1 ;******************************************************************************
2 ;* MMX optimized discrete wavelet trasnform
3 ;* Copyright (c) 2010 David Conrad
5 ;* This file is part of FFmpeg.
7 ;* FFmpeg is free software; you can redistribute it and/or
8 ;* modify it under the terms of the GNU Lesser General Public
9 ;* License as published by the Free Software Foundation; either
10 ;* version 2.1 of the License, or (at your option) any later version.
12 ;* FFmpeg is distributed in the hope that it will be useful,
13 ;* but WITHOUT ANY WARRANTY; without even the implied warranty of
14 ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 ;* Lesser General Public License for more details.
17 ;* You should have received a copy of the GNU Lesser General Public
18 ;* License along with FFmpeg; if not, write to the Free Software
19 ;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 ;******************************************************************************
22 %include "libavutil/x86/x86util.asm"
29 pw_1991: times 4 dw 9,-1
33 ; %1 -= (%2 + %3 + 2)>>2 %4 is pw_2
34 %macro COMPOSE_53iL0 4
41 ; m1 = %1 + (-m0 + 9*m1 + 9*%2 -%3 + 8)>>4
42 ; if %4 is supplied, %1 is loaded unaligned from there
43 ; m2: clobbered m3: pw_8 m4: pw_1991
44 %macro COMPOSE_DD97iH0 3-4
62 %macro COMPOSE_VERTICAL 1
63 ; void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
65 cglobal vertical_compose53iL0_%1, 4,4,1, b0, b1, b2, width
72 mova m1, [b0q+2*widthq]
73 mova m0, [b1q+2*widthq]
74 COMPOSE_53iL0 m0, m1, [b2q+2*widthq], m2
75 mova [b1q+2*widthq], m0
79 ; void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
81 cglobal vertical_compose_dirac53iH0_%1, 4,4,1, b0, b1, b2, width
88 mova m0, [b0q+2*widthq]
89 paddw m0, [b2q+2*widthq]
92 paddw m0, [b1q+2*widthq]
93 mova [b1q+2*widthq], m0
97 ; void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
98 ; IDWTELEM *b3, IDWTELEM *b4, int width)
99 cglobal vertical_compose_dd97iH0_%1, 6,6,5, b0, b1, b2, b3, b4, width
107 mova m0, [b0q+2*widthq]
108 mova m1, [b1q+2*widthq]
109 COMPOSE_DD97iH0 [b2q+2*widthq], [b3q+2*widthq], [b4q+2*widthq]
110 mova [b2q+2*widthq], m1
114 ; void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
115 ; IDWTELEM *b3, IDWTELEM *b4, int width)
116 cglobal vertical_compose_dd137iL0_%1, 6,6,6, b0, b1, b2, b3, b4, width
124 mova m0, [b0q+2*widthq]
125 mova m1, [b1q+2*widthq]
126 mova m5, [b2q+2*widthq]
127 paddw m0, [b4q+2*widthq]
128 paddw m1, [b3q+2*widthq]
139 mova [b2q+2*widthq], m5
143 ; void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
144 cglobal vertical_compose_haar_%1, 3,4,3, b0, b1, width
151 mova m1, [b1q+2*widthq]
152 mova m0, [b0q+2*widthq]
157 mova [b0q+2*widthq], m0
159 mova [b1q+2*widthq], m2
164 ; extend the left and right edges of the tmp array by %1 and %2 respectively
165 %macro EDGE_EXTENSION 3
172 mov %3, [tmpq+2*w2q-2]
175 mov [tmpq+2*w2q+2*%%i], %3
181 %macro HAAR_HORIZONTAL 2
182 ; void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *tmp, int width)
183 cglobal horizontal_compose_haar%2i_%1, 3,6,4, b, tmp, w, x, w2, b_w2
190 movu m1, [b_w2q + 2*xq]
195 mova [tmpq + 2*xq], m0
201 and w2q, ~(mmsize/2 - 1)
206 movu m1, [b_w2q + 2*xq]
207 mova m0, [tmpq + 2*xq]
210 ; shift and interleave
221 mova [bq+4*xq+mmsize], m2
232 ; void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int width)
233 cglobal horizontal_compose_dd97i_ssse3, 3,6,8, b, tmp, w, x, w2, b_w2
242 movu m1, [b_w2q + 2*xq]
247 COMPOSE_53iL0 m0, m1, m2, m7
248 mova [tmpq + 2*xq], m0
253 EDGE_EXTENSION 1, 2, xw
254 ; leave the last up to 7 (sse) or 3 (mmx) values for C
256 and w2d, ~(mmsize/2 - 1)
260 mova m7, [tmpq-mmsize]
268 mova m7, [tmpq + 2*xq + 16]
273 COMPOSE_DD97iH0 m0, m6, m2, [b_w2q + 2*xq]
277 ; shift and interleave
286 mova [bq+4*xq+mmsize], m2
298 HAAR_HORIZONTAL mmx, 0
299 HAAR_HORIZONTAL mmx, 1
304 COMPOSE_VERTICAL sse2
305 HAAR_HORIZONTAL sse2, 0
306 HAAR_HORIZONTAL sse2, 1