1 ;*****************************************************************************
2 ;* SSE2-optimized HEVC deblocking code
3 ;*****************************************************************************
4 ;* Copyright (C) 2013 VTT
6 ;* Authors: Seppo Tomperi <seppo.tomperi@vtt.fi>
8 ;* This file is part of FFmpeg.
10 ;* FFmpeg is free software; you can redistribute it and/or
11 ;* modify it under the terms of the GNU Lesser General Public
12 ;* License as published by the Free Software Foundation; either
13 ;* version 2.1 of the License, or (at your option) any later version.
15 ;* FFmpeg is distributed in the hope that it will be useful,
16 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
17 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 ;* Lesser General Public License for more details.
20 ;* You should have received a copy of the GNU Lesser General Public
21 ;* License along with FFmpeg; if not, write to the Free Software
22 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 ;******************************************************************************
25 %include "libavutil/x86/x86util.asm"
29 pw_pixel_max: times 8 dw ((1 << 10)-1)
40 ; expands to [base],...,[base+7*stride]
41 %define PASS8ROWS(base, base3, stride, stride3) \
42 [base], [base+stride], [base+stride*2], [base3], \
43 [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
45 ; in: 8 rows of 4 bytes in %4..%11
46 ; out: 4 rows of 8 words in m0..m3
47 %macro TRANSPOSE4x8B_LOAD 8
76 ; in: 4 rows of 8 words in m0..m3
77 ; out: 8 rows of 4 bytes in %1..%8
78 %macro TRANSPOSE8x4B_STORE 8
107 ; in: 8 rows of 4 words in %4..%11
108 ; out: 4 rows of 8 words in m0..m3
109 %macro TRANSPOSE4x8W_LOAD 8
130 punpckhqdq m1, m0, m4
132 punpckhqdq m3, m2, m6
137 ; in: 4 rows of 8 words in m0..m3
138 ; out: 8 rows of 4 words in %1..%8
139 %macro TRANSPOSE8x4W_STORE 8
140 pxor m5, m5; zeros reg
141 CLIPW m0, m5, [pw_pixel_max]
142 CLIPW m1, m5, [pw_pixel_max]
143 CLIPW m2, m5, [pw_pixel_max]
144 CLIPW m3, m5, [pw_pixel_max]
167 ; in: 8 rows of 8 bytes in %1..%8
168 ; out: 8 rows of 8 words in m0..m7
169 %macro TRANSPOSE8x8B_LOAD 8
190 punpckldq m1, m3, m9; 0, 1
191 punpckhdq m3, m9; 2, 3
193 punpckldq m5, m7, m4; 4, 5
194 punpckhdq m7, m4; 6, 7
198 punpcklbw m0, m1, m13; 0 in 16 bit
199 punpckhbw m1, m13; 1 in 16 bit
201 punpcklbw m2, m3, m13; 2
204 punpcklbw m4, m5, m13; 4
207 punpcklbw m6, m7, m13; 6
212 ; in: 8 rows of 8 words in m0..m8
213 ; out: 8 rows of 8 bytes in %1..%8
214 %macro TRANSPOSE8x8B_STORE 8
236 punpckhdq m10, m0, m4; 2, 3
237 punpckldq m0, m4; 0, 1
239 punpckldq m11, m8, m9; 4, 5
240 punpckhdq m8, m9; 6, 7
251 ; in: 8 rows of 8 words in %1..%8
252 ; out: 8 rows of 8 words in m0..m7
253 %macro TRANSPOSE8x8W_LOAD 8
262 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
265 ; in: 8 rows of 8 words in m0..m8
266 ; out: 8 rows of 8 words in %1..%8
267 %macro TRANSPOSE8x8W_STORE 8
268 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
271 CLIPW m0, m8, [pw_pixel_max]
272 CLIPW m1, m8, [pw_pixel_max]
273 CLIPW m2, m8, [pw_pixel_max]
274 CLIPW m3, m8, [pw_pixel_max]
275 CLIPW m4, m8, [pw_pixel_max]
276 CLIPW m5, m8, [pw_pixel_max]
277 CLIPW m6, m8, [pw_pixel_max]
278 CLIPW m7, m8, [pw_pixel_max]
296 pand %2, m11 ; and mask
297 pandn m10, m11, %1; and -mask
304 ; mask in %3, will be clobbered
305 %macro MASKED_COPY2 3
306 pand %2, %3 ; and mask
307 pandn %3, %1; and -mask
313 ; input in m0 ... m3 and tcs in r2. Output in m1 and m2
314 %macro CHROMA_DEBLOCK_BODY 1
315 psubw m4, m2, m1; q0 - p0
316 psubw m5, m0, m3; p1 - q1
323 movd m4, [tcq+4]; tc1
325 shufps m6, m4, 0; tc0, tc1
326 pmullw m4, m6, [pw_m1]; -tc0, -tc1
333 psllw m4, %1-8; << (BIT_DEPTH - 8)
334 psllw m6, %1-8; << (BIT_DEPTH - 8)
338 paddw m1, m5; p0 + delta0
339 psubw m2, m5; q0 - delta0
342 ; input in m0 ... m7, beta in r2 tcs in r3. Output in m1...m6
343 %macro LUMA_DEBLOCK_BODY 2
347 ABS1 m10, m11 ; 0dp0, 0dp3 , 1dp0, 1dp3
352 ABS1 m11, m13 ; 0dq0, 0dq3 , 1dq0, 1dq3
360 ;end beta calculations
362 paddw m9, m10, m11; 0d0, 0d3 , 1d0, 1d3
364 pshufhw m14, m9, 0x0f ;0b00001111; 0d3 0d3 0d0 0d0 in high
365 pshuflw m14, m14, 0x0f ;0b00001111; 1d3 1d3 1d0 1d0 in low
367 pshufhw m9, m9, 0xf0 ;0b11110000; 0d0 0d0 0d3 0d3
368 pshuflw m9, m9, 0xf0 ;0b11110000; 1d0 1d0 1d3 1d3
370 paddw m14, m9; 0d0+0d3, 1d0+1d3
373 pcmpgtw m15, m13, m14
374 movmskps r13, m15 ;filtering mask 0d0 + 0d3 < beta0 (bit 2 or 3) , 1d0 + 1d3 < beta1 (bit 0 or 1)
378 ;weak / strong decision compare to beta_2
379 psraw m15, m13, 2; beta >> 2
381 pcmpgtw m15, m8; (d0 << 1) < beta_2, (d3 << 1) < beta_2
383 ;end weak / strong decision
385 ; weak filter nd_p/q calculation
390 and r7, 0xffff; 1dp0 + 1dp3
393 and r8, 0xffff; 0dp0 + 0dp3
399 and r9, 0xffff; 1dq0 + 1dq3
402 and r10, 0xffff; 0dq0 + 0dq3
403 ; end calc for weak filter
415 pcmpeqd m11, [pd_1]; filtering mask
417 ;decide between strong and weak filtering
430 add r11d, r3d; tc0 + tc1
434 shufps m8, m9, 0; tc0, tc1
437 pavgw m8, m9; tc25 = ((tc * 5 + 1) >> 1)
438 ;end tc25 calculations
440 ;----beta_3 comparison-----
441 psubw m12, m0, m3; p3 - p0
442 ABS1 m12, m14; abs(p3 - p0)
444 psubw m15, m7, m4; q3 - q0
445 ABS1 m15, m14; abs(q3 - q0)
447 paddw m12, m15; abs(p3 - p0) + abs(q3 - q0)
449 pshufhw m12, m12, 0xf0 ;0b11110000;
450 pshuflw m12, m12, 0xf0 ;0b11110000;
452 psraw m13, 3; beta >> 3
455 and r14, r11; strong mask , beta_2 and beta_3 comparisons
456 ;----beta_3 comparison end-----
457 ;----tc25 comparison---
458 psubw m12, m3, m4; p0 - q0
459 ABS1 m12, m14; abs(p0 - q0)
461 pshufhw m12, m12, 0xf0 ;0b11110000;
462 pshuflw m12, m12, 0xf0 ;0b11110000;
464 pcmpgtw m8, m12; tc25 comparisons
466 and r14, r11; strong mask, beta_2, beta_3 and tc25 comparisons
467 ;----tc25 comparison end---
470 and r14, r11; strong mask, bits 2 and 0
472 pmullw m14, m9, [pw_m2]; -tc * 2
476 mov r11, r14; strong mask
478 movd m12, r14d; store to xmm for mask generation
481 movd m10, r11d; store to xmm for mask generation
482 or r14, r11; final strong mask, bits 1 and 0
486 pcmpeqd m10, [pd_1]; strong mask
488 mova m13, [pw_4]; 4 in every cell
489 pand m11, m10; combine filtering mask and strong mask
490 paddw m12, m2, m3; p1 + p0
491 paddw m12, m4; p1 + p0 + q0
493 paddw m12, m12; 2*p1 + 2*p0 + 2*q0
494 paddw m12, m1; p2 + 2*p1 + 2*p0 + 2*q0
495 paddw m12, m5; p2 + 2*p1 + 2*p0 + 2*q0 + q1
496 paddw m12, m13; p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4
497 psraw m12, 3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3)
498 psubw m12, m3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3) - p0
500 pminsw m12, m9; av_clip( , -2 * tc, 2 * tc)
503 paddw m15, m1, m10; p2 + p1 + p0 + q0
504 psrlw m13, 1; 2 in every cell
505 paddw m15, m13; p2 + p1 + p0 + q0 + 2
506 psraw m15, 2; (p2 + p1 + p0 + q0 + 2) >> 2
507 psubw m15, m2;((p2 + p1 + p0 + q0 + 2) >> 2) - p1
509 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
512 paddw m8, m1, m0; p3 + p2
513 paddw m8, m8; 2*p3 + 2*p2
514 paddw m8, m1; 2*p3 + 3*p2
515 paddw m8, m10; 2*p3 + 3*p2 + p1 + p0 + q0
517 paddw m8, m13; 2*p3 + 3*p2 + p1 + p0 + q0 + 4
518 psraw m8, 3; (2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3
519 psubw m8, m1; ((2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3) - p2
521 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
525 paddw m8, m3, m4; p0 + q0
526 paddw m8, m5; p0 + q0 + q1
527 paddw m8, m8; 2*p0 + 2*q0 + 2*q1
528 paddw m8, m2; p1 + 2*p0 + 2*q0 + 2*q1
529 paddw m8, m6; p1 + 2*p0 + 2*q0 + 2*q1 + q2
530 paddw m8, m13; p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4
531 psraw m8, 3; (p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4) >>3
534 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
538 paddw m15, m3, m4; p0 + q0
539 paddw m15, m5; p0 + q0 + q1
541 paddw m15, m6; p0 + q0 + q1 + q2
542 psrlw m13, 1; 2 in every cell
543 paddw m15, m13; p0 + q0 + q1 + q2 + 2
544 psraw m15, 2; (p0 + q0 + q1 + q2 + 2) >> 2
545 psubw m15, m5; ((p0 + q0 + q1 + q2 + 2) >> 2) - q1
547 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
550 paddw m13, m7; q3 + 2
551 paddw m13, m6; q3 + q2 + 2
552 paddw m13, m13; 2*q3 + 2*q2 + 4
553 paddw m13, m6; 2*q3 + 3*q2 + 4
554 paddw m13, m10; 2*q3 + 3*q2 + q1 + q0 + p0 + 4
555 psraw m13, 3; (2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3
556 psubw m13, m6; ((2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3) - q2
558 pminsw m13, m9; av_clip( , -2 * tc, 2 * tc)
567 not r14; strong mask -> weak mask
568 and r14, r13; final weak filtering mask, bits 0 and 1
571 ; weak filtering mask
578 pcmpeqd m11, [pd_1]; filtering mask
583 shr betaq, 3; ((beta + (beta >> 1)) >> 3))
586 psubw m12, m4, m3 ; q0 - p0
587 psllw m10, m12, 3; 8 * (q0 - p0)
588 paddw m12, m10 ; 9 * (q0 - p0)
590 psubw m10, m5, m2 ; q1 - p1
591 psllw m8, m10, 1; 2 * ( q1 - p1 )
592 paddw m10, m8; 3 * ( q1 - p1 )
593 psubw m12, m10; 9 * (q0 - p0) - 3 * ( q1 - p1 )
595 psraw m12, 4; >> 4 , delta0
596 PABSW m13, m12; abs(delta0)
599 psllw m10, m9, 2; 8 * tc
600 paddw m10, m9; 10 * tc
604 psraw m9, 1; tc * 2 -> tc
605 psraw m14, 1; -tc * 2 -> -tc
608 pminsw m12, m9; av_clip(delta0, -tc, tc)
610 psraw m9, 1; tc -> tc / 2
611 pmullw m14, m9, [pw_m1]; -tc / 2
613 pavgw m15, m1, m3; (p2 + p0 + 1) >> 1
614 psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1
615 paddw m15, m12; ((p2 + p0 + 1) >> 1) - p1 + delta0
616 psraw m15, 1; (((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1
618 pminsw m15, m9; av_clip(deltap1, -tc/2, tc/2)
625 movd m13, r7d; 1dp0 + 1dp3
626 movd m8, r8d; 0dp0 + 0dp3
632 ;end beta calculations
633 MASKED_COPY2 m2, m15, m8; write p1'
635 pavgw m8, m6, m4; (q2 + q0 + 1) >> 1
636 psubw m8, m5; ((q2 + q0 + 1) >> 1) - q1
637 psubw m8, m12; ((q2 + q0 + 1) >> 1) - q1 - delta0)
638 psraw m8, 1; ((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1
640 pminsw m8, m9; av_clip(deltaq1, -tc/2, tc/2)
647 shufps m13, m15, 0; dq0 + dq3
649 pcmpgtw m10, m13; compare to ((beta+(beta>>1))>>3)
651 MASKED_COPY2 m5, m8, m10; write q1'
653 paddw m15, m3, m12 ; p0 + delta0
656 psubw m8, m4, m12 ; q0 - delta0
661 ;-----------------------------------------------------------------------------
662 ; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q)
663 ;-----------------------------------------------------------------------------
664 cglobal hevc_v_loop_filter_chroma_8, 3, 5, 7, pix, stride, tc, pix0, r3stride
666 lea r3strideq, [3*strideq]
669 TRANSPOSE4x8B_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
670 CHROMA_DEBLOCK_BODY 8
671 TRANSPOSE8x4B_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
674 cglobal hevc_v_loop_filter_chroma_10, 3, 5, 7, pix, stride, tc, pix0, r3stride
676 lea r3strideq, [3*strideq]
679 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
680 CHROMA_DEBLOCK_BODY 10
681 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
684 ;-----------------------------------------------------------------------------
685 ; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q
686 ;-----------------------------------------------------------------------------
687 cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
692 movq m1, [pix0q+strideq]; p0
694 movq m3, [pixq+strideq]; q1
695 pxor m5, m5; zeros reg
700 CHROMA_DEBLOCK_BODY 8
702 movh[pix0q+strideq], m1
706 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
711 movu m1, [pix0q+strideq]; p0
713 movu m3, [pixq+strideq]; q1
714 CHROMA_DEBLOCK_BODY 10
715 pxor m5, m5; zeros reg
716 CLIPW m1, m5, [pw_pixel_max]
717 CLIPW m2, m5, [pw_pixel_max]
718 movu [pix0q+strideq], m1
723 %macro LOOP_FILTER_LUMA 0
724 ;-----------------------------------------------------------------------------
725 ; void ff_hevc_v_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int *_beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
726 ;-----------------------------------------------------------------------------
727 cglobal hevc_v_loop_filter_luma_8, 4, 15, 16, pix, stride, beta, tc
732 TRANSPOSE8x8B_LOAD PASS8ROWS(r6, r0, r1, r5)
733 LUMA_DEBLOCK_BODY 8, v
735 TRANSPOSE8x8B_STORE PASS8ROWS(r6, r0, r1, r5)
739 cglobal hevc_v_loop_filter_luma_10, 4, 15, 16, pix, stride, beta, tc
741 lea r5, [3 * strideq]
744 TRANSPOSE8x8W_LOAD PASS8ROWS(r6, pixq, strideq, r5)
745 LUMA_DEBLOCK_BODY 10, v
747 TRANSPOSE8x8W_STORE PASS8ROWS(r6, r0, r1, r5)
751 ;-----------------------------------------------------------------------------
752 ; void ff_hevc_h_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int *_beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
753 ;-----------------------------------------------------------------------------
754 cglobal hevc_h_loop_filter_luma_8, 4, 15, 16, pix, stride, beta, tc, count, pix0, src3stride
755 lea src3strideq, [3 * strideq]
757 sub pix0q, src3strideq
759 movdqu m0, [pix0q]; p3
760 movdqu m1, [pix0q + strideq]; p2
761 movdqu m2, [pix0q + 2 * strideq]; p1
762 movdqu m3, [pix0q + src3strideq]; p0
763 movdqu m4, [pixq]; q0
764 movdqu m5, [pixq + strideq]; q1
765 movdqu m6, [pixq + 2 * strideq]; q2
766 movdqu m7, [pixq + src3strideq]; q3
776 LUMA_DEBLOCK_BODY 8, h
782 movhps [r5 + 2 * r1], m1
786 movhps [r0 + 2 * r1], m5
790 cglobal hevc_h_loop_filter_luma_10, 4, 15, 16, pix, stride, beta, tc, count, pix0, src3stride
791 lea src3strideq, [3 * strideq]
793 sub pix0q, src3strideq
795 movdqu m0, [pix0q]; p3
796 movdqu m1, [pix0q + strideq]; p2
797 movdqu m2, [pix0q + 2 * strideq]; p1
798 movdqu m3, [pix0q + src3strideq]; p0
799 movdqu m4, [pixq]; q0
800 movdqu m5, [pixq + strideq]; q1
801 movdqu m6, [pixq + 2 * strideq]; q2
802 movdqu m7, [pixq + src3strideq]; q3
803 LUMA_DEBLOCK_BODY 10, h
805 pxor m8, m8; zeros reg
806 CLIPW m1, m8, [pw_pixel_max]
807 CLIPW m2, m8, [pw_pixel_max]
808 CLIPW m3, m8, [pw_pixel_max]
809 CLIPW m4, m8, [pw_pixel_max]
810 CLIPW m5, m8, [pw_pixel_max]
811 CLIPW m6, m8, [pw_pixel_max]
812 movdqu [pix0q + strideq], m1; p2
813 movdqu [pix0q + 2 * strideq], m2; p1
814 movdqu [pix0q + src3strideq], m3; p0
815 movdqu [pixq ], m4; q0
816 movdqu [pixq + strideq], m5; q1
817 movdqu [pixq + 2 * strideq], m6; q2