1;******************************************************************************
2;* x86 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_1991: times 4 dw 9,-1
26
27cextern pw_1
28cextern pw_2
29cextern pw_8
30cextern pw_16
31
32SECTION .text
33
34; %1 -= (%2 + %3 + 2)>>2     %4 is pw_2
35%macro COMPOSE_53iL0 4
36    paddw   %2, %3
37    paddw   %2, %4
38    psraw   %2, 2
39    psubw   %1, %2
40%endm
41
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
46    paddw   m0, %3
47    paddw   m1, %2
48    psubw   m0, m3
49    mova    m2, m1
50    punpcklwd m1, m0
51    punpckhwd m2, m0
52    pmaddwd m1, m4
53    pmaddwd m2, m4
54%if %0 > 3
55    movu    %1, %4
56%endif
57    psrad   m1, 4
58    psrad   m2, 4
59    packssdw m1, m2
60    paddw   m1, %1
61%endm
62
63%macro COMPOSE_VERTICAL 1
64; void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
65;                                  int width)
66cglobal vertical_compose53iL0_%1, 4,4,1, b0, b1, b2, width
67    mova    m2, [pw_2]
68%if ARCH_X86_64
69    mov     widthd, widthd
70%endif
71.loop:
72    sub     widthq, mmsize/2
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
77    jg      .loop
78    REP_RET
79
80; void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
81;                                  int width)
82cglobal vertical_compose_dirac53iH0_%1, 4,4,1, b0, b1, b2, width
83    mova    m1, [pw_1]
84%if ARCH_X86_64
85    mov     widthd, widthd
86%endif
87.loop:
88    sub     widthq, mmsize/2
89    mova    m0, [b0q+2*widthq]
90    paddw   m0, [b2q+2*widthq]
91    paddw   m0, m1
92    psraw   m0, 1
93    paddw   m0, [b1q+2*widthq]
94    mova    [b1q+2*widthq], m0
95    jg      .loop
96    REP_RET
97
98; void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
99;                               IDWTELEM *b3, IDWTELEM *b4, int width)
100cglobal vertical_compose_dd97iH0_%1, 6,6,5, b0, b1, b2, b3, b4, width
101    mova    m3, [pw_8]
102    mova    m4, [pw_1991]
103%if ARCH_X86_64
104    mov     widthd, widthd
105%endif
106.loop:
107    sub     widthq, mmsize/2
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
112    jg      .loop
113    REP_RET
114
115; void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
116;                                IDWTELEM *b3, IDWTELEM *b4, int width)
117cglobal vertical_compose_dd137iL0_%1, 6,6,6, b0, b1, b2, b3, b4, width
118    mova    m3, [pw_16]
119    mova    m4, [pw_1991]
120%if ARCH_X86_64
121    mov     widthd, widthd
122%endif
123.loop:
124    sub     widthq, mmsize/2
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]
130    psubw   m0, m3
131    mova    m2, m1
132    punpcklwd m1, m0
133    punpckhwd m2, m0
134    pmaddwd m1, m4
135    pmaddwd m2, m4
136    psrad   m1, 5
137    psrad   m2, 5
138    packssdw m1, m2
139    psubw   m5, m1
140    mova    [b2q+2*widthq], m5
141    jg      .loop
142    REP_RET
143
144; void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
145cglobal vertical_compose_haar_%1, 3,4,3, b0, b1, width
146    mova    m3, [pw_1]
147%if ARCH_X86_64
148    mov     widthd, widthd
149%endif
150.loop:
151    sub     widthq, mmsize/2
152    mova    m1, [b1q+2*widthq]
153    mova    m0, [b0q+2*widthq]
154    mova    m2, m1
155    paddw   m1, m3
156    psraw   m1, 1
157    psubw   m0, m1
158    mova    [b0q+2*widthq], m0
159    paddw   m2, m0
160    mova    [b1q+2*widthq], m2
161    jg      .loop
162    REP_RET
163%endmacro
164
165; extend the left and right edges of the tmp array by %1 and %2 respectively
166%macro EDGE_EXTENSION 3
167    mov     %3, [tmpq]
168%assign %%i 1
169%rep %1
170    mov     [tmpq-2*%%i], %3
171    %assign %%i %%i+1
172%endrep
173    mov     %3, [tmpq+2*w2q-2]
174%assign %%i 0
175%rep %2
176    mov     [tmpq+2*w2q+2*%%i], %3
177    %assign %%i %%i+1
178%endrep
179%endmacro
180
181
182%macro HAAR_HORIZONTAL 2
183; void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *tmp, int width)
184cglobal horizontal_compose_haar%2i_%1, 3,6,4, b, tmp, w, x, w2, b_w2
185    mov    w2d, wd
186    xor     xq, xq
187    shr    w2d, 1
188    lea  b_w2q, [bq+wq]
189    mova    m3, [pw_1]
190.lowpass_loop:
191    movu    m1, [b_w2q + 2*xq]
192    mova    m0, [bq    + 2*xq]
193    paddw   m1, m3
194    psraw   m1, 1
195    psubw   m0, m1
196    mova    [tmpq + 2*xq], m0
197    add     xq, mmsize/2
198    cmp     xq, w2q
199    jl      .lowpass_loop
200
201    xor     xq, xq
202    and    w2q, ~(mmsize/2 - 1)
203    cmp    w2q, mmsize/2
204    jl      .end
205
206.highpass_loop:
207    movu    m1, [b_w2q + 2*xq]
208    mova    m0, [tmpq  + 2*xq]
209    paddw   m1, m0
210
211    ; shift and interleave
212%if %2 == 1
213    paddw   m0, m3
214    paddw   m1, m3
215    psraw   m0, 1
216    psraw   m1, 1
217%endif
218    mova    m2, m0
219    punpcklwd m0, m1
220    punpckhwd m2, m1
221    mova    [bq+4*xq], m0
222    mova    [bq+4*xq+mmsize], m2
223
224    add     xq, mmsize/2
225    cmp     xq, w2q
226    jl      .highpass_loop
227.end:
228    REP_RET
229%endmacro
230
231
232INIT_XMM
233; void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int width)
234cglobal horizontal_compose_dd97i_ssse3, 3,6,8, b, tmp, w, x, w2, b_w2
235    mov    w2d, wd
236    xor     xd, xd
237    shr    w2d, 1
238    lea  b_w2q, [bq+wq]
239    movu    m4, [bq+wq]
240    mova    m7, [pw_2]
241    pslldq  m4, 14
242.lowpass_loop:
243    movu    m1, [b_w2q + 2*xq]
244    mova    m0, [bq    + 2*xq]
245    mova    m2, m1
246    palignr m1, m4, 14
247    mova    m4, m2
248    COMPOSE_53iL0 m0, m1, m2, m7
249    mova    [tmpq + 2*xq], m0
250    add     xd, mmsize/2
251    cmp     xd, w2d
252    jl      .lowpass_loop
253
254    EDGE_EXTENSION 1, 2, xw
255    ; leave the last up to 7 (sse) or 3 (mmx) values for C
256    xor     xd, xd
257    and    w2d, ~(mmsize/2 - 1)
258    cmp    w2d, mmsize/2
259    jl      .end
260
261    mova    m7, [tmpq-mmsize]
262    mova    m0, [tmpq]
263    mova    m5, [pw_1]
264    mova    m3, [pw_8]
265    mova    m4, [pw_1991]
266.highpass_loop:
267    mova    m6, m0
268    palignr m0, m7, 14
269    mova    m7, [tmpq + 2*xq + 16]
270    mova    m1, m7
271    mova    m2, m7
272    palignr m1, m6, 2
273    palignr m2, m6, 4
274    COMPOSE_DD97iH0 m0, m6, m2, [b_w2q + 2*xq]
275    mova    m0, m7
276    mova    m7, m6
277
278    ; shift and interleave
279    paddw   m6, m5
280    paddw   m1, m5
281    psraw   m6, 1
282    psraw   m1, 1
283    mova    m2, m6
284    punpcklwd m6, m1
285    punpckhwd m2, m1
286    mova    [bq+4*xq], m6
287    mova    [bq+4*xq+mmsize], m2
288
289    add     xd, mmsize/2
290    cmp     xd, w2d
291    jl      .highpass_loop
292.end:
293    REP_RET
294
295
296%if ARCH_X86_64 == 0
297INIT_MMX
298COMPOSE_VERTICAL mmx
299HAAR_HORIZONTAL mmx, 0
300HAAR_HORIZONTAL mmx, 1
301%endif
302
303;;INIT_XMM
304INIT_XMM
305COMPOSE_VERTICAL sse2
306HAAR_HORIZONTAL sse2, 0
307HAAR_HORIZONTAL sse2, 1
308