root / target-i386 / ops_sse_header.h @ b8b6a50b
History | View | Annotate | Download (8.3 kB)
1 |
/*
|
---|---|
2 |
* MMX/3DNow!/SSE/SSE2/SSE3/PNI support
|
3 |
*
|
4 |
* Copyright (c) 2005 Fabrice Bellard
|
5 |
*
|
6 |
* This library is free software; you can redistribute it and/or
|
7 |
* modify it under the terms of the GNU Lesser General Public
|
8 |
* License as published by the Free Software Foundation; either
|
9 |
* version 2 of the License, or (at your option) any later version.
|
10 |
*
|
11 |
* This library is distributed in the hope that it will be useful,
|
12 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
14 |
* Lesser General Public License for more details.
|
15 |
*
|
16 |
* You should have received a copy of the GNU Lesser General Public
|
17 |
* License along with this library; if not, write to the Free Software
|
18 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
19 |
*/
|
20 |
#if SHIFT == 0 |
21 |
#define Reg MMXReg
|
22 |
#define SUFFIX _mmx
|
23 |
#else
|
24 |
#define Reg XMMReg
|
25 |
#define SUFFIX _xmm
|
26 |
#endif
|
27 |
|
28 |
void glue(helper_psrlw, SUFFIX)(Reg *d, Reg *s);
|
29 |
void glue(helper_psraw, SUFFIX)(Reg *d, Reg *s);
|
30 |
void glue(helper_psllw, SUFFIX)(Reg *d, Reg *s);
|
31 |
void glue(helper_psrld, SUFFIX)(Reg *d, Reg *s);
|
32 |
void glue(helper_psrad, SUFFIX)(Reg *d, Reg *s);
|
33 |
void glue(helper_pslld, SUFFIX)(Reg *d, Reg *s);
|
34 |
void glue(helper_psrlq, SUFFIX)(Reg *d, Reg *s);
|
35 |
void glue(helper_psllq, SUFFIX)(Reg *d, Reg *s);
|
36 |
|
37 |
#if SHIFT == 1 |
38 |
void glue(helper_psrldq, SUFFIX)(Reg *d, Reg *s);
|
39 |
void glue(helper_pslldq, SUFFIX)(Reg *d, Reg *s);
|
40 |
#endif
|
41 |
|
42 |
#define SSE_HELPER_B(name, F)\
|
43 |
void glue(name, SUFFIX) (Reg *d, Reg *s);
|
44 |
|
45 |
#define SSE_HELPER_W(name, F)\
|
46 |
void glue(name, SUFFIX) (Reg *d, Reg *s);
|
47 |
|
48 |
#define SSE_HELPER_L(name, F)\
|
49 |
void glue(name, SUFFIX) (Reg *d, Reg *s);
|
50 |
|
51 |
#define SSE_HELPER_Q(name, F)\
|
52 |
void glue(name, SUFFIX) (Reg *d, Reg *s);
|
53 |
|
54 |
SSE_HELPER_B(helper_paddb, FADD); |
55 |
SSE_HELPER_W(helper_paddw, FADD); |
56 |
SSE_HELPER_L(helper_paddl, FADD); |
57 |
SSE_HELPER_Q(helper_paddq, FADD); |
58 |
|
59 |
SSE_HELPER_B(helper_psubb, FSUB); |
60 |
SSE_HELPER_W(helper_psubw, FSUB); |
61 |
SSE_HELPER_L(helper_psubl, FSUB); |
62 |
SSE_HELPER_Q(helper_psubq, FSUB); |
63 |
|
64 |
SSE_HELPER_B(helper_paddusb, FADDUB); |
65 |
SSE_HELPER_B(helper_paddsb, FADDSB); |
66 |
SSE_HELPER_B(helper_psubusb, FSUBUB); |
67 |
SSE_HELPER_B(helper_psubsb, FSUBSB); |
68 |
|
69 |
SSE_HELPER_W(helper_paddusw, FADDUW); |
70 |
SSE_HELPER_W(helper_paddsw, FADDSW); |
71 |
SSE_HELPER_W(helper_psubusw, FSUBUW); |
72 |
SSE_HELPER_W(helper_psubsw, FSUBSW); |
73 |
|
74 |
SSE_HELPER_B(helper_pminub, FMINUB); |
75 |
SSE_HELPER_B(helper_pmaxub, FMAXUB); |
76 |
|
77 |
SSE_HELPER_W(helper_pminsw, FMINSW); |
78 |
SSE_HELPER_W(helper_pmaxsw, FMAXSW); |
79 |
|
80 |
SSE_HELPER_Q(helper_pand, FAND); |
81 |
SSE_HELPER_Q(helper_pandn, FANDN); |
82 |
SSE_HELPER_Q(helper_por, FOR); |
83 |
SSE_HELPER_Q(helper_pxor, FXOR); |
84 |
|
85 |
SSE_HELPER_B(helper_pcmpgtb, FCMPGTB); |
86 |
SSE_HELPER_W(helper_pcmpgtw, FCMPGTW); |
87 |
SSE_HELPER_L(helper_pcmpgtl, FCMPGTL); |
88 |
|
89 |
SSE_HELPER_B(helper_pcmpeqb, FCMPEQ); |
90 |
SSE_HELPER_W(helper_pcmpeqw, FCMPEQ); |
91 |
SSE_HELPER_L(helper_pcmpeql, FCMPEQ); |
92 |
|
93 |
SSE_HELPER_W(helper_pmullw, FMULLW); |
94 |
#if SHIFT == 0 |
95 |
SSE_HELPER_W(helper_pmulhrw, FMULHRW); |
96 |
#endif
|
97 |
SSE_HELPER_W(helper_pmulhuw, FMULHUW); |
98 |
SSE_HELPER_W(helper_pmulhw, FMULHW); |
99 |
|
100 |
SSE_HELPER_B(helper_pavgb, FAVG); |
101 |
SSE_HELPER_W(helper_pavgw, FAVG); |
102 |
|
103 |
void glue(helper_pmuludq, SUFFIX) (Reg *d, Reg *s);
|
104 |
void glue(helper_pmaddwd, SUFFIX) (Reg *d, Reg *s);
|
105 |
|
106 |
void glue(helper_psadbw, SUFFIX) (Reg *d, Reg *s);
|
107 |
void glue(helper_maskmov, SUFFIX) (Reg *d, Reg *s, target_ulong a0);
|
108 |
void glue(helper_movl_mm_T0, SUFFIX) (Reg *d, uint32_t val);
|
109 |
#ifdef TARGET_X86_64
|
110 |
void glue(helper_movq_mm_T0, SUFFIX) (Reg *d, uint64_t val);
|
111 |
#endif
|
112 |
|
113 |
#if SHIFT == 0 |
114 |
void glue(helper_pshufw, SUFFIX) (Reg *d, Reg *s, int order); |
115 |
#else
|
116 |
void helper_shufps(Reg *d, Reg *s, int order); |
117 |
void helper_shufpd(Reg *d, Reg *s, int order); |
118 |
void glue(helper_pshufd, SUFFIX) (Reg *d, Reg *s, int order); |
119 |
void glue(helper_pshuflw, SUFFIX) (Reg *d, Reg *s, int order); |
120 |
void glue(helper_pshufhw, SUFFIX) (Reg *d, Reg *s, int order); |
121 |
#endif
|
122 |
|
123 |
#if SHIFT == 1 |
124 |
/* FPU ops */
|
125 |
/* XXX: not accurate */
|
126 |
|
127 |
#define SSE_HELPER_S(name, F)\
|
128 |
void helper_ ## name ## ps (Reg *d, Reg *s); \ |
129 |
void helper_ ## name ## ss (Reg *d, Reg *s); \ |
130 |
void helper_ ## name ## pd (Reg *d, Reg *s); \ |
131 |
void helper_ ## name ## sd (Reg *d, Reg *s); |
132 |
|
133 |
SSE_HELPER_S(add, FPU_ADD); |
134 |
SSE_HELPER_S(sub, FPU_SUB); |
135 |
SSE_HELPER_S(mul, FPU_MUL); |
136 |
SSE_HELPER_S(div, FPU_DIV); |
137 |
SSE_HELPER_S(min, FPU_MIN); |
138 |
SSE_HELPER_S(max, FPU_MAX); |
139 |
SSE_HELPER_S(sqrt, FPU_SQRT); |
140 |
|
141 |
|
142 |
void helper_cvtps2pd(Reg *d, Reg *s);
|
143 |
void helper_cvtpd2ps(Reg *d, Reg *s);
|
144 |
void helper_cvtss2sd(Reg *d, Reg *s);
|
145 |
void helper_cvtsd2ss(Reg *d, Reg *s);
|
146 |
void helper_cvtdq2ps(Reg *d, Reg *s);
|
147 |
void helper_cvtdq2pd(Reg *d, Reg *s);
|
148 |
void helper_cvtpi2ps(XMMReg *d, MMXReg *s);
|
149 |
void helper_cvtpi2pd(XMMReg *d, MMXReg *s);
|
150 |
void helper_cvtsi2ss(XMMReg *d, uint32_t val);
|
151 |
void helper_cvtsi2sd(XMMReg *d, uint32_t val);
|
152 |
|
153 |
#ifdef TARGET_X86_64
|
154 |
void helper_cvtsq2ss(XMMReg *d, uint64_t val);
|
155 |
void helper_cvtsq2sd(XMMReg *d, uint64_t val);
|
156 |
#endif
|
157 |
|
158 |
void helper_cvtps2dq(XMMReg *d, XMMReg *s);
|
159 |
void helper_cvtpd2dq(XMMReg *d, XMMReg *s);
|
160 |
void helper_cvtps2pi(MMXReg *d, XMMReg *s);
|
161 |
void helper_cvtpd2pi(MMXReg *d, XMMReg *s);
|
162 |
int32_t helper_cvtss2si(XMMReg *s); |
163 |
int32_t helper_cvtsd2si(XMMReg *s); |
164 |
#ifdef TARGET_X86_64
|
165 |
int64_t helper_cvtss2sq(XMMReg *s); |
166 |
int64_t helper_cvtsd2sq(XMMReg *s); |
167 |
#endif
|
168 |
|
169 |
void helper_cvttps2dq(XMMReg *d, XMMReg *s);
|
170 |
void helper_cvttpd2dq(XMMReg *d, XMMReg *s);
|
171 |
void helper_cvttps2pi(MMXReg *d, XMMReg *s);
|
172 |
void helper_cvttpd2pi(MMXReg *d, XMMReg *s);
|
173 |
int32_t helper_cvttss2si(XMMReg *s); |
174 |
int32_t helper_cvttsd2si(XMMReg *s); |
175 |
#ifdef TARGET_X86_64
|
176 |
int64_t helper_cvttss2sq(XMMReg *s); |
177 |
int64_t helper_cvttsd2sq(XMMReg *s); |
178 |
#endif
|
179 |
|
180 |
void helper_rsqrtps(XMMReg *d, XMMReg *s);
|
181 |
void helper_rsqrtss(XMMReg *d, XMMReg *s);
|
182 |
void helper_rcpps(XMMReg *d, XMMReg *s);
|
183 |
void helper_rcpss(XMMReg *d, XMMReg *s);
|
184 |
void helper_haddps(XMMReg *d, XMMReg *s);
|
185 |
void helper_haddpd(XMMReg *d, XMMReg *s);
|
186 |
void helper_hsubps(XMMReg *d, XMMReg *s);
|
187 |
void helper_hsubpd(XMMReg *d, XMMReg *s);
|
188 |
void helper_addsubps(XMMReg *d, XMMReg *s);
|
189 |
void helper_addsubpd(XMMReg *d, XMMReg *s);
|
190 |
|
191 |
#define SSE_HELPER_CMP(name, F)\
|
192 |
void helper_ ## name ## ps (Reg *d, Reg *s); \ |
193 |
void helper_ ## name ## ss (Reg *d, Reg *s); \ |
194 |
void helper_ ## name ## pd (Reg *d, Reg *s); \ |
195 |
void helper_ ## name ## sd (Reg *d, Reg *s); |
196 |
|
197 |
SSE_HELPER_CMP(cmpeq, FPU_CMPEQ); |
198 |
SSE_HELPER_CMP(cmplt, FPU_CMPLT); |
199 |
SSE_HELPER_CMP(cmple, FPU_CMPLE); |
200 |
SSE_HELPER_CMP(cmpunord, FPU_CMPUNORD); |
201 |
SSE_HELPER_CMP(cmpneq, FPU_CMPNEQ); |
202 |
SSE_HELPER_CMP(cmpnlt, FPU_CMPNLT); |
203 |
SSE_HELPER_CMP(cmpnle, FPU_CMPNLE); |
204 |
SSE_HELPER_CMP(cmpord, FPU_CMPORD); |
205 |
|
206 |
void helper_ucomiss(Reg *d, Reg *s);
|
207 |
void helper_comiss(Reg *d, Reg *s);
|
208 |
void helper_ucomisd(Reg *d, Reg *s);
|
209 |
void helper_comisd(Reg *d, Reg *s);
|
210 |
uint32_t helper_movmskps(Reg *s); |
211 |
uint32_t helper_movmskpd(Reg *s); |
212 |
#endif
|
213 |
|
214 |
uint32_t glue(helper_pmovmskb, SUFFIX)(Reg *s); |
215 |
void glue(helper_packsswb, SUFFIX) (Reg *d, Reg *s);
|
216 |
void glue(helper_packuswb, SUFFIX) (Reg *d, Reg *s);
|
217 |
void glue(helper_packssdw, SUFFIX) (Reg *d, Reg *s);
|
218 |
#define UNPCK_OP(base_name, base) \
|
219 |
void glue(helper_punpck ## base_name ## bw, SUFFIX) (Reg *d, Reg *s); \ |
220 |
void glue(helper_punpck ## base_name ## wd, SUFFIX) (Reg *d, Reg *s); \ |
221 |
void glue(helper_punpck ## base_name ## dq, SUFFIX) (Reg *d, Reg *s); |
222 |
|
223 |
UNPCK_OP(l, 0);
|
224 |
UNPCK_OP(h, 1);
|
225 |
|
226 |
#if SHIFT == 1 |
227 |
void glue(helper_punpcklqdq, SUFFIX) (Reg *d, Reg *s);
|
228 |
void glue(helper_punpckhqdq, SUFFIX) (Reg *d, Reg *s);
|
229 |
#endif
|
230 |
|
231 |
/* 3DNow! float ops */
|
232 |
#if SHIFT == 0 |
233 |
void helper_pi2fd(MMXReg *d, MMXReg *s);
|
234 |
void helper_pi2fw(MMXReg *d, MMXReg *s);
|
235 |
void helper_pf2id(MMXReg *d, MMXReg *s);
|
236 |
void helper_pf2iw(MMXReg *d, MMXReg *s);
|
237 |
void helper_pfacc(MMXReg *d, MMXReg *s);
|
238 |
void helper_pfadd(MMXReg *d, MMXReg *s);
|
239 |
void helper_pfcmpeq(MMXReg *d, MMXReg *s);
|
240 |
void helper_pfcmpge(MMXReg *d, MMXReg *s);
|
241 |
void helper_pfcmpgt(MMXReg *d, MMXReg *s);
|
242 |
void helper_pfmax(MMXReg *d, MMXReg *s);
|
243 |
void helper_pfmin(MMXReg *d, MMXReg *s);
|
244 |
void helper_pfmul(MMXReg *d, MMXReg *s);
|
245 |
void helper_pfnacc(MMXReg *d, MMXReg *s);
|
246 |
void helper_pfpnacc(MMXReg *d, MMXReg *s);
|
247 |
void helper_pfrcp(MMXReg *d, MMXReg *s);
|
248 |
void helper_pfrsqrt(MMXReg *d, MMXReg *s);
|
249 |
void helper_pfsub(MMXReg *d, MMXReg *s);
|
250 |
void helper_pfsubr(MMXReg *d, MMXReg *s);
|
251 |
void helper_pswapd(MMXReg *d, MMXReg *s);
|
252 |
#endif
|
253 |
|
254 |
#undef SHIFT
|
255 |
#undef Reg
|
256 |
#undef SUFFIX
|
257 |
|
258 |
#undef SSE_HELPER_B
|
259 |
#undef SSE_HELPER_W
|
260 |
#undef SSE_HELPER_L
|
261 |
#undef SSE_HELPER_Q
|
262 |
#undef SSE_HELPER_S
|
263 |
#undef SSE_HELPER_CMP
|
264 |
#undef UNPCK_OP
|