Mercurial > libavcodec.hg

comparison i386/mpegvideo_mmx_template.c @ 2979:bfabfdf9ce55 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
COSMETICS: tabs --> spaces, some prettyprinting
author diego
date Thu, 22 Dec 2005 01:10:11 +0000
parents ef2149182f1c
children 0b546eab515d
comparison
equal deleted inserted replaced
2978:403183bbb505 2979:bfabfdf9ce55
19 */ 19 */
20 #undef SPREADW 20 #undef SPREADW
21 #undef PMAXW 21 #undef PMAXW
22 #ifdef HAVE_MMX2 22 #ifdef HAVE_MMX2
23 #define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t" 23 #define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t"
24 #define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t" 24 #define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t"
25 #define PMAX(a,b) \ 25 #define PMAX(a,b) \
26 "pshufw $0x0E," #a ", " #b " \n\t"\ 26 "pshufw $0x0E," #a ", " #b " \n\t"\
27 PMAXW(b, a)\ 27 PMAXW(b, a)\
28 "pshufw $0x01," #a ", " #b " \n\t"\ 28 "pshufw $0x01," #a ", " #b " \n\t"\
29 PMAXW(b, a) 29 PMAXW(b, a)
30 #else 30 #else
31 #define SPREADW(a) \ 31 #define SPREADW(a) \
32 "punpcklwd " #a ", " #a " \n\t"\ 32 "punpcklwd " #a ", " #a " \n\t"\
33 "punpcklwd " #a ", " #a " \n\t" 33 "punpcklwd " #a ", " #a " \n\t"
34 #define PMAXW(a,b) \ 34 #define PMAXW(a,b) \
35 "psubusw " #a ", " #b " \n\t"\ 35 "psubusw " #a ", " #b " \n\t"\
36 "paddw " #a ", " #b " \n\t" 36 "paddw " #a ", " #b " \n\t"
37 #define PMAX(a,b) \ 37 #define PMAX(a,b) \
38 "movq " #a ", " #b " \n\t"\ 38 "movq " #a ", " #b " \n\t"\
39 "psrlq $32, " #a " \n\t"\ 39 "psrlq $32, " #a " \n\t"\
40 PMAXW(b, a)\ 40 PMAXW(b, a)\
41 "movq " #a ", " #b " \n\t"\ 41 "movq " #a ", " #b " \n\t"\
42 "psrlq $16, " #a " \n\t"\ 42 "psrlq $16, " #a " \n\t"\
43 PMAXW(b, a) 43 PMAXW(b, a)
44 44
45 #endif 45 #endif
46 46
47 static int RENAME(dct_quantize)(MpegEncContext *s, 47 static int RENAME(dct_quantize)(MpegEncContext *s,
48 DCTELEM *block, int n, 48 DCTELEM *block, int n,
69 q = s->c_dc_scale; 69 q = s->c_dc_scale;
70 /* note: block[0] is assumed to be positive */ 70 /* note: block[0] is assumed to be positive */
71 if (!s->h263_aic) { 71 if (!s->h263_aic) {
72 #if 1 72 #if 1
73 asm volatile ( 73 asm volatile (
74 "mul %%ecx \n\t" 74 "mul %%ecx \n\t"
75 : "=d" (level), "=a"(dummy) 75 : "=d" (level), "=a"(dummy)
76 : "a" ((block[0]>>2) + q), "c" (inverse[q<<1]) 76 : "a" ((block[0]>>2) + q), "c" (inverse[q<<1])
77 ); 77 );
78 #else 78 #else
79 asm volatile ( 79 asm volatile (
80 "xorl %%edx, %%edx \n\t" 80 "xorl %%edx, %%edx \n\t"
81 "divw %%cx \n\t" 81 "divw %%cx \n\t"
82 "movzwl %%ax, %%eax \n\t" 82 "movzwl %%ax, %%eax \n\t"
83 : "=a" (level) 83 : "=a" (level)
84 : "a" ((block[0]>>2) + q), "c" (q<<1) 84 : "a" ((block[0]>>2) + q), "c" (q<<1)
85 : "%edx" 85 : "%edx"
86 ); 86 );
87 #endif 87 #endif
88 } else 88 } else
89 /* For AIC we skip quant/dequant of INTRADC */ 89 /* For AIC we skip quant/dequant of INTRADC */
90 level = (block[0] + 4)>>3; 90 level = (block[0] + 4)>>3;
101 } 101 }
102 102
103 if((s->out_format == FMT_H263 || s->out_format == FMT_H261) && s->mpeg_quant==0){ 103 if((s->out_format == FMT_H263 || s->out_format == FMT_H261) && s->mpeg_quant==0){
104 104
105 asm volatile( 105 asm volatile(
106 "movd %%"REG_a", %%mm3 \n\t" // last_non_zero_p1 106 "movd %%"REG_a", %%mm3 \n\t" // last_non_zero_p1
107 SPREADW(%%mm3) 107 SPREADW(%%mm3)
108 "pxor %%mm7, %%mm7 \n\t" // 0 108 "pxor %%mm7, %%mm7 \n\t" // 0
109 "pxor %%mm4, %%mm4 \n\t" // 0 109 "pxor %%mm4, %%mm4 \n\t" // 0
110 "movq (%2), %%mm5 \n\t" // qmat[0] 110 "movq (%2), %%mm5 \n\t" // qmat[0]
111 "pxor %%mm6, %%mm6 \n\t" 111 "pxor %%mm6, %%mm6 \n\t"
112 "psubw (%3), %%mm6 \n\t" // -bias[0] 112 "psubw (%3), %%mm6 \n\t" // -bias[0]
113 "mov $-128, %%"REG_a" \n\t" 113 "mov $-128, %%"REG_a" \n\t"
114 ".balign 16 \n\t" 114 ".balign 16 \n\t"
115 "1: \n\t" 115 "1: \n\t"
116 "pxor %%mm1, %%mm1 \n\t" // 0 116 "pxor %%mm1, %%mm1 \n\t" // 0
117 "movq (%1, %%"REG_a"), %%mm0 \n\t" // block[i] 117 "movq (%1, %%"REG_a"), %%mm0 \n\t" // block[i]
118 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 118 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00
119 "pxor %%mm1, %%mm0 \n\t" 119 "pxor %%mm1, %%mm0 \n\t"
120 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) 120 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i])
121 "psubusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] 121 "psubusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0]
122 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16 122 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16
123 "por %%mm0, %%mm4 \n\t" 123 "por %%mm0, %%mm4 \n\t"
124 "pxor %%mm1, %%mm0 \n\t" 124 "pxor %%mm1, %%mm0 \n\t"
125 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) 125 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
126 "movq %%mm0, (%5, %%"REG_a") \n\t" 126 "movq %%mm0, (%5, %%"REG_a") \n\t"
127 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 127 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00
128 "movq (%4, %%"REG_a"), %%mm1 \n\t" 128 "movq (%4, %%"REG_a"), %%mm1 \n\t"
129 "movq %%mm7, (%1, %%"REG_a") \n\t" // 0 129 "movq %%mm7, (%1, %%"REG_a") \n\t" // 0
130 "pandn %%mm1, %%mm0 \n\t" 130 "pandn %%mm1, %%mm0 \n\t"
131 PMAXW(%%mm0, %%mm3) 131 PMAXW(%%mm0, %%mm3)
132 "add $8, %%"REG_a" \n\t" 132 "add $8, %%"REG_a" \n\t"
133 " js 1b \n\t" 133 " js 1b \n\t"
134 PMAX(%%mm3, %%mm0) 134 PMAX(%%mm3, %%mm0)
135 "movd %%mm3, %%"REG_a" \n\t" 135 "movd %%mm3, %%"REG_a" \n\t"
136 "movzb %%al, %%"REG_a" \n\t" // last_non_zero_p1 136 "movzb %%al, %%"REG_a" \n\t" // last_non_zero_p1
137 : "+a" (last_non_zero_p1) 137 : "+a" (last_non_zero_p1)
138 : "r" (block+64), "r" (qmat), "r" (bias), 138 : "r" (block+64), "r" (qmat), "r" (bias),
139 "r" (inv_zigzag_direct16+64), "r" (temp_block+64) 139 "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
140 ); 140 );
141 // note the asm is split cuz gcc doesnt like that many operands ... 141 // note the asm is split cuz gcc doesnt like that many operands ...
142 asm volatile( 142 asm volatile(
143 "movd %1, %%mm1 \n\t" // max_qcoeff 143 "movd %1, %%mm1 \n\t" // max_qcoeff
144 SPREADW(%%mm1) 144 SPREADW(%%mm1)
145 "psubusw %%mm1, %%mm4 \n\t" 145 "psubusw %%mm1, %%mm4 \n\t"
146 "packuswb %%mm4, %%mm4 \n\t" 146 "packuswb %%mm4, %%mm4 \n\t"
147 "movd %%mm4, %0 \n\t" // *overflow 147 "movd %%mm4, %0 \n\t" // *overflow
148 : "=g" (*overflow) 148 : "=g" (*overflow)
149 : "g" (s->max_qcoeff) 149 : "g" (s->max_qcoeff)
150 ); 150 );
151 }else{ // FMT_H263 151 }else{ // FMT_H263
152 asm volatile( 152 asm volatile(
153 "movd %%"REG_a", %%mm3 \n\t" // last_non_zero_p1 153 "movd %%"REG_a", %%mm3 \n\t" // last_non_zero_p1
154 SPREADW(%%mm3) 154 SPREADW(%%mm3)
155 "pxor %%mm7, %%mm7 \n\t" // 0 155 "pxor %%mm7, %%mm7 \n\t" // 0
156 "pxor %%mm4, %%mm4 \n\t" // 0 156 "pxor %%mm4, %%mm4 \n\t" // 0
157 "mov $-128, %%"REG_a" \n\t" 157 "mov $-128, %%"REG_a" \n\t"
158 ".balign 16 \n\t" 158 ".balign 16 \n\t"
159 "1: \n\t" 159 "1: \n\t"
160 "pxor %%mm1, %%mm1 \n\t" // 0 160 "pxor %%mm1, %%mm1 \n\t" // 0
161 "movq (%1, %%"REG_a"), %%mm0 \n\t" // block[i] 161 "movq (%1, %%"REG_a"), %%mm0 \n\t" // block[i]
162 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00 162 "pcmpgtw %%mm0, %%mm1 \n\t" // block[i] <= 0 ? 0xFF : 0x00
163 "pxor %%mm1, %%mm0 \n\t" 163 "pxor %%mm1, %%mm0 \n\t"
164 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i]) 164 "psubw %%mm1, %%mm0 \n\t" // ABS(block[i])
165 "movq (%3, %%"REG_a"), %%mm6 \n\t" // bias[0] 165 "movq (%3, %%"REG_a"), %%mm6 \n\t" // bias[0]
166 "paddusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0] 166 "paddusw %%mm6, %%mm0 \n\t" // ABS(block[i]) + bias[0]
167 "movq (%2, %%"REG_a"), %%mm5 \n\t" // qmat[i] 167 "movq (%2, %%"REG_a"), %%mm5 \n\t" // qmat[i]
168 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16 168 "pmulhw %%mm5, %%mm0 \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16
169 "por %%mm0, %%mm4 \n\t" 169 "por %%mm0, %%mm4 \n\t"
170 "pxor %%mm1, %%mm0 \n\t" 170 "pxor %%mm1, %%mm0 \n\t"
171 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i]) 171 "psubw %%mm1, %%mm0 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
172 "movq %%mm0, (%5, %%"REG_a") \n\t" 172 "movq %%mm0, (%5, %%"REG_a") \n\t"
173 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00 173 "pcmpeqw %%mm7, %%mm0 \n\t" // out==0 ? 0xFF : 0x00
174 "movq (%4, %%"REG_a"), %%mm1 \n\t" 174 "movq (%4, %%"REG_a"), %%mm1 \n\t"
175 "movq %%mm7, (%1, %%"REG_a") \n\t" // 0 175 "movq %%mm7, (%1, %%"REG_a") \n\t" // 0
176 "pandn %%mm1, %%mm0 \n\t" 176 "pandn %%mm1, %%mm0 \n\t"
177 PMAXW(%%mm0, %%mm3) 177 PMAXW(%%mm0, %%mm3)
178 "add $8, %%"REG_a" \n\t" 178 "add $8, %%"REG_a" \n\t"
179 " js 1b \n\t" 179 " js 1b \n\t"
180 PMAX(%%mm3, %%mm0) 180 PMAX(%%mm3, %%mm0)
181 "movd %%mm3, %%"REG_a" \n\t" 181 "movd %%mm3, %%"REG_a" \n\t"
182 "movzb %%al, %%"REG_a" \n\t" // last_non_zero_p1 182 "movzb %%al, %%"REG_a" \n\t" // last_non_zero_p1
183 : "+a" (last_non_zero_p1) 183 : "+a" (last_non_zero_p1)
184 : "r" (block+64), "r" (qmat+64), "r" (bias+64), 184 : "r" (block+64), "r" (qmat+64), "r" (bias+64),
185 "r" (inv_zigzag_direct16+64), "r" (temp_block+64) 185 "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
186 ); 186 );
187 // note the asm is split cuz gcc doesnt like that many operands ... 187 // note the asm is split cuz gcc doesnt like that many operands ...
188 asm volatile( 188 asm volatile(
189 "movd %1, %%mm1 \n\t" // max_qcoeff 189 "movd %1, %%mm1 \n\t" // max_qcoeff
190 SPREADW(%%mm1) 190 SPREADW(%%mm1)
191 "psubusw %%mm1, %%mm4 \n\t" 191 "psubusw %%mm1, %%mm4 \n\t"
192 "packuswb %%mm4, %%mm4 \n\t" 192 "packuswb %%mm4, %%mm4 \n\t"
193 "movd %%mm4, %0 \n\t" // *overflow 193 "movd %%mm4, %0 \n\t" // *overflow
194 : "=g" (*overflow) 194 : "=g" (*overflow)
195 : "g" (s->max_qcoeff) 195 : "g" (s->max_qcoeff)
196 ); 196 );
197 } 197 }
198 198