2 # This file is part of the GROMACS molecular simulation package.
4 # Copyright (c) 2017,2018,2019,2020, by the GROMACS development team, led by
5 # Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 # and including many others, as listed in the AUTHORS file in the
7 # top-level source directory and at http://www.gromacs.org.
9 # GROMACS is free software; you can redistribute it and/or
10 # modify it under the terms of the GNU Lesser General Public License
11 # as published by the Free Software Foundation; either version 2.1
12 # of the License, or (at your option) any later version.
14 # GROMACS is distributed in the hope that it will be useful,
15 # but WITHOUT ANY WARRANTY; without even the implied warranty of
16 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 # Lesser General Public License for more details.
19 # You should have received a copy of the GNU Lesser General Public
20 # License along with GROMACS; if not, see
21 # http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 # Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 # If you want to redistribute modifications to GROMACS, please
25 # consider that scientific software is very special. Version
26 # control is crucial - bugs must be traceable. We will be happy to
27 # consider code for inclusion in the official distribution, but
28 # derived work must not be called official GROMACS. Details are found
29 # in the README & COPYING files - if they are missing, get the
30 # official version at http://www.gromacs.org.
32 # To help us fund GROMACS development, we humbly ask that you cite
33 # the research papers on the package. Check out http://www.gromacs.org.
35 include(gmxFindFlagsForSource)
37 # Macro that manages setting the respective C and C++ toolchain
38 # variables so that subsequent tests for SIMD support can work.
39 macro(find_x86_toolchain_flags TOOLCHAIN_C_FLAGS_VARIABLE TOOLCHAIN_CXX_FLAGS_VARIABLE)
40 # On OS X, we often want to use gcc instead of clang, since gcc
41 # supports OpenMP (until clang 3.8, or so, plus whenever Apple
42 # support it in their version). However, by default gcc uses the
43 # external system assembler, which does not support AVX, so we
44 # need to tell the linker to use the clang compilers assembler
45 # instead - and this has to happen before we detect AVX flags.
46 if(APPLE AND CMAKE_C_COMPILER_ID STREQUAL "GNU")
47 gmx_test_cflag(GNU_C_USE_CLANG_AS "-Wa,-q" ${TOOLCHAIN_C_FLAGS_VARIABLE})
49 if(APPLE AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
50 gmx_test_cxxflag(GNU_CXX_USE_CLANG_AS "-Wa,-q" ${TOOLCHAIN_CXX_FLAGS_VARIABLE})
54 # Macro that manages setting the respective C and C++ toolchain
55 # variables so that subsequent tests for SIMD support can work.
56 macro(find_power_vsx_toolchain_flags TOOLCHAIN_C_FLAGS_VARIABLE TOOLCHAIN_CXX_FLAGS_VARIABLE)
57 if(${CMAKE_CXX_COMPILER_ID} MATCHES "GNU" OR ${CMAKE_C_COMPILER_ID} MATCHES "GNU")
58 # VSX uses the same function API as Altivec/VMX, so make sure we tune for the current CPU and not VMX.
59 # By putting these flags here rather than in the general compiler flags file we can safely assume
60 # that we are at least on Power7 since that is when VSX appeared.
62 # NOTE: Enabling instruction fusion on Power8/9 using -mpower8-fusion/-mpower9-fusion
63 # seems to produce buggy code (see #2747, #2746, #2734).
64 # Note that instruction fusion does have considerable performance benefits
65 # (up to 8% measured with gcc 8) if the issue is resolved the flag can be re-enabled.
66 gmx_run_cpu_detection(brand)
67 if(CPU_DETECTION_BRAND MATCHES "POWER7")
68 gmx_test_cflag(GNU_C_VSX_POWER7 "-mcpu=power7 -mtune=power7" ${TOOLCHAIN_C_FLAGS_VARIABLE})
69 gmx_test_cflag(GNU_CXX_VSX_POWER7 "-mcpu=power7 -mtune=power7" ${TOOLCHAIN_CXX_FLAGS_VARIABLE})
70 elseif(CPU_DETECTION_BRAND MATCHES "POWER8")
71 # Enable power8 vector extensions on such platforms.
72 gmx_test_cflag(GNU_C_VSX_POWER8 "-mcpu=power8 -mpower8-vector" ${TOOLCHAIN_C_FLAGS_VARIABLE})
73 gmx_test_cflag(GNU_CXX_VSX_POWER8 "-mcpu=power8 -mpower8-vector" ${TOOLCHAIN_CXX_FLAGS_VARIABLE})
74 elseif(CPU_DETECTION_BRAND MATCHES "POWER9")
75 # Enable power9 vector extensions on such platforms.
76 # TODO consider whether adding " -mpower9-vector -mpower9-fusion"
78 gmx_test_cflag(GNU_C_VSX_POWER9 "-mcpu=power9 -mtune=power9" ${TOOLCHAIN_C_FLAGS_VARIABLE})
79 gmx_test_cflag(GNU_CXX_VSX_POWER9 "-mcpu=power9 -mtune=power9" ${TOOLCHAIN_CXX_FLAGS_VARIABLE})
81 # Don't add arch-specific flags for unknown architectures.
84 if(${CMAKE_CXX_COMPILER_ID} MATCHES "XL" OR ${CMAKE_C_COMPILER_ID} MATCHES "XL")
85 if(CMAKE_CXX_COMPILER_VERSION VERSION_LESS "13.1.5" OR CMAKE_C_COMPILER_VERSION VERSION_LESS "13.1.5")
86 message(FATAL_ERROR "Using VSX SIMD requires XL compiler version 13.1.5 or later.")
92 function(gmx_find_simd_sse2_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
93 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
94 gmx_find_flags(SIMD_SSE2_C_FLAGS_RESULT SIMD_SSE2_CXX_FLAGS_RESULT
95 "#include<xmmintrin.h>
96 int main(){__m128 x=_mm_set1_ps(0.5);x=_mm_rsqrt_ps(x);return _mm_movemask_ps(x);}"
97 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
98 SIMD_SSE2_C_FLAGS SIMD_SSE2_CXX_FLAGS
99 "-msse2" "/arch:SSE2" "-hgnu")
101 if(${SIMD_SSE2_C_FLAGS_RESULT})
102 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_SSE2_C_FLAGS}" CACHE INTERNAL "C flags required for SSE2 instructions")
104 if(${SIMD_SSE2_CXX_FLAGS_RESULT})
105 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_SSE2_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for SSE2 instructions")
107 set(${C_FLAGS_RESULT} ${SIMD_SSE2_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for SSE2 C flags" FORCE)
108 set(${CXX_FLAGS_RESULT} ${SIMD_SSE2_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for SSE2 C++ flags" FORCE)
112 function(gmx_find_simd_sse4_1_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
113 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
114 # Note: MSVC enables SSE4.1 with the SSE2 flag, so we include that in testing.
115 gmx_find_flags(SIMD_SSE4_1_C_FLAGS_RESULT SIMD_SSE4_1_CXX_FLAGS_RESULT
116 "#include<smmintrin.h>
117 int main(){__m128 x=_mm_set1_ps(0.5);x=_mm_dp_ps(x,x,0x77);return _mm_movemask_ps(x);}"
118 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
119 SIMD_SSE4_1_C_FLAGS SIMD_SSE4_1_CXX_FLAGS
120 "-msse4.1" "/arch:SSE4.1" "/arch:SSE2" "-hgnu")
122 if(${SIMD_SSE4_1_C_FLAGS_RESULT})
123 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_SSE4_1_C_FLAGS}" CACHE INTERNAL "C flags required for SSE4.1 instructions")
125 if(${SIMD_SSE4_1_CXX_FLAGS_RESULT})
126 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_SSE4_1_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for SSE4.1 instructions")
128 set(${C_FLAGS_RESULT} ${SIMD_SSE4_1_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for SSE4.1 C flags" FORCE)
129 set(${CXX_FLAGS_RESULT} ${SIMD_SSE4_1_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for SSE4.1 C++ flags" FORCE)
132 # AVX, but using only 128-bit instructions and FMA (AMD XOP processors)
133 function(gmx_find_simd_avx_128_fma_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
134 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
136 # AVX128/FMA on AMD is a bit complicated. We need to do detection in three stages:
137 # 1) Find the flags required for generic AVX support
138 # 2) Find the flags necessary to enable fused-multiply add support
139 # 3) Optional: Find a flag to enable the AMD XOP instructions
141 ### STAGE 1: Find the generic AVX flag, but stick to 128-bit instructions
142 gmx_find_flags(SIMD_AVX_128_FMA_C_FLAGS_RESULT SIMD_AVX_128_FMA_CXX_FLAGS_RESULT
143 "#include<immintrin.h>
144 int main(){__m128 x=_mm_set1_ps(0.5);x=_mm_permute_ps(x,1);return 0;}"
145 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
146 SIMD_AVX_GENERIC_C_FLAGS SIMD_AVX_GENERIC_CXX_FLAGS
147 "-mavx" "/arch:AVX" "-hgnu")
149 if(SIMD_AVX_128_FMA_C_FLAGS_RESULT AND SIMD_AVX_128_FMA_CXX_FLAGS_RESULT)
150 set(MERGED_C_FLAGS "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_GENERIC_C_FLAGS}")
151 set(MERGED_CXX_FLAGS "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_GENERIC_CXX_FLAGS}")
153 ### STAGE 2: Find the fused-multiply add flag.
154 # GCC requires x86intrin.h for FMA support. MSVC 2010 requires intrin.h for FMA support.
155 check_include_file(x86intrin.h HAVE_X86INTRIN_H ${SIMD_C_FLAGS})
156 check_include_file(intrin.h HAVE_INTRIN_H ${SIMD_C_FLAGS})
158 set(INCLUDE_X86INTRIN_H "#include <x86intrin.h>")
161 set(INCLUDE_INTRIN_H "#include <xintrin.h>")
164 gmx_find_flags(SIMD_AVX_128_FMA_C_FLAGS_RESULT SIMD_AVX_128_FMA_CXX_FLAGS_RESULT
165 "#include<immintrin.h>
166 ${INCLUDE_X86INTRIN_H}
168 int main(){__m128 x=_mm_set1_ps(0.5);x=_mm_macc_ps(x,x,x);return _mm_movemask_ps(x);}"
169 MERGED_C_FLAGS MERGED_CXX_FLAGS
170 SIMD_AVX_AMD_FMA_C_FLAGS SIMD_AVX_AMD_FMA_CXX_FLAGS
173 if(SIMD_AVX_128_FMA_C_FLAGS_RESULT AND SIMD_AVX_128_FMA_CXX_FLAGS_RESULT)
174 set(MERGED_C_FLAGS "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_AMD_FMA_C_FLAGS}")
175 set(MERGED_CXX_FLAGS "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_AMD_FMA_CXX_FLAGS}")
176 ### STAGE 3: Find the XOP instruction flag. This is optional.
177 gmx_find_flags(SIMD_AVX_XOP_C_FLAGS_RESULT SIMD_AVX_XOP_CXX_FLAGS_RESULT
178 "#include<immintrin.h>
179 ${INCLUDE_X86INTRIN_H}
181 int main(){__m128 x=_mm_set1_ps(0.5);x=_mm_frcz_ps(x);return _mm_movemask_ps(x);}"
182 MERGED_C_FLAGS MERGED_CXX_FLAGS
183 SIMD_AVX_XOP_C_FLAGS SIMD_AVX_XOP_CXX_FLAGS
188 if(${SIMD_AVX_128_FMA_C_FLAGS_RESULT})
189 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_GENERIC_C_FLAGS} ${SIMD_AVX_AMD_FMA_C_FLAGS} ${SIMD_AVX_XOP_C_FLAGS}" CACHE INTERNAL "C flags required for 128-bit AVX with AMD FMA instructions")
191 if(${SIMD_AVX_128_FMA_CXX_FLAGS_RESULT})
192 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_GENERIC_CXX_FLAGS} ${SIMD_AVX_AMD_FMA_CXX_FLAGS} ${SIMD_AVX_XOP_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for 128-bit AVX with AMD FMA instructions")
194 set(${C_FLAGS_RESULT} ${SIMD_AVX_128_FMA_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for 128-bit AVX with AMD FMA C flags" FORCE)
195 set(${CXX_FLAGS_RESULT} ${SIMD_AVX_128_FMA_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for 128-bit AVX with AMD FMA C++ flags" FORCE)
199 # AVX (no AMD extensions)
200 function(gmx_find_simd_avx_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
201 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
202 gmx_find_flags(SIMD_AVX_C_FLAGS_RESULT SIMD_AVX_CXX_FLAGS_RESULT
203 "#include<immintrin.h>
204 int main(){__m256 x=_mm256_set1_ps(0.5);x=_mm256_add_ps(x,x);return _mm256_movemask_ps(x);}"
205 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
206 SIMD_AVX_C_FLAGS SIMD_AVX_CXX_FLAGS
207 "-mavx" "/arch:AVX" "-hgnu")
209 if(${SIMD_AVX_C_FLAGS_RESULT})
210 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_C_FLAGS}" CACHE INTERNAL "C flags required for AVX instructions")
212 if(${SIMD_AVX_CXX_FLAGS_RESULT})
213 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for AVX instructions")
215 set(${C_FLAGS_RESULT} ${SIMD_AVX_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX C flags" FORCE)
216 set(${CXX_FLAGS_RESULT} ${SIMD_AVX_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX C++ flags" FORCE)
220 function(gmx_find_simd_avx2_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
221 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
222 # For our "AVX2_256" support we would ideally want to enable the instructions
223 # we want to use, '-mavx2 -mfma'. icc (v16-18) does not allow doing that and
224 # instead, it requires the '-march=core-avx2' flag to be used. Annoyingly, it does
225 # however accept the former flags but it is not silent about it issuing warnings
226 # that can't be disabled.
227 # At the same time Intel's -march=core-avx2 flag is not rejected by gcc/clang either
228 # (though they're at least silent). However, -march=core-avx2 is an undocumented
229 # flag with unclear behavior in gcc/clang (and might enable some arch-specific optimizations).
230 # For this reason, and because we can't distinguish compilers just based on checking flag
231 # compatibility, we need to treat the Intel and gcc/clang separately.
232 if (CMAKE_C_COMPILER_ID MATCHES "Intel")
233 set(TOOLCHAIN_FLAG_FOR_AVX2 "-march=core-avx2")
235 set(TOOLCHAIN_FLAG_FOR_AVX2 "-mavx2 -mfma")
237 gmx_find_flags(SIMD_AVX2_C_FLAGS_RESULT SIMD_AVX2_CXX_FLAGS_RESULT
238 "#include<immintrin.h>
239 int main(){__m256i x=_mm256_set1_epi32(5);x=_mm256_add_epi32(x,x);return _mm256_movemask_epi8(x);}"
240 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
241 SIMD_AVX2_C_FLAGS SIMD_AVX2_CXX_FLAGS
242 "${TOOLCHAIN_FLAG_FOR_AVX2}" "-mavx2" "/arch:AVX2" "-hgnu")
244 if(${SIMD_AVX2_C_FLAGS_RESULT})
245 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX2_C_FLAGS}" CACHE INTERNAL "C flags required for AVX2 instructions")
247 if(${SIMD_AVX2_CXX_FLAGS_RESULT})
248 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX2_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for AVX2 instructions")
250 set(${C_FLAGS_RESULT} ${SIMD_AVX2_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX2 C flags" FORCE)
251 set(${CXX_FLAGS_RESULT} ${SIMD_AVX2_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX2 C++ flags" FORCE)
255 # AVX-512F (Skylake-X)
256 function(gmx_find_simd_avx_512_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
257 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
259 gmx_find_flags(SIMD_AVX_512_C_FLAGS_RESULT SIMD_AVX_512_CXX_FLAGS_RESULT
260 "#include<immintrin.h>
261 int main(){__m512 x=_mm512_set1_ps(0.5); __m512 y=_mm512_fmadd_ps(x,x,x);
262 __m512i i = _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
263 __mmask16 mask = (short)(0xffff);
264 int idata[16]; i = _mm512_maskz_permutexvar_epi32(mask, i, i);
265 _mm512_storeu_si512(idata, i);
266 return idata[0]*(int)(_mm512_cmp_ps_mask(x,y,_CMP_LT_OS));}"
267 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
268 SIMD_AVX_512_C_FLAGS SIMD_AVX_512_CXX_FLAGS
269 "-xCORE-AVX512 -qopt-zmm-usage=high" "-xCORE-AVX512" "-mavx512f -mfma" "-mavx512f" "/arch:AVX512" "-hgnu") #ICC should use ZMM if code anyhow uses ZMM
271 if(${SIMD_AVX_512_C_FLAGS_RESULT})
272 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_512_C_FLAGS}" CACHE INTERNAL "C flags required for AVX-512 instructions")
274 if(${SIMD_AVX_512_CXX_FLAGS_RESULT})
275 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_512_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for AVX-512 instructions")
277 set(${C_FLAGS_RESULT} ${SIMD_AVX_512_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX-512 C flags" FORCE)
278 set(${CXX_FLAGS_RESULT} ${SIMD_AVX_512_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX-512 C++ flags" FORCE)
283 function(gmx_find_simd_avx_512_knl_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
284 find_x86_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
286 gmx_find_flags(SIMD_AVX_512_KNL_C_FLAGS_RESULT SIMD_AVX_512_KNL_CXX_FLAGS_RESULT
287 "#include<immintrin.h>
288 int main(){__m512 y,x=_mm512_set1_ps(0.5);y=_mm512_rsqrt28_ps(x);return (int)_mm512_cmp_ps_mask(x,y,_CMP_LT_OS);}"
289 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
290 SIMD_AVX_512_KNL_C_FLAGS SIMD_AVX_512_KNL_CXX_FLAGS
291 "-xMIC-AVX512" "-mavx512er -mfma" "-mavx512er" "/arch:AVX" "-hgnu") # no AVX_512ER flags known for MSVC yet
293 if(${SIMD_AVX_512_KNL_C_FLAGS_RESULT})
294 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_AVX_512_KNL_C_FLAGS}" CACHE INTERNAL "C flags required for AVX-512 for KNL instructions")
296 if(${SIMD_AVX_512_KNL_CXX_FLAGS_RESULT})
297 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_AVX_512_KNL_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for AVX-512 for KNL instructions")
299 set(${C_FLAGS_RESULT} ${SIMD_AVX_512_KNL_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX-512 for KNL C flags" FORCE)
300 set(${CXX_FLAGS_RESULT} ${SIMD_AVX_512_KNL_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for AVX-512 for KNL C++ flags" FORCE)
304 # Arm Neon (32-bit ARM)
305 function(gmx_find_simd_arm_neon_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
307 gmx_find_flags(SIMD_ARM_NEON_C_FLAGS_RESULT SIMD_ARM_NEON_CXX_FLAGS_RESULT
308 "#include<arm_neon.h>
309 int main(){float32x4_t x=vdupq_n_f32(0.5);x=vmlaq_f32(x,x,x);return vgetq_lane_f32(x,0)>0;}"
310 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
311 SIMD_ARM_NEON_C_FLAGS SIMD_ARM_NEON_CXX_FLAGS
312 "-mfpu=neon-vfpv4" "-mfpu=neon" "")
314 if(${SIMD_ARM_NEON_C_FLAGS_RESULT})
315 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_ARM_NEON_C_FLAGS}" CACHE INTERNAL "C flags required for Arm Neon instructions")
317 if(${SIMD_ARM_NEON_CXX_FLAGS_RESULT})
318 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_ARM_NEON_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for Arm Neon instructions")
320 set(${C_FLAGS_RESULT} ${SIMD_ARM_NEON_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for Arm Neon C flags" FORCE)
321 set(${CXX_FLAGS_RESULT} ${SIMD_ARM_NEON_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for Arm Neon C++ flags" FORCE)
324 # Arm Neon Asimd (64-bit ARM)
325 function(gmx_find_simd_arm_neon_asimd_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
327 gmx_find_flags(SIMD_ARM_NEON_ASIMD_C_FLAGS_RESULT SIMD_ARM_NEON_ASIMD_CXX_FLAGS_RESULT
328 "#include<arm_neon.h>
329 int main(){float64x2_t x=vdupq_n_f64(0.5);x=vfmaq_f64(x,x,x);x=vrndnq_f64(x);return vgetq_lane_f64(x,0)>0;}"
330 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
331 SIMD_ARM_NEON_ASIMD_C_FLAGS SIMD_ARM_NEON_ASIMD_CXX_FLAGS
334 if(${SIMD_ARM_NEON_ASIMD_C_FLAGS_RESULT})
335 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_ARM_NEON_ASIMD_C_FLAGS}" CACHE INTERNAL "C flags required for Arm Neon Asimd instructions")
337 if(${SIMD_ARM_NEON_ASIMD_CXX_FLAGS_RESULT})
338 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_ARM_NEON_ASIMD_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for Arm Neon Asimd instructions")
340 set(${C_FLAGS_RESULT} ${SIMD_ARM_NEON_ASIMD_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for Arm Neon Asimd C flags" FORCE)
341 set(${CXX_FLAGS_RESULT} ${SIMD_ARM_NEON_ASIMD_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for Arm Neon Asimd C++ flags" FORCE)
345 function(gmx_find_simd_ibm_vmx_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
347 gmx_find_flags(SIMD_IBM_VMX_C_FLAGS_RESULT SIMD_IBM_VMX_CXX_FLAGS_RESULT
349 int main(){vector float x,y=vec_ctf(vec_splat_s32(1),0);x=vec_madd(y,y,y);return vec_all_ge(y,x);}"
350 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
351 SIMD_IBM_VMX_C_FLAGS SIMD_IBM_VMX_CXX_FLAGS
352 "-maltivec -mabi=altivec" "-qarch=auto -qaltivec")
354 if(${SIMD_IBM_VMX_C_FLAGS_RESULT})
355 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_IBM_VMX_C_FLAGS}" CACHE INTERNAL "C flags required for IBM VMX instructions")
357 if(${SIMD_IBM_VMX_CXX_FLAGS_RESULT})
358 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_IBM_VMX_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for IBM VMX instructions")
360 set(${C_FLAGS_RESULT} ${SIMD_IBM_VMX_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for IBM VMX C flags" FORCE)
361 set(${CXX_FLAGS_RESULT} ${SIMD_IBM_VMX_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for IBM VMX C++ flags" FORCE)
364 # IBM VSX (power7 and later)
365 function(gmx_find_simd_ibm_vsx_flags C_FLAGS_RESULT CXX_FLAGS_RESULT C_FLAGS_VARIABLE CXX_FLAGS_VARIABLE)
366 find_power_vsx_toolchain_flags(TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS)
367 gmx_find_flags(SIMD_IBM_VSX_C_FLAGS_RESULT SIMD_IBM_VSX_CXX_FLAGS_RESULT
369 int main(){vector double x,y=vec_splats(1.0);x=vec_madd(y,y,y);return vec_all_ge(y,x);}"
370 TOOLCHAIN_C_FLAGS TOOLCHAIN_CXX_FLAGS
371 SIMD_IBM_VSX_C_FLAGS SIMD_IBM_VSX_CXX_FLAGS
372 "-mvsx" "-maltivec -mabi=altivec" "-qarch=auto -qaltivec")
374 if(${SIMD_IBM_VSX_C_FLAGS_RESULT})
375 set(${C_FLAGS_VARIABLE} "${TOOLCHAIN_C_FLAGS} ${SIMD_IBM_VSX_C_FLAGS}" CACHE INTERNAL "C flags required for IBM VSX instructions")
377 if(${SIMD_IBM_VSX_CXX_FLAGS_RESULT})
378 set(${CXX_FLAGS_VARIABLE} "${TOOLCHAIN_CXX_FLAGS} ${SIMD_IBM_VSX_CXX_FLAGS}" CACHE INTERNAL "C++ flags required for IBM VSX instructions")
380 set(${C_FLAGS_RESULT} ${SIMD_IBM_VSX_C_FLAGS_RESULT} CACHE INTERNAL "Result of test for IBM VSX C flags" FORCE)
381 set(${CXX_FLAGS_RESULT} ${SIMD_IBM_VSX_CXX_FLAGS_RESULT} CACHE INTERNAL "Result of test for IBM VSX C++ flags" FORCE)