1 ;******************************************************************************
2 ;* MMX/SSSE3-optimized functions for H264 chroma MC
3 ;* Copyright (c) 2005 Zoltan Hidvegi <hzoli -a- hzoli -d- com>,
4 ;* 2005-2008 Loren Merritt
6 ;* This file is part of Libav.
8 ;* Libav is free software; you can redistribute it and/or
9 ;* modify it under the terms of the GNU Lesser General Public
10 ;* License as published by the Free Software Foundation; either
11 ;* version 2.1 of the License, or (at your option) any later version.
13 ;* Libav is distributed in the hope that it will be useful,
14 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
15 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 ;* Lesser General Public License for more details.
18 ;* You should have received a copy of the GNU Lesser General Public
19 ;* License along with Libav; if not, write to the Free Software
20 ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 ;******************************************************************************
23 %include "libavutil/x86/x86util.asm"
27 rnd_rv40_2d_tbl: times 4 dw 0
43 rnd_rv40_1d_tbl: times 4 dw 0
69 %macro mv0_pixels_mc8 0
75 CHROMAMC_AVG mm0, [r0 ]
76 CHROMAMC_AVG mm1, [r0+r2]
83 CHROMAMC_AVG mm0, [r0 ]
84 CHROMAMC_AVG mm1, [r0+r2]
92 %macro chroma_mc8_mmx_func 2-3
95 %define rnd_1d_rv40 r8
96 %define rnd_2d_rv40 r8
99 %define rnd_1d_rv40 rnd_rv40_1d_tbl
100 %define rnd_2d_rv40 rnd_rv40_2d_tbl
106 ; put/avg_h264_chroma_mc8_*(uint8_t *dst /*align 8*/, uint8_t *src /*align 1*/,
107 ; int stride, int h, int mx, int my)
108 cglobal %1_%2_chroma_mc8%3, 6, 7 + extra_regs, 0
114 jne .at_least_one_non_zero
115 ; mx == 0 AND my == 0 - no filter needed
119 .at_least_one_non_zero:
123 and r7, 6 ; &~1 for mx/my=[0,7]
130 and r0, 6 ; &~1 for mx/my=[0,7]
145 mov r6, r2 ; dxy = x ? 1 : stride
148 ; mx == 0 XOR my == 0 - 1 dimensional filter only
153 lea r8, [rnd_rv40_1d_tbl]
162 movq m6, [rnd_1d_%2+rnd_bias*8] ; mm6 = rnd >> 3
164 punpckldq m5, m5 ; mm5 = B = x
166 psubw m4, m5 ; mm4 = A = 8-x
169 movq m0, [r1 ] ; mm0 = src[0..7]
170 movq m2, [r1+r6] ; mm1 = src[1..8]
178 pmullw m0, m4 ; [mm0,mm1] = A * src[0..7]
180 pmullw m2, m5 ; [mm2,mm3] = B * src[1..8]
190 CHROMAMC_AVG m0, [dest_reg]
191 movq [dest_reg], m0 ; dst[0..7] = (A * src[0..7] + B * src[1..8] + (rnd >> 3)) >> 3
199 .both_non_zero: ; general case, bilinear
204 lea r8, [rnd_rv40_2d_tbl]
210 mov r6, rsp ; backup stack pointer
211 and rsp, ~(mmsize-1) ; align stack
212 sub rsp, 16 ; AA and DD
216 punpckldq m4, m4 ; mm4 = x words
217 punpckldq m6, m6 ; mm6 = y words
219 pmullw m4, m6 ; mm4 = x * y
224 movq [rsp+8], m4 ; DD = x * y
225 psubw m5, m4 ; mm5 = B = 8x - xy
226 psubw m6, m4 ; mm6 = C = 8y - xy
228 psubw m4, m7 ; mm4 = A = xy - (8x+8y) + 64
232 movq m0, [r1 ] ; mm0 = src[0..7]
233 movq m1, [r1+1] ; mm1 = src[1..8]
247 paddw m2, m1 ; mm2 = A * src[0..3] + B * src[1..4]
248 paddw m3, m0 ; mm3 = A * src[4..7] + B * src[5..8]
257 paddw m3, m1 ; [mm2,mm3] += C * src[0..7]
267 paddw m3, m4 ; [mm2,mm3] += D * src[1..8]
270 paddw m2, [rnd_2d_%2+rnd_bias*8]
271 paddw m3, [rnd_2d_%2+rnd_bias*8]
275 CHROMAMC_AVG m2, [dest_reg]
276 movq [dest_reg], m2 ; dst[0..7] = ([mm2,mm3] + rnd) >> 6
281 mov rsp, r6 ; restore stack pointer
285 %macro chroma_mc4_mmx_func 2
292 cglobal %1_%2_chroma_mc4, 6, 6 + extra_regs, 0
310 lea r6, [rnd_rv40_2d_tbl]
311 %define rnd_2d_rv40 r6
313 %define rnd_2d_rv40 rnd_rv40_2d_tbl
315 and r5, 6 ; &~1 for mx/my=[0,7]
345 paddw m6, [rnd_2d_%2+rnd_bias*8]
349 CHROMAMC_AVG4 m1, m6, [r0]
364 paddw m0, [rnd_2d_%2+rnd_bias*8]
368 CHROMAMC_AVG4 m1, m0, [r0]
376 %macro chroma_mc2_mmx_func 2
377 cglobal %1_%2_chroma_mc2, 6, 7, 0
386 imul r5d, r4d ; x*y<<16 | y*(8-x)
388 sub r4d, r5d ; x*(8-y)<<16 | (8-x)*(8-y)
392 punpckldq m5, m5 ; mm5 = {A,B,A,B}
393 punpckldq m6, m6 ; mm6 = {C,D,C,D}
397 pshufw m2, m2, 0x94 ; mm0 = src[0,1,1,2]
402 pmaddwd m1, m5 ; mm1 = A * src[0,1] + B * src[1,2]
405 pshufw m0, m0, 0x94 ; mm0 = src[0,1,1,2]
408 paddw m1, [rnd_2d_%2]
409 paddw m1, m0 ; mm1 += C * src[0,1] + D * src[1,2]
413 CHROMAMC_AVG4 m1, m3, [r0]
422 %define rnd_1d_h264 pw_4
423 %define rnd_2d_h264 pw_32
424 %define rnd_1d_vc1 pw_3
425 %define rnd_2d_vc1 pw_28
438 %define CHROMAMC_AVG NOTHING
439 %define CHROMAMC_AVG4 NOTHING
440 chroma_mc8_mmx_func put, h264, _rnd
441 chroma_mc8_mmx_func put, vc1, _nornd
442 chroma_mc8_mmx_func put, rv40
443 chroma_mc4_mmx_func put, h264
444 chroma_mc4_mmx_func put, rv40
447 chroma_mc2_mmx_func put, h264
449 %define CHROMAMC_AVG DIRECT_AVG
450 %define CHROMAMC_AVG4 COPY_AVG
452 chroma_mc8_mmx_func avg, h264, _rnd
453 chroma_mc8_mmx_func avg, vc1, _nornd
454 chroma_mc8_mmx_func avg, rv40
455 chroma_mc4_mmx_func avg, h264
456 chroma_mc4_mmx_func avg, rv40
457 chroma_mc2_mmx_func avg, h264
461 chroma_mc8_mmx_func avg, h264, _rnd
462 chroma_mc8_mmx_func avg, vc1, _nornd
463 chroma_mc8_mmx_func avg, rv40
464 chroma_mc4_mmx_func avg, h264
465 chroma_mc4_mmx_func avg, rv40
467 %macro chroma_mc8_ssse3_func 2-3
468 cglobal %1_%2_chroma_mc8%3, 6, 7, 8
474 jne .at_least_one_non_zero
475 ; mx == 0 AND my == 0 - no filter needed
479 .at_least_one_non_zero:
485 ; general case, bilinear
490 add r4, 8 ; x*288+8 = x<<8 | (8-x)
492 imul r6, r4 ; (8-y)*(x*255+8) = (8-y)*x<<8 | (8-y)*(8-x)
493 imul r4d, r5d ; y *(x*255+8) = y *x<<8 | y *(8-x)
497 movdqa m5, [rnd_2d_%2]
544 sub r4, r5 ; 255*x+8 = x<<8 | (8-x)
546 movdqa m6, [rnd_1d_%2]
581 sub r5, r4 ; 255*y+8 = y<<8 | (8-y)
583 movdqa m6, [rnd_1d_%2]
615 %macro chroma_mc4_ssse3_func 2
616 cglobal %1_%2_chroma_mc4, 6, 7, 0
626 imul r6d, r4d ; (8-y)*(x*255+8) = (8-y)*x<<8 | (8-y)*(8-x)
627 imul r4d, r5d ; y *(x*255+8) = y *x<<8 | y *(8-x)
640 punpcklbw m1, [r1+r2*1+1]
641 punpcklbw m3, [r1+r2*2+1]
658 CHROMAMC_AVG m1, [r0 ]
659 CHROMAMC_AVG m3, [r0+r2]
668 %define CHROMAMC_AVG NOTHING
670 chroma_mc8_ssse3_func put, h264, _rnd
671 chroma_mc8_ssse3_func put, vc1, _nornd
673 chroma_mc4_ssse3_func put, h264
675 %define CHROMAMC_AVG DIRECT_AVG
678 chroma_mc8_ssse3_func avg, h264, _rnd
679 chroma_mc8_ssse3_func avg, vc1, _nornd
681 chroma_mc4_ssse3_func avg, h264