Initial support for AVX-512{VL,BW,DQ}
[official-gcc.git] / gcc / testsuite / gcc.target / i386 / avx-1.c
blobfd390f9157c2d88446be848fe84db57246564e86
1 /* { dg-do compile } */
2 /* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -m3dnow -mavx -mavx2 -maes -mpclmul" } */
4 #include <mm_malloc.h>
6 /* Test that the intrinsics compile with 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. */
12 #define extern
13 #define __inline
15 /* Following intrinsics require immediate arguments. */
17 /* ammintrin.h */
18 #define __builtin_ia32_extrqi(X, I, L) __builtin_ia32_extrqi(X, 1, 1)
19 #define __builtin_ia32_insertqi(X, Y, I, L) __builtin_ia32_insertqi(X, Y, 1, 1)
21 /* immintrin.h */
22 #define __builtin_ia32_blendpd256(X, Y, M) __builtin_ia32_blendpd256(X, Y, 1)
23 #define __builtin_ia32_blendps256(X, Y, M) __builtin_ia32_blendps256(X, Y, 1)
24 #define __builtin_ia32_dpps256(X, Y, M) __builtin_ia32_dpps256(X, Y, 1)
25 #define __builtin_ia32_shufpd256(X, Y, M) __builtin_ia32_shufpd256(X, Y, 1)
26 #define __builtin_ia32_shufps256(X, Y, M) __builtin_ia32_shufps256(X, Y, 1)
27 #define __builtin_ia32_cmpsd(X, Y, O) __builtin_ia32_cmpsd(X, Y, 1)
28 #define __builtin_ia32_cmpss(X, Y, O) __builtin_ia32_cmpss(X, Y, 1)
29 #define __builtin_ia32_cmppd(X, Y, O) __builtin_ia32_cmppd(X, Y, 1)
30 #define __builtin_ia32_cmpps(X, Y, O) __builtin_ia32_cmpps(X, Y, 1)
31 #define __builtin_ia32_cmppd256(X, Y, O) __builtin_ia32_cmppd256(X, Y, 1)
32 #define __builtin_ia32_cmpps256(X, Y, O) __builtin_ia32_cmpps256(X, Y, 1)
33 #define __builtin_ia32_vextractf128_pd256(X, N) __builtin_ia32_vextractf128_pd256(X, 1)
34 #define __builtin_ia32_vextractf128_ps256(X, N) __builtin_ia32_vextractf128_ps256(X, 1)
35 #define __builtin_ia32_vextractf128_si256(X, N) __builtin_ia32_vextractf128_si256(X, 1)
36 #define __builtin_ia32_vpermilpd(X, N) __builtin_ia32_vpermilpd(X, 1)
37 #define __builtin_ia32_vpermilpd256(X, N) __builtin_ia32_vpermilpd256(X, 1)
38 #define __builtin_ia32_vpermilps(X, N) __builtin_ia32_vpermilps(X, 1)
39 #define __builtin_ia32_vpermilps256(X, N) __builtin_ia32_vpermilps256(X, 1)
40 #define __builtin_ia32_vpermil2pd(X, Y, C, I) __builtin_ia32_vpermil2pd(X, Y, C, 1)
41 #define __builtin_ia32_vpermil2pd256(X, Y, C, I) __builtin_ia32_vpermil2pd256(X, Y, C, 1)
42 #define __builtin_ia32_vpermil2ps(X, Y, C, I) __builtin_ia32_vpermil2ps(X, Y, C, 1)
43 #define __builtin_ia32_vpermil2ps256(X, Y, C, I) __builtin_ia32_vpermil2ps256(X, Y, C, 1)
44 #define __builtin_ia32_vperm2f128_pd256(X, Y, C) __builtin_ia32_vperm2f128_pd256(X, Y, 1)
45 #define __builtin_ia32_vperm2f128_ps256(X, Y, C) __builtin_ia32_vperm2f128_ps256(X, Y, 1)
46 #define __builtin_ia32_vperm2f128_si256(X, Y, C) __builtin_ia32_vperm2f128_si256(X, Y, 1)
47 #define __builtin_ia32_vinsertf128_pd256(X, Y, C) __builtin_ia32_vinsertf128_pd256(X, Y, 1)
48 #define __builtin_ia32_vinsertf128_ps256(X, Y, C) __builtin_ia32_vinsertf128_ps256(X, Y, 1)
49 #define __builtin_ia32_vinsertf128_si256(X, Y, C) __builtin_ia32_vinsertf128_si256(X, Y, 1)
50 #define __builtin_ia32_roundpd256(V, M) __builtin_ia32_roundpd256(V, 1)
51 #define __builtin_ia32_roundps256(V, M) __builtin_ia32_roundps256(V, 1)
52 #define __builtin_ia32_mpsadbw256(X, Y, M) __builtin_ia32_mpsadbw256(X, Y, 1)
53 #define __builtin_ia32_palignr256(X, Y, M) __builtin_ia32_palignr256(X, Y, 8)
54 #define __builtin_ia32_pblendw256(X, Y, M) __builtin_ia32_pblendw256(X, Y, 8)
55 #define __builtin_ia32_pshufd256(X, M) __builtin_ia32_pshufd256(X, 8)
56 #define __builtin_ia32_pshufhw256(X, M) __builtin_ia32_pshufhw256(X, 8)
57 #define __builtin_ia32_pshuflw256(X, M) __builtin_ia32_pshuflw256(X, 8)
58 #define __builtin_ia32_pslldqi256(X, M) __builtin_ia32_pslldqi256(X, 8)
59 #define __builtin_ia32_psrldqi256(X, M) __builtin_ia32_psrldqi256(X, 8)
60 #define __builtin_ia32_pblendd128(X, Y, M) __builtin_ia32_pblendd128(X, Y, 1)
61 #define __builtin_ia32_pblendd256(X, Y, M) __builtin_ia32_pblendd256(X, Y, 1)
62 #define __builtin_ia32_permdf256(X, M) __builtin_ia32_permdf256(X, 1)
63 #define __builtin_ia32_permdi256(X, M) __builtin_ia32_permdi256(X, 1)
64 #define __builtin_ia32_permti256(X, Y, M) __builtin_ia32_permti256(X, Y, 1)
65 #define __builtin_ia32_extract128i256(X, M) __builtin_ia32_extract128i256(X, 1)
66 #define __builtin_ia32_insert128i256(X, Y, M) __builtin_ia32_insert128i256(X, Y, 1)
67 #define __builtin_ia32_gathersiv2df(A, B, C, D, M) __builtin_ia32_gathersiv2df(A, B, C, D, 1)
68 #define __builtin_ia32_gathersiv4df(A, B, C, D, M) __builtin_ia32_gathersiv4df(A, B, C, D, 1)
69 #define __builtin_ia32_gatherdiv2df(A, B, C, D, M) __builtin_ia32_gatherdiv2df(A, B, C, D, 1)
70 #define __builtin_ia32_gatherdiv4df(A, B, C, D, M) __builtin_ia32_gatherdiv4df(A, B, C, D, 1)
71 #define __builtin_ia32_gathersiv4sf(A, B, C, D, M) __builtin_ia32_gathersiv4sf(A, B, C, D, 1)
72 #define __builtin_ia32_gathersiv8sf(A, B, C, D, M) __builtin_ia32_gathersiv8sf(A, B, C, D, 1)
73 #define __builtin_ia32_gatherdiv4sf(A, B, C, D, M) __builtin_ia32_gatherdiv4sf(A, B, C, D, 1)
74 #define __builtin_ia32_gatherdiv4sf256(A, B, C, D, M) \
75 __builtin_ia32_gatherdiv4sf256(A, B, C, D, 1)
76 #define __builtin_ia32_gathersiv2di(A, B, C, D, M) __builtin_ia32_gathersiv2di(A, B, C, D, 1)
77 #define __builtin_ia32_gathersiv4di(A, B, C, D, M) __builtin_ia32_gathersiv4di(A, B, C, D, 1)
78 #define __builtin_ia32_gatherdiv2di(A, B, C, D, M) __builtin_ia32_gatherdiv2di(A, B, C, D, 1)
79 #define __builtin_ia32_gatherdiv4di(A, B, C, D, M) __builtin_ia32_gatherdiv4di(A, B, C, D, 1)
80 #define __builtin_ia32_gathersiv4si(A, B, C, D, M) __builtin_ia32_gathersiv4si(A, B, C, D, 1)
81 #define __builtin_ia32_gathersiv8si(A, B, C, D, M) __builtin_ia32_gathersiv8si(A, B, C, D, 1)
82 #define __builtin_ia32_gatherdiv4si(A, B, C, D, M) __builtin_ia32_gatherdiv4si(A, B, C, D, 1)
83 #define __builtin_ia32_gatherdiv4si256(A, B, C, D, M) \
84 __builtin_ia32_gatherdiv4si256(A, B, C, D, 1)
86 /* wmmintrin.h */
87 #define __builtin_ia32_aeskeygenassist128(X, C) __builtin_ia32_aeskeygenassist128(X, 1)
88 #define __builtin_ia32_pclmulqdq128(X, Y, I) __builtin_ia32_pclmulqdq128(X, Y, 1)
90 /* smmintrin.h */
91 #define __builtin_ia32_roundpd(V, M) __builtin_ia32_roundpd(V, 1)
92 #define __builtin_ia32_roundsd(D, V, M) __builtin_ia32_roundsd(D, V, 1)
93 #define __builtin_ia32_roundps(V, M) __builtin_ia32_roundps(V, 1)
94 #define __builtin_ia32_roundss(D, V, M) __builtin_ia32_roundss(D, V, 1)
96 #define __builtin_ia32_pblendw128(X, Y, M) __builtin_ia32_pblendw128 (X, Y, 1)
97 #define __builtin_ia32_blendps(X, Y, M) __builtin_ia32_blendps(X, Y, 1)
98 #define __builtin_ia32_blendpd(X, Y, M) __builtin_ia32_blendpd(X, Y, 1)
99 #define __builtin_ia32_dpps(X, Y, M) __builtin_ia32_dpps(X, Y, 1)
100 #define __builtin_ia32_dppd(X, Y, M) __builtin_ia32_dppd(X, Y, 1)
101 #define __builtin_ia32_insertps128(D, S, N) __builtin_ia32_insertps128(D, S, 1)
102 #define __builtin_ia32_vec_ext_v4sf(X, N) __builtin_ia32_vec_ext_v4sf(X, 1)
103 #define __builtin_ia32_vec_set_v16qi(D, S, N) __builtin_ia32_vec_set_v16qi(D, S, 1)
104 #define __builtin_ia32_vec_set_v4si(D, S, N) __builtin_ia32_vec_set_v4si(D, S, 1)
105 #define __builtin_ia32_vec_set_v2di(D, S, N) __builtin_ia32_vec_set_v2di(D, S, 1)
106 #define __builtin_ia32_vec_ext_v16qi(X, N) __builtin_ia32_vec_ext_v16qi(X, 1)
107 #define __builtin_ia32_vec_ext_v4si(X, N) __builtin_ia32_vec_ext_v4si(X, 1)
108 #define __builtin_ia32_vec_ext_v2di(X, N) __builtin_ia32_vec_ext_v2di(X, 1)
109 #define __builtin_ia32_mpsadbw128(X, Y, M) __builtin_ia32_mpsadbw128(X, Y, 1)
110 #define __builtin_ia32_pcmpistrm128(X, Y, M) \
111 __builtin_ia32_pcmpistrm128(X, Y, 1)
112 #define __builtin_ia32_pcmpistri128(X, Y, M) \
113 __builtin_ia32_pcmpistri128(X, Y, 1)
114 #define __builtin_ia32_pcmpestrm128(X, LX, Y, LY, M) \
115 __builtin_ia32_pcmpestrm128(X, LX, Y, LY, 1)
116 #define __builtin_ia32_pcmpestri128(X, LX, Y, LY, M) \
117 __builtin_ia32_pcmpestri128(X, LX, Y, LY, 1)
118 #define __builtin_ia32_pcmpistria128(X, Y, M) \
119 __builtin_ia32_pcmpistria128(X, Y, 1)
120 #define __builtin_ia32_pcmpistric128(X, Y, M) \
121 __builtin_ia32_pcmpistric128(X, Y, 1)
122 #define __builtin_ia32_pcmpistrio128(X, Y, M) \
123 __builtin_ia32_pcmpistrio128(X, Y, 1)
124 #define __builtin_ia32_pcmpistris128(X, Y, M) \
125 __builtin_ia32_pcmpistris128(X, Y, 1)
126 #define __builtin_ia32_pcmpistriz128(X, Y, M) \
127 __builtin_ia32_pcmpistriz128(X, Y, 1)
128 #define __builtin_ia32_pcmpestria128(X, LX, Y, LY, M) \
129 __builtin_ia32_pcmpestria128(X, LX, Y, LY, 1)
130 #define __builtin_ia32_pcmpestric128(X, LX, Y, LY, M) \
131 __builtin_ia32_pcmpestric128(X, LX, Y, LY, 1)
132 #define __builtin_ia32_pcmpestrio128(X, LX, Y, LY, M) \
133 __builtin_ia32_pcmpestrio128(X, LX, Y, LY, 1)
134 #define __builtin_ia32_pcmpestris128(X, LX, Y, LY, M) \
135 __builtin_ia32_pcmpestris128(X, LX, Y, LY, 1)
136 #define __builtin_ia32_pcmpestriz128(X, LX, Y, LY, M) \
137 __builtin_ia32_pcmpestriz128(X, LX, Y, LY, 1)
139 /* tmmintrin.h */
140 #define __builtin_ia32_palignr128(X, Y, N) __builtin_ia32_palignr128(X, Y, 8)
141 #define __builtin_ia32_palignr(X, Y, N) __builtin_ia32_palignr(X, Y, 8)
143 /* emmintrin.h */
144 #define __builtin_ia32_psrldqi128(A, B) __builtin_ia32_psrldqi128(A, 8)
145 #define __builtin_ia32_pslldqi128(A, B) __builtin_ia32_pslldqi128(A, 8)
146 #define __builtin_ia32_pshufhw(A, N) __builtin_ia32_pshufhw(A, 0)
147 #define __builtin_ia32_pshuflw(A, N) __builtin_ia32_pshuflw(A, 0)
148 #define __builtin_ia32_pshufd(A, N) __builtin_ia32_pshufd(A, 0)
149 #define __builtin_ia32_vec_set_v8hi(A, D, N) \
150 __builtin_ia32_vec_set_v8hi(A, D, 0)
151 #define __builtin_ia32_vec_ext_v8hi(A, N) __builtin_ia32_vec_ext_v8hi(A, 0)
152 #define __builtin_ia32_shufpd(A, B, N) __builtin_ia32_shufpd(A, B, 0)
154 /* xmmintrin.h */
155 #define __builtin_prefetch(P, A, I) __builtin_prefetch(P, 0, _MM_HINT_NTA)
156 #define __builtin_ia32_pshufw(A, N) __builtin_ia32_pshufw(A, 0)
157 #define __builtin_ia32_vec_set_v4hi(A, D, N) \
158 __builtin_ia32_vec_set_v4hi(A, D, 0)
159 #define __builtin_ia32_vec_ext_v4hi(A, N) __builtin_ia32_vec_ext_v4hi(A, 0)
160 #define __builtin_ia32_shufps(A, B, N) __builtin_ia32_shufps(A, B, 0)
162 /* f16cintrin.h */
163 #define __builtin_ia32_vcvtps2ph(A, I) __builtin_ia32_vcvtps2ph(A, 0)
164 #define __builtin_ia32_vcvtps2ph256(A, I) __builtin_ia32_vcvtps2ph256(A, 0)
166 /* rtmintrin.h */
167 #define __builtin_ia32_xabort(I) __builtin_ia32_xabort(0)
169 /* avx512fintrin.h */
170 #define __builtin_ia32_addpd512_mask(A, B, C, D, E) __builtin_ia32_addpd512_mask(A, B, C, D, 8)
171 #define __builtin_ia32_addps512_mask(A, B, C, D, E) __builtin_ia32_addps512_mask(A, B, C, D, 8)
172 #define __builtin_ia32_addsd_round(A, B, C) __builtin_ia32_addsd_round(A, B, 8)
173 #define __builtin_ia32_addss_round(A, B, C) __builtin_ia32_addss_round(A, B, 8)
174 #define __builtin_ia32_alignd512_mask(A, B, F, D, E) __builtin_ia32_alignd512_mask(A, B, 1, D, E)
175 #define __builtin_ia32_alignq512_mask(A, B, F, D, E) __builtin_ia32_alignq512_mask(A, B, 1, D, E)
176 #define __builtin_ia32_cmpd512_mask(A, B, E, D) __builtin_ia32_cmpd512_mask(A, B, 1, D)
177 #define __builtin_ia32_cmppd512_mask(A, B, F, D, E) __builtin_ia32_cmppd512_mask(A, B, 1, D, 8)
178 #define __builtin_ia32_cmpps512_mask(A, B, F, D, E) __builtin_ia32_cmpps512_mask(A, B, 1, D, 8)
179 #define __builtin_ia32_cmpq512_mask(A, B, E, D) __builtin_ia32_cmpq512_mask(A, B, 1, D)
180 #define __builtin_ia32_cmpsd_mask(A, B, F, D, E) __builtin_ia32_cmpsd_mask(A, B, 1, D, 8)
181 #define __builtin_ia32_cmpss_mask(A, B, F, D, E) __builtin_ia32_cmpss_mask(A, B, 1, D, 8)
182 #define __builtin_ia32_cvtdq2ps512_mask(A, B, C, D) __builtin_ia32_cvtdq2ps512_mask(A, B, C, 8)
183 #define __builtin_ia32_cvtpd2dq512_mask(A, B, C, D) __builtin_ia32_cvtpd2dq512_mask(A, B, C, 8)
184 #define __builtin_ia32_cvtpd2ps512_mask(A, B, C, D) __builtin_ia32_cvtpd2ps512_mask(A, B, C, 8)
185 #define __builtin_ia32_cvtpd2udq512_mask(A, B, C, D) __builtin_ia32_cvtpd2udq512_mask(A, B, C, 8)
186 #define __builtin_ia32_cvtps2dq512_mask(A, B, C, D) __builtin_ia32_cvtps2dq512_mask(A, B, C, 8)
187 #define __builtin_ia32_cvtps2pd512_mask(A, B, C, D) __builtin_ia32_cvtps2pd512_mask(A, B, C, 8)
188 #define __builtin_ia32_cvtps2udq512_mask(A, B, C, D) __builtin_ia32_cvtps2udq512_mask(A, B, C, 8)
189 #define __builtin_ia32_cvtsd2ss_round(A, B, C) __builtin_ia32_cvtsd2ss_round(A, B, 8)
190 #define __builtin_ia32_cvtss2sd_round(A, B, C) __builtin_ia32_cvtss2sd_round(A, B, 4)
191 #define __builtin_ia32_cvtsi2sd64(A, B, C) __builtin_ia32_cvtsi2sd64(A, B, 8)
192 #define __builtin_ia32_cvtsi2ss32(A, B, C) __builtin_ia32_cvtsi2ss32(A, B, 8)
193 #define __builtin_ia32_cvtsi2ss64(A, B, C) __builtin_ia32_cvtsi2ss64(A, B, 8)
194 #define __builtin_ia32_cvttpd2dq512_mask(A, B, C, D) __builtin_ia32_cvttpd2dq512_mask(A, B, C, 8)
195 #define __builtin_ia32_cvttpd2udq512_mask(A, B, C, D) __builtin_ia32_cvttpd2udq512_mask(A, B, C, 8)
196 #define __builtin_ia32_cvttps2dq512_mask(A, B, C, D) __builtin_ia32_cvttps2dq512_mask(A, B, C, 8)
197 #define __builtin_ia32_cvttps2udq512_mask(A, B, C, D) __builtin_ia32_cvttps2udq512_mask(A, B, C, 8)
198 #define __builtin_ia32_cvtudq2ps512_mask(A, B, C, D) __builtin_ia32_cvtudq2ps512_mask(A, B, C, 8)
199 #define __builtin_ia32_cvtusi2sd64(A, B, C) __builtin_ia32_cvtusi2sd64(A, B, 8)
200 #define __builtin_ia32_cvtusi2ss32(A, B, C) __builtin_ia32_cvtusi2ss32(A, B, 8)
201 #define __builtin_ia32_cvtusi2ss64(A, B, C) __builtin_ia32_cvtusi2ss64(A, B, 8)
202 #define __builtin_ia32_divpd512_mask(A, B, C, D, E) __builtin_ia32_divpd512_mask(A, B, C, D, 8)
203 #define __builtin_ia32_divps512_mask(A, B, C, D, E) __builtin_ia32_divps512_mask(A, B, C, D, 8)
204 #define __builtin_ia32_divsd_round(A, B, C) __builtin_ia32_divsd_round(A, B, 8)
205 #define __builtin_ia32_divss_round(A, B, C) __builtin_ia32_divss_round(A, B, 8)
206 #define __builtin_ia32_extractf32x4_mask(A, E, C, D) __builtin_ia32_extractf32x4_mask(A, 1, C, D)
207 #define __builtin_ia32_extractf64x4_mask(A, E, C, D) __builtin_ia32_extractf64x4_mask(A, 1, C, D)
208 #define __builtin_ia32_extracti32x4_mask(A, E, C, D) __builtin_ia32_extracti32x4_mask(A, 1, C, D)
209 #define __builtin_ia32_extracti64x4_mask(A, E, C, D) __builtin_ia32_extracti64x4_mask(A, 1, C, D)
210 #define __builtin_ia32_fixupimmpd512_mask(A, B, C, I, E, F) __builtin_ia32_fixupimmpd512_mask(A, B, C, 1, E, 8)
211 #define __builtin_ia32_fixupimmpd512_maskz(A, B, C, I, E, F) __builtin_ia32_fixupimmpd512_maskz(A, B, C, 1, E, 8)
212 #define __builtin_ia32_fixupimmps512_mask(A, B, C, I, E, F) __builtin_ia32_fixupimmps512_mask(A, B, C, 1, E, 8)
213 #define __builtin_ia32_fixupimmps512_maskz(A, B, C, I, E, F) __builtin_ia32_fixupimmps512_maskz(A, B, C, 1, E, 8)
214 #define __builtin_ia32_fixupimmsd_mask(A, B, C, I, E, F) __builtin_ia32_fixupimmsd_mask(A, B, C, 1, E, 8)
215 #define __builtin_ia32_fixupimmsd_maskz(A, B, C, I, E, F) __builtin_ia32_fixupimmsd_maskz(A, B, C, 1, E, 8)
216 #define __builtin_ia32_fixupimmss_mask(A, B, C, I, E, F) __builtin_ia32_fixupimmss_mask(A, B, C, 1, E, 8)
217 #define __builtin_ia32_fixupimmss_maskz(A, B, C, I, E, F) __builtin_ia32_fixupimmss_maskz(A, B, C, 1, E, 8)
218 #define __builtin_ia32_gatherdiv8df(A, B, C, D, F) __builtin_ia32_gatherdiv8df(A, B, C, D, 8)
219 #define __builtin_ia32_gatherdiv8di(A, B, C, D, F) __builtin_ia32_gatherdiv8di(A, B, C, D, 8)
220 #define __builtin_ia32_gatherdiv16sf(A, B, C, D, F) __builtin_ia32_gatherdiv16sf(A, B, C, D, 8)
221 #define __builtin_ia32_gatherdiv16si(A, B, C, D, F) __builtin_ia32_gatherdiv16si(A, B, C, D, 8)
222 #define __builtin_ia32_gathersiv16sf(A, B, C, D, F) __builtin_ia32_gathersiv16sf(A, B, C, D, 8)
223 #define __builtin_ia32_gathersiv16si(A, B, C, D, F) __builtin_ia32_gathersiv16si(A, B, C, D, 8)
224 #define __builtin_ia32_gathersiv8df(A, B, C, D, F) __builtin_ia32_gathersiv8df(A, B, C, D, 8)
225 #define __builtin_ia32_gathersiv8di(A, B, C, D, F) __builtin_ia32_gathersiv8di(A, B, C, D, 8)
226 #define __builtin_ia32_getexppd512_mask(A, B, C, D) __builtin_ia32_getexppd512_mask(A, B, C, 8)
227 #define __builtin_ia32_getexpps512_mask(A, B, C, D) __builtin_ia32_getexpps512_mask(A, B, C, 8)
228 #define __builtin_ia32_getexpsd128_round(A, B, C) __builtin_ia32_getexpsd128_round(A, B, 4)
229 #define __builtin_ia32_getexpss128_round(A, B, C) __builtin_ia32_getexpss128_round(A, B, 4)
230 #define __builtin_ia32_getmantpd512_mask(A, F, C, D, E) __builtin_ia32_getmantpd512_mask(A, 1, C, D, 8)
231 #define __builtin_ia32_getmantps512_mask(A, F, C, D, E) __builtin_ia32_getmantps512_mask(A, 1, C, D, 8)
232 #define __builtin_ia32_getmantsd_round(A, B, C, D) __builtin_ia32_getmantsd_round(A, B, 1, 4)
233 #define __builtin_ia32_getmantss_round(A, B, C, D) __builtin_ia32_getmantss_round(A, B, 1, 4)
234 #define __builtin_ia32_insertf32x4_mask(A, B, F, D, E) __builtin_ia32_insertf32x4_mask(A, B, 1, D, E)
235 #define __builtin_ia32_insertf64x4_mask(A, B, F, D, E) __builtin_ia32_insertf64x4_mask(A, B, 1, D, E)
236 #define __builtin_ia32_inserti32x4_mask(A, B, F, D, E) __builtin_ia32_inserti32x4_mask(A, B, 1, D, E)
237 #define __builtin_ia32_inserti64x4_mask(A, B, F, D, E) __builtin_ia32_inserti64x4_mask(A, B, 1, D, E)
238 #define __builtin_ia32_maxpd512_mask(A, B, C, D, E) __builtin_ia32_maxpd512_mask(A, B, C, D, 8)
239 #define __builtin_ia32_maxps512_mask(A, B, C, D, E) __builtin_ia32_maxps512_mask(A, B, C, D, 8)
240 #define __builtin_ia32_maxsd_round(A, B, C) __builtin_ia32_maxsd_round(A, B, 4)
241 #define __builtin_ia32_maxss_round(A, B, C) __builtin_ia32_maxss_round(A, B, 4)
242 #define __builtin_ia32_minpd512_mask(A, B, C, D, E) __builtin_ia32_minpd512_mask(A, B, C, D, 8)
243 #define __builtin_ia32_minps512_mask(A, B, C, D, E) __builtin_ia32_minps512_mask(A, B, C, D, 8)
244 #define __builtin_ia32_minsd_round(A, B, C) __builtin_ia32_minsd_round(A, B, 4)
245 #define __builtin_ia32_minss_round(A, B, C) __builtin_ia32_minss_round(A, B, 4)
246 #define __builtin_ia32_mulpd512_mask(A, B, C, D, E) __builtin_ia32_mulpd512_mask(A, B, C, D, 8)
247 #define __builtin_ia32_mulps512_mask(A, B, C, D, E) __builtin_ia32_mulps512_mask(A, B, C, D, 8)
248 #define __builtin_ia32_mulsd_round(A, B, C) __builtin_ia32_mulsd_round(A, B, 8)
249 #define __builtin_ia32_mulss_round(A, B, C) __builtin_ia32_mulss_round(A, B, 8)
250 #define __builtin_ia32_permdf512_mask(A, E, C, D) __builtin_ia32_permdf512_mask(A, 1, C, D)
251 #define __builtin_ia32_permdi512_mask(A, E, C, D) __builtin_ia32_permdi512_mask(A, 1, C, D)
252 #define __builtin_ia32_prold512_mask(A, E, C, D) __builtin_ia32_prold512_mask(A, 1, C, D)
253 #define __builtin_ia32_prolq512_mask(A, E, C, D) __builtin_ia32_prolq512_mask(A, 1, C, D)
254 #define __builtin_ia32_prord512_mask(A, E, C, D) __builtin_ia32_prord512_mask(A, 1, C, D)
255 #define __builtin_ia32_prorq512_mask(A, E, C, D) __builtin_ia32_prorq512_mask(A, 1, C, D)
256 #define __builtin_ia32_pshufd512_mask(A, E, C, D) __builtin_ia32_pshufd512_mask(A, 1, C, D)
257 #define __builtin_ia32_pslldi512_mask(A, E, C, D) __builtin_ia32_pslldi512_mask(A, 1, C, D)
258 #define __builtin_ia32_psllqi512_mask(A, E, C, D) __builtin_ia32_psllqi512_mask(A, 1, C, D)
259 #define __builtin_ia32_psradi512_mask(A, E, C, D) __builtin_ia32_psradi512_mask(A, 1, C, D)
260 #define __builtin_ia32_psraqi512_mask(A, E, C, D) __builtin_ia32_psraqi512_mask(A, 1, C, D)
261 #define __builtin_ia32_psrldi512_mask(A, E, C, D) __builtin_ia32_psrldi512_mask(A, 1, C, D)
262 #define __builtin_ia32_psrlqi512_mask(A, E, C, D) __builtin_ia32_psrlqi512_mask(A, 1, C, D)
263 #define __builtin_ia32_pternlogd512_mask(A, B, C, F, E) __builtin_ia32_pternlogd512_mask(A, B, C, 1, E)
264 #define __builtin_ia32_pternlogd512_maskz(A, B, C, F, E) __builtin_ia32_pternlogd512_maskz(A, B, C, 1, E)
265 #define __builtin_ia32_pternlogq512_mask(A, B, C, F, E) __builtin_ia32_pternlogq512_mask(A, B, C, 1, E)
266 #define __builtin_ia32_pternlogq512_maskz(A, B, C, F, E) __builtin_ia32_pternlogq512_maskz(A, B, C, 1, E)
267 #define __builtin_ia32_rndscalepd_mask(A, F, C, D, E) __builtin_ia32_rndscalepd_mask(A, 1, C, D, 8)
268 #define __builtin_ia32_rndscaleps_mask(A, F, C, D, E) __builtin_ia32_rndscaleps_mask(A, 1, C, D, 8)
269 #define __builtin_ia32_rndscalesd_round(A, B, C, D) __builtin_ia32_rndscalesd_round(A, B, 1, 4)
270 #define __builtin_ia32_rndscaless_round(A, B, C, D) __builtin_ia32_rndscaless_round(A, B, 1, 4)
271 #define __builtin_ia32_scalefpd512_mask(A, B, C, D, E) __builtin_ia32_scalefpd512_mask(A, B, C, D, 8)
272 #define __builtin_ia32_scalefps512_mask(A, B, C, D, E) __builtin_ia32_scalefps512_mask(A, B, C, D, 8)
273 #define __builtin_ia32_scalefsd_round(A, B, C) __builtin_ia32_scalefsd_round(A, B, 8)
274 #define __builtin_ia32_scalefss_round(A, B, C) __builtin_ia32_scalefss_round(A, B, 8)
275 #define __builtin_ia32_scatterdiv8df(A, B, C, D, F) __builtin_ia32_scatterdiv8df(A, B, C, D, 8)
276 #define __builtin_ia32_scatterdiv8di(A, B, C, D, F) __builtin_ia32_scatterdiv8di(A, B, C, D, 8)
277 #define __builtin_ia32_scatterdiv16sf(A, B, C, D, F) __builtin_ia32_scatterdiv16sf(A, B, C, D, 8)
278 #define __builtin_ia32_scatterdiv16si(A, B, C, D, F) __builtin_ia32_scatterdiv16si(A, B, C, D, 8)
279 #define __builtin_ia32_scattersiv16sf(A, B, C, D, F) __builtin_ia32_scattersiv16sf(A, B, C, D, 8)
280 #define __builtin_ia32_scattersiv16si(A, B, C, D, F) __builtin_ia32_scattersiv16si(A, B, C, D, 8)
281 #define __builtin_ia32_scattersiv8df(A, B, C, D, F) __builtin_ia32_scattersiv8df(A, B, C, D, 8)
282 #define __builtin_ia32_scattersiv8di(A, B, C, D, F) __builtin_ia32_scattersiv8di(A, B, C, D, 8)
283 #define __builtin_ia32_shuf_f32x4_mask(A, B, F, D, E) __builtin_ia32_shuf_f32x4_mask(A, B, 1, D, E)
284 #define __builtin_ia32_shuf_f64x2_mask(A, B, F, D, E) __builtin_ia32_shuf_f64x2_mask(A, B, 1, D, E)
285 #define __builtin_ia32_shuf_i32x4_mask(A, B, F, D, E) __builtin_ia32_shuf_i32x4_mask(A, B, 1, D, E)
286 #define __builtin_ia32_shuf_i64x2_mask(A, B, F, D, E) __builtin_ia32_shuf_i64x2_mask(A, B, 1, D, E)
287 #define __builtin_ia32_shufpd512_mask(A, B, F, D, E) __builtin_ia32_shufpd512_mask(A, B, 1, D, E)
288 #define __builtin_ia32_shufps512_mask(A, B, F, D, E) __builtin_ia32_shufps512_mask(A, B, 1, D, E)
289 #define __builtin_ia32_sqrtpd512_mask(A, B, C, D) __builtin_ia32_sqrtpd512_mask(A, B, C, 8)
290 #define __builtin_ia32_sqrtps512_mask(A, B, C, D) __builtin_ia32_sqrtps512_mask(A, B, C, 8)
291 #define __builtin_ia32_sqrtss_round(A, B, C) __builtin_ia32_sqrtss_round(A, B, 8)
292 #define __builtin_ia32_sqrtsd_round(A, B, C) __builtin_ia32_sqrtsd_round(A, B, 8)
293 #define __builtin_ia32_subpd512_mask(A, B, C, D, E) __builtin_ia32_subpd512_mask(A, B, C, D, 8)
294 #define __builtin_ia32_subps512_mask(A, B, C, D, E) __builtin_ia32_subps512_mask(A, B, C, D, 8)
295 #define __builtin_ia32_subsd_round(A, B, C) __builtin_ia32_subsd_round(A, B, 8)
296 #define __builtin_ia32_subss_round(A, B, C) __builtin_ia32_subss_round(A, B, 8)
297 #define __builtin_ia32_ucmpd512_mask(A, B, E, D) __builtin_ia32_ucmpd512_mask(A, B, 1, D)
298 #define __builtin_ia32_ucmpq512_mask(A, B, E, D) __builtin_ia32_ucmpq512_mask(A, B, 1, D)
299 #define __builtin_ia32_vcomisd(A, B, C, D) __builtin_ia32_vcomisd(A, B, 1, 8)
300 #define __builtin_ia32_vcomiss(A, B, C, D) __builtin_ia32_vcomiss(A, B, 1, 8)
301 #define __builtin_ia32_vcvtph2ps512_mask(A, B, C, D) __builtin_ia32_vcvtph2ps512_mask(A, B, C, 8)
302 #define __builtin_ia32_vcvtps2ph512_mask(A, E, C, D) __builtin_ia32_vcvtps2ph512_mask(A, 1, C, D)
303 #define __builtin_ia32_vcvtsd2si32(A, B) __builtin_ia32_vcvtsd2si32(A, 8)
304 #define __builtin_ia32_vcvtsd2si64(A, B) __builtin_ia32_vcvtsd2si64(A, 8)
305 #define __builtin_ia32_vcvtsd2usi32(A, B) __builtin_ia32_vcvtsd2usi32(A, 8)
306 #define __builtin_ia32_vcvtsd2usi64(A, B) __builtin_ia32_vcvtsd2usi64(A, 8)
307 #define __builtin_ia32_vcvtss2si32(A, B) __builtin_ia32_vcvtss2si32(A, 8)
308 #define __builtin_ia32_vcvtss2si64(A, B) __builtin_ia32_vcvtss2si64(A, 8)
309 #define __builtin_ia32_vcvtss2usi32(A, B) __builtin_ia32_vcvtss2usi32(A, 8)
310 #define __builtin_ia32_vcvtss2usi64(A, B) __builtin_ia32_vcvtss2usi64(A, 8)
311 #define __builtin_ia32_vcvttsd2si32(A, B) __builtin_ia32_vcvttsd2si32(A, 8)
312 #define __builtin_ia32_vcvttsd2si64(A, B) __builtin_ia32_vcvttsd2si64(A, 8)
313 #define __builtin_ia32_vcvttsd2usi32(A, B) __builtin_ia32_vcvttsd2usi32(A, 8)
314 #define __builtin_ia32_vcvttsd2usi64(A, B) __builtin_ia32_vcvttsd2usi64(A, 8)
315 #define __builtin_ia32_vcvttss2si32(A, B) __builtin_ia32_vcvttss2si32(A, 8)
316 #define __builtin_ia32_vcvttss2si64(A, B) __builtin_ia32_vcvttss2si64(A, 8)
317 #define __builtin_ia32_vcvttss2usi32(A, B) __builtin_ia32_vcvttss2usi32(A, 8)
318 #define __builtin_ia32_vcvttss2usi64(A, B) __builtin_ia32_vcvttss2usi64(A, 8)
319 #define __builtin_ia32_vfmaddpd512_mask(A, B, C, D, E) __builtin_ia32_vfmaddpd512_mask(A, B, C, D, 8)
320 #define __builtin_ia32_vfmaddpd512_mask3(A, B, C, D, E) __builtin_ia32_vfmaddpd512_mask3(A, B, C, D, 8)
321 #define __builtin_ia32_vfmaddpd512_maskz(A, B, C, D, E) __builtin_ia32_vfmaddpd512_maskz(A, B, C, D, 8)
322 #define __builtin_ia32_vfmaddps512_mask(A, B, C, D, E) __builtin_ia32_vfmaddps512_mask(A, B, C, D, 8)
323 #define __builtin_ia32_vfmaddps512_mask3(A, B, C, D, E) __builtin_ia32_vfmaddps512_mask3(A, B, C, D, 8)
324 #define __builtin_ia32_vfmaddps512_maskz(A, B, C, D, E) __builtin_ia32_vfmaddps512_maskz(A, B, C, D, 8)
325 #define __builtin_ia32_vfmaddsd3_round(A, B, C, D) __builtin_ia32_vfmaddsd3_round(A, B, C, 8)
326 #define __builtin_ia32_vfmaddss3_round(A, B, C, D) __builtin_ia32_vfmaddss3_round(A, B, C, 8)
327 #define __builtin_ia32_vfmaddsubpd512_mask(A, B, C, D, E) __builtin_ia32_vfmaddsubpd512_mask(A, B, C, D, 8)
328 #define __builtin_ia32_vfmaddsubpd512_mask3(A, B, C, D, E) __builtin_ia32_vfmaddsubpd512_mask3(A, B, C, D, 8)
329 #define __builtin_ia32_vfmaddsubpd512_maskz(A, B, C, D, E) __builtin_ia32_vfmaddsubpd512_maskz(A, B, C, D, 8)
330 #define __builtin_ia32_vfmaddsubps512_mask(A, B, C, D, E) __builtin_ia32_vfmaddsubps512_mask(A, B, C, D, 8)
331 #define __builtin_ia32_vfmaddsubps512_mask3(A, B, C, D, E) __builtin_ia32_vfmaddsubps512_mask3(A, B, C, D, 8)
332 #define __builtin_ia32_vfmaddsubps512_maskz(A, B, C, D, E) __builtin_ia32_vfmaddsubps512_maskz(A, B, C, D, 8)
333 #define __builtin_ia32_vfmsubaddpd512_mask3(A, B, C, D, E) __builtin_ia32_vfmsubaddpd512_mask3(A, B, C, D, 8)
334 #define __builtin_ia32_vfmsubaddps512_mask3(A, B, C, D, E) __builtin_ia32_vfmsubaddps512_mask3(A, B, C, D, 8)
335 #define __builtin_ia32_vfmsubpd512_mask3(A, B, C, D, E) __builtin_ia32_vfmsubpd512_mask3(A, B, C, D, 8)
336 #define __builtin_ia32_vfmsubps512_mask3(A, B, C, D, E) __builtin_ia32_vfmsubps512_mask3(A, B, C, D, 8)
337 #define __builtin_ia32_vfmsubsd3_mask3(A, B, C, D, E) __builtin_ia32_vfmsubsd3_mask3(A, B, C, D, 8)
338 #define __builtin_ia32_vfmsubss3_mask3(A, B, C, D, E) __builtin_ia32_vfmsubss3_mask3(A, B, C, D, 8)
339 #define __builtin_ia32_vfnmaddpd512_mask(A, B, C, D, E) __builtin_ia32_vfnmaddpd512_mask(A, B, C, D, 8)
340 #define __builtin_ia32_vfnmaddps512_mask(A, B, C, D, E) __builtin_ia32_vfnmaddps512_mask(A, B, C, D, 8)
341 #define __builtin_ia32_vfnmsubpd512_mask(A, B, C, D, E) __builtin_ia32_vfnmsubpd512_mask(A, B, C, D, 8)
342 #define __builtin_ia32_vfnmsubpd512_mask3(A, B, C, D, E) __builtin_ia32_vfnmsubpd512_mask3(A, B, C, D, 8)
343 #define __builtin_ia32_vfnmsubps512_mask(A, B, C, D, E) __builtin_ia32_vfnmsubps512_mask(A, B, C, D, 8)
344 #define __builtin_ia32_vfnmsubps512_mask3(A, B, C, D, E) __builtin_ia32_vfnmsubps512_mask3(A, B, C, D, 8)
345 #define __builtin_ia32_vpermilpd512_mask(A, E, C, D) __builtin_ia32_vpermilpd512_mask(A, 1, C, D)
346 #define __builtin_ia32_vpermilps512_mask(A, E, C, D) __builtin_ia32_vpermilps512_mask(A, 1, C, D)
348 /* avx512erintrin.h */
349 #define __builtin_ia32_exp2ps_mask(A, B, C, D) __builtin_ia32_exp2ps_mask(A, B, C, 8)
350 #define __builtin_ia32_exp2pd_mask(A, B, C, D) __builtin_ia32_exp2pd_mask(A, B, C, 8)
351 #define __builtin_ia32_rcp28ps_mask(A, B, C, D) __builtin_ia32_rcp28ps_mask(A, B, C, 8)
352 #define __builtin_ia32_rcp28pd_mask(A, B, C, D) __builtin_ia32_rcp28pd_mask(A, B, C, 8)
353 #define __builtin_ia32_rsqrt28ps_mask(A, B, C, D) __builtin_ia32_rsqrt28ps_mask(A, B, C, 8)
354 #define __builtin_ia32_rsqrt28pd_mask(A, B, C, D) __builtin_ia32_rsqrt28pd_mask(A, B, C, 8)
355 #define __builtin_ia32_rcp28ss_round(A, B, C) __builtin_ia32_rcp28ss_round(A, B, 8)
356 #define __builtin_ia32_rcp28sd_round(A, B, C) __builtin_ia32_rcp28sd_round(A, B, 8)
357 #define __builtin_ia32_rsqrt28ss_round(A, B, C) __builtin_ia32_rsqrt28ss_round(A, B, 8)
358 #define __builtin_ia32_rsqrt28sd_round(A, B, C) __builtin_ia32_rsqrt28sd_round(A, B, 8)
360 /* avx512pfintrin.h */
361 #define __builtin_ia32_gatherpfdps(A, B, C, D, E) __builtin_ia32_gatherpfdps(A, B, C, 1, _MM_HINT_T0)
362 #define __builtin_ia32_gatherpfqps(A, B, C, D, E) __builtin_ia32_gatherpfqps(A, B, C, 1, _MM_HINT_T0)
363 #define __builtin_ia32_scatterpfdps(A, B, C, D, E) __builtin_ia32_scatterpfdps(A, B, C, 1, _MM_HINT_T0)
364 #define __builtin_ia32_scatterpfqps(A, B, C, D, E) __builtin_ia32_scatterpfqps(A, B, C, 1, _MM_HINT_T0)
365 #define __builtin_ia32_gatherpfdpd(A, B, C, D, E) __builtin_ia32_gatherpfdpd(A, B, C, 1, _MM_HINT_T0)
366 #define __builtin_ia32_gatherpfqpd(A, B, C, D, E) __builtin_ia32_gatherpfqpd(A, B, C, 1, _MM_HINT_T0)
367 #define __builtin_ia32_scatterpfdpd(A, B, C, D, E) __builtin_ia32_scatterpfdpd(A, B, C, 1, _MM_HINT_T0)
368 #define __builtin_ia32_scatterpfqpd(A, B, C, D, E) __builtin_ia32_scatterpfqpd(A, B, C, 1, _MM_HINT_T0)
370 /* shaintrin.h */
371 #define __builtin_ia32_sha1rnds4(A, B, C) __builtin_ia32_sha1rnds4(A, B, 1)
373 /* TODO split */
374 #define __builtin_ia32_pslldq512(A, B) __builtin_ia32_pslldq512(A, 8)
375 #define __builtin_ia32_psrldq512(A, B) __builtin_ia32_psrldq512(A, 8)
376 #define __builtin_ia32_alignd128_mask(A, B, F, D, E) __builtin_ia32_alignd128_mask(A, B, 1, D, E)
377 #define __builtin_ia32_alignd256_mask(A, B, F, D, E) __builtin_ia32_alignd256_mask(A, B, 1, D, E)
378 #define __builtin_ia32_alignq128_mask(A, B, F, D, E) __builtin_ia32_alignq128_mask(A, B, 1, D, E)
379 #define __builtin_ia32_alignq256_mask(A, B, F, D, E) __builtin_ia32_alignq256_mask(A, B, 1, D, E)
380 #define __builtin_ia32_cmpb128_mask(A, B, E, D) __builtin_ia32_cmpb128_mask(A, B, 1, D)
381 #define __builtin_ia32_cmpb256_mask(A, B, E, D) __builtin_ia32_cmpb256_mask(A, B, 1, D)
382 #define __builtin_ia32_cmpb512_mask(A, B, E, D) __builtin_ia32_cmpb512_mask(A, B, 1, D)
383 #define __builtin_ia32_cmpd128_mask(A, B, E, D) __builtin_ia32_cmpd128_mask(A, B, 1, D)
384 #define __builtin_ia32_cmpd256_mask(A, B, E, D) __builtin_ia32_cmpd256_mask(A, B, 1, D)
385 #define __builtin_ia32_cmppd128_mask(A, B, E, D) __builtin_ia32_cmppd128_mask(A, B, 1, D)
386 #define __builtin_ia32_cmppd256_mask(A, B, E, D) __builtin_ia32_cmppd256_mask(A, B, 1, D)
387 #define __builtin_ia32_cmpps128_mask(A, B, E, D) __builtin_ia32_cmpps128_mask(A, B, 1, D)
388 #define __builtin_ia32_cmpps256_mask(A, B, E, D) __builtin_ia32_cmpps256_mask(A, B, 1, D)
389 #define __builtin_ia32_cmpq128_mask(A, B, E, D) __builtin_ia32_cmpq128_mask(A, B, 1, D)
390 #define __builtin_ia32_cmpq256_mask(A, B, E, D) __builtin_ia32_cmpq256_mask(A, B, 1, D)
391 #define __builtin_ia32_cmpw128_mask(A, B, E, D) __builtin_ia32_cmpw128_mask(A, B, 1, D)
392 #define __builtin_ia32_cmpw256_mask(A, B, E, D) __builtin_ia32_cmpw256_mask(A, B, 1, D)
393 #define __builtin_ia32_cmpw512_mask(A, B, E, D) __builtin_ia32_cmpw512_mask(A, B, 1, D)
394 #define __builtin_ia32_cvtpd2qq512_mask(A, B, C, D) __builtin_ia32_cvtpd2qq512_mask(A, B, C, 8)
395 #define __builtin_ia32_cvtpd2uqq512_mask(A, B, C, D) __builtin_ia32_cvtpd2uqq512_mask(A, B, C, 8)
396 #define __builtin_ia32_cvtps2qq512_mask(A, B, C, D) __builtin_ia32_cvtps2qq512_mask(A, B, C, 8)
397 #define __builtin_ia32_cvtps2uqq512_mask(A, B, C, D) __builtin_ia32_cvtps2uqq512_mask(A, B, C, 8)
398 #define __builtin_ia32_cvtqq2pd512_mask(A, B, C, D) __builtin_ia32_cvtqq2pd512_mask(A, B, C, 8)
399 #define __builtin_ia32_cvtqq2ps512_mask(A, B, C, D) __builtin_ia32_cvtqq2ps512_mask(A, B, C, 8)
400 #define __builtin_ia32_cvttpd2qq512_mask(A, B, C, D) __builtin_ia32_cvttpd2qq512_mask(A, B, C, 8)
401 #define __builtin_ia32_cvttpd2uqq512_mask(A, B, C, D) __builtin_ia32_cvttpd2uqq512_mask(A, B, C, 8)
402 #define __builtin_ia32_cvttps2qq512_mask(A, B, C, D) __builtin_ia32_cvttps2qq512_mask(A, B, C, 8)
403 #define __builtin_ia32_cvttps2uqq512_mask(A, B, C, D) __builtin_ia32_cvttps2uqq512_mask(A, B, C, 8)
404 #define __builtin_ia32_cvtuqq2pd512_mask(A, B, C, D) __builtin_ia32_cvtuqq2pd512_mask(A, B, C, 8)
405 #define __builtin_ia32_cvtuqq2ps512_mask(A, B, C, D) __builtin_ia32_cvtuqq2ps512_mask(A, B, C, 8)
406 #define __builtin_ia32_dbpsadbw128_mask(A, B, F, D, E) __builtin_ia32_dbpsadbw128_mask(A, B, 1, D, E)
407 #define __builtin_ia32_dbpsadbw256_mask(A, B, F, D, E) __builtin_ia32_dbpsadbw256_mask(A, B, 1, D, E)
408 #define __builtin_ia32_dbpsadbw512_mask(A, B, F, D, E) __builtin_ia32_dbpsadbw512_mask(A, B, 1, D, E)
409 #define __builtin_ia32_extractf32x4_256_mask(A, E, C, D) __builtin_ia32_extractf32x4_256_mask(A, 1, C, D)
410 #define __builtin_ia32_extractf32x8_mask(A, E, C, D) __builtin_ia32_extractf32x8_mask(A, 1, C, D)
411 #define __builtin_ia32_extractf64x2_256_mask(A, E, C, D) __builtin_ia32_extractf64x2_256_mask(A, 1, C, D)
412 #define __builtin_ia32_extractf64x2_512_mask(A, E, C, D) __builtin_ia32_extractf64x2_512_mask(A, 1, C, D)
413 #define __builtin_ia32_extracti32x4_256_mask(A, E, C, D) __builtin_ia32_extracti32x4_256_mask(A, 1, C, D)
414 #define __builtin_ia32_extracti32x8_mask(A, E, C, D) __builtin_ia32_extracti32x8_mask(A, 1, C, D)
415 #define __builtin_ia32_extracti64x2_256_mask(A, E, C, D) __builtin_ia32_extracti64x2_256_mask(A, 1, C, D)
416 #define __builtin_ia32_extracti64x2_512_mask(A, E, C, D) __builtin_ia32_extracti64x2_512_mask(A, 1, C, D)
417 #define __builtin_ia32_fixupimmpd128(A, B, C, E) __builtin_ia32_fixupimmpd128(A, B, C, 1)
418 #define __builtin_ia32_fixupimmpd128_mask(A, B, C, F, E) __builtin_ia32_fixupimmpd128_mask(A, B, C, 1, E)
419 #define __builtin_ia32_fixupimmpd128_maskz(A, B, C, F, E) __builtin_ia32_fixupimmpd128_maskz(A, B, C, 1, E)
420 #define __builtin_ia32_fixupimmpd256(A, B, C, E) __builtin_ia32_fixupimmpd256(A, B, C, 1)
421 #define __builtin_ia32_fixupimmpd256_mask(A, B, C, F, E) __builtin_ia32_fixupimmpd256_mask(A, B, C, 1, E)
422 #define __builtin_ia32_fixupimmpd256_maskz(A, B, C, F, E) __builtin_ia32_fixupimmpd256_maskz(A, B, C, 1, E)
423 #define __builtin_ia32_fixupimmps128(A, B, C, E) __builtin_ia32_fixupimmps128(A, B, C, 1)
424 #define __builtin_ia32_fixupimmps128_mask(A, B, C, F, E) __builtin_ia32_fixupimmps128_mask(A, B, C, 1, E)
425 #define __builtin_ia32_fixupimmps128_maskz(A, B, C, F, E) __builtin_ia32_fixupimmps128_maskz(A, B, C, 1, E)
426 #define __builtin_ia32_fixupimmps256(A, B, C, E) __builtin_ia32_fixupimmps256(A, B, C, 1)
427 #define __builtin_ia32_fixupimmps256_mask(A, B, C, F, E) __builtin_ia32_fixupimmps256_mask(A, B, C, 1, E)
428 #define __builtin_ia32_fixupimmps256_maskz(A, B, C, F, E) __builtin_ia32_fixupimmps256_maskz(A, B, C, 1, E)
429 #define __builtin_ia32_fpclasspd128_mask(A, D, C) __builtin_ia32_fpclasspd128_mask(A, 1, C)
430 #define __builtin_ia32_fpclasspd256_mask(A, D, C) __builtin_ia32_fpclasspd256_mask(A, 1, C)
431 #define __builtin_ia32_fpclasspd512_mask(A, D, C) __builtin_ia32_fpclasspd512_mask(A, 1, C)
432 #define __builtin_ia32_fpclassps128_mask(A, D, C) __builtin_ia32_fpclassps128_mask(A, 1, C)
433 #define __builtin_ia32_fpclassps256_mask(A, D, C) __builtin_ia32_fpclassps256_mask(A, 1, C)
434 #define __builtin_ia32_fpclassps512_mask(A, D, C) __builtin_ia32_fpclassps512_mask(A, 1, C)
435 #define __builtin_ia32_fpclasssd(A, D) __builtin_ia32_fpclasssd(A, 1)
436 #define __builtin_ia32_fpclassss(A, D) __builtin_ia32_fpclassss(A, 1)
437 #define __builtin_ia32_gather3div2df(A, B, C, D, F) __builtin_ia32_gather3div2df(A, B, C, D, 1)
438 #define __builtin_ia32_gather3div2di(A, B, C, D, F) __builtin_ia32_gather3div2di(A, B, C, D, 1)
439 #define __builtin_ia32_gather3div4df(A, B, C, D, F) __builtin_ia32_gather3div4df(A, B, C, D, 1)
440 #define __builtin_ia32_gather3div4di(A, B, C, D, F) __builtin_ia32_gather3div4di(A, B, C, D, 1)
441 #define __builtin_ia32_gather3div4sf(A, B, C, D, F) __builtin_ia32_gather3div4sf(A, B, C, D, 1)
442 #define __builtin_ia32_gather3div4si(A, B, C, D, F) __builtin_ia32_gather3div4si(A, B, C, D, 1)
443 #define __builtin_ia32_gather3div8sf(A, B, C, D, F) __builtin_ia32_gather3div8sf(A, B, C, D, 1)
444 #define __builtin_ia32_gather3div8si(A, B, C, D, F) __builtin_ia32_gather3div8si(A, B, C, D, 1)
445 #define __builtin_ia32_gather3siv2df(A, B, C, D, F) __builtin_ia32_gather3siv2df(A, B, C, D, 1)
446 #define __builtin_ia32_gather3siv2di(A, B, C, D, F) __builtin_ia32_gather3siv2di(A, B, C, D, 1)
447 #define __builtin_ia32_gather3siv4df(A, B, C, D, F) __builtin_ia32_gather3siv4df(A, B, C, D, 1)
448 #define __builtin_ia32_gather3siv4di(A, B, C, D, F) __builtin_ia32_gather3siv4di(A, B, C, D, 1)
449 #define __builtin_ia32_gather3siv4sf(A, B, C, D, F) __builtin_ia32_gather3siv4sf(A, B, C, D, 1)
450 #define __builtin_ia32_gather3siv4si(A, B, C, D, F) __builtin_ia32_gather3siv4si(A, B, C, D, 1)
451 #define __builtin_ia32_gather3siv8sf(A, B, C, D, F) __builtin_ia32_gather3siv8sf(A, B, C, D, 1)
452 #define __builtin_ia32_gather3siv8si(A, B, C, D, F) __builtin_ia32_gather3siv8si(A, B, C, D, 1)
453 #define __builtin_ia32_getmantpd128_mask(A, E, C, D) __builtin_ia32_getmantpd128_mask(A, 1, C, D)
454 #define __builtin_ia32_getmantpd256_mask(A, E, C, D) __builtin_ia32_getmantpd256_mask(A, 1, C, D)
455 #define __builtin_ia32_getmantps128_mask(A, E, C, D) __builtin_ia32_getmantps128_mask(A, 1, C, D)
456 #define __builtin_ia32_getmantps256_mask(A, E, C, D) __builtin_ia32_getmantps256_mask(A, 1, C, D)
457 #define __builtin_ia32_insertf32x4_256_mask(A, B, F, D, E) __builtin_ia32_insertf32x4_256_mask(A, B, 1, D, E)
458 #define __builtin_ia32_insertf32x8_mask(A, B, F, D, E) __builtin_ia32_insertf32x8_mask(A, B, 1, D, E)
459 #define __builtin_ia32_insertf64x2_256_mask(A, B, F, D, E) __builtin_ia32_insertf64x2_256_mask(A, B, 1, D, E)
460 #define __builtin_ia32_insertf64x2_512_mask(A, B, F, D, E) __builtin_ia32_insertf64x2_512_mask(A, B, 1, D, E)
461 #define __builtin_ia32_inserti32x4_256_mask(A, B, F, D, E) __builtin_ia32_inserti32x4_256_mask(A, B, 1, D, E)
462 #define __builtin_ia32_inserti32x8_mask(A, B, F, D, E) __builtin_ia32_inserti32x8_mask(A, B, 1, D, E)
463 #define __builtin_ia32_inserti64x2_256_mask(A, B, F, D, E) __builtin_ia32_inserti64x2_256_mask(A, B, 1, D, E)
464 #define __builtin_ia32_inserti64x2_512_mask(A, B, F, D, E) __builtin_ia32_inserti64x2_512_mask(A, B, 1, D, E)
465 #define __builtin_ia32_palignr128_mask(A, B, F, D, E) __builtin_ia32_palignr128_mask(A, B, 8, D, E)
466 #define __builtin_ia32_palignr256_mask(A, B, F, D, E) __builtin_ia32_palignr256_mask(A, B, 8, D, E)
467 #define __builtin_ia32_palignr512(A, B, D) __builtin_ia32_palignr512(A, B, 8)
468 #define __builtin_ia32_palignr512_mask(A, B, F, D, E) __builtin_ia32_palignr512_mask(A, B, 8, D, E)
469 #define __builtin_ia32_permdf256_mask(A, E, C, D) __builtin_ia32_permdf256_mask(A, 1, C, D)
470 #define __builtin_ia32_permdi256_mask(A, E, C, D) __builtin_ia32_permdi256_mask(A, 1, C, D)
471 #define __builtin_ia32_prold128_mask(A, E, C, D) __builtin_ia32_prold128_mask(A, 1, C, D)
472 #define __builtin_ia32_prold256_mask(A, E, C, D) __builtin_ia32_prold256_mask(A, 1, C, D)
473 #define __builtin_ia32_prolq128_mask(A, E, C, D) __builtin_ia32_prolq128_mask(A, 1, C, D)
474 #define __builtin_ia32_prolq256_mask(A, E, C, D) __builtin_ia32_prolq256_mask(A, 1, C, D)
475 #define __builtin_ia32_prord128_mask(A, E, C, D) __builtin_ia32_prord128_mask(A, 1, C, D)
476 #define __builtin_ia32_prord256_mask(A, E, C, D) __builtin_ia32_prord256_mask(A, 1, C, D)
477 #define __builtin_ia32_prorq128_mask(A, E, C, D) __builtin_ia32_prorq128_mask(A, 1, C, D)
478 #define __builtin_ia32_prorq256_mask(A, E, C, D) __builtin_ia32_prorq256_mask(A, 1, C, D)
479 #define __builtin_ia32_pshufd128_mask(A, E, C, D) __builtin_ia32_pshufd128_mask(A, 1, C, D)
480 #define __builtin_ia32_pshufd256_mask(A, E, C, D) __builtin_ia32_pshufd256_mask(A, 1, C, D)
481 #define __builtin_ia32_pshufhw128_mask(A, E, C, D) __builtin_ia32_pshufhw128_mask(A, 1, C, D)
482 #define __builtin_ia32_pshufhw256_mask(A, E, C, D) __builtin_ia32_pshufhw256_mask(A, 1, C, D)
483 #define __builtin_ia32_pshufhw512_mask(A, E, C, D) __builtin_ia32_pshufhw512_mask(A, 1, C, D)
484 #define __builtin_ia32_pshuflw128_mask(A, E, C, D) __builtin_ia32_pshuflw128_mask(A, 1, C, D)
485 #define __builtin_ia32_pshuflw256_mask(A, E, C, D) __builtin_ia32_pshuflw256_mask(A, 1, C, D)
486 #define __builtin_ia32_pshuflw512_mask(A, E, C, D) __builtin_ia32_pshuflw512_mask(A, 1, C, D)
487 #define __builtin_ia32_pslldi128_mask(A, E, C, D) __builtin_ia32_pslldi128_mask(A, 1, C, D)
488 #define __builtin_ia32_pslldi256_mask(A, E, C, D) __builtin_ia32_pslldi256_mask(A, 1, C, D)
489 #define __builtin_ia32_psllqi128_mask(A, E, C, D) __builtin_ia32_psllqi128_mask(A, 1, C, D)
490 #define __builtin_ia32_psllqi256_mask(A, E, C, D) __builtin_ia32_psllqi256_mask(A, 1, C, D)
491 #define __builtin_ia32_psllwi128_mask(A, E, C, D) __builtin_ia32_psllwi128_mask(A, 1, C, D)
492 #define __builtin_ia32_psllwi256_mask(A, E, C, D) __builtin_ia32_psllwi256_mask(A, 1, C, D)
493 #define __builtin_ia32_psllwi512_mask(A, E, C, D) __builtin_ia32_psllwi512_mask(A, 1, C, D)
494 #define __builtin_ia32_psradi128_mask(A, E, C, D) __builtin_ia32_psradi128_mask(A, 1, C, D)
495 #define __builtin_ia32_psradi256_mask(A, E, C, D) __builtin_ia32_psradi256_mask(A, 1, C, D)
496 #define __builtin_ia32_psraqi128_mask(A, E, C, D) __builtin_ia32_psraqi128_mask(A, 1, C, D)
497 #define __builtin_ia32_psraqi256_mask(A, E, C, D) __builtin_ia32_psraqi256_mask(A, 1, C, D)
498 #define __builtin_ia32_psrawi128_mask(A, E, C, D) __builtin_ia32_psrawi128_mask(A, 1, C, D)
499 #define __builtin_ia32_psrawi256_mask(A, E, C, D) __builtin_ia32_psrawi256_mask(A, 1, C, D)
500 #define __builtin_ia32_psrawi512_mask(A, E, C, D) __builtin_ia32_psrawi512_mask(A, 1, C, D)
501 #define __builtin_ia32_psrldi128_mask(A, E, C, D) __builtin_ia32_psrldi128_mask(A, 1, C, D)
502 #define __builtin_ia32_psrldi256_mask(A, E, C, D) __builtin_ia32_psrldi256_mask(A, 1, C, D)
503 #define __builtin_ia32_psrlqi128_mask(A, E, C, D) __builtin_ia32_psrlqi128_mask(A, 1, C, D)
504 #define __builtin_ia32_psrlqi256_mask(A, E, C, D) __builtin_ia32_psrlqi256_mask(A, 1, C, D)
505 #define __builtin_ia32_psrlwi128_mask(A, E, C, D) __builtin_ia32_psrlwi128_mask(A, 1, C, D)
506 #define __builtin_ia32_psrlwi256_mask(A, E, C, D) __builtin_ia32_psrlwi256_mask(A, 1, C, D)
507 #define __builtin_ia32_psrlwi512_mask(A, E, C, D) __builtin_ia32_psrlwi512_mask(A, 1, C, D)
508 #define __builtin_ia32_pternlogd128_mask(A, B, C, F, E) __builtin_ia32_pternlogd128_mask(A, B, C, 1, E)
509 #define __builtin_ia32_pternlogd128_maskz(A, B, C, F, E) __builtin_ia32_pternlogd128_maskz(A, B, C, 1, E)
510 #define __builtin_ia32_pternlogd256_mask(A, B, C, F, E) __builtin_ia32_pternlogd256_mask(A, B, C, 1, E)
511 #define __builtin_ia32_pternlogd256_maskz(A, B, C, F, E) __builtin_ia32_pternlogd256_maskz(A, B, C, 1, E)
512 #define __builtin_ia32_pternlogq128_mask(A, B, C, F, E) __builtin_ia32_pternlogq128_mask(A, B, C, 1, E)
513 #define __builtin_ia32_pternlogq128_maskz(A, B, C, F, E) __builtin_ia32_pternlogq128_maskz(A, B, C, 1, E)
514 #define __builtin_ia32_pternlogq256_mask(A, B, C, F, E) __builtin_ia32_pternlogq256_mask(A, B, C, 1, E)
515 #define __builtin_ia32_pternlogq256_maskz(A, B, C, F, E) __builtin_ia32_pternlogq256_maskz(A, B, C, 1, E)
516 #define __builtin_ia32_rangepd128_mask(A, B, F, D, E) __builtin_ia32_rangepd128_mask(A, B, 1, D, E)
517 #define __builtin_ia32_rangepd256_mask(A, B, F, D, E) __builtin_ia32_rangepd256_mask(A, B, 1, D, E)
518 #define __builtin_ia32_rangepd512_mask(A, B, I, D, E, F) __builtin_ia32_rangepd512_mask(A, B, 1, D, E, 8)
519 #define __builtin_ia32_rangeps128_mask(A, B, F, D, E) __builtin_ia32_rangeps128_mask(A, B, 1, D, E)
520 #define __builtin_ia32_rangeps256_mask(A, B, F, D, E) __builtin_ia32_rangeps256_mask(A, B, 1, D, E)
521 #define __builtin_ia32_rangeps512_mask(A, B, I, D, E, F) __builtin_ia32_rangeps512_mask(A, B, 1, D, E, 8)
522 #define __builtin_ia32_rangesd128_round(A, B, I, F) __builtin_ia32_rangesd128_round(A, B, 1, 8)
523 #define __builtin_ia32_rangess128_round(A, B, I, F) __builtin_ia32_rangess128_round(A, B, 1, 8)
524 #define __builtin_ia32_reducepd128_mask(A, E, C, D) __builtin_ia32_reducepd128_mask(A, 1, C, D)
525 #define __builtin_ia32_reducepd256_mask(A, E, C, D) __builtin_ia32_reducepd256_mask(A, 1, C, D)
526 #define __builtin_ia32_reducepd512_mask(A, E, C, D) __builtin_ia32_reducepd512_mask(A, 1, C, D)
527 #define __builtin_ia32_reduceps128_mask(A, E, C, D) __builtin_ia32_reduceps128_mask(A, 1, C, D)
528 #define __builtin_ia32_reduceps256_mask(A, E, C, D) __builtin_ia32_reduceps256_mask(A, 1, C, D)
529 #define __builtin_ia32_reduceps512_mask(A, E, C, D) __builtin_ia32_reduceps512_mask(A, 1, C, D)
530 #define __builtin_ia32_reducesd(A, B, F) __builtin_ia32_reducesd(A, B, 1)
531 #define __builtin_ia32_reducess(A, B, F) __builtin_ia32_reducess(A, B, 1)
532 #define __builtin_ia32_rndscalepd_128_mask(A, E, C, D) __builtin_ia32_rndscalepd_128_mask(A, 1, C, D)
533 #define __builtin_ia32_rndscalepd_256_mask(A, E, C, D) __builtin_ia32_rndscalepd_256_mask(A, 1, C, D)
534 #define __builtin_ia32_rndscaleps_128_mask(A, E, C, D) __builtin_ia32_rndscaleps_128_mask(A, 1, C, D)
535 #define __builtin_ia32_rndscaleps_256_mask(A, E, C, D) __builtin_ia32_rndscaleps_256_mask(A, 1, C, D)
536 #define __builtin_ia32_scatterdiv2df(A, B, C, D, F) __builtin_ia32_scatterdiv2df(A, B, C, D, 1)
537 #define __builtin_ia32_scatterdiv2di(A, B, C, D, F) __builtin_ia32_scatterdiv2di(A, B, C, D, 1)
538 #define __builtin_ia32_scatterdiv4df(A, B, C, D, F) __builtin_ia32_scatterdiv4df(A, B, C, D, 1)
539 #define __builtin_ia32_scatterdiv4di(A, B, C, D, F) __builtin_ia32_scatterdiv4di(A, B, C, D, 1)
540 #define __builtin_ia32_scatterdiv4sf(A, B, C, D, F) __builtin_ia32_scatterdiv4sf(A, B, C, D, 1)
541 #define __builtin_ia32_scatterdiv4si(A, B, C, D, F) __builtin_ia32_scatterdiv4si(A, B, C, D, 1)
542 #define __builtin_ia32_scatterdiv8sf(A, B, C, D, F) __builtin_ia32_scatterdiv8sf(A, B, C, D, 1)
543 #define __builtin_ia32_scatterdiv8si(A, B, C, D, F) __builtin_ia32_scatterdiv8si(A, B, C, D, 1)
544 #define __builtin_ia32_scattersiv2df(A, B, C, D, F) __builtin_ia32_scattersiv2df(A, B, C, D, 1)
545 #define __builtin_ia32_scattersiv2di(A, B, C, D, F) __builtin_ia32_scattersiv2di(A, B, C, D, 1)
546 #define __builtin_ia32_scattersiv4df(A, B, C, D, F) __builtin_ia32_scattersiv4df(A, B, C, D, 1)
547 #define __builtin_ia32_scattersiv4di(A, B, C, D, F) __builtin_ia32_scattersiv4di(A, B, C, D, 1)
548 #define __builtin_ia32_scattersiv4sf(A, B, C, D, F) __builtin_ia32_scattersiv4sf(A, B, C, D, 1)
549 #define __builtin_ia32_scattersiv4si(A, B, C, D, F) __builtin_ia32_scattersiv4si(A, B, C, D, 1)
550 #define __builtin_ia32_scattersiv8sf(A, B, C, D, F) __builtin_ia32_scattersiv8sf(A, B, C, D, 1)
551 #define __builtin_ia32_scattersiv8si(A, B, C, D, F) __builtin_ia32_scattersiv8si(A, B, C, D, 1)
552 #define __builtin_ia32_shuf_f32x4_256_mask(A, B, F, D, E) __builtin_ia32_shuf_f32x4_256_mask(A, B, 1, D, E)
553 #define __builtin_ia32_shuf_f64x2_256_mask(A, B, F, D, E) __builtin_ia32_shuf_f64x2_256_mask(A, B, 1, D, E)
554 #define __builtin_ia32_shuf_i32x4_256_mask(A, B, F, D, E) __builtin_ia32_shuf_i32x4_256_mask(A, B, 1, D, E)
555 #define __builtin_ia32_shuf_i64x2_256_mask(A, B, F, D, E) __builtin_ia32_shuf_i64x2_256_mask(A, B, 1, D, E)
556 #define __builtin_ia32_shufpd128_mask(A, B, F, D, E) __builtin_ia32_shufpd128_mask(A, B, 1, D, E)
557 #define __builtin_ia32_shufpd256_mask(A, B, F, D, E) __builtin_ia32_shufpd256_mask(A, B, 1, D, E)
558 #define __builtin_ia32_shufps128_mask(A, B, F, D, E) __builtin_ia32_shufps128_mask(A, B, 1, D, E)
559 #define __builtin_ia32_shufps256_mask(A, B, F, D, E) __builtin_ia32_shufps256_mask(A, B, 1, D, E)
560 #define __builtin_ia32_ucmpb128_mask(A, B, E, D) __builtin_ia32_ucmpb128_mask(A, B, 1, D)
561 #define __builtin_ia32_ucmpb256_mask(A, B, E, D) __builtin_ia32_ucmpb256_mask(A, B, 1, D)
562 #define __builtin_ia32_ucmpb512_mask(A, B, E, D) __builtin_ia32_ucmpb512_mask(A, B, 1, D)
563 #define __builtin_ia32_ucmpd128_mask(A, B, E, D) __builtin_ia32_ucmpd128_mask(A, B, 1, D)
564 #define __builtin_ia32_ucmpd256_mask(A, B, E, D) __builtin_ia32_ucmpd256_mask(A, B, 1, D)
565 #define __builtin_ia32_ucmpq128_mask(A, B, E, D) __builtin_ia32_ucmpq128_mask(A, B, 1, D)
566 #define __builtin_ia32_ucmpq256_mask(A, B, E, D) __builtin_ia32_ucmpq256_mask(A, B, 1, D)
567 #define __builtin_ia32_ucmpw128_mask(A, B, E, D) __builtin_ia32_ucmpw128_mask(A, B, 1, D)
568 #define __builtin_ia32_ucmpw256_mask(A, B, E, D) __builtin_ia32_ucmpw256_mask(A, B, 1, D)
569 #define __builtin_ia32_ucmpw512_mask(A, B, E, D) __builtin_ia32_ucmpw512_mask(A, B, 1, D)
570 #define __builtin_ia32_vcvtps2ph256_mask(A, E, C, D) __builtin_ia32_vcvtps2ph256_mask(A, 1, C, D)
571 #define __builtin_ia32_vcvtps2ph_mask(A, E, C, D) __builtin_ia32_vcvtps2ph_mask(A, 1, C, D)
572 #define __builtin_ia32_vpermilpd256_mask(A, E, C, D) __builtin_ia32_vpermilpd256_mask(A, 1, C, D)
573 #define __builtin_ia32_vpermilpd_mask(A, E, C, D) __builtin_ia32_vpermilpd_mask(A, 1, C, D)
574 #define __builtin_ia32_vpermilps256_mask(A, E, C, D) __builtin_ia32_vpermilps256_mask(A, 1, C, D)
575 #define __builtin_ia32_vpermilps_mask(A, E, C, D) __builtin_ia32_vpermilps_mask(A, 1, C, D)
577 #include <wmmintrin.h>
578 #include <immintrin.h>
579 #include <mm3dnow.h>