00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021 #ifndef FFMPEG_MMX_H
00022 #define FFMPEG_MMX_H
00023
00024
00025
00026
00027
00028
00029
00030 typedef union {
00031 long long q;
00032 unsigned long long uq;
00033 int d[2];
00034 unsigned int ud[2];
00035 short w[4];
00036 unsigned short uw[4];
00037 char b[8];
00038 unsigned char ub[8];
00039 float s[2];
00040 } mmx_t;
00041
00042
00043 #define mmx_i2r(op,imm,reg) \
00044 __asm__ __volatile__ (#op " %0, %%" #reg \
00045 : \
00046 : "i" (imm) )
00047
00048 #define mmx_a2r(op,a_off,a_reg,reg) \
00049 __asm__ __volatile__ (#op " " #a_off "(%0), %%" #reg \
00050 : \
00051 : "r" (a_reg))
00052
00053 #define mmx_m2r(op,mem,reg) \
00054 __asm__ __volatile__ (#op " %0, %%" #reg \
00055 : \
00056 : "m" (mem))
00057
00058 #define mmx_a2r(op,a_off,a_reg,reg) \
00059 __asm__ __volatile__ (#op " " #a_off "(%0), %%" #reg \
00060 : \
00061 : "r" (a_reg))
00062
00063 #define mmx_r2m(op,reg,mem) \
00064 __asm__ __volatile__ (#op " %%" #reg ", %0" \
00065 : "=m" (mem) \
00066 : )
00067
00068 #define mmx_r2a(op,reg,a_off,a_reg) \
00069 __asm__ __volatile__ (#op " %%" #reg ", " #a_off "(%0)" \
00070 : \
00071 : "r"(a_reg))
00072
00073 #define mmx_r2r(op,regs,regd) \
00074 __asm__ __volatile__ (#op " %" #regs ", %" #regd)
00075
00076
00077 #define emms() __asm__ __volatile__ ("emms")
00078
00079 #define movd_m2r(var,reg) mmx_m2r (movd, var, reg)
00080 #define movd_r2m(reg,var) mmx_r2m (movd, reg, var)
00081 #define movd_r2r(regs,regd) mmx_r2r (movd, regs, regd)
00082 #define movd_v2r(var,reg) __asm__ __volatile__ ("movd %0, %%" #reg \
00083 : \
00084 : "rm" (var))
00085 #define movd_r2v(reg,var) __asm__ __volatile__ ("movd %%" #reg ", %0" \
00086 : "=rm" (var) \
00087 : )
00088
00089 #define movq_m2r(var,reg) mmx_m2r (movq, var, reg)
00090 #define movq_a2r(off,var,reg) mmx_a2r (movq, off, var, reg)
00091 #define movq_r2m(reg,var) mmx_r2m (movq, reg, var)
00092 #define movq_r2r(regs,regd) mmx_r2r (movq, regs, regd)
00093 #define movq_r2a(reg,off,var) mmx_r2a (movq, reg, off, var)
00094
00095 #define packssdw_m2r(var,reg) mmx_m2r (packssdw, var, reg)
00096 #define packssdw_r2r(regs,regd) mmx_r2r (packssdw, regs, regd)
00097 #define packsswb_m2r(var,reg) mmx_m2r (packsswb, var, reg)
00098 #define packsswb_r2r(regs,regd) mmx_r2r (packsswb, regs, regd)
00099
00100 #define packuswb_m2r(var,reg) mmx_m2r (packuswb, var, reg)
00101 #define packuswb_r2r(regs,regd) mmx_r2r (packuswb, regs, regd)
00102
00103 #define paddb_m2r(var,reg) mmx_m2r (paddb, var, reg)
00104 #define paddb_r2r(regs,regd) mmx_r2r (paddb, regs, regd)
00105 #define paddd_m2r(var,reg) mmx_m2r (paddd, var, reg)
00106 #define paddd_r2r(regs,regd) mmx_r2r (paddd, regs, regd)
00107 #define paddw_m2r(var,reg) mmx_m2r (paddw, var, reg)
00108 #define paddw_r2r(regs,regd) mmx_r2r (paddw, regs, regd)
00109
00110 #define paddsb_m2r(var,reg) mmx_m2r (paddsb, var, reg)
00111 #define paddsb_r2r(regs,regd) mmx_r2r (paddsb, regs, regd)
00112 #define paddsw_m2r(var,reg) mmx_m2r (paddsw, var, reg)
00113 #define paddsw_r2r(regs,regd) mmx_r2r (paddsw, regs, regd)
00114
00115 #define paddusb_m2r(var,reg) mmx_m2r (paddusb, var, reg)
00116 #define paddusb_r2r(regs,regd) mmx_r2r (paddusb, regs, regd)
00117 #define paddusw_m2r(var,reg) mmx_m2r (paddusw, var, reg)
00118 #define paddusw_r2r(regs,regd) mmx_r2r (paddusw, regs, regd)
00119
00120 #define pand_m2r(var,reg) mmx_m2r (pand, var, reg)
00121 #define pand_r2r(regs,regd) mmx_r2r (pand, regs, regd)
00122
00123 #define pandn_m2r(var,reg) mmx_m2r (pandn, var, reg)
00124 #define pandn_r2r(regs,regd) mmx_r2r (pandn, regs, regd)
00125
00126 #define pcmpeqb_m2r(var,reg) mmx_m2r (pcmpeqb, var, reg)
00127 #define pcmpeqb_r2r(regs,regd) mmx_r2r (pcmpeqb, regs, regd)
00128 #define pcmpeqd_m2r(var,reg) mmx_m2r (pcmpeqd, var, reg)
00129 #define pcmpeqd_r2r(regs,regd) mmx_r2r (pcmpeqd, regs, regd)
00130 #define pcmpeqw_m2r(var,reg) mmx_m2r (pcmpeqw, var, reg)
00131 #define pcmpeqw_r2r(regs,regd) mmx_r2r (pcmpeqw, regs, regd)
00132
00133 #define pcmpgtb_m2r(var,reg) mmx_m2r (pcmpgtb, var, reg)
00134 #define pcmpgtb_r2r(regs,regd) mmx_r2r (pcmpgtb, regs, regd)
00135 #define pcmpgtd_m2r(var,reg) mmx_m2r (pcmpgtd, var, reg)
00136 #define pcmpgtd_r2r(regs,regd) mmx_r2r (pcmpgtd, regs, regd)
00137 #define pcmpgtw_m2r(var,reg) mmx_m2r (pcmpgtw, var, reg)
00138 #define pcmpgtw_r2r(regs,regd) mmx_r2r (pcmpgtw, regs, regd)
00139
00140 #define pmaddwd_m2r(var,reg) mmx_m2r (pmaddwd, var, reg)
00141 #define pmaddwd_a2r(of,var,reg) mmx_a2r (pmaddwd, of, var, reg)
00142 #define pmaddwd_r2r(regs,regd) mmx_r2r (pmaddwd, regs, regd)
00143
00144 #define pmulhw_m2r(var,reg) mmx_m2r (pmulhw, var, reg)
00145 #define pmulhw_r2r(regs,regd) mmx_r2r (pmulhw, regs, regd)
00146
00147 #define pmullw_m2r(var,reg) mmx_m2r (pmullw, var, reg)
00148 #define pmullw_r2r(regs,regd) mmx_r2r (pmullw, regs, regd)
00149
00150 #define por_m2r(var,reg) mmx_m2r (por, var, reg)
00151 #define por_r2r(regs,regd) mmx_r2r (por, regs, regd)
00152
00153 #define pslld_i2r(imm,reg) mmx_i2r (pslld, imm, reg)
00154 #define pslld_m2r(var,reg) mmx_m2r (pslld, var, reg)
00155 #define pslld_r2r(regs,regd) mmx_r2r (pslld, regs, regd)
00156 #define psllq_i2r(imm,reg) mmx_i2r (psllq, imm, reg)
00157 #define psllq_m2r(var,reg) mmx_m2r (psllq, var, reg)
00158 #define psllq_r2r(regs,regd) mmx_r2r (psllq, regs, regd)
00159 #define psllw_i2r(imm,reg) mmx_i2r (psllw, imm, reg)
00160 #define psllw_m2r(var,reg) mmx_m2r (psllw, var, reg)
00161 #define psllw_r2r(regs,regd) mmx_r2r (psllw, regs, regd)
00162
00163 #define psrad_i2r(imm,reg) mmx_i2r (psrad, imm, reg)
00164 #define psrad_m2r(var,reg) mmx_m2r (psrad, var, reg)
00165 #define psrad_r2r(regs,regd) mmx_r2r (psrad, regs, regd)
00166 #define psraw_i2r(imm,reg) mmx_i2r (psraw, imm, reg)
00167 #define psraw_m2r(var,reg) mmx_m2r (psraw, var, reg)
00168 #define psraw_r2r(regs,regd) mmx_r2r (psraw, regs, regd)
00169
00170 #define psrld_i2r(imm,reg) mmx_i2r (psrld, imm, reg)
00171 #define psrld_m2r(var,reg) mmx_m2r (psrld, var, reg)
00172 #define psrld_r2r(regs,regd) mmx_r2r (psrld, regs, regd)
00173 #define psrlq_i2r(imm,reg) mmx_i2r (psrlq, imm, reg)
00174 #define psrlq_m2r(var,reg) mmx_m2r (psrlq, var, reg)
00175 #define psrlq_r2r(regs,regd) mmx_r2r (psrlq, regs, regd)
00176 #define psrlw_i2r(imm,reg) mmx_i2r (psrlw, imm, reg)
00177 #define psrlw_m2r(var,reg) mmx_m2r (psrlw, var, reg)
00178 #define psrlw_r2r(regs,regd) mmx_r2r (psrlw, regs, regd)
00179
00180 #define psubb_m2r(var,reg) mmx_m2r (psubb, var, reg)
00181 #define psubb_r2r(regs,regd) mmx_r2r (psubb, regs, regd)
00182 #define psubd_m2r(var,reg) mmx_m2r (psubd, var, reg)
00183 #define psubd_r2r(regs,regd) mmx_r2r (psubd, regs, regd)
00184 #define psubw_m2r(var,reg) mmx_m2r (psubw, var, reg)
00185 #define psubw_r2r(regs,regd) mmx_r2r (psubw, regs, regd)
00186
00187 #define psubsb_m2r(var,reg) mmx_m2r (psubsb, var, reg)
00188 #define psubsb_r2r(regs,regd) mmx_r2r (psubsb, regs, regd)
00189 #define psubsw_m2r(var,reg) mmx_m2r (psubsw, var, reg)
00190 #define psubsw_r2r(regs,regd) mmx_r2r (psubsw, regs, regd)
00191
00192 #define psubusb_m2r(var,reg) mmx_m2r (psubusb, var, reg)
00193 #define psubusb_r2r(regs,regd) mmx_r2r (psubusb, regs, regd)
00194 #define psubusw_m2r(var,reg) mmx_m2r (psubusw, var, reg)
00195 #define psubusw_r2r(regs,regd) mmx_r2r (psubusw, regs, regd)
00196
00197 #define punpckhbw_m2r(var,reg) mmx_m2r (punpckhbw, var, reg)
00198 #define punpckhbw_r2r(regs,regd) mmx_r2r (punpckhbw, regs, regd)
00199 #define punpckhdq_m2r(var,reg) mmx_m2r (punpckhdq, var, reg)
00200 #define punpckhdq_r2r(regs,regd) mmx_r2r (punpckhdq, regs, regd)
00201 #define punpckhwd_m2r(var,reg) mmx_m2r (punpckhwd, var, reg)
00202 #define punpckhwd_r2r(regs,regd) mmx_r2r (punpckhwd, regs, regd)
00203
00204 #define punpcklbw_m2r(var,reg) mmx_m2r (punpcklbw, var, reg)
00205 #define punpcklbw_r2r(regs,regd) mmx_r2r (punpcklbw, regs, regd)
00206 #define punpckldq_m2r(var,reg) mmx_m2r (punpckldq, var, reg)
00207 #define punpckldq_r2r(regs,regd) mmx_r2r (punpckldq, regs, regd)
00208 #define punpcklwd_m2r(var,reg) mmx_m2r (punpcklwd, var, reg)
00209 #define punpcklwd_r2r(regs,regd) mmx_r2r (punpcklwd, regs, regd)
00210
00211 #define pxor_m2r(var,reg) mmx_m2r (pxor, var, reg)
00212 #define pxor_r2r(regs,regd) mmx_r2r (pxor, regs, regd)
00213
00214
00215
00216
00217 #define pavgusb_m2r(var,reg) mmx_m2r (pavgusb, var, reg)
00218 #define pavgusb_r2r(regs,regd) mmx_r2r (pavgusb, regs, regd)
00219
00220
00221
00222
00223
00224 #define mmx_m2ri(op,mem,reg,imm) \
00225 __asm__ __volatile__ (#op " %1, %0, %%" #reg \
00226 : \
00227 : "m" (mem), "i" (imm))
00228 #define mmx_r2ri(op,regs,regd,imm) \
00229 __asm__ __volatile__ (#op " %0, %%" #regs ", %%" #regd \
00230 : \
00231 : "i" (imm) )
00232
00233 #define mmx_fetch(mem,hint) \
00234 __asm__ __volatile__ ("prefetch" #hint " %0" \
00235 : \
00236 : "m" (mem))
00237
00238
00239 #define maskmovq(regs,maskreg) mmx_r2ri (maskmovq, regs, maskreg)
00240
00241 #define movntq_r2m(mmreg,var) mmx_r2m (movntq, mmreg, var)
00242
00243 #define pavgb_m2r(var,reg) mmx_m2r (pavgb, var, reg)
00244 #define pavgb_r2r(regs,regd) mmx_r2r (pavgb, regs, regd)
00245 #define pavgw_m2r(var,reg) mmx_m2r (pavgw, var, reg)
00246 #define pavgw_r2r(regs,regd) mmx_r2r (pavgw, regs, regd)
00247
00248 #define pextrw_r2r(mmreg,reg,imm) mmx_r2ri (pextrw, mmreg, reg, imm)
00249
00250 #define pinsrw_r2r(reg,mmreg,imm) mmx_r2ri (pinsrw, reg, mmreg, imm)
00251
00252 #define pmaxsw_m2r(var,reg) mmx_m2r (pmaxsw, var, reg)
00253 #define pmaxsw_r2r(regs,regd) mmx_r2r (pmaxsw, regs, regd)
00254
00255 #define pmaxub_m2r(var,reg) mmx_m2r (pmaxub, var, reg)
00256 #define pmaxub_r2r(regs,regd) mmx_r2r (pmaxub, regs, regd)
00257
00258 #define pminsw_m2r(var,reg) mmx_m2r (pminsw, var, reg)
00259 #define pminsw_r2r(regs,regd) mmx_r2r (pminsw, regs, regd)
00260
00261 #define pminub_m2r(var,reg) mmx_m2r (pminub, var, reg)
00262 #define pminub_r2r(regs,regd) mmx_r2r (pminub, regs, regd)
00263
00264 #define pmovmskb(mmreg,reg) \
00265 __asm__ __volatile__ ("movmskps %" #mmreg ", %" #reg)
00266
00267 #define pmulhuw_m2r(var,reg) mmx_m2r (pmulhuw, var, reg)
00268 #define pmulhuw_r2r(regs,regd) mmx_r2r (pmulhuw, regs, regd)
00269
00270 #define prefetcht0(mem) mmx_fetch (mem, t0)
00271 #define prefetcht1(mem) mmx_fetch (mem, t1)
00272 #define prefetcht2(mem) mmx_fetch (mem, t2)
00273 #define prefetchnta(mem) mmx_fetch (mem, nta)
00274
00275 #define psadbw_m2r(var,reg) mmx_m2r (psadbw, var, reg)
00276 #define psadbw_r2r(regs,regd) mmx_r2r (psadbw, regs, regd)
00277
00278 #define pshufw_m2r(var,reg,imm) mmx_m2ri(pshufw, var, reg, imm)
00279 #define pshufw_r2r(regs,regd,imm) mmx_r2ri(pshufw, regs, regd, imm)
00280
00281 #define sfence() __asm__ __volatile__ ("sfence\n\t")
00282
00283
00284 #define pshufhw_m2r(var,reg,imm) mmx_m2ri(pshufhw, var, reg, imm)
00285 #define pshufhw_r2r(regs,regd,imm) mmx_r2ri(pshufhw, regs, regd, imm)
00286 #define pshuflw_m2r(var,reg,imm) mmx_m2ri(pshuflw, var, reg, imm)
00287 #define pshuflw_r2r(regs,regd,imm) mmx_r2ri(pshuflw, regs, regd, imm)
00288
00289 #define pshufd_r2r(regs,regd,imm) mmx_r2ri(pshufd, regs, regd, imm)
00290
00291 #define movdqa_m2r(var,reg) mmx_m2r (movdqa, var, reg)
00292 #define movdqa_r2m(reg,var) mmx_r2m (movdqa, reg, var)
00293 #define movdqa_r2r(regs,regd) mmx_r2r (movdqa, regs, regd)
00294 #define movdqu_m2r(var,reg) mmx_m2r (movdqu, var, reg)
00295 #define movdqu_r2m(reg,var) mmx_r2m (movdqu, reg, var)
00296 #define movdqu_r2r(regs,regd) mmx_r2r (movdqu, regs, regd)
00297
00298 #define pmullw_r2m(reg,var) mmx_r2m (pmullw, reg, var)
00299
00300 #define pslldq_i2r(imm,reg) mmx_i2r (pslldq, imm, reg)
00301 #define psrldq_i2r(imm,reg) mmx_i2r (psrldq, imm, reg)
00302
00303 #define punpcklqdq_r2r(regs,regd) mmx_r2r (punpcklqdq, regs, regd)
00304 #define punpckhqdq_r2r(regs,regd) mmx_r2r (punpckhqdq, regs, regd)
00305
00306
00307 #endif