Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / vp3dsp_mmx.c @ be449fca

History | View | Annotate | Download (12 KB)

1
/*
2
 * Copyright (C) 2004 the ffmpeg project
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20

    
21
/**
22
 * @file vp3dsp_mmx.c
23
 * MMX-optimized functions cribbed from the original VP3 source code.
24
 */
25

    
26
#include "libavcodec/dsputil.h"
27
#include "dsputil_mmx.h"
28

    
29
extern const uint16_t ff_vp3_idct_data[];
30

    
31
/* from original comments: The Macro does IDct on 4 1-D Dcts */
32
#define BeginIDCT() \
33
    "movq   "I(3)", %%mm2 \n\t" \
34
    "movq   "C(3)", %%mm6 \n\t" \
35
    "movq    %%mm2, %%mm4 \n\t" \
36
    "movq   "J(5)", %%mm7 \n\t" \
37
    "pmulhw  %%mm6, %%mm4 \n\t"    /* r4 = c3*i3 - i3 */ \
38
    "movq   "C(5)", %%mm1 \n\t" \
39
    "pmulhw  %%mm7, %%mm6 \n\t"    /* r6 = c3*i5 - i5 */ \
40
    "movq    %%mm1, %%mm5 \n\t" \
41
    "pmulhw  %%mm2, %%mm1 \n\t"    /* r1 = c5*i3 - i3 */ \
42
    "movq   "I(1)", %%mm3 \n\t" \
43
    "pmulhw  %%mm7, %%mm5 \n\t"    /* r5 = c5*i5 - i5 */ \
44
    "movq   "C(1)", %%mm0 \n\t" \
45
    "paddw   %%mm2, %%mm4 \n\t"    /* r4 = c3*i3 */ \
46
    "paddw   %%mm7, %%mm6 \n\t"    /* r6 = c3*i5 */ \
47
    "paddw   %%mm1, %%mm2 \n\t"    /* r2 = c5*i3 */ \
48
    "movq   "J(7)", %%mm1 \n\t" \
49
    "paddw   %%mm5, %%mm7 \n\t"    /* r7 = c5*i5 */ \
50
    "movq    %%mm0, %%mm5 \n\t"    /* r5 = c1 */ \
51
    "pmulhw  %%mm3, %%mm0 \n\t"    /* r0 = c1*i1 - i1 */ \
52
    "paddsw  %%mm7, %%mm4 \n\t"    /* r4 = C = c3*i3 + c5*i5 */ \
53
    "pmulhw  %%mm1, %%mm5 \n\t"    /* r5 = c1*i7 - i7 */ \
54
    "movq   "C(7)", %%mm7 \n\t" \
55
    "psubsw  %%mm2, %%mm6 \n\t"    /* r6 = D = c3*i5 - c5*i3 */ \
56
    "paddw   %%mm3, %%mm0 \n\t"    /* r0 = c1*i1 */ \
57
    "pmulhw  %%mm7, %%mm3 \n\t"    /* r3 = c7*i1 */ \
58
    "movq   "I(2)", %%mm2 \n\t" \
59
    "pmulhw  %%mm1, %%mm7 \n\t"    /* r7 = c7*i7 */ \
60
    "paddw   %%mm1, %%mm5 \n\t"    /* r5 = c1*i7 */ \
61
    "movq    %%mm2, %%mm1 \n\t"    /* r1 = i2 */ \
62
    "pmulhw "C(2)", %%mm2 \n\t"    /* r2 = c2*i2 - i2 */ \
63
    "psubsw  %%mm5, %%mm3 \n\t"    /* r3 = B = c7*i1 - c1*i7 */ \
64
    "movq   "J(6)", %%mm5 \n\t" \
65
    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = A = c1*i1 + c7*i7 */ \
66
    "movq    %%mm5, %%mm7 \n\t"    /* r7 = i6 */ \
67
    "psubsw  %%mm4, %%mm0 \n\t"    /* r0 = A - C */ \
68
    "pmulhw "C(2)", %%mm5 \n\t"    /* r5 = c2*i6 - i6 */ \
69
    "paddw   %%mm1, %%mm2 \n\t"    /* r2 = c2*i2 */ \
70
    "pmulhw "C(6)", %%mm1 \n\t"    /* r1 = c6*i2 */ \
71
    "paddsw  %%mm4, %%mm4 \n\t"    /* r4 = C + C */ \
72
    "paddsw  %%mm0, %%mm4 \n\t"    /* r4 = C. = A + C */ \
73
    "psubsw  %%mm6, %%mm3 \n\t"    /* r3 = B - D */ \
74
    "paddw   %%mm7, %%mm5 \n\t"    /* r5 = c2*i6 */ \
75
    "paddsw  %%mm6, %%mm6 \n\t"    /* r6 = D + D */ \
76
    "pmulhw "C(6)", %%mm7 \n\t"    /* r7 = c6*i6 */ \
77
    "paddsw  %%mm3, %%mm6 \n\t"    /* r6 = D. = B + D */ \
78
    "movq    %%mm4, "I(1)"\n\t"    /* save C. at I(1) */ \
79
    "psubsw  %%mm5, %%mm1 \n\t"    /* r1 = H = c6*i2 - c2*i6 */ \
80
    "movq   "C(4)", %%mm4 \n\t" \
81
    "movq    %%mm3, %%mm5 \n\t"    /* r5 = B - D */ \
82
    "pmulhw  %%mm4, %%mm3 \n\t"    /* r3 = (c4 - 1) * (B - D) */ \
83
    "paddsw  %%mm2, %%mm7 \n\t"    /* r3 = (c4 - 1) * (B - D) */ \
84
    "movq    %%mm6, "I(2)"\n\t"    /* save D. at I(2) */ \
85
    "movq    %%mm0, %%mm2 \n\t"    /* r2 = A - C */ \
86
    "movq   "I(0)", %%mm6 \n\t" \
87
    "pmulhw  %%mm4, %%mm0 \n\t"    /* r0 = (c4 - 1) * (A - C) */ \
88
    "paddw   %%mm3, %%mm5 \n\t"    /* r5 = B. = c4 * (B - D) */ \
89
    "movq   "J(4)", %%mm3 \n\t" \
90
    "psubsw  %%mm1, %%mm5 \n\t"    /* r5 = B.. = B. - H */ \
91
    "paddw   %%mm0, %%mm2 \n\t"    /* r0 = A. = c4 * (A - C) */ \
92
    "psubsw  %%mm3, %%mm6 \n\t"    /* r6 = i0 - i4 */ \
93
    "movq    %%mm6, %%mm0 \n\t" \
94
    "pmulhw  %%mm4, %%mm6 \n\t"    /* r6 = (c4 - 1) * (i0 - i4) */ \
95
    "paddsw  %%mm3, %%mm3 \n\t"    /* r3 = i4 + i4 */ \
96
    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H + H */ \
97
    "paddsw  %%mm0, %%mm3 \n\t"    /* r3 = i0 + i4 */ \
98
    "paddsw  %%mm5, %%mm1 \n\t"    /* r1 = H. = B + H */ \
99
    "pmulhw  %%mm3, %%mm4 \n\t"    /* r4 = (c4 - 1) * (i0 + i4) */ \
100
    "paddsw  %%mm0, %%mm6 \n\t"    /* r6 = F = c4 * (i0 - i4) */ \
101
    "psubsw  %%mm2, %%mm6 \n\t"    /* r6 = F. = F - A. */ \
102
    "paddsw  %%mm2, %%mm2 \n\t"    /* r2 = A. + A. */ \
103
    "movq   "I(1)", %%mm0 \n\t"    /* r0 = C. */ \
104
    "paddsw  %%mm6, %%mm2 \n\t"    /* r2 = A.. = F + A. */ \
105
    "paddw   %%mm3, %%mm4 \n\t"    /* r4 = E = c4 * (i0 + i4) */ \
106
    "psubsw  %%mm1, %%mm2 \n\t"    /* r2 = R2 = A.. - H. */
107

    
108
/* RowIDCT gets ready to transpose */
109
#define RowIDCT() \
110
    BeginIDCT() \
111
    "movq   "I(2)", %%mm3 \n\t"    /* r3 = D. */ \
112
    "psubsw  %%mm7, %%mm4 \n\t"    /* r4 = E. = E - G */ \
113
    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H. + H. */ \
114
    "paddsw  %%mm7, %%mm7 \n\t"    /* r7 = G + G */ \
115
    "paddsw  %%mm2, %%mm1 \n\t"    /* r1 = R1 = A.. + H. */ \
116
    "paddsw  %%mm4, %%mm7 \n\t"    /* r1 = R1 = A.. + H. */ \
117
    "psubsw  %%mm3, %%mm4 \n\t"    /* r4 = R4 = E. - D. */ \
118
    "paddsw  %%mm3, %%mm3 \n\t" \
119
    "psubsw  %%mm5, %%mm6 \n\t"    /* r6 = R6 = F. - B.. */ \
120
    "paddsw  %%mm5, %%mm5 \n\t" \
121
    "paddsw  %%mm4, %%mm3 \n\t"    /* r3 = R3 = E. + D. */ \
122
    "paddsw  %%mm6, %%mm5 \n\t"    /* r5 = R5 = F. + B.. */ \
123
    "psubsw  %%mm0, %%mm7 \n\t"    /* r7 = R7 = G. - C. */ \
124
    "paddsw  %%mm0, %%mm0 \n\t" \
125
    "movq    %%mm1, "I(1)"\n\t"    /* save R1 */ \
126
    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = R0 = G. + C. */
127

    
128
/* Column IDCT normalizes and stores final results */
129
#define ColumnIDCT() \
130
    BeginIDCT() \
131
    "paddsw "OC_8", %%mm2 \n\t"    /* adjust R2 (and R1) for shift */ \
132
    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H. + H. */ \
133
    "paddsw  %%mm2, %%mm1 \n\t"    /* r1 = R1 = A.. + H. */ \
134
    "psraw      $4, %%mm2 \n\t"    /* r2 = NR2 */ \
135
    "psubsw  %%mm7, %%mm4 \n\t"    /* r4 = E. = E - G */ \
136
    "psraw      $4, %%mm1 \n\t"    /* r1 = NR1 */ \
137
    "movq   "I(2)", %%mm3 \n\t"    /* r3 = D. */ \
138
    "paddsw  %%mm7, %%mm7 \n\t"    /* r7 = G + G */ \
139
    "movq    %%mm2, "I(2)"\n\t"    /* store NR2 at I2 */ \
140
    "paddsw  %%mm4, %%mm7 \n\t"    /* r7 = G. = E + G */ \
141
    "movq    %%mm1, "I(1)"\n\t"    /* store NR1 at I1 */ \
142
    "psubsw  %%mm3, %%mm4 \n\t"    /* r4 = R4 = E. - D. */ \
143
    "paddsw "OC_8", %%mm4 \n\t"    /* adjust R4 (and R3) for shift */ \
144
    "paddsw  %%mm3, %%mm3 \n\t"    /* r3 = D. + D. */ \
145
    "paddsw  %%mm4, %%mm3 \n\t"    /* r3 = R3 = E. + D. */ \
146
    "psraw      $4, %%mm4 \n\t"    /* r4 = NR4 */ \
147
    "psubsw  %%mm5, %%mm6 \n\t"    /* r6 = R6 = F. - B.. */ \
148
    "psraw      $4, %%mm3 \n\t"    /* r3 = NR3 */ \
149
    "paddsw "OC_8", %%mm6 \n\t"    /* adjust R6 (and R5) for shift */ \
150
    "paddsw  %%mm5, %%mm5 \n\t"    /* r5 = B.. + B.. */ \
151
    "paddsw  %%mm6, %%mm5 \n\t"    /* r5 = R5 = F. + B.. */ \
152
    "psraw      $4, %%mm6 \n\t"    /* r6 = NR6 */ \
153
    "movq    %%mm4, "J(4)"\n\t"    /* store NR4 at J4 */ \
154
    "psraw      $4, %%mm5 \n\t"    /* r5 = NR5 */ \
155
    "movq    %%mm3, "I(3)"\n\t"    /* store NR3 at I3 */ \
156
    "psubsw  %%mm0, %%mm7 \n\t"    /* r7 = R7 = G. - C. */ \
157
    "paddsw "OC_8", %%mm7 \n\t"    /* adjust R7 (and R0) for shift */ \
158
    "paddsw  %%mm0, %%mm0 \n\t"    /* r0 = C. + C. */ \
159
    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = R0 = G. + C. */ \
160
    "psraw      $4, %%mm7 \n\t"    /* r7 = NR7 */ \
161
    "movq    %%mm6, "J(6)"\n\t"    /* store NR6 at J6 */ \
162
    "psraw      $4, %%mm0 \n\t"    /* r0 = NR0 */ \
163
    "movq    %%mm5, "J(5)"\n\t"    /* store NR5 at J5 */ \
164
    "movq    %%mm7, "J(7)"\n\t"    /* store NR7 at J7 */ \
165
    "movq    %%mm0, "I(0)"\n\t"    /* store NR0 at I0 */
166

    
167
/* Following macro does two 4x4 transposes in place.
168

169
  At entry (we assume):
170

171
    r0 = a3 a2 a1 a0
172
    I(1) = b3 b2 b1 b0
173
    r2 = c3 c2 c1 c0
174
    r3 = d3 d2 d1 d0
175

176
    r4 = e3 e2 e1 e0
177
    r5 = f3 f2 f1 f0
178
    r6 = g3 g2 g1 g0
179
    r7 = h3 h2 h1 h0
180

181
  At exit, we have:
182

183
    I(0) = d0 c0 b0 a0
184
    I(1) = d1 c1 b1 a1
185
    I(2) = d2 c2 b2 a2
186
    I(3) = d3 c3 b3 a3
187

188
    J(4) = h0 g0 f0 e0
189
    J(5) = h1 g1 f1 e1
190
    J(6) = h2 g2 f2 e2
191
    J(7) = h3 g3 f3 e3
192

193
   I(0) I(1) I(2) I(3)  is the transpose of r0 I(1) r2 r3.
194
   J(4) J(5) J(6) J(7)  is the transpose of r4 r5 r6 r7.
195

196
   Since r1 is free at entry, we calculate the Js first. */
197
#define Transpose() \
198
    "movq       %%mm4, %%mm1 \n\t"    /* r1 = e3 e2 e1 e0 */ \
199
    "punpcklwd  %%mm5, %%mm4 \n\t"    /* r4 = f1 e1 f0 e0 */ \
200
    "movq       %%mm0, "I(0)"\n\t"    /* save a3 a2 a1 a0 */ \
201
    "punpckhwd  %%mm5, %%mm1 \n\t"    /* r1 = f3 e3 f2 e2 */ \
202
    "movq       %%mm6, %%mm0 \n\t"    /* r0 = g3 g2 g1 g0 */ \
203
    "punpcklwd  %%mm7, %%mm6 \n\t"    /* r6 = h1 g1 h0 g0 */ \
204
    "movq       %%mm4, %%mm5 \n\t"    /* r5 = f1 e1 f0 e0 */ \
205
    "punpckldq  %%mm6, %%mm4 \n\t"    /* r4 = h0 g0 f0 e0 = R4 */ \
206
    "punpckhdq  %%mm6, %%mm5 \n\t"    /* r5 = h1 g1 f1 e1 = R5 */ \
207
    "movq       %%mm1, %%mm6 \n\t"    /* r6 = f3 e3 f2 e2 */ \
208
    "movq       %%mm4, "J(4)"\n\t" \
209
    "punpckhwd  %%mm7, %%mm0 \n\t"    /* r0 = h3 g3 h2 g2 */ \
210
    "movq       %%mm5, "J(5)"\n\t" \
211
    "punpckhdq  %%mm0, %%mm6 \n\t"    /* r6 = h3 g3 f3 e3 = R7 */ \
212
    "movq      "I(0)", %%mm4 \n\t"    /* r4 = a3 a2 a1 a0 */ \
213
    "punpckldq  %%mm0, %%mm1 \n\t"    /* r1 = h2 g2 f2 e2 = R6 */ \
214
    "movq      "I(1)", %%mm5 \n\t"    /* r5 = b3 b2 b1 b0 */ \
215
    "movq       %%mm4, %%mm0 \n\t"    /* r0 = a3 a2 a1 a0 */ \
216
    "movq       %%mm6, "J(7)"\n\t" \
217
    "punpcklwd  %%mm5, %%mm0 \n\t"    /* r0 = b1 a1 b0 a0 */ \
218
    "movq       %%mm1, "J(6)"\n\t" \
219
    "punpckhwd  %%mm5, %%mm4 \n\t"    /* r4 = b3 a3 b2 a2 */ \
220
    "movq       %%mm2, %%mm5 \n\t"    /* r5 = c3 c2 c1 c0 */ \
221
    "punpcklwd  %%mm3, %%mm2 \n\t"    /* r2 = d1 c1 d0 c0 */ \
222
    "movq       %%mm0, %%mm1 \n\t"    /* r1 = b1 a1 b0 a0 */ \
223
    "punpckldq  %%mm2, %%mm0 \n\t"    /* r0 = d0 c0 b0 a0 = R0 */ \
224
    "punpckhdq  %%mm2, %%mm1 \n\t"    /* r1 = d1 c1 b1 a1 = R1 */ \
225
    "movq       %%mm4, %%mm2 \n\t"    /* r2 = b3 a3 b2 a2 */ \
226
    "movq       %%mm0, "I(0)"\n\t" \
227
    "punpckhwd  %%mm3, %%mm5 \n\t"    /* r5 = d3 c3 d2 c2 */ \
228
    "movq       %%mm1, "I(1)"\n\t" \
229
    "punpckhdq  %%mm5, %%mm4 \n\t"    /* r4 = d3 c3 b3 a3 = R3 */ \
230
    "punpckldq  %%mm5, %%mm2 \n\t"    /* r2 = d2 c2 b2 a2 = R2 */ \
231
    "movq       %%mm4, "I(3)"\n\t" \
232
    "movq       %%mm2, "I(2)"\n\t"
233

    
234
void ff_vp3_idct_mmx(int16_t *output_data)
235
{
236
    /* eax = quantized input
237
     * ebx = dequantizer matrix
238
     * ecx = IDCT constants
239
     *  M(I) = ecx + MaskOffset(0) + I * 8
240
     *  C(I) = ecx + CosineOffset(32) + (I-1) * 8
241
     * edx = output
242
     * r0..r7 = mm0..mm7
243
     */
244

    
245
#define C(x) AV_STRINGIFY(16*(x-1))"(%1)"
246
#define OC_8 "%2"
247

    
248
    /* at this point, function has completed dequantization + dezigzag +
249
     * partial transposition; now do the idct itself */
250
#define I(x) AV_STRINGIFY(16* x       )"(%0)"
251
#define J(x) AV_STRINGIFY(16*(x-4) + 8)"(%0)"
252

    
253
    __asm__ volatile (
254
        RowIDCT()
255
        Transpose()
256

    
257
#undef I
258
#undef J
259
#define I(x) AV_STRINGIFY(16* x    + 64)"(%0)"
260
#define J(x) AV_STRINGIFY(16*(x-4) + 72)"(%0)"
261

    
262
        RowIDCT()
263
        Transpose()
264

    
265
#undef I
266
#undef J
267
#define I(x) AV_STRINGIFY(16*x)"(%0)"
268
#define J(x) AV_STRINGIFY(16*x)"(%0)"
269

    
270
        ColumnIDCT()
271

    
272
#undef I
273
#undef J
274
#define I(x) AV_STRINGIFY(16*x + 8)"(%0)"
275
#define J(x) AV_STRINGIFY(16*x + 8)"(%0)"
276

    
277
        ColumnIDCT()
278
        :: "r"(output_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8)
279
    );
280
#undef I
281
#undef J
282

    
283
}
284

    
285
void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
286
{
287
    ff_vp3_idct_mmx(block);
288
    put_signed_pixels_clamped_mmx(block, dest, line_size);
289
}
290

    
291
void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
292
{
293
    ff_vp3_idct_mmx(block);
294
    add_pixels_clamped_mmx(block, dest, line_size);
295
}