1 /*===---- mm3dnow.h - 3DNow! intrinsics ------------------------------------===
2  *
3  * Permission is hereby granted, free of charge, to any person obtaining a copy
4  * of this software and associated documentation files (the "Software"), to deal
5  * in the Software without restriction, including without limitation the rights
6  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7  * copies of the Software, and to permit persons to whom the Software is
8  * furnished to do so, subject to the following conditions:
9  *
10  * The above copyright notice and this permission notice shall be included in
11  * all copies or substantial portions of the Software.
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19  * THE SOFTWARE.
20  *
21  *===-----------------------------------------------------------------------===
22  */
23 
24 #ifndef _MM3DNOW_H_INCLUDED
25 #define _MM3DNOW_H_INCLUDED
26 
27 #include <mmintrin.h>
28 #include <prfchwintrin.h>
29 
30 typedef float __v2sf __attribute__((__vector_size__(8)));
31 
32 /* Define the default attributes for the functions in this file. */
33 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnow"), __min_vector_width__(64)))
34 
35 static __inline__ void __attribute__((__always_inline__, __nodebug__, __target__("3dnow")))
_m_femms(void)36 _m_femms(void) {
37   __builtin_ia32_femms();
38 }
39 
40 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pavgusb(__m64 __m1,__m64 __m2)41 _m_pavgusb(__m64 __m1, __m64 __m2) {
42   return (__m64)__builtin_ia32_pavgusb((__v8qi)__m1, (__v8qi)__m2);
43 }
44 
45 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pf2id(__m64 __m)46 _m_pf2id(__m64 __m) {
47   return (__m64)__builtin_ia32_pf2id((__v2sf)__m);
48 }
49 
50 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfacc(__m64 __m1,__m64 __m2)51 _m_pfacc(__m64 __m1, __m64 __m2) {
52   return (__m64)__builtin_ia32_pfacc((__v2sf)__m1, (__v2sf)__m2);
53 }
54 
55 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfadd(__m64 __m1,__m64 __m2)56 _m_pfadd(__m64 __m1, __m64 __m2) {
57   return (__m64)__builtin_ia32_pfadd((__v2sf)__m1, (__v2sf)__m2);
58 }
59 
60 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpeq(__m64 __m1,__m64 __m2)61 _m_pfcmpeq(__m64 __m1, __m64 __m2) {
62   return (__m64)__builtin_ia32_pfcmpeq((__v2sf)__m1, (__v2sf)__m2);
63 }
64 
65 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpge(__m64 __m1,__m64 __m2)66 _m_pfcmpge(__m64 __m1, __m64 __m2) {
67   return (__m64)__builtin_ia32_pfcmpge((__v2sf)__m1, (__v2sf)__m2);
68 }
69 
70 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpgt(__m64 __m1,__m64 __m2)71 _m_pfcmpgt(__m64 __m1, __m64 __m2) {
72   return (__m64)__builtin_ia32_pfcmpgt((__v2sf)__m1, (__v2sf)__m2);
73 }
74 
75 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmax(__m64 __m1,__m64 __m2)76 _m_pfmax(__m64 __m1, __m64 __m2) {
77   return (__m64)__builtin_ia32_pfmax((__v2sf)__m1, (__v2sf)__m2);
78 }
79 
80 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmin(__m64 __m1,__m64 __m2)81 _m_pfmin(__m64 __m1, __m64 __m2) {
82   return (__m64)__builtin_ia32_pfmin((__v2sf)__m1, (__v2sf)__m2);
83 }
84 
85 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmul(__m64 __m1,__m64 __m2)86 _m_pfmul(__m64 __m1, __m64 __m2) {
87   return (__m64)__builtin_ia32_pfmul((__v2sf)__m1, (__v2sf)__m2);
88 }
89 
90 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcp(__m64 __m)91 _m_pfrcp(__m64 __m) {
92   return (__m64)__builtin_ia32_pfrcp((__v2sf)__m);
93 }
94 
95 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcpit1(__m64 __m1,__m64 __m2)96 _m_pfrcpit1(__m64 __m1, __m64 __m2) {
97   return (__m64)__builtin_ia32_pfrcpit1((__v2sf)__m1, (__v2sf)__m2);
98 }
99 
100 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcpit2(__m64 __m1,__m64 __m2)101 _m_pfrcpit2(__m64 __m1, __m64 __m2) {
102   return (__m64)__builtin_ia32_pfrcpit2((__v2sf)__m1, (__v2sf)__m2);
103 }
104 
105 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrsqrt(__m64 __m)106 _m_pfrsqrt(__m64 __m) {
107   return (__m64)__builtin_ia32_pfrsqrt((__v2sf)__m);
108 }
109 
110 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrsqrtit1(__m64 __m1,__m64 __m2)111 _m_pfrsqrtit1(__m64 __m1, __m64 __m2) {
112   return (__m64)__builtin_ia32_pfrsqit1((__v2sf)__m1, (__v2sf)__m2);
113 }
114 
115 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfsub(__m64 __m1,__m64 __m2)116 _m_pfsub(__m64 __m1, __m64 __m2) {
117   return (__m64)__builtin_ia32_pfsub((__v2sf)__m1, (__v2sf)__m2);
118 }
119 
120 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfsubr(__m64 __m1,__m64 __m2)121 _m_pfsubr(__m64 __m1, __m64 __m2) {
122   return (__m64)__builtin_ia32_pfsubr((__v2sf)__m1, (__v2sf)__m2);
123 }
124 
125 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pi2fd(__m64 __m)126 _m_pi2fd(__m64 __m) {
127   return (__m64)__builtin_ia32_pi2fd((__v2si)__m);
128 }
129 
130 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pmulhrw(__m64 __m1,__m64 __m2)131 _m_pmulhrw(__m64 __m1, __m64 __m2) {
132   return (__m64)__builtin_ia32_pmulhrw((__v4hi)__m1, (__v4hi)__m2);
133 }
134 
135 /* Handle the 3dnowa instructions here. */
136 #undef __DEFAULT_FN_ATTRS
137 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnowa"), __min_vector_width__(64)))
138 
139 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pf2iw(__m64 __m)140 _m_pf2iw(__m64 __m) {
141   return (__m64)__builtin_ia32_pf2iw((__v2sf)__m);
142 }
143 
144 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfnacc(__m64 __m1,__m64 __m2)145 _m_pfnacc(__m64 __m1, __m64 __m2) {
146   return (__m64)__builtin_ia32_pfnacc((__v2sf)__m1, (__v2sf)__m2);
147 }
148 
149 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfpnacc(__m64 __m1,__m64 __m2)150 _m_pfpnacc(__m64 __m1, __m64 __m2) {
151   return (__m64)__builtin_ia32_pfpnacc((__v2sf)__m1, (__v2sf)__m2);
152 }
153 
154 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pi2fw(__m64 __m)155 _m_pi2fw(__m64 __m) {
156   return (__m64)__builtin_ia32_pi2fw((__v2si)__m);
157 }
158 
159 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pswapdsf(__m64 __m)160 _m_pswapdsf(__m64 __m) {
161   return (__m64)__builtin_ia32_pswapdsf((__v2sf)__m);
162 }
163 
164 static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pswapdsi(__m64 __m)165 _m_pswapdsi(__m64 __m) {
166   return (__m64)__builtin_ia32_pswapdsi((__v2si)__m);
167 }
168 
169 #undef __DEFAULT_FN_ATTRS
170 
171 #endif
172