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"
30 %define pw_pixel_max_10 pw_1023
31 pw_pixel_max_12: times 8 dw ((1 << 12)-1)
42 ; expands to [base],...,[base+7*stride]
43 %define PASS8ROWS(base, base3, stride, stride3) \
44 [base], [base+stride], [base+stride*2], [base3], \
45 [base3+stride], [base3+stride*2], [base3+stride3], [base3+stride*4]
47 ; in: 8 rows of 4 bytes in %4..%11
48 ; out: 4 rows of 8 words in m0..m3
49 %macro TRANSPOSE4x8B_LOAD 8
78 ; in: 4 rows of 8 words in m0..m3
79 ; out: 8 rows of 4 bytes in %1..%8
80 %macro TRANSPOSE8x4B_STORE 8
83 SBUTTERFLY bw, 0, 1, 2
84 SBUTTERFLY wd, 0, 1, 2
103 ; in: 8 rows of 4 words in %4..%11
104 ; out: 4 rows of 8 words in m0..m3
105 %macro TRANSPOSE4x8W_LOAD 8
126 punpckhqdq m1, m0, m4
128 punpckhqdq m3, m2, m6
133 ; in: 4 rows of 8 words in m0..m3
134 ; out: 8 rows of 4 words in %1..%8
135 %macro TRANSPOSE8x4W_STORE 9
136 TRANSPOSE4x4W 0, 1, 2, 3, 4
138 pxor m5, m5; zeros reg
154 ; in: 8 rows of 8 bytes in %1..%8
155 ; out: 8 rows of 8 words in m0..m7
156 %macro TRANSPOSE8x8B_LOAD 8
177 punpckldq m1, m3, m9; 0, 1
178 punpckhdq m3, m9; 2, 3
180 punpckldq m5, m7, m4; 4, 5
181 punpckhdq m7, m4; 6, 7
185 punpcklbw m0, m1, m13; 0 in 16 bit
186 punpckhbw m1, m13; 1 in 16 bit
188 punpcklbw m2, m3, m13; 2
191 punpcklbw m4, m5, m13; 4
194 punpcklbw m6, m7, m13; 6
199 ; in: 8 rows of 8 words in m0..m8
200 ; out: 8 rows of 8 bytes in %1..%8
201 %macro TRANSPOSE8x8B_STORE 8
206 TRANSPOSE2x4x4B 0, 1, 2, 3, 4
218 ; in: 8 rows of 8 words in %1..%8
219 ; out: 8 rows of 8 words in m0..m7
220 %macro TRANSPOSE8x8W_LOAD 8
229 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
232 ; in: 8 rows of 8 words in m0..m8
233 ; out: 8 rows of 8 words in %1..%8
234 %macro TRANSPOSE8x8W_STORE 9
235 TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
263 pand %2, m11 ; and mask
264 pandn m10, m11, %1; and -mask
271 ; mask in %3, will be clobbered
272 %macro MASKED_COPY2 3
273 pand %2, %3 ; and mask
274 pandn %3, %1; and -mask
280 ; input in m0 ... m3 and tcs in r2. Output in m1 and m2
281 %macro CHROMA_DEBLOCK_BODY 1
282 psubw m4, m2, m1; q0 - p0
283 psubw m5, m0, m3; p1 - q1
290 pshufd m6, m6, 0xA0; tc0, tc1
292 psignw m4, m6, [pw_m1]; -tc0, -tc1
294 pmullw m4, m6, [pw_m1]; -tc0, -tc1
302 psllw m4, %1-8; << (BIT_DEPTH - 8)
303 psllw m6, %1-8; << (BIT_DEPTH - 8)
307 paddw m1, m5; p0 + delta0
308 psubw m2, m5; q0 - delta0
311 ; input in m0 ... m7, beta in r2 tcs in r3. Output in m1...m6
312 %macro LUMA_DEBLOCK_BODY 2
316 ABS1 m10, m11 ; 0dp0, 0dp3 , 1dp0, 1dp3
321 ABS1 m11, m13 ; 0dq0, 0dq3 , 1dq0, 1dq3
329 ;end beta calculations
331 paddw m9, m10, m11; 0d0, 0d3 , 1d0, 1d3
333 pshufhw m14, m9, 0x0f ;0b00001111; 0d3 0d3 0d0 0d0 in high
334 pshuflw m14, m14, 0x0f ;0b00001111; 1d3 1d3 1d0 1d0 in low
336 pshufhw m9, m9, 0xf0 ;0b11110000; 0d0 0d0 0d3 0d3
337 pshuflw m9, m9, 0xf0 ;0b11110000; 1d0 1d0 1d3 1d3
339 paddw m14, m9; 0d0+0d3, 1d0+1d3
342 pcmpgtw m15, m13, m14
343 movmskps r13, m15 ;filtering mask 0d0 + 0d3 < beta0 (bit 2 or 3) , 1d0 + 1d3 < beta1 (bit 0 or 1)
347 ;weak / strong decision compare to beta_2
348 psraw m15, m13, 2; beta >> 2
350 pcmpgtw m15, m8; (d0 << 1) < beta_2, (d3 << 1) < beta_2
352 ;end weak / strong decision
354 ; weak filter nd_p/q calculation
368 ; end calc for weak filter
380 pcmpeqd m11, [pd_1]; filtering mask
382 ;decide between strong and weak filtering
393 add r11d, r3d; tc0 + tc1
398 shufps m8, m9, 0; tc0, tc1
401 pavgw m8, m9; tc25 = ((tc * 5 + 1) >> 1)
402 ;end tc25 calculations
404 ;----beta_3 comparison-----
405 psubw m12, m0, m3; p3 - p0
406 ABS1 m12, m14; abs(p3 - p0)
408 psubw m15, m7, m4; q3 - q0
409 ABS1 m15, m14; abs(q3 - q0)
411 paddw m12, m15; abs(p3 - p0) + abs(q3 - q0)
413 pshufhw m12, m12, 0xf0 ;0b11110000;
414 pshuflw m12, m12, 0xf0 ;0b11110000;
416 psraw m13, 3; beta >> 3
419 and r6, r11; strong mask , beta_2 and beta_3 comparisons
420 ;----beta_3 comparison end-----
421 ;----tc25 comparison---
422 psubw m12, m3, m4; p0 - q0
423 ABS1 m12, m14; abs(p0 - q0)
425 pshufhw m12, m12, 0xf0 ;0b11110000;
426 pshuflw m12, m12, 0xf0 ;0b11110000;
428 pcmpgtw m8, m12; tc25 comparisons
430 and r6, r11; strong mask, beta_2, beta_3 and tc25 comparisons
431 ;----tc25 comparison end---
434 and r6, r11; strong mask, bits 2 and 0
436 pmullw m14, m9, [pw_m2]; -tc * 2
440 mov r11, r6; strong mask
442 movd m12, r6d; store to xmm for mask generation
445 movd m10, r11d; store to xmm for mask generation
446 or r6, r11; final strong mask, bits 1 and 0
450 pcmpeqd m10, [pd_1]; strong mask
452 mova m13, [pw_4]; 4 in every cell
453 pand m11, m10; combine filtering mask and strong mask
454 paddw m12, m2, m3; p1 + p0
455 paddw m12, m4; p1 + p0 + q0
457 paddw m12, m12; 2*p1 + 2*p0 + 2*q0
458 paddw m12, m1; p2 + 2*p1 + 2*p0 + 2*q0
459 paddw m12, m5; p2 + 2*p1 + 2*p0 + 2*q0 + q1
460 paddw m12, m13; p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4
461 psraw m12, 3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3)
462 psubw m12, m3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3) - p0
464 pminsw m12, m9; av_clip( , -2 * tc, 2 * tc)
467 paddw m15, m1, m10; p2 + p1 + p0 + q0
468 psrlw m13, 1; 2 in every cell
469 paddw m15, m13; p2 + p1 + p0 + q0 + 2
470 psraw m15, 2; (p2 + p1 + p0 + q0 + 2) >> 2
471 psubw m15, m2;((p2 + p1 + p0 + q0 + 2) >> 2) - p1
473 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
476 paddw m8, m1, m0; p3 + p2
477 paddw m8, m8; 2*p3 + 2*p2
478 paddw m8, m1; 2*p3 + 3*p2
479 paddw m8, m10; 2*p3 + 3*p2 + p1 + p0 + q0
481 paddw m8, m13; 2*p3 + 3*p2 + p1 + p0 + q0 + 4
482 psraw m8, 3; (2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3
483 psubw m8, m1; ((2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3) - p2
485 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
489 paddw m8, m3, m4; p0 + q0
490 paddw m8, m5; p0 + q0 + q1
491 paddw m8, m8; 2*p0 + 2*q0 + 2*q1
492 paddw m8, m2; p1 + 2*p0 + 2*q0 + 2*q1
493 paddw m8, m6; p1 + 2*p0 + 2*q0 + 2*q1 + q2
494 paddw m8, m13; p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4
495 psraw m8, 3; (p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4) >>3
498 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
502 paddw m15, m3, m4; p0 + q0
503 paddw m15, m5; p0 + q0 + q1
505 paddw m15, m6; p0 + q0 + q1 + q2
506 psrlw m13, 1; 2 in every cell
507 paddw m15, m13; p0 + q0 + q1 + q2 + 2
508 psraw m15, 2; (p0 + q0 + q1 + q2 + 2) >> 2
509 psubw m15, m5; ((p0 + q0 + q1 + q2 + 2) >> 2) - q1
511 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
514 paddw m13, m7; q3 + 2
515 paddw m13, m6; q3 + q2 + 2
516 paddw m13, m13; 2*q3 + 2*q2 + 4
517 paddw m13, m6; 2*q3 + 3*q2 + 4
518 paddw m13, m10; 2*q3 + 3*q2 + q1 + q0 + p0 + 4
519 psraw m13, 3; (2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3
520 psubw m13, m6; ((2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3) - q2
522 pminsw m13, m9; av_clip( , -2 * tc, 2 * tc)
531 not r6; strong mask -> weak mask
532 and r6, r13; final weak filtering mask, bits 0 and 1
535 ; weak filtering mask
542 pcmpeqd m11, [pd_1]; filtering mask
547 shr betaq, 3; ((beta + (beta >> 1)) >> 3))
550 psubw m12, m4, m3 ; q0 - p0
551 psllw m10, m12, 3; 8 * (q0 - p0)
552 paddw m12, m10 ; 9 * (q0 - p0)
554 psubw m10, m5, m2 ; q1 - p1
555 psllw m8, m10, 1; 2 * ( q1 - p1 )
556 paddw m10, m8; 3 * ( q1 - p1 )
557 psubw m12, m10; 9 * (q0 - p0) - 3 * ( q1 - p1 )
559 psraw m12, 4; >> 4 , delta0
560 PABSW m13, m12; abs(delta0)
563 psllw m10, m9, 2; 8 * tc
564 paddw m10, m9; 10 * tc
568 psraw m9, 1; tc * 2 -> tc
569 psraw m14, 1; -tc * 2 -> -tc
572 pminsw m12, m9; av_clip(delta0, -tc, tc)
574 psraw m9, 1; tc -> tc / 2
576 psignw m14, m9, [pw_m1]; -tc / 2
578 pmullw m14, m9, [pw_m1]; -tc / 2
581 pavgw m15, m1, m3; (p2 + p0 + 1) >> 1
582 psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1
583 paddw m15, m12; ((p2 + p0 + 1) >> 1) - p1 + delta0
584 psraw m15, 1; (((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1
586 pminsw m15, m9; av_clip(deltap1, -tc/2, tc/2)
593 movd m13, r7d; 1dp0 + 1dp3
594 movd m8, r8d; 0dp0 + 0dp3
600 ;end beta calculations
601 MASKED_COPY2 m2, m15, m8; write p1'
603 pavgw m8, m6, m4; (q2 + q0 + 1) >> 1
604 psubw m8, m5; ((q2 + q0 + 1) >> 1) - q1
605 psubw m8, m12; ((q2 + q0 + 1) >> 1) - q1 - delta0)
606 psraw m8, 1; ((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1
608 pminsw m8, m9; av_clip(deltaq1, -tc/2, tc/2)
615 shufps m13, m15, 0; dq0 + dq3
617 pcmpgtw m10, m13; compare to ((beta+(beta>>1))>>3)
619 MASKED_COPY2 m5, m8, m10; write q1'
621 paddw m15, m3, m12 ; p0 + delta0
624 psubw m8, m4, m12 ; q0 - delta0
628 ;-----------------------------------------------------------------------------
629 ; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int32_t *tc,
630 ; uint8_t *_no_p, uint8_t *_no_q);
631 ;-----------------------------------------------------------------------------
632 %macro LOOP_FILTER_CHROMA 0
633 cglobal hevc_v_loop_filter_chroma_8, 3, 5, 7, pix, stride, tc, pix0, r3stride
635 lea r3strideq, [3*strideq]
638 TRANSPOSE4x8B_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
639 CHROMA_DEBLOCK_BODY 8
640 TRANSPOSE8x4B_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
643 cglobal hevc_v_loop_filter_chroma_10, 3, 5, 7, pix, stride, tc, pix0, r3stride
645 lea r3strideq, [3*strideq]
648 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
649 CHROMA_DEBLOCK_BODY 10
650 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq), [pw_pixel_max_10]
653 cglobal hevc_v_loop_filter_chroma_12, 3, 5, 7, pix, stride, tc, pix0, r3stride
655 lea r3strideq, [3*strideq]
658 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
659 CHROMA_DEBLOCK_BODY 12
660 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq), [pw_pixel_max_12]
663 ;-----------------------------------------------------------------------------
664 ; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int32_t *tc,
665 ; uint8_t *_no_p, uint8_t *_no_q);
666 ;-----------------------------------------------------------------------------
667 cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
672 movq m1, [pix0q+strideq]; p0
674 movq m3, [pixq+strideq]; q1
675 pxor m5, m5; zeros reg
680 CHROMA_DEBLOCK_BODY 8
682 movh[pix0q+strideq], m1
686 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
691 movu m1, [pix0q+strideq]; p0
693 movu m3, [pixq+strideq]; q1
694 CHROMA_DEBLOCK_BODY 10
695 pxor m5, m5; zeros reg
696 CLIPW m1, m5, [pw_pixel_max_10]
697 CLIPW m2, m5, [pw_pixel_max_10]
698 movu [pix0q+strideq], m1
702 cglobal hevc_h_loop_filter_chroma_12, 3, 4, 7, pix, stride, tc, pix0
707 movu m1, [pix0q+strideq]; p0
709 movu m3, [pixq+strideq]; q1
710 CHROMA_DEBLOCK_BODY 12
711 pxor m5, m5; zeros reg
712 CLIPW m1, m5, [pw_pixel_max_12]
713 CLIPW m2, m5, [pw_pixel_max_12]
714 movu [pix0q+strideq], m1
725 %macro LOOP_FILTER_LUMA 0
726 ;-----------------------------------------------------------------------------
727 ; void ff_hevc_v_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int beta,
728 ; int32_t *tc, uint8_t *_no_p, uint8_t *_no_q);
729 ;-----------------------------------------------------------------------------
730 cglobal hevc_v_loop_filter_luma_8, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
733 mov src3strideq, pixq
735 TRANSPOSE8x8B_LOAD PASS8ROWS(src3strideq, pixq, r1, pix0q)
736 LUMA_DEBLOCK_BODY 8, v
738 TRANSPOSE8x8B_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q)
742 cglobal hevc_v_loop_filter_luma_10, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
744 lea pix0q, [3 * strideq]
745 mov src3strideq, pixq
747 TRANSPOSE8x8W_LOAD PASS8ROWS(src3strideq, pixq, strideq, pix0q)
748 LUMA_DEBLOCK_BODY 10, v
750 TRANSPOSE8x8W_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q), [pw_pixel_max_10]
754 cglobal hevc_v_loop_filter_luma_12, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
756 lea pix0q, [3 * strideq]
757 mov src3strideq, pixq
759 TRANSPOSE8x8W_LOAD PASS8ROWS(src3strideq, pixq, strideq, pix0q)
760 LUMA_DEBLOCK_BODY 12, v
762 TRANSPOSE8x8W_STORE PASS8ROWS(src3strideq, pixq, r1, pix0q), [pw_pixel_max_12]
766 ;-----------------------------------------------------------------------------
767 ; void ff_hevc_h_loop_filter_luma(uint8_t *_pix, ptrdiff_t _stride, int beta,
768 ; int32_t *tc, uint8_t *_no_p, uint8_t *_no_q);
769 ;-----------------------------------------------------------------------------
770 cglobal hevc_h_loop_filter_luma_8, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
771 lea src3strideq, [3 * strideq]
773 sub pix0q, src3strideq
776 movq m1, [pix0q + strideq]; p2
777 movq m2, [pix0q + 2 * strideq]; p1
778 movq m3, [pix0q + src3strideq]; p0
780 movq m5, [pixq + strideq]; q1
781 movq m6, [pixq + 2 * strideq]; q2
782 movq m7, [pixq + src3strideq]; q3
792 LUMA_DEBLOCK_BODY 8, h
797 movh [pix0q + strideq], m1
798 movhps [pix0q + 2 * strideq], m1
799 movh [pix0q + src3strideq], m3
801 movh [pixq + strideq], m5
802 movhps [pixq + 2 * strideq], m5
806 cglobal hevc_h_loop_filter_luma_10, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
807 lea src3strideq, [3 * strideq]
809 sub pix0q, src3strideq
811 movdqu m0, [pix0q]; p3
812 movdqu m1, [pix0q + strideq]; p2
813 movdqu m2, [pix0q + 2 * strideq]; p1
814 movdqu m3, [pix0q + src3strideq]; p0
815 movdqu m4, [pixq]; q0
816 movdqu m5, [pixq + strideq]; q1
817 movdqu m6, [pixq + 2 * strideq]; q2
818 movdqu m7, [pixq + src3strideq]; q3
819 LUMA_DEBLOCK_BODY 10, h
821 pxor m8, m8; zeros reg
822 CLIPW m1, m8, [pw_pixel_max_10]
823 CLIPW m2, m8, [pw_pixel_max_10]
824 CLIPW m3, m8, [pw_pixel_max_10]
825 CLIPW m4, m8, [pw_pixel_max_10]
826 CLIPW m5, m8, [pw_pixel_max_10]
827 CLIPW m6, m8, [pw_pixel_max_10]
828 movdqu [pix0q + strideq], m1; p2
829 movdqu [pix0q + 2 * strideq], m2; p1
830 movdqu [pix0q + src3strideq], m3; p0
831 movdqu [pixq ], m4; q0
832 movdqu [pixq + strideq], m5; q1
833 movdqu [pixq + 2 * strideq], m6; q2
837 cglobal hevc_h_loop_filter_luma_12, 4, 14, 16, pix, stride, beta, tc, pix0, src3stride
838 lea src3strideq, [3 * strideq]
840 sub pix0q, src3strideq
842 movdqu m0, [pix0q]; p3
843 movdqu m1, [pix0q + strideq]; p2
844 movdqu m2, [pix0q + 2 * strideq]; p1
845 movdqu m3, [pix0q + src3strideq]; p0
846 movdqu m4, [pixq]; q0
847 movdqu m5, [pixq + strideq]; q1
848 movdqu m6, [pixq + 2 * strideq]; q2
849 movdqu m7, [pixq + src3strideq]; q3
850 LUMA_DEBLOCK_BODY 12, h
852 pxor m8, m8; zeros reg
853 CLIPW m1, m8, [pw_pixel_max_12]
854 CLIPW m2, m8, [pw_pixel_max_12]
855 CLIPW m3, m8, [pw_pixel_max_12]
856 CLIPW m4, m8, [pw_pixel_max_12]
857 CLIPW m5, m8, [pw_pixel_max_12]
858 CLIPW m6, m8, [pw_pixel_max_12]
859 movdqu [pix0q + strideq], m1; p2
860 movdqu [pix0q + 2 * strideq], m2; p1
861 movdqu [pix0q + src3strideq], m3; p0
862 movdqu [pixq ], m4; q0
863 movdqu [pixq + strideq], m5; q1
864 movdqu [pixq + 2 * strideq], m6; q2