pextrd [dstq+r3 ], xm1, 2
1102 cmp hd, 4
1103 je .ret
1104 lea dstq, [dstq+strideq*4]
1105 pextrd [dstq+strideq*0], xm0, 1
1106 pextrd [dstq+strideq*1], xm1, 1
1107 pextrd [dstq+strideq*2], xm0, 3
1108 pextrd [dstq+r3 ], xm1, 3
1109 lea dstq, [dstq+strideq*4]
1110 sub hd, 8
1111 jg .w4_loop
1112 .ret:
1113 RET
1114 ALIGN function_align
1115 .w8:
1116 %assign stack_offset stack_offset - stack_size_padded
1117 WIN64_SPILL_XMM 12
1118 mova m10, [base+ipred_h_shuf]
1119 vbroadcasti128 m11, [base+smooth_weights+8*2]
1120 mova m7, [base+ipred_v_shuf]
1121 vpbroadcastq m8, [tlq+1]
1122 sub tlq, 4
1123 lea r3, [strideq*3]
1124 sub tlq, hq
1125 punpcklbw m8, m0
1126 pshufd m6, m7, q0000
1127 pshufd m7, m7, q1111
1128 pmaddubsw m9, m8, m5
1129 paddw m3, m8
1130 paddw m9, m3
1131 .w8_loop:
1132 vpbroadcastd m1, [tlq+hq]
1133 pshufb m1, m10
1134 punpcklbw m0, m1, m4
1135 punpckhbw m1, m4
1136 pmaddubsw m2, m0, m5
1137 pmaddubsw m3, m1, m5
1138 paddw m2, m0
1139 paddw m3, m1
1140 pmaddubsw m0, m11
1141 pmaddubsw m1, m11
1142 paddw m2, m0
1143 paddw m3, m1
1144 vpbroadcastq m1, [v_weightsq]
1145 add v_weightsq, 8
1146 pshufb m0, m1, m6
1147 pshufb m1, m7
1148 SMOOTH_2D_END 0, 1, 8, 8, 9, 9
1149 vextracti128 xm1, m0, 1
1150 movq [dstq+strideq*0], xm0
1151 movq [dstq+strideq*1], xm1
1152 movhps [dstq+strideq*2], xm0
1153 movhps [dstq+r3 ], xm1
1154 lea dstq, [dstq+strideq*4]
1155 sub hd, 4
1156 jg .w8_loop
1157 RET
1158 ALIGN function_align
1159 .w16:
1160 SETUP_STACK_FRAME 32*4, 7, 14
1161 vbroadcasti128 m11, [tlq+1]
1162 lea r3, [rsp+64*2-4]
1163 punpcklbw m10, m11, m0 ; top, bottom
1164 punpckhbw m11, m0
1165 call .prep_v
1166 sub tlq, 2
1167 pmaddubsw m12, m10, m5
1168 pmaddubsw m13, m11, m5
1169 vpbroadcastd xm5, [base+pb_1]
1170 mova m9, [base+ipred_v_shuf]
1171 vbroadcasti128 m6, [base+smooth_weights+16*2]
1172 vbroadcasti128 m7, [base+smooth_weights+16*3]
1173 vperm2i128 m8, m9, m9, 0x01
1174 paddw m0, m10, m3
1175 paddw m3, m11
1176 paddw m12, m0
1177 paddw m13, m3
1178 .w16_loop:
1179 vpbroadcastd m3, [tlq+hq]
1180 vpbroadcastd m0, [r3+hq*2]
1181 vpbroadcastd m1, [v_weightsq]
1182 add v_weightsq, 4
1183 pshufb m3, m5
1184 punpcklbw m3, m4 ; left, right
1185 pmaddubsw m2, m3, m6
1186 pmaddubsw m3, m7
1187 pshufb m0, m8
1188 pshufb m1, m9
1189 paddw m2, m0
1190 paddw m3, m0
1191 SMOOTH_2D_END 1, 1, 10, 11, 12, 13
1192 mova [dstq+strideq*0], xm0
1193 vextracti128 [dstq+strideq*1], m0, 1
1194 lea dstq, [dstq+strideq*2]
1195 sub hd, 2
1196 jg .w16_loop
1197 RET
1198 ALIGN function_align
1199 .w32:
1200 SETUP_STACK_FRAME 32*4, 7, 11
1201 movu m8, [tlq+1]
1202 lea r3, [rsp+64*2-2]
1203 punpcklbw m7, m8, m0
1204 punpckhbw m8, m0
1205 call .prep_v
1206 dec tlq
1207 pmaddubsw m9, m7, m5
1208 pmaddubsw m10, m8, m5
1209 mova xm5, [base+smooth_weights+16*4]
1210 vinserti128 m5, [base+smooth_weights+16*6], 1
1211 mova xm6, [base+smooth_weights+16*5]
1212 vinserti128 m6, [base+smooth_weights+16*7], 1
1213 paddw m0, m7, m3
1214 paddw m3, m8
1215 paddw m9, m0
1216 paddw m10, m3
1217 .w32_loop:
1218 vpbroadcastb m3, [tlq+hq]
1219 punpcklbw m3, m4
1220 vpbroadcastw m0, [r3+hq*2]
1221 vpbroadcastw m1, [v_weightsq]
1222 add v_weightsq, 2
1223 pmaddubsw m2, m3, m5
1224 pmaddubsw m3, m6
1225 paddw m2, m0
1226 paddw m3, m0
1227 SMOOTH_2D_END 1, 1, 7, 8, 9, 10
1228 mova [dstq], m0
1229 add dstq, strideq
1230 dec hd
1231 jg .w32_loop
1232 RET
1233 ALIGN function_align
1234 .w64:
1235 SETUP_STACK_FRAME 32*8, 7, 16
1236 movu m13, [tlq+1 ]
1237 movu m15, [tlq+33]
1238 add r6, smooth_weights+16*15-ipred_smooth_avx2_table
1239 lea r3, [rsp+64*2-2]
1240 punpcklbw m12, m13, m0
1241 punpckhbw m13, m0
1242 punpcklbw m14, m15, m0
1243 punpckhbw m15, m0
1244 call .prep_v
1245 dec tlq
1246 pmaddubsw m0, m12, m5
1247 pmaddubsw m1, m13, m5
1248 pmaddubsw m2, m14, m5
1249 pmaddubsw m5, m15, m5
1250 mova xm8, [r6-16*7]
1251 vinserti128 m8, [r6-16*5], 1
1252 mova xm9, [r6-16*6]
1253 vinserti128 m9, [r6-16*4], 1
1254 mova xm10, [r6-16*3]
1255 vinserti128 m10, [r6-16*1], 1
1256 mova xm11, [r6-16*2]
1257 vinserti128 m11, [r6-16*0], 1
1258 lea r6, [rsp+32*4]
1259 paddw m0, m3
1260 paddw m1, m3
1261 paddw m2, m3
1262 paddw m3, m5
1263 paddw m0, m12
1264 paddw m1, m13
1265 paddw m2, m14
1266 paddw m3, m15
1267 mova [r6+32*0], m0
1268 mova [r6+32*1], m1
1269 mova [r6+32*2], m2
1270 mova [r6+32*3], m3
1271 .w64_loop:
1272 vpbroadcastb m5, [tlq+hq]
1273 punpcklbw m5, m4
1274 vpbroadcastw m6, [r3+hq*2]
1275 vpbroadcastw m7, [v_weightsq]
1276 add v_weightsq, 2
1277 pmaddubsw m2, m5, m8
1278 pmaddubsw m3, m5, m9
1279 paddw m2, m6
1280 paddw m3, m6
1281 SMOOTH_2D_END 7, 7, 12, 13, [r6+32*0], [r6+32*1]
1282 mova [dstq+32*0], m0
1283 pmaddubsw m2, m5, m10
1284 pmaddubsw m3, m5, m11
1285 paddw m2, m6
1286 paddw m3, m6
1287 SMOOTH_2D_END 7, 7, 14, 15, [r6+32*2], [r6+32*3]
1288 mova [dstq+32*1], m0
1289 add dstq, strideq
1290 dec hd
1291 jg .w64_loop
1292 RET
1293 ALIGN function_align
1294 .prep_v:
1295 vpermq m2, [tlq-32*1], q3120
1296 punpckhbw m1, m2, m4
1297 punpcklbw m2, m4
1298 pmaddubsw m0, m1, m5 ; 127 * left - 127 * right
1299 paddw m0, m1 ; 128 * left + 129 * right
1300 pmaddubsw m1, m2, m5
1301 paddw m1, m2
1302 vpermq m2, [tlq-32*2], q3120
1303 mova [rsp+gprsize+32*3], m0
1304 mova [rsp+gprsize+32*2], m1
1305 punpckhbw m1, m2, m4
1306 punpcklbw m2, m4
1307 pmaddubsw m0, m1, m5
1308 paddw m0, m1
1309 pmaddubsw m1, m2, m5
1310 paddw m1, m2
1311 mova [rsp+gprsize+32*1], m0
1312 mova [rsp+gprsize+32*0], m1
1313 sub r3, hq
1314 sub tlq, hq
1315 sub r3, hq
1316 ret
1317
1318 cglobal ipred_z1_8bpc, 3, 8, 0, dst, stride, tl, w, h, angle, dx, maxbase
1319 %assign org_stack_offset stack_offset
1320 lea r6, [ipred_z1_avx2_table]
1321 tzcnt wd, wm
1322 movifnidn angled, anglem
1323 movifnidn hd, hm
1324 lea r7, [dr_intra_derivative]
1325 inc tlq
1326 movsxd wq, [r6+wq*4]
1327 add wq, r6
1328 mov dxd, angled
1329 and dxd, 0x7e
1330 add angled, 165 ; ~90
1331 movzx dxd, word [r7+dxq]
1332 xor angled, 0x4ff ; d = 90 - angle
1333 vpbroadcastd m3, [pw_512]
1334 vpbroadcastd m4, [pw_62]
1335 vpbroadcastd m5, [pw_64]
1336 jmp wq
1337 .w4:
1338 cmp angleb, 40
1339 jae .w4_no_upsample
1340 lea r3d, [angleq-1024]
1341 sar r3d, 7
1342 add r3d, hd
1343 jg .w4_no_upsample ; !enable_intra_edge_filter || h > 8 || (h == 8 && is_sm)
1344 ALLOC_STACK -32, 8
1345 mova xm1, [tlq-1]
1346 pshufb xm0, xm1, [z_upsample1]
1347 pshufb xm1, [z_upsample2]
1348 vpbroadcastd xm2, [pb_36_m4] ; upshifted by 2 to be able to reuse
1349 add dxd, dxd ; pw_512 (which is already in m3)
1350 pmaddubsw xm0, xm2 ; for rounding instead of pw_2048
1351 pextrd [rsp+16], xm1, 3 ; top[max_base_x]
1352 pmaddubsw xm1, xm2
1353 movd xm7, dxd
1354 mov r3d, dxd ; xpos
1355 vpbroadcastw m7, xm7
1356 paddw xm1, xm0
1357 movq xm0, [tlq]
1358 pmulhrsw xm1, xm3
1359 pslldq m6, m7, 8
1360 paddw xm2, xm7, xm7
1361 lea r2, [strideq*3]
1362 paddw m6, m7
1363 packuswb xm1, xm1
1364 paddw m6, m2 ; xpos2 xpos3 xpos0 xpos1
1365 punpcklbw xm0, xm1
1366 psllw m7, 2
1367 mova [rsp], xm0
1368 .w4_upsample_loop:
1369 lea r5d, [r3+dxq]
1370 shr r3d, 6 ; base0
1371 vpbroadcastq m1, [rsp+r3]
1372 lea r3d, [r5+dxq]
1373 shr r5d, 6 ; base1
1374 vpbroadcastq m2, [rsp+r5]
1375 lea r5d, [r3+dxq]
1376 shr r3d, 6 ; base2
1377 movq xm0, [rsp+r3]
1378 lea r3d, [r5+dxq]
1379 shr r5d, 6 ; base3
1380 movhps xm0, [rsp+r5]
1381 vpblendd m1, m2, 0xc0
1382 pand m2, m4, m6 ; frac
1383 vpblendd m0, m1, 0xf0
1384 psubw m1, m5, m2 ; 64-frac
1385 psllw m2, 8
1386 por m1, m2 ; 64-frac, frac
1387 pmaddubsw m0, m1
1388 paddw m6, m7 ; xpos += dx
1389 pmulhrsw m0, m3
1390 packuswb m0, m0
1391 vextracti128 xm1, m0, 1
1392 movd [dstq+strideq*2], xm0
1393 pextrd [dstq+r2 ], xm0, 1
1394 movd [dstq+strideq*0], xm1
1395 pextrd [dstq+strideq*1], xm1, 1
1396 lea dstq, [dstq+strideq*4]
1397 sub hd, 4
1398 jg .w4_upsample_loop
1399 RET
1400 ALIGN function_align
1401 .filter_strength: ; w4/w8/w16
1402 ; The C version uses a lot of branches, but we can do all the comparisons
1403 ; in parallel and use popcnt to get the final filter strength value.
1404 %define base r3-z_filter_t0
1405 lea r3, [z_filter_t0]
1406 movd xm0, maxbased
1407 movd xm2, angled
1408 shr angled, 8 ; is_sm << 1
1409 vpbroadcastb m0, xm0
1410 vpbroadcastb m2, xm2
1411 pcmpeqb m1, m0, [base+z_filter_wh]
1412 pand m1, m2
1413 mova xm2, [r3+angleq*8] ; upper ymm half zero in both cases
1414 pcmpgtb m1, m2
1415 pmovmskb r5d, m1
1416 ret
1417 .w4_no_upsample:
1418 %assign stack_offset org_stack_offset
1419 ALLOC_STACK -16, 11
1420 mov maxbased, 7
1421 test angled, 0x400 ; !enable_intra_edge_filter
1422 jnz .w4_main
1423 lea maxbased, [hq+3]
1424 call .filter_strength
1425 mov maxbased, 7
1426 test r5d, r5d
1427 jz .w4_main ; filter_strength == 0
1428 popcnt r5d, r5d
1429 vpbroadcastd m7, [base+pb_8]
1430 vbroadcasti128 m2, [tlq-1]
1431 pminub m1, m7, [base+z_filter_s]
1432 vpbroadcastd m8, [base+z_filter_k-4+r5*4+12*0]
1433 pminub m7, [base+z_filter_s+8]
1434 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*1]
1435 vpbroadcastd m10, [base+z_filter_k-4+r5*4+12*2]
1436 pshufb m0, m2, m1
1437 shufps m1, m7, q2121
1438 pmaddubsw m0, m8
1439 pshufb m1, m2, m1
1440 pmaddubsw m1, m9
1441 pshufb m2, m7
1442 pmaddubsw m2, m10
1443 paddw m0, m1
1444 paddw m0, m2
1445 pmulhrsw m0, m3
1446 mov r3d, 9
1447 mov tlq, rsp
1448 cmp hd, 4
1449 cmovne maxbased, r3d
1450 vextracti128 xm1, m0, 1
1451 packuswb xm0, xm1
1452 mova [tlq], xm0
1453 .w4_main:
1454 movd xm6, dxd
1455 vpbroadcastq m0, [z_base_inc] ; base_inc << 6
1456 vpbroadcastb m7, [tlq+maxbaseq]
1457 shl maxbased, 6
1458 vpbroadcastw m6, xm6
1459 mov r3d, dxd ; xpos
1460 movd xm9, maxbased
1461 vpbroadcastw m9, xm9
1462 vbroadcasti128 m8, [z1_shuf_w4]
1463 psrlw m7, 8 ; top[max_base_x]
1464 paddw m10, m6, m6
1465 psubw m9, m0 ; max_base_x
1466 vpblendd m6, m10, 0xcc
1467 mova xm0, xm10
1468 paddw m6, m0 ; xpos2 xpos3 xpos0 xpos1
1469 paddw m10, m10
1470 .w4_loop:
1471 lea r5d, [r3+dxq]
1472 shr r3d, 6 ; base0
1473 vpbroadcastq m1, [tlq+r3]
1474 lea r3d, [r5+dxq]
1475 shr r5d, 6 ; base1
1476 vpbroadcastq m2, [tlq+r5]
1477 lea r5d, [r3+dxq]
1478 shr r3d, 6 ; base2
1479 movq xm0, [tlq+r3]
1480 lea r3d, [r5+dxq]
1481 shr r5d, 6 ; base3
1482 movhps xm0, [tlq+r5]
1483 vpblendd m1, m2, 0xc0
1484 pand m2, m4, m6 ; frac
1485 vpblendd m0, m1, 0xf0
1486 psubw m1, m5, m2 ; 64-frac
1487 psllw m2, 8
1488 pshufb m0, m8
1489 por m1, m2 ; 64-frac, frac
1490 pmaddubsw m0, m1
1491 pcmpgtw m1, m9, m6 ; base < max_base_x
1492 pmulhrsw m0, m3
1493 paddw m6, m10 ; xpos += dx
1494 lea r5, [dstq+strideq*2]
1495 vpblendvb m0, m7, m0, m1
1496 packuswb m0, m0
1497 vextracti128 xm1, m0, 1
1498 movd [r5 +strideq*0], xm0
1499 pextrd [r5 +strideq*1], xm0, 1
1500 movd [dstq+strideq*0], xm1
1501 pextrd [dstq+strideq*1], xm1, 1
1502 sub hd, 4
1503 jz .w4_end
1504 lea dstq, [dstq+strideq*4]
1505 cmp r3d, maxbased
1506 jb .w4_loop
1507 packuswb xm7, xm7
1508 lea r6, [strideq*3]
1509 .w4_end_loop:
1510 movd [dstq+strideq*0], xm7
1511 movd [dstq+strideq*1], xm7
1512 movd [dstq+strideq*2], xm7
1513 movd [dstq+r6 ], xm7
1514 lea dstq, [dstq+strideq*4]
1515 sub hd, 4
1516 jg .w4_end_loop
1517 .w4_end:
1518 RET
1519 ALIGN function_align
1520 .w8:
1521 lea r3d, [angleq+216]
1522 mov r3b, hb
1523 cmp r3d, 8
1524 ja .w8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || h > 8
1525 %assign stack_offset org_stack_offset
1526 ALLOC_STACK -32, 8
1527 movu xm2, [z_filter_s+6]
1528 mova xm0, [tlq-1]
1529 movd xm6, hd
1530 vinserti128 m0, [tlq+7], 1
1531 vpbroadcastb xm6, xm6
1532 vbroadcasti128 m1, [z_upsample1]
1533 pminub xm6, xm2
1534 vpbroadcastd m7, [pb_36_m4]
1535 vinserti128 m2, xm6, 1
1536 add dxd, dxd
1537 pshufb m1, m0, m1
1538 pshufb m2, m0, m2
1539 movd xm6, dxd
1540 pmaddubsw m1, m7
1541 pmaddubsw m2, m7
1542 vpbroadcastw m6, xm6
1543 mov r3d, dxd
1544 psrldq m0, 1
1545 lea r2, [strideq*3]
1546 paddw m7, m6, m6
1547 paddw m1, m2
1548 vpblendd m6, m7, 0xf0
1549 pmulhrsw m1, m3
1550 pslldq m2, m7, 8
1551 paddw m7, m7
1552 paddw m6, m2
1553 packuswb m1, m1
1554 punpcklbw m0, m1
1555 mova [rsp], m0
1556 .w8_upsample_loop:
1557 lea r5d, [r3+dxq]
1558 shr r3d, 6 ; base0
1559 movu xm0, [rsp+r3]
1560 lea r3d, [r5+dxq]
1561 shr r5d, 6 ; base1
1562 vinserti128 m0, [rsp+r5], 1
1563 lea r5d, [r3+dxq]
1564 shr r3d, 6 ; base2
1565 pand m1, m4, m6
1566 psubw m2, m5, m1
1567 psllw m1, 8
1568 por m2, m1
1569 punpcklqdq m1, m2, m2 ; frac0 frac1
1570 pmaddubsw m0, m1
1571 movu xm1, [rsp+r3]
1572 lea r3d, [r5+dxq]
1573 shr r5d, 6 ; base3
1574 vinserti128 m1, [rsp+r5], 1
1575 punpckhqdq m2, m2 ; frac2 frac3
1576 pmaddubsw m1, m2
1577 pmulhrsw m0, m3
1578 paddw m6, m7
1579 pmulhrsw m1, m3
1580 packuswb m0, m1
1581 vextracti128 xm1, m0, 1
1582 movq [dstq+strideq*0], xm0
1583 movhps [dstq+strideq*2], xm0
1584 movq [dstq+strideq*1], xm1
1585 movhps [dstq+r2 ], xm1
1586 lea dstq, [dstq+strideq*4]
1587 sub hd, 4
1588 jg .w8_upsample_loop
1589 RET
1590 .w8_no_intra_edge_filter:
1591 and maxbased, 7
1592 or maxbased, 8 ; imin(h+7, 15)
1593 jmp .w8_main
1594 .w8_no_upsample:
1595 %assign stack_offset org_stack_offset
1596 ALLOC_STACK -32, 10
1597 lea maxbased, [hq+7]
1598 test angled, 0x400
1599 jnz .w8_no_intra_edge_filter
1600 call .filter_strength
1601 test r5d, r5d
1602 jz .w8_main ; filter_strength == 0
1603 popcnt r5d, r5d
1604 movu xm2, [tlq]
1605 pminub xm1, xm0, [base+z_filter_s+14]
1606 vinserti128 m2, [tlq-1], 1
1607 vinserti128 m1, [base+z_filter_s+ 0], 1
1608 vpbroadcastd m7, [base+z_filter_k-4+r5*4+12*0]
1609 pminub xm0, [base+z_filter_s+22]
1610 vinserti128 m0, [base+z_filter_s+ 8], 1
1611 pshufb m6, m2, m1
1612 pmaddubsw m6, m7
1613 vpbroadcastd m7, [base+z_filter_k-4+r5*4+12*1]
1614 movzx r3d, byte [tlq+15]
1615 shufps m1, m0, q2121
1616 pshufb m1, m2, m1
1617 pmaddubsw m1, m7
1618 paddw m1, m6
1619 sub r5d, 3
1620 jnz .w8_3tap
1621 ; filter_strength == 3 uses a 5-tap filter instead of a 3-tap one,
1622 ; which also results in an awkward edge case where out[w*2] is
1623 ; slightly different from out[max_base_x] when h > w.
1624 vpbroadcastd m7, [z_filter_k+4*8]
1625 movzx r2d, byte [tlq+14]
1626 pshufb m2, m0
1627 pmaddubsw m2, m7
1628 sub r2d, r3d
1629 lea r2d, [r2+r3*8+4]
1630 shr r2d, 3 ; (tlq[w*2-2] + tlq[w*2-1]*7 + 4) >> 3
1631 mov [rsp+16], r2b
1632 paddw m1, m2
1633 .w8_3tap:
1634 pmulhrsw m1, m3
1635 sar r5d, 1
1636 mov tlq, rsp
1637 add r5d, 17 ; w*2 + (filter_strength == 3)
1638 cmp hd, 16
1639 cmovns maxbased, r5d
1640 mov [tlq+r5], r3b
1641 vextracti128 xm0, m1, 1
1642 packuswb xm0, xm1
1643 mova [tlq], xm0
1644 .w8_main:
1645 movd xm2, dxd
1646 vbroadcasti128 m0, [z_base_inc]
1647 vpbroadcastw m2, xm2
1648 vpbroadcastb m7, [tlq+maxbaseq]
1649 shl maxbased, 6
1650 movd xm9, maxbased
1651 vbroadcasti128 m8, [z_filter_s+2]
1652 vpbroadcastw m9, xm9
1653 psrlw m7, 8
1654 psubw m9, m0
1655 mov r3d, dxd
1656 paddw m6, m2, m2
1657 vpblendd m2, m6, 0xf0
1658 .w8_loop:
1659 lea r5d, [r3+dxq]
1660 shr r3d, 6
1661 pand m0, m4, m2
1662 psubw m1, m5, m0
1663 psllw m0, 8
1664 por m1, m0
1665 movu xm0, [tlq+r3]
1666 lea r3d, [r5+dxq]
1667 shr r5d, 6 ; base1
1668 vinserti128 m0, [tlq+r5], 1
1669 pshufb m0, m8
1670 pmaddubsw m0, m1
1671 pcmpgtw m1, m9, m2
1672 paddw m2, m6
1673 pmulhrsw m0, m3
1674 vpblendvb m0, m7, m0, m1
1675 vextracti128 xm1, m0, 1
1676 packuswb xm0, xm1
1677 movq [dstq+strideq*0], xm0
1678 movhps [dstq+strideq*1], xm0
1679 sub hd, 2
1680 jz .w8_end
1681 lea dstq, [dstq+strideq*2]
1682 cmp r3d, maxbased
1683 jb .w8_loop
1684 packuswb xm7, xm7
1685 .w8_end_loop:
1686 movq [dstq+strideq*0], xm7
1687 movq [dstq+strideq*1], xm7
1688 lea dstq, [dstq+strideq*2]
1689 sub hd, 2
1690 jg .w8_end_loop
1691 .w8_end:
1692 RET
1693 .w16_no_intra_edge_filter:
1694 and maxbased, 15
1695 or maxbased, 16 ; imin(h+15, 31)
1696 jmp .w16_main
1697 ALIGN function_align
1698 .w16:
1699 %assign stack_offset org_stack_offset
1700 ALLOC_STACK -64, 12
1701 lea maxbased, [hq+15]
1702 test angled, 0x400
1703 jnz .w16_no_intra_edge_filter
1704 call .filter_strength
1705 test r5d, r5d
1706 jz .w16_main ; filter_strength == 0
1707 popcnt r5d, r5d
1708 vpbroadcastd m1, [base+pb_12]
1709 vbroadcasti128 m6, [base+z_filter_s+8]
1710 vinserti128 m2, m6, [base+z_filter_s], 0
1711 vinserti128 m6, [base+z_filter_s+16], 1
1712 mova xm10, [tlq-1]
1713 vinserti128 m10, [tlq+3], 1
1714 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*0]
1715 vbroadcasti128 m7, [base+z_filter_s+14]
1716 vinserti128 m8, m7, [base+z_filter_s+6], 0
1717 vinserti128 m7, [base+z_filter_s+22], 1
1718 psubw m0, m1
1719 movu xm11, [tlq+12]
1720 vinserti128 m11, [tlq+16], 1
1721 pminub m8, m0
1722 pminub m7, m0
1723 pshufb m0, m10, m2
1724 shufps m2, m6, q2121
1725 pmaddubsw m0, m9
1726 pshufb m1, m11, m8
1727 shufps m8, m7, q2121
1728 pmaddubsw m1, m9
1729 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*1]
1730 movzx r3d, byte [tlq+31]
1731 pshufb m2, m10, m2
1732 pmaddubsw m2, m9
1733 pshufb m8, m11, m8
1734 pmaddubsw m8, m9
1735 paddw m0, m2
1736 paddw m1, m8
1737 sub r5d, 3
1738 jnz .w16_3tap
1739 vpbroadcastd m9, [z_filter_k+4*8]
1740 movzx r2d, byte [tlq+30]
1741 pshufb m10, m6
1742 pmaddubsw m10, m9
1743 pshufb m11, m7
1744 pmaddubsw m11, m9
1745 sub r2d, r3d
1746 lea r2d, [r2+r3*8+4]
1747 shr r2d, 3
1748 mov [rsp+32], r2b
1749 paddw m0, m10
1750 paddw m1, m11
1751 .w16_3tap:
1752 pmulhrsw m0, m3
1753 pmulhrsw m1, m3
1754 sar r5d, 1
1755 mov tlq, rsp
1756 add r5d, 33
1757 cmp hd, 32
1758 cmovns maxbased, r5d
1759 mov [tlq+r5], r3b
1760 packuswb m0, m1
1761 vpermq m0, m0, q3120
1762 mova [tlq], m0
1763 .w16_main:
1764 movd xm6, dxd
1765 vbroadcasti128 m0, [z_base_inc]
1766 vpbroadcastb m7, [tlq+maxbaseq]
1767 shl maxbased, 6
1768 vpbroadcastw m6, xm6
1769 movd xm9, maxbased
1770 vbroadcasti128 m8, [z_filter_s+2]
1771 vpbroadcastw m9, xm9
1772 mov r3d, dxd
1773 psubw m9, m0
1774 paddw m11, m6, m6
1775 psubw m10, m9, m3 ; 64*8
1776 vpblendd m6, m11, 0xf0
1777 .w16_loop:
1778 lea r5d, [r3+dxq]
1779 shr r3d, 6 ; base0
1780 pand m1, m4, m6
1781 psubw m2, m5, m1
1782 psllw m1, 8
1783 por m2, m1
1784 movu xm0, [tlq+r3+0]
1785 movu xm1, [tlq+r3+8]
1786 lea r3d, [r5+dxq]
1787 shr r5d, 6 ; base1
1788 vinserti128 m0, [tlq+r5+0], 1
1789 vinserti128 m1, [tlq+r5+8], 1
1790 pshufb m0, m8
1791 pshufb m1, m8
1792 pmaddubsw m0, m2
1793 pmaddubsw m1, m2
1794 pmulhrsw m0, m3
1795 pmulhrsw m1, m3
1796 packuswb m0, m1
1797 pcmpgtw m1, m9, m6
1798 pcmpgtw m2, m10, m6
1799 packsswb m1, m2
1800 paddw m6, m11
1801 vpblendvb m0, m7, m0, m1
1802 mova [dstq+strideq*0], xm0
1803 vextracti128 [dstq+strideq*1], m0, 1
1804 sub hd, 2
1805 jz .w16_end
1806 lea dstq, [dstq+strideq*2]
1807 cmp r3d, maxbased
1808 jb .w16_loop
1809 .w16_end_loop:
1810 mova [dstq+strideq*0], xm7
1811 mova [dstq+strideq*1], xm7
1812 lea dstq, [dstq+strideq*2]
1813 sub hd, 2
1814 jg .w16_end_loop
1815 .w16_end:
1816 RET
1817 ALIGN function_align
1818 .w32:
1819 %assign stack_offset org_stack_offset
1820 ALLOC_STACK -96, 15
1821 lea r3d, [hq+31]
1822 mov maxbased, 63
1823 cmp hd, 32
1824 cmovs maxbased, r3d
1825 test angled, 0x400 ; !enable_intra_edge_filter
1826 jnz .w32_main
1827 vbroadcasti128 m0, [pb_0to15]
1828 sub r3d, 29 ; h+2
1829 movu xm13, [tlq+29] ; 32-39
1830 movd xm1, r3d
1831 movu xm14, [tlq+37] ; 40-47
1832 sub r3d, 8 ; h-6
1833 vinserti128 m14, [tlq+51], 1 ; 56-63
1834 vpbroadcastb xm1, xm1
1835 mova xm11, [tlq- 1] ; 0- 7
1836 vinserti128 m11, [tlq+13], 1 ; 16-23
1837 movd xm2, r3d
1838 movu xm12, [tlq+ 5] ; 8-15
1839 vinserti128 m12, [tlq+19], 1 ; 24-31
1840 pminub xm1, xm0 ; clip 32x8
1841 mova m7, [z_filter_s+0]
1842 pshufb xm13, xm1
1843 vpbroadcastd m1, [pb_12]
1844 vpbroadcastb xm2, xm2
1845 vinserti128 m13, [tlq+43], 1 ; 48-55
1846 vinserti128 m8, m7, [z_filter_s+4], 1
1847 vpblendd m2, m1, 0xf0
1848 vinserti128 m7, [z_filter_s+12], 0
1849 pminub m2, m0 ; clip 32x16 and 32x(32|64)
1850 vpbroadcastd m9, [z_filter_k+4*2+12*0]
1851 pshufb m14, m2
1852 pshufb m0, m11, m8
1853 shufps m8, m7, q1021
1854 pmaddubsw m0, m9
1855 pshufb m2, m12, m8
1856 pmaddubsw m2, m9
1857 pshufb m1, m13, m8
1858 pmaddubsw m1, m9
1859 pshufb m6, m14, m8
1860 pmaddubsw m6, m9
1861 vpbroadcastd m9, [z_filter_k+4*2+12*1]
1862 pshufb m10, m11, m8
1863 shufps m8, m7, q2121
1864 pmaddubsw m10, m9
1865 paddw m0, m10
1866 pshufb m10, m12, m8
1867 pmaddubsw m10, m9
1868 paddw m2, m10
1869 pshufb m10, m13, m8
1870 pmaddubsw m10, m9
1871 paddw m1, m10
1872 pshufb m10, m14, m8
1873 pmaddubsw m10, m9
1874 paddw m6, m10
1875 vpbroadcastd m9, [z_filter_k+4*2+12*2]
1876 pshufb m11, m8
1877 pmaddubsw m11, m9
1878 pshufb m12, m7
1879 pmaddubsw m12, m9
1880 movzx r3d, byte [tlq+63]
1881 movzx r2d, byte [tlq+62]
1882 paddw m0, m11
1883 paddw m2, m12
1884 pshufb m13, m7
1885 pmaddubsw m13, m9
1886 pshufb m14, m7
1887 pmaddubsw m14, m9
1888 paddw m1, m13
1889 paddw m6, m14
1890 sub r2d, r3d
1891 lea r2d, [r2+r3*8+4] ; edge case for 32x64
1892 pmulhrsw m0, m3
1893 pmulhrsw m2, m3
1894 pmulhrsw m1, m3
1895 pmulhrsw m6, m3
1896 shr r2d, 3
1897 mov [rsp+64], r2b
1898 mov tlq, rsp
1899 mov [tlq+65], r3b
1900 mov r3d, 65
1901 cmp hd, 64
1902 cmove maxbased, r3d
1903 packuswb m0, m2
1904 packuswb m1, m6
1905 mova [tlq+ 0], m0
1906 mova [tlq+32], m1
1907 .w32_main:
1908 movd xm6, dxd
1909 vpbroadcastb m7, [tlq+maxbaseq]
1910 shl maxbased, 6
1911 vpbroadcastw m6, xm6
1912 movd xm9, maxbased
1913 vbroadcasti128 m8, [z_filter_s+2]
1914 vpbroadcastw m9, xm9
1915 mov r5d, dxd
1916 psubw m9, [z_base_inc]
1917 mova m11, m6
1918 psubw m10, m9, m3 ; 64*8
1919 .w32_loop:
1920 mov r3d, r5d
1921 shr r3d, 6
1922 pand m1, m4, m6
1923 psubw m2, m5, m1
1924 psllw m1, 8
1925 por m2, m1
1926 movu m0, [tlq+r3+0]
1927 movu m1, [tlq+r3+8]
1928 add r5d, dxd
1929 pshufb m0, m8
1930 pshufb m1, m8
1931 pmaddubsw m0, m2
1932 pmaddubsw m1, m2
1933 pmulhrsw m0, m3
1934 pmulhrsw m1, m3
1935 packuswb m0, m1
1936 pcmpgtw m1, m9, m6
1937 pcmpgtw m2, m10, m6
1938 packsswb m1, m2
1939 paddw m6, m11
1940 vpblendvb m0, m7, m0, m1
1941 mova [dstq], m0
1942 dec hd
1943 jz .w32_end
1944 add dstq, strideq
1945 cmp r5d, maxbased
1946 jb .w32_loop
1947 test hb, 1
1948 jz .w32_end_loop
1949 mova [dstq], m7
1950 add dstq, strideq
1951 dec hd
1952 jz .w32_end
1953 .w32_end_loop:
1954 mova [dstq+strideq*0], m7
1955 mova [dstq+strideq*1], m7
1956 lea dstq, [dstq+strideq*2]
1957 sub hd, 2
1958 jg .w32_end_loop
1959 .w32_end:
1960 RET
1961 ALIGN function_align
1962 .w64:
1963 %assign stack_offset org_stack_offset
1964 ALLOC_STACK -128, 16
1965 lea maxbased, [hq+63]
1966 test angled, 0x400 ; !enable_intra_edge_filter
1967 jnz .w64_main
1968 mova xm11, [tlq- 1] ; 0- 7
1969 vinserti128 m11, [tlq+13], 1 ; 16-23
1970 movu xm12, [tlq+ 5] ; 8-15
1971 vinserti128 m12, [tlq+19], 1 ; 24-31
1972 mova m7, [z_filter_s+0]
1973 vinserti128 m8, m7, [z_filter_s+4], 1
1974 vinserti128 m7, [z_filter_s+12], 0
1975 vpbroadcastd m9, [z_filter_k+4*2+12*0]
1976 movu xm13, [tlq+29] ; 32-39
1977 vinserti128 m13, [tlq+43], 1 ; 48-55
1978 movu xm14, [tlq+37] ; 40-47
1979 vinserti128 m14, [tlq+51], 1 ; 56-63
1980 pshufb m0, m11, m8
1981 shufps m8, m7, q1021
1982 pmaddubsw m0, m9
1983 pshufb m2, m12, m8
1984 pmaddubsw m2, m9
1985 pshufb m1, m13, m8
1986 pmaddubsw m1, m9
1987 pshufb m6, m14, m8
1988 pmaddubsw m6, m9
1989 vpbroadcastd m9, [z_filter_k+4*2+12*1]
1990 pshufb m10, m11, m8
1991 shufps m15, m8, m7, q2121
1992 pmaddubsw m10, m9
1993 paddw m0, m10
1994 pshufb m10, m12, m15
1995 pmaddubsw m10, m9
1996 paddw m2, m10
1997 pshufb m10, m13, m15
1998 pmaddubsw m10, m9
1999 paddw m1, m10
2000 pshufb m10, m14, m15
2001 pmaddubsw m10, m9
2002 paddw m6, m10
2003 vpbroadcastd m10, [z_filter_k+4*2+12*2]
2004 pshufb m11, m15
2005 pmaddubsw m11, m10
2006 pshufb m12, m7
2007 pmaddubsw m12, m10
2008 pshufb m13, m7
2009 pmaddubsw m13, m10
2010 pshufb m14, m7
2011 pmaddubsw m14, m10
2012 paddw m0, m11
2013 paddw m2, m12
2014 paddw m1, m13
2015 paddw m6, m14
2016 movu xm11, [tlq+ 61] ; 64- 71
2017 vinserti128 m11, [tlq+ 75], 1 ; 80- 87
2018 movu xm12, [tlq+ 69] ; 72- 79
2019 vinserti128 m12, [tlq+ 83], 1 ; 88- 95
2020 movu xm13, [tlq+ 93] ; 96-103
2021 vinserti128 m13, [tlq+107], 1 ; 112-119
2022 movu xm14, [tlq+101] ; 104-111
2023 vinserti128 m14, [tlq+115], 1 ; 120-127
2024 pmulhrsw m0, m3
2025 pmulhrsw m2, m3
2026 pmulhrsw m1, m3
2027 pmulhrsw m6, m3
2028 lea r3d, [hq-20]
2029 mov tlq, rsp
2030 packuswb m0, m2
2031 packuswb m1, m6
2032 vpbroadcastd xm2, [pb_14]
2033 vbroadcasti128 m6, [pb_0to15]
2034 mova [tlq+32*0], m0
2035 mova [tlq+32*1], m1
2036 movd xm0, r3d
2037 vpbroadcastd m1, [pb_12]
2038 vpbroadcastb m0, xm0
2039 paddb m0, m2
2040 pminub m0, m6 ; clip 64x16 and 64x32
2041 pshufb m12, m0
2042 pminub m1, m6 ; clip 64x64
2043 pshufb m14, m1
2044 pshufb m0, m11, m7
2045 pmaddubsw m0, m10
2046 pshufb m2, m12, m7
2047 pmaddubsw m2, m10
2048 pshufb m1, m13, m7
2049 pmaddubsw m1, m10
2050 pshufb m6, m14, m7
2051 pmaddubsw m6, m10
2052 pshufb m7, m11, m15
2053 pmaddubsw m7, m9
2054 pshufb m10, m12, m15
2055 pmaddubsw m10, m9
2056 paddw m0, m7
2057 pshufb m7, m13, m15
2058 pmaddubsw m7, m9
2059 paddw m2, m10
2060 pshufb m10, m14, m15
2061 pmaddubsw m10, m9
2062 paddw m1, m7
2063 paddw m6, m10
2064 vpbroadcastd m9, [z_filter_k+4*2+12*0]
2065 pshufb m11, m8
2066 pmaddubsw m11, m9
2067 pshufb m12, m8
2068 pmaddubsw m12, m9
2069 pshufb m13, m8
2070 pmaddubsw m13, m9
2071 pshufb m14, m8
2072 pmaddubsw m14, m9
2073 paddw m0, m11
2074 paddw m2, m12
2075 paddw m1, m13
2076 paddw m6, m14
2077 pmulhrsw m0, m3
2078 pmulhrsw m2, m3
2079 pmulhrsw m1, m3
2080 pmulhrsw m6, m3
2081 packuswb m0, m2
2082 packuswb m1, m6
2083 mova [tlq+32*2], m0
2084 mova [tlq+32*3], m1
2085 .w64_main:
2086 movd xm12, dxd
2087 vpbroadcastb m7, [tlq+maxbaseq]
2088 lea r3d, [dxq-64]
2089 shl maxbased, 6
2090 vpbroadcastw m12, xm12
2091 sub r3d, maxbased
2092 vbroadcasti128 m8, [z_filter_s+2]
2093 movd xm6, r3d
2094 mov r5d, dxd
2095 mova m10, [pb_1to32]
2096 vpbroadcastd m11, [pb_32]
2097 vpbroadcastw m6, xm6
2098 .w64_loop:
2099 mov r3d, r5d
2100 shr r3d, 6
2101 movu m0, [tlq+r3+ 0]
2102 movu m1, [tlq+r3+ 8]
2103 pand m2, m4, m6
2104 psubw m9, m5, m2
2105 psllw m2, 8
2106 por m9, m2
2107 pshufb m0, m8
2108 pshufb m1, m8
2109 pmaddubsw m0, m9
2110 pmaddubsw m1, m9
2111 psraw m2, m6, 6
2112 pmulhrsw m0, m3
2113 pmulhrsw m1, m3
2114 packsswb m2, m2
2115 paddb m2, m10
2116 packuswb m0, m1
2117 vpblendvb m0, m7, m0, m2
2118 mova [dstq+ 0], m0
2119 movu m0, [tlq+r3+32]
2120 movu m1, [tlq+r3+40]
2121 add r5d, dxd
2122 pshufb m0, m8
2123 pshufb m1, m8
2124 pmaddubsw m0, m9
2125 pmaddubsw m1, m9
2126 paddb m2, m11
2127 pmulhrsw m0, m3
2128 pmulhrsw m1, m3
2129 paddw m6, m12
2130 packuswb m0, m1
2131 vpblendvb m0, m7, m0, m2
2132 mova [dstq+32], m0
2133 dec hd
2134 jz .w64_end
2135 add dstq, strideq
2136 cmp r5d, maxbased
2137 jb .w64_loop
2138 .w64_end_loop:
2139 mova [dstq+ 0], m7
2140 mova [dstq+32], m7
2141 add dstq, strideq
2142 dec hd
2143 jg .w64_end_loop
2144 .w64_end:
2145 RET
2146
2147 cglobal ipred_z2_8bpc, 3, 10, 16, 224, dst, stride, tl, w, h, angle, dx, dy
2148 %define base r9-z_filter_t0
2149 lea r9, [ipred_z2_avx2_table]
2150 tzcnt wd, wm
2151 movifnidn angled, anglem
2152 movifnidn hd, hm
2153 lea dxq, [dr_intra_derivative-90]
2154 movsxd wq, [r9+wq*4]
2155 movzx dyd, angleb
2156 xor angled, 0x400
2157 mov r8, dxq
2158 sub dxq, dyq
2159 add wq, r9
2160 add r9, z_filter_t0-ipred_z2_avx2_table
2161 mova m2, [tlq-64]
2162 mova m0, [tlq-32]
2163 mova m1, [tlq]
2164 and dyd, ~1
2165 and dxq, ~1
2166 movzx dyd, word [r8+dyq] ; angle - 90
2167 movzx dxd, word [dxq+270] ; 180 - angle
2168 vpbroadcastd m13, [base+pw_512]
2169 vpbroadcastd m14, [base+pw_62]
2170 vpbroadcastd m15, [base+pw_64]
2171 mova [rsp+ 0], m2
2172 mova [rsp+32], m0
2173 mova [rsp+64], m1
2174 neg dxd
2175 neg dyd
2176 jmp wq
2177 .w4:
2178 vpbroadcastq m6, [base+z2_base_inc] ; base_inc << 6
2179 vbroadcasti128 m10, [base+z1_shuf_w4]
2180 vbroadcasti128 m11, [base+z2_shuf_h4]
2181 lea r2d, [dxq+(65<<6)] ; xpos
2182 movd xm5, dyd
2183 mov r8d, (63-4)<<6
2184 mov dyq, -4
2185 pshuflw xm5, xm5, q0000
2186 pmullw xm5, [base+z2_ymul]
2187 test angled, 0x400
2188 jnz .w4_main ; !enable_intra_edge_filter
2189 lea r3d, [hq+2]
2190 add angled, 1022
2191 shl r3d, 6
2192 test r3d, angled
2193 jnz .w4_no_upsample_above ; angle >= 130 || h > 8 || (is_sm && h == 8)
2194 vpbroadcastd xm3, [base+pb_4]
2195 call .upsample_above
2196 sub angled, 1075 ; angle - 53
2197 lea r3d, [hq+3]
2198 xor angled, 0x7f ; 180 - angle
2199 call .filter_strength
2200 jmp .w4_filter_left
2201 ALIGN function_align
2202 .filter_strength:
2203 movd xm8, r3d
2204 mov r3d, angled
2205 movd xm7, angled
2206 vpbroadcastb m8, xm8
2207 shr r3d, 8 ; is_sm << 1
2208 vpbroadcastb m7, xm7
2209 pcmpeqb m8, [base+z_filter_wh]
2210 mova xm9, [r9+r3*8]
2211 pand m0, m8, m7
2212 pcmpgtb m0, m9
2213 pmovmskb r3d, m0
2214 ret
2215 ALIGN function_align
2216 .upsample_above: ; w4/w8
2217 pshufb xm2, xm1, [base+z_upsample1-2]
2218 pminub xm3, [base+z_filter_s+4]
2219 vpbroadcastd xm4, [base+pb_36_m4]
2220 vbroadcasti128 m10, [base+pb_0to15]
2221 pshufb xm3, xm1, xm3
2222 pmaddubsw xm2, xm4
2223 pmaddubsw xm3, xm4
2224 lea r2d, [r2+dxq+(1<<6)]
2225 add dxd, dxd
2226 paddw xm2, xm3
2227 pmulhrsw xm2, xm13
2228 sub r8d, 3<<6
2229 paddw m6, m6
2230 packuswb xm2, xm2
2231 punpcklbw xm1, xm2
2232 mova [rsp+gprsize+64], xm1
2233 ret
2234 ALIGN function_align
2235 .upsample_left: ; h4/h8
2236 mov r3d, hd
2237 and r3d, 4
2238 movd xm2, [rsp+gprsize+64]
2239 movddup xm0, [rsp+gprsize+56]
2240 movd xm1, r3d
2241 palignr xm2, xm0, 1
2242 vpbroadcastb xm1, xm1
2243 pshufb xm2, [base+z_filter_s+18]
2244 vpbroadcastd xm3, [base+pb_36_m4]
2245 pmaxub xm1, [base+z_upsample1-2]
2246 pshufb xm1, xm0, xm1
2247 pmaddubsw xm2, xm3
2248 pmaddubsw xm1, xm3
2249 paddw xm5, xm5
2250 add dyq, dyq
2251 paddw xm1, xm2
2252 pmulhrsw xm1, xm13
2253 vbroadcasti128 m11, [base+z2_upsample]
2254 paddw xm5, xm15
2255 packuswb xm1, xm1
2256 punpcklbw xm0, xm1
2257 mova [rsp+gprsize+48], xm0
2258 ret
2259 .w4_no_upsample_above:
2260 lea r3d, [hq+3]
2261 sub angled, 1112 ; angle - 90
2262 call .filter_strength
2263 test r3d, r3d
2264 jz .w4_no_filter_above
2265 popcnt r3d, r3d
2266 vpbroadcastd xm2, [base+pb_4]
2267 pminub xm2, [base+z_filter_s]
2268 vpbroadcastd xm0, [base+z_filter_k-4+r3*4+12*0]
2269 vpbroadcastd xm4, [base+z_filter_k-4+r3*4+12*1]
2270 pshufb xm3, xm1, xm2 ; 00 01 12 23
2271 pshufd xm2, xm2, q0321
2272 pmaddubsw xm0, xm3, xm0
2273 pshufb xm2, xm1, xm2 ; 12 23 34 44
2274 pmaddubsw xm2, xm4
2275 vpbroadcastd xm4, [base+z_filter_k-4+r3*4+12*2]
2276 punpckhqdq xm3, xm3 ; 34 44 44 44
2277 pmaddubsw xm3, xm4
2278 movd xm4, r6m ; max_width
2279 pminsw xm4, xm15
2280 vpbroadcastb xm4, xm4
2281 paddw xm0, xm2
2282 paddw xm0, xm3
2283 pmulhrsw xm0, xm13
2284 psubb xm4, [base+pb_1to32]
2285 psrlq xm1, 8
2286 packuswb xm0, xm0
2287 vpblendvb xm0, xm1, xm4
2288 movd [rsp+65], xm0
2289 .w4_no_filter_above:
2290 lea r3d, [hq+2]
2291 add angled, 973 ; angle + 883
2292 shl r3d, 6
2293 test r3d, angled
2294 jz .w4_upsample_left ; angle <= 140 || h > 8 || (is_sm && h == 8)
2295 vpbroadcastd xm0, [base+pb_90]
2296 psubb xm0, xm7 ; 180 - angle
2297 pand xm0, xm8 ; reuse from previous filter_strength call
2298 pcmpgtb xm0, xm9
2299 pmovmskb r3d, xm0
2300 .w4_filter_left:
2301 test r3d, r3d
2302 jz .w4_main
2303 popcnt r3d, r3d
2304 mov r5d, 10
2305 cmp hd, 16
2306 movu xm2, [rsp+49]
2307 vinserti128 m2, [rsp+43], 1
2308 cmovs r5d, hd
2309 xor r5d, 15 ; h == 16 ? 5 : 15 - h
2310 movd xm0, r5d
2311 vbroadcasti128 m1, [base+z_filter_s+12]
2312 vbroadcasti128 m4, [base+z_filter_s+16]
2313 vinserti128 m3, m1, [z_filter_s+8], 1 ; 56 67 78 89 9a ab bc cd 55 55 56 67 78 89 9a ab
2314 vpblendd m1, m4, 0x0f ; 78 89 9a ab bc cd de ef 56 67 78 89 9a ab bc cd
2315 vinserti128 m4, [base+z_filter_s+20], 0 ; 9a ab bc cd de ef ff ff 78 89 9a ab bc cd de ef
2316 vpbroadcastb m0, xm0
2317 pmaxub m0, m3
2318 vpbroadcastd m3, [base+z_filter_k-4+r3*4+12*0]
2319 pshufb m0, m2, m0
2320 pmaddubsw m0, m3
2321 vpbroadcastd m3, [base+z_filter_k-4+r3*4+12*1]
2322 pshufb m1, m2, m1
2323 pmaddubsw m1, m3
2324 vpbroadcastd m3, [base+z_filter_k-4+r3*4+12*2]
2325 pshufb m2, m4
2326 pmaddubsw m2, m3
2327 movd xm4, r7m ; max_height
2328 pminsw xm4, xm15
2329 vpbroadcastb xm4, xm4
2330 psubb xm4, [base+pb_16to1]
2331 paddw m1, m0
2332 paddw m1, m2
2333 pmulhrsw m1, m13
2334 vextracti128 xm0, m1, 1
2335 packuswb xm0, xm1
2336 vpblendvb xm0, [rsp+48], xm4
2337 mova [rsp+48], xm0
2338 jmp .w4_main
2339 .w4_upsample_left:
2340 call .upsample_left
2341 .w4_main:
2342 movd xm0, dxd
2343 mova m12, [base+z2_y_shuf_h4]
2344 lea r5, [rsp+56] ; left-7
2345 vpbroadcastw m0, xm0
2346 lea r9, [strideq*3]
2347 psraw xm1, xm5, 6
2348 pand xm5, xm14 ; frac_y
2349 pxor xm2, xm2
2350 paddw m7, m0, m0
2351 psubw xm4, xm2, xm1 ; base_y
2352 vpblendd m0, m7, 0xcc
2353 mova xm1, xm7
2354 punpcklwd xm4, xm2
2355 paddw m0, m1 ; xpos2 xpos3 xpos0 xpos1
2356 psubw xm1, xm15, xm5 ; 64-frac_y
2357 psllw xm5, 8
2358 paddw m7, m7
2359 paddw m6, m0
2360 por xm5, xm1 ; 64-frac_y, frac_y
2361 vpbroadcastq m5, xm5
2362 .w4_loop:
2363 lea r3d, [r2+dxq]
2364 shr r2d, 6 ; base_x0
2365 vpbroadcastq m1, [rsp+r2]
2366 lea r2d, [r3+dxq]
2367 shr r3d, 6 ; base_x1
2368 vpbroadcastq m2, [rsp+r3]
2369 lea r3d, [r2+dxq]
2370 shr r2d, 6 ; base_x2
2371 movq xm0, [rsp+r2]
2372 lea r2d, [r3+dxq]
2373 shr r3d, 6 ; base_x3
2374 movhps xm0, [rsp+r3]
2375 vpblendd m1, m2, 0xc0
2376 pand m2, m14, m6 ; frac_x
2377 vpblendd m0, m1, 0xf0
2378 psubw m1, m15, m2 ; 64-frac_x
2379 psllw m2, 8
2380 pshufb m0, m10
2381 por m1, m2 ; 64-frac_x, frac_x
2382 pmaddubsw m0, m1
2383 cmp r3d, 64
2384 jge .w4_toponly
2385 mova m1, m7 ; arbitrary negative value
2386 vpgatherdq m3, [r5+xm4], m1
2387 pshufb m1, m3, m11
2388 vpermd m1, m12, m1
2389 pmaddubsw m1, m5
2390 psraw m2, m6, 15 ; base_x < topleft
2391 vpblendvb m0, m1, m2
2392 .w4_toponly:
2393 pmulhrsw m0, m13
2394 paddw m6, m7 ; xpos += dx
2395 add r5, dyq
2396 packuswb m0, m0
2397 vextracti128 xm1, m0, 1
2398 movd [dstq+strideq*2], xm0
2399 pextrd [dstq+r9 ], xm0, 1
2400 movd [dstq+strideq*0], xm1
2401 pextrd [dstq+strideq*1], xm1, 1
2402 sub hd, 4
2403 jz .w4_end
2404 lea dstq, [dstq+strideq*4]
2405 cmp r2d, r8d
2406 jge .w4_loop
2407 .w4_leftonly_loop:
2408 mova m1, m7
2409 vpgatherdq m2, [r5+xm4], m1
2410 add r5, dyq
2411 pshufb m0, m2, m11
2412 vpermd m0, m12, m0
2413 pmaddubsw m0, m5
2414 pmulhrsw m0, m13
2415 packuswb m0, m0
2416 vextracti128 xm1, m0, 1
2417 movd [dstq+strideq*2], xm0
2418 pextrd [dstq+r9 ], xm0, 1
2419 movd [dstq+strideq*0], xm1
2420 pextrd [dstq+strideq*1], xm1, 1
2421 lea dstq, [dstq+strideq*4]
2422 sub hd, 4
2423 jg .w4_leftonly_loop
2424 .w4_end:
2425 RET
2426 .w8:
2427 vbroadcasti128 m6, [base+z2_base_inc] ; base_inc << 6
2428 movd xm5, dyd
2429 vbroadcasti128 m10, [base+z_filter_s+2]
2430 vbroadcasti128 m11, [base+z2_shuf_h4]
2431 lea r2d, [dxq+(65<<6)] ; xpos
2432 vpbroadcastw xm5, xm5
2433 mov r8d, (63-8)<<6
2434 mov dyq, -4
2435 pmullw xm5, [base+z2_ymul]
2436 test angled, 0x400
2437 jnz .w8_main
2438 lea r3d, [angleq+126]
2439 mov r3b, hb
2440 cmp r3d, 8
2441 ja .w8_no_upsample_above ; angle >= 130 || h > 8 || is_sm
2442 vpbroadcastd xm3, [base+pb_8]
2443 movhps [rsp+80], xm1
2444 call .upsample_above
2445 sub angled, 53 ; angle - 53
2446 lea r3d, [hq+7]
2447 xor angled, 0x7f ; 180 - angle
2448 call .filter_strength
2449 jmp .w8_filter_left
2450 .w8_no_upsample_above:
2451 lea r3d, [hq+7]
2452 sub angled, 90 ; angle - 90
2453 call .filter_strength
2454 test r3d, r3d
2455 jz .w8_no_filter_above
2456 popcnt r3d, r3d
2457 vpbroadcastd xm3, [base+pb_8]
2458 pminub xm3, [base+z_filter_s+8]
2459 vpbroadcastd xm0, [base+z_filter_k-4+r3*4+12*0]
2460 vpbroadcastd xm4, [base+z_filter_k-4+r3*4+12*1]
2461 pshufb xm2, xm1, [base+z_filter_s] ; 00 01 12 23 34 45 56 67
2462 pmaddubsw xm0, xm2, xm0
2463 pshufb xm3, xm1, xm3 ; 34 45 56 67 78 88 88 88
2464 shufps xm2, xm3, q2121 ; 12 23 34 45 56 67 78 88
2465 pmaddubsw xm2, xm4
2466 vpbroadcastd xm4, [base+z_filter_k-4+r3*4+12*2]
2467 pmaddubsw xm3, xm4
2468 movd xm4, r6m ; max_width
2469 pminuw xm4, xm15
2470 vpbroadcastb xm4, xm4
2471 paddw xm0, xm2
2472 paddw xm0, xm3
2473 pmulhrsw xm0, xm13
2474 psubb xm4, [base+pb_1to32]
2475 psrldq xm1, 1
2476 packuswb xm0, xm0
2477 vpblendvb xm0, xm1, xm4
2478 movq [rsp+65], xm0
2479 .w8_no_filter_above:
2480 lea r3d, [angleq-51]
2481 mov r3b, hb
2482 cmp r3d, 8
2483 jbe .w8_upsample_left ; angle > 140 && h <= 8 && !is_sm
2484 vpbroadcastd m0, [base+pb_90]
2485 psubb m0, m7
2486 pand m0, m8
2487 pcmpgtb m0, m9
2488 pmovmskb r3d, m0
2489 .w8_filter_left:
2490 test r3d, r3d
2491 jz .w8_main
2492 popcnt r3d, r3d
2493 vpbroadcastd m7, [base+z_filter_k-4+r3*4+12*0]
2494 vpbroadcastd m8, [base+z_filter_k-4+r3*4+12*1]
2495 vpbroadcastd m9, [base+z_filter_k-4+r3*4+12*2]
2496 cmp hd, 32
2497 jne .w8_filter_left_h16
2498 movu xm2, [rsp+27]
2499 vinserti128 m2, [rsp+35], 1
2500 vpbroadcastd xm0, [base+pb_5]
2501 vbroadcasti128 m3, [base+z_filter_s+ 8]
2502 vbroadcasti128 m1, [base+z_filter_s+12]
2503 vbroadcasti128 m4, [base+z_filter_s+16]
2504 pmaxub m3, m0
2505 pshufb m3, m2, m3
2506 pmaddubsw m3, m7
2507 pshufb m1, m2, m1
2508 pmaddubsw m1, m8
2509 pshufb m2, m4
2510 pmaddubsw m2, m9
2511 paddw m3, m1
2512 paddw m3, m2
2513 pmulhrsw m3, m13
2514 jmp .w8_filter_left_top16
2515 .w8_filter_left_h16:
2516 mov r5d, 10
2517 cmp hd, 16
2518 cmovs r5d, hd
2519 xor r5d, 15 ; h == 16 ? 5 : 15 - h
2520 movd xm0, r5d
2521 vpbroadcastb m0, xm0
2522 .w8_filter_left_top16:
2523 vbroadcasti128 m1, [base+z_filter_s+12]
2524 vinserti128 m2, m1, [base+z_filter_s+8], 1 ; 56 67 78 89 9a ab bc cd 55 55 56 67 78 89 9a ab
2525 vbroadcasti128 m4, [base+z_filter_s+16]
2526 vpblendd m1, m4, 0x0f ; 78 89 9a ab bc cd de ef 56 67 78 89 9a ab bc cd
2527 vinserti128 m4, [base+z_filter_s+20], 0 ; 9a ab bc cd de ef ff ff 78 89 9a ab bc cd de ef
2528 pmaxub m0, m2
2529 movu xm2, [rsp+49]
2530 vinserti128 m2, [rsp+43], 1
2531 pshufb m0, m2, m0
2532 pmaddubsw m0, m7
2533 movd xm7, r7m ; max_height
2534 pshufb m1, m2, m1
2535 pmaddubsw m1, m8
2536 pshufb m2, m4
2537 pmaddubsw m2, m9
2538 pminsw xm7, xm15
2539 paddw m1, m0
2540 vpbroadcastb m7, xm7
2541 paddw m1, m2
2542 pmulhrsw m1, m13
2543 psubb m7, [base+pb_32to1]
2544 packuswb m3, m1
2545 vpermq m3, m3, q1320
2546 vpblendvb m3, [rsp+32], m7
2547 mova [rsp+32], m3
2548 jmp .w8_main
2549 .w8_upsample_left:
2550 call .upsample_left
2551 .w8_main:
2552 movd xm3, dxd
2553 lea r5, [rsp+56] ; left-7
2554 pshufd xm1, xm5, q3120
2555 pand xm5, xm14
2556 vpbroadcastw m3, xm3
2557 pxor xm0, xm0
2558 psubw xm2, xm15, xm5
2559 psraw xm1, 6
2560 lea r9, [strideq*3]
2561 paddw m7, m3, m3
2562 psubw xm9, xm0, xm1 ; base_y
2563 psllw xm5, 8
2564 punpcklwd xm8, xm9, xm0 ; base_y 0, 1, 4, 5
2565 vpblendd m3, m7, 0xf0 ; xpos0 xpos1
2566 por xm5, xm2 ; 64-frac_y, frac_y
2567 punpckhwd xm9, xm0 ; base_y 2, 3, 6, 7
2568 paddw m6, m3
2569 vinserti128 m12, m5, xm5, 1
2570 .w8_loop:
2571 lea r3d, [r2+dxq]
2572 shr r2d, 6 ; base_x0
2573 movu xm0, [rsp+r2]
2574 lea r2d, [r3+dxq]
2575 shr r3d, 6 ; base_x1
2576 vinserti128 m0, [rsp+r3], 1
2577 lea r3d, [r2+dxq]
2578 shr r2d, 6 ; base_x2
2579 movu xm1, [rsp+r2]
2580 lea r2d, [r3+dxq]
2581 shr r3d, 6 ; base_x3
2582 vinserti128 m1, [rsp+r3], 1
2583 pand m2, m14, m6
2584 paddsw m4, m6, m7
2585 psubw m5, m15, m2
2586 psllw m2, 8
2587 pshufb m0, m10
2588 por m2, m5
2589 pmaddubsw m0, m2
2590 pand m2, m14, m4
2591 psubw m5, m15, m2
2592 psllw m2, 8
2593 pshufb m1, m10
2594 por m2, m5
2595 pmaddubsw m1, m2
2596 cmp r3d, 64
2597 jge .w8_toponly
2598 mova m5, m7
2599 vpgatherdq m3, [r5+xm9], m7
2600 mova m7, m5
2601 vpgatherdq m2, [r5+xm8], m5
2602 pshufb m3, m11
2603 pshufb m2, m11
2604 punpckldq m5, m2, m3 ; a0 b0 c0 d0 a1 b1 c1 d1 e0 f0 g0 h0 e1 f1 g1 h1
2605 punpckhdq m2, m3 ; a2 b2 c2 d2 a3 b3 c3 d3 e2 f2 g2 h2 e3 f3 g3 h3
2606 vpermq m5, m5, q3120 ; y0 y1
2607 vpermq m2, m2, q3120 ; y2 y3
2608 pmaddubsw m5, m12
2609 pmaddubsw m2, m12
2610 psraw m6, 15 ; base_x < topleft
2611 vpblendvb m0, m5, m6
2612 psraw m3, m4, 15
2613 vpblendvb m1, m2, m3
2614 .w8_toponly:
2615 pmulhrsw m0, m13
2616 pmulhrsw m1, m13
2617 paddw m6, m4, m7 ; xpos += dx
2618 add r5, dyq
2619 packuswb m0, m1
2620 vextracti128 xm1, m0, 1
2621 movq [dstq+strideq*0], xm0
2622 movhps [dstq+strideq*2], xm0
2623 movq [dstq+strideq*1], xm1
2624 movhps [dstq+r9 ], xm1
2625 sub hd, 4
2626 jz .w8_end
2627 lea dstq, [dstq+strideq*4]
2628 cmp r2d, r8d
2629 jge .w8_loop
2630 .w8_leftonly_loop:
2631 mova m0, m7
2632 vpgatherdq m5, [r5+xm9], m7
2633 mova m7, m0
2634 vpgatherdq m3, [r5+xm8], m0
2635 add r5, dyq
2636 pshufb m2, m5, m11
2637 pshufb m1, m3, m11
2638 punpckldq m0, m1, m2
2639 punpckhdq m1, m2
2640 vpermq m0, m0, q3120
2641 vpermq m1, m1, q3120
2642 pmaddubsw m0, m12
2643 pmaddubsw m1, m12
2644 pmulhrsw m0, m13
2645 pmulhrsw m1, m13
2646 packuswb m0, m1
2647 vextracti128 xm1, m0, 1
2648 movq [dstq+strideq*0], xm0
2649 movhps [dstq+strideq*2], xm0
2650 movq [dstq+strideq*1], xm1
2651 movhps [dstq+r9 ], xm1
2652 lea dstq, [dstq+strideq*4]
2653 sub hd, 4
2654 jg .w8_leftonly_loop
2655 .w8_end:
2656 RET
2657 .w16:
2658 mov r8d, hd
2659 test angled, 0x400
2660 jnz .w16_main
2661 lea r3d, [hq+15]
2662 sub angled, 90
2663 call .filter_strength
2664 test r3d, r3d
2665 jz .w16_no_filter_above
2666 popcnt r3d, r3d
2667 vbroadcasti128 m6, [tlq+1]
2668 mova xm2, [base+z_filter_s]
2669 vinserti128 m2, [base+z_filter_s+14], 1 ; 00 01 12 23 34 45 56 67 67 78 89 9a ab bc cd de
2670 movu xm3, [base+z_filter_s+8]
2671 vinserti128 m3, [base+z_filter_s+22], 1 ; 34 45 56 67 78 89 9a ab ab bc cd de ef ff ff ff
2672 vpblendd m1, m6, 0xf0
2673 vpbroadcastd m0, [base+z_filter_k-4+r3*4+12*0]
2674 vpbroadcastd m4, [base+z_filter_k-4+r3*4+12*1]
2675 vpbroadcastd m5, [base+z_filter_k-4+r3*4+12*2]
2676 pshufb m2, m1, m2
2677 pshufb m1, m3
2678 pmaddubsw m0, m2, m0
2679 shufps m2, m1, q2121 ; 12 23 34 45 56 67 78 89 89 9a ab bc cd de ef ff
2680 pmaddubsw m2, m4
2681 pmaddubsw m1, m5
2682 movd xm4, r6m ; max_width
2683 pminsw xm4, xm15
2684 vpbroadcastb xm4, xm4
2685 paddw m0, m2
2686 paddw m0, m1
2687 pmulhrsw m0, m13
2688 psubb xm4, [base+pb_1to32]
2689 vextracti128 xm2, m0, 1
2690 packuswb xm0, xm2
2691 vpblendvb xm0, xm6, xm4
2692 movu [rsp+65], xm0
2693 .w16_no_filter_above:
2694 vpbroadcastd m0, [base+pb_90]
2695 psubb m0, m7
2696 pand m0, m8
2697 pcmpgtb m0, m9
2698 pmovmskb r3d, m0
2699 test r3d, r3d
2700 jz .w16_main
2701 popcnt r3d, r3d
2702 vpbroadcastd m7, [base+z_filter_k-4+r3*4+12*0]
2703 vpbroadcastd m8, [base+z_filter_k-4+r3*4+12*1]
2704 vpbroadcastd m9, [base+z_filter_k-4+r3*4+12*2]
2705 .w16_filter_left:
2706 movd xm6, r7m ; max_height
2707 pminsw xm6, xm15
2708 vpbroadcastb m6, xm6
2709 cmp hd, 32
2710 jl .w16_filter_left_h16
2711 vpbroadcastd xm0, [base+pb_5]
2712 vbroadcasti128 m10, [base+z_filter_s+ 8]
2713 vbroadcasti128 m11, [base+z_filter_s+12]
2714 vbroadcasti128 m12, [base+z_filter_s+16]
2715 je .w16_filter_left_h32
2716 movu m3, [tlq-69]
2717 movu m5, [tlq-61]
2718 pmaxub m1, m10, m0
2719 pshufb m1, m3, m1
2720 pmaddubsw m1, m7
2721 pshufb m2, m3, m11
2722 pmaddubsw m2, m8
2723 pshufb m3, m12
2724 pmaddubsw m3, m9
2725 paddw m1, m2
2726 pshufb m2, m5, m10
2727 pmaddubsw m2, m7
2728 pshufb m4, m5, m11
2729 pmaddubsw m4, m8
2730 pshufb m5, m12
2731 pmaddubsw m5, m9
2732 paddw m1, m3
2733 vpbroadcastd m3, [base+pb_32]
2734 paddb m3, [base+pb_32to1]
2735 paddw m2, m4
2736 paddw m2, m5
2737 pmulhrsw m1, m13
2738 pmulhrsw m2, m13
2739 psubb m3, m6, m3
2740 packuswb m1, m2
2741 vpblendvb m1, [tlq-64], m3
2742 mova [rsp], m1
2743 jmp .w16_filter_left_top32
2744 .w16_filter_left_h32:
2745 pmaxub m10, m0
2746 .w16_filter_left_top32:
2747 movu xm2, [tlq-37]
2748 vinserti128 m2, [tlq-29], 1
2749 pshufb m3, m2, m10
2750 pshufb m1, m2, m11
2751 pshufb m2, m12
2752 pmaddubsw m3, m7
2753 pmaddubsw m1, m8
2754 pmaddubsw m2, m9
2755 paddw m3, m1
2756 paddw m3, m2
2757 pmulhrsw m3, m13
2758 jmp .w16_filter_left_top16
2759 .w16_filter_left_h16:
2760 mov r5d, 10
2761 cmp hd, 16
2762 cmovs r5d, hd
2763 xor r5d, 15 ; h == 16 ? 5 : 15 - h
2764 movd xm0, r5d
2765 vpbroadcastb m0, xm0
2766 .w16_filter_left_top16:
2767 movu xm2, [tlq-15]
2768 vinserti128 m2, [tlq-21], 1
2769 vbroadcasti128 m1, [base+z_filter_s+12]
2770 vbroadcasti128 m4, [base+z_filter_s+16]
2771 vinserti128 m5, m1, [base+z_filter_s+8], 1 ; 56 67 78 89 9a ab bc cd 34 45 56 67 78 89 9a ab
2772 vpblendd m1, m4, 0x0f ; 78 89 9a ab bc cd de ef 56 67 78 89 9a ab bc cd
2773 vinserti128 m4, [base+z_filter_s+20], 0 ; 9a ab bc cd de ef ff ff 78 89 9a ab bc cd de ef
2774 pmaxub m0, m5
2775 pshufb m0, m2, m0
2776 pmaddubsw m0, m7
2777 pshufb m1, m2, m1
2778 pmaddubsw m1, m8
2779 pshufb m2, m4
2780 pmaddubsw m2, m9
2781 psubb m6, [base+pb_32to1]
2782 paddw m1, m0
2783 paddw m1, m2
2784 pmulhrsw m1, m13
2785 packuswb m3, m1
2786 vpermq m3, m3, q1320
2787 vpblendvb m3, [tlq-32], m6
2788 mova [rsp+32], m3
2789 .w16_main:
2790 movd xm1, dyd
2791 vbroadcasti128 m10, [base+z_filter_s+2]
2792 movd xm7, dxd
2793 vbroadcasti128 m11, [base+z2_shuf_h2]
2794 vpbroadcastw m1, xm1
2795 vpbroadcastw m7, xm7
2796 mov r7, dstq
2797 pmullw m0, m1, [base+z2_ymul]
2798 psllw xm1, 4
2799 paddw m6, m7, [base+z2_base_inc]
2800 lea r9d, [dxq+(65<<6)] ; xpos
2801 movd [rsp+156], xm1
2802 .w16_loop0:
2803 mov r2d, r9d
2804 mova [rsp+160], m0
2805 lea r5, [rsp+60] ; left-3
2806 mova [rsp+192], m6
2807 pxor m1, m1
2808 psraw m2, m0, 6
2809 pand m0, m14
2810 psubw m9, m1, m2 ; base_y
2811 psubw m12, m15, m0
2812 punpcklwd m8, m9, m1 ; base_y 0, 1, 2, 3, 8, 9, 10, 11
2813 psllw m0, 8
2814 punpckhwd m9, m1 ; base_y 4, 5, 6, 7, 12, 13, 14, 15
2815 por m12, m0 ; 64-frac_y, frac_y
2816 .w16_loop:
2817 lea r3d, [r2+dxq]
2818 shr r2d, 6 ; base_x0
2819 movu xm0, [rsp+r2]
2820 vinserti128 m0, [rsp+r2+8], 1
2821 lea r2d, [r3+dxq]
2822 shr r3d, 6 ; base_x1
2823 movu xm1, [rsp+r3]
2824 vinserti128 m1, [rsp+r3+8], 1
2825 pand m2, m14, m6
2826 paddsw m5, m6, m7
2827 psubw m3, m15, m2
2828 psllw m2, 8
2829 pshufb m0, m10
2830 por m2, m3
2831 pmaddubsw m0, m2
2832 pand m2, m14, m5
2833 psubw m3, m15, m2
2834 psllw m2, 8
2835 pshufb m1, m10
2836 por m2, m3
2837 pmaddubsw m1, m2
2838 cmp r3d, 64
2839 jge .w16_toponly
2840 punpckhwd m2, m5, m5 ; mask out unnecessary loads
2841 vpgatherdd m4, [r5+m9], m2
2842 punpcklwd m2, m5, m5
2843 vpgatherdd m3, [r5+m8], m2
2844 pshufb m4, m11 ; e0 f0 g0 h0 e1 f1 g1 h1 m0 n0 o0 p0 m1 n1 o1 p1
2845 pshufb m3, m11 ; a0 b0 c0 d0 a1 b1 c1 d1 i0 j0 k0 l0 i1 j1 k1 l1
2846 punpcklqdq m2, m3, m4 ; y0
2847 punpckhqdq m3, m4 ; y1
2848 pmaddubsw m2, m12
2849 pmaddubsw m3, m12
2850 psraw m6, 15 ; base_x < topleft
2851 vpblendvb m0, m2, m6
2852 psraw m6, m5, 15
2853 vpblendvb m1, m3, m6
2854 .w16_toponly:
2855 pmulhrsw m0, m13
2856 pmulhrsw m1, m13
2857 paddw m6, m5, m7 ; xpos += dx
2858 sub r5, 2
2859 packuswb m0, m1
2860 vpermq m0, m0, q3120
2861 mova [dstq+strideq*0], xm0
2862 vextracti128 [dstq+strideq*1], m0, 1
2863 sub hd, 2
2864 jz .w16_end
2865 lea dstq, [dstq+strideq*2]
2866 cmp r2d, (63-16)<<6
2867 jge .w16_loop
2868 .w16_leftonly_loop:
2869 mova m0, m7
2870 vpgatherdd m4, [r5+m9], m7
2871 mova m7, m0
2872 vpgatherdd m3, [r5+m8], m0
2873 sub r5, 2
2874 pshufb m2, m4, m11
2875 pshufb m1, m3, m11
2876 punpcklqdq m0, m1, m2
2877 punpckhqdq m1, m2
2878 pmaddubsw m0, m12
2879 pmaddubsw m1, m12
2880 pmulhrsw m0, m13
2881 pmulhrsw m1, m13
2882 packuswb m0, m1
2883 vpermq m0, m0, q3120
2884 mova [dstq+strideq*0], xm0
2885 vextracti128 [dstq+strideq*1], m0, 1
2886 lea dstq, [dstq+strideq*2]
2887 sub hd, 2
2888 jg .w16_leftonly_loop
2889 .w16_end:
2890 sub r8d, 1<<8
2891 jl .w16_ret
2892 vpbroadcastd m0, [rsp+156]
2893 paddw m0, [rsp+160] ; base_y += 16*dy
2894 paddw m6, m13, [rsp+192]
2895 add r7, 16
2896 add r9d, 16<<6
2897 movzx hd, r8b
2898 mov dstq, r7
2899 paddw m6, m13 ; base_x += 16*64
2900 jmp .w16_loop0
2901 .w16_ret:
2902 RET
2903 .w32:
2904 mova m2, [tlq+32]
2905 lea r8d, [hq+(1<<8)]
2906 mova [rsp+96], m2
2907 test angled, 0x400
2908 jnz .w16_main
2909 vpbroadcastd m7, [base+z_filter_k+4*2+12*0]
2910 vpbroadcastd m8, [base+z_filter_k+4*2+12*1]
2911 vpbroadcastd m9, [base+z_filter_k+4*2+12*2]
2912 mova xm5, [base+z_filter_s]
2913 vinserti128 m5, [base+z_filter_s+10], 1 ; 00 01 12 23 34 45 56 67 45 56 67 78 89 9a ab bc
2914 vinserti128 m1, [tlq+11], 1
2915 movu xm6, [base+z_filter_s+12]
2916 vinserti128 m6, [base+z_filter_s+22], 1 ; 56 67 78 89 9a ab bc cd ab bc cd de ef ff ff ff
2917 movu xm3, [tlq+ 6]
2918 vinserti128 m3, [tlq+17], 1
2919 movd xm0, r6m ; max_width
2920 pminsw xm0, xm15
2921 vpbroadcastb m10, xm0
2922 .w32_filter_above:
2923 pshufb m0, m1, m5
2924 shufps m4, m5, m6, q1021 ; 12 23 34 45 56 67 78 89 67 78 89 9a ab bc cd de
2925 pmaddubsw m0, m7
2926 pshufb m2, m1, m4
2927 shufps m5, m6, q2132 ; 34 45 56 67 78 89 9a ab 89 9a ab bc cd de ef ff
2928 pmaddubsw m2, m8
2929 pshufb m1, m5
2930 pmaddubsw m1, m9
2931 paddw m0, m2
2932 paddw m0, m1
2933 pshufb m1, m3, m4
2934 pmaddubsw m1, m7
2935 pshufb m2, m3, m5
2936 pmaddubsw m2, m8
2937 pshufb m3, m6
2938 pmaddubsw m3, m9
2939 paddw m1, m2
2940 paddw m1, m3
2941 pmulhrsw m0, m13
2942 pmulhrsw m1, m13
2943 psubb m10, [base+pb_1to32]
2944 packuswb m0, m1
2945 vpblendvb m0, [tlq+1], m10
2946 movu [rsp+65], m0
2947 jmp .w16_filter_left
2948 .w64:
2949 mova m2, [tlq+32]
2950 mov r3d, [tlq+64]
2951 lea r8d, [hq+(3<<8)]
2952 mova [rsp+ 96], m2
2953 mov [rsp+128], r3d
2954 test angled, 0x400
2955 jnz .w16_main
2956 vpbroadcastd m7, [base+z_filter_k+4*2+12*0]
2957 vpbroadcastd m8, [base+z_filter_k+4*2+12*1]
2958 vpbroadcastd m9, [base+z_filter_k+4*2+12*2]
2959 movu xm6, [base+z_filter_s+ 4]
2960 vinserti128 m6, [base+z_filter_s+10], 1 ; 12 23 34 45 56 67 78 89 45 56 67 78 89 9a ab bc
2961 movu xm3, [tlq+30]
2962 vinserti128 m3, [tlq+43], 1
2963 movu xm5, [base+z_filter_s+16]
2964 vinserti128 m5, [base+z_filter_s+22], 1 ; 78 89 9a ab bc cd de ef ab bc cd de ef ff ff ff
2965 pshufb m0, m3, m6
2966 shufps m4, m6, m5, q1021 ; 34 45 56 67 78 89 9a ab 67 78 89 9a ab bc cd de
2967 pmaddubsw m0, m7
2968 pshufb m2, m3, m4
2969 shufps m6, m5, q2132 ; 56 67 78 89 9a ab bc cd 89 9a ab bc cd de ef ff
2970 pmaddubsw m2, m8
2971 pshufb m3, m6
2972 pmaddubsw m3, m9
2973 paddw m0, m2
2974 paddw m0, m3
2975 movu xm2, [tlq+36]
2976 vinserti128 m2, [tlq+49], 1
2977 pshufb m4, m2, m4
2978 pmaddubsw m4, m7
2979 pshufb m3, m2, m6
2980 pmaddubsw m3, m8
2981 pshufb m2, m5
2982 pmaddubsw m2, m9
2983 movd xm5, r6m ; max_width
2984 pminsw xm5, xm15
2985 vpbroadcastb m10, xm5
2986 paddw m3, m4
2987 paddw m2, m3
2988 vpbroadcastd m3, [base+pb_32]
2989 pmulhrsw m0, m13
2990 pmulhrsw m2, m13
2991 mova xm5, [base+z_filter_s]
2992 vinserti128 m5, [base+z_filter_s+6], 1
2993 psubb m3, m10, m3
2994 psubb m3, [base+pb_1to32]
2995 vinserti128 m1, [tlq+13], 1
2996 packuswb m0, m2
2997 vpblendvb m0, [tlq+33], m3
2998 movu xm3, [tlq+ 6]
2999 vinserti128 m3, [tlq+19], 1
3000 movu [rsp+97], m0
3001 jmp .w32_filter_above
3002
3003 cglobal ipred_z3_8bpc, 4, 9, 0, dst, stride, tl, w, h, angle, dy, org_w, maxbase
3004 %assign org_stack_offset stack_offset
3005 lea r6, [ipred_z3_avx2_table]
3006 tzcnt hd, hm
3007 movifnidn angled, anglem
3008 lea r7, [dr_intra_derivative+45*2-1]
3009 dec tlq
3010 movsxd hq, [r6+hq*4]
3011 sub angled, 180
3012 add hq, r6
3013 mov dyd, angled
3014 neg dyd
3015 xor angled, 0x400
3016 or dyq, ~0x7e
3017 movzx dyd, word [r7+dyq]
3018 vpbroadcastd m3, [pw_512]
3019 vpbroadcastd m4, [pw_62]
3020 vpbroadcastd m5, [pw_64]
3021 mov org_wd, wd
3022 jmp hq
3023 .h4:
3024 lea r7, [strideq*3]
3025 cmp angleb, 40
3026 jae .h4_no_upsample
3027 lea r4d, [angleq-1024]
3028 sar r4d, 7
3029 add r4d, wd
3030 jg .h4_no_upsample ; !enable_intra_edge_filter || w > 8 || (w == 8 && is_sm)
3031 ALLOC_STACK -32, 9
3032 movu xm8, [tlq-7]
3033 pshufb xm0, xm8, [z_upsample1-4]
3034 vpbroadcastb xm2, xm8
3035 pshufb xm1, xm8, [z_filter_s+2]
3036 mova [rsp+16], xm2 ; top[max_base_y]
3037 vpbroadcastd xm2, [pb_36_m4]
3038 add dyd, dyd
3039 pmaddubsw xm0, xm2
3040 pmaddubsw xm1, xm2
3041 movd xm7, dyd
3042 mov r2d, dyd
3043 vpbroadcastw m7, xm7
3044 paddw xm1, xm0
3045 pmulhrsw xm1, xm3
3046 pslldq m6, m7, 8
3047 paddw xm2, xm7, xm7
3048 paddw m6, m7
3049 packuswb xm1, xm1
3050 paddw m6, m2
3051 punpcklbw xm1, xm8
3052 mova xm8, [z_transpose4]
3053 psllw m7, 2
3054 pshufb xm1, [pb_15to0]
3055 mova [rsp], xm1
3056 .h4_upsample_loop:
3057 lea r4d, [r2+dyq]
3058 shr r2d, 6
3059 vpbroadcastq m1, [rsp+r2]
3060 lea r2d, [r4+dyq]
3061 shr r4d, 6
3062 vpbroadcastq m2, [rsp+r4]
3063 lea r4d, [r2+dyq]
3064 shr r2d, 6
3065 movq xm0, [rsp+r2]
3066 lea r2d, [r4+dyq]
3067 shr r4d, 6
3068 movhps xm0, [rsp+r4]
3069 vpblendd m1, m2, 0xc0
3070 pand m2, m4, m6
3071 vpblendd m0, m1, 0xf0
3072 psubw m1, m5, m2
3073 psllw m2, 8
3074 por m1, m2
3075 pmaddubsw m0, m1
3076 paddw m6, m7
3077 pmulhrsw m0, m3
3078 vextracti128 xm1, m0, 1
3079 packuswb xm1, xm0
3080 pshufb xm1, xm8
3081 movd [dstq+strideq*0], xm1
3082 pextrd [dstq+strideq*1], xm1, 1
3083 pextrd [dstq+strideq*2], xm1, 2
3084 pextrd [dstq+r7 ], xm1, 3
3085 add dstq, 4
3086 sub wd, 4
3087 jg .h4_upsample_loop
3088 RET
3089 ALIGN function_align
3090 .filter_strength: ; h4/h8/h16
3091 %define base r4-z_filter_t0
3092 lea r4, [z_filter_t0]
3093 movd xm0, maxbased
3094 movd xm2, angled
3095 shr angled, 8 ; is_sm << 1
3096 vpbroadcastb m0, xm0
3097 vpbroadcastb m2, xm2
3098 pcmpeqb m1, m0, [base+z_filter_wh]
3099 pand m1, m2
3100 mova xm2, [r4+angleq*8]
3101 pcmpgtb m1, m2
3102 pmovmskb r5d, m1
3103 ret
3104 .h4_no_upsample:
3105 %assign stack_offset org_stack_offset
3106 ALLOC_STACK -16, 12
3107 mov maxbased, 7
3108 test angled, 0x400 ; !enable_intra_edge_filter
3109 jnz .h4_main
3110 lea maxbased, [wq+3]
3111 call .filter_strength
3112 mov maxbased, 7
3113 test r5d, r5d
3114 jz .h4_main ; filter_strength == 0
3115 popcnt r5d, r5d
3116 vpbroadcastd m7, [base+pb_7]
3117 vbroadcasti128 m2, [tlq-14]
3118 pmaxub m1, m7, [base+z_filter_s-4]
3119 vpbroadcastd m8, [base+z_filter_k-4+r5*4+12*0]
3120 pmaxub m7, [base+z_filter_s+4]
3121 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*1]
3122 vpbroadcastd m10, [base+z_filter_k-4+r5*4+12*2]
3123 pshufb m0, m2, m1
3124 shufps m1, m7, q2121
3125 pmaddubsw m0, m8
3126 pshufb m1, m2, m1
3127 pmaddubsw m1, m9
3128 pshufb m2, m7
3129 pmaddubsw m2, m10
3130 paddw m0, m1
3131 paddw m0, m2
3132 pmulhrsw m0, m3
3133 mov r4d, 9
3134 lea tlq, [rsp+15]
3135 cmp wd, 4
3136 cmovne maxbased, r4d
3137 vextracti128 xm1, m0, 1
3138 packuswb xm0, xm1
3139 mova [rsp], xm0
3140 .h4_main:
3141 movd xm6, dyd
3142 vpbroadcastq m0, [z_base_inc] ; base_inc << 6
3143 mov r4, tlq
3144 sub tlq, 4
3145 neg dyq
3146 vpbroadcastw m6, xm6
3147 sub r4, maxbaseq
3148 shl maxbased, 6
3149 vpbroadcastb m7, [r4]
3150 lea r4, [dyq+63] ; ypos
3151 movd xm9, maxbased
3152 not maxbased
3153 vbroadcasti128 m8, [z3_shuf_w4]
3154 add maxbased, 64
3155 vpbroadcastw m9, xm9
3156 psrlw m7, 8 ; top[max_base_y]
3157 paddw m10, m6, m6
3158 psubw m9, m0 ; max_base_y
3159 vpblendd m6, m10, 0xcc
3160 mova xm0, xm10
3161 paddw m6, m0 ; ypos2 ypos3 ypos0 ypos1
3162 paddw m10, m10
3163 mova xm11, [z_transpose4]
3164 .h4_loop:
3165 lea r5, [r4+dyq]
3166 sar r4, 6 ; base0
3167 vpbroadcastq m1, [tlq+r4]
3168 lea r4, [r5+dyq]
3169 sar r5, 6 ; base1
3170 vpbroadcastq m2, [tlq+r5]
3171 lea r5, [r4+dyq]
3172 sar r4, 6 ; base2
3173 movq xm0, [tlq+r4]
3174 lea r4, [r5+dyq]
3175 sar r5, 6 ; base3
3176 movhps xm0, [tlq+r5]
3177 vpblendd m1, m2, 0xc0
3178 pand m2, m4, m6 ; frac
3179 vpblendd m0, m1, 0xf0
3180 psubw m1, m5, m2 ; 64-frac
3181 psllw m2, 8
3182 pshufb m0, m8
3183 por m1, m2 ; 64-frac, frac
3184 pmaddubsw m0, m1
3185 pcmpgtw m1, m9, m6 ; base < max_base_y
3186 pmulhrsw m0, m3
3187 paddw m6, m10 ; ypos += dy
3188 vpblendvb m0, m7, m0, m1
3189 vextracti128 xm1, m0, 1
3190 packuswb xm1, xm0
3191 pshufb xm1, xm11 ; transpose
3192 movd [dstq+strideq*0], xm1
3193 pextrd [dstq+strideq*1], xm1, 1
3194 pextrd [dstq+strideq*2], xm1, 2
3195 pextrd [dstq+r7 ], xm1, 3
3196 sub wd, 4
3197 jz .h4_end
3198 add dstq, 4
3199 cmp r4d, maxbased
3200 jg .h4_loop
3201 packuswb xm7, xm7
3202 .h4_end_loop:
3203 movd [dstq+strideq*0], xm7
3204 movd [dstq+strideq*1], xm7
3205 movd [dstq+strideq*2], xm7
3206 movd [dstq+r7 ], xm7
3207 add dstq, 4
3208 sub wd, 4
3209 jg .h4_end_loop
3210 .h4_end:
3211 RET
3212 ALIGN function_align
3213 .h8:
3214 lea r4d, [angleq+216]
3215 mov r4b, wb
3216 cmp r4d, 8
3217 ja .h8_no_upsample ; !enable_intra_edge_filter || is_sm || d >= 40 || w > 8
3218 %assign stack_offset org_stack_offset
3219 ALLOC_STACK -32, 8
3220 and r4d, 4
3221 mova xm0, [tlq-15]
3222 vinserti128 m0, [tlq- 9], 1
3223 movd xm1, r4d
3224 movu xm2, [z_filter_s+2]
3225 vinserti128 m2, [z_filter_s+6], 1
3226 vpbroadcastb xm1, xm1 ; w & 4
3227 vpbroadcastd m7, [pb_36_m4]
3228 pmaxub xm1, [z_upsample1-4] ; clip 4x8
3229 vinserti128 m1, [z_upsample1], 1
3230 add dyd, dyd
3231 pshufb m1, m0, m1
3232 pshufb m2, m0, m2
3233 vinserti128 m0, [tlq-7], 1
3234 movd xm6, dyd
3235 pmaddubsw m1, m7
3236 pmaddubsw m2, m7
3237 vpbroadcastw m6, xm6
3238 mov r2d, dyd
3239 lea r5, [strideq*3]
3240 paddw m7, m6, m6
3241 paddw m1, m2
3242 vpblendd m6, m7, 0xf0
3243 pmulhrsw m1, m3
3244 pslldq m2, m7, 8
3245 paddw m7, m7
3246 paddw m6, m2
3247 vbroadcasti128 m2, [pb_15to0]
3248 packuswb m1, m1
3249 punpcklbw m1, m0
3250 pshufb m1, m2
3251 vextracti128 [rsp+ 0], m1, 1
3252 mova [rsp+16], xm1
3253 .h8_upsample_loop:
3254 lea r4d, [r2+dyq]
3255 shr r2d, 6 ; base0
3256 movu xm0, [rsp+r2]
3257 lea r2d, [r4+dyq]
3258 shr r4d, 6 ; base1
3259 vinserti128 m0, [rsp+r4], 1
3260 lea r4d, [r2+dyq]
3261 shr r2d, 6 ; base2
3262 pand m1, m4, m6
3263 psubw m2, m5, m1
3264 psllw m1, 8
3265 por m2, m1
3266 punpcklqdq m1, m2, m2 ; frac0 frac1
3267 pmaddubsw m0, m1
3268 movu xm1, [rsp+r2]
3269 lea r2d, [r4+dyq]
3270 shr r4d, 6 ; base3
3271 vinserti128 m1, [rsp+r4], 1
3272 punpckhqdq m2, m2 ; frac2 frac3
3273 pmaddubsw m1, m2
3274 pmulhrsw m0, m3
3275 paddw m6, m7
3276 pmulhrsw m1, m3
3277 lea r4, [dstq+strideq*4]
3278 psllw m1, 8
3279 por m0, m1
3280 vextracti128 xm1, m0, 1
3281 punpcklbw xm2, xm0, xm1
3282 punpckhbw xm0, xm1
3283 movd [dstq+strideq*0], xm2
3284 pextrd [dstq+strideq*1], xm2, 1
3285 pextrd [dstq+strideq*2], xm2, 2
3286 pextrd [dstq+r5 ], xm2, 3
3287 movd [r4 +strideq*0], xm0
3288 pextrd [r4 +strideq*1], xm0, 1
3289 pextrd [r4 +strideq*2], xm0, 2
3290 pextrd [r4 +r5 ], xm0, 3
3291 add dstq, 4
3292 sub wd, 4
3293 jg .h8_upsample_loop
3294 RET
3295 .h8_no_intra_edge_filter:
3296 and maxbased, 7
3297 or maxbased, 8 ; imin(w+7, 15)
3298 jmp .h8_main
3299 .h8_no_upsample:
3300 %assign stack_offset org_stack_offset
3301 ALLOC_STACK -32, 10
3302 lea maxbased, [wq+7]
3303 test angled, 0x400
3304 jnz .h8_no_intra_edge_filter
3305 call .filter_strength
3306 test r5d, r5d
3307 jz .h8_main ; filter_strength == 0
3308 popcnt r5d, r5d
3309 vpbroadcastd xm6, [base+pb_15]
3310 pcmpeqb xm1, xm1
3311 psubusb xm6, xm0
3312 psubb xm6, xm1 ; w == 4 ? 5 : 1
3313 movu xm2, [tlq-16]
3314 pmaxub xm1, xm6, [base+z_filter_s]
3315 vinserti128 m2, [tlq-14], 1
3316 vinserti128 m1, [base+z_filter_s+12], 1
3317 vpbroadcastd m7, [base+z_filter_k-4+r5*4+12*0]
3318 pmaxub xm6, [base+z_filter_s+ 8]
3319 vinserti128 m6, [base+z_filter_s+20], 1
3320 pshufb m0, m2, m1
3321 pmaddubsw m0, m7
3322 vpbroadcastd m7, [base+z_filter_k-4+r5*4+12*1]
3323 movzx r4d, byte [tlq-15]
3324 shufps m1, m6, q2121
3325 pshufb m1, m2, m1
3326 pmaddubsw m1, m7
3327 paddw m0, m1
3328 sub r5d, 3
3329 jnz .h8_3tap
3330 vpbroadcastd m7, [z_filter_k+4*8]
3331 movzx r2d, byte [tlq-14]
3332 pshufb m2, m6
3333 pmaddubsw m2, m7
3334 sub r2d, r4d
3335 lea r2d, [r2+r4*8+4]
3336 shr r2d, 3
3337 mov [rsp+15], r2b
3338 paddw m0, m2
3339 .h8_3tap:
3340 pmulhrsw m0, m3
3341 sar r5d, 1
3342 lea tlq, [rsp+31]
3343 add r5d, 17
3344 cmp wd, 16
3345 cmovns maxbased, r5d
3346 neg r5
3347 mov [tlq+r5], r4b
3348 vextracti128 xm1, m0, 1
3349 packuswb xm0, xm1
3350 mova [tlq-15], xm0
3351 .h8_main:
3352 movd xm2, dyd
3353 vbroadcasti128 m0, [z_base_inc]
3354 mov r4, tlq
3355 sub tlq, 8
3356 neg dyq
3357 vpbroadcastw m2, xm2
3358 sub r4, maxbaseq
3359 shl maxbased, 6
3360 vpbroadcastb m7, [r4]
3361 lea r4, [dyq+63]
3362 movd xm9, maxbased
3363 not maxbased
3364 vbroadcasti128 m8, [z3_shuf]
3365 add maxbased, 64
3366 vpbroadcastw m9, xm9
3367 psrlw m7, 8
3368 psubw m9, m0
3369 paddw m6, m2, m2
3370 vpblendd m2, m6, 0x0f
3371 .h8_loop:
3372 lea r5, [r4+dyq]
3373 sar r4, 6
3374 pand m0, m4, m2
3375 psubw m1, m5, m0
3376 psllw m0, 8
3377 por m1, m0
3378 vbroadcasti128 m0, [tlq+r4]
3379 lea r4, [r5+dyq]
3380 sar r5, 6
3381 vinserti128 m0, [tlq+r5], 0
3382 sub rsp, 8*2
3383 pshufb m0, m8
3384 pmaddubsw m0, m1
3385 pcmpgtw m1, m9, m2
3386 paddw m2, m6
3387 pmulhrsw m0, m3
3388 vpblendvb m0, m7, m0, m1
3389 vextracti128 xm1, m0, 1
3390 psllw xm0, 8
3391 por xm0, xm1 ; interleave rows (partial transpose)
3392 mova [rsp], xm0
3393 sub wd, 2
3394 jz .h8_transpose
3395 cmp r4d, maxbased
3396 jg .h8_loop
3397 packuswb xm0, xm7, xm7
3398 .h8_end_loop:
3399 sub rsp, 8*2
3400 mova [rsp], xm0
3401 sub wd, 2
3402 jg .h8_end_loop
3403 .h8_transpose:
3404 mova xm2, [rsp+16*1]
3405 sub org_wd, 8
3406 lea r2, [strideq*3]
3407 lea r6, [dstq+org_wq]
3408 cmovns dstq, r6
3409 punpcklwd xm1, xm2, xm0
3410 punpckhwd xm2, xm0
3411 lea r6, [dstq+strideq*4]
3412 jge .h8_w8
3413 add rsp, 16*2
3414 movd [dstq+strideq*0], xm1
3415 pextrd [dstq+strideq*1], xm1, 1
3416 pextrd [dstq+strideq*2], xm1, 2
3417 pextrd [dstq+r2 ], xm1, 3
3418 movd [r6 +strideq*0], xm2
3419 pextrd [r6 +strideq*1], xm2, 1
3420 pextrd [r6 +strideq*2], xm2, 2
3421 pextrd [r6 +r2 ], xm2, 3
3422 jmp .h8_end
3423 .h8_w8_loop:
3424 mova xm0, [rsp+16*0]
3425 mova xm2, [rsp+16*1]
3426 punpcklwd xm1, xm2, xm0
3427 punpckhwd xm2, xm0
3428 .h8_w8: ; w8/w16/w32
3429 mova xm0, [rsp+16*2]
3430 mova xm4, [rsp+16*3]
3431 add rsp, 16*4
3432 punpcklwd xm3, xm4, xm0
3433 punpckhwd xm4, xm0
3434 punpckldq xm0, xm3, xm1
3435 punpckhdq xm3, xm1
3436 punpckldq xm1, xm4, xm2
3437 punpckhdq xm4, xm2
3438 movq [dstq+strideq*0], xm0
3439 movhps [dstq+strideq*1], xm0
3440 movq [dstq+strideq*2], xm3
3441 movhps [dstq+r2 ], xm3
3442 movq [r6 +strideq*0], xm1
3443 movhps [r6 +strideq*1], xm1
3444 movq [r6 +strideq*2], xm4
3445 movhps [r6 +r2 ], xm4
3446 sub dstq, 8
3447 sub r6, 8
3448 sub org_wd, 8
3449 jge .h8_w8_loop
3450 .h8_end:
3451 RET
3452 .h16_no_intra_edge_filter:
3453 and maxbased, 15
3454 or maxbased, 16 ; imin(w+15, 31)
3455 jmp .h16_main
3456 ALIGN function_align
3457 .h16:
3458 %assign stack_offset org_stack_offset
3459 ALLOC_STACK -64, 12
3460 lea maxbased, [wq+15]
3461 test angled, 0x400
3462 jnz .h16_no_intra_edge_filter
3463 call .filter_strength
3464 test r5d, r5d
3465 jz .h16_main ; filter_strength == 0
3466 popcnt r5d, r5d
3467 vpbroadcastd m11, [base+pb_27]
3468 vpbroadcastd m1, [base+pb_1]
3469 vbroadcasti128 m6, [base+z_filter_s+12]
3470 vinserti128 m2, m6, [base+z_filter_s+4], 0
3471 vinserti128 m6, [base+z_filter_s+20], 1
3472 movu xm10, [tlq-18]
3473 vinserti128 m10, [tlq-14], 1
3474 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*0]
3475 vbroadcasti128 m7, [base+z_filter_s+8]
3476 vinserti128 m8, m7, [base+z_filter_s+0], 0
3477 vinserti128 m7, [base+z_filter_s+16], 1
3478 psubusb m11, m0
3479 por m1, m11
3480 movu xm11, [tlq-32]
3481 vinserti128 m11, [tlq-28], 1
3482 pmaxub m8, m1
3483 pmaxub m7, m1
3484 pshufb m0, m10, m2
3485 shufps m2, m6, q2121
3486 pmaddubsw m0, m9
3487 pshufb m1, m11, m8
3488 shufps m8, m7, q2121
3489 pmaddubsw m1, m9
3490 vpbroadcastd m9, [base+z_filter_k-4+r5*4+12*1]
3491 movzx r4d, byte [tlq-31]
3492 pshufb m2, m10, m2
3493 pmaddubsw m2, m9
3494 pshufb m8, m11, m8
3495 pmaddubsw m8, m9
3496 paddw m0, m2
3497 paddw m1, m8
3498 sub r5d, 3
3499 jnz .h16_3tap
3500 vpbroadcastd m9, [z_filter_k+4*8]
3501 movzx r2d, byte [tlq-30]
3502 pshufb m10, m6
3503 pmaddubsw m10, m9
3504 pshufb m11, m7
3505 pmaddubsw m11, m9
3506 sub r2d, r4d
3507 lea r2d, [r2+r4*8+4]
3508 shr r2d, 3
3509 mov [rsp+31], r2b
3510 paddw m0, m10
3511 paddw m1, m11
3512 .h16_3tap:
3513 pmulhrsw m0, m3
3514 pmulhrsw m1, m3
3515 sar r5d, 1
3516 lea tlq, [rsp+63]
3517 add r5d, 33
3518 cmp wd, 32
3519 cmovns maxbased, r5d
3520 neg r5
3521 mov [tlq+r5], r4b
3522 packuswb m0, m1
3523 vpermq m0, m0, q2031
3524 mova [tlq-31], m0
3525 .h16_main:
3526 movd xm6, dyd
3527 vbroadcasti128 m0, [z_base_inc]
3528 mov r4, tlq
3529 sub tlq, 8
3530 neg dyq
3531 vpbroadcastw m6, xm6
3532 sub r4, maxbaseq
3533 shl maxbased, 6
3534 vpbroadcastb m7, [r4]
3535 lea r4, [dyq+63]
3536 movd xm9, maxbased
3537 not maxbased
3538 vbroadcasti128 m8, [z3_shuf]
3539 add maxbased, 64
3540 vpbroadcastw m9, xm9
3541 psubw m9, m0
3542 paddw m11, m6, m6
3543 psubw m10, m9, m3 ; 64*8
3544 vpblendd m6, m11, 0xf0
3545 .h16_loop:
3546 lea r5, [r4+dyq]
3547 sar r4, 6
3548 pand m1, m4, m6
3549 psubw m2, m5, m1
3550 psllw m1, 8
3551 por m2, m1
3552 movu xm0, [tlq+r4-0]
3553 movu xm1, [tlq+r4-8]
3554 lea r4, [r5+dyq]
3555 sar r5, 6
3556 vinserti128 m0, [tlq+r5-0], 1
3557 vinserti128 m1, [tlq+r5-8], 1
3558 sub rsp, 32
3559 pshufb m0, m8
3560 pshufb m1, m8
3561 pmaddubsw m0, m2
3562 pmaddubsw m1, m2
3563 pmulhrsw m0, m3
3564 pmulhrsw m1, m3
3565 packuswb m0, m1
3566 pcmpgtw m1, m9, m6
3567 pcmpgtw m2, m10, m6
3568 packsswb m1, m2
3569 paddw m6, m11
3570 vpblendvb m0, m7, m0, m1
3571 vpermq m0, m0, q3120
3572 mova [rsp], m0
3573 sub wd, 2
3574 jz .h16_transpose
3575 cmp r4d, maxbased
3576 jg .h16_loop
3577 mova m0, m7
3578 .h16_end_loop:
3579 sub rsp, 32
3580 mova [rsp], m7
3581 sub wd, 2
3582 jg .h16_end_loop
3583 .h16_transpose:
3584 mova m2, [rsp+32*1]
3585 sub org_wd, 8
3586 lea r2, [strideq*3]
3587 lea r6, [dstq+org_wq]
3588 cmovns dstq, r6
3589 punpcklbw m1, m2, m0
3590 punpckhbw m2, m0
3591 lea r3, [strideq*5]
3592 punpcklbw m0, m1, m2
3593 punpckhbw m1, m2
3594 lea r4, [strideq+r2*2] ; stride*7
3595 jge .h16_w8
3596 add rsp, 32*2
3597 movd [dstq+strideq*0], xm0
3598 pextrd [dstq+strideq*1], xm0, 1
3599 pextrd [dstq+strideq*2], xm0, 2
3600 pextrd [dstq+r2 ], xm0, 3
3601 vextracti128 xm0, m0, 1
3602 movd [dstq+strideq*4], xm1
3603 pextrd [dstq+r3 ], xm1, 1
3604 pextrd [dstq+r2*2 ], xm1, 2
3605 pextrd [dstq+r4 ], xm1, 3
3606 lea dstq, [dstq+strideq*8]
3607 vextracti128 xm1, m1, 1
3608 movd [dstq+strideq*0], xm0
3609 pextrd [dstq+strideq*1], xm0, 1
3610 pextrd [dstq+strideq*2], xm0, 2
3611 pextrd [dstq+r2 ], xm0, 3
3612 movd [dstq+strideq*4], xm1
3613 pextrd [dstq+r3 ], xm1, 1
3614 pextrd [dstq+r2*2 ], xm1, 2
3615 pextrd [dstq+r4 ], xm1, 3
3616 jmp .h16_end
3617 .h16_w8_loop:
3618 mova m0, [rsp+32*0]
3619 mova m2, [rsp+32*1]
3620 punpcklbw m1, m2, m0
3621 punpckhbw m2, m0
3622 punpcklbw m0, m1, m2
3623 punpckhbw m1, m2
3624 .h16_w8:
3625 mova m2, [rsp+32*2]
3626 mova m4, [rsp+32*3]
3627 lea r6, [dstq+strideq*8]
3628 add rsp, 32*4
3629 punpcklbw m3, m4, m2
3630 punpckhbw m4, m2
3631 punpcklbw m2, m3, m4
3632 punpckhbw m3, m4
3633 punpckldq m4, m2, m0
3634 punpckhdq m2, m0
3635 punpckldq m0, m3, m1
3636 punpckhdq m3, m1
3637 movq [dstq+strideq*0], xm4
3638 movhps [dstq+strideq*1], xm4
3639 vextracti128 xm4, m4, 1
3640 movq [dstq+strideq*2], xm2
3641 movhps [dstq+r2 ], xm2
3642 vextracti128 xm2, m2, 1
3643 movq [dstq+strideq*4], xm0
3644 movhps [dstq+r3 ], xm0
3645 vextracti128 xm0, m0, 1
3646 movq [dstq+r2*2 ], xm3
3647 movhps [dstq+r4 ], xm3
3648 vextracti128 xm3, m3, 1
3649 movq [r6+strideq*0], xm4
3650 movhps [r6+strideq*1], xm4
3651 movq [r6+strideq*2], xm2
3652 movhps [r6+r2 ], xm2
3653 movq [r6+strideq*4], xm0
3654 movhps [r6+r3 ], xm0
3655 movq [r6+r2*2 ], xm3
3656 movhps [r6+r4 ], xm3
3657 sub dstq, 8
3658 sub org_wd, 8
3659 jge .h16_w8_loop
3660 .h16_end:
3661 RET
3662 ALIGN function_align
3663 .h32:
3664 %assign stack_offset org_stack_offset
3665 ALLOC_STACK -96, 15
3666 lea maxbased, [wq+31]
3667 and maxbased, 31
3668 or maxbased, 32 ; imin(w+31, 63)
3669 test angled, 0x400 ; !enable_intra_edge_filter
3670 jnz .h32_main
3671 vbroadcasti128 m0, [pb_0to15]
3672 mov r4d, 21
3673 mov r5d, 3
3674 movu xm11, [tlq-66] ; 56-63
3675 vinserti128 m11, [tlq-52], 1 ; 40-47
3676 sub r4d, wd ; 21-w
3677 cmovns r5d, r4d
3678 movu xm12, [tlq-58] ; 48-55
3679 vinserti128 m12, [tlq-44], 1 ; 32-39
3680 sub r4d, 8 ; 13-w
3681 movd xm1, r5d
3682 movu xm13, [tlq-34] ; 24-31
3683 vinserti128 m13, [tlq-20], 1 ; 8-15
3684 movd xm2, r4d
3685 vpbroadcastb m1, xm1
3686 movu xm14, [tlq-28] ; 16-23
3687 vinserti128 m14, [tlq-14], 1 ; 0- 7
3688 vpbroadcastb m2, xm2
3689 pmaxsb m1, m0 ; clip 16x32 and (32|64)x32
3690 movu m7, [z_filter_s+4]
3691 pshufb m11, m1
3692 vinserti128 m8, m7, [z_filter_s+8], 1
3693 vinserti128 m7, [z_filter_s+16], 0
3694 pmaxsb m2, m0 ; clip 8x32
3695 vpbroadcastd m9, [z_filter_k+4*2+12*0]
3696 pshufb m12, m2
3697 pshufb m0, m11, m8
3698 pmaddubsw m0, m9
3699 pshufb m2, m12, m8
3700 pmaddubsw m2, m9
3701 pshufb m1, m13, m8
3702 pmaddubsw m1, m9
3703 shufps m8, m7, q1021
3704 pshufb m6, m14, m8
3705 pmaddubsw m6, m9
3706 vpbroadcastd m9, [z_filter_k+4*2+12*1]
3707 pshufb m10, m11, m8
3708 pmaddubsw m10, m9
3709 paddw m0, m10
3710 pshufb m10, m12, m8
3711 pmaddubsw m10, m9
3712 paddw m2, m10
3713 pshufb m10, m13, m8
3714 pmaddubsw m10, m9
3715 shufps m8, m7, q2121
3716 paddw m1, m10
3717 pshufb m10, m14, m8
3718 pmaddubsw m10, m9
3719 paddw m6, m10
3720 vpbroadcastd m9, [z_filter_k+4*2+12*2]
3721 pshufb m11, m8
3722 pmaddubsw m11, m9
3723 pshufb m12, m8
3724 pmaddubsw m12, m9
3725 movzx r4d, byte [tlq-63]
3726 movzx r2d, byte [tlq-62]
3727 paddw m0, m11
3728 paddw m2, m12
3729 pshufb m13, m8
3730 pmaddubsw m13, m9
3731 pshufb m14, m7
3732 pmaddubsw m14, m9
3733 paddw m1, m13
3734 paddw m6, m14
3735 sub r2d, r4d
3736 lea r2d, [r2+r4*8+4] ; edge case for 64x32
3737 pmulhrsw m0, m3
3738 pmulhrsw m2, m3
3739 pmulhrsw m1, m3
3740 pmulhrsw m6, m3
3741 shr r2d, 3
3742 mov [rsp+31], r2b
3743 lea tlq, [rsp+95]
3744 mov [tlq-65], r4b
3745 mov r4d, 65
3746 cmp wd, 64
3747 cmove maxbased, r4d
3748 packuswb m0, m2
3749 packuswb m1, m6
3750 mova [tlq-63], m0
3751 mova [tlq-31], m1
3752 .h32_main:
3753 movd xm6, dyd
3754 mov r4, tlq
3755 sub tlq, 8
3756 neg dyq
3757 vpbroadcastw m6, xm6
3758 sub r4, maxbaseq
3759 shl maxbased, 6
3760 vpbroadcastb m7, [r4]
3761 lea r4, [dyq+63]
3762 movd xm9, maxbased
3763 not maxbased
3764 vbroadcasti128 m8, [z3_shuf]
3765 add maxbased, 64
3766 vpbroadcastw m9, xm9
3767 psubw m9, [z_base_inc]
3768 mova m11, m6
3769 psubw m10, m9, m3 ; 64*8
3770 .h32_loop:
3771 mov r5, r4
3772 sar r5, 6
3773 pand m1, m4, m6
3774 psubw m2, m5, m1
3775 psllw m1, 8
3776 por m2, m1
3777 movu xm0, [tlq+r5- 0]
3778 vinserti128 m0, [tlq+r5-16], 1
3779 movu xm1, [tlq+r5- 8]
3780 vinserti128 m1, [tlq+r5-24], 1
3781 sub rsp, 32
3782 add r4, dyq
3783 pshufb m0, m8
3784 pshufb m1, m8
3785 pmaddubsw m0, m2
3786 pmaddubsw m1, m2
3787 pmulhrsw m0, m3
3788 pmulhrsw m1, m3
3789 packuswb m0, m1
3790 pcmpgtw m1, m9, m6
3791 pcmpgtw m2, m10, m6
3792 packsswb m1, m2
3793 paddw m6, m11
3794 vpblendvb m0, m7, m0, m1
3795 mova [rsp], m0
3796 dec wd
3797 jz .h32_transpose
3798 cmp r4d, maxbased
3799 jg .h32_loop
3800 .h32_end_loop:
3801 sub rsp, 32
3802 mova [rsp], m7
3803 dec wd
3804 jg .h32_end_loop
3805 .h32_transpose:
3806 lea dstq, [dstq+org_wq-8]
3807 lea r2, [strideq*3]
3808 lea r3, [strideq*5]
3809 lea r4, [strideq+r2*2] ; stride*7
3810 .h32_w8_loop:
3811 mova m7, [rsp+32*0]
3812 mova m6, [rsp+32*1]
3813 mova m5, [rsp+32*2]
3814 mova m4, [rsp+32*3]
3815 mova m3, [rsp+32*4]
3816 mova m2, [rsp+32*5]
3817 mova m1, [rsp+32*6]
3818 mova m0, [rsp+32*7]
3819 lea r6, [dstq+strideq*8]
3820 add rsp, 32*8
3821 punpcklbw m8, m0, m1
3822 punpckhbw m0, m1
3823 punpcklbw m1, m2, m3
3824 punpckhbw m2, m3
3825 punpcklbw m3, m4, m5
3826 punpckhbw m4, m5
3827 punpcklbw m5, m6, m7
3828 punpckhbw m6, m7
3829 punpcklwd m7, m8, m1
3830 punpckhwd m8, m1
3831 punpcklwd m1, m0, m2
3832 punpckhwd m0, m2
3833 punpcklwd m2, m3, m5
3834 punpckhwd m3, m5
3835 punpcklwd m5, m4, m6
3836 punpckhwd m4, m6
3837 punpckldq m6, m7, m2
3838 punpckhdq m7, m2
3839 punpckldq m2, m8, m3
3840 punpckhdq m8, m3
3841 punpckldq m3, m1, m5
3842 punpckhdq m1, m5
3843 punpckldq m5, m0, m4
3844 punpckhdq m0, m4
3845 movq [dstq+strideq*0], xm6
3846 movhps [dstq+strideq*1], xm6
3847 vextracti128 xm6, m6, 1
3848 movq [dstq+strideq*2], xm7
3849 movhps [dstq+r2 ], xm7
3850 vextracti128 xm7, m7, 1
3851 movq [dstq+strideq*4], xm2
3852 movhps [dstq+r3 ], xm2
3853 vextracti128 xm2, m2, 1
3854 movq [dstq+r2*2 ], xm8
3855 movhps [dstq+r4 ], xm8
3856 vextracti128 xm8, m8, 1
3857 movq [r6+strideq*0], xm3
3858 movhps [r6+strideq*1], xm3
3859 vextracti128 xm3, m3, 1
3860 movq [r6+strideq*2], xm1
3861 movhps [r6+r2 ], xm1
3862 vextracti128 xm1, m1, 1
3863 movq [r6+strideq*4], xm5
3864 movhps [r6+r3 ], xm5
3865 vextracti128 xm5, m5, 1
3866 movq [r6+r2*2 ], xm0
3867 movhps [r6+r4 ], xm0
3868 lea r6, [r6+strideq*8]
3869 vextracti128 xm0, m0, 1
3870 movq [r6+strideq*0], xm6
3871 movhps [r6+strideq*1], xm6
3872 movq [r6+strideq*2], xm7
3873 movhps [r6+r2 ], xm7
3874 movq [r6+strideq*4], xm2
3875 movhps [r6+r3 ], xm2
3876 movq [r6+r2*2 ], xm8
3877 movhps [r6+r4 ], xm8
3878 lea r6, [r6+strideq*8]
3879 movq [r6+strideq*0], xm3
3880 movhps [r6+strideq*1], xm3
3881 movq [r6+strideq*2], xm1
3882 movhps [r6+r2 ], xm1
3883 movq [r6+strideq*4], xm5
3884 movhps [r6+r3 ], xm5
3885 movq [r6+r2*2 ], xm0
3886 movhps [r6+r4 ], xm0
3887 sub dstq, 8
3888 sub org_wd, 8
3889 jg .h32_w8_loop
3890 RET
3891 ALIGN function_align
3892 .h64:
3893 %assign stack_offset org_stack_offset
3894 ALLOC_STACK -128, 16
3895 lea maxbased, [wq+63]
3896 test angled, 0x400 ; !enable_intra_edge_filter
3897 jnz .h64_main
3898 mov r4d, 21
3899 vpbroadcastb xm11, [tlq-127]
3900 vpblendd xm11, [tlq-130], 0x0e ; 120-127
3901 sub r4d, wd ; 21-w
3902 mov r5d, 3
3903 vinserti128 m11, [tlq-116], 1 ; 104-111
3904 movu m7, [z_filter_s+4]
3905 cmp wd, 32
3906 cmove r4d, r5d
3907 vinserti128 m8, m7, [z_filter_s+8], 1
3908 vbroadcasti128 m6, [pb_0to15]
3909 movd xm1, r4d
3910 vpbroadcastd m9, [z_filter_k+4*2+12*0]
3911 movu xm12, [tlq-122] ; 112-119
3912 vinserti128 m12, [tlq-108], 1 ; 96-103
3913 vpbroadcastb m1, xm1
3914 movu xm13, [tlq- 98] ; 88- 95
3915 vinserti128 m13, [tlq- 84], 1 ; 72- 79
3916 movu xm14, [tlq- 90] ; 80- 87
3917 vinserti128 m14, [tlq- 76], 1 ; 64- 71
3918 vinserti128 m7, [z_filter_s+16], 0
3919 pshufb m0, m11, m8
3920 pmaddubsw m0, m9
3921 pshufb m2, m12, m8
3922 pmaddubsw m2, m9
3923 pmaxsb m1, m6 ; clip (16|32)x64
3924 pshufb m13, m1
3925 pshufb m1, m13, m8
3926 pmaddubsw m1, m9
3927 pshufb m6, m14, m8
3928 pmaddubsw m6, m9
3929 vpbroadcastd m9, [z_filter_k+4*2+12*1]
3930 shufps m15, m8, m7, q1021
3931 pshufb m10, m11, m15
3932 pmaddubsw m10, m9
3933 paddw m0, m10
3934 pshufb m10, m12, m15
3935 pmaddubsw m10, m9
3936 paddw m2, m10
3937 pshufb m10, m13, m15
3938 pmaddubsw m10, m9
3939 paddw m1, m10
3940 pshufb m10, m14, m15
3941 pmaddubsw m10, m9
3942 paddw m6, m10
3943 vpbroadcastd m9, [z_filter_k+4*2+12*2]
3944 shufps m10, m8, m7, q2132
3945 pshufb m11, m10
3946 pmaddubsw m11, m9
3947 pshufb m12, m10
3948 pmaddubsw m12, m9
3949 pshufb m13, m10
3950 pmaddubsw m13, m9
3951 pshufb m14, m10
3952 pmaddubsw m14, m9
3953 paddw m0, m11
3954 paddw m2, m12
3955 paddw m1, m13
3956 paddw m6, m14
3957 movu xm11, [tlq-66] ; 56-63
3958 vinserti128 m11, [tlq-52], 1 ; 40-47
3959 movu xm12, [tlq-58] ; 48-55
3960 vinserti128 m12, [tlq-44], 1 ; 32-39
3961 movu xm13, [tlq-34] ; 24-31
3962 vinserti128 m13, [tlq-20], 1 ; 8-15
3963 movu xm14, [tlq-28] ; 16-23
3964 vinserti128 m14, [tlq-14], 1 ; 0- 7
3965 pmulhrsw m0, m3
3966 pmulhrsw m2, m3
3967 pmulhrsw m1, m3
3968 pmulhrsw m6, m3
3969 lea tlq, [rsp+127]
3970 packuswb m0, m2
3971 packuswb m1, m6
3972 mova [tlq-127], m0
3973 mova [tlq- 95], m1
3974 pshufb m0, m11, m10
3975 pmaddubsw m0, m9
3976 pshufb m2, m12, m10
3977 pmaddubsw m2, m9
3978 pshufb m1, m13, m10
3979 pmaddubsw m1, m9
3980 pshufb m6, m14, m7
3981 pmaddubsw m6, m9
3982 vpbroadcastd m9, [z_filter_k+4*2+12*1]
3983 pshufb m7, m11, m15
3984 pmaddubsw m7, m9
3985 paddw m0, m7
3986 pshufb m7, m12, m15
3987 pmaddubsw m7, m9
3988 paddw m2, m7
3989 pshufb m7, m13, m15
3990 pmaddubsw m7, m9
3991 paddw m1, m7
3992 pshufb m7, m14, m10
3993 pmaddubsw m7, m9
3994 paddw m6, m7
3995 vpbroadcastd m9, [z_filter_k+4*2+12*0]
3996 pshufb m11, m8
3997 pmaddubsw m11, m9
3998 pshufb m12, m8
3999 pmaddubsw m12, m9
4000 pshufb m13, m8
4001 pmaddubsw m13, m9
4002 pshufb m14, m15
4003 pmaddubsw m14, m9
4004 paddw m0, m11
4005 paddw m2, m12
4006 paddw m1, m13
4007 paddw m6, m14
4008 pmulhrsw m0, m3
4009 pmulhrsw m2, m3
4010 pmulhrsw m1, m3
4011 pmulhrsw m6, m3
4012 packuswb m0, m2
4013 packuswb m1, m6
4014 mova [tlq-63], m0
4015 mova [tlq-31], m1
4016 .h64_main:
4017 movd xm12, dyd
4018 neg maxbaseq
4019 vbroadcasti128 m8, [z3_shuf]
4020 vpbroadcastb m7, [tlq+maxbaseq]
4021 shl maxbased, 6
4022 vpbroadcastw m12, xm12
4023 lea r5d, [dyq+maxbaseq-64]
4024 neg dyq
4025 or maxbased, 63
4026 lea r4, [dyq+63]
4027 movd xm6, r5d
4028 mova xm10, [pb_1to32+16]
4029 vinserti128 m10, [pb_1to32], 1
4030 vpbroadcastd m11, [pb_32]
4031 vpbroadcastw m6, xm6
4032 .h64_loop:
4033 mov r5, r4
4034 sar r5, 6
4035 movu m0, [tlq+r5-24]
4036 movu m1, [tlq+r5-32]
4037 pand m2, m4, m6
4038 psubw m9, m5, m2
4039 psllw m2, 8
4040 por m9, m2
4041 pshufb m0, m8
4042 pshufb m1, m8
4043 pmaddubsw m0, m9
4044 pmaddubsw m1, m9
4045 psraw m2, m6, 6
4046 sub rsp, 64
4047 pmulhrsw m0, m3
4048 pmulhrsw m1, m3
4049 packsswb m2, m2
4050 paddb m2, m10
4051 packuswb m0, m1
4052 vpblendvb m0, m7, m0, m2
4053 mova [rsp+32], m0
4054 movu m0, [tlq+r5-56]
4055 movu m1, [tlq+r5-64]
4056 add r4, dyq
4057 pshufb m0, m8
4058 pshufb m1, m8
4059 pmaddubsw m0, m9
4060 pmaddubsw m1, m9
4061 paddb m2, m11
4062 pmulhrsw m0, m3
4063 pmulhrsw m1, m3
4064 paddw m6, m12
4065 packuswb m0, m1
4066 vpblendvb m0, m7, m0, m2
4067 mova [rsp], m0
4068 dec wd
4069 jz .h64_transpose
4070 cmp r4d, maxbased
4071 jg .h64_loop
4072 .h64_end_loop:
4073 sub rsp, 64
4074 mova [rsp+32], m7
4075 mova [rsp+ 0], m7
4076 dec wd
4077 jg .h64_end_loop
4078 .h64_transpose:
4079 lea r2, [strideq*3]
4080 lea r3, [strideq*5]
4081 imul r5, strideq, -8
4082 lea dstq, [dstq+org_wq-16]
4083 lea r4, [strideq+r2*2] ; stride*7
4084 .h64_transpose_loop0:
4085 lea r6, [rsp+16*3]
4086 .h64_transpose_loop:
4087 mova xm0, [r6+64*15]
4088 vinserti128 m0, [r6+64* 7], 1
4089 mova xm1, [r6+64*14]
4090 vinserti128 m1, [r6+64* 6], 1
4091 mova xm2, [r6+64*13]
4092 vinserti128 m2, [r6+64* 5], 1
4093 mova xm3, [r6+64*12]
4094 vinserti128 m3, [r6+64* 4], 1
4095 mova xm4, [r6+64*11]
4096 vinserti128 m4, [r6+64* 3], 1
4097 mova xm5, [r6+64*10]
4098 vinserti128 m5, [r6+64* 2], 1
4099 mova xm6, [r6+64* 9]
4100 vinserti128 m6, [r6+64* 1], 1
4101 mova xm7, [r6+64* 8]
4102 vinserti128 m7, [r6+64* 0], 1
4103 sub r6, 16
4104 punpcklbw m8, m0, m1
4105 punpckhbw m0, m1
4106 punpcklbw m1, m2, m3
4107 punpckhbw m2, m3
4108 punpcklbw m3, m4, m5
4109 punpckhbw m4, m5
4110 punpcklbw m5, m6, m7
4111 punpckhbw m6, m7
4112 punpcklwd m7, m8, m1
4113 punpckhwd m8, m1
4114 punpcklwd m1, m0, m2
4115 punpckhwd m0, m2
4116 punpcklwd m2, m3, m5
4117 punpckhwd m3, m5
4118 punpcklwd m5, m4, m6
4119 punpckhwd m4, m6
4120 punpckldq m6, m7, m2
4121 punpckhdq m7, m2
4122 punpckldq m2, m8, m3
4123 punpckhdq m8, m3
4124 punpckldq m3, m1, m5
4125 punpckhdq m1, m5
4126 punpckldq m5, m0, m4
4127 punpckhdq m0, m4
4128 vpermq m6, m6, q3120
4129 vpermq m7, m7, q3120
4130 vpermq m2, m2, q3120
4131 vpermq m8, m8, q3120
4132 vpermq m3, m3, q3120
4133 vpermq m1, m1, q3120
4134 vpermq m5, m5, q3120
4135 vpermq m0, m0, q3120
4136 mova [dstq+strideq*0], xm6
4137 vextracti128 [dstq+strideq*1], m6, 1
4138 mova [dstq+strideq*2], xm7
4139 vextracti128 [dstq+r2 ], m7, 1
4140 mova [dstq+strideq*4], xm2
4141 vextracti128 [dstq+r3 ], m2, 1
4142 mova [dstq+r2*2 ], xm8
4143 vextracti128 [dstq+r4 ], m8, 1
4144 sub dstq, r5
4145 mova [dstq+strideq*0], xm3
4146 vextracti128 [dstq+strideq*1], m3, 1
4147 mova [dstq+strideq*2], xm1
4148 vextracti128 [dstq+r2 ], m1, 1
4149 mova [dstq+strideq*4], xm5
4150 vextracti128 [dstq+r3 ], m5, 1
4151 mova [dstq+r2*2 ], xm0
4152 vextracti128 [dstq+r4 ], m0, 1
4153 sub dstq, r5
4154 cmp r6, rsp
4155 jae .h64_transpose_loop
4156 add rsp, 64*16
4157 lea dstq, [dstq+r5*8-16]
4158 sub org_wd, 16
4159 jg .h64_transpose_loop0
4160 .h64_end:
4161 RET
4162
4163 %macro FILTER_XMM 4 ; dst, src, tmp, shuf
4164 %ifnum %4
4165 pshufb xm%2, xm%4
4166 %else
4167 pshufb xm%2, %4
4168 %endif
4169 pshufd xm%1, xm%2, q0000 ; p0 p1
4170 pmaddubsw xm%1, xm2
4171 pshufd xm%3, xm%2, q1111 ; p2 p3
4172 pmaddubsw xm%3, xm3
4173 paddw xm%1, xm1
4174 paddw xm%1, xm%3
4175 pshufd xm%3, xm%2, q2222 ; p4 p5
4176 pmaddubsw xm%3, xm4
4177 paddw xm%1, xm%3
4178 pshufd xm%3, xm%2, q3333 ; p6 __
4179 pmaddubsw xm%3, xm5
4180 paddw xm%1, xm%3
4181 psraw xm%1, 4
4182 packuswb xm%1, xm%1
4183 %endmacro
4184
4185 %macro FILTER_YMM 4 ; dst, src, tmp, shuf
4186 pshufb m%2, m%4
4187 pshufd m%1, m%2, q0000
4188 pmaddubsw m%1, m2
4189 pshufd m%3, m%2, q1111
4190 pmaddubsw m%3, m3
4191 paddw m%1, m1
4192 paddw m%1, m%3
4193 pshufd m%3, m%2, q2222
4194 pmaddubsw m%3, m4
4195 paddw m%1, m%3
4196 pshufd m%3, m%2, q3333
4197 pmaddubsw m%3, m5
4198 paddw m%1, m%3
4199 psraw m%1, 4
4200 vperm2i128 m%3, m%1, m%1, 0x01
4201 packuswb m%1, m%3
4202 %endmacro
4203
4204 ; The ipred_filter SIMD processes 4x2 blocks in the following order which
4205 ; increases parallelism compared to doing things row by row. One redundant
4206 ; block is calculated for w8 and w16, two for w32.
4207 ; w4 w8 w16 w32
4208 ; 1 1 2 1 2 3 5 1 2 3 5 b c d f
4209 ; 2 2 3 2 4 5 7 2 4 5 7 c e f h
4210 ; 3 3 4 4 6 7 9 4 6 7 9 e g h j
4211 ; ___ 4 ___ 4 5 ___ 6 8 9 a ___ 6 8 9 a g i j k ___
4212 ; 5 8 8 i
4213
4214 cglobal ipred_filter_8bpc, 3, 7, 0, dst, stride, tl, w, h, filter
4215 %define base r6-ipred_filter_avx2_table
4216 lea r6, [filter_intra_taps]
4217 tzcnt wd, wm
4218 %ifidn filterd, filterm
4219 movzx filterd, filterb
4220 %else
4221 movzx filterd, byte filterm
4222 %endif
4223 shl filterd, 6
4224 add filterq, r6
4225 lea r6, [ipred_filter_avx2_table]
4226 movq xm0, [tlq-3] ; _ 6 5 0 1 2 3 4
4227 movsxd wq, [r6+wq*4]
4228 vpbroadcastd m1, [base+pw_8]
4229 vbroadcasti128 m2, [filterq+16*0]
4230 vbroadcasti128 m3, [filterq+16*1]
4231 vbroadcasti128 m4, [filterq+16*2]
4232 vbroadcasti128 m5, [filterq+16*3]
4233 add wq, r6
4234 mov hd, hm
4235 jmp wq
4236 .w4:
4237 WIN64_SPILL_XMM 9
4238 mova xm8, [base+filter_shuf2]
4239 sub tlq, 3
4240 sub tlq, hq
4241 jmp .w4_loop_start
4242 .w4_loop:
4243 pinsrd xm0, xm6, [tlq+hq], 0
4244 lea dstq, [dstq+strideq*2]
4245 .w4_loop_start:
4246 FILTER_XMM 6, 0, 7, 8
4247 movd [dstq+strideq*0], xm6
4248 pextrd [dstq+strideq*1], xm6, 1
4249 sub hd, 2
4250 jg .w4_loop
4251 RET
4252 ALIGN function_align
4253 .w8:
4254 %assign stack_offset stack_offset - stack_size_padded
4255 WIN64_SPILL_XMM 10
4256 mova m8, [base+filter_shuf1]
4257 FILTER_XMM 7, 0, 6, [base+filter_shuf2]
4258 vpbroadcastd m0, [tlq+4]
4259 vpbroadcastd m6, [tlq+5]
4260 sub tlq, 4
4261 sub tlq, hq
4262 vpbroadcastq m7, xm7
4263 vpblendd m7, m6, 0x20
4264 .w8_loop:
4265 vpbroadcastd xm6, [tlq+hq]
4266 palignr m6, m0, 12
4267 vpblendd m0, m6, m7, 0xeb ; _ _ _ _ 1 2 3 4 6 5 0 _ _ _ _ _
4268 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4269 mova xm6, xm7
4270 call .main
4271 vpblendd xm6, xm7, 0x0c
4272 pshufd xm6, xm6, q3120
4273 movq [dstq+strideq*0], xm6
4274 movhps [dstq+strideq*1], xm6
4275 lea dstq, [dstq+strideq*2]
4276 sub hd, 2
4277 jg .w8_loop
4278 RET
4279 ALIGN function_align
4280 .w16:
4281 %if WIN64
4282 %assign stack_offset stack_offset - stack_size_padded
4283 %assign xmm_regs_used 15
4284 %assign stack_size_padded 0x98
4285 SUB rsp, stack_size_padded
4286 %endif
4287 sub hd, 2
4288 TAIL_CALL .w16_main, 0
4289 .w16_main:
4290 %if WIN64
4291 movaps [rsp+0xa8], xmm6
4292 movaps [rsp+0xb8], xmm7
4293 movaps [rsp+0x28], xmm8
4294 movaps [rsp+0x38], xmm9
4295 movaps [rsp+0x48], xmm10
4296 movaps [rsp+0x58], xmm11
4297 movaps [rsp+0x68], xmm12
4298 movaps [rsp+0x78], xmm13
4299 movaps [rsp+0x88], xmm14
4300 %endif
4301 FILTER_XMM 12, 0, 7, [base+filter_shuf2]
4302 vpbroadcastd m0, [tlq+5]
4303 vpblendd m0, [tlq-12], 0x14
4304 mova m8, [base+filter_shuf1]
4305 vpbroadcastq m7, xm12
4306 vpblendd m0, m7, 0xc2 ; _ _ _ _ 1 2 3 4 6 5 0 _ _ _ _ _
4307 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4308 call .main ; c0 d0 a1 b1 a1 b1 c0 d0
4309 movlps xm9, xm7, [tlq+5] ; _ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
4310 vinserti128 m14, m8, [base+filter_shuf3], 0
4311 vpblendd xm12, xm7, 0x0c ; a0 b0 a1 b1
4312 FILTER_XMM 6, 9, 10, 14
4313 vpbroadcastq m6, xm6 ; a2 b2 __ __ __ __ a2 b2
4314 vpbroadcastd m9, [tlq+13]
4315 vpbroadcastd m10, [tlq+12]
4316 psrld m11, m8, 4
4317 vpblendd m6, m9, 0x20 ; top
4318 sub tlq, 6
4319 sub tlq, hq
4320 .w16_loop:
4321 vpbroadcastd xm9, [tlq+hq]
4322 palignr m9, m0, 12
4323 vpblendd m0, m9, m7, 0xe2 ; _ _ _ _ 1 2 3 4 6 5 0 _ _ _ _ _
4324 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4325 mova xm13, xm7
4326 call .main ; e0 f0 c1 d1 c1 d1 e0 f0
4327 vpblendd m9, m12, m10, 0xf0
4328 vpblendd m12, m6, 0xc0
4329 pshufd m9, m9, q3333
4330 vpblendd m9, m6, 0xee
4331 vpblendd m10, m9, m7, 0x0c ; _ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
4332 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4333 FILTER_YMM 6, 10, 9, 14 ; c2 d2 a3 b3 a3 b3 c2 d2
4334 vpblendd m12, m6, 0x30 ; a0 b0 a1 b1 a3 b3 a2 b2
4335 vpermd m9, m11, m12 ; a0 a1 a2 a3 b0 b1 b2 b3
4336 vpblendd xm12, xm13, xm7, 0x0c ; c0 d0 c1 d1
4337 mova [dstq+strideq*0], xm9
4338 vextracti128 [dstq+strideq*1], m9, 1
4339 lea dstq, [dstq+strideq*2]
4340 sub hd, 2
4341 jg .w16_loop
4342 vpblendd xm7, xm6, xm10, 0x04 ; _ _ _ 5 _ _ _ 6 0 _ _ _ 1 2 3 4
4343 pshufd xm7, xm7, q1032 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4344 FILTER_XMM 0, 7, 9, [base+filter_shuf1+16]
4345 vpblendd xm6, xm0, 0x0c ; c2 d2 c3 d3
4346 shufps xm0, xm12, xm6, q2020 ; c0 c1 c2 c3
4347 shufps xm6, xm12, xm6, q3131 ; d0 d1 d2 d3
4348 mova [dstq+strideq*0], xm0
4349 mova [dstq+strideq*1], xm6
4350 ret
4351 ALIGN function_align
4352 .w32:
4353 sub rsp, stack_size_padded
4354 sub hd, 2
4355 lea r3, [dstq+16]
4356 lea r5d, [hq-2]
4357 call .w16_main
4358 add tlq, r5
4359 mov dstq, r3
4360 lea r3, [strideq-4]
4361 lea r4, [r3+strideq*2]
4362 movq xm0, [tlq+21]
4363 pinsrd xm0, [dstq-4], 2
4364 pinsrd xm0, [dstq+r3*1], 3
4365 FILTER_XMM 12, 0, 7, 14 ; a0 b0 a0 b0
4366 movq xm7, [dstq+r3*2]
4367 pinsrd xm7, [dstq+r4], 2
4368 palignr xm7, xm0, 12 ; 0 _ _ _ _ _ _ _ _ _ _ 5 _ _ _ 6
4369 vpbroadcastd m0, [tlq+28]
4370 vpbroadcastd m9, [tlq+29]
4371 vbroadcasti128 m8, [base+filter_shuf1+16]
4372 vpblendd m0, m9, 0x20
4373 vpblendd m0, m7, 0x0f
4374 vpbroadcastq m7, xm12
4375 vpblendd m0, m7, 0xc2 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4376 call .main ; c0 d0 a1 b1 a1 b1 c0 d0
4377 add r3, 2
4378 lea r4, [r4+strideq*2]
4379 movlps xm9, xm7, [tlq+29] ; _ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
4380 vpblendd xm12, xm7, 0x0c ; a0 b0 a1 b1
4381 FILTER_XMM 6, 9, 10, 14
4382 vpbroadcastq m6, xm6 ; a2 b2 __ __ __ __ a2 b2
4383 vpbroadcastd m9, [tlq+37]
4384 vpbroadcastd m10, [tlq+36]
4385 vpblendd m6, m9, 0x20 ; top
4386 .w32_loop:
4387 movq xm9, [dstq+r3*4]
4388 pinsrd xm9, [dstq+r4], 2
4389 .w32_loop_last:
4390 palignr m9, m0, 12
4391 vpblendd m0, m9, m7, 0xe2 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4392 mova xm13, xm7 ; c0 d0
4393 call .main ; e0 f0 c1 d1 c1 d1 e0 f0
4394 vpblendd m9, m12, m10, 0xf0
4395 vpblendd m12, m6, 0xc0
4396 pshufd m9, m9, q3333
4397 vpblendd m9, m6, 0xee
4398 vpblendd m10, m9, m7, 0x0c ; _ _ _ 0 1 2 3 4 _ _ _ 5 _ _ _ 6
4399 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4400 FILTER_YMM 6, 10, 9, 14 ; c2 d2 a3 b3 a3 b3 c2 d2
4401 vpblendd m12, m6, 0x30 ; a0 b0 a1 b1 a3 b3 a2 b2
4402 vpermd m9, m11, m12 ; a0 a1 a2 a3 b0 b1 b2 b3
4403 vpblendd xm12, xm13, xm7, 0x0c ; c0 d0 c1 d1
4404 mova [dstq+strideq*0], xm9
4405 vextracti128 [dstq+strideq*1], m9, 1
4406 lea dstq, [dstq+strideq*2]
4407 sub r5d, 2
4408 jg .w32_loop
4409 jz .w32_loop_last
4410 vpblendd xm7, xm6, xm10, 0x04 ; _ _ _ 5 _ _ _ 6 0 _ _ _ 1 2 3 4
4411 pshufd xm7, xm7, q1032 ; 0 _ _ _ 1 2 3 4 _ _ _ 5 _ _ _ 6
4412 FILTER_XMM 0, 7, 9, [base+filter_shuf1+16]
4413 vpblendd xm6, xm0, 0x0c ; c2 d2 c3 d3
4414 shufps xm0, xm12, xm6, q2020 ; c0 c1 c2 c3
4415 shufps xm6, xm12, xm6, q3131 ; d0 d1 d2 d3
4416 mova [dstq+strideq*0], xm0
4417 mova [dstq+strideq*1], xm6
4418 RET
4419 ALIGN function_align
4420 .main:
4421 FILTER_YMM 7, 0, 9, 8
4422 ret
4423
4424 %if WIN64
4425 DECLARE_REG_TMP 5
4426 %else
4427 DECLARE_REG_TMP 7
4428 %endif
4429
4430 %macro IPRED_CFL 1 ; ac in, unpacked pixels out
4431 psignw m3, m%1, m1
4432 pabsw m%1, m%1
4433 pmulhrsw m%1, m2
4434 psignw m%1, m3
4435 paddw m%1, m0
4436 %endmacro
4437
4438 cglobal ipred_cfl_top_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
4439 lea t0, [ipred_cfl_left_avx2_table]
4440 tzcnt wd, wm
4441 inc tlq
4442 movu m0, [tlq]
4443 movifnidn hd, hm
4444 mov r6d, 0x8000
4445 shrx r6d, r6d, wd
4446 movd xm3, r6d
4447 movsxd r6, [t0+wq*4]
4448 pcmpeqd m2, m2
4449 pmaddubsw m0, m2
4450 add r6, t0
4451 add t0, ipred_cfl_splat_avx2_table-ipred_cfl_left_avx2_table
4452 movsxd wq, [t0+wq*4]
4453 add wq, t0
4454 movifnidn acq, acmp
4455 jmp r6
4456
4457 cglobal ipred_cfl_left_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
4458 mov hd, hm ; zero upper half
4459 tzcnt r6d, hd
4460 sub tlq, hq
4461 tzcnt wd, wm
4462 movu m0, [tlq]
4463 mov t0d, 0x8000
4464 shrx t0d, t0d, r6d
4465 movd xm3, t0d
4466 lea t0, [ipred_cfl_left_avx2_table]
4467 movsxd r6, [t0+r6*4]
4468 pcmpeqd m2, m2
4469 pmaddubsw m0, m2
4470 add r6, t0
4471 add t0, ipred_cfl_splat_avx2_table-ipred_cfl_left_avx2_table
4472 movsxd wq, [t0+wq*4]
4473 add wq, t0
4474 movifnidn acq, acmp
4475 jmp r6
4476 .h32:
4477 vextracti128 xm1, m0, 1
4478 paddw xm0, xm1
4479 .h16:
4480 punpckhqdq xm1, xm0, xm0
4481 paddw xm0, xm1
4482 .h8:
4483 psrlq xm1, xm0, 32
4484 paddw xm0, xm1
4485 .h4:
4486 pmaddwd xm0, xm2
4487 pmulhrsw xm0, xm3
4488 vpbroadcastw m0, xm0
4489 jmp wq
4490
4491 cglobal ipred_cfl_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
4492 movifnidn hd, hm
4493 movifnidn wd, wm
4494 tzcnt r6d, hd
4495 lea t0d, [wq+hq]
4496 movd xm4, t0d
4497 tzcnt t0d, t0d
4498 movd xm5, t0d
4499 lea t0, [ipred_cfl_avx2_table]
4500 tzcnt wd, wd
4501 movsxd r6, [t0+r6*4]
4502 movsxd wq, [t0+wq*4+4*4]
4503 pcmpeqd m3, m3
4504 psrlw xm4, 1
4505 add r6, t0
4506 add wq, t0
4507 movifnidn acq, acmp
4508 jmp r6
4509 .h4:
4510 movd xm0, [tlq-4]
4511 pmaddubsw xm0, xm3
4512 jmp wq
4513 .w4:
4514 movd xm1, [tlq+1]
4515 pmaddubsw xm1, xm3
4516 psubw xm0, xm4
4517 paddw xm0, xm1
4518 pmaddwd xm0, xm3
4519 cmp hd, 4
4520 jg .w4_mul
4521 psrlw xm0, 3
4522 jmp .w4_end
4523 .w4_mul:
4524 punpckhqdq xm1, xm0, xm0
4525 lea r2d, [hq*2]
4526 mov r6d, 0x55563334
4527 paddw xm0, xm1
4528 shrx r6d, r6d, r2d
4529 psrlq xm1, xm0, 32
4530 paddw xm0, xm1
4531 movd xm1, r6d
4532 psrlw xm0, 2
4533 pmulhuw xm0, xm1
4534 .w4_end:
4535 vpbroadcastw m0, xm0
4536 .s4:
4537 vpbroadcastw m1, alpham
4538 lea r6, [strideq*3]
4539 pabsw m2, m1
4540 psllw m2, 9
4541 .s4_loop:
4542 mova m4, [acq]
4543 IPRED_CFL 4
4544 packuswb m4, m4
4545 vextracti128 xm5, m4, 1
4546 movd [dstq+strideq*0], xm4
4547 pextrd [dstq+strideq*1], xm4, 1
4548 movd [dstq+strideq*2], xm5
4549 pextrd [dstq+r6 ], xm5, 1
4550 lea dstq, [dstq+strideq*4]
4551 add acq, 32
4552 sub hd, 4
4553 jg .s4_loop
4554 RET
4555 ALIGN function_align
4556 .h8:
4557 movq xm0, [tlq-8]
4558 pmaddubsw xm0, xm3
4559 jmp wq
4560 .w8:
4561 movq xm1, [tlq+1]
4562 vextracti128 xm2, m0, 1
4563 pmaddubsw xm1, xm3
4564 psubw xm0, xm4
4565 paddw xm0, xm2
4566 punpckhqdq xm2, xm0, xm0
4567 paddw xm0, xm2
4568 paddw xm0, xm1
4569 psrlq xm1, xm0, 32
4570 paddw xm0, xm1
4571 pmaddwd xm0, xm3
4572 psrlw xm0, xm5
4573 cmp hd, 8
4574 je .w8_end
4575 mov r6d, 0x5556
4576 mov r2d, 0x3334
4577 cmp hd, 32
4578 cmove r6d, r2d
4579 movd xm1, r6d
4580 pmulhuw xm0, xm1
4581 .w8_end:
4582 vpbroadcastw m0, xm0
4583 .s8:
4584 vpbroadcastw m1, alpham
4585 lea r6, [strideq*3]
4586 pabsw m2, m1
4587 psllw m2, 9
4588 .s8_loop:
4589 mova m4, [acq]
4590 mova m5, [acq+32]
4591 IPRED_CFL 4
4592 IPRED_CFL 5
4593 packuswb m4, m5
4594 vextracti128 xm5, m4, 1
4595 movq [dstq+strideq*0], xm4
4596 movq [dstq+strideq*1], xm5
4597 movhps [dstq+strideq*2], xm4
4598 movhps [dstq+r6 ], xm5
4599 lea dstq, [dstq+strideq*4]
4600 add acq, 64
4601 sub hd, 4
4602 jg .s8_loop
4603 RET
4604 ALIGN function_align
4605 .h16:
4606 mova xm0, [tlq-16]
4607 pmaddubsw xm0, xm3
4608 jmp wq
4609 .w16:
4610 movu xm1, [tlq+1]
4611 vextracti128 xm2, m0, 1
4612 pmaddubsw xm1, xm3
4613 psubw xm0, xm4
4614 paddw xm0, xm2
4615 paddw xm0, xm1
4616 punpckhqdq xm1, xm0, xm0
4617 paddw xm0, xm1
4618 psrlq xm1, xm0, 32
4619 paddw xm0, xm1
4620 pmaddwd xm0, xm3
4621 psrlw xm0, xm5
4622 cmp hd, 16
4623 je .w16_end
4624 mov r6d, 0x5556
4625 mov r2d, 0x3334
4626 test hb, 8|32
4627 cmovz r6d, r2d
4628 movd xm1, r6d
4629 pmulhuw xm0, xm1
4630 .w16_end:
4631 vpbroadcastw m0, xm0
4632 .s16:
4633 vpbroadcastw m1, alpham
4634 pabsw m2, m1
4635 psllw m2, 9
4636 .s16_loop:
4637 mova m4, [acq]
4638 mova m5, [acq+32]
4639 IPRED_CFL 4
4640 IPRED_CFL 5
4641 packuswb m4, m5
4642 vpermq m4, m4, q3120
4643 mova [dstq+strideq*0], xm4
4644 vextracti128 [dstq+strideq*1], m4, 1
4645 lea dstq, [dstq+strideq*2]
4646 add acq, 64
4647 sub hd, 2
4648 jg .s16_loop
4649 RET
4650 ALIGN function_align
4651 .h32:
4652 mova m0, [tlq-32]
4653 pmaddubsw m0, m3
4654 jmp wq
4655 .w32:
4656 movu m1, [tlq+1]
4657 pmaddubsw m1, m3
4658 paddw m0, m1
4659 vextracti128 xm1, m0, 1
4660 psubw xm0, xm4
4661 paddw xm0, xm1
4662 punpckhqdq xm1, xm0, xm0
4663 paddw xm0, xm1
4664 psrlq xm1, xm0, 32
4665 paddw xm0, xm1
4666 pmaddwd xm0, xm3
4667 psrlw xm0, xm5
4668 cmp hd, 32
4669 je .w32_end
4670 lea r2d, [hq*2]
4671 mov r6d, 0x33345556
4672 shrx r6d, r6d, r2d
4673 movd xm1, r6d
4674 pmulhuw xm0, xm1
4675 .w32_end:
4676 vpbroadcastw m0, xm0
4677 .s32:
4678 vpbroadcastw m1, alpham
4679 pabsw m2, m1
4680 psllw m2, 9
4681 .s32_loop:
4682 mova m4, [acq]
4683 mova m5, [acq+32]
4684 IPRED_CFL 4
4685 IPRED_CFL 5
4686 packuswb m4, m5
4687 vpermq m4, m4, q3120
4688 mova [dstq], m4
4689 add dstq, strideq
4690 add acq, 64
4691 dec hd
4692 jg .s32_loop
4693 RET
4694
4695 cglobal ipred_cfl_128_8bpc, 3, 7, 6, dst, stride, tl, w, h, ac, alpha
4696 lea t0, [ipred_cfl_splat_avx2_table]
4697 tzcnt wd, wm
4698 movifnidn hd, hm
4699 movsxd wq, [t0+wq*4]
4700 vpbroadcastd m0, [t0-ipred_cfl_splat_avx2_table+pw_128]
4701 add wq, t0
4702 movifnidn acq, acmp
4703 jmp wq
4704
4705 cglobal ipred_cfl_ac_420_8bpc, 4, 9, 5, ac, y, stride, wpad, hpad, w, h, sz, ac_bak
4706 movifnidn hpadd, hpadm
4707 movifnidn wd, wm
4708 mov hd, hm
4709 mov szd, wd
4710 mov ac_bakq, acq
4711 imul szd, hd
4712 shl hpadd, 2
4713 sub hd, hpadd
4714 vpbroadcastd m2, [pb_2]
4715 pxor m4, m4
4716 cmp wd, 8
4717 jg .w16
4718 je .w8
4719 ; fall-through
4720
4721 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, sz, ac_bak
4722 .w4:
4723 lea stride3q, [strideq*3]
4724 .w4_loop:
4725 movq xm0, [yq]
4726 movq xm1, [yq+strideq]
4727 movhps xm0, [yq+strideq*2]
4728 movhps xm1, [yq+stride3q]
4729 pmaddubsw xm0, xm2
4730 pmaddubsw xm1, xm2
4731 paddw xm0, xm1
4732 mova [acq], xm0
4733 paddw xm4, xm0
4734 lea yq, [yq+strideq*4]
4735 add acq, 16
4736 sub hd, 2
4737 jg .w4_loop
4738 test hpadd, hpadd
4739 jz .calc_avg
4740 vpermq m0, m0, q1111
4741 .w4_hpad_loop:
4742 mova [acq], m0
4743 paddw m4, m0
4744 add acq, 32
4745 sub hpadd, 4
4746 jg .w4_hpad_loop
4747 jmp .calc_avg
4748
4749 .w8:
4750 lea stride3q, [strideq*3]
4751 test wpadd, wpadd
4752 jnz .w8_wpad
4753 .w8_loop:
4754 mova xm0, [yq]
4755 mova xm1, [yq+strideq]
4756 vinserti128 m0, [yq+strideq*2], 1
4757 vinserti128 m1, [yq+stride3q], 1
4758 pmaddubsw m0, m2
4759 pmaddubsw m1, m2
4760 paddw m0, m1
4761 mova [acq], m0
4762 paddw m4, m0
4763 lea yq, [yq+strideq*4]
4764 add acq, 32
4765 sub hd, 2
4766 jg .w8_loop
4767 test hpadd, hpadd
4768 jz .calc_avg
4769 jmp .w8_hpad
4770 .w8_wpad:
4771 vbroadcasti128 m3, [cfl_ac_w8_pad1_shuffle]
4772 .w8_wpad_loop:
4773 movq xm0, [yq]
4774 movq xm1, [yq+strideq]
4775 vinserti128 m0, [yq+strideq*2], 1
4776 vinserti128 m1, [yq+stride3q], 1
4777 pmaddubsw m0, m2
4778 pmaddubsw m1, m2
4779 paddw m0, m1
4780 pshufb m0, m3
4781 mova [acq], m0
4782 paddw m4, m0
4783 lea yq, [yq+strideq*4]
4784 add acq, 32
4785 sub hd, 2
4786 jg .w8_wpad_loop
4787 test hpadd, hpadd
4788 jz .calc_avg
4789 .w8_hpad:
4790 vpermq m0, m0, q3232
4791 .w8_hpad_loop:
4792 mova [acq], m0
4793 paddw m4, m0
4794 add acq, 32
4795 sub hpadd, 2
4796 jg .w8_hpad_loop
4797 jmp .calc_avg
4798
4799 .w16:
4800 test wpadd, wpadd
4801 jnz .w16_wpad
4802 .w16_loop:
4803 mova m0, [yq]
4804 mova m1, [yq+strideq]
4805 pmaddubsw m0, m2
4806 pmaddubsw m1, m2
4807 paddw m0, m1
4808 mova [acq], m0
4809 paddw m4, m0
4810 lea yq, [yq+strideq*2]
4811 add acq, 32
4812 dec hd
4813 jg .w16_loop
4814 test hpadd, hpadd
4815 jz .calc_avg
4816 jmp .w16_hpad_loop
4817 .w16_wpad:
4818 DEFINE_ARGS ac, y, stride, wpad, hpad, iptr, h, sz, ac_bak
4819 lea iptrq, [ipred_cfl_ac_420_avx2_table]
4820 shl wpadd, 2
4821 mova m3, [iptrq+cfl_ac_w16_pad_shuffle- \
4822 ipred_cfl_ac_420_avx2_table+wpadq*8-32]
4823 movsxd wpadq, [iptrq+wpadq+4]
4824 add iptrq, wpadq
4825 jmp iptrq
4826 .w16_pad3:
4827 vpbroadcastq m0, [yq]
4828 vpbroadcastq m1, [yq+strideq]
4829 jmp .w16_wpad_end
4830 .w16_pad2:
4831 vbroadcasti128 m0, [yq]
4832 vbroadcasti128 m1, [yq+strideq]
4833 jmp .w16_wpad_end
4834 .w16_pad1:
4835 mova m0, [yq]
4836 mova m1, [yq+strideq]
4837 ; fall-through
4838 .w16_wpad_end:
4839 pmaddubsw m0, m2
4840 pmaddubsw m1, m2
4841 paddw m0, m1
4842 pshufb m0, m3
4843 mova [acq], m0
4844 paddw m4, m0
4845 lea yq, [yq+strideq*2]
4846 add acq, 32
4847 dec hd
4848 jz .w16_wpad_done
4849 jmp iptrq
4850 .w16_wpad_done:
4851 test hpadd, hpadd
4852 jz .calc_avg
4853 .w16_hpad_loop:
4854 mova [acq], m0
4855 paddw m4, m0
4856 add acq, 32
4857 dec hpadd
4858 jg .w16_hpad_loop
4859 ; fall-through
4860
4861 .calc_avg:
4862 vpbroadcastd m2, [pw_1]
4863 pmaddwd m0, m4, m2
4864 vextracti128 xm1, m0, 1
4865 tzcnt r1d, szd
4866 paddd xm0, xm1
4867 movd xm2, r1d
4868 movd xm3, szd
4869 punpckhqdq xm1, xm0, xm0
4870 paddd xm0, xm1
4871 psrad xm3, 1
4872 psrlq xm1, xm0, 32
4873 paddd xm0, xm3
4874 paddd xm0, xm1
4875 psrad xm0, xm2
4876 vpbroadcastw m0, xm0
4877 .sub_loop:
4878 mova m1, [ac_bakq]
4879 psubw m1, m0
4880 mova [ac_bakq], m1
4881 add ac_bakq, 32
4882 sub szd, 16
4883 jg .sub_loop
4884 RET
4885
4886 cglobal ipred_cfl_ac_422_8bpc, 4, 9, 6, ac, y, stride, wpad, hpad, w, h, sz, ac_bak
4887 movifnidn hpadd, hpadm
4888 movifnidn wd, wm
4889 mov hd, hm
4890 mov szd, wd
4891 mov ac_bakq, acq
4892 imul szd, hd
4893 shl hpadd, 2
4894 sub hd, hpadd
4895 vpbroadcastd m2, [pb_4]
4896 pxor m4, m4
4897 pxor m5, m5
4898 cmp wd, 8
4899 jg .w16
4900 je .w8
4901 ; fall-through
4902
4903 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, sz, ac_bak
4904 .w4:
4905 lea stride3q, [strideq*3]
4906 .w4_loop:
4907 movq xm1, [yq]
4908 movhps xm1, [yq+strideq]
4909 movq xm0, [yq+strideq*2]
4910 movhps xm0, [yq+stride3q]
4911 pmaddubsw xm0, xm2
4912 pmaddubsw xm1, xm2
4913 mova [acq], xm1
4914 mova [acq+16], xm0
4915 paddw xm4, xm0
4916 paddw xm5, xm1
4917 lea yq, [yq+strideq*4]
4918 add acq, 32
4919 sub hd, 4
4920 jg .w4_loop
4921 test hpadd, hpadd
4922 jz .calc_avg
4923 vpermq m0, m0, q1111
4924 .w4_hpad_loop:
4925 mova [acq], m0
4926 paddw m4, m0
4927 add acq, 32
4928 sub hpadd, 4
4929 jg .w4_hpad_loop
4930 jmp .calc_avg
4931
4932 .w8:
4933 lea stride3q, [strideq*3]
4934 test wpadd, wpadd
4935 jnz .w8_wpad
4936 .w8_loop:
4937 mova xm1, [yq]
4938 vinserti128 m1, [yq+strideq], 1
4939 mova xm0, [yq+strideq*2]
4940 vinserti128 m0, [yq+stride3q], 1
4941 pmaddubsw m0, m2
4942 pmaddubsw m1, m2
4943 mova [acq], m1
4944 mova [acq+32], m0
4945 paddw m4, m0
4946 paddw m5, m1
4947 lea yq, [yq+strideq*4]
4948 add acq, 64
4949 sub hd, 4
4950 jg .w8_loop
4951 test hpadd, hpadd
4952 jz .calc_avg
4953 jmp .w8_hpad
4954 .w8_wpad:
4955 vbroadcasti128 m3, [cfl_ac_w8_pad1_shuffle]
4956 .w8_wpad_loop:
4957 movq xm1, [yq]
4958 vinserti128 m1, [yq+strideq], 1
4959 movq xm0, [yq+strideq*2]
4960 vinserti128 m0, [yq+stride3q], 1
4961 pmaddubsw m0, m2
4962 pmaddubsw m1, m2
4963 pshufb m0, m3
4964 pshufb m1, m3
4965 mova [acq], m1
4966 mova [acq+32], m0
4967 paddw m4, m0
4968 paddw m5, m1
4969 lea yq, [yq+strideq*4]
4970 add acq, 64
4971 sub hd, 4
4972 jg .w8_wpad_loop
4973 test hpadd, hpadd
4974 jz .calc_avg
4975 .w8_hpad:
4976 vpermq m0, m0, q3232
4977 .w8_hpad_loop:
4978 mova [acq], m0
4979 paddw m4, m0
4980 add acq, 32
4981 sub hpadd, 2
4982 jg .w8_hpad_loop
4983 jmp .calc_avg
4984
4985 .w16:
4986 test wpadd, wpadd
4987 jnz .w16_wpad
4988 .w16_loop:
4989 mova m1, [yq]
4990 mova m0, [yq+strideq]
4991 pmaddubsw m0, m2
4992 pmaddubsw m1, m2
4993 mova [acq], m1
4994 mova [acq+32], m0
4995 paddw m4, m0
4996 paddw m5, m1
4997 lea yq, [yq+strideq*2]
4998 add acq, 64
4999 sub hd, 2
5000 jg .w16_loop
5001 test hpadd, hpadd
5002 jz .calc_avg
5003 jmp .w16_hpad_loop
5004 .w16_wpad:
5005 DEFINE_ARGS ac, y, stride, wpad, hpad, iptr, h, sz, ac_bak
5006 lea iptrq, [ipred_cfl_ac_422_avx2_table]
5007 shl wpadd, 2
5008 mova m3, [iptrq+cfl_ac_w16_pad_shuffle- \
5009 ipred_cfl_ac_422_avx2_table+wpadq*8-32]
5010 movsxd wpadq, [iptrq+wpadq+4]
5011 add iptrq, wpadq
5012 jmp iptrq
5013 .w16_pad3:
5014 vpbroadcastq m1, [yq]
5015 vpbroadcastq m0, [yq+strideq]
5016 jmp .w16_wpad_end
5017 .w16_pad2:
5018 vbroadcasti128 m1, [yq]
5019 vbroadcasti128 m0, [yq+strideq]
5020 jmp .w16_wpad_end
5021 .w16_pad1:
5022 mova m1, [yq]
5023 mova m0, [yq+strideq]
5024 ; fall-through
5025 .w16_wpad_end:
5026 pmaddubsw m0, m2
5027 pmaddubsw m1, m2
5028 pshufb m0, m3
5029 pshufb m1, m3
5030 mova [acq], m1
5031 mova [acq+32], m0
5032 paddw m4, m0
5033 paddw m5, m1
5034 lea yq, [yq+strideq*2]
5035 add acq, 64
5036 sub hd, 2
5037 jz .w16_wpad_done
5038 jmp iptrq
5039 .w16_wpad_done:
5040 test hpadd, hpadd
5041 jz .calc_avg
5042 .w16_hpad_loop:
5043 mova [acq], m0
5044 mova [acq+32], m0
5045 paddw m4, m0
5046 paddw m5, m0
5047 add acq, 64
5048 sub hpadd, 2
5049 jg .w16_hpad_loop
5050 ; fall-through
5051
5052 .calc_avg:
5053 vpbroadcastd m2, [pw_1]
5054 pmaddwd m5, m5, m2
5055 pmaddwd m0, m4, m2
5056 paddd m0, m5
5057 vextracti128 xm1, m0, 1
5058 tzcnt r1d, szd
5059 paddd xm0, xm1
5060 movd xm2, r1d
5061 movd xm3, szd
5062 punpckhqdq xm1, xm0, xm0
5063 paddd xm0, xm1
5064 psrad xm3, 1
5065 psrlq xm1, xm0, 32
5066 paddd xm0, xm3
5067 paddd xm0, xm1
5068 psrad xm0, xm2
5069 vpbroadcastw m0, xm0
5070 .sub_loop:
5071 mova m1, [ac_bakq]
5072 psubw m1, m0
5073 mova [ac_bakq], m1
5074 add ac_bakq, 32
5075 sub szd, 16
5076 jg .sub_loop
5077 RET
5078
5079 cglobal ipred_cfl_ac_444_8bpc, 4, 9, 6, ac, y, stride, wpad, hpad, w, h, sz, ac_bak
5080 movifnidn hpadd, hpadm
5081 movifnidn wd, wm
5082 mov hd, hm
5083 mov szd, wd
5084 imul szd, hd
5085 shl hpadd, 2
5086 sub hd, hpadd
5087 pxor m4, m4
5088 vpbroadcastd m5, [pw_1]
5089 tzcnt r8d, wd
5090 lea r5, [ipred_cfl_ac_444_avx2_table]
5091 movsxd r8, [r5+r8*4+12]
5092 add r5, r8
5093
5094 DEFINE_ARGS ac, y, stride, wpad, hpad, stride3, h, sz, ac_bak
5095 mov ac_bakq, acq
5096 jmp r5
5097
5098 .w4:
5099 lea stride3q, [strideq*3]
5100 pxor xm2, xm2
5101 .w4_loop:
5102 movd xm1, [yq]
5103 movd xm0, [yq+strideq*2]
5104 pinsrd xm1, [yq+strideq], 1
5105 pinsrd xm0, [yq+stride3q], 1
5106 punpcklbw xm1, xm2
5107 punpcklbw xm0, xm2
5108 psllw xm1, 3
5109 psllw xm0, 3
5110 mova [acq], xm1
5111 mova [acq+16], xm0
5112 paddw xm1, xm0
5113 paddw xm4, xm1
5114 lea yq, [yq+strideq*4]
5115 add acq, 32
5116 sub hd, 4
5117 jg .w4_loop
5118 test hpadd, hpadd
5119 jz .calc_avg_mul
5120 pshufd xm0, xm0, q3232
5121 paddw xm1, xm0, xm0
5122 .w4_hpad_loop:
5123 mova [acq], xm0
5124 mova [acq+16], xm0
5125 paddw xm4, xm1
5126 add acq, 32
5127 sub hpadd, 4
5128 jg .w4_hpad_loop
5129 jmp .calc_avg_mul
5130
5131 .w8:
5132 lea stride3q, [strideq*3]
5133 pxor m2, m2
5134 .w8_loop:
5135 movq xm1, [yq]
5136 movq xm0, [yq+strideq*2]
5137 vinserti128 m1, [yq+strideq], 1
5138 vinserti128 m0, [yq+stride3q], 1
5139 punpcklbw m1, m2
5140 punpcklbw m0, m2
5141 psllw m1, 3
5142 psllw m0, 3
5143 mova [acq], m1
5144 mova [acq+32], m0
5145 paddw m1, m0
5146 paddw m4, m1
5147 lea yq, [yq+strideq*4]
5148 add acq, 64
5149 sub hd, 4
5150 jg .w8_loop
5151 test hpadd, hpadd
5152 jz .calc_avg_mul
5153 vpermq m0, m0, q3232
5154 paddw m1, m0, m0
5155 .w8_hpad_loop:
5156 mova [acq], m0
5157 mova [acq+32], m0
5158 paddw m4, m1
5159 add acq, 64
5160 sub hpadd, 4
5161 jg .w8_hpad_loop
5162 jmp .calc_avg_mul
5163
5164 .w16:
5165 test wpadd, wpadd
5166 jnz .w16_wpad
5167 .w16_loop:
5168 pmovzxbw m1, [yq]
5169 pmovzxbw m0, [yq+strideq]
5170 psllw m1, 3
5171 psllw m0, 3
5172 mova [acq], m1
5173 mova [acq+32], m0
5174 paddw m1, m0
5175 pmaddwd m1, m5
5176 paddd m4, m1
5177 lea yq, [yq+strideq*2]
5178 add acq, 64
5179 sub hd, 2
5180 jg .w16_loop
5181 test hpadd, hpadd
5182 jz .calc_avg
5183 jmp .w16_hpad
5184 .w16_wpad:
5185 mova m3, [cfl_ac_444_w16_pad1_shuffle]
5186 .w16_wpad_loop:
5187 vpbroadcastq m1, [yq]
5188 vpbroadcastq m0, [yq+strideq]
5189 pshufb m1, m3
5190 pshufb m0, m3
5191 psllw m1, 3
5192 psllw m0, 3
5193 mova [acq], m1
5194 mova [acq+32], m0
5195 paddw m1, m0
5196 pmaddwd m1, m5
5197 paddd m4, m1
5198 lea yq, [yq+strideq*2]
5199 add acq, 64
5200 sub hd, 2
5201 jg .w16_wpad_loop
5202 test hpadd, hpadd
5203 jz .calc_avg
5204 .w16_hpad:
5205 paddw m1, m0, m0
5206 pmaddwd m1, m5
5207 .w16_hpad_loop:
5208 mova [acq], m0
5209 mova [acq+32], m0
5210 paddd m4, m1
5211 add acq, 64
5212 sub hpadd, 2
5213 jg .w16_hpad_loop
5214 jmp .calc_avg
5215
5216 .w32:
5217 test wpadd, wpadd
5218 jnz .w32_wpad
5219 .w32_loop:
5220 pmovzxbw m1, [yq]
5221 pmovzxbw m0, [yq+16]
5222 psllw m1, 3
5223 psllw m0, 3
5224 mova [acq], m1
5225 mova [acq+32], m0
5226 paddw m2, m1, m0
5227 pmaddwd m2, m5
5228 paddd m4, m2
5229 add yq, strideq
5230 add acq, 64
5231 dec hd
5232 jg .w32_loop
5233 test hpadd, hpadd
5234 jz .calc_avg
5235 jmp .w32_hpad_loop
5236 .w32_wpad:
5237 DEFINE_ARGS ac, y, stride, wpad, hpad, iptr, h, sz, ac_bak
5238 lea iptrq, [ipred_cfl_ac_444_avx2_table]
5239 add wpadd, wpadd
5240 mova m3, [iptrq+cfl_ac_444_w16_pad1_shuffle-ipred_cfl_ac_444_avx2_table]
5241 movsxd wpadq, [iptrq+wpadq+4]
5242 add iptrq, wpadq
5243 jmp iptrq
5244 .w32_pad3:
5245 vpbroadcastq m1, [yq]
5246 pshufb m1, m3
5247 vpermq m0, m1, q3232
5248 jmp .w32_wpad_end
5249 .w32_pad2:
5250 pmovzxbw m1, [yq]
5251 pshufhw m0, m1, q3333
5252 vpermq m0, m0, q3333
5253 jmp .w32_wpad_end
5254 .w32_pad1:
5255 pmovzxbw m1, [yq]
5256 vpbroadcastq m0, [yq+16]
5257 pshufb m0, m3
5258 ; fall-through
5259 .w32_wpad_end:
5260 psllw m1, 3
5261 psllw m0, 3
5262 mova [acq], m1
5263 mova [acq+32], m0
5264 paddw m2, m1, m0
5265 pmaddwd m2, m5
5266 paddd m4, m2
5267 add yq, strideq
5268 add acq, 64
5269 dec hd
5270 jz .w32_wpad_done
5271 jmp iptrq
5272 .w32_wpad_done:
5273 test hpadd, hpadd
5274 jz .calc_avg
5275 .w32_hpad_loop:
5276 mova [acq], m1
5277 mova [acq+32], m0
5278 paddd m4, m2
5279 add acq, 64
5280 dec hpadd
5281 jg .w32_hpad_loop
5282 jmp .calc_avg
5283
5284 .calc_avg_mul:
5285 pmaddwd m4, m5
5286 .calc_avg:
5287 vextracti128 xm1, m4, 1
5288 tzcnt r1d, szd
5289 paddd xm0, xm4, xm1
5290 movd xm2, r1d
5291 movd xm3, szd
5292 punpckhqdq xm1, xm0, xm0
5293 paddd xm0, xm1
5294 psrad xm3, 1
5295 psrlq xm1, xm0, 32
5296 paddd xm0, xm3
5297 paddd xm0, xm1
5298 psrad xm0, xm2
5299 vpbroadcastw m0, xm0
5300 .sub_loop:
5301 mova m1, [ac_bakq]
5302 psubw m1, m0
5303 mova [ac_bakq], m1
5304 add ac_bakq, 32
5305 sub szd, 16
5306 jg .sub_loop
5307 RET
5308
5309 cglobal pal_pred_8bpc, 4, 6, 5, dst, stride, pal, idx, w, h
5310 vbroadcasti128 m4, [palq]
5311 lea r2, [pal_pred_avx2_table]
5312 tzcnt wd, wm
5313 movifnidn hd, hm
5314 movsxd wq, [r2+wq*4]
5315 packuswb m4, m4
5316 add wq, r2
5317 lea r2, [strideq*3]
5318 jmp wq
5319 .w4:
5320 pshufb xm0, xm4, [idxq]
5321 add idxq, 16
5322 movd [dstq+strideq*0], xm0
5323 pextrd [dstq+strideq*1], xm0, 1
5324 pextrd [dstq+strideq*2], xm0, 2
5325 pextrd [dstq+r2 ], xm0, 3
5326 lea dstq, [dstq+strideq*4]
5327 sub hd, 4
5328 jg .w4
5329 RET
5330 ALIGN function_align
5331 .w8:
5332 pshufb xm0, xm4, [idxq+16*0]
5333 pshufb xm1, xm4, [idxq+16*1]
5334 add idxq, 16*2
5335 movq [dstq+strideq*0], xm0
5336 movhps [dstq+strideq*1], xm0
5337 movq [dstq+strideq*2], xm1
5338 movhps [dstq+r2 ], xm1
5339 lea dstq, [dstq+strideq*4]
5340 sub hd, 4
5341 jg .w8
5342 RET
5343 ALIGN function_align
5344 .w16:
5345 pshufb m0, m4, [idxq+32*0]
5346 pshufb m1, m4, [idxq+32*1]
5347 add idxq, 32*2
5348 mova [dstq+strideq*0], xm0
5349 vextracti128 [dstq+strideq*1], m0, 1
5350 mova [dstq+strideq*2], xm1
5351 vextracti128 [dstq+r2 ], m1, 1
5352 lea dstq, [dstq+strideq*4]
5353 sub hd, 4
5354 jg .w16
5355 RET
5356 ALIGN function_align
5357 .w32:
5358 pshufb m0, m4, [idxq+32*0]
5359 pshufb m1, m4, [idxq+32*1]
5360 pshufb m2, m4, [idxq+32*2]
5361 pshufb m3, m4, [idxq+32*3]
5362 add idxq, 32*4
5363 mova [dstq+strideq*0], m0
5364 mova [dstq+strideq*1], m1
5365 mova [dstq+strideq*2], m2
5366 mova [dstq+r2 ], m3
5367 lea dstq, [dstq+strideq*4]
5368 sub hd, 4
5369 jg .w32
5370 RET
5371 ALIGN function_align
5372 .w64:
5373 pshufb m0, m4, [idxq+32*0]
5374 pshufb m1, m4, [idxq+32*1]
5375 pshufb m2, m4, [idxq+32*2]
5376 pshufb m3, m4, [idxq+32*3]
5377 add idxq, 32*4
5378 mova [dstq+strideq*0+32*0], m0
5379 mova [dstq+strideq*0+32*1], m1
5380 mova [dstq+strideq*1+32*0], m2
5381 mova [dstq+strideq*1+32*1], m3
5382 lea dstq, [dstq+strideq*2]
5383 sub hd, 2
5384 jg .w64
5385 RET
5386
5387 %endif