Libav
mpegvideoenc_template.c
Go to the documentation of this file.
1 /*
2  * MPEG video MMX templates
3  *
4  * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * This file is part of Libav.
7  *
8  * Libav is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * Libav is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with Libav; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22 
23 #include <stdint.h>
24 
25 #include "libavutil/internal.h"
26 #include "libavutil/x86/asm.h"
27 #include "libavcodec/mpegutils.h"
28 #include "libavcodec/mpegvideo.h"
29 #include "fdct.h"
30 
31 #undef MMREG_WIDTH
32 #undef MM
33 #undef MOVQ
34 #undef SPREADW
35 #undef PMAXW
36 #undef PMAX
37 #undef SAVE_SIGN
38 #undef RESTORE_SIGN
39 
40 #if COMPILE_TEMPLATE_SSE2
41 #define MMREG_WIDTH "16"
42 #define MM "%%xmm"
43 #define MOVQ "movdqa"
44 #define SPREADW(a) \
45  "pshuflw $0, "a", "a" \n\t"\
46  "punpcklwd "a", "a" \n\t"
47 #define PMAXW(a,b) "pmaxsw "a", "b" \n\t"
48 #define PMAX(a,b) \
49  "movhlps "a", "b" \n\t"\
50  PMAXW(b, a)\
51  "pshuflw $0x0E, "a", "b" \n\t"\
52  PMAXW(b, a)\
53  "pshuflw $0x01, "a", "b" \n\t"\
54  PMAXW(b, a)
55 #else
56 #define MMREG_WIDTH "8"
57 #define MM "%%mm"
58 #define MOVQ "movq"
59 #if COMPILE_TEMPLATE_MMXEXT
60 #define SPREADW(a) "pshufw $0, "a", "a" \n\t"
61 #define PMAXW(a,b) "pmaxsw "a", "b" \n\t"
62 #define PMAX(a,b) \
63  "pshufw $0x0E, "a", "b" \n\t"\
64  PMAXW(b, a)\
65  "pshufw $0x01, "a", "b" \n\t"\
66  PMAXW(b, a)
67 #else
68 #define SPREADW(a) \
69  "punpcklwd "a", "a" \n\t"\
70  "punpcklwd "a", "a" \n\t"
71 #define PMAXW(a,b) \
72  "psubusw "a", "b" \n\t"\
73  "paddw "a", "b" \n\t"
74 #define PMAX(a,b) \
75  "movq "a", "b" \n\t"\
76  "psrlq $32, "a" \n\t"\
77  PMAXW(b, a)\
78  "movq "a", "b" \n\t"\
79  "psrlq $16, "a" \n\t"\
80  PMAXW(b, a)
81 
82 #endif
83 #endif
84 
85 #if COMPILE_TEMPLATE_SSSE3
86 #define SAVE_SIGN(a,b) \
87  "movdqa "b", "a" \n\t"\
88  "pabsw "b", "b" \n\t"
89 #define RESTORE_SIGN(a,b) \
90  "psignw "a", "b" \n\t"
91 #else
92 #define SAVE_SIGN(a,b) \
93  "pxor "a", "a" \n\t"\
94  "pcmpgtw "b", "a" \n\t" /* block[i] <= 0 ? 0xFF : 0x00 */\
95  "pxor "a", "b" \n\t"\
96  "psubw "a", "b" \n\t" /* ABS(block[i]) */
97 #define RESTORE_SIGN(a,b) \
98  "pxor "a", "b" \n\t"\
99  "psubw "a", "b" \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
100 #endif
101 
103  int16_t *block, int n,
104  int qscale, int *overflow)
105 {
106  x86_reg last_non_zero_p1;
107  int level=0, q; //=0 is because gcc says uninitialized ...
108  const uint16_t *qmat, *bias;
109  LOCAL_ALIGNED_16(int16_t, temp_block, [64]);
110 
111  assert((7&(int)(&temp_block[0])) == 0); //did gcc align it correctly?
112 
113  //s->fdct (block);
114  RENAME_FDCT(ff_fdct)(block); // cannot be anything else ...
115 
116  if(s->dct_error_sum)
117  s->denoise_dct(s, block);
118 
119  if (s->mb_intra) {
120  int dummy;
121  if (n < 4)
122  q = s->y_dc_scale;
123  else
124  q = s->c_dc_scale;
125  /* note: block[0] is assumed to be positive */
126  if (!s->h263_aic) {
127  __asm__ volatile (
128  "mul %%ecx \n\t"
129  : "=d" (level), "=a"(dummy)
130  : "a" ((block[0]>>2) + q), "c" (ff_inverse[q<<1])
131  );
132  } else
133  /* For AIC we skip quant/dequant of INTRADC */
134  level = (block[0] + 4)>>3;
135 
136  block[0]=0; //avoid fake overflow
137 // temp_block[0] = (block[0] + (q >> 1)) / q;
138  last_non_zero_p1 = 1;
139  bias = s->q_intra_matrix16[qscale][1];
140  qmat = s->q_intra_matrix16[qscale][0];
141  } else {
142  last_non_zero_p1 = 0;
143  bias = s->q_inter_matrix16[qscale][1];
144  qmat = s->q_inter_matrix16[qscale][0];
145  }
146 
147  if((s->out_format == FMT_H263 || s->out_format == FMT_H261) && s->mpeg_quant==0){
148 
149  __asm__ volatile(
150  "movd %%"FF_REG_a", "MM"3 \n\t" // last_non_zero_p1
151  SPREADW(MM"3")
152  "pxor "MM"7, "MM"7 \n\t" // 0
153  "pxor "MM"4, "MM"4 \n\t" // 0
154  MOVQ" (%2), "MM"5 \n\t" // qmat[0]
155  "pxor "MM"6, "MM"6 \n\t"
156  "psubw (%3), "MM"6 \n\t" // -bias[0]
157  "mov $-128, %%"FF_REG_a" \n\t"
158  ".p2align 4 \n\t"
159  "1: \n\t"
160  MOVQ" (%1, %%"FF_REG_a"), "MM"0 \n\t" // block[i]
161  SAVE_SIGN(MM"1", MM"0") // ABS(block[i])
162  "psubusw "MM"6, "MM"0 \n\t" // ABS(block[i]) + bias[0]
163  "pmulhw "MM"5, "MM"0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16
164  "por "MM"0, "MM"4 \n\t"
165  RESTORE_SIGN(MM"1", MM"0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
166  MOVQ" "MM"0, (%5, %%"FF_REG_a") \n\t"
167  "pcmpeqw "MM"7, "MM"0 \n\t" // out==0 ? 0xFF : 0x00
168  MOVQ" (%4, %%"FF_REG_a"), "MM"1 \n\t"
169  MOVQ" "MM"7, (%1, %%"FF_REG_a") \n\t" // 0
170  "pandn "MM"1, "MM"0 \n\t"
171  PMAXW(MM"0", MM"3")
172  "add $"MMREG_WIDTH", %%"FF_REG_a" \n\t"
173  " js 1b \n\t"
174  PMAX(MM"3", MM"0")
175  "movd "MM"3, %%"FF_REG_a" \n\t"
176  "movzb %%al, %%"FF_REG_a" \n\t" // last_non_zero_p1
177  : "+a" (last_non_zero_p1)
178  : "r" (block+64), "r" (qmat), "r" (bias),
179  "r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64)
180  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
181  "%xmm4", "%xmm5", "%xmm6", "%xmm7")
182  );
183  }else{ // FMT_H263
184  __asm__ volatile(
185  "movd %%"FF_REG_a", "MM"3 \n\t" // last_non_zero_p1
186  SPREADW(MM"3")
187  "pxor "MM"7, "MM"7 \n\t" // 0
188  "pxor "MM"4, "MM"4 \n\t" // 0
189  "mov $-128, %%"FF_REG_a" \n\t"
190  ".p2align 4 \n\t"
191  "1: \n\t"
192  MOVQ" (%1, %%"FF_REG_a"), "MM"0 \n\t" // block[i]
193  SAVE_SIGN(MM"1", MM"0") // ABS(block[i])
194  MOVQ" (%3, %%"FF_REG_a"), "MM"6 \n\t" // bias[0]
195  "paddusw "MM"6, "MM"0 \n\t" // ABS(block[i]) + bias[0]
196  MOVQ" (%2, %%"FF_REG_a"), "MM"5 \n\t" // qmat[i]
197  "pmulhw "MM"5, "MM"0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16
198  "por "MM"0, "MM"4 \n\t"
199  RESTORE_SIGN(MM"1", MM"0") // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
200  MOVQ" "MM"0, (%5, %%"FF_REG_a") \n\t"
201  "pcmpeqw "MM"7, "MM"0 \n\t" // out==0 ? 0xFF : 0x00
202  MOVQ" (%4, %%"FF_REG_a"), "MM"1 \n\t"
203  MOVQ" "MM"7, (%1, %%"FF_REG_a") \n\t" // 0
204  "pandn "MM"1, "MM"0 \n\t"
205  PMAXW(MM"0", MM"3")
206  "add $"MMREG_WIDTH", %%"FF_REG_a" \n\t"
207  " js 1b \n\t"
208  PMAX(MM"3", MM"0")
209  "movd "MM"3, %%"FF_REG_a" \n\t"
210  "movzb %%al, %%"FF_REG_a" \n\t" // last_non_zero_p1
211  : "+a" (last_non_zero_p1)
212  : "r" (block+64), "r" (qmat+64), "r" (bias+64),
213  "r" (inv_zigzag_direct16 + 64), "r" (temp_block + 64)
214  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
215  "%xmm4", "%xmm5", "%xmm6", "%xmm7")
216  );
217  }
218  __asm__ volatile(
219  "movd %1, "MM"1 \n\t" // max_qcoeff
220  SPREADW(MM"1")
221  "psubusw "MM"1, "MM"4 \n\t"
222  "packuswb "MM"4, "MM"4 \n\t"
223 #if COMPILE_TEMPLATE_SSE2
224  "packuswb "MM"4, "MM"4 \n\t"
225 #endif
226  "movd "MM"4, %0 \n\t" // *overflow
227  : "=g" (*overflow)
228  : "g" (s->max_qcoeff)
229  );
230 
231  if(s->mb_intra) block[0]= level;
232  else block[0]= temp_block[0];
233 
234  if (s->idsp.perm_type == FF_IDCT_PERM_SIMPLE) {
235  if(last_non_zero_p1 <= 1) goto end;
236  block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08];
237  block[0x20] = temp_block[0x10];
238  if(last_non_zero_p1 <= 4) goto end;
239  block[0x18] = temp_block[0x09]; block[0x04] = temp_block[0x02];
240  block[0x09] = temp_block[0x03];
241  if(last_non_zero_p1 <= 7) goto end;
242  block[0x14] = temp_block[0x0A]; block[0x28] = temp_block[0x11];
243  block[0x12] = temp_block[0x18]; block[0x02] = temp_block[0x20];
244  if(last_non_zero_p1 <= 11) goto end;
245  block[0x1A] = temp_block[0x19]; block[0x24] = temp_block[0x12];
246  block[0x19] = temp_block[0x0B]; block[0x01] = temp_block[0x04];
247  block[0x0C] = temp_block[0x05];
248  if(last_non_zero_p1 <= 16) goto end;
249  block[0x11] = temp_block[0x0C]; block[0x29] = temp_block[0x13];
250  block[0x16] = temp_block[0x1A]; block[0x0A] = temp_block[0x21];
251  block[0x30] = temp_block[0x28]; block[0x22] = temp_block[0x30];
252  block[0x38] = temp_block[0x29]; block[0x06] = temp_block[0x22];
253  if(last_non_zero_p1 <= 24) goto end;
254  block[0x1B] = temp_block[0x1B]; block[0x21] = temp_block[0x14];
255  block[0x1C] = temp_block[0x0D]; block[0x05] = temp_block[0x06];
256  block[0x0D] = temp_block[0x07]; block[0x15] = temp_block[0x0E];
257  block[0x2C] = temp_block[0x15]; block[0x13] = temp_block[0x1C];
258  if(last_non_zero_p1 <= 32) goto end;
259  block[0x0B] = temp_block[0x23]; block[0x34] = temp_block[0x2A];
260  block[0x2A] = temp_block[0x31]; block[0x32] = temp_block[0x38];
261  block[0x3A] = temp_block[0x39]; block[0x26] = temp_block[0x32];
262  block[0x39] = temp_block[0x2B]; block[0x03] = temp_block[0x24];
263  if(last_non_zero_p1 <= 40) goto end;
264  block[0x1E] = temp_block[0x1D]; block[0x25] = temp_block[0x16];
265  block[0x1D] = temp_block[0x0F]; block[0x2D] = temp_block[0x17];
266  block[0x17] = temp_block[0x1E]; block[0x0E] = temp_block[0x25];
267  block[0x31] = temp_block[0x2C]; block[0x2B] = temp_block[0x33];
268  if(last_non_zero_p1 <= 48) goto end;
269  block[0x36] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B];
270  block[0x23] = temp_block[0x34]; block[0x3C] = temp_block[0x2D];
271  block[0x07] = temp_block[0x26]; block[0x1F] = temp_block[0x1F];
272  block[0x0F] = temp_block[0x27]; block[0x35] = temp_block[0x2E];
273  if(last_non_zero_p1 <= 56) goto end;
274  block[0x2E] = temp_block[0x35]; block[0x33] = temp_block[0x3C];
275  block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36];
276  block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37];
277  block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
278  }else{
279  if(last_non_zero_p1 <= 1) goto end;
280  block[0x01] = temp_block[0x01];
281  block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10];
282  if(last_non_zero_p1 <= 4) goto end;
283  block[0x09] = temp_block[0x09]; block[0x02] = temp_block[0x02];
284  block[0x03] = temp_block[0x03];
285  if(last_non_zero_p1 <= 7) goto end;
286  block[0x0A] = temp_block[0x0A]; block[0x11] = temp_block[0x11];
287  block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20];
288  if(last_non_zero_p1 <= 11) goto end;
289  block[0x19] = temp_block[0x19];
290  block[0x12] = temp_block[0x12]; block[0x0B] = temp_block[0x0B];
291  block[0x04] = temp_block[0x04]; block[0x05] = temp_block[0x05];
292  if(last_non_zero_p1 <= 16) goto end;
293  block[0x0C] = temp_block[0x0C]; block[0x13] = temp_block[0x13];
294  block[0x1A] = temp_block[0x1A]; block[0x21] = temp_block[0x21];
295  block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30];
296  block[0x29] = temp_block[0x29]; block[0x22] = temp_block[0x22];
297  if(last_non_zero_p1 <= 24) goto end;
298  block[0x1B] = temp_block[0x1B]; block[0x14] = temp_block[0x14];
299  block[0x0D] = temp_block[0x0D]; block[0x06] = temp_block[0x06];
300  block[0x07] = temp_block[0x07]; block[0x0E] = temp_block[0x0E];
301  block[0x15] = temp_block[0x15]; block[0x1C] = temp_block[0x1C];
302  if(last_non_zero_p1 <= 32) goto end;
303  block[0x23] = temp_block[0x23]; block[0x2A] = temp_block[0x2A];
304  block[0x31] = temp_block[0x31]; block[0x38] = temp_block[0x38];
305  block[0x39] = temp_block[0x39]; block[0x32] = temp_block[0x32];
306  block[0x2B] = temp_block[0x2B]; block[0x24] = temp_block[0x24];
307  if(last_non_zero_p1 <= 40) goto end;
308  block[0x1D] = temp_block[0x1D]; block[0x16] = temp_block[0x16];
309  block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17];
310  block[0x1E] = temp_block[0x1E]; block[0x25] = temp_block[0x25];
311  block[0x2C] = temp_block[0x2C]; block[0x33] = temp_block[0x33];
312  if(last_non_zero_p1 <= 48) goto end;
313  block[0x3A] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B];
314  block[0x34] = temp_block[0x34]; block[0x2D] = temp_block[0x2D];
315  block[0x26] = temp_block[0x26]; block[0x1F] = temp_block[0x1F];
316  block[0x27] = temp_block[0x27]; block[0x2E] = temp_block[0x2E];
317  if(last_non_zero_p1 <= 56) goto end;
318  block[0x35] = temp_block[0x35]; block[0x3C] = temp_block[0x3C];
319  block[0x3D] = temp_block[0x3D]; block[0x36] = temp_block[0x36];
320  block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37];
321  block[0x3E] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
322  }
323  end:
324  return last_non_zero_p1 - 1;
325 }
#define PMAXW(a, b)
#define RENAME(a)
static int RENAME() dct_quantize(MpegEncContext *s, int16_t *block, int n, int qscale, int *overflow)
#define MOVQ
mpegvideo header.
static uint16_t inv_zigzag_direct16[64]
Definition: mpegvideoenc.c:31
static int16_t block[64]
Definition: dct.c:97
#define PMAX(a, b)
const uint32_t ff_inverse[257]
Definition: mathtables.c:25
int x86_reg
Definition: asm.h:71
#define XMM_CLOBBERS_ONLY(...)
Definition: asm.h:98
#define RESTORE_SIGN(a, b)
#define MM
common internal API header
#define SAVE_SIGN(a, b)
uint8_t level
Definition: svq3.c:204
MpegEncContext.
Definition: mpegvideo.h:76
#define SPREADW(a)
#define LOCAL_ALIGNED_16(t, v,...)
Definition: internal.h:111
#define MMREG_WIDTH