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 ;******************************************************************************
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
69 mova m1, [b0q+2*widthq]
70 mova m0, [b1q+2*widthq]
71 COMPOSE_53iL0 m0, m1, [b2q+2*widthq], m2
72 mova [b1q+2*widthq], m0
76 ; void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
78 cglobal vertical_compose_dirac53iH0_%1, 4,4,1, b0, b1, b2, width
82 mova m0, [b0q+2*widthq]
83 paddw m0, [b2q+2*widthq]
86 paddw m0, [b1q+2*widthq]
87 mova [b1q+2*widthq], m0
91 ; void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
92 ; IDWTELEM *b3, IDWTELEM *b4, int width)
93 cglobal vertical_compose_dd97iH0_%1, 6,6,5, b0, b1, b2, b3, b4, width
98 mova m0, [b0q+2*widthq]
99 mova m1, [b1q+2*widthq]
100 COMPOSE_DD97iH0 [b2q+2*widthq], [b3q+2*widthq], [b4q+2*widthq]
101 mova [b2q+2*widthq], m1
105 ; void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
106 ; IDWTELEM *b3, IDWTELEM *b4, int width)
107 cglobal vertical_compose_dd137iL0_%1, 6,6,6, b0, b1, b2, b3, b4, width
112 mova m0, [b0q+2*widthq]
113 mova m1, [b1q+2*widthq]
114 mova m5, [b2q+2*widthq]
115 paddw m0, [b4q+2*widthq]
116 paddw m1, [b3q+2*widthq]
127 mova [b2q+2*widthq], m5
131 ; void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
132 cglobal vertical_compose_haar_%1, 3,4,3, b0, b1, width
136 mova m1, [b1q+2*widthq]
137 mova m0, [b0q+2*widthq]
142 mova [b0q+2*widthq], m0
144 mova [b1q+2*widthq], m2
149 ; extend the left and right edges of the tmp array by %1 and %2 respectively
150 %macro EDGE_EXTENSION 3
157 mov %3, [tmpq+2*w2q-2]
160 mov [tmpq+2*w2q+2*%%i], %3
166 %macro HAAR_HORIZONTAL 2
167 ; void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *tmp, int width)
168 cglobal horizontal_compose_haar%2i_%1, 3,6,4, b, tmp, w, x, w2, b_w2
175 movu m1, [b_w2q + 2*xq]
180 mova [tmpq + 2*xq], m0
186 and w2q, ~(mmsize/2 - 1)
191 movu m1, [b_w2q + 2*xq]
192 mova m0, [tmpq + 2*xq]
195 ; shift and interleave
206 mova [bq+4*xq+mmsize], m2
217 ; void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int width)
218 cglobal horizontal_compose_dd97i_ssse3, 3,6,8, b, tmp, w, x, w2, b_w2
227 movu m1, [b_w2q + 2*xq]
232 COMPOSE_53iL0 m0, m1, m2, m7
233 mova [tmpq + 2*xq], m0
238 EDGE_EXTENSION 1, 2, xw
239 ; leave the last up to 7 (sse) or 3 (mmx) values for C
241 and w2d, ~(mmsize/2 - 1)
245 mova m7, [tmpq-mmsize]
253 mova m7, [tmpq + 2*xq + 16]
258 COMPOSE_DD97iH0 m0, m6, m2, [b_w2q + 2*xq]
262 ; shift and interleave
271 mova [bq+4*xq+mmsize], m2
283 HAAR_HORIZONTAL mmx, 0
284 HAAR_HORIZONTAL mmx, 1
289 COMPOSE_VERTICAL sse2
290 HAAR_HORIZONTAL sse2, 0
291 HAAR_HORIZONTAL sse2, 1