1;******************************************************************************
2;* MMX optimized discrete wavelet trasnform
3;* Copyright (c) 2010 David Conrad
4;*
5;* This file is part of FFmpeg.
6;*
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.
11;*
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.
16;*
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;******************************************************************************
21
22%include "libavutil/x86/x86util.asm"
23
24SECTION_RODATA
25pw_1: times 8 dw 1
26pw_2: times 8 dw 2
27pw_8: times 8 dw 8
28pw_16: times 8 dw 16
29pw_1991: times 4 dw 9,-1
30
31section .text
32
33; %1 -= (%2 + %3 + 2)>>2     %4 is pw_2
34%macro COMPOSE_53iL0 4
35    paddw   %2, %3
36    paddw   %2, %4
37    psraw   %2, 2
38    psubw   %1, %2
39%endm
40
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
45    paddw   m0, %3
46    paddw   m1, %2
47    psubw   m0, m3
48    mova    m2, m1
49    punpcklwd m1, m0
50    punpckhwd m2, m0
51    pmaddwd m1, m4
52    pmaddwd m2, m4
53%if %0 > 3
54    movu    %1, %4
55%endif
56    psrad   m1, 4
57    psrad   m2, 4
58    packssdw m1, m2
59    paddw   m1, %1
60%endm
61
62%macro COMPOSE_VERTICAL 1
63; void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
64;                                  int width)
65cglobal vertical_compose53iL0_%1, 4,4,1, b0, b1, b2, width
66    mova    m2, [pw_2]
67%if ARCH_X86_64
68    mov     widthd, widthd
69%endif
70.loop:
71    sub     widthq, mmsize/2
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
76    jg      .loop
77    REP_RET
78
79; void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
80;                                  int width)
81cglobal vertical_compose_dirac53iH0_%1, 4,4,1, b0, b1, b2, width
82    mova    m1, [pw_1]
83%if ARCH_X86_64
84    mov     widthd, widthd
85%endif
86.loop:
87    sub     widthq, mmsize/2
88    mova    m0, [b0q+2*widthq]
89    paddw   m0, [b2q+2*widthq]
90    paddw   m0, m1
91    psraw   m0, 1
92    paddw   m0, [b1q+2*widthq]
93    mova    [b1q+2*widthq], m0
94    jg      .loop
95    REP_RET
96
97; void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
98;                               IDWTELEM *b3, IDWTELEM *b4, int width)
99cglobal vertical_compose_dd97iH0_%1, 6,6,5, b0, b1, b2, b3, b4, width
100    mova    m3, [pw_8]
101    mova    m4, [pw_1991]
102%if ARCH_X86_64
103    mov     widthd, widthd
104%endif
105.loop:
106    sub     widthq, mmsize/2
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
111    jg      .loop
112    REP_RET
113
114; void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
115;                                IDWTELEM *b3, IDWTELEM *b4, int width)
116cglobal vertical_compose_dd137iL0_%1, 6,6,6, b0, b1, b2, b3, b4, width
117    mova    m3, [pw_16]
118    mova    m4, [pw_1991]
119%if ARCH_X86_64
120    mov     widthd, widthd
121%endif
122.loop:
123    sub     widthq, mmsize/2
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]
129    psubw   m0, m3
130    mova    m2, m1
131    punpcklwd m1, m0
132    punpckhwd m2, m0
133    pmaddwd m1, m4
134    pmaddwd m2, m4
135    psrad   m1, 5
136    psrad   m2, 5
137    packssdw m1, m2
138    psubw   m5, m1
139    mova    [b2q+2*widthq], m5
140    jg      .loop
141    REP_RET
142
143; void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
144cglobal vertical_compose_haar_%1, 3,4,3, b0, b1, width
145    mova    m3, [pw_1]
146%if ARCH_X86_64
147    mov     widthd, widthd
148%endif
149.loop:
150    sub     widthq, mmsize/2
151    mova    m1, [b1q+2*widthq]
152    mova    m0, [b0q+2*widthq]
153    mova    m2, m1
154    paddw   m1, m3
155    psraw   m1, 1
156    psubw   m0, m1
157    mova    [b0q+2*widthq], m0
158    paddw   m2, m0
159    mova    [b1q+2*widthq], m2
160    jg      .loop
161    REP_RET
162%endmacro
163
164; extend the left and right edges of the tmp array by %1 and %2 respectively
165%macro EDGE_EXTENSION 3
166    mov     %3, [tmpq]
167%assign %%i 1
168%rep %1
169    mov     [tmpq-2*%%i], %3
170    %assign %%i %%i+1
171%endrep
172    mov     %3, [tmpq+2*w2q-2]
173%assign %%i 0
174%rep %2
175    mov     [tmpq+2*w2q+2*%%i], %3
176    %assign %%i %%i+1
177%endrep
178%endmacro
179
180
181%macro HAAR_HORIZONTAL 2
182; void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *tmp, int width)
183cglobal horizontal_compose_haar%2i_%1, 3,6,4, b, tmp, w, x, w2, b_w2
184    mov    w2d, wd
185    xor     xq, xq
186    shr    w2d, 1
187    lea  b_w2q, [bq+wq]
188    mova    m3, [pw_1]
189.lowpass_loop:
190    movu    m1, [b_w2q + 2*xq]
191    mova    m0, [bq    + 2*xq]
192    paddw   m1, m3
193    psraw   m1, 1
194    psubw   m0, m1
195    mova    [tmpq + 2*xq], m0
196    add     xq, mmsize/2
197    cmp     xq, w2q
198    jl      .lowpass_loop
199
200    xor     xq, xq
201    and    w2q, ~(mmsize/2 - 1)
202    cmp    w2q, mmsize/2
203    jl      .end
204
205.highpass_loop:
206    movu    m1, [b_w2q + 2*xq]
207    mova    m0, [tmpq  + 2*xq]
208    paddw   m1, m0
209
210    ; shift and interleave
211%if %2 == 1
212    paddw   m0, m3
213    paddw   m1, m3
214    psraw   m0, 1
215    psraw   m1, 1
216%endif
217    mova    m2, m0
218    punpcklwd m0, m1
219    punpckhwd m2, m1
220    mova    [bq+4*xq], m0
221    mova    [bq+4*xq+mmsize], m2
222
223    add     xq, mmsize/2
224    cmp     xq, w2q
225    jl      .highpass_loop
226.end:
227    REP_RET
228%endmacro
229
230
231INIT_XMM
232; void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int width)
233cglobal horizontal_compose_dd97i_ssse3, 3,6,8, b, tmp, w, x, w2, b_w2
234    mov    w2d, wd
235    xor     xd, xd
236    shr    w2d, 1
237    lea  b_w2q, [bq+wq]
238    movu    m4, [bq+wq]
239    mova    m7, [pw_2]
240    pslldq  m4, 14
241.lowpass_loop:
242    movu    m1, [b_w2q + 2*xq]
243    mova    m0, [bq    + 2*xq]
244    mova    m2, m1
245    palignr m1, m4, 14
246    mova    m4, m2
247    COMPOSE_53iL0 m0, m1, m2, m7
248    mova    [tmpq + 2*xq], m0
249    add     xd, mmsize/2
250    cmp     xd, w2d
251    jl      .lowpass_loop
252
253    EDGE_EXTENSION 1, 2, xw
254    ; leave the last up to 7 (sse) or 3 (mmx) values for C
255    xor     xd, xd
256    and    w2d, ~(mmsize/2 - 1)
257    cmp    w2d, mmsize/2
258    jl      .end
259
260    mova    m7, [tmpq-mmsize]
261    mova    m0, [tmpq]
262    mova    m5, [pw_1]
263    mova    m3, [pw_8]
264    mova    m4, [pw_1991]
265.highpass_loop:
266    mova    m6, m0
267    palignr m0, m7, 14
268    mova    m7, [tmpq + 2*xq + 16]
269    mova    m1, m7
270    mova    m2, m7
271    palignr m1, m6, 2
272    palignr m2, m6, 4
273    COMPOSE_DD97iH0 m0, m6, m2, [b_w2q + 2*xq]
274    mova    m0, m7
275    mova    m7, m6
276
277    ; shift and interleave
278    paddw   m6, m5
279    paddw   m1, m5
280    psraw   m6, 1
281    psraw   m1, 1
282    mova    m2, m6
283    punpcklwd m6, m1
284    punpckhwd m2, m1
285    mova    [bq+4*xq], m6
286    mova    [bq+4*xq+mmsize], m2
287
288    add     xd, mmsize/2
289    cmp     xd, w2d
290    jl      .highpass_loop
291.end:
292    REP_RET
293
294
295%if ARCH_X86_64 == 0
296INIT_MMX
297COMPOSE_VERTICAL mmx
298HAAR_HORIZONTAL mmx, 0
299HAAR_HORIZONTAL mmx, 1
300%endif
301
302;;INIT_XMM
303INIT_XMM
304COMPOSE_VERTICAL sse2
305HAAR_HORIZONTAL sse2, 0
306HAAR_HORIZONTAL sse2, 1
307