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