1 /*
2  * mmx.h
3  * Copyright (C) 1997-2001 H. Dietz and R. Fisher
4  */
5 #ifndef I386MMX_H
6 #define I386MMX_H
7 
8 /*
9  * The type of an value that fits in an MMX register (note that long
10  * long constant values MUST be suffixed by LL and unsigned long long
11  * values by ULL, lest they be truncated by the compiler)
12  */
13 
14 typedef    union {
15     long long             q;    /* Quadword (64-bit) value */
16     unsigned long long   uq;    /* Unsigned Quadword */
17     int                d[2];    /* 2 Doubleword (32-bit) values */
18     unsigned int      ud[2];    /* 2 Unsigned Doubleword */
19     short              w[4];    /* 4 Word (16-bit) values */
20     unsigned short    uw[4];    /* 4 Unsigned Word */
21     char               b[8];    /* 8 Byte (8-bit) values */
22     unsigned char     ub[8];    /* 8 Unsigned Byte */
23     float              s[2];    /* Single-precision (32-bit) value */
24 } mmx_t;    /* On an 8-byte (64-bit) boundary */
25 
26 
27 #define    mmx_i2r(op,imm,reg) \
28     __asm__ __volatile__ (#op " %0, %%" #reg \
29                   : /* nothing */ \
30                   : "i" (imm))
31 
32 #define    mmx_m2r(op,mem,reg) \
33     __asm__ __volatile__ (#op " %0, %%" #reg \
34                   : /* nothing */ \
35                   : "m" (mem))
36 
37 #define    mmx_r2m(op,reg,mem) \
38     __asm__ __volatile__ (#op " %%" #reg ", %0" \
39                   : "=m" (mem) \
40                   : /* nothing */ )
41 
42 #define    mmx_r2r(op,regs,regd) \
43     __asm__ __volatile__ (#op " %" #regs ", %" #regd)
44 
45 
46 #define    emms() __asm__ __volatile__ ("emms")
47 
48 #define    movd_m2r(var,reg)            mmx_m2r (movd, var, reg)
49 #define    movd_r2m(reg,var)            mmx_r2m (movd, reg, var)
50 #define    movd_r2r(regs,regd)          mmx_r2r (movd, regs, regd)
51 
52 #define    movq_m2r(var,reg)            mmx_m2r (movq, var, reg)
53 #define    movq_r2m(reg,var)            mmx_r2m (movq, reg, var)
54 #define    movq_r2r(regs,regd)          mmx_r2r (movq, regs, regd)
55 
56 #define    packssdw_m2r(var,reg)        mmx_m2r (packssdw, var, reg)
57 #define    packssdw_r2r(regs,regd)      mmx_r2r (packssdw, regs, regd)
58 #define    packsswb_m2r(var,reg)        mmx_m2r (packsswb, var, reg)
59 #define    packsswb_r2r(regs,regd)      mmx_r2r (packsswb, regs, regd)
60 
61 #define    packuswb_m2r(var,reg)        mmx_m2r (packuswb, var, reg)
62 #define    packuswb_r2r(regs,regd)      mmx_r2r (packuswb, regs, regd)
63 
64 #define    paddb_m2r(var,reg)           mmx_m2r (paddb, var, reg)
65 #define    paddb_r2r(regs,regd)         mmx_r2r (paddb, regs, regd)
66 #define    paddd_m2r(var,reg)           mmx_m2r (paddd, var, reg)
67 #define    paddd_r2r(regs,regd)         mmx_r2r (paddd, regs, regd)
68 #define    paddw_m2r(var,reg)           mmx_m2r (paddw, var, reg)
69 #define    paddw_r2r(regs,regd)         mmx_r2r (paddw, regs, regd)
70 
71 #define    paddsb_m2r(var,reg)          mmx_m2r (paddsb, var, reg)
72 #define    paddsb_r2r(regs,regd)        mmx_r2r (paddsb, regs, regd)
73 #define    paddsw_m2r(var,reg)          mmx_m2r (paddsw, var, reg)
74 #define    paddsw_r2r(regs,regd)        mmx_r2r (paddsw, regs, regd)
75 
76 #define    paddusb_m2r(var,reg)         mmx_m2r (paddusb, var, reg)
77 #define    paddusb_r2r(regs,regd)       mmx_r2r (paddusb, regs, regd)
78 #define    paddusw_m2r(var,reg)         mmx_m2r (paddusw, var, reg)
79 #define    paddusw_r2r(regs,regd)       mmx_r2r (paddusw, regs, regd)
80 
81 #define    pand_m2r(var,reg)            mmx_m2r (pand, var, reg)
82 #define    pand_r2r(regs,regd)          mmx_r2r (pand, regs, regd)
83 
84 #define    pandn_m2r(var,reg)           mmx_m2r (pandn, var, reg)
85 #define    pandn_r2r(regs,regd)         mmx_r2r (pandn, regs, regd)
86 
87 #define    pcmpeqb_m2r(var,reg)         mmx_m2r (pcmpeqb, var, reg)
88 #define    pcmpeqb_r2r(regs,regd)       mmx_r2r (pcmpeqb, regs, regd)
89 #define    pcmpeqd_m2r(var,reg)         mmx_m2r (pcmpeqd, var, reg)
90 #define    pcmpeqd_r2r(regs,regd)       mmx_r2r (pcmpeqd, regs, regd)
91 #define    pcmpeqw_m2r(var,reg)         mmx_m2r (pcmpeqw, var, reg)
92 #define    pcmpeqw_r2r(regs,regd)       mmx_r2r (pcmpeqw, regs, regd)
93 
94 #define    pcmpgtb_m2r(var,reg)         mmx_m2r (pcmpgtb, var, reg)
95 #define    pcmpgtb_r2r(regs,regd)       mmx_r2r (pcmpgtb, regs, regd)
96 #define    pcmpgtd_m2r(var,reg)         mmx_m2r (pcmpgtd, var, reg)
97 #define    pcmpgtd_r2r(regs,regd)       mmx_r2r (pcmpgtd, regs, regd)
98 #define    pcmpgtw_m2r(var,reg)         mmx_m2r (pcmpgtw, var, reg)
99 #define    pcmpgtw_r2r(regs,regd)       mmx_r2r (pcmpgtw, regs, regd)
100 
101 #define    pmaddwd_m2r(var,reg)         mmx_m2r (pmaddwd, var, reg)
102 #define    pmaddwd_r2r(regs,regd)       mmx_r2r (pmaddwd, regs, regd)
103 
104 #define    pmulhw_m2r(var,reg)          mmx_m2r (pmulhw, var, reg)
105 #define    pmulhw_r2r(regs,regd)        mmx_r2r (pmulhw, regs, regd)
106 
107 #define    pmullw_m2r(var,reg)          mmx_m2r (pmullw, var, reg)
108 #define    pmullw_r2r(regs,regd)        mmx_r2r (pmullw, regs, regd)
109 
110 #define    por_m2r(var,reg)             mmx_m2r (por, var, reg)
111 #define    por_r2r(regs,regd)           mmx_r2r (por, regs, regd)
112 
113 #define    pslld_i2r(imm,reg)           mmx_i2r (pslld, imm, reg)
114 #define    pslld_m2r(var,reg)           mmx_m2r (pslld, var, reg)
115 #define    pslld_r2r(regs,regd)         mmx_r2r (pslld, regs, regd)
116 #define    psllq_i2r(imm,reg)           mmx_i2r (psllq, imm, reg)
117 #define    psllq_m2r(var,reg)           mmx_m2r (psllq, var, reg)
118 #define    psllq_r2r(regs,regd)         mmx_r2r (psllq, regs, regd)
119 #define    psllw_i2r(imm,reg)           mmx_i2r (psllw, imm, reg)
120 #define    psllw_m2r(var,reg)           mmx_m2r (psllw, var, reg)
121 #define    psllw_r2r(regs,regd)         mmx_r2r (psllw, regs, regd)
122 
123 #define    psrad_i2r(imm,reg)           mmx_i2r (psrad, imm, reg)
124 #define    psrad_m2r(var,reg)           mmx_m2r (psrad, var, reg)
125 #define    psrad_r2r(regs,regd)         mmx_r2r (psrad, regs, regd)
126 #define    psraw_i2r(imm,reg)           mmx_i2r (psraw, imm, reg)
127 #define    psraw_m2r(var,reg)           mmx_m2r (psraw, var, reg)
128 #define    psraw_r2r(regs,regd)         mmx_r2r (psraw, regs, regd)
129 
130 #define    psrld_i2r(imm,reg)           mmx_i2r (psrld, imm, reg)
131 #define    psrld_m2r(var,reg)           mmx_m2r (psrld, var, reg)
132 #define    psrld_r2r(regs,regd)         mmx_r2r (psrld, regs, regd)
133 #define    psrlq_i2r(imm,reg)           mmx_i2r (psrlq, imm, reg)
134 #define    psrlq_m2r(var,reg)           mmx_m2r (psrlq, var, reg)
135 #define    psrlq_r2r(regs,regd)         mmx_r2r (psrlq, regs, regd)
136 #define    psrlw_i2r(imm,reg)           mmx_i2r (psrlw, imm, reg)
137 #define    psrlw_m2r(var,reg)           mmx_m2r (psrlw, var, reg)
138 #define    psrlw_r2r(regs,regd)         mmx_r2r (psrlw, regs, regd)
139 
140 #define    psubb_m2r(var,reg)           mmx_m2r (psubb, var, reg)
141 #define    psubb_r2r(regs,regd)         mmx_r2r (psubb, regs, regd)
142 #define    psubd_m2r(var,reg)           mmx_m2r (psubd, var, reg)
143 #define    psubd_r2r(regs,regd)         mmx_r2r (psubd, regs, regd)
144 #define    psubw_m2r(var,reg)           mmx_m2r (psubw, var, reg)
145 #define    psubw_r2r(regs,regd)         mmx_r2r (psubw, regs, regd)
146 
147 #define    psubsb_m2r(var,reg)          mmx_m2r (psubsb, var, reg)
148 #define    psubsb_r2r(regs,regd)        mmx_r2r (psubsb, regs, regd)
149 #define    psubsw_m2r(var,reg)          mmx_m2r (psubsw, var, reg)
150 #define    psubsw_r2r(regs,regd)        mmx_r2r (psubsw, regs, regd)
151 
152 #define    psubusb_m2r(var,reg)         mmx_m2r (psubusb, var, reg)
153 #define    psubusb_r2r(regs,regd)       mmx_r2r (psubusb, regs, regd)
154 #define    psubusw_m2r(var,reg)         mmx_m2r (psubusw, var, reg)
155 #define    psubusw_r2r(regs,regd)       mmx_r2r (psubusw, regs, regd)
156 
157 #define    punpckhbw_m2r(var,reg)       mmx_m2r (punpckhbw, var, reg)
158 #define    punpckhbw_r2r(regs,regd)     mmx_r2r (punpckhbw, regs, regd)
159 #define    punpckhdq_m2r(var,reg)       mmx_m2r (punpckhdq, var, reg)
160 #define    punpckhdq_r2r(regs,regd)     mmx_r2r (punpckhdq, regs, regd)
161 #define    punpckhwd_m2r(var,reg)       mmx_m2r (punpckhwd, var, reg)
162 #define    punpckhwd_r2r(regs,regd)     mmx_r2r (punpckhwd, regs, regd)
163 
164 #define    punpcklbw_m2r(var,reg)       mmx_m2r (punpcklbw, var, reg)
165 #define    punpcklbw_r2r(regs,regd)     mmx_r2r (punpcklbw, regs, regd)
166 #define    punpckldq_m2r(var,reg)       mmx_m2r (punpckldq, var, reg)
167 #define    punpckldq_r2r(regs,regd)     mmx_r2r (punpckldq, regs, regd)
168 #define    punpcklwd_m2r(var,reg)       mmx_m2r (punpcklwd, var, reg)
169 #define    punpcklwd_r2r(regs,regd)     mmx_r2r (punpcklwd, regs, regd)
170 
171 #define    pxor_m2r(var,reg)            mmx_m2r (pxor, var, reg)
172 #define    pxor_r2r(regs,regd)          mmx_r2r (pxor, regs, regd)
173 
174 
175 /* 3DNOW extensions */
176 
177 #define pavgusb_m2r(var,reg)            mmx_m2r (pavgusb, var, reg)
178 #define pavgusb_r2r(regs,regd)          mmx_r2r (pavgusb, regs, regd)
179 
180 
181 /* AMD MMX extensions - also available in intel SSE */
182 
183 
184 #define mmx_m2ri(op,mem,reg,imm) \
185         __asm__ __volatile__ (#op " %1, %0, %%" #reg \
186                               : /* nothing */ \
187                               : "X" (mem), "X" (imm))
188 #define mmx_r2ri(op,regs,regd,imm) \
189         __asm__ __volatile__ (#op " %0, %%" #regs ", %%" #regd \
190                               : /* nothing */ \
191                               : "X" (imm))
192 
193 #define    mmx_fetch(mem,hint) \
194     __asm__ __volatile__ ("prefetch" #hint " %0" \
195                   : /* nothing */ \
196                   : "X" (mem))
197 
198 
199 #define    maskmovq(regs,maskreg)       mmx_r2ri (maskmovq, regs, maskreg)
200 
201 #define    movntq_r2m(mmreg,var)        mmx_r2m (movntq, mmreg, var)
202 
203 #define    pavgb_m2r(var,reg)           mmx_m2r (pavgb, var, reg)
204 #define    pavgb_r2r(regs,regd)         mmx_r2r (pavgb, regs, regd)
205 #define    pavgw_m2r(var,reg)           mmx_m2r (pavgw, var, reg)
206 #define    pavgw_r2r(regs,regd)         mmx_r2r (pavgw, regs, regd)
207 
208 #define    pextrw_r2r(mmreg,reg,imm)    mmx_r2ri (pextrw, mmreg, reg, imm)
209 
210 #define    pinsrw_r2r(reg,mmreg,imm)    mmx_r2ri (pinsrw, reg, mmreg, imm)
211 
212 #define    pmaxsw_m2r(var,reg)          mmx_m2r (pmaxsw, var, reg)
213 #define    pmaxsw_r2r(regs,regd)        mmx_r2r (pmaxsw, regs, regd)
214 
215 #define    pmaxub_m2r(var,reg)          mmx_m2r (pmaxub, var, reg)
216 #define    pmaxub_r2r(regs,regd)        mmx_r2r (pmaxub, regs, regd)
217 
218 #define    pminsw_m2r(var,reg)          mmx_m2r (pminsw, var, reg)
219 #define    pminsw_r2r(regs,regd)        mmx_r2r (pminsw, regs, regd)
220 
221 #define    pminub_m2r(var,reg)          mmx_m2r (pminub, var, reg)
222 #define    pminub_r2r(regs,regd)        mmx_r2r (pminub, regs, regd)
223 
224 #define    pmovmskb(mmreg,reg) \
225     __asm__ __volatile__ ("movmskps %" #mmreg ", %" #reg)
226 
227 #define    pmulhuw_m2r(var,reg)         mmx_m2r (pmulhuw, var, reg)
228 #define    pmulhuw_r2r(regs,regd)       mmx_r2r (pmulhuw, regs, regd)
229 
230 #define    prefetcht0(mem)              mmx_fetch (mem, t0)
231 #define    prefetcht1(mem)              mmx_fetch (mem, t1)
232 #define    prefetcht2(mem)              mmx_fetch (mem, t2)
233 #define    prefetchnta(mem)             mmx_fetch (mem, nta)
234 
235 #define    psadbw_m2r(var,reg)          mmx_m2r (psadbw, var, reg)
236 #define    psadbw_r2r(regs,regd)        mmx_r2r (psadbw, regs, regd)
237 
238 #define    pshufw_m2r(var,reg,imm)      mmx_m2ri(pshufw, var, reg, imm)
239 #define    pshufw_r2r(regs,regd,imm)    mmx_r2ri(pshufw, regs, regd, imm)
240 
241 #define    sfence() __asm__ __volatile__ ("sfence\n\t")
242 
243 /* SSE2 */
244 #define    pshufhw_m2r(var,reg,imm)     mmx_m2ri(pshufhw, var, reg, imm)
245 #define    pshufhw_r2r(regs,regd,imm)   mmx_r2ri(pshufhw, regs, regd, imm)
246 #define    pshuflw_m2r(var,reg,imm)     mmx_m2ri(pshuflw, var, reg, imm)
247 #define    pshuflw_r2r(regs,regd,imm)   mmx_r2ri(pshuflw, regs, regd, imm)
248 
249 #define    pshufd_r2r(regs,regd,imm)    mmx_r2ri(pshufd, regs, regd, imm)
250 
251 #define    movdqa_m2r(var,reg)          mmx_m2r (movdqa, var, reg)
252 #define    movdqa_r2m(reg,var)          mmx_r2m (movdqa, reg, var)
253 #define    movdqa_r2r(regs,regd)        mmx_r2r (movdqa, regs, regd)
254 #define    movdqu_m2r(var,reg)          mmx_m2r (movdqu, var, reg)
255 #define    movdqu_r2m(reg,var)          mmx_r2m (movdqu, reg, var)
256 #define    movdqu_r2r(regs,regd)        mmx_r2r (movdqu, regs, regd)
257 
258 #define    pmullw_r2m(reg,var)          mmx_r2m (pmullw, reg, var)
259 
260 #define    pslldq_i2r(imm,reg)          mmx_i2r (pslldq, imm, reg)
261 #define    psrldq_i2r(imm,reg)          mmx_i2r (psrldq, imm, reg)
262 
263 #define    punpcklqdq_r2r(regs,regd)    mmx_r2r (punpcklqdq, regs, regd)
264 #define    punpckhqdq_r2r(regs,regd)    mmx_r2r (punpckhqdq, regs, regd)
265 
266 
267 #endif /* I386MMX_H */
268