1*67e74705SXin Li /*===---- mm3dnow.h - 3DNow! intrinsics ------------------------------------===
2*67e74705SXin Li *
3*67e74705SXin Li * Permission is hereby granted, free of charge, to any person obtaining a copy
4*67e74705SXin Li * of this software and associated documentation files (the "Software"), to deal
5*67e74705SXin Li * in the Software without restriction, including without limitation the rights
6*67e74705SXin Li * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7*67e74705SXin Li * copies of the Software, and to permit persons to whom the Software is
8*67e74705SXin Li * furnished to do so, subject to the following conditions:
9*67e74705SXin Li *
10*67e74705SXin Li * The above copyright notice and this permission notice shall be included in
11*67e74705SXin Li * all copies or substantial portions of the Software.
12*67e74705SXin Li *
13*67e74705SXin Li * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14*67e74705SXin Li * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15*67e74705SXin Li * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16*67e74705SXin Li * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17*67e74705SXin Li * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18*67e74705SXin Li * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19*67e74705SXin Li * THE SOFTWARE.
20*67e74705SXin Li *
21*67e74705SXin Li *===-----------------------------------------------------------------------===
22*67e74705SXin Li */
23*67e74705SXin Li
24*67e74705SXin Li #ifndef _MM3DNOW_H_INCLUDED
25*67e74705SXin Li #define _MM3DNOW_H_INCLUDED
26*67e74705SXin Li
27*67e74705SXin Li #include <mmintrin.h>
28*67e74705SXin Li #include <prfchwintrin.h>
29*67e74705SXin Li
30*67e74705SXin Li typedef float __v2sf __attribute__((__vector_size__(8)));
31*67e74705SXin Li
32*67e74705SXin Li /* Define the default attributes for the functions in this file. */
33*67e74705SXin Li #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnow")))
34*67e74705SXin Li
35*67e74705SXin Li static __inline__ void __DEFAULT_FN_ATTRS
_m_femms(void)36*67e74705SXin Li _m_femms(void) {
37*67e74705SXin Li __builtin_ia32_femms();
38*67e74705SXin Li }
39*67e74705SXin Li
40*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pavgusb(__m64 __m1,__m64 __m2)41*67e74705SXin Li _m_pavgusb(__m64 __m1, __m64 __m2) {
42*67e74705SXin Li return (__m64)__builtin_ia32_pavgusb((__v8qi)__m1, (__v8qi)__m2);
43*67e74705SXin Li }
44*67e74705SXin Li
45*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pf2id(__m64 __m)46*67e74705SXin Li _m_pf2id(__m64 __m) {
47*67e74705SXin Li return (__m64)__builtin_ia32_pf2id((__v2sf)__m);
48*67e74705SXin Li }
49*67e74705SXin Li
50*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfacc(__m64 __m1,__m64 __m2)51*67e74705SXin Li _m_pfacc(__m64 __m1, __m64 __m2) {
52*67e74705SXin Li return (__m64)__builtin_ia32_pfacc((__v2sf)__m1, (__v2sf)__m2);
53*67e74705SXin Li }
54*67e74705SXin Li
55*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfadd(__m64 __m1,__m64 __m2)56*67e74705SXin Li _m_pfadd(__m64 __m1, __m64 __m2) {
57*67e74705SXin Li return (__m64)__builtin_ia32_pfadd((__v2sf)__m1, (__v2sf)__m2);
58*67e74705SXin Li }
59*67e74705SXin Li
60*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpeq(__m64 __m1,__m64 __m2)61*67e74705SXin Li _m_pfcmpeq(__m64 __m1, __m64 __m2) {
62*67e74705SXin Li return (__m64)__builtin_ia32_pfcmpeq((__v2sf)__m1, (__v2sf)__m2);
63*67e74705SXin Li }
64*67e74705SXin Li
65*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpge(__m64 __m1,__m64 __m2)66*67e74705SXin Li _m_pfcmpge(__m64 __m1, __m64 __m2) {
67*67e74705SXin Li return (__m64)__builtin_ia32_pfcmpge((__v2sf)__m1, (__v2sf)__m2);
68*67e74705SXin Li }
69*67e74705SXin Li
70*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfcmpgt(__m64 __m1,__m64 __m2)71*67e74705SXin Li _m_pfcmpgt(__m64 __m1, __m64 __m2) {
72*67e74705SXin Li return (__m64)__builtin_ia32_pfcmpgt((__v2sf)__m1, (__v2sf)__m2);
73*67e74705SXin Li }
74*67e74705SXin Li
75*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmax(__m64 __m1,__m64 __m2)76*67e74705SXin Li _m_pfmax(__m64 __m1, __m64 __m2) {
77*67e74705SXin Li return (__m64)__builtin_ia32_pfmax((__v2sf)__m1, (__v2sf)__m2);
78*67e74705SXin Li }
79*67e74705SXin Li
80*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmin(__m64 __m1,__m64 __m2)81*67e74705SXin Li _m_pfmin(__m64 __m1, __m64 __m2) {
82*67e74705SXin Li return (__m64)__builtin_ia32_pfmin((__v2sf)__m1, (__v2sf)__m2);
83*67e74705SXin Li }
84*67e74705SXin Li
85*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfmul(__m64 __m1,__m64 __m2)86*67e74705SXin Li _m_pfmul(__m64 __m1, __m64 __m2) {
87*67e74705SXin Li return (__m64)__builtin_ia32_pfmul((__v2sf)__m1, (__v2sf)__m2);
88*67e74705SXin Li }
89*67e74705SXin Li
90*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcp(__m64 __m)91*67e74705SXin Li _m_pfrcp(__m64 __m) {
92*67e74705SXin Li return (__m64)__builtin_ia32_pfrcp((__v2sf)__m);
93*67e74705SXin Li }
94*67e74705SXin Li
95*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcpit1(__m64 __m1,__m64 __m2)96*67e74705SXin Li _m_pfrcpit1(__m64 __m1, __m64 __m2) {
97*67e74705SXin Li return (__m64)__builtin_ia32_pfrcpit1((__v2sf)__m1, (__v2sf)__m2);
98*67e74705SXin Li }
99*67e74705SXin Li
100*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrcpit2(__m64 __m1,__m64 __m2)101*67e74705SXin Li _m_pfrcpit2(__m64 __m1, __m64 __m2) {
102*67e74705SXin Li return (__m64)__builtin_ia32_pfrcpit2((__v2sf)__m1, (__v2sf)__m2);
103*67e74705SXin Li }
104*67e74705SXin Li
105*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrsqrt(__m64 __m)106*67e74705SXin Li _m_pfrsqrt(__m64 __m) {
107*67e74705SXin Li return (__m64)__builtin_ia32_pfrsqrt((__v2sf)__m);
108*67e74705SXin Li }
109*67e74705SXin Li
110*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfrsqrtit1(__m64 __m1,__m64 __m2)111*67e74705SXin Li _m_pfrsqrtit1(__m64 __m1, __m64 __m2) {
112*67e74705SXin Li return (__m64)__builtin_ia32_pfrsqit1((__v2sf)__m1, (__v2sf)__m2);
113*67e74705SXin Li }
114*67e74705SXin Li
115*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfsub(__m64 __m1,__m64 __m2)116*67e74705SXin Li _m_pfsub(__m64 __m1, __m64 __m2) {
117*67e74705SXin Li return (__m64)__builtin_ia32_pfsub((__v2sf)__m1, (__v2sf)__m2);
118*67e74705SXin Li }
119*67e74705SXin Li
120*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfsubr(__m64 __m1,__m64 __m2)121*67e74705SXin Li _m_pfsubr(__m64 __m1, __m64 __m2) {
122*67e74705SXin Li return (__m64)__builtin_ia32_pfsubr((__v2sf)__m1, (__v2sf)__m2);
123*67e74705SXin Li }
124*67e74705SXin Li
125*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pi2fd(__m64 __m)126*67e74705SXin Li _m_pi2fd(__m64 __m) {
127*67e74705SXin Li return (__m64)__builtin_ia32_pi2fd((__v2si)__m);
128*67e74705SXin Li }
129*67e74705SXin Li
130*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pmulhrw(__m64 __m1,__m64 __m2)131*67e74705SXin Li _m_pmulhrw(__m64 __m1, __m64 __m2) {
132*67e74705SXin Li return (__m64)__builtin_ia32_pmulhrw((__v4hi)__m1, (__v4hi)__m2);
133*67e74705SXin Li }
134*67e74705SXin Li
135*67e74705SXin Li /* Handle the 3dnowa instructions here. */
136*67e74705SXin Li #undef __DEFAULT_FN_ATTRS
137*67e74705SXin Li #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("3dnowa")))
138*67e74705SXin Li
139*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pf2iw(__m64 __m)140*67e74705SXin Li _m_pf2iw(__m64 __m) {
141*67e74705SXin Li return (__m64)__builtin_ia32_pf2iw((__v2sf)__m);
142*67e74705SXin Li }
143*67e74705SXin Li
144*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfnacc(__m64 __m1,__m64 __m2)145*67e74705SXin Li _m_pfnacc(__m64 __m1, __m64 __m2) {
146*67e74705SXin Li return (__m64)__builtin_ia32_pfnacc((__v2sf)__m1, (__v2sf)__m2);
147*67e74705SXin Li }
148*67e74705SXin Li
149*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pfpnacc(__m64 __m1,__m64 __m2)150*67e74705SXin Li _m_pfpnacc(__m64 __m1, __m64 __m2) {
151*67e74705SXin Li return (__m64)__builtin_ia32_pfpnacc((__v2sf)__m1, (__v2sf)__m2);
152*67e74705SXin Li }
153*67e74705SXin Li
154*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pi2fw(__m64 __m)155*67e74705SXin Li _m_pi2fw(__m64 __m) {
156*67e74705SXin Li return (__m64)__builtin_ia32_pi2fw((__v2si)__m);
157*67e74705SXin Li }
158*67e74705SXin Li
159*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pswapdsf(__m64 __m)160*67e74705SXin Li _m_pswapdsf(__m64 __m) {
161*67e74705SXin Li return (__m64)__builtin_ia32_pswapdsf((__v2sf)__m);
162*67e74705SXin Li }
163*67e74705SXin Li
164*67e74705SXin Li static __inline__ __m64 __DEFAULT_FN_ATTRS
_m_pswapdsi(__m64 __m)165*67e74705SXin Li _m_pswapdsi(__m64 __m) {
166*67e74705SXin Li return (__m64)__builtin_ia32_pswapdsi((__v2si)__m);
167*67e74705SXin Li }
168*67e74705SXin Li
169*67e74705SXin Li #undef __DEFAULT_FN_ATTRS
170*67e74705SXin Li
171*67e74705SXin Li #endif
172