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_12: times 8 dw ((1 << 12)-1)
30 pw_pixel_max_10: times 8 dw ((1 << 10)-1)
41 ; expands to [base],...,[base+7*stride]
42 %define PASS8ROWS(base, base3, stride, stride3) \
43 [base], [base+stride], [base+stride*2], [base3], \
44 [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
46 ; in: 8 rows of 4 bytes in %4..%11
47 ; out: 4 rows of 8 words in m0..m3
48 %macro TRANSPOSE4x8B_LOAD 8
77 ; in: 4 rows of 8 words in m0..m3
78 ; out: 8 rows of 4 bytes in %1..%8
79 %macro TRANSPOSE8x4B_STORE 8
82 SBUTTERFLY bw, 0, 1, 2
83 SBUTTERFLY wd, 0, 1, 2
102 ; in: 8 rows of 4 words in %4..%11
103 ; out: 4 rows of 8 words in m0..m3
104 %macro TRANSPOSE4x8W_LOAD 8
125 punpckhqdq m1, m0, m4
127 punpckhqdq m3, m2, m6
132 ; in: 4 rows of 8 words in m0..m3
133 ; out: 8 rows of 4 words in %1..%8
134 %macro TRANSPOSE8x4W_STORE 9
135 TRANSPOSE4x4W 0, 1, 2, 3, 4
137 pxor m5, m5; zeros reg
153 ; in: 8 rows of 8 bytes in %1..%8
154 ; out: 8 rows of 8 words in m0..m7
155 %macro TRANSPOSE8x8B_LOAD 8
176 punpckldq m1, m3, m9; 0, 1
177 punpckhdq m3, m9; 2, 3
179 punpckldq m5, m7, m4; 4, 5
180 punpckhdq m7, m4; 6, 7
184 punpcklbw m0, m1, m13; 0 in 16 bit
185 punpckhbw m1, m13; 1 in 16 bit
187 punpcklbw m2, m3, m13; 2
190 punpcklbw m4, m5, m13; 4
193 punpcklbw m6, m7, m13; 6
198 ; in: 8 rows of 8 words in m0..m8
199 ; out: 8 rows of 8 bytes in %1..%8
200 %macro TRANSPOSE8x8B_STORE 8
205 TRANSPOSE2x4x4B 0, 1, 2, 3, 4
217 ; in: 8 rows of 8 words in %1..%8
218 ; out: 8 rows of 8 words in m0..m7
219 %macro TRANSPOSE8x8W_LOAD 8
228 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
231 ; in: 8 rows of 8 words in m0..m8
232 ; out: 8 rows of 8 words in %1..%8
233 %macro TRANSPOSE8x8W_STORE 9
234 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
262 pand %2, m11 ; and mask
263 pandn m10, m11, %1; and -mask
270 ; mask in %3, will be clobbered
271 %macro MASKED_COPY2 3
272 pand %2, %3 ; and mask
273 pandn %3, %1; and -mask
279 ; input in m0 ... m3 and tcs in r2. Output in m1 and m2
280 %macro CHROMA_DEBLOCK_BODY 1
281 psubw m4, m2, m1; q0 - p0
282 psubw m5, m0, m3; p1 - q1
289 pshufd m6, m6, 0xA0; tc0, tc1
291 psignw m4, m6, [pw_m1]; -tc0, -tc1
293 pmullw m4, m6, [pw_m1]; -tc0, -tc1
301 psllw m4, %1-8; << (BIT_DEPTH - 8)
302 psllw m6, %1-8; << (BIT_DEPTH - 8)
306 paddw m1, m5; p0 + delta0
307 psubw m2, m5; q0 - delta0
310 ; input in m0 ... m7, beta in r2 tcs in r3. Output in m1...m6
311 %macro LUMA_DEBLOCK_BODY 2
315 ABS1 m10, m11 ; 0dp0, 0dp3 , 1dp0, 1dp3
320 ABS1 m11, m13 ; 0dq0, 0dq3 , 1dq0, 1dq3
328 ;end beta calculations
330 paddw m9, m10, m11; 0d0, 0d3 , 1d0, 1d3
332 pshufhw m14, m9, 0x0f ;0b00001111; 0d3 0d3 0d0 0d0 in high
333 pshuflw m14, m14, 0x0f ;0b00001111; 1d3 1d3 1d0 1d0 in low
335 pshufhw m9, m9, 0xf0 ;0b11110000; 0d0 0d0 0d3 0d3
336 pshuflw m9, m9, 0xf0 ;0b11110000; 1d0 1d0 1d3 1d3
338 paddw m14, m9; 0d0+0d3, 1d0+1d3
341 pcmpgtw m15, m13, m14
342 movmskps r13, m15 ;filtering mask 0d0 + 0d3 < beta0 (bit 2 or 3) , 1d0 + 1d3 < beta1 (bit 0 or 1)
346 ;weak / strong decision compare to beta_2
347 psraw m15, m13, 2; beta >> 2
349 pcmpgtw m15, m8; (d0 << 1) < beta_2, (d3 << 1) < beta_2
351 ;end weak / strong decision
353 ; weak filter nd_p/q calculation
367 ; end calc for weak filter
379 pcmpeqd m11, [pd_1]; filtering mask
381 ;decide between strong and weak filtering
392 add r11d, r3d; tc0 + tc1
397 shufps m8, m9, 0; tc0, tc1
400 pavgw m8, m9; tc25 = ((tc * 5 + 1) >> 1)
401 ;end tc25 calculations
403 ;----beta_3 comparison-----
404 psubw m12, m0, m3; p3 - p0
405 ABS1 m12, m14; abs(p3 - p0)
407 psubw m15, m7, m4; q3 - q0
408 ABS1 m15, m14; abs(q3 - q0)
410 paddw m12, m15; abs(p3 - p0) + abs(q3 - q0)
412 pshufhw m12, m12, 0xf0 ;0b11110000;
413 pshuflw m12, m12, 0xf0 ;0b11110000;
415 psraw m13, 3; beta >> 3
418 and r6, r11; strong mask , beta_2 and beta_3 comparisons
419 ;----beta_3 comparison end-----
420 ;----tc25 comparison---
421 psubw m12, m3, m4; p0 - q0
422 ABS1 m12, m14; abs(p0 - q0)
424 pshufhw m12, m12, 0xf0 ;0b11110000;
425 pshuflw m12, m12, 0xf0 ;0b11110000;
427 pcmpgtw m8, m12; tc25 comparisons
429 and r6, r11; strong mask, beta_2, beta_3 and tc25 comparisons
430 ;----tc25 comparison end---
433 and r6, r11; strong mask, bits 2 and 0
435 pmullw m14, m9, [pw_m2]; -tc * 2
439 mov r11, r6; strong mask
441 movd m12, r6d; store to xmm for mask generation
444 movd m10, r11d; store to xmm for mask generation
445 or r6, r11; final strong mask, bits 1 and 0
449 pcmpeqd m10, [pd_1]; strong mask
451 mova m13, [pw_4]; 4 in every cell
452 pand m11, m10; combine filtering mask and strong mask
453 paddw m12, m2, m3; p1 + p0
454 paddw m12, m4; p1 + p0 + q0
456 paddw m12, m12; 2*p1 + 2*p0 + 2*q0
457 paddw m12, m1; p2 + 2*p1 + 2*p0 + 2*q0
458 paddw m12, m5; p2 + 2*p1 + 2*p0 + 2*q0 + q1
459 paddw m12, m13; p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4
460 psraw m12, 3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3)
461 psubw m12, m3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3) - p0
463 pminsw m12, m9; av_clip( , -2 * tc, 2 * tc)
466 paddw m15, m1, m10; p2 + p1 + p0 + q0
467 psrlw m13, 1; 2 in every cell
468 paddw m15, m13; p2 + p1 + p0 + q0 + 2
469 psraw m15, 2; (p2 + p1 + p0 + q0 + 2) >> 2
470 psubw m15, m2;((p2 + p1 + p0 + q0 + 2) >> 2) - p1
472 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
475 paddw m8, m1, m0; p3 + p2
476 paddw m8, m8; 2*p3 + 2*p2
477 paddw m8, m1; 2*p3 + 3*p2
478 paddw m8, m10; 2*p3 + 3*p2 + p1 + p0 + q0
480 paddw m8, m13; 2*p3 + 3*p2 + p1 + p0 + q0 + 4
481 psraw m8, 3; (2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3
482 psubw m8, m1; ((2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3) - p2
484 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
488 paddw m8, m3, m4; p0 + q0
489 paddw m8, m5; p0 + q0 + q1
490 paddw m8, m8; 2*p0 + 2*q0 + 2*q1
491 paddw m8, m2; p1 + 2*p0 + 2*q0 + 2*q1
492 paddw m8, m6; p1 + 2*p0 + 2*q0 + 2*q1 + q2
493 paddw m8, m13; p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4
494 psraw m8, 3; (p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4) >>3
497 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
501 paddw m15, m3, m4; p0 + q0
502 paddw m15, m5; p0 + q0 + q1
504 paddw m15, m6; p0 + q0 + q1 + q2
505 psrlw m13, 1; 2 in every cell
506 paddw m15, m13; p0 + q0 + q1 + q2 + 2
507 psraw m15, 2; (p0 + q0 + q1 + q2 + 2) >> 2
508 psubw m15, m5; ((p0 + q0 + q1 + q2 + 2) >> 2) - q1
510 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
513 paddw m13, m7; q3 + 2
514 paddw m13, m6; q3 + q2 + 2
515 paddw m13, m13; 2*q3 + 2*q2 + 4
516 paddw m13, m6; 2*q3 + 3*q2 + 4
517 paddw m13, m10; 2*q3 + 3*q2 + q1 + q0 + p0 + 4
518 psraw m13, 3; (2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3
519 psubw m13, m6; ((2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3) - q2
521 pminsw m13, m9; av_clip( , -2 * tc, 2 * tc)
530 not r6; strong mask -> weak mask
531 and r6, r13; final weak filtering mask, bits 0 and 1
534 ; weak filtering mask
541 pcmpeqd m11, [pd_1]; filtering mask
546 shr betaq, 3; ((beta + (beta >> 1)) >> 3))
549 psubw m12, m4, m3 ; q0 - p0
550 psllw m10, m12, 3; 8 * (q0 - p0)
551 paddw m12, m10 ; 9 * (q0 - p0)
553 psubw m10, m5, m2 ; q1 - p1
554 psllw m8, m10, 1; 2 * ( q1 - p1 )
555 paddw m10, m8; 3 * ( q1 - p1 )
556 psubw m12, m10; 9 * (q0 - p0) - 3 * ( q1 - p1 )
558 psraw m12, 4; >> 4 , delta0
559 PABSW m13, m12; abs(delta0)
562 psllw m10, m9, 2; 8 * tc
563 paddw m10, m9; 10 * tc
567 psraw m9, 1; tc * 2 -> tc
568 psraw m14, 1; -tc * 2 -> -tc
571 pminsw m12, m9; av_clip(delta0, -tc, tc)
573 psraw m9, 1; tc -> tc / 2
575 psignw m14, m9, [pw_m1]; -tc / 2
577 pmullw m14, m9, [pw_m1]; -tc / 2
580 pavgw m15, m1, m3; (p2 + p0 + 1) >> 1
581 psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1
582 paddw m15, m12; ((p2 + p0 + 1) >> 1) - p1 + delta0
583 psraw m15, 1; (((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1
585 pminsw m15, m9; av_clip(deltap1, -tc/2, tc/2)
592 movd m13, r7d; 1dp0 + 1dp3
593 movd m8, r8d; 0dp0 + 0dp3
599 ;end beta calculations
600 MASKED_COPY2 m2, m15, m8; write p1'
602 pavgw m8, m6, m4; (q2 + q0 + 1) >> 1
603 psubw m8, m5; ((q2 + q0 + 1) >> 1) - q1
604 psubw m8, m12; ((q2 + q0 + 1) >> 1) - q1 - delta0)
605 psraw m8, 1; ((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1
607 pminsw m8, m9; av_clip(deltaq1, -tc/2, tc/2)
614 shufps m13, m15, 0; dq0 + dq3
616 pcmpgtw m10, m13; compare to ((beta+(beta>>1))>>3)
618 MASKED_COPY2 m5, m8, m10; write q1'
620 paddw m15, m3, m12 ; p0 + delta0
623 psubw m8, m4, m12 ; q0 - delta0
627 ;-----------------------------------------------------------------------------
628 ; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int32_t *tc,
629 ; uint8_t *_no_p, uint8_t *_no_q);
630 ;-----------------------------------------------------------------------------
631 %macro LOOP_FILTER_CHROMA 0
632 cglobal hevc_v_loop_filter_chroma_8, 3, 5, 7, pix, stride, tc, pix0, r3stride
634 lea r3strideq, [3*strideq]
637 TRANSPOSE4x8B_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
638 CHROMA_DEBLOCK_BODY 8
639 TRANSPOSE8x4B_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
642 cglobal hevc_v_loop_filter_chroma_10, 3, 5, 7, pix, stride, tc, pix0, r3stride
644 lea r3strideq, [3*strideq]
647 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
648 CHROMA_DEBLOCK_BODY 10
649 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq), [pw_pixel_max_10]
652 cglobal hevc_v_loop_filter_chroma_12, 3, 5, 7, pix, stride, tc, pix0, r3stride
654 lea r3strideq, [3*strideq]
657 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
658 CHROMA_DEBLOCK_BODY 12
659 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq), [pw_pixel_max_12]
662 ;-----------------------------------------------------------------------------
663 ; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int32_t *tc,
664 ; uint8_t *_no_p, uint8_t *_no_q);
665 ;-----------------------------------------------------------------------------
666 cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
671 movq m1, [pix0q+strideq]; p0
673 movq m3, [pixq+strideq]; q1
674 pxor m5, m5; zeros reg
679 CHROMA_DEBLOCK_BODY 8
681 movh[pix0q+strideq], m1
685 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
690 movu m1, [pix0q+strideq]; p0
692 movu m3, [pixq+strideq]; q1
693 CHROMA_DEBLOCK_BODY 10
694 pxor m5, m5; zeros reg
695 CLIPW m1, m5, [pw_pixel_max_10]
696 CLIPW m2, m5, [pw_pixel_max_10]
697 movu [pix0q+strideq], m1
701 cglobal hevc_h_loop_filter_chroma_12, 3, 4, 7, pix, stride, tc, pix0
706 movu m1, [pix0q+strideq]; p0
708 movu m3, [pixq+strideq]; q1
709 CHROMA_DEBLOCK_BODY 12
710 pxor m5, m5; zeros reg
711 CLIPW m1, m5, [pw_pixel_max_12]
712 CLIPW m2, m5, [pw_pixel_max_12]
713 movu [pix0q+strideq], m1
724 %macro LOOP_FILTER_LUMA 0
725 ;-----------------------------------------------------------------------------
726 ; void ff_hevc_v_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int beta,
727 ; int32_t *tc, uint8_t *_no_p, uint8_t *_no_q);
728 ;-----------------------------------------------------------------------------
729 cglobal hevc_v_loop_filter_luma_8, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
732 mov src3strideq, pixq
734 TRANSPOSE8x8B_LOAD PASS8ROWS(src3strideq, pixq, r1, pix0q)
735 LUMA_DEBLOCK_BODY 8, v
737 TRANSPOSE8x8B_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q)
741 cglobal hevc_v_loop_filter_luma_10, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
743 lea pix0q, [3 * strideq]
744 mov src3strideq, pixq
746 TRANSPOSE8x8W_LOAD PASS8ROWS(src3strideq, pixq, strideq, pix0q)
747 LUMA_DEBLOCK_BODY 10, v
749 TRANSPOSE8x8W_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q), [pw_pixel_max_10]
753 cglobal hevc_v_loop_filter_luma_12, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
755 lea pix0q, [3 * strideq]
756 mov src3strideq, pixq
758 TRANSPOSE8x8W_LOAD PASS8ROWS(src3strideq, pixq, strideq, pix0q)
759 LUMA_DEBLOCK_BODY 12, v
761 TRANSPOSE8x8W_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q), [pw_pixel_max_12]
765 ;-----------------------------------------------------------------------------
766 ; void ff_hevc_h_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int beta,
767 ; int32_t *tc, uint8_t *_no_p, uint8_t *_no_q);
768 ;-----------------------------------------------------------------------------
769 cglobal hevc_h_loop_filter_luma_8, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
770 lea src3strideq, [3 * strideq]
772 sub pix0q, src3strideq
775 movq m1, [pix0q + strideq]; p2
776 movq m2, [pix0q + 2 * strideq]; p1
777 movq m3, [pix0q + src3strideq]; p0
779 movq m5, [pixq + strideq]; q1
780 movq m6, [pixq + 2 * strideq]; q2
781 movq m7, [pixq + src3strideq]; q3
791 LUMA_DEBLOCK_BODY 8, h
796 movh [pix0q + strideq], m1
797 movhps [pix0q + 2 * strideq], m1
798 movh [pix0q + src3strideq], m3
800 movh [pixq + strideq], m5
801 movhps [pixq + 2 * strideq], m5
805 cglobal hevc_h_loop_filter_luma_10, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
806 lea src3strideq, [3 * strideq]
808 sub pix0q, src3strideq
810 movdqu m0, [pix0q]; p3
811 movdqu m1, [pix0q + strideq]; p2
812 movdqu m2, [pix0q + 2 * strideq]; p1
813 movdqu m3, [pix0q + src3strideq]; p0
814 movdqu m4, [pixq]; q0
815 movdqu m5, [pixq + strideq]; q1
816 movdqu m6, [pixq + 2 * strideq]; q2
817 movdqu m7, [pixq + src3strideq]; q3
818 LUMA_DEBLOCK_BODY 10, h
820 pxor m8, m8; zeros reg
821 CLIPW m1, m8, [pw_pixel_max_10]
822 CLIPW m2, m8, [pw_pixel_max_10]
823 CLIPW m3, m8, [pw_pixel_max_10]
824 CLIPW m4, m8, [pw_pixel_max_10]
825 CLIPW m5, m8, [pw_pixel_max_10]
826 CLIPW m6, m8, [pw_pixel_max_10]
827 movdqu [pix0q + strideq], m1; p2
828 movdqu [pix0q + 2 * strideq], m2; p1
829 movdqu [pix0q + src3strideq], m3; p0
830 movdqu [pixq ], m4; q0
831 movdqu [pixq + strideq], m5; q1
832 movdqu [pixq + 2 * strideq], m6; q2
836 cglobal hevc_h_loop_filter_luma_12, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
837 lea src3strideq, [3 * strideq]
839 sub pix0q, src3strideq
841 movdqu m0, [pix0q]; p3
842 movdqu m1, [pix0q + strideq]; p2
843 movdqu m2, [pix0q + 2 * strideq]; p1
844 movdqu m3, [pix0q + src3strideq]; p0
845 movdqu m4, [pixq]; q0
846 movdqu m5, [pixq + strideq]; q1
847 movdqu m6, [pixq + 2 * strideq]; q2
848 movdqu m7, [pixq + src3strideq]; q3
849 LUMA_DEBLOCK_BODY 12, h
851 pxor m8, m8; zeros reg
852 CLIPW m1, m8, [pw_pixel_max_12]
853 CLIPW m2, m8, [pw_pixel_max_12]
854 CLIPW m3, m8, [pw_pixel_max_12]
855 CLIPW m4, m8, [pw_pixel_max_12]
856 CLIPW m5, m8, [pw_pixel_max_12]
857 CLIPW m6, m8, [pw_pixel_max_12]
858 movdqu [pix0q + strideq], m1; p2
859 movdqu [pix0q + 2 * strideq], m2; p1
860 movdqu [pix0q + src3strideq], m3; p0
861 movdqu [pixq ], m4; q0
862 movdqu [pixq + strideq], m5; q1
863 movdqu [pixq + 2 * strideq], m6; q2