consistent include usage
[libav.git] / libavcodec / i386 / mpegvideo_mmx_template.c
CommitLineData
2f349de2
MN
1/*
2 Copyright (C) 2002 Michael Niedermayer <michaelni@gmx.at>
3
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17*/
18
19#undef SPREADW
20#undef PMAXW
21#ifdef HAVE_MMX2
22#define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t"
23#define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t"
24
25#else
26#define SPREADW(a) \
27 "punpcklwd " #a ", " #a " \n\t"\
28 "punpcklwd " #a ", " #a " \n\t"
29#define PMAXW(a,b) \
30 "psubusw " #a ", " #b " \n\t"\
31 "paddw " #a ", " #b " \n\t"
32#endif
33
34static int RENAME(dct_quantize)(MpegEncContext *s,
35 DCTELEM *block, int n,
d7e9533a 36 int qscale, int *overflow)
2f349de2 37{
d7e9533a
MN
38 int level=0, last_non_zero_p1, q; //=0 is cuz gcc says uninitalized ...
39 const UINT16 *qmat, *bias;
2f349de2 40 static __align8 INT16 temp_block[64];
2f349de2
MN
41
42 av_fdct (block);
d7e9533a 43
2f349de2
MN
44 if (s->mb_intra) {
45 int dummy;
46 if (n < 4)
47 q = s->y_dc_scale;
48 else
49 q = s->c_dc_scale;
2f349de2 50 /* note: block[0] is assumed to be positive */
ef5b1b5a 51 if (!s->h263_aic) {
2f349de2 52#if 1
d7e9533a
MN
53 asm volatile (
54 "xorl %%edx, %%edx \n\t"
55 "mul %%ecx \n\t"
56 : "=d" (level), "=a"(dummy)
57 : "a" (block[0] + (q >> 1)), "c" (inverse[q])
58 );
2f349de2 59#else
d7e9533a
MN
60 asm volatile (
61 "xorl %%edx, %%edx \n\t"
62 "divw %%cx \n\t"
63 "movzwl %%ax, %%eax \n\t"
64 : "=a" (level)
65 : "a" (block[0] + (q >> 1)), "c" (q)
66 : "%edx"
67 );
2f349de2 68#endif
ef5b1b5a
J
69 } else
70 /* For AIC we skip quant/dequant of INTRADC */
71 level = block[0];
72
d7e9533a 73 block[0]=0; //avoid fake overflow
2f349de2 74// temp_block[0] = (block[0] + (q >> 1)) / q;
2f349de2 75 last_non_zero_p1 = 1;
d7e9533a
MN
76 bias = s->q_intra_matrix16_bias[qscale];
77 qmat = s->q_intra_matrix16[qscale];
2f349de2 78 } else {
2f349de2 79 last_non_zero_p1 = 0;
d7e9533a
MN
80 bias = s->q_inter_matrix16_bias[qscale];
81 qmat = s->q_inter_matrix16[qscale];
2f349de2
MN
82 }
83
d7e9533a
MN
84 if(s->out_format == FMT_H263){
85
86 asm volatile(
87 "movd %%eax, %%mm3 \n\t" // last_non_zero_p1
88 SPREADW(%%mm3)
89 "pxor %%mm7, %%mm7 \n\t" // 0
90 "pxor %%mm4, %%mm4 \n\t" // 0
91 "movq (%2), %%mm5 \n\t" // qmat[0]
92 "pxor %%mm6, %%mm6 \n\t"
93 "psubw (%3), %%mm6 \n\t" // -bias[0]
94 "movl $-128, %%eax \n\t"
95 ".balign 16 \n\t"
96 "1: \n\t"
97 "pxor %%mm1, %%mm1 \n\t" // 0
98 "movq (%1, %%eax), %%mm0 \n\t" // block[i]
99 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00
100 "pxor %%mm1, %%mm0 \n\t"
101 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i])
102 "psubusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0]
103 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16
104 "por %%mm0, %%mm4 \n\t"
105 "pxor %%mm1, %%mm0 \n\t"
106 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
107 "movq %%mm0, (%5, %%eax) \n\t"
108 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00
109 "movq (%4, %%eax), %%mm1 \n\t"
110 "movq %%mm7, (%1, %%eax) \n\t" // 0
111 "pandn %%mm1, %%mm0 \n\t"
112 PMAXW(%%mm0, %%mm3)
113 "addl $8, %%eax \n\t"
114 " js 1b \n\t"
115 "movq %%mm3, %%mm0 \n\t"
116 "psrlq $32, %%mm3 \n\t"
117 PMAXW(%%mm0, %%mm3)
118 "movq %%mm3, %%mm0 \n\t"
119 "psrlq $16, %%mm3 \n\t"
120 PMAXW(%%mm0, %%mm3)
121 "movd %%mm3, %%eax \n\t"
122 "movzbl %%al, %%eax \n\t" // last_non_zero_p1
123 : "+a" (last_non_zero_p1)
124 : "r" (block+64), "r" (qmat), "r" (bias),
125 "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
126 );
127 // note the asm is split cuz gcc doesnt like that many operands ...
128 asm volatile(
129 "movd %1, %%mm1 \n\t" // max_qcoeff
130 SPREADW(%%mm1)
131 "psubusw %%mm1, %%mm4 \n\t"
132 "packuswb %%mm4, %%mm4 \n\t"
133 "movd %%mm4, %0 \n\t" // *overflow
134 : "=g" (*overflow)
135 : "g" (s->max_qcoeff)
136 );
137 }else{ // FMT_H263
138 asm volatile(
139 "movd %%eax, %%mm3 \n\t" // last_non_zero_p1
140 SPREADW(%%mm3)
141 "pxor %%mm7, %%mm7 \n\t" // 0
142 "pxor %%mm4, %%mm4 \n\t" // 0
143 "movl $-128, %%eax \n\t"
144 ".balign 16 \n\t"
145 "1: \n\t"
146 "pxor %%mm1, %%mm1 \n\t" // 0
147 "movq (%1, %%eax), %%mm0 \n\t" // block[i]
148 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00
149 "pxor %%mm1, %%mm0 \n\t"
150 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i])
151 "movq (%3, %%eax), %%mm6 \n\t" // bias[0]
152 "paddusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0]
153 "movq (%2, %%eax), %%mm5 \n\t" // qmat[i]
154 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16
155 "por %%mm0, %%mm4 \n\t"
156 "pxor %%mm1, %%mm0 \n\t"
157 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
158 "movq %%mm0, (%5, %%eax) \n\t"
159 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00
160 "movq (%4, %%eax), %%mm1 \n\t"
161 "movq %%mm7, (%1, %%eax) \n\t" // 0
162 "pandn %%mm1, %%mm0 \n\t"
163 PMAXW(%%mm0, %%mm3)
164 "addl $8, %%eax \n\t"
165 " js 1b \n\t"
166 "movq %%mm3, %%mm0 \n\t"
167 "psrlq $32, %%mm3 \n\t"
168 PMAXW(%%mm0, %%mm3)
169 "movq %%mm3, %%mm0 \n\t"
170 "psrlq $16, %%mm3 \n\t"
171 PMAXW(%%mm0, %%mm3)
172 "movd %%mm3, %%eax \n\t"
173 "movzbl %%al, %%eax \n\t" // last_non_zero_p1
174 : "+a" (last_non_zero_p1)
175 : "r" (block+64), "r" (qmat+64), "r" (bias+64),
176 "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
177 );
178 // note the asm is split cuz gcc doesnt like that many operands ...
179 asm volatile(
180 "movd %1, %%mm1 \n\t" // max_qcoeff
181 SPREADW(%%mm1)
182 "psubusw %%mm1, %%mm4 \n\t"
183 "packuswb %%mm4, %%mm4 \n\t"
184 "movd %%mm4, %0 \n\t" // *overflow
185 : "=g" (*overflow)
186 : "g" (s->max_qcoeff)
187 );
188 }
189
190 if(s->mb_intra) temp_block[0]= level; //FIXME move afer permute
2f349de2
MN
191// last_non_zero_p1=64;
192 /* permute for IDCT */
193 asm volatile(
d7e9533a 194 "movl %0, %%eax \n\t"
2f349de2
MN
195 "pushl %%ebp \n\t"
196 "movl %%esp, " MANGLE(esp_temp) "\n\t"
197 "1: \n\t"
198 "movzbl (%1, %%eax), %%ebx \n\t"
199 "movzbl 1(%1, %%eax), %%ebp \n\t"
200 "movw (%2, %%ebx, 2), %%cx \n\t"
201 "movw (%2, %%ebp, 2), %%sp \n\t"
202 "movzbl " MANGLE(permutation) "(%%ebx), %%ebx\n\t"
203 "movzbl " MANGLE(permutation) "(%%ebp), %%ebp\n\t"
204 "movw %%cx, (%3, %%ebx, 2) \n\t"
205 "movw %%sp, (%3, %%ebp, 2) \n\t"
206 "addl $2, %%eax \n\t"
207 " js 1b \n\t"
208 "movl " MANGLE(esp_temp) ", %%esp\n\t"
209 "popl %%ebp \n\t"
210 :
211 : "g" (-last_non_zero_p1), "d" (zigzag_direct_noperm+last_non_zero_p1), "S" (temp_block), "D" (block)
212 : "%eax", "%ebx", "%ecx"
213 );
214/*
215 for(i=0; i<last_non_zero_p1; i++)
216 {
217 int j= zigzag_direct_noperm[i];
218 block[block_permute_op(j)]= temp_block[j];
219 }
220*/
221//block_permute(block);
d7e9533a 222
2f349de2
MN
223 return last_non_zero_p1 - 1;
224}