1 ;******************************************************************************
2 ;* x86 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"
25 pw_1991: times 4 dw 9,-1
34 ; %1 -= (%2 + %3 + 2)>>2 %4 is pw_2
35 %macro COMPOSE_53iL0 4
42 ; m1 = %1 + (-m0 + 9*m1 + 9*%2 -%3 + 8)>>4
43 ; if %4 is supplied, %1 is loaded unaligned from there
44 ; m2: clobbered m3: pw_8 m4: pw_1991
45 %macro COMPOSE_DD97iH0 3-4
63 %macro COMPOSE_VERTICAL 1
64 ; void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
66 cglobal vertical_compose53iL0_%1, 4,4,1, b0, b1, b2, width
73 mova m1, [b0q+2*widthq]
74 mova m0, [b1q+2*widthq]
75 COMPOSE_53iL0 m0, m1, [b2q+2*widthq], m2
76 mova [b1q+2*widthq], m0
80 ; void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
82 cglobal vertical_compose_dirac53iH0_%1, 4,4,1, b0, b1, b2, width
89 mova m0, [b0q+2*widthq]
90 paddw m0, [b2q+2*widthq]
93 paddw m0, [b1q+2*widthq]
94 mova [b1q+2*widthq], m0
98 ; void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
99 ; IDWTELEM *b3, IDWTELEM *b4, int width)
100 cglobal vertical_compose_dd97iH0_%1, 6,6,5, b0, b1, b2, b3, b4, width
108 mova m0, [b0q+2*widthq]
109 mova m1, [b1q+2*widthq]
110 COMPOSE_DD97iH0 [b2q+2*widthq], [b3q+2*widthq], [b4q+2*widthq]
111 mova [b2q+2*widthq], m1
115 ; void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
116 ; IDWTELEM *b3, IDWTELEM *b4, int width)
117 cglobal vertical_compose_dd137iL0_%1, 6,6,6, b0, b1, b2, b3, b4, width
125 mova m0, [b0q+2*widthq]
126 mova m1, [b1q+2*widthq]
127 mova m5, [b2q+2*widthq]
128 paddw m0, [b4q+2*widthq]
129 paddw m1, [b3q+2*widthq]
140 mova [b2q+2*widthq], m5
144 ; void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
145 cglobal vertical_compose_haar_%1, 3,4,3, b0, b1, width
152 mova m1, [b1q+2*widthq]
153 mova m0, [b0q+2*widthq]
158 mova [b0q+2*widthq], m0
160 mova [b1q+2*widthq], m2
165 ; extend the left and right edges of the tmp array by %1 and %2 respectively
166 %macro EDGE_EXTENSION 3
173 mov %3, [tmpq+2*w2q-2]
176 mov [tmpq+2*w2q+2*%%i], %3
182 %macro HAAR_HORIZONTAL 2
183 ; void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *tmp, int width)
184 cglobal horizontal_compose_haar%2i_%1, 3,6,4, b, tmp, w, x, w2, b_w2
191 movu m1, [b_w2q + 2*xq]
196 mova [tmpq + 2*xq], m0
202 and w2q, ~(mmsize/2 - 1)
207 movu m1, [b_w2q + 2*xq]
208 mova m0, [tmpq + 2*xq]
211 ; shift and interleave
222 mova [bq+4*xq+mmsize], m2
233 ; void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int width)
234 cglobal horizontal_compose_dd97i_ssse3, 3,6,8, b, tmp, w, x, w2, b_w2
243 movu m1, [b_w2q + 2*xq]
248 COMPOSE_53iL0 m0, m1, m2, m7
249 mova [tmpq + 2*xq], m0
254 EDGE_EXTENSION 1, 2, xw
255 ; leave the last up to 7 (sse) or 3 (mmx) values for C
257 and w2d, ~(mmsize/2 - 1)
261 mova m7, [tmpq-mmsize]
269 mova m7, [tmpq + 2*xq + 16]
274 COMPOSE_DD97iH0 m0, m6, m2, [b_w2q + 2*xq]
278 ; shift and interleave
287 mova [bq+4*xq+mmsize], m2
299 HAAR_HORIZONTAL mmx, 0
300 HAAR_HORIZONTAL mmx, 1
305 COMPOSE_VERTICAL sse2
306 HAAR_HORIZONTAL sse2, 0
307 HAAR_HORIZONTAL sse2, 1