1/* { dg-do compile } */
2/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -m3dnow -mavx -msse4a -maes -mpclmul" } */
3
4#include <mm_malloc.h>
5
6/* Test that the intrinsics compile without optimization.  All of them are
7   defined as inline functions in {,x,e,p,t,s,w,g,a,b}mmintrin.h and
8   mm3dnow.h that reference the proper builtin functions.  Defining away
9   "extern" and "__inline" results in all of them being compiled as proper
10   functions.  */
11
12#define extern
13#define __inline
14
15#include <wmmintrin.h>
16#include <immintrin.h>
17#include <ammintrin.h>
18#include <mm3dnow.h>
19
20#define _CONCAT(x,y) x ## y
21
22#define test_1(func, type, op1_type, imm)				\
23  type _CONCAT(_,func) (op1_type A, int const I)			\
24  { return func (A, imm); }
25
26#define test_1x(func, type, op1_type, imm1, imm2)			\
27  type _CONCAT(_,func) (op1_type A, int const I, int const L)		\
28  { return func (A, imm1, imm2); }
29
30#define test_2(func, type, op1_type, op2_type, imm)			\
31  type _CONCAT(_,func) (op1_type A, op2_type B, int const I)		\
32  { return func (A, B, imm); }
33
34#define test_2x(func, type, op1_type, op2_type, imm1, imm2)		\
35  type _CONCAT(_,func) (op1_type A, op2_type B, int const I, int const L) \
36  { return func (A, B, imm1, imm2); }
37
38#define test_3(func, type, op1_type, op2_type, op3_type, imm)	\
39  type _CONCAT(_,func) (op1_type A, op2_type B,				\
40			op3_type C, int const I)		\
41  { return func (A, B, C, imm); }
42
43#define test_4(func, type, op1_type, op2_type, op3_type, op4_type, imm)	\
44  type _CONCAT(_,func) (op1_type A, op2_type B,				\
45			op3_type C, op4_type D, int const I)		\
46  { return func (A, B, C, D, imm); }
47
48
49/* Following intrinsics require immediate arguments.  They
50   are defined as macros for non-optimized compilations. */
51
52/* ammintrin.h */
53test_1x (_mm_extracti_si64, __m128i, __m128i, 1, 1)
54test_2x (_mm_inserti_si64, __m128i, __m128i, __m128i, 1, 1)
55
56/* immintrin.h */
57test_2 (_mm256_blend_pd, __m256d, __m256d, __m256d, 1)
58test_2 (_mm256_blend_ps, __m256, __m256, __m256, 1)
59test_2 (_mm256_dp_ps, __m256, __m256, __m256, 1)
60test_2 (_mm256_shuffle_pd, __m256d, __m256d, __m256d, 1)
61test_2 (_mm256_shuffle_ps, __m256, __m256, __m256, 1)
62test_2 (_mm_cmp_sd, __m128d, __m128d, __m128d, 1)
63test_2 (_mm_cmp_ss, __m128, __m128, __m128, 1)
64test_2 (_mm_cmp_pd, __m128d, __m128d, __m128d, 1)
65test_2 (_mm_cmp_ps, __m128, __m128, __m128, 1)
66test_2 (_mm256_cmp_pd, __m256d, __m256d, __m256d, 1)
67test_2 (_mm256_cmp_ps, __m256, __m256, __m256, 1)
68test_1 (_mm256_extractf128_pd, __m128d, __m256d, 1)
69test_1 (_mm256_extractf128_ps, __m128, __m256, 1)
70test_1 (_mm256_extractf128_si256, __m128i, __m256i, 1)
71test_1 (_mm256_extract_epi8, int, __m256i, 20)
72test_1 (_mm256_extract_epi16, int, __m256i, 13)
73test_1 (_mm256_extract_epi32, int, __m256i, 6)
74#ifdef __x86_64__
75test_1 (_mm256_extract_epi64, long long, __m256i, 2)
76#endif
77test_1 (_mm_permute_pd, __m128d, __m128d, 1)
78test_1 (_mm256_permute_pd, __m256d, __m256d, 1)
79test_1 (_mm_permute_ps, __m128, __m128, 1)
80test_1 (_mm256_permute_ps, __m256, __m256, 1)
81test_2 (_mm256_permute2f128_pd, __m256d, __m256d, __m256d, 1)
82test_2 (_mm256_permute2f128_ps, __m256, __m256, __m256, 1)
83test_2 (_mm256_permute2f128_si256, __m256i, __m256i, __m256i, 1)
84test_2 (_mm256_insertf128_pd, __m256d, __m256d, __m128d, 1)
85test_2 (_mm256_insertf128_ps, __m256, __m256, __m128, 1)
86test_2 (_mm256_insertf128_si256, __m256i, __m256i, __m128i, 1)
87test_2 (_mm256_insert_epi8, __m256i, __m256i, int, 30)
88test_2 (_mm256_insert_epi16, __m256i, __m256i, int, 7)
89test_2 (_mm256_insert_epi32, __m256i, __m256i, int, 3)
90#ifdef __x86_64__
91test_2 (_mm256_insert_epi64, __m256i, __m256i, long long, 1)
92#endif
93test_1 (_mm256_round_pd, __m256d, __m256d, 1)
94test_1 (_mm256_round_ps, __m256, __m256, 1)
95
96/* wmmintrin.h */
97test_1 (_mm_aeskeygenassist_si128, __m128i, __m128i, 1)
98test_2 (_mm_clmulepi64_si128, __m128i, __m128i, __m128i, 1)
99
100/* smmintrin.h */
101test_1 (_mm_round_pd, __m128d, __m128d, 1)
102test_1 (_mm_round_ps, __m128, __m128, 1)
103test_2 (_mm_round_sd, __m128d, __m128d, __m128d, 1)
104test_2 (_mm_round_ss, __m128, __m128, __m128, 1)
105
106test_2 (_mm_blend_epi16, __m128i, __m128i, __m128i, 1)
107test_2 (_mm_blend_ps, __m128, __m128, __m128, 1)
108test_2 (_mm_blend_pd, __m128d, __m128d, __m128d, 1)
109test_2 (_mm_dp_ps, __m128, __m128, __m128, 1)
110test_2 (_mm_dp_pd, __m128d, __m128d, __m128d, 1)
111test_2 (_mm_insert_ps, __m128, __m128, __m128, 1)
112test_1 (_mm_extract_ps, int, __m128, 1)
113test_2 (_mm_insert_epi8, __m128i, __m128i, int, 1)
114test_2 (_mm_insert_epi32, __m128i, __m128i, int, 1)
115#ifdef __x86_64__
116test_2 (_mm_insert_epi64, __m128i, __m128i, long long, 1)
117#endif
118test_1 (_mm_extract_epi8, int, __m128i, 1)
119test_1 (_mm_extract_epi32, int, __m128i, 1)
120#ifdef __x86_64__
121test_1 (_mm_extract_epi64, long long, __m128i, 1)
122#endif
123test_2 (_mm_mpsadbw_epu8, __m128i, __m128i, __m128i, 1)
124test_2 (_mm_cmpistrm, __m128i, __m128i, __m128i, 1)
125test_2 (_mm_cmpistri, int, __m128i, __m128i, 1)
126test_4 (_mm_cmpestrm, __m128i, __m128i, int, __m128i, int, 1)
127test_4 (_mm_cmpestri, int, __m128i, int, __m128i, int, 1)
128test_2 (_mm_cmpistra, int, __m128i, __m128i, 1)
129test_2 (_mm_cmpistrc, int, __m128i, __m128i, 1)
130test_2 (_mm_cmpistro, int, __m128i, __m128i, 1)
131test_2 (_mm_cmpistrs, int, __m128i, __m128i, 1)
132test_2 (_mm_cmpistrz, int, __m128i, __m128i, 1)
133test_4 (_mm_cmpestra, int, __m128i, int, __m128i, int, 1)
134test_4 (_mm_cmpestrc, int, __m128i, int, __m128i, int, 1)
135test_4 (_mm_cmpestro, int, __m128i, int, __m128i, int, 1)
136test_4 (_mm_cmpestrs, int, __m128i, int, __m128i, int, 1)
137test_4 (_mm_cmpestrz, int, __m128i, int, __m128i, int, 1)
138
139/* tmmintrin.h */
140test_2 (_mm_alignr_epi8, __m128i, __m128i, __m128i, 1)
141test_2 (_mm_alignr_pi8, __m64, __m64, __m64, 1)
142
143/* emmintrin.h */
144test_2 (_mm_shuffle_pd, __m128d, __m128d, __m128d, 1)
145test_1 (_mm_srli_si128, __m128i, __m128i, 1)
146test_1 (_mm_slli_si128, __m128i, __m128i, 1)
147test_1 (_mm_extract_epi16, int, __m128i, 1)
148test_2 (_mm_insert_epi16, __m128i, __m128i, int, 1)
149test_1 (_mm_shufflehi_epi16, __m128i, __m128i, 1)
150test_1 (_mm_shufflelo_epi16, __m128i, __m128i, 1)
151test_1 (_mm_shuffle_epi32, __m128i, __m128i, 1)
152
153/* xmmintrin.h */
154test_2 (_mm_shuffle_ps, __m128, __m128, __m128, 1)
155test_1 (_mm_extract_pi16, int, __m64, 1)
156test_1 (_m_pextrw, int, __m64, 1)
157test_2 (_mm_insert_pi16, __m64, __m64, int, 1)
158test_2 (_m_pinsrw, __m64, __m64, int, 1)
159test_1 (_mm_shuffle_pi16, __m64, __m64, 1)
160test_1 (_m_pshufw, __m64, __m64, 1)
161test_1 (_mm_prefetch, void, void *, _MM_HINT_NTA)
162
163