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
23 * MMX-optimized functions cribbed from the original VP3 source code.
24 */
25
26#include "libavutil/x86_cpu.h"
27#include "libavcodec/dsputil.h"
28#include "dsputil_mmx.h"
29#include "vp3dsp_mmx.h"
30
31extern const uint16_t ff_vp3_idct_data[];
32
33// this is off by one or two for some cases when filter_limit is greater than 63
34// in:  p0 in mm6, p1 in mm4, p2 in mm2, p3 in mm1
35// out: p1 in mm4, p2 in mm3
36#define VP3_LOOP_FILTER(flim) \
37    "movq       %%mm6, %%mm7 \n\t" \
38    "pand    "MANGLE(ff_pb_7 )", %%mm6 \n\t" /* p0&7 */ \
39    "psrlw         $3, %%mm7 \n\t" \
40    "pand    "MANGLE(ff_pb_1F)", %%mm7 \n\t" /* p0>>3 */ \
41    "movq       %%mm2, %%mm3 \n\t" /* mm3 = p2 */ \
42    "pxor       %%mm4, %%mm2 \n\t" \
43    "pand    "MANGLE(ff_pb_1 )", %%mm2 \n\t" /* (p2^p1)&1 */ \
44    "movq       %%mm2, %%mm5 \n\t" \
45    "paddb      %%mm2, %%mm2 \n\t" \
46    "paddb      %%mm5, %%mm2 \n\t" /* 3*(p2^p1)&1 */ \
47    "paddb      %%mm6, %%mm2 \n\t" /* extra bits lost in shifts */ \
48    "pcmpeqb    %%mm0, %%mm0 \n\t" \
49    "pxor       %%mm0, %%mm1 \n\t" /* 255 - p3 */ \
50    "pavgb      %%mm2, %%mm1 \n\t" /* (256 - p3 + extrabits) >> 1 */ \
51    "pxor       %%mm4, %%mm0 \n\t" /* 255 - p1 */ \
52    "pavgb      %%mm3, %%mm0 \n\t" /* (256 + p2-p1) >> 1 */ \
53    "paddb   "MANGLE(ff_pb_3 )", %%mm1 \n\t" \
54    "pavgb      %%mm0, %%mm1 \n\t" /* 128+2+(   p2-p1  - p3) >> 2 */ \
55    "pavgb      %%mm0, %%mm1 \n\t" /* 128+1+(3*(p2-p1) - p3) >> 3 */ \
56    "paddusb    %%mm1, %%mm7 \n\t" /* d+128+1 */ \
57    "movq    "MANGLE(ff_pb_81)", %%mm6 \n\t" \
58    "psubusb    %%mm7, %%mm6 \n\t" \
59    "psubusb "MANGLE(ff_pb_81)", %%mm7 \n\t" \
60\
61    "movq     "#flim", %%mm5 \n\t" \
62    "pminub     %%mm5, %%mm6 \n\t" \
63    "pminub     %%mm5, %%mm7 \n\t" \
64    "movq       %%mm6, %%mm0 \n\t" \
65    "movq       %%mm7, %%mm1 \n\t" \
66    "paddb      %%mm6, %%mm6 \n\t" \
67    "paddb      %%mm7, %%mm7 \n\t" \
68    "pminub     %%mm5, %%mm6 \n\t" \
69    "pminub     %%mm5, %%mm7 \n\t" \
70    "psubb      %%mm0, %%mm6 \n\t" \
71    "psubb      %%mm1, %%mm7 \n\t" \
72    "paddusb    %%mm7, %%mm4 \n\t" \
73    "psubusb    %%mm6, %%mm4 \n\t" \
74    "psubusb    %%mm7, %%mm3 \n\t" \
75    "paddusb    %%mm6, %%mm3 \n\t"
76
77#define STORE_4_WORDS(dst0, dst1, dst2, dst3, mm) \
78    "movd "#mm", %0        \n\t" \
79    "movw   %w0, -1"#dst0" \n\t" \
80    "psrlq  $32, "#mm"     \n\t" \
81    "shr    $16, %0        \n\t" \
82    "movw   %w0, -1"#dst1" \n\t" \
83    "movd "#mm", %0        \n\t" \
84    "movw   %w0, -1"#dst2" \n\t" \
85    "shr    $16, %0        \n\t" \
86    "movw   %w0, -1"#dst3" \n\t"
87
88void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values)
89{
90    __asm__ volatile(
91        "movq          %0, %%mm6 \n\t"
92        "movq          %1, %%mm4 \n\t"
93        "movq          %2, %%mm2 \n\t"
94        "movq          %3, %%mm1 \n\t"
95
96        VP3_LOOP_FILTER(%4)
97
98        "movq       %%mm4, %1    \n\t"
99        "movq       %%mm3, %2    \n\t"
100
101        : "+m" (*(uint64_t*)(src - 2*stride)),
102          "+m" (*(uint64_t*)(src - 1*stride)),
103          "+m" (*(uint64_t*)(src + 0*stride)),
104          "+m" (*(uint64_t*)(src + 1*stride))
105        : "m"(*(uint64_t*)(bounding_values+129))
106    );
107}
108
109void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values)
110{
111    x86_reg tmp;
112
113    __asm__ volatile(
114        "movd -2(%1),      %%mm6 \n\t"
115        "movd -2(%1,%3),   %%mm0 \n\t"
116        "movd -2(%1,%3,2), %%mm1 \n\t"
117        "movd -2(%1,%4),   %%mm4 \n\t"
118
119        TRANSPOSE8x4(%%mm6, %%mm0, %%mm1, %%mm4, -2(%2), -2(%2,%3), -2(%2,%3,2), -2(%2,%4), %%mm2)
120        VP3_LOOP_FILTER(%5)
121        SBUTTERFLY(%%mm4, %%mm3, %%mm5, bw, q)
122
123        STORE_4_WORDS((%1), (%1,%3), (%1,%3,2), (%1,%4), %%mm4)
124        STORE_4_WORDS((%2), (%2,%3), (%2,%3,2), (%2,%4), %%mm5)
125
126        : "=&r"(tmp)
127        : "r"(src), "r"(src+4*stride), "r"((x86_reg)stride), "r"((x86_reg)3*stride),
128          "m"(*(uint64_t*)(bounding_values+129))
129        : "memory"
130    );
131}
132
133/* from original comments: The Macro does IDct on 4 1-D Dcts */
134#define BeginIDCT() \
135    "movq   "I(3)", %%mm2 \n\t" \
136    "movq   "C(3)", %%mm6 \n\t" \
137    "movq    %%mm2, %%mm4 \n\t" \
138    "movq   "J(5)", %%mm7 \n\t" \
139    "pmulhw  %%mm6, %%mm4 \n\t"    /* r4 = c3*i3 - i3 */ \
140    "movq   "C(5)", %%mm1 \n\t" \
141    "pmulhw  %%mm7, %%mm6 \n\t"    /* r6 = c3*i5 - i5 */ \
142    "movq    %%mm1, %%mm5 \n\t" \
143    "pmulhw  %%mm2, %%mm1 \n\t"    /* r1 = c5*i3 - i3 */ \
144    "movq   "I(1)", %%mm3 \n\t" \
145    "pmulhw  %%mm7, %%mm5 \n\t"    /* r5 = c5*i5 - i5 */ \
146    "movq   "C(1)", %%mm0 \n\t" \
147    "paddw   %%mm2, %%mm4 \n\t"    /* r4 = c3*i3 */ \
148    "paddw   %%mm7, %%mm6 \n\t"    /* r6 = c3*i5 */ \
149    "paddw   %%mm1, %%mm2 \n\t"    /* r2 = c5*i3 */ \
150    "movq   "J(7)", %%mm1 \n\t" \
151    "paddw   %%mm5, %%mm7 \n\t"    /* r7 = c5*i5 */ \
152    "movq    %%mm0, %%mm5 \n\t"    /* r5 = c1 */ \
153    "pmulhw  %%mm3, %%mm0 \n\t"    /* r0 = c1*i1 - i1 */ \
154    "paddsw  %%mm7, %%mm4 \n\t"    /* r4 = C = c3*i3 + c5*i5 */ \
155    "pmulhw  %%mm1, %%mm5 \n\t"    /* r5 = c1*i7 - i7 */ \
156    "movq   "C(7)", %%mm7 \n\t" \
157    "psubsw  %%mm2, %%mm6 \n\t"    /* r6 = D = c3*i5 - c5*i3 */ \
158    "paddw   %%mm3, %%mm0 \n\t"    /* r0 = c1*i1 */ \
159    "pmulhw  %%mm7, %%mm3 \n\t"    /* r3 = c7*i1 */ \
160    "movq   "I(2)", %%mm2 \n\t" \
161    "pmulhw  %%mm1, %%mm7 \n\t"    /* r7 = c7*i7 */ \
162    "paddw   %%mm1, %%mm5 \n\t"    /* r5 = c1*i7 */ \
163    "movq    %%mm2, %%mm1 \n\t"    /* r1 = i2 */ \
164    "pmulhw "C(2)", %%mm2 \n\t"    /* r2 = c2*i2 - i2 */ \
165    "psubsw  %%mm5, %%mm3 \n\t"    /* r3 = B = c7*i1 - c1*i7 */ \
166    "movq   "J(6)", %%mm5 \n\t" \
167    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = A = c1*i1 + c7*i7 */ \
168    "movq    %%mm5, %%mm7 \n\t"    /* r7 = i6 */ \
169    "psubsw  %%mm4, %%mm0 \n\t"    /* r0 = A - C */ \
170    "pmulhw "C(2)", %%mm5 \n\t"    /* r5 = c2*i6 - i6 */ \
171    "paddw   %%mm1, %%mm2 \n\t"    /* r2 = c2*i2 */ \
172    "pmulhw "C(6)", %%mm1 \n\t"    /* r1 = c6*i2 */ \
173    "paddsw  %%mm4, %%mm4 \n\t"    /* r4 = C + C */ \
174    "paddsw  %%mm0, %%mm4 \n\t"    /* r4 = C. = A + C */ \
175    "psubsw  %%mm6, %%mm3 \n\t"    /* r3 = B - D */ \
176    "paddw   %%mm7, %%mm5 \n\t"    /* r5 = c2*i6 */ \
177    "paddsw  %%mm6, %%mm6 \n\t"    /* r6 = D + D */ \
178    "pmulhw "C(6)", %%mm7 \n\t"    /* r7 = c6*i6 */ \
179    "paddsw  %%mm3, %%mm6 \n\t"    /* r6 = D. = B + D */ \
180    "movq    %%mm4, "I(1)"\n\t"    /* save C. at I(1) */ \
181    "psubsw  %%mm5, %%mm1 \n\t"    /* r1 = H = c6*i2 - c2*i6 */ \
182    "movq   "C(4)", %%mm4 \n\t" \
183    "movq    %%mm3, %%mm5 \n\t"    /* r5 = B - D */ \
184    "pmulhw  %%mm4, %%mm3 \n\t"    /* r3 = (c4 - 1) * (B - D) */ \
185    "paddsw  %%mm2, %%mm7 \n\t"    /* r3 = (c4 - 1) * (B - D) */ \
186    "movq    %%mm6, "I(2)"\n\t"    /* save D. at I(2) */ \
187    "movq    %%mm0, %%mm2 \n\t"    /* r2 = A - C */ \
188    "movq   "I(0)", %%mm6 \n\t" \
189    "pmulhw  %%mm4, %%mm0 \n\t"    /* r0 = (c4 - 1) * (A - C) */ \
190    "paddw   %%mm3, %%mm5 \n\t"    /* r5 = B. = c4 * (B - D) */ \
191    "movq   "J(4)", %%mm3 \n\t" \
192    "psubsw  %%mm1, %%mm5 \n\t"    /* r5 = B.. = B. - H */ \
193    "paddw   %%mm0, %%mm2 \n\t"    /* r0 = A. = c4 * (A - C) */ \
194    "psubsw  %%mm3, %%mm6 \n\t"    /* r6 = i0 - i4 */ \
195    "movq    %%mm6, %%mm0 \n\t" \
196    "pmulhw  %%mm4, %%mm6 \n\t"    /* r6 = (c4 - 1) * (i0 - i4) */ \
197    "paddsw  %%mm3, %%mm3 \n\t"    /* r3 = i4 + i4 */ \
198    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H + H */ \
199    "paddsw  %%mm0, %%mm3 \n\t"    /* r3 = i0 + i4 */ \
200    "paddsw  %%mm5, %%mm1 \n\t"    /* r1 = H. = B + H */ \
201    "pmulhw  %%mm3, %%mm4 \n\t"    /* r4 = (c4 - 1) * (i0 + i4) */ \
202    "paddsw  %%mm0, %%mm6 \n\t"    /* r6 = F = c4 * (i0 - i4) */ \
203    "psubsw  %%mm2, %%mm6 \n\t"    /* r6 = F. = F - A. */ \
204    "paddsw  %%mm2, %%mm2 \n\t"    /* r2 = A. + A. */ \
205    "movq   "I(1)", %%mm0 \n\t"    /* r0 = C. */ \
206    "paddsw  %%mm6, %%mm2 \n\t"    /* r2 = A.. = F + A. */ \
207    "paddw   %%mm3, %%mm4 \n\t"    /* r4 = E = c4 * (i0 + i4) */ \
208    "psubsw  %%mm1, %%mm2 \n\t"    /* r2 = R2 = A.. - H. */
209
210/* RowIDCT gets ready to transpose */
211#define RowIDCT() \
212    BeginIDCT() \
213    "movq   "I(2)", %%mm3 \n\t"    /* r3 = D. */ \
214    "psubsw  %%mm7, %%mm4 \n\t"    /* r4 = E. = E - G */ \
215    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H. + H. */ \
216    "paddsw  %%mm7, %%mm7 \n\t"    /* r7 = G + G */ \
217    "paddsw  %%mm2, %%mm1 \n\t"    /* r1 = R1 = A.. + H. */ \
218    "paddsw  %%mm4, %%mm7 \n\t"    /* r1 = R1 = A.. + H. */ \
219    "psubsw  %%mm3, %%mm4 \n\t"    /* r4 = R4 = E. - D. */ \
220    "paddsw  %%mm3, %%mm3 \n\t" \
221    "psubsw  %%mm5, %%mm6 \n\t"    /* r6 = R6 = F. - B.. */ \
222    "paddsw  %%mm5, %%mm5 \n\t" \
223    "paddsw  %%mm4, %%mm3 \n\t"    /* r3 = R3 = E. + D. */ \
224    "paddsw  %%mm6, %%mm5 \n\t"    /* r5 = R5 = F. + B.. */ \
225    "psubsw  %%mm0, %%mm7 \n\t"    /* r7 = R7 = G. - C. */ \
226    "paddsw  %%mm0, %%mm0 \n\t" \
227    "movq    %%mm1, "I(1)"\n\t"    /* save R1 */ \
228    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = R0 = G. + C. */
229
230/* Column IDCT normalizes and stores final results */
231#define ColumnIDCT() \
232    BeginIDCT() \
233    "paddsw "OC_8", %%mm2 \n\t"    /* adjust R2 (and R1) for shift */ \
234    "paddsw  %%mm1, %%mm1 \n\t"    /* r1 = H. + H. */ \
235    "paddsw  %%mm2, %%mm1 \n\t"    /* r1 = R1 = A.. + H. */ \
236    "psraw      $4, %%mm2 \n\t"    /* r2 = NR2 */ \
237    "psubsw  %%mm7, %%mm4 \n\t"    /* r4 = E. = E - G */ \
238    "psraw      $4, %%mm1 \n\t"    /* r1 = NR1 */ \
239    "movq   "I(2)", %%mm3 \n\t"    /* r3 = D. */ \
240    "paddsw  %%mm7, %%mm7 \n\t"    /* r7 = G + G */ \
241    "movq    %%mm2, "I(2)"\n\t"    /* store NR2 at I2 */ \
242    "paddsw  %%mm4, %%mm7 \n\t"    /* r7 = G. = E + G */ \
243    "movq    %%mm1, "I(1)"\n\t"    /* store NR1 at I1 */ \
244    "psubsw  %%mm3, %%mm4 \n\t"    /* r4 = R4 = E. - D. */ \
245    "paddsw "OC_8", %%mm4 \n\t"    /* adjust R4 (and R3) for shift */ \
246    "paddsw  %%mm3, %%mm3 \n\t"    /* r3 = D. + D. */ \
247    "paddsw  %%mm4, %%mm3 \n\t"    /* r3 = R3 = E. + D. */ \
248    "psraw      $4, %%mm4 \n\t"    /* r4 = NR4 */ \
249    "psubsw  %%mm5, %%mm6 \n\t"    /* r6 = R6 = F. - B.. */ \
250    "psraw      $4, %%mm3 \n\t"    /* r3 = NR3 */ \
251    "paddsw "OC_8", %%mm6 \n\t"    /* adjust R6 (and R5) for shift */ \
252    "paddsw  %%mm5, %%mm5 \n\t"    /* r5 = B.. + B.. */ \
253    "paddsw  %%mm6, %%mm5 \n\t"    /* r5 = R5 = F. + B.. */ \
254    "psraw      $4, %%mm6 \n\t"    /* r6 = NR6 */ \
255    "movq    %%mm4, "J(4)"\n\t"    /* store NR4 at J4 */ \
256    "psraw      $4, %%mm5 \n\t"    /* r5 = NR5 */ \
257    "movq    %%mm3, "I(3)"\n\t"    /* store NR3 at I3 */ \
258    "psubsw  %%mm0, %%mm7 \n\t"    /* r7 = R7 = G. - C. */ \
259    "paddsw "OC_8", %%mm7 \n\t"    /* adjust R7 (and R0) for shift */ \
260    "paddsw  %%mm0, %%mm0 \n\t"    /* r0 = C. + C. */ \
261    "paddsw  %%mm7, %%mm0 \n\t"    /* r0 = R0 = G. + C. */ \
262    "psraw      $4, %%mm7 \n\t"    /* r7 = NR7 */ \
263    "movq    %%mm6, "J(6)"\n\t"    /* store NR6 at J6 */ \
264    "psraw      $4, %%mm0 \n\t"    /* r0 = NR0 */ \
265    "movq    %%mm5, "J(5)"\n\t"    /* store NR5 at J5 */ \
266    "movq    %%mm7, "J(7)"\n\t"    /* store NR7 at J7 */ \
267    "movq    %%mm0, "I(0)"\n\t"    /* store NR0 at I0 */
268
269/* Following macro does two 4x4 transposes in place.
270
271  At entry (we assume):
272
273    r0 = a3 a2 a1 a0
274    I(1) = b3 b2 b1 b0
275    r2 = c3 c2 c1 c0
276    r3 = d3 d2 d1 d0
277
278    r4 = e3 e2 e1 e0
279    r5 = f3 f2 f1 f0
280    r6 = g3 g2 g1 g0
281    r7 = h3 h2 h1 h0
282
283  At exit, we have:
284
285    I(0) = d0 c0 b0 a0
286    I(1) = d1 c1 b1 a1
287    I(2) = d2 c2 b2 a2
288    I(3) = d3 c3 b3 a3
289
290    J(4) = h0 g0 f0 e0
291    J(5) = h1 g1 f1 e1
292    J(6) = h2 g2 f2 e2
293    J(7) = h3 g3 f3 e3
294
295   I(0) I(1) I(2) I(3)  is the transpose of r0 I(1) r2 r3.
296   J(4) J(5) J(6) J(7)  is the transpose of r4 r5 r6 r7.
297
298   Since r1 is free at entry, we calculate the Js first. */
299#define Transpose() \
300    "movq       %%mm4, %%mm1 \n\t"    /* r1 = e3 e2 e1 e0 */ \
301    "punpcklwd  %%mm5, %%mm4 \n\t"    /* r4 = f1 e1 f0 e0 */ \
302    "movq       %%mm0, "I(0)"\n\t"    /* save a3 a2 a1 a0 */ \
303    "punpckhwd  %%mm5, %%mm1 \n\t"    /* r1 = f3 e3 f2 e2 */ \
304    "movq       %%mm6, %%mm0 \n\t"    /* r0 = g3 g2 g1 g0 */ \
305    "punpcklwd  %%mm7, %%mm6 \n\t"    /* r6 = h1 g1 h0 g0 */ \
306    "movq       %%mm4, %%mm5 \n\t"    /* r5 = f1 e1 f0 e0 */ \
307    "punpckldq  %%mm6, %%mm4 \n\t"    /* r4 = h0 g0 f0 e0 = R4 */ \
308    "punpckhdq  %%mm6, %%mm5 \n\t"    /* r5 = h1 g1 f1 e1 = R5 */ \
309    "movq       %%mm1, %%mm6 \n\t"    /* r6 = f3 e3 f2 e2 */ \
310    "movq       %%mm4, "J(4)"\n\t" \
311    "punpckhwd  %%mm7, %%mm0 \n\t"    /* r0 = h3 g3 h2 g2 */ \
312    "movq       %%mm5, "J(5)"\n\t" \
313    "punpckhdq  %%mm0, %%mm6 \n\t"    /* r6 = h3 g3 f3 e3 = R7 */ \
314    "movq      "I(0)", %%mm4 \n\t"    /* r4 = a3 a2 a1 a0 */ \
315    "punpckldq  %%mm0, %%mm1 \n\t"    /* r1 = h2 g2 f2 e2 = R6 */ \
316    "movq      "I(1)", %%mm5 \n\t"    /* r5 = b3 b2 b1 b0 */ \
317    "movq       %%mm4, %%mm0 \n\t"    /* r0 = a3 a2 a1 a0 */ \
318    "movq       %%mm6, "J(7)"\n\t" \
319    "punpcklwd  %%mm5, %%mm0 \n\t"    /* r0 = b1 a1 b0 a0 */ \
320    "movq       %%mm1, "J(6)"\n\t" \
321    "punpckhwd  %%mm5, %%mm4 \n\t"    /* r4 = b3 a3 b2 a2 */ \
322    "movq       %%mm2, %%mm5 \n\t"    /* r5 = c3 c2 c1 c0 */ \
323    "punpcklwd  %%mm3, %%mm2 \n\t"    /* r2 = d1 c1 d0 c0 */ \
324    "movq       %%mm0, %%mm1 \n\t"    /* r1 = b1 a1 b0 a0 */ \
325    "punpckldq  %%mm2, %%mm0 \n\t"    /* r0 = d0 c0 b0 a0 = R0 */ \
326    "punpckhdq  %%mm2, %%mm1 \n\t"    /* r1 = d1 c1 b1 a1 = R1 */ \
327    "movq       %%mm4, %%mm2 \n\t"    /* r2 = b3 a3 b2 a2 */ \
328    "movq       %%mm0, "I(0)"\n\t" \
329    "punpckhwd  %%mm3, %%mm5 \n\t"    /* r5 = d3 c3 d2 c2 */ \
330    "movq       %%mm1, "I(1)"\n\t" \
331    "punpckhdq  %%mm5, %%mm4 \n\t"    /* r4 = d3 c3 b3 a3 = R3 */ \
332    "punpckldq  %%mm5, %%mm2 \n\t"    /* r2 = d2 c2 b2 a2 = R2 */ \
333    "movq       %%mm4, "I(3)"\n\t" \
334    "movq       %%mm2, "I(2)"\n\t"
335
336void ff_vp3_idct_mmx(int16_t *output_data)
337{
338    /* eax = quantized input
339     * ebx = dequantizer matrix
340     * ecx = IDCT constants
341     *  M(I) = ecx + MaskOffset(0) + I * 8
342     *  C(I) = ecx + CosineOffset(32) + (I-1) * 8
343     * edx = output
344     * r0..r7 = mm0..mm7
345     */
346
347#define C(x) AV_STRINGIFY(16*(x-1))"(%1)"
348#define OC_8 "%2"
349
350    /* at this point, function has completed dequantization + dezigzag +
351     * partial transposition; now do the idct itself */
352#define I(x) AV_STRINGIFY(16* x       )"(%0)"
353#define J(x) AV_STRINGIFY(16*(x-4) + 8)"(%0)"
354
355    __asm__ volatile (
356        RowIDCT()
357        Transpose()
358
359#undef I
360#undef J
361#define I(x) AV_STRINGIFY(16* x    + 64)"(%0)"
362#define J(x) AV_STRINGIFY(16*(x-4) + 72)"(%0)"
363
364        RowIDCT()
365        Transpose()
366
367#undef I
368#undef J
369#define I(x) AV_STRINGIFY(16*x)"(%0)"
370#define J(x) AV_STRINGIFY(16*x)"(%0)"
371
372        ColumnIDCT()
373
374#undef I
375#undef J
376#define I(x) AV_STRINGIFY(16*x + 8)"(%0)"
377#define J(x) AV_STRINGIFY(16*x + 8)"(%0)"
378
379        ColumnIDCT()
380        :: "r"(output_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8)
381    );
382#undef I
383#undef J
384
385}
386
387void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
388{
389    ff_vp3_idct_mmx(block);
390    put_signed_pixels_clamped_mmx(block, dest, line_size);
391}
392
393void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
394{
395    ff_vp3_idct_mmx(block);
396    add_pixels_clamped_mmx(block, dest, line_size);
397}
398
399void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int linesize, const DCTELEM *block)
400{
401    int dc = block[0];
402    dc = (46341*dc)>>16;
403    dc = (46341*dc + (8<<16))>>20;
404
405    __asm__ volatile(
406        "movd          %3, %%mm0 \n\t"
407        "pshufw $0, %%mm0, %%mm0 \n\t"
408        "pxor       %%mm1, %%mm1 \n\t"
409        "psubw      %%mm0, %%mm1 \n\t"
410        "packuswb   %%mm0, %%mm0 \n\t"
411        "packuswb   %%mm1, %%mm1 \n\t"
412
413#define DC_ADD \
414        "movq        (%0), %%mm2 \n\t" \
415        "movq     (%0,%1), %%mm3 \n\t" \
416        "paddusb    %%mm0, %%mm2 \n\t" \
417        "movq   (%0,%1,2), %%mm4 \n\t" \
418        "paddusb    %%mm0, %%mm3 \n\t" \
419        "movq     (%0,%2), %%mm5 \n\t" \
420        "paddusb    %%mm0, %%mm4 \n\t" \
421        "paddusb    %%mm0, %%mm5 \n\t" \
422        "psubusb    %%mm1, %%mm2 \n\t" \
423        "psubusb    %%mm1, %%mm3 \n\t" \
424        "movq       %%mm2, (%0)  \n\t" \
425        "psubusb    %%mm1, %%mm4 \n\t" \
426        "movq       %%mm3, (%0,%1) \n\t" \
427        "psubusb    %%mm1, %%mm5 \n\t" \
428        "movq       %%mm4, (%0,%1,2) \n\t" \
429        "movq       %%mm5, (%0,%2) \n\t"
430
431        DC_ADD
432        "lea    (%0,%1,4), %0 \n\t"
433        DC_ADD
434
435        : "+r"(dest)
436        : "r"((x86_reg)linesize), "r"((x86_reg)3*linesize), "r"(dc)
437    );
438}
439