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 tc (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, betas 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
359 movd m13, r11d; beta0
366 movd m14, r12d; beta1
368 pshufd m13, m14, 0; beta0, beta1
369 ;end beta calculations
371 paddw m9, m10, m11; 0d0, 0d3 , 1d0, 1d3
373 pshufhw m14, m9, 0x0f ;0b00001111; 0d3 0d3 0d0 0d0 in high
374 pshuflw m14, m14, 0x0f ;0b00001111; 1d3 1d3 1d0 1d0 in low
376 pshufhw m9, m9, 0xf0 ;0b11110000; 0d0 0d0 0d3 0d3
377 pshuflw m9, m9, 0xf0 ;0b11110000; 1d0 1d0 1d3 1d3
379 paddw m14, m9; 0d0+0d3, 1d0+1d3
382 pcmpgtw m15, m13, m14; beta0, beta1
383 movmskps r13, m15 ;filtering mask 0d0 + 0d3 < beta0 (bit 2 or 3) , 1d0 + 1d3 < beta1 (bit 0 or 1)
387 ;weak / strong decision compare to beta_2
388 psraw m15, m13, 2; beta >> 2
390 pcmpgtw m15, m8; (d0 << 1) < beta_2, (d3 << 1) < beta_2
392 ;end weak / strong decision
394 ; weak filter nd_p/q calculation
399 and r7, 0xffff; 1dp0 + 1dp3
402 and r8, 0xffff; 0dp0 + 0dp3
408 and r9, 0xffff; 1dq0 + 1dq3
411 and r10, 0xffff; 0dq0 + 0dq3
412 ; end calc for weak filter
424 pcmpeqd m11, [pd_1]; filtering mask
426 ;decide between strong and weak filtering
439 add r2d, r3d; tc0 + tc1
443 shufps m8, m9, 0; tc0, tc1
446 pavgw m8, m9; tc25 = ((tc * 5 + 1) >> 1)
447 ;end tc25 calculations
449 ;----beta_3 comparison-----
450 psubw m12, m0, m3; p3 - p0
451 ABS1 m12, m14; abs(p3 - p0)
453 psubw m15, m7, m4; q3 - q0
454 ABS1 m15, m14; abs(q3 - q0)
456 paddw m12, m15; abs(p3 - p0) + abs(q3 - q0)
458 pshufhw m12, m12, 0xf0 ;0b11110000;
459 pshuflw m12, m12, 0xf0 ;0b11110000;
461 psraw m13, 3; beta >> 3
464 and r14, r2; strong mask , beta_2 and beta_3 comparisons
465 ;----beta_3 comparison end-----
466 ;----tc25 comparison---
467 psubw m12, m3, m4; p0 - q0
468 ABS1 m12, m14; abs(p0 - q0)
470 pshufhw m12, m12, 0xf0 ;0b11110000;
471 pshuflw m12, m12, 0xf0 ;0b11110000;
473 pcmpgtw m8, m12; tc25 comparisons
475 and r14, r2; strong mask, beta_2, beta_3 and tc25 comparisons
476 ;----tc25 comparison end---
479 and r14, r2; strong mask, bits 2 and 0
481 pmullw m14, m9, [pw_m2]; -tc * 2
485 mov r2, r14; strong mask
487 movd m12, r14d; store to xmm for mask generation
490 movd m10, r2d; store to xmm for mask generation
491 or r14, r2; final strong mask, bits 1 and 0
495 pcmpeqd m10, [pd_1]; strong mask
497 mova m13, [pw_4]; 4 in every cell
498 pand m11, m10; combine filtering mask and strong mask
499 paddw m12, m2, m3; p1 + p0
500 paddw m12, m4; p1 + p0 + q0
502 paddw m12, m12; 2*p1 + 2*p0 + 2*q0
503 paddw m12, m1; p2 + 2*p1 + 2*p0 + 2*q0
504 paddw m12, m5; p2 + 2*p1 + 2*p0 + 2*q0 + q1
505 paddw m12, m13; p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4
506 psraw m12, 3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3)
507 psubw m12, m3; ((p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4) >> 3) - p0
509 pminsw m12, m9; av_clip( , -2 * tc, 2 * tc)
512 paddw m15, m1, m10; p2 + p1 + p0 + q0
513 psrlw m13, 1; 2 in every cell
514 paddw m15, m13; p2 + p1 + p0 + q0 + 2
515 psraw m15, 2; (p2 + p1 + p0 + q0 + 2) >> 2
516 psubw m15, m2;((p2 + p1 + p0 + q0 + 2) >> 2) - p1
518 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
521 paddw m8, m1, m0; p3 + p2
522 paddw m8, m8; 2*p3 + 2*p2
523 paddw m8, m1; 2*p3 + 3*p2
524 paddw m8, m10; 2*p3 + 3*p2 + p1 + p0 + q0
526 paddw m8, m13; 2*p3 + 3*p2 + p1 + p0 + q0 + 4
527 psraw m8, 3; (2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3
528 psubw m8, m1; ((2*p3 + 3*p2 + p1 + p0 + q0 + 4) >> 3) - p2
530 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
534 paddw m8, m3, m4; p0 + q0
535 paddw m8, m5; p0 + q0 + q1
536 paddw m8, m8; 2*p0 + 2*q0 + 2*q1
537 paddw m8, m2; p1 + 2*p0 + 2*q0 + 2*q1
538 paddw m8, m6; p1 + 2*p0 + 2*q0 + 2*q1 + q2
539 paddw m8, m13; p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4
540 psraw m8, 3; (p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4) >>3
543 pminsw m8, m9; av_clip( , -2 * tc, 2 * tc)
547 paddw m15, m3, m4; p0 + q0
548 paddw m15, m5; p0 + q0 + q1
550 paddw m15, m6; p0 + q0 + q1 + q2
551 psrlw m13, 1; 2 in every cell
552 paddw m15, m13; p0 + q0 + q1 + q2 + 2
553 psraw m15, 2; (p0 + q0 + q1 + q2 + 2) >> 2
554 psubw m15, m5; ((p0 + q0 + q1 + q2 + 2) >> 2) - q1
556 pminsw m15, m9; av_clip( , -2 * tc, 2 * tc)
559 paddw m13, m7; q3 + 2
560 paddw m13, m6; q3 + q2 + 2
561 paddw m13, m13; 2*q3 + 2*q2 + 4
562 paddw m13, m6; 2*q3 + 3*q2 + 4
563 paddw m13, m10; 2*q3 + 3*q2 + q1 + q0 + p0 + 4
564 psraw m13, 3; (2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3
565 psubw m13, m6; ((2*q3 + 3*q2 + q1 + q0 + p0 + 4) >> 3) - q2
567 pminsw m13, m9; av_clip( , -2 * tc, 2 * tc)
576 not r14; strong mask -> weak mask
577 and r14, r13; final weak filtering mask, bits 0 and 1
580 ; weak filtering mask
587 pcmpeqd m11, [pd_1]; filtering mask
592 shr r11, 3; ((beta0+(beta0>>1))>>3))
597 shr r12, 3; ((beta1+(beta1>>1))>>3))
600 psubw m12, m4, m3 ; q0 - p0
601 psllw m10, m12, 3; 8 * (q0 - p0)
602 paddw m12, m10 ; 9 * (q0 - p0)
604 psubw m10, m5, m2 ; q1 - p1
605 psllw m8, m10, 1; 2 * ( q1 - p1 )
606 paddw m10, m8; 3 * ( q1 - p1 )
607 psubw m12, m10; 9 * (q0 - p0) - 3 * ( q1 - p1 )
609 psraw m12, 4; >> 4 , delta0
610 PABSW m13, m12; abs(delta0)
613 psllw m10, m9, 2; 8 * tc
614 paddw m10, m9; 10 * tc
618 psraw m9, 1; tc * 2 -> tc
619 psraw m14, 1; -tc * 2 -> -tc
622 pminsw m12, m9; av_clip(delta0, -tc, tc)
624 psraw m9, 1; tc -> tc / 2
625 pmullw m14, m9, [pw_m1]; -tc / 2
627 pavgw m15, m1, m3; (p2 + p0 + 1) >> 1
628 psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1
629 paddw m15, m12; ((p2 + p0 + 1) >> 1) - p1 + delta0
630 psraw m15, 1; (((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1
632 pminsw m15, m9; av_clip(deltap1, -tc/2, tc/2)
636 movd m10, r11d; beta0
638 movd m13, r12d; beta1
640 shufps m10, m13, 0; betax0, betax1
642 movd m13, r7d; 1dp0 + 1dp3
643 movd m8, r8d; 0dp0 + 0dp3
649 ;end beta calculations
650 MASKED_COPY2 m2, m15, m8; write p1'
652 pavgw m8, m6, m4; (q2 + q0 + 1) >> 1
653 psubw m8, m5; ((q2 + q0 + 1) >> 1) - q1
654 psubw m8, m12; ((q2 + q0 + 1) >> 1) - q1 - delta0)
655 psraw m8, 1; ((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1
657 pminsw m8, m9; av_clip(deltaq1, -tc/2, tc/2)
664 shufps m13, m15, 0; dq0 + dq3
666 pcmpgtw m10, m13; compare to ((beta+(beta>>1))>>3)
668 MASKED_COPY2 m5, m8, m10; write q1'
670 paddw m15, m3, m12 ; p0 + delta0
673 psubw m8, m4, m12 ; q0 - delta0
678 ;-----------------------------------------------------------------------------
679 ; void ff_hevc_v_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q)
680 ;-----------------------------------------------------------------------------
681 cglobal hevc_v_loop_filter_chroma_8, 3, 5, 7, pix, stride, tc, pix0, r3stride
683 lea r3strideq, [3*strideq]
686 TRANSPOSE4x8B_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
687 CHROMA_DEBLOCK_BODY 8
688 TRANSPOSE8x4B_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
691 cglobal hevc_v_loop_filter_chroma_10, 3, 5, 7, pix, stride, tc, pix0, r3stride
693 lea r3strideq, [3*strideq]
696 TRANSPOSE4x8W_LOAD PASS8ROWS(pix0q, pixq, strideq, r3strideq)
697 CHROMA_DEBLOCK_BODY 10
698 TRANSPOSE8x4W_STORE PASS8ROWS(pix0q, pixq, strideq, r3strideq)
701 ;-----------------------------------------------------------------------------
702 ; void ff_hevc_h_loop_filter_chroma(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q
703 ;-----------------------------------------------------------------------------
704 cglobal hevc_h_loop_filter_chroma_8, 3, 4, 7, pix, stride, tc, pix0
709 movq m1, [pix0q+strideq]; p0
711 movq m3, [pixq+strideq]; q1
712 pxor m5, m5; zeros reg
717 CHROMA_DEBLOCK_BODY 8
718 packuswb m1, m1 ; p0' packed in bytes on low quadword
719 packuswb m2, m2 ; q0' packed in bytes on low quadword
720 movq [pix0q+strideq], m1
724 cglobal hevc_h_loop_filter_chroma_10, 3, 4, 7, pix, stride, tc, pix0
729 movu m1, [pix0q+strideq]; p0
731 movu m3, [pixq+strideq]; q1
732 CHROMA_DEBLOCK_BODY 10
733 pxor m5, m5; zeros reg
734 CLIPW m1, m5, [pw_pixel_max]
735 CLIPW m2, m5, [pw_pixel_max]
736 movu [pix0q+strideq], m1
741 %macro LOOP_FILTER_LUMA 0
742 ;-----------------------------------------------------------------------------
743 ; 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);
744 ;-----------------------------------------------------------------------------
745 cglobal hevc_v_loop_filter_luma_8, 4, 15, 16, pix, stride, beta, tc
750 TRANSPOSE8x8B_LOAD PASS8ROWS(r6, r0, r1, r5)
751 LUMA_DEBLOCK_BODY 8, v
753 TRANSPOSE8x8B_STORE PASS8ROWS(r6, r0, r1, r5)
757 cglobal hevc_v_loop_filter_luma_10, 4, 15, 16, pix, stride, beta, tc
759 lea r5, [3 * strideq]
762 TRANSPOSE8x8W_LOAD PASS8ROWS(r6, pixq, strideq, r5)
763 LUMA_DEBLOCK_BODY 10, v
765 TRANSPOSE8x8W_STORE PASS8ROWS(r6, r0, r1, r5)
769 ;-----------------------------------------------------------------------------
770 ; 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);
771 ;-----------------------------------------------------------------------------
772 cglobal hevc_h_loop_filter_luma_8, 4, 15, 16, pix, stride, beta, tc, count, pix0, src3stride
773 lea src3strideq, [3 * strideq]
775 sub pix0q, src3strideq
777 movdqu m0, [pix0q]; p3
778 movdqu m1, [pix0q + strideq]; p2
779 movdqu m2, [pix0q + 2 * strideq]; p1
780 movdqu m3, [pix0q + src3strideq]; p0
781 movdqu m4, [pixq]; q0
782 movdqu m5, [pixq + strideq]; q1
783 movdqu m6, [pixq + 2 * strideq]; q2
784 movdqu m7, [pixq + src3strideq]; q3
794 LUMA_DEBLOCK_BODY 8, h
803 movq [r5+2*r1], m2; p1
807 movq [r0+2*r1], m6; q2
811 cglobal hevc_h_loop_filter_luma_10, 4, 15, 16, pix, stride, beta, tc, count, pix0, src3stride
812 lea src3strideq, [3 * strideq]
814 sub pix0q, src3strideq
816 movdqu m0, [pix0q]; p3
817 movdqu m1, [pix0q + strideq]; p2
818 movdqu m2, [pix0q + 2 * strideq]; p1
819 movdqu m3, [pix0q + src3strideq]; p0
820 movdqu m4, [pixq]; q0
821 movdqu m5, [pixq + strideq]; q1
822 movdqu m6, [pixq + 2 * strideq]; q2
823 movdqu m7, [pixq + src3strideq]; q3
824 LUMA_DEBLOCK_BODY 10, h
826 pxor m8, m8; zeros reg
827 CLIPW m1, m8, [pw_pixel_max]
828 CLIPW m2, m8, [pw_pixel_max]
829 CLIPW m3, m8, [pw_pixel_max]
830 CLIPW m4, m8, [pw_pixel_max]
831 CLIPW m5, m8, [pw_pixel_max]
832 CLIPW m6, m8, [pw_pixel_max]
833 movdqu [pix0q + strideq], m1; p2
834 movdqu [pix0q + 2 * strideq], m2; p1
835 movdqu [pix0q + src3strideq], m3; p0
836 movdqu [pixq ], m4; q0
837 movdqu [pixq + strideq], m5; q1
838 movdqu [pixq + 2 * strideq], m6; q2