2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2013,2014, 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.
37 * \defgroup module_simd SIMD intrinsics interface (simd)
38 * \ingroup group_utilitymodules
40 * \brief Provides an architecture-independent way of doing SIMD coding.
42 * Overview of the SIMD implementation is provided in \ref page_simd.
43 * The details are documented in simd.h and the reference implementation
46 * \author Erik Lindahl <erik.lindahl@scilifelab.se>
49 #ifndef GMX_SIMD_SIMD_H
50 #define GMX_SIMD_SIMD_H
52 /*! \libinternal \file
54 * \brief Definitions, capabilities, and wrappers for SIMD module.
56 * The macros in this file are intended to be used for writing
57 * architecture-independent SIMD intrinsics code.
58 * To support a new architecture, adding a new sub-include with macros here
59 * should be (nearly) all that is needed.
61 * The defines in this top-level file will set default Gromacs real precision
62 * operations to either single or double precision based on whether
63 * GMX_DOUBLE is defined. The actual implementation - including e.g.
64 * conversion operations specifically between single and double - is documented
65 * in impl_reference.h.
67 * \author Erik Lindahl <erik.lindahl@scilifelab.se>
70 * \ingroup module_simd
77 #include "gromacs/utility/basedefinitions.h"
79 /* Forward declarations so memory allocation can be used in implementations */
80 static gmx_inline float * gmx_simd_align_f(float *p);
81 static gmx_inline double * gmx_simd_align_d(double *p);
82 static gmx_inline int * gmx_simd_align_fi(int *p);
83 static gmx_inline int * gmx_simd_align_di(int *p);
84 static gmx_inline float * gmx_simd4_align_f(float *p);
85 static gmx_inline double * gmx_simd4_align_d(double *p);
88 /*! \addtogroup module_simd */
91 /*! \name SIMD predefined macros to describe high-level capabilities
93 * These macros are used to describe the features available in default
94 * Gromacs real precision. They are set from the lower-level implementation
95 * files that have macros describing single and double precision individually,
96 * as well as the implementation details.
101 * GMX_SIMD indicates that some sort of SIMD support is present in software.
103 * It is disabled if no architecture, neither reference SIMD, has been selected.
108 /* Intel MIC is a bit special since it is a co-processor. This means the rest
109 * of GROMACS (which runs on the CPU) should use a default SIMD set like AVX,
110 * while the part running on the coprocessor defines __MIC__. All functions in
111 * this SIMD module are static, so it will work perfectly fine to include this
112 * file with different SIMD definitions for different files.
115 # include "gromacs/simd/impl_intel_mic/impl_intel_mic.h"
116 #elif defined GMX_SIMD_X86_AVX2_256
117 # include "gromacs/simd/impl_x86_avx2_256/impl_x86_avx2_256.h"
118 #elif defined GMX_SIMD_X86_AVX_256
119 # include "gromacs/simd/impl_x86_avx_256/impl_x86_avx_256.h"
120 #elif defined GMX_SIMD_X86_AVX_128_FMA
121 # include "gromacs/simd/impl_x86_avx_128_fma/impl_x86_avx_128_fma.h"
122 #elif defined GMX_SIMD_X86_SSE4_1
123 # include "gromacs/simd/impl_x86_sse4_1/impl_x86_sse4_1.h"
124 #elif defined GMX_SIMD_X86_SSE2
125 # include "gromacs/simd/impl_x86_sse2/impl_x86_sse2.h"
126 #elif defined GMX_SIMD_IBM_QPX
127 # include "gromacs/simd/impl_ibm_qpx/impl_ibm_qpx.h"
128 #elif (defined GMX_SIMD_REFERENCE) || (defined DOXYGEN)
129 /* Plain C SIMD reference implementation, also serves as documentation.
130 * For now this code path will also be taken for Sparc64_HPC_ACE since we have
131 * not yet added the verlet kernel extensions there. The group kernels do not
132 * depend on this file, so they will still be accelerated with SIMD.
134 # include "gromacs/simd/impl_reference/impl_reference.h"
136 /* Turn off the GMX_SIMD flag if we do not even have reference support */
141 * SIMD4 width is always 4, but use this for clarity in definitions.
143 * It improves code readability to allocate e.g. 2*GMX_SIMD4_WIDTH instead of 8.
145 #define GMX_SIMD4_WIDTH 4
149 /*! \name SIMD memory alignment operations
154 * Align a float pointer for usage with SIMD instructions.
156 * You should typically \a not call this function directly (unless you explicitly
157 * want single precision even when GMX_DOUBLE is set), but use the
158 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
160 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FLOAT_WIDTH extra elements.
162 * \return Aligned pointer (>=p) suitable for loading/storing float fp SIMD.
163 * If \ref GMX_SIMD_HAVE_FLOAT is not set, p will be returned unchanged.
165 * Start by allocating an extra \ref GMX_SIMD_FLOAT_WIDTH float elements of memory,
166 * and then call this function. The returned pointer will be greater or equal
167 * to the one you provided, and point to an address inside your provided memory
168 * that is aligned to the SIMD width.
170 static gmx_inline float *
171 gmx_simd_align_f(float *p)
173 # ifdef GMX_SIMD_HAVE_FLOAT
174 return (float *)(((size_t)((p)+GMX_SIMD_FLOAT_WIDTH-1)) & (~((size_t)(GMX_SIMD_FLOAT_WIDTH*sizeof(float)-1))));
181 * Align a double pointer for usage with SIMD instructions.
183 * You should typically \a not call this function directly (unless you explicitly
184 * want double precision even when GMX_DOUBLE is not set), but use the
185 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
187 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DOUBLE_WIDTH extra elements.
189 * \return Aligned pointer (>=p) suitable for loading/storing double fp SIMD.
190 * If \ref GMX_SIMD_HAVE_DOUBLE is not set, p will be returned unchanged.
192 * Start by allocating an extra \ref GMX_SIMD_DOUBLE_WIDTH double elements of memory,
193 * and then call this function. The returned pointer will be greater or equal
194 * to the one you provided, and point to an address inside your provided memory
195 * that is aligned to the SIMD width.
197 static gmx_inline double *
198 gmx_simd_align_d(double *p)
200 # ifdef GMX_SIMD_HAVE_DOUBLE
201 return (double *)(((size_t)((p)+GMX_SIMD_DOUBLE_WIDTH-1)) & (~((size_t)(GMX_SIMD_DOUBLE_WIDTH*sizeof(double)-1))));
208 * Align a (float) integer pointer for usage with SIMD instructions.
210 * You should typically \a not call this function directly (unless you explicitly
211 * want integers corresponding to single precision even when GMX_DOUBLE is
212 * set), but use the \ref gmx_simd_align_i macro to align integer memory
213 * corresponding to Gromacs default floating-point precision.
215 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FINT32_WIDTH extra elements.
217 * \return Aligned pointer (>=p) suitable for loading/storing float-integer SIMD.
218 * If \ref GMX_SIMD_HAVE_FINT32 is not set, p will be returned unchanged.
220 * This routine provides aligned memory for usage with \ref gmx_simd_fint32_t. You
221 * should have allocated an extra \ref GMX_SIMD_FINT32_WIDTH * sizeof(int) bytes. The
222 * reason why we need to separate float-integer vs. double-integer is that the
223 * width of registers after conversions from the floating-point types might not
224 * be identical, or even supported, in both cases.
226 static gmx_inline int *
227 gmx_simd_align_fi(int *p)
229 # ifdef GMX_SIMD_HAVE_FINT32
230 return (int *)(((size_t)((p)+GMX_SIMD_FINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_FINT32_WIDTH*sizeof(int)-1))));
237 * Align a (double) integer pointer for usage with SIMD instructions.
239 * You should typically \a not call this function directly (unless you explicitly
240 * want integers corresponding to doublele precision even when GMX_DOUBLE is
241 * not set), but use the \ref gmx_simd_align_i macro to align integer memory
242 * corresponding to Gromacs default floating-point precision.
244 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DINT32_WIDTH extra elements.
246 * \return Aligned pointer (>=p) suitable for loading/storing double-integer SIMD.
247 * If \ref GMX_SIMD_HAVE_DINT32 is not set, p will be returned unchanged.
249 * This routine provides aligned memory for usage with \ref gmx_simd_dint32_t. You
250 * should have allocated an extra \ref GMX_SIMD_DINT32_WIDTH*sizeof(int) bytes. The
251 * reason why we need to separate float-integer vs. double-integer is that the
252 * width of registers after conversions from the floating-point types might not
253 * be identical, or even supported, in both cases.
255 static gmx_inline int *
256 gmx_simd_align_di(int *p)
258 # ifdef GMX_SIMD_HAVE_DINT32
259 return (int *)(((size_t)((p)+GMX_SIMD_DINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_DINT32_WIDTH*sizeof(int)-1))));
266 * Align a float pointer for usage with SIMD4 instructions.
268 * You should typically \a not call this function directly (unless you explicitly
269 * want single precision even when GMX_DOUBLE is set), but use the
270 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
272 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
274 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
275 * If \ref GMX_SIMD4_HAVE_FLOAT is not set, p will be returned unchanged.
277 * This routine provides aligned memory for usage with \ref gmx_simd4_float_t.
278 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(float) bytes.
280 static gmx_inline float *
281 gmx_simd4_align_f(float *p)
283 # ifdef GMX_SIMD4_HAVE_FLOAT
284 return (float *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(float)-1))));
291 * Align a double pointer for usage with SIMD4 instructions.
293 * You should typically \a not call this function directly (unless you explicitly
294 * want double precision even when GMX_DOUBLE is not set), but use the
295 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
297 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
299 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
300 * If \ref GMX_SIMD4_HAVE_DOUBLE is not set, p will be returned unchanged.
302 * This routine provides aligned memory for usage with \ref gmx_simd4_double_t.
303 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(double) bytes.
305 static gmx_inline double *
306 gmx_simd4_align_d(double *p)
308 # ifdef GMX_SIMD4_HAVE_DOUBLE
309 return (double *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(double)-1))));
318 /* Define Gromacs "real" precision macros depending on Gromacs config. Note
319 * that conversions float-to-double and v.v. are not included here since they
320 * are not precision-dependent - find them in the implementation files.
323 /* Double floating-point. The documentation is in the float part below */
324 # define gmx_simd_real_t gmx_simd_double_t
325 # define gmx_simd_load_r gmx_simd_load_d
326 # define gmx_simd_load1_r gmx_simd_load1_d
327 # define gmx_simd_set1_r gmx_simd_set1_d
328 # define gmx_simd_store_r gmx_simd_store_d
329 # define gmx_simd_loadu_r gmx_simd_loadu_d
330 # define gmx_simd_storeu_r gmx_simd_storeu_d
331 # define gmx_simd_setzero_r gmx_simd_setzero_d
332 # define gmx_simd_add_r gmx_simd_add_d
333 # define gmx_simd_sub_r gmx_simd_sub_d
334 # define gmx_simd_mul_r gmx_simd_mul_d
335 # define gmx_simd_fmadd_r gmx_simd_fmadd_d
336 # define gmx_simd_fmsub_r gmx_simd_fmsub_d
337 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_d
338 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_d
339 # define gmx_simd_and_r gmx_simd_and_d
340 # define gmx_simd_andnot_r gmx_simd_andnot_d
341 # define gmx_simd_or_r gmx_simd_or_d
342 # define gmx_simd_xor_r gmx_simd_xor_d
343 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_d
344 # define gmx_simd_rcp_r gmx_simd_rcp_d
345 # define gmx_simd_fabs_r gmx_simd_fabs_d
346 # define gmx_simd_fneg_r gmx_simd_fneg_d
347 # define gmx_simd_max_r gmx_simd_max_d
348 # define gmx_simd_min_r gmx_simd_min_d
349 # define gmx_simd_round_r gmx_simd_round_d
350 # define gmx_simd_trunc_r gmx_simd_trunc_d
351 # define gmx_simd_fraction_r gmx_simd_fraction_d
352 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_d
353 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_d
354 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_d
355 /* Double integer and conversions */
356 # define gmx_simd_int32_t gmx_simd_dint32_t
357 # define gmx_simd_load_i gmx_simd_load_di
358 # define gmx_simd_set1_i gmx_simd_set1_di
359 # define gmx_simd_store_i gmx_simd_store_di
360 # define gmx_simd_loadu_i gmx_simd_loadu_di
361 # define gmx_simd_storeu_i gmx_simd_storeu_di
362 # define gmx_simd_setzero_i gmx_simd_setzero_di
363 # define gmx_simd_cvt_r2i gmx_simd_cvt_d2i
364 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_d2i
365 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2d
366 # define gmx_simd_extract_i gmx_simd_extract_di
367 # define gmx_simd_slli_i gmx_simd_slli_di
368 # define gmx_simd_srli_i gmx_simd_srli_di
369 # define gmx_simd_and_i gmx_simd_and_di
370 # define gmx_simd_andnot_i gmx_simd_andnot_di
371 # define gmx_simd_or_i gmx_simd_or_di
372 # define gmx_simd_xor_i gmx_simd_xor_di
373 # define gmx_simd_add_i gmx_simd_add_di
374 # define gmx_simd_sub_i gmx_simd_sub_di
375 # define gmx_simd_mul_i gmx_simd_mul_di
376 /* Double booleans and selection */
377 # define gmx_simd_bool_t gmx_simd_dbool_t
378 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_d
379 # define gmx_simd_cmplt_r gmx_simd_cmplt_d
380 # define gmx_simd_cmple_r gmx_simd_cmple_d
381 # define gmx_simd_and_b gmx_simd_and_db
382 # define gmx_simd_or_b gmx_simd_or_db
383 # define gmx_simd_anytrue_b gmx_simd_anytrue_db
384 # define gmx_simd_blendzero_r gmx_simd_blendzero_d
385 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_d
386 # define gmx_simd_blendv_r gmx_simd_blendv_d
387 # define gmx_simd_reduce_r gmx_simd_reduce_d
388 # define gmx_simd_ibool_t gmx_simd_dibool_t
389 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_di
390 # define gmx_simd_cmplt_i gmx_simd_cmplt_di
391 # define gmx_simd_and_ib gmx_simd_and_dib
392 # define gmx_simd_or_ib gmx_simd_or_dib
393 # define gmx_simd_anytrue_ib gmx_simd_anytrue_dib
394 # define gmx_simd_blendzero_i gmx_simd_blendzero_di
395 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_di
396 # define gmx_simd_blendv_i gmx_simd_blendv_di
397 /* Conversions between integer and double floating-point booleans */
398 # define gmx_simd_cvt_b2ib gmx_simd_cvt_db2dib
399 # define gmx_simd_cvt_ib2b gmx_simd_cvt_dib2db
401 /* SIMD4 double fp - we only support a subset of SIMD instructions for SIMD4 */
402 # define gmx_simd4_real_t gmx_simd4_double_t
403 # define gmx_simd4_load_r gmx_simd4_load_d
404 # define gmx_simd4_load1_r gmx_simd4_load1_d
405 # define gmx_simd4_set1_r gmx_simd4_set1_d
406 # define gmx_simd4_store_r gmx_simd4_store_d
407 # define gmx_simd4_loadu_r gmx_simd4_loadu_d
408 # define gmx_simd4_storeu_r gmx_simd4_storeu_d
409 # define gmx_simd4_setzero_r gmx_simd4_setzero_d
410 # define gmx_simd4_add_r gmx_simd4_add_d
411 # define gmx_simd4_sub_r gmx_simd4_sub_d
412 # define gmx_simd4_mul_r gmx_simd4_mul_d
413 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_d
414 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_d
415 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_d
416 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_d
417 # define gmx_simd4_and_r gmx_simd4_and_d
418 # define gmx_simd4_andnot_r gmx_simd4_andnot_d
419 # define gmx_simd4_or_r gmx_simd4_or_d
420 # define gmx_simd4_xor_r gmx_simd4_xor_d
421 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_d
422 # define gmx_simd4_fabs_r gmx_simd4_fabs_d
423 # define gmx_simd4_fneg_r gmx_simd4_fneg_d
424 # define gmx_simd4_max_r gmx_simd4_max_d
425 # define gmx_simd4_min_r gmx_simd4_min_d
426 # define gmx_simd4_round_r gmx_simd4_round_d
427 # define gmx_simd4_trunc_r gmx_simd4_trunc_d
428 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_d
429 # define gmx_simd4_bool_t gmx_simd4_dbool_t
430 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_d
431 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_d
432 # define gmx_simd4_cmple_r gmx_simd4_cmple_d
433 # define gmx_simd4_and_b gmx_simd4_and_db
434 # define gmx_simd4_or_b gmx_simd4_or_db
435 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_db
436 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_d
437 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_d
438 # define gmx_simd4_blendv_r gmx_simd4_blendv_d
439 # define gmx_simd4_reduce_r gmx_simd4_reduce_d
441 /* Memory allocation */
442 # define gmx_simd_align_r gmx_simd_align_d
443 # define gmx_simd_align_i gmx_simd_align_di
444 # define gmx_simd4_align_r gmx_simd4_align_d
446 # ifdef GMX_SIMD_HAVE_DOUBLE
447 # define GMX_SIMD_HAVE_REAL
448 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_DOUBLE_WIDTH
450 # ifdef GMX_SIMD_HAVE_DINT32
451 # define GMX_SIMD_HAVE_INT32
452 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_DINT32_WIDTH
454 # ifdef GMX_SIMD_HAVE_DINT32_EXTRACT
455 # define GMX_SIMD_HAVE_INT32_EXTRACT
457 # ifdef GMX_SIMD_HAVE_DINT32_LOGICAL
458 # define GMX_SIMD_HAVE_INT32_LOGICAL
460 # ifdef GMX_SIMD_HAVE_DINT32_ARITHMETICS
461 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
463 # ifdef GMX_SIMD4_HAVE_DOUBLE
464 # define GMX_SIMD4_HAVE_REAL
467 #else /* GMX_DOUBLE */
469 /*! \name SIMD data types
471 * The actual storage of these types is implementation dependent. The
472 * documentation is generated from the reference implementation, but for
473 * normal usage this will likely not be what you are using.
476 /*! \brief Real precision floating-point SIMD datatype.
478 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
480 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_double_t
481 * internally, otherwise \ref gmx_simd_float_t.
483 # define gmx_simd_real_t gmx_simd_float_t
485 /*! \brief 32-bit integer SIMD type.
487 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
489 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dint32_t
490 * internally, otherwise \ref gmx_simd_fint32_t. This might seem a strange
491 * implementation detail, but it is because some SIMD implementations use
492 * different types/widths of integers registers when converting from
493 * double vs. single precision floating point. As long as you just use
494 * this type you will not have to worry about precision.
496 # define gmx_simd_int32_t gmx_simd_fint32_t
498 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_real_t.
500 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
502 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dbool_t
503 * internally, otherwise \ref gmx_simd_fbool_t. This is necessary since some
504 * SIMD implementations use bitpatterns for marking truth, so single-
505 * vs. double precision booleans are not necessarily exchangable.
506 * As long as you just use this type you will not have to worry about precision.
508 * See \ref gmx_simd_ibool_t for an explanation of real vs. integer booleans.
510 # define gmx_simd_bool_t gmx_simd_fbool_t
512 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_int32_t.
514 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
516 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dibool_t
517 * internally, otherwise \ref gmx_simd_fibool_t. This is necessary since some
518 * SIMD implementations use bitpatterns for marking truth, so single-
519 * vs. double precision booleans are not necessarily exchangable, and while
520 * a double-precision boolean might be represented with a 64-bit mask, the
521 * corresponding integer might only use a 32-bit mask.
523 * We provide conversion routines for these cases, so the only thing you need to
524 * keep in mind is to use \ref gmx_simd_bool_t when working with
525 * \ref gmx_simd_real_t while you pick \ref gmx_simd_ibool_t when working with
526 * \ref gmx_simd_int32_t.
528 * To convert between them, use \ref gmx_simd_cvt_b2ib and \ref gmx_simd_cvt_ib2b.
530 # define gmx_simd_ibool_t gmx_simd_fibool_t
534 * \name SIMD load/store operations on gmx_simd_real_t
536 * \note Unaligned load/stores are only available when
537 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
541 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from aligned memory to \ref gmx_simd_real_t
543 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_d,
544 * otherwise \ref gmx_simd_load_f.
546 * \copydetails gmx_simd_load_f
548 # define gmx_simd_load_r gmx_simd_load_f
550 /*! \brief Set all elements in \ref gmx_simd_real_t from single value in memory.
552 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load1_d,
553 * otherwise \ref gmx_simd_load1_f.
555 * \copydetails gmx_simd_load1_f
557 # define gmx_simd_load1_r gmx_simd_load1_f
559 /*! \brief Set all elements in \ref gmx_simd_real_t from a scalar.
561 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_d,
562 * otherwise \ref gmx_simd_set1_f.
564 * \copydetails gmx_simd_set1_f
566 # define gmx_simd_set1_r gmx_simd_set1_f
568 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to aligned memory.
570 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_d,
571 * otherwise \ref gmx_simd_store_f.
573 * \copydetails gmx_simd_store_f
575 # define gmx_simd_store_r gmx_simd_store_f
577 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_real_t.
579 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_d,
580 * otherwise \ref gmx_simd_loadu_f.
582 * \copydetails gmx_simd_loadu_f
584 # define gmx_simd_loadu_r gmx_simd_loadu_f
586 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to unaligned memory.
588 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_d,
589 * otherwise \ref gmx_simd_storeu_f.
591 * \copydetails gmx_simd_storeu_f
593 # define gmx_simd_storeu_r gmx_simd_storeu_f
595 /*! \brief Set all elements in \ref gmx_simd_real_t to 0.0.
597 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_setzero_d,
598 * otherwise \ref gmx_simd_setzero_f.
600 * \copydetails gmx_simd_setzero_f
602 # define gmx_simd_setzero_r gmx_simd_setzero_f
605 * \name SIMD load/store operations on gmx_simd_int32_t
607 * \note Unaligned load/stores are only available when
608 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
612 /*! \brief Load \ref GMX_SIMD_INT32_WIDTH values from aligned memory to \ref gmx_simd_int32_t .
614 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_di ,
615 * otherwise \ref gmx_simd_load_fi .
617 * \copydetails gmx_simd_load_fi
619 # define gmx_simd_load_i gmx_simd_load_fi
621 /*! \brief Set all elements in \ref gmx_simd_int32_t from a single integer.
623 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_di ,
624 * otherwise \ref gmx_simd_set1_fi .
626 * \copydetails gmx_simd_set1_fi
628 # define gmx_simd_set1_i gmx_simd_set1_fi
630 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to aligned memory.
632 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_di ,
633 * otherwise \ref gmx_simd_store_fi .
635 * \copydetails gmx_simd_store_fi
637 # define gmx_simd_store_i gmx_simd_store_fi
639 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_int32_t.
641 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_di ,
642 * otherwise \ref gmx_simd_loadu_fi .
644 * \copydetails gmx_simd_loadu_fi
646 # define gmx_simd_loadu_i gmx_simd_loadu_fi
648 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to unaligned memory.
650 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_di ,
651 * otherwise \ref gmx_simd_storeu_fi .
653 * \copydetails gmx_simd_storeu_fi
655 # define gmx_simd_storeu_i gmx_simd_storeu_fi
657 /*! \brief Extract single integer from \ref gmx_simd_int32_t element.
659 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_extract_di ,
660 * otherwise \ref gmx_simd_extract_fi .
662 * \copydetails gmx_simd_extract_fi
664 # define gmx_simd_extract_i gmx_simd_extract_fi
666 /*! \brief Set all elements in \ref gmx_simd_int32_t to 0.
668 * If GMX_DOUBLE is defined, it will be aliased to \ref gmx_simd_setzero_di ,
669 * otherwise \ref gmx_simd_setzero_fi .
671 * \copydetails gmx_simd_setzero_fi
673 # define gmx_simd_setzero_i gmx_simd_setzero_fi
677 * \name SIMD floating-point logical operations on gmx_simd_real_t
679 * These instructions are available if \ref GMX_SIMD_HAVE_LOGICAL is defined.
683 /*! \brief Bitwise \a and on two \ref gmx_simd_real_t.
685 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_d,
686 * otherwise \ref gmx_simd_and_f.
688 * \copydetails gmx_simd_and_f
690 # define gmx_simd_and_r gmx_simd_and_f
692 /*! \brief Bitwise \a and-not on two \ref gmx_simd_real_t; 1st arg is complemented.
694 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_d,
695 * otherwise \ref gmx_simd_andnot_f.
697 * \copydetails gmx_simd_andnot_f
699 # define gmx_simd_andnot_r gmx_simd_andnot_f
701 /*! \brief Bitwise \a or on two \ref gmx_simd_real_t.
703 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_d,
704 * otherwise \ref gmx_simd_or_f.
706 * \copydetails gmx_simd_or_f
708 # define gmx_simd_or_r gmx_simd_or_f
710 /*! \brief Bitwise \a exclusive-or on two \ref gmx_simd_real_t.
712 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_d,
713 * otherwise \ref gmx_simd_xor_f.
715 * \copydetails gmx_simd_xor_f
717 # define gmx_simd_xor_r gmx_simd_xor_f
720 * \name SIMD floating-point arithmetic operations on gmx_simd_real_t
724 /*! \brief SIMD a+b for two \ref gmx_simd_real_t.
726 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_d,
727 * otherwise \ref gmx_simd_add_f.
729 * \copydetails gmx_simd_add_f
731 # define gmx_simd_add_r gmx_simd_add_f
733 /*! \brief SIMD a-b for two \ref gmx_simd_real_t.
735 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_d,
736 * otherwise \ref gmx_simd_sub_f.
738 * \copydetails gmx_simd_sub_f
740 # define gmx_simd_sub_r gmx_simd_sub_f
742 /*! \brief SIMD a*b for two \ref gmx_simd_real_t.
744 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_d,
745 * otherwise \ref gmx_simd_mul_f.
747 * \copydetails gmx_simd_mul_f
749 # define gmx_simd_mul_r gmx_simd_mul_f
751 /*! \brief SIMD a*b+c for three \ref gmx_simd_real_t.
753 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmadd_d,
754 * otherwise \ref gmx_simd_fmadd_f.
756 * \copydetails gmx_simd_fmadd_f
758 # define gmx_simd_fmadd_r gmx_simd_fmadd_f
760 /*! \brief SIMD a*b-c for three \ref gmx_simd_real_t.
762 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmsub_d,
763 * otherwise \ref gmx_simd_fmsub_f.
765 * \copydetails gmx_simd_fmsub_f
767 # define gmx_simd_fmsub_r gmx_simd_fmsub_f
769 /*! \brief SIMD -a*b+c for three \ref gmx_simd_real_t.
771 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmadd_d,
772 * otherwise \ref gmx_simd_fnmadd_f.
774 * \copydetails gmx_simd_fnmadd_f
776 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_f
778 /*! \brief SIMD -a*b-c for three \ref gmx_simd_real_t.
780 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmsub_d,
781 * otherwise \ref gmx_simd_fnmsub_f.
783 * \copydetails gmx_simd_fnmsub_f
785 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_f
787 /*! \brief SIMD table lookup for 1/sqrt(x) approximation.
789 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rsqrt_d,
790 * otherwise \ref gmx_simd_rsqrt_f.
792 * \copydetails gmx_simd_rsqrt_f
794 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_f
796 /*! \brief SIMD table lookup for 1/x approximation.
798 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rcp_d,
799 * otherwise \ref gmx_simd_rcp_f.
801 * \copydetails gmx_simd_rcp_f
803 # define gmx_simd_rcp_r gmx_simd_rcp_f
805 /*! \brief SIMD fabs(x) for \ref gmx_simd_real_t.
807 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fabs_d,
808 * otherwise \ref gmx_simd_fabs_f.
810 * \copydetails gmx_simd_fabs_f
812 # define gmx_simd_fabs_r gmx_simd_fabs_f
814 /*! \brief SIMD -x for \ref gmx_simd_real_t.
816 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fneg_d,
817 * otherwise \ref gmx_simd_fneg_f.
819 * \copydetails gmx_simd_fneg_f
821 # define gmx_simd_fneg_r gmx_simd_fneg_f
823 /*! \brief SIMD max(a,b) for each element in \ref gmx_simd_real_t.
825 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_max_d,
826 * otherwise \ref gmx_simd_max_f.
828 * \copydetails gmx_simd_max_f
830 # define gmx_simd_max_r gmx_simd_max_f
832 /*! \brief SIMD min(a,b) for each element in \ref gmx_simd_real_t.
834 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_min_d,
835 * otherwise \ref gmx_simd_min_f.
837 * \copydetails gmx_simd_min_f
839 # define gmx_simd_min_r gmx_simd_min_f
841 /*! \brief Round \ref gmx_simd_real_t to nearest int, return \ref gmx_simd_real_t.
843 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_round_d,
844 * otherwise \ref gmx_simd_round_f.
846 * \copydetails gmx_simd_round_f
848 # define gmx_simd_round_r gmx_simd_round_f
850 /*! \brief Truncate \ref gmx_simd_real_t towards 0, return \ref gmx_simd_real_t.
852 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_trunc_d,
853 * otherwise \ref gmx_simd_trunc_f.
855 * \copydetails gmx_simd_trunc_f
857 # define gmx_simd_trunc_r gmx_simd_trunc_f
859 /*! \brief SIMD Fraction, i.e. x-trunc(x) for \ref gmx_simd_real_t.
861 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fraction_d,
862 * otherwise \ref gmx_simd_fraction_f.
864 * \copydetails gmx_simd_fraction_f
866 # define gmx_simd_fraction_r gmx_simd_fraction_f
868 /*! \brief Return the FP exponent of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
870 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_exponent_d,
871 * otherwise \ref gmx_simd_get_exponent_f.
873 * \copydetails gmx_simd_exponent_f
875 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_f
877 /*! \brief Return the FP mantissa of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
879 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_mantissa_d,
880 * otherwise \ref gmx_simd_get_mantissa_f.
882 * \copydetails gmx_simd_mantissa_f
884 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_f
886 /*! \brief Set the exponent of a SIMD \ref gmx_simd_real_t from a \ref gmx_simd_real_t.
888 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set_exponent_d,
889 * otherwise \ref gmx_simd_set_exponent_f.
891 * \copydetails gmx_simd_set_exponent_f
893 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_f
896 * \name SIMD comparison, boolean, and select operations for gmx_simd_real_t
900 /*! \brief SIMD a==b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
902 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_d,
903 * otherwise \ref gmx_simd_cmpeq_f.
905 * \copydetails gmx_simd_cmpeq_f
907 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_f
909 /*! \brief SIMD a<b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
911 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_d,
912 * otherwise \ref gmx_simd_cmplt_f.
914 * \copydetails gmx_simd_cmplt_f
916 # define gmx_simd_cmplt_r gmx_simd_cmplt_f
918 /*! \brief SIMD a<=b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
920 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmple_d,
921 * otherwise \ref gmx_simd_cmple_f.
923 * \copydetails gmx_simd_cmple_f
925 # define gmx_simd_cmple_r gmx_simd_cmple_f
927 /*! \brief For each element, the result boolean is true if both arguments are true
929 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_db,
930 * otherwise \ref gmx_simd_and_fb.
932 * \copydetails gmx_simd_and_fb
934 # define gmx_simd_and_b gmx_simd_and_fb
936 /*! \brief For each element, the result boolean is true if either argument is true
938 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_db,
939 * otherwise \ref gmx_simd_or_fb.
941 * \copydetails gmx_simd_or_fn
943 # define gmx_simd_or_b gmx_simd_or_fb
945 /*! \brief Return nonzero if any element in gmx_simd_bool_t is true, otherwise 0.
947 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_db,
948 * otherwise \ref gmx_simd_anytrue_fb.
950 * \copydetails gmx_simd_anytrue_fb
952 # define gmx_simd_anytrue_b gmx_simd_anytrue_fb
954 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is true, otherwise 0.
956 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_d,
957 * otherwise \ref gmx_simd_blendzero_f.
959 * \copydetails gmx_simd_blendzero_f
961 * \sa gmx_simd_blendzero_i
963 # define gmx_simd_blendzero_r gmx_simd_blendzero_f
965 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is false, otherwise 0.
967 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_d,
968 * otherwise \ref gmx_simd_blendnotzero_f.
970 * \copydetails gmx_simd_blendnotzero_f
972 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_f
974 /*! \brief Selects from 2nd real SIMD arg where boolean is true, otherwise 1st arg.
976 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_d,
977 * otherwise \ref gmx_simd_blendv_f.
979 * \copydetails gmx_simd_blendv_f
981 # define gmx_simd_blendv_r gmx_simd_blendv_f
983 /*! \brief Return sum of all elements in SIMD floating-point variable.
985 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_reduce_d,
986 * otherwise \ref gmx_simd_reduce_f.
988 * \copydetails gmx_simd_reduce_f
990 # define gmx_simd_reduce_r gmx_simd_reduce_f
993 * \name SIMD integer logical operations on gmx_simd_int32_t
995 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_LOGICAL is defined.
999 /*! \brief Shift each element in \ref gmx_simd_int32_t left by immediate
1001 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_slli_di,
1002 * otherwise \ref gmx_simd_slli_fi.
1004 * \copydetails gmx_simd_slli_fi
1006 # define gmx_simd_slli_i gmx_simd_slli_fi
1008 /*! \brief Shift each element in \ref gmx_simd_int32_t right by immediate
1010 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_srli_di,
1011 * otherwise \ref gmx_simd_srli_fi.
1013 * \copydetails gmx_simd_srli_fi
1015 # define gmx_simd_srli_i gmx_simd_srli_fi
1017 /*! \brief Bitwise \a and on two \ref gmx_simd_int32_t.
1019 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_di,
1020 * otherwise \ref gmx_simd_and_fi.
1022 * \copydetails gmx_simd_and_fi
1024 # define gmx_simd_and_i gmx_simd_and_fi
1026 /*! \brief Bitwise \a and-not on two \ref gmx_simd_int32_t; 1st arg is complemented.
1028 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_di,
1029 * otherwise \ref gmx_simd_andnot_fi.
1031 * \copydetails gmx_simd_andnot_fi
1033 # define gmx_simd_andnot_i gmx_simd_andnot_fi
1035 /*! \brief Bitwise \a or on two \ref gmx_simd_int32_t.
1037 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_di,
1038 * otherwise \ref gmx_simd_or_fi.
1040 * \copydetails gmx_simd_or_fi
1042 # define gmx_simd_or_i gmx_simd_or_fi
1044 /*! \brief Bitwise \a xor on two \ref gmx_simd_int32_t.
1046 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_di,
1047 * otherwise \ref gmx_simd_xor_fi.
1049 * \copydetails gmx_simd_xor_fi
1051 # define gmx_simd_xor_i gmx_simd_xor_fi
1054 * \name SIMD integer arithmetic operations on gmx_simd_int32_t
1056 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1060 /*! \brief SIMD a+b for two \ref gmx_simd_int32_t.
1062 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_di,
1063 * otherwise \ref gmx_simd_add_fi.
1065 * \copydetails gmx_simd_add_fi
1067 # define gmx_simd_add_i gmx_simd_add_fi
1069 /*! \brief SIMD a-b for two \ref gmx_simd_int32_t.
1071 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_di,
1072 * otherwise \ref gmx_simd_sub_fi.
1074 * \copydetails gmx_simd_sub_fi
1076 # define gmx_simd_sub_i gmx_simd_sub_fi
1078 /*! \brief SIMD a*b for two \ref gmx_simd_int32_t.
1080 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_di,
1081 * otherwise \ref gmx_simd_mul_fi.
1083 * \copydetails gmx_simd_mul_fi
1085 # define gmx_simd_mul_i gmx_simd_mul_fi
1088 * \name SIMD integer comparison, booleans, and selection on gmx_simd_int32_t
1090 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1094 /*! \brief Returns boolean describing whether a==b, for \ref gmx_simd_int32_t
1096 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_di,
1097 * otherwise \ref gmx_simd_cmpeq_fi.
1099 * \copydetails gmx_simd_cmpeq_fi
1101 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_fi
1103 /*! \brief Returns boolean describing whether a<b, for \ref gmx_simd_int32_t
1105 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_di,
1106 * otherwise \ref gmx_simd_cmplt_fi.
1108 * \copydetails gmx_simd_cmplt_fi
1110 # define gmx_simd_cmplt_i gmx_simd_cmplt_fi
1112 /*! \brief For each element, the result boolean is true if both arguments are true
1114 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_dib,
1115 * otherwise \ref gmx_simd_and_fib.
1117 * \copydetails gmx_simd_and_fib
1119 # define gmx_simd_and_ib gmx_simd_and_fib
1121 /*! \brief For each element, the result boolean is true if either argument is true.
1123 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_dib,
1124 * otherwise \ref gmx_simd_or_fib.
1126 * \copydetails gmx_simd_or_fib
1128 # define gmx_simd_or_ib gmx_simd_or_fib
1130 /*! \brief Return nonzero if any element in gmx_simd_ibool_t is true, otherwise 0.
1132 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_dib,
1133 * otherwise \ref gmx_simd_anytrue_fib.
1135 * \copydetails gmx_simd_anytrue_fib
1137 # define gmx_simd_anytrue_ib gmx_simd_anytrue_fib
1139 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is true, otherwise 0.
1141 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_di,
1142 * otherwise \ref gmx_simd_blendzero_fi.
1144 * \copydetails gmx_simd_blendzero_fi
1146 # define gmx_simd_blendzero_i gmx_simd_blendzero_fi
1148 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is false, otherwise 0.
1150 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_di,
1151 * otherwise \ref gmx_simd_blendnotzero_fi.
1153 * \copydetails gmx_simd_blendnotzero_fi
1155 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_fi
1157 /*! \brief Selects from 2nd int SIMD arg where boolean is true, otherwise 1st arg.
1159 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_di,
1160 * otherwise \ref gmx_simd_blendv_fi.
1162 * \copydetails gmx_simd_blendv_fi
1164 # define gmx_simd_blendv_i gmx_simd_blendv_fi
1167 * \name SIMD conversion operations
1169 * These instructions are available when both types involved in the conversion
1170 * are defined, e.g. \ref GMX_SIMD_HAVE_REAL and \ref GMX_SIMD_HAVE_INT32
1171 * for real-to-integer conversion.
1175 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, round to nearest integer.
1177 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_d2i,
1178 * otherwise \ref gmx_simd_cvt_f2i.
1180 * \copydetails gmx_simd_cvt_f2i
1182 # define gmx_simd_cvt_r2i gmx_simd_cvt_f2i
1184 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, truncate towards zero
1186 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvtt_d2i,
1187 * otherwise \ref gmx_simd_cvtt_f2i.
1189 * \copydetails gmx_simd_cvtt_f2i
1191 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_f2i
1193 /*! \brief Convert gmx_simd_int32_t to gmx_simd_real_t
1195 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_i2d,
1196 * otherwise \ref gmx_simd_cvt_i2f.
1198 * \copydetails gmx_simd_cvt_i2f
1200 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2f
1202 /*! \brief Convert from gmx_simd_bool_t to gmx_simd_ibool_t
1204 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_db2dib,
1205 * otherwise \ref gmx_simd_cvt_fb2fib.
1207 * \copydetails gmx_simd_cvt_fb2fib
1209 # define gmx_simd_cvt_b2ib gmx_simd_cvt_fb2fib
1211 /*! \brief Convert from gmx_simd_ibool_t to gmx_simd_bool_t
1213 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_dib2db,
1214 * otherwise \ref gmx_simd_cvt_fib2fb.
1216 * \copydetails gmx_simd_cvt_fib2fb
1218 # define gmx_simd_cvt_ib2b gmx_simd_cvt_fib2fb
1222 * \name SIMD memory alignment operations
1226 /*! \brief Align real memory for SIMD usage.
1228 * This routine will only align memory if \ref GMX_SIMD_HAVE_REAL is defined.
1229 * Otherwise the original pointer will be returned.
1231 * Start by allocating an extra \ref GMX_SIMD_REAL_WIDTH float elements of memory,
1232 * and then call this function. The returned pointer will be greater or equal
1233 * to the one you provided, and point to an address inside your provided memory
1234 * that is aligned to the SIMD width.
1236 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_d,
1237 * otherwise \ref gmx_simd_align_f. For detailed documentation, see the
1238 * precision-specific implementation routines.
1240 # define gmx_simd_align_r gmx_simd_align_f
1242 /*! \brief Align integer memory for SIMD usage.
1244 * This routine will only align memory if \ref GMX_SIMD_HAVE_INT32 is defined.
1245 * Otherwise the original pointer will be returned.
1247 * Start by allocating an extra \ref GMX_SIMD_INT32_WIDTH elements of memory,
1248 * and then call this function. The returned pointer will be greater or equal
1249 * to the one you provided, and point to an address inside your provided memory
1250 * that is aligned to the SIMD width.
1252 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_di,
1253 * otherwise \ref gmx_simd_align_fi. For detailed documentation, see the
1254 * precision-specific implementation routines.
1256 # define gmx_simd_align_i gmx_simd_align_fi
1260 /*! \name SIMD4 - constant width-four SIMD datatypes
1262 * These operations are only meant to be used for a few coordinate
1263 * manipulation and grid interpolation routines, so we only support a subset
1264 * of operations for SIMD4. To avoid repeating all the documentation from
1265 * the generic width SIMD routines, we only provide brief documentation for
1266 * these operations. Follow the link to the implementation documentation or the
1267 * reference to the corresponding generic SIMD routine. The format will be
1268 * exactly the same, but they have SIMD replaced with SIMD4.
1272 /*! \brief SIMD real datatype guaranteed to be 4 elements wide, if available.
1274 * All the SIMD4 datatypes and operations behave like their counterparts for
1275 * the generic SIMD implementation, but they might be implemented with different
1276 * registers, or not supported at all. It is important that you check the
1277 * define \ref GMX_SIMD4_HAVE_REAL before using it.
1279 * Just as the normal SIMD operations, all SIMD4 types and routines will
1280 * be aliased to either single or double precision ones based on whether
1281 * GMX_DOUBLE is defined.
1283 * \note There is no support for integer or math operations in SIMD4.
1285 # define gmx_simd4_real_t gmx_simd4_float_t
1287 /*! \brief Boolean for \ref gmx_simd4_real_t comparision/selection */
1288 # define gmx_simd4_bool_t gmx_simd4_fbool_t
1290 /*! \brief Load aligned data to gmx_simd4_real_t.
1292 * \copydetails gmx_simd4_load_f
1294 # define gmx_simd4_load_r gmx_simd4_load_f
1296 /*! \brief Load single element to gmx_simd4_real_t
1298 * \copydetails gmx_simd4_load1_f
1300 # define gmx_simd4_load1_r gmx_simd4_load1_f
1302 /*! \brief Set gmx_simd4_real_t from scalar value
1304 * \copydetails gmx_simd4_set1_f
1306 # define gmx_simd4_set1_r gmx_simd4_set1_f
1308 /*! \brief store aligned data from gmx_simd4_real_t
1310 * \copydetails gmx_simd4_store_f
1312 # define gmx_simd4_store_r gmx_simd4_store_f
1314 /*! \brief Load unaligned data to gmx_simd4_real_t
1316 * \copydetails gmx_simd4_loadu_f
1318 # define gmx_simd4_loadu_r gmx_simd4_loadu_f
1320 /*! \brief Store unaligned data from gmx_simd4_real_t
1322 * \copydetails gmx_simd4_storeu_f
1324 # define gmx_simd4_storeu_r gmx_simd4_storeu_f
1326 /*! \brief Set all elements in gmx_simd4_real_t to 0.0
1328 * \copydetails gmx_simd4_setzero_f
1330 # define gmx_simd4_setzero_r gmx_simd4_setzero_f
1332 /*! \brief Bitwise and for two gmx_simd4_real_t
1334 * \copydetails gmx_simd4_and_f
1336 # define gmx_simd4_and_r gmx_simd4_and_f
1338 /*! \brief Bitwise and-not for two gmx_simd4_real_t. 1st arg is complemented.
1340 * \copydetails gmx_simd4_andnot_f
1342 # define gmx_simd4_andnot_r gmx_simd4_andnot_f
1344 /*! \brief Bitwise or for two gmx_simd4_real_t
1346 * \copydetails gmx_simd4_or_f
1348 # define gmx_simd4_or_r gmx_simd4_or_f
1350 /*! \brief Bitwise xor for two gmx_simd4_real_t
1352 * \copydetails gmx_simd4_xor_f
1354 # define gmx_simd4_xor_r gmx_simd4_xor_f
1356 /*! \brief a+b for \ref gmx_simd4_real_t
1358 * \copydetails gmx_simd4_add_f
1360 # define gmx_simd4_add_r gmx_simd4_add_f
1362 /*! \brief a-b for \ref gmx_simd4_real_t
1364 * \copydetails gmx_simd4_sub_f
1366 # define gmx_simd4_sub_r gmx_simd4_sub_f
1368 /*! \brief a*b for \ref gmx_simd4_real_t
1370 * \copydetails gmx_simd4_mul_f
1372 # define gmx_simd4_mul_r gmx_simd4_mul_f
1374 /*! \brief a*b+c for \ref gmx_simd4_real_t
1376 * \copydetails gmx_simd4_fmadd_f
1378 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_f
1380 /*! \brief a*b-c for \ref gmx_simd4_real_t
1382 * \copydetails gmx_simd4_fmsub_f
1384 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_f
1386 /*! \brief -a*b+c for \ref gmx_simd4_real_t
1388 * \copydetails gmx_simd4_fnmadd_f
1390 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_f
1392 /*! \brief -a*b-c for \ref gmx_simd4_real_t
1394 * \copydetails gmx_simd4_fnmsub_f
1396 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_f
1398 /*! \brief 1/sqrt(x) approximate lookup for \ref gmx_simd4_real_t
1400 * \copydetails gmx_simd4_rsqrt_f
1402 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_f
1404 /*! \brief fabs(x) for \ref gmx_simd4_real_t
1406 * \copydetails gmx_simd4_fabs_f
1408 # define gmx_simd4_fabs_r gmx_simd4_fabs_f
1410 /*! \brief Change sign (-x) for \ref gmx_simd4_real_t
1412 * \copydetails gmx_simd4_fneg_f
1414 # define gmx_simd4_fneg_r gmx_simd4_fneg_f
1416 /*! \brief Select maximum of each pair of elements from args for \ref gmx_simd4_real_t
1418 * \copydetails gmx_simd4_max_f
1420 # define gmx_simd4_max_r gmx_simd4_max_f
1422 /*! \brief Select minimum of each pair of elements from args for \ref gmx_simd4_real_t
1424 * \copydetails gmx_simd4_min_f
1426 # define gmx_simd4_min_r gmx_simd4_min_f
1428 /*! \brief Round \ref gmx_simd4_real_t to nearest integer, return \ref gmx_simd4_real_t
1430 * \copydetails gmx_simd4_round_f
1432 # define gmx_simd4_round_r gmx_simd4_round_f
1434 /*! \brief Truncate \ref gmx_simd4_real_t towards zero, return \ref gmx_simd4_real_t
1436 * \copydetails gmx_simd4_trunc_f
1438 # define gmx_simd4_trunc_r gmx_simd4_trunc_f
1440 /*! \brief Scalar product of first three elements of two \ref gmx_simd4_real_t *
1442 * \copydetails gmx_simd4_dotproduct3_f
1444 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_f
1446 /*! \brief Return booleans whether a==b for each element two \ref gmx_simd4_real_t
1448 * \copydetails gmx_simd4_cmpeq_f
1450 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_f
1451 /*! \brief Return booleans whether a<b for each element two \ref gmx_simd4_real_t
1453 * \copydetails gmx_simd4_cmplt_f
1455 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_f
1456 /*! \brief Return booleans whether a<=b for each element two \ref gmx_simd4_real_t
1458 * \copydetails gmx_simd4_cmple_f
1460 # define gmx_simd4_cmple_r gmx_simd4_cmple_f
1462 /*! \brief Logical and for two \ref gmx_simd4_bool_t
1464 * \copydetails gmx_simd4_and_fb
1466 # define gmx_simd4_and_b gmx_simd4_and_fb
1467 /*! \brief Logical or for two \ref gmx_simd4_bool_t
1469 * \copydetails gmx_simd4_or_fb
1471 # define gmx_simd4_or_b gmx_simd4_or_fb
1473 /*! \brief Return nonzero if any element in \ref gmx_simd4_bool_t is true, otherwise 0
1475 * \copydetails gmx_simd4_anytrue_fb
1477 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_fb
1479 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1481 * \copydetails gmx_simd4_blendzero_f
1483 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_f
1485 /*! \brief Selects from 2nd real SIMD4 arg where boolean is false, otherwise 1st arg
1487 * \copydetails gmx_simd4_blendnotzero_f
1489 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_f
1491 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1493 * \copydetails gmx_simd4_blendv_f
1495 # define gmx_simd4_blendv_r gmx_simd4_blendv_f
1497 /*! \brief Return sum of all elements in SIMD4 floating-point variable.
1499 * \copydetails gmx_simd4_reduce_f
1501 # define gmx_simd4_reduce_r gmx_simd4_reduce_f
1503 /*! \brief Align real memory for SIMD4 usage.
1505 * \copydetails gmx_simd4_align_f
1507 # define gmx_simd4_align_r gmx_simd4_align_f
1511 /*! \name SIMD predefined macros to describe high-level capabilities
1515 # if (defined GMX_SIMD_HAVE_FLOAT) || (defined DOXYGEN)
1516 /*! \brief Defined if gmx_simd_real_t is available.
1518 * if GMX_DOUBLE is defined, this will be aliased to
1519 * \ref GMX_SIMD_HAVE_DOUBLE, otherwise GMX_SIMD_HAVE_FLOAT.
1521 # define GMX_SIMD_HAVE_REAL
1522 /*! \brief Width of gmx_simd_real_t.
1524 * if GMX_DOUBLE is defined, this will be aliased to
1525 * \ref GMX_SIMD_DOUBLE_WIDTH, otherwise GMX_SIMD_FLOAT_WIDTH.
1527 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_FLOAT_WIDTH
1529 # if (defined GMX_SIMD_HAVE_FINT32) || (defined DOXYGEN)
1530 /*! \brief Defined if gmx_simd_int32_t is available.
1532 * if GMX_DOUBLE is defined, this will be aliased to
1533 * \ref GMX_SIMD_HAVE_DINT32, otherwise GMX_SIMD_HAVE_FINT32.
1535 # define GMX_SIMD_HAVE_INT32
1536 /*! \brief Width of gmx_simd_int32_t.
1538 * if GMX_DOUBLE is defined, this will be aliased to
1539 * \ref GMX_SIMD_DINT32_WIDTH, otherwise GMX_SIMD_FINT32_WIDTH.
1541 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_FINT32_WIDTH
1543 # if (defined GMX_SIMD_HAVE_FINT32_EXTRACT) || (defined DOXYGEN)
1544 /*! \brief Defined if gmx_simd_extract_i() is available.
1546 * if GMX_DOUBLE is defined, this will be aliased to
1547 * \ref GMX_SIMD_HAVE_DINT32_EXTRACT, otherwise GMX_SIMD_HAVE_FINT32_EXTRACT.
1549 # define GMX_SIMD_HAVE_INT32_EXTRACT
1551 # if (defined GMX_SIMD_HAVE_FINT32_LOGICAL) || (defined DOXYGEN)
1552 /*! \brief Defined if logical ops are supported on gmx_simd_int32_t.
1554 * if GMX_DOUBLE is defined, this will be aliased to
1555 * \ref GMX_SIMD_HAVE_DINT32_LOGICAL, otherwise GMX_SIMD_HAVE_FINT32_LOGICAL.
1557 # define GMX_SIMD_HAVE_INT32_LOGICAL
1559 # if (defined GMX_SIMD_HAVE_FINT32_ARITHMETICS) || (defined DOXYGEN)
1560 /*! \brief Defined if arithmetic ops are supported on gmx_simd_int32_t.
1562 * if GMX_DOUBLE is defined, this will be aliased to
1563 * \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS, otherwise GMX_SIMD_HAVE_FINT32_ARITHMETICS.
1565 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
1567 # if (defined GMX_SIMD4_HAVE_FLOAT) || (defined DOXYGEN)
1568 /*! \brief Defined if gmx_simd4_real_t is available.
1570 * if GMX_DOUBLE is defined, this will be aliased to
1571 * \ref GMX_SIMD4_HAVE_DOUBLE, otherwise GMX_SIMD4_HAVE_FLOAT.
1573 # define GMX_SIMD4_HAVE_REAL
1578 #endif /* GMX_DOUBLE */
1583 #endif /* GMX_SIMD_SIMD_H */