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 "impl_intel_mic/impl_intel_mic.h"
116 #elif defined GMX_SIMD_X86_AVX2_256
117 # include "impl_x86_avx2_256/impl_x86_avx2_256.h"
118 #elif defined GMX_SIMD_X86_AVX_256
119 # include "impl_x86_avx_256/impl_x86_avx_256.h"
120 #elif defined GMX_SIMD_X86_AVX_128_FMA
121 # include "impl_x86_avx_128_fma/impl_x86_avx_128_fma.h"
122 #elif defined GMX_SIMD_X86_SSE4_1
123 # include "impl_x86_sse4_1/impl_x86_sse4_1.h"
124 #elif defined GMX_SIMD_X86_SSE2
125 # include "impl_x86_sse2/impl_x86_sse2.h"
126 #elif defined GMX_SIMD_ARM_NEON
127 # include "impl_arm_neon/impl_arm_neon.h"
128 #elif defined GMX_SIMD_ARM_NEON_ASIMD
129 # include "impl_arm_neon_asimd/impl_arm_neon_asimd.h"
130 #elif defined GMX_SIMD_IBM_QPX
131 # include "impl_ibm_qpx/impl_ibm_qpx.h"
132 #elif defined GMX_SIMD_SPARC64_HPC_ACE
133 # include "impl_sparc64_hpc_ace/impl_sparc64_hpc_ace.h"
134 #elif (defined GMX_SIMD_REFERENCE) || (defined DOXYGEN)
135 /* Plain C SIMD reference implementation, also serves as documentation.
136 * For now this code path will also be taken for Sparc64_HPC_ACE since we have
137 * not yet added the verlet kernel extensions there. The group kernels do not
138 * depend on this file, so they will still be accelerated with SIMD.
140 # include "impl_reference/impl_reference.h"
142 /* Turn off the GMX_SIMD flag if we do not even have reference support */
147 * SIMD4 width is always 4, but use this for clarity in definitions.
149 * It improves code readability to allocate e.g. 2*GMX_SIMD4_WIDTH instead of 8.
151 #define GMX_SIMD4_WIDTH 4
155 /*! \name SIMD memory alignment operations
160 * Align a float pointer for usage with SIMD instructions.
162 * You should typically \a not call this function directly (unless you explicitly
163 * want single precision even when GMX_DOUBLE is set), but use the
164 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
166 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FLOAT_WIDTH extra elements.
168 * \return Aligned pointer (>=p) suitable for loading/storing float fp SIMD.
169 * If \ref GMX_SIMD_HAVE_FLOAT is not set, p will be returned unchanged.
171 * Start by allocating an extra \ref GMX_SIMD_FLOAT_WIDTH float elements of memory,
172 * and then call this function. The returned pointer will be greater or equal
173 * to the one you provided, and point to an address inside your provided memory
174 * that is aligned to the SIMD width.
176 static gmx_inline float *
177 gmx_simd_align_f(float *p)
179 # ifdef GMX_SIMD_HAVE_FLOAT
180 return (float *)(((size_t)((p)+GMX_SIMD_FLOAT_WIDTH-1)) & (~((size_t)(GMX_SIMD_FLOAT_WIDTH*sizeof(float)-1))));
187 * Align a double pointer for usage with SIMD instructions.
189 * You should typically \a not call this function directly (unless you explicitly
190 * want double precision even when GMX_DOUBLE is not set), but use the
191 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
193 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DOUBLE_WIDTH extra elements.
195 * \return Aligned pointer (>=p) suitable for loading/storing double fp SIMD.
196 * If \ref GMX_SIMD_HAVE_DOUBLE is not set, p will be returned unchanged.
198 * Start by allocating an extra \ref GMX_SIMD_DOUBLE_WIDTH double elements of memory,
199 * and then call this function. The returned pointer will be greater or equal
200 * to the one you provided, and point to an address inside your provided memory
201 * that is aligned to the SIMD width.
203 static gmx_inline double *
204 gmx_simd_align_d(double *p)
206 # ifdef GMX_SIMD_HAVE_DOUBLE
207 return (double *)(((size_t)((p)+GMX_SIMD_DOUBLE_WIDTH-1)) & (~((size_t)(GMX_SIMD_DOUBLE_WIDTH*sizeof(double)-1))));
214 * Align a (float) integer pointer for usage with SIMD instructions.
216 * You should typically \a not call this function directly (unless you explicitly
217 * want integers corresponding to single precision even when GMX_DOUBLE is
218 * set), but use the \ref gmx_simd_align_i macro to align integer memory
219 * corresponding to Gromacs default floating-point precision.
221 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FINT32_WIDTH extra elements.
223 * \return Aligned pointer (>=p) suitable for loading/storing float-integer SIMD.
224 * If \ref GMX_SIMD_HAVE_FINT32 is not set, p will be returned unchanged.
226 * This routine provides aligned memory for usage with \ref gmx_simd_fint32_t. You
227 * should have allocated an extra \ref GMX_SIMD_FINT32_WIDTH * sizeof(int) bytes. The
228 * reason why we need to separate float-integer vs. double-integer is that the
229 * width of registers after conversions from the floating-point types might not
230 * be identical, or even supported, in both cases.
232 static gmx_inline int *
233 gmx_simd_align_fi(int *p)
235 # ifdef GMX_SIMD_HAVE_FINT32
236 return (int *)(((size_t)((p)+GMX_SIMD_FINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_FINT32_WIDTH*sizeof(int)-1))));
243 * Align a (double) integer pointer for usage with SIMD instructions.
245 * You should typically \a not call this function directly (unless you explicitly
246 * want integers corresponding to doublele precision even when GMX_DOUBLE is
247 * not set), but use the \ref gmx_simd_align_i macro to align integer memory
248 * corresponding to Gromacs default floating-point precision.
250 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DINT32_WIDTH extra elements.
252 * \return Aligned pointer (>=p) suitable for loading/storing double-integer SIMD.
253 * If \ref GMX_SIMD_HAVE_DINT32 is not set, p will be returned unchanged.
255 * This routine provides aligned memory for usage with \ref gmx_simd_dint32_t. You
256 * should have allocated an extra \ref GMX_SIMD_DINT32_WIDTH*sizeof(int) bytes. The
257 * reason why we need to separate float-integer vs. double-integer is that the
258 * width of registers after conversions from the floating-point types might not
259 * be identical, or even supported, in both cases.
261 static gmx_inline int *
262 gmx_simd_align_di(int *p)
264 # ifdef GMX_SIMD_HAVE_DINT32
265 return (int *)(((size_t)((p)+GMX_SIMD_DINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_DINT32_WIDTH*sizeof(int)-1))));
272 * Align a float pointer for usage with SIMD4 instructions.
274 * You should typically \a not call this function directly (unless you explicitly
275 * want single precision even when GMX_DOUBLE is set), but use the
276 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
278 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
280 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
281 * If \ref GMX_SIMD4_HAVE_FLOAT is not set, p will be returned unchanged.
283 * This routine provides aligned memory for usage with \ref gmx_simd4_float_t.
284 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(float) bytes.
286 static gmx_inline float *
287 gmx_simd4_align_f(float *p)
289 # ifdef GMX_SIMD4_HAVE_FLOAT
290 return (float *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(float)-1))));
297 * Align a double pointer for usage with SIMD4 instructions.
299 * You should typically \a not call this function directly (unless you explicitly
300 * want double precision even when GMX_DOUBLE is not set), but use the
301 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
303 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
305 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
306 * If \ref GMX_SIMD4_HAVE_DOUBLE is not set, p will be returned unchanged.
308 * This routine provides aligned memory for usage with \ref gmx_simd4_double_t.
309 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(double) bytes.
311 static gmx_inline double *
312 gmx_simd4_align_d(double *p)
314 # ifdef GMX_SIMD4_HAVE_DOUBLE
315 return (double *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(double)-1))));
324 /* Define Gromacs "real" precision macros depending on Gromacs config. Note
325 * that conversions float-to-double and v.v. are not included here since they
326 * are not precision-dependent - find them in the implementation files.
329 /* Double floating-point. The documentation is in the float part below */
330 # define gmx_simd_real_t gmx_simd_double_t
331 # define gmx_simd_load_r gmx_simd_load_d
332 # define gmx_simd_load1_r gmx_simd_load1_d
333 # define gmx_simd_set1_r gmx_simd_set1_d
334 # define gmx_simd_store_r gmx_simd_store_d
335 # define gmx_simd_loadu_r gmx_simd_loadu_d
336 # define gmx_simd_storeu_r gmx_simd_storeu_d
337 # define gmx_simd_setzero_r gmx_simd_setzero_d
338 # define gmx_simd_add_r gmx_simd_add_d
339 # define gmx_simd_sub_r gmx_simd_sub_d
340 # define gmx_simd_mul_r gmx_simd_mul_d
341 # define gmx_simd_fmadd_r gmx_simd_fmadd_d
342 # define gmx_simd_fmsub_r gmx_simd_fmsub_d
343 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_d
344 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_d
345 # define gmx_simd_and_r gmx_simd_and_d
346 # define gmx_simd_andnot_r gmx_simd_andnot_d
347 # define gmx_simd_or_r gmx_simd_or_d
348 # define gmx_simd_xor_r gmx_simd_xor_d
349 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_d
350 # define gmx_simd_rcp_r gmx_simd_rcp_d
351 # define gmx_simd_fabs_r gmx_simd_fabs_d
352 # define gmx_simd_fneg_r gmx_simd_fneg_d
353 # define gmx_simd_max_r gmx_simd_max_d
354 # define gmx_simd_min_r gmx_simd_min_d
355 # define gmx_simd_round_r gmx_simd_round_d
356 # define gmx_simd_trunc_r gmx_simd_trunc_d
357 # define gmx_simd_fraction_r gmx_simd_fraction_d
358 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_d
359 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_d
360 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_d
361 /* Double integer and conversions */
362 # define gmx_simd_int32_t gmx_simd_dint32_t
363 # define gmx_simd_load_i gmx_simd_load_di
364 # define gmx_simd_set1_i gmx_simd_set1_di
365 # define gmx_simd_store_i gmx_simd_store_di
366 # define gmx_simd_loadu_i gmx_simd_loadu_di
367 # define gmx_simd_storeu_i gmx_simd_storeu_di
368 # define gmx_simd_setzero_i gmx_simd_setzero_di
369 # define gmx_simd_cvt_r2i gmx_simd_cvt_d2i
370 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_d2i
371 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2d
372 # define gmx_simd_extract_i gmx_simd_extract_di
373 # define gmx_simd_slli_i gmx_simd_slli_di
374 # define gmx_simd_srli_i gmx_simd_srli_di
375 # define gmx_simd_and_i gmx_simd_and_di
376 # define gmx_simd_andnot_i gmx_simd_andnot_di
377 # define gmx_simd_or_i gmx_simd_or_di
378 # define gmx_simd_xor_i gmx_simd_xor_di
379 # define gmx_simd_add_i gmx_simd_add_di
380 # define gmx_simd_sub_i gmx_simd_sub_di
381 # define gmx_simd_mul_i gmx_simd_mul_di
382 /* Double booleans and selection */
383 # define gmx_simd_bool_t gmx_simd_dbool_t
384 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_d
385 # define gmx_simd_cmplt_r gmx_simd_cmplt_d
386 # define gmx_simd_cmple_r gmx_simd_cmple_d
387 # define gmx_simd_and_b gmx_simd_and_db
388 # define gmx_simd_or_b gmx_simd_or_db
389 # define gmx_simd_anytrue_b gmx_simd_anytrue_db
390 # define gmx_simd_blendzero_r gmx_simd_blendzero_d
391 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_d
392 # define gmx_simd_blendv_r gmx_simd_blendv_d
393 # define gmx_simd_reduce_r gmx_simd_reduce_d
394 # define gmx_simd_ibool_t gmx_simd_dibool_t
395 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_di
396 # define gmx_simd_cmplt_i gmx_simd_cmplt_di
397 # define gmx_simd_and_ib gmx_simd_and_dib
398 # define gmx_simd_or_ib gmx_simd_or_dib
399 # define gmx_simd_anytrue_ib gmx_simd_anytrue_dib
400 # define gmx_simd_blendzero_i gmx_simd_blendzero_di
401 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_di
402 # define gmx_simd_blendv_i gmx_simd_blendv_di
403 /* Conversions between integer and double floating-point booleans */
404 # define gmx_simd_cvt_b2ib gmx_simd_cvt_db2dib
405 # define gmx_simd_cvt_ib2b gmx_simd_cvt_dib2db
407 /* SIMD4 double fp - we only support a subset of SIMD instructions for SIMD4 */
408 # define gmx_simd4_real_t gmx_simd4_double_t
409 # define gmx_simd4_load_r gmx_simd4_load_d
410 # define gmx_simd4_load1_r gmx_simd4_load1_d
411 # define gmx_simd4_set1_r gmx_simd4_set1_d
412 # define gmx_simd4_store_r gmx_simd4_store_d
413 # define gmx_simd4_loadu_r gmx_simd4_loadu_d
414 # define gmx_simd4_storeu_r gmx_simd4_storeu_d
415 # define gmx_simd4_setzero_r gmx_simd4_setzero_d
416 # define gmx_simd4_add_r gmx_simd4_add_d
417 # define gmx_simd4_sub_r gmx_simd4_sub_d
418 # define gmx_simd4_mul_r gmx_simd4_mul_d
419 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_d
420 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_d
421 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_d
422 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_d
423 # define gmx_simd4_and_r gmx_simd4_and_d
424 # define gmx_simd4_andnot_r gmx_simd4_andnot_d
425 # define gmx_simd4_or_r gmx_simd4_or_d
426 # define gmx_simd4_xor_r gmx_simd4_xor_d
427 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_d
428 # define gmx_simd4_fabs_r gmx_simd4_fabs_d
429 # define gmx_simd4_fneg_r gmx_simd4_fneg_d
430 # define gmx_simd4_max_r gmx_simd4_max_d
431 # define gmx_simd4_min_r gmx_simd4_min_d
432 # define gmx_simd4_round_r gmx_simd4_round_d
433 # define gmx_simd4_trunc_r gmx_simd4_trunc_d
434 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_d
435 # define gmx_simd4_bool_t gmx_simd4_dbool_t
436 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_d
437 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_d
438 # define gmx_simd4_cmple_r gmx_simd4_cmple_d
439 # define gmx_simd4_and_b gmx_simd4_and_db
440 # define gmx_simd4_or_b gmx_simd4_or_db
441 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_db
442 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_d
443 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_d
444 # define gmx_simd4_blendv_r gmx_simd4_blendv_d
445 # define gmx_simd4_reduce_r gmx_simd4_reduce_d
447 /* Memory allocation */
448 # define gmx_simd_align_r gmx_simd_align_d
449 # define gmx_simd_align_i gmx_simd_align_di
450 # define gmx_simd4_align_r gmx_simd4_align_d
452 # ifdef GMX_SIMD_HAVE_DOUBLE
453 # define GMX_SIMD_HAVE_REAL
454 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_DOUBLE_WIDTH
456 # ifdef GMX_SIMD_HAVE_DINT32
457 # define GMX_SIMD_HAVE_INT32
458 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_DINT32_WIDTH
460 # ifdef GMX_SIMD_HAVE_DINT32_EXTRACT
461 # define GMX_SIMD_HAVE_INT32_EXTRACT
463 # ifdef GMX_SIMD_HAVE_DINT32_LOGICAL
464 # define GMX_SIMD_HAVE_INT32_LOGICAL
466 # ifdef GMX_SIMD_HAVE_DINT32_ARITHMETICS
467 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
469 # ifdef GMX_SIMD4_HAVE_DOUBLE
470 # define GMX_SIMD4_HAVE_REAL
473 #else /* GMX_DOUBLE */
475 /*! \name SIMD data types
477 * The actual storage of these types is implementation dependent. The
478 * documentation is generated from the reference implementation, but for
479 * normal usage this will likely not be what you are using.
482 /*! \brief Real precision floating-point SIMD datatype.
484 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
486 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_double_t
487 * internally, otherwise \ref gmx_simd_float_t.
489 # define gmx_simd_real_t gmx_simd_float_t
491 /*! \brief 32-bit integer SIMD type.
493 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
495 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dint32_t
496 * internally, otherwise \ref gmx_simd_fint32_t. This might seem a strange
497 * implementation detail, but it is because some SIMD implementations use
498 * different types/widths of integers registers when converting from
499 * double vs. single precision floating point. As long as you just use
500 * this type you will not have to worry about precision.
502 # define gmx_simd_int32_t gmx_simd_fint32_t
504 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_real_t.
506 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
508 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dbool_t
509 * internally, otherwise \ref gmx_simd_fbool_t. This is necessary since some
510 * SIMD implementations use bitpatterns for marking truth, so single-
511 * vs. double precision booleans are not necessarily exchangable.
512 * As long as you just use this type you will not have to worry about precision.
514 * See \ref gmx_simd_ibool_t for an explanation of real vs. integer booleans.
516 # define gmx_simd_bool_t gmx_simd_fbool_t
518 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_int32_t.
520 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
522 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dibool_t
523 * internally, otherwise \ref gmx_simd_fibool_t. This is necessary since some
524 * SIMD implementations use bitpatterns for marking truth, so single-
525 * vs. double precision booleans are not necessarily exchangable, and while
526 * a double-precision boolean might be represented with a 64-bit mask, the
527 * corresponding integer might only use a 32-bit mask.
529 * We provide conversion routines for these cases, so the only thing you need to
530 * keep in mind is to use \ref gmx_simd_bool_t when working with
531 * \ref gmx_simd_real_t while you pick \ref gmx_simd_ibool_t when working with
532 * \ref gmx_simd_int32_t.
534 * To convert between them, use \ref gmx_simd_cvt_b2ib and \ref gmx_simd_cvt_ib2b.
536 # define gmx_simd_ibool_t gmx_simd_fibool_t
540 * \name SIMD load/store operations on gmx_simd_real_t
542 * \note Unaligned load/stores are only available when
543 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
547 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from aligned memory to \ref gmx_simd_real_t
549 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_d,
550 * otherwise \ref gmx_simd_load_f.
552 * \copydetails gmx_simd_load_f
554 # define gmx_simd_load_r gmx_simd_load_f
556 /*! \brief Set all elements in \ref gmx_simd_real_t from single value in memory.
558 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load1_d,
559 * otherwise \ref gmx_simd_load1_f.
561 * \copydetails gmx_simd_load1_f
563 # define gmx_simd_load1_r gmx_simd_load1_f
565 /*! \brief Set all elements in \ref gmx_simd_real_t from a scalar.
567 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_d,
568 * otherwise \ref gmx_simd_set1_f.
570 * \copydetails gmx_simd_set1_f
572 # define gmx_simd_set1_r gmx_simd_set1_f
574 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to aligned memory.
576 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_d,
577 * otherwise \ref gmx_simd_store_f.
579 * \copydetails gmx_simd_store_f
581 # define gmx_simd_store_r gmx_simd_store_f
583 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_real_t.
585 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_d,
586 * otherwise \ref gmx_simd_loadu_f.
588 * \copydetails gmx_simd_loadu_f
590 # define gmx_simd_loadu_r gmx_simd_loadu_f
592 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to unaligned memory.
594 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_d,
595 * otherwise \ref gmx_simd_storeu_f.
597 * \copydetails gmx_simd_storeu_f
599 # define gmx_simd_storeu_r gmx_simd_storeu_f
601 /*! \brief Set all elements in \ref gmx_simd_real_t to 0.0.
603 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_setzero_d,
604 * otherwise \ref gmx_simd_setzero_f.
606 * \copydetails gmx_simd_setzero_f
608 # define gmx_simd_setzero_r gmx_simd_setzero_f
611 * \name SIMD load/store operations on gmx_simd_int32_t
613 * \note Unaligned load/stores are only available when
614 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
618 /*! \brief Load \ref GMX_SIMD_INT32_WIDTH values from aligned memory to \ref gmx_simd_int32_t .
620 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_di ,
621 * otherwise \ref gmx_simd_load_fi .
623 * \copydetails gmx_simd_load_fi
625 # define gmx_simd_load_i gmx_simd_load_fi
627 /*! \brief Set all elements in \ref gmx_simd_int32_t from a single integer.
629 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_di ,
630 * otherwise \ref gmx_simd_set1_fi .
632 * \copydetails gmx_simd_set1_fi
634 # define gmx_simd_set1_i gmx_simd_set1_fi
636 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to aligned memory.
638 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_di ,
639 * otherwise \ref gmx_simd_store_fi .
641 * \copydetails gmx_simd_store_fi
643 # define gmx_simd_store_i gmx_simd_store_fi
645 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_int32_t.
647 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_di ,
648 * otherwise \ref gmx_simd_loadu_fi .
650 * \copydetails gmx_simd_loadu_fi
652 # define gmx_simd_loadu_i gmx_simd_loadu_fi
654 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to unaligned memory.
656 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_di ,
657 * otherwise \ref gmx_simd_storeu_fi .
659 * \copydetails gmx_simd_storeu_fi
661 # define gmx_simd_storeu_i gmx_simd_storeu_fi
663 /*! \brief Extract single integer from \ref gmx_simd_int32_t element.
665 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_extract_di ,
666 * otherwise \ref gmx_simd_extract_fi .
668 * \copydetails gmx_simd_extract_fi
670 # define gmx_simd_extract_i gmx_simd_extract_fi
672 /*! \brief Set all elements in \ref gmx_simd_int32_t to 0.
674 * If GMX_DOUBLE is defined, it will be aliased to \ref gmx_simd_setzero_di ,
675 * otherwise \ref gmx_simd_setzero_fi .
677 * \copydetails gmx_simd_setzero_fi
679 # define gmx_simd_setzero_i gmx_simd_setzero_fi
683 * \name SIMD floating-point logical operations on gmx_simd_real_t
685 * These instructions are available if \ref GMX_SIMD_HAVE_LOGICAL is defined.
689 /*! \brief Bitwise \a and on two \ref gmx_simd_real_t.
691 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_d,
692 * otherwise \ref gmx_simd_and_f.
694 * \copydetails gmx_simd_and_f
696 # define gmx_simd_and_r gmx_simd_and_f
698 /*! \brief Bitwise \a and-not on two \ref gmx_simd_real_t; 1st arg is complemented.
700 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_d,
701 * otherwise \ref gmx_simd_andnot_f.
703 * \copydetails gmx_simd_andnot_f
705 # define gmx_simd_andnot_r gmx_simd_andnot_f
707 /*! \brief Bitwise \a or on two \ref gmx_simd_real_t.
709 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_d,
710 * otherwise \ref gmx_simd_or_f.
712 * \copydetails gmx_simd_or_f
714 # define gmx_simd_or_r gmx_simd_or_f
716 /*! \brief Bitwise \a exclusive-or on two \ref gmx_simd_real_t.
718 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_d,
719 * otherwise \ref gmx_simd_xor_f.
721 * \copydetails gmx_simd_xor_f
723 # define gmx_simd_xor_r gmx_simd_xor_f
726 * \name SIMD floating-point arithmetic operations on gmx_simd_real_t
730 /*! \brief SIMD a+b for two \ref gmx_simd_real_t.
732 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_d,
733 * otherwise \ref gmx_simd_add_f.
735 * \copydetails gmx_simd_add_f
737 # define gmx_simd_add_r gmx_simd_add_f
739 /*! \brief SIMD a-b for two \ref gmx_simd_real_t.
741 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_d,
742 * otherwise \ref gmx_simd_sub_f.
744 * \copydetails gmx_simd_sub_f
746 # define gmx_simd_sub_r gmx_simd_sub_f
748 /*! \brief SIMD a*b for two \ref gmx_simd_real_t.
750 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_d,
751 * otherwise \ref gmx_simd_mul_f.
753 * \copydetails gmx_simd_mul_f
755 # define gmx_simd_mul_r gmx_simd_mul_f
757 /*! \brief SIMD a*b+c for three \ref gmx_simd_real_t.
759 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmadd_d,
760 * otherwise \ref gmx_simd_fmadd_f.
762 * \copydetails gmx_simd_fmadd_f
764 # define gmx_simd_fmadd_r gmx_simd_fmadd_f
766 /*! \brief SIMD a*b-c for three \ref gmx_simd_real_t.
768 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmsub_d,
769 * otherwise \ref gmx_simd_fmsub_f.
771 * \copydetails gmx_simd_fmsub_f
773 # define gmx_simd_fmsub_r gmx_simd_fmsub_f
775 /*! \brief SIMD -a*b+c for three \ref gmx_simd_real_t.
777 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmadd_d,
778 * otherwise \ref gmx_simd_fnmadd_f.
780 * \copydetails gmx_simd_fnmadd_f
782 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_f
784 /*! \brief SIMD -a*b-c for three \ref gmx_simd_real_t.
786 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmsub_d,
787 * otherwise \ref gmx_simd_fnmsub_f.
789 * \copydetails gmx_simd_fnmsub_f
791 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_f
793 /*! \brief SIMD table lookup for 1/sqrt(x) approximation.
795 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rsqrt_d,
796 * otherwise \ref gmx_simd_rsqrt_f.
798 * \copydetails gmx_simd_rsqrt_f
800 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_f
802 /*! \brief SIMD table lookup for 1/x approximation.
804 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rcp_d,
805 * otherwise \ref gmx_simd_rcp_f.
807 * \copydetails gmx_simd_rcp_f
809 # define gmx_simd_rcp_r gmx_simd_rcp_f
811 /*! \brief SIMD fabs(x) for \ref gmx_simd_real_t.
813 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fabs_d,
814 * otherwise \ref gmx_simd_fabs_f.
816 * \copydetails gmx_simd_fabs_f
818 # define gmx_simd_fabs_r gmx_simd_fabs_f
820 /*! \brief SIMD -x for \ref gmx_simd_real_t.
822 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fneg_d,
823 * otherwise \ref gmx_simd_fneg_f.
825 * \copydetails gmx_simd_fneg_f
827 # define gmx_simd_fneg_r gmx_simd_fneg_f
829 /*! \brief SIMD max(a,b) for each element in \ref gmx_simd_real_t.
831 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_max_d,
832 * otherwise \ref gmx_simd_max_f.
834 * \copydetails gmx_simd_max_f
836 # define gmx_simd_max_r gmx_simd_max_f
838 /*! \brief SIMD min(a,b) for each element in \ref gmx_simd_real_t.
840 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_min_d,
841 * otherwise \ref gmx_simd_min_f.
843 * \copydetails gmx_simd_min_f
845 # define gmx_simd_min_r gmx_simd_min_f
847 /*! \brief Round \ref gmx_simd_real_t to nearest int, return \ref gmx_simd_real_t.
849 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_round_d,
850 * otherwise \ref gmx_simd_round_f.
852 * \copydetails gmx_simd_round_f
854 # define gmx_simd_round_r gmx_simd_round_f
856 /*! \brief Truncate \ref gmx_simd_real_t towards 0, return \ref gmx_simd_real_t.
858 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_trunc_d,
859 * otherwise \ref gmx_simd_trunc_f.
861 * \copydetails gmx_simd_trunc_f
863 # define gmx_simd_trunc_r gmx_simd_trunc_f
865 /*! \brief SIMD Fraction, i.e. x-trunc(x) for \ref gmx_simd_real_t.
867 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fraction_d,
868 * otherwise \ref gmx_simd_fraction_f.
870 * \copydetails gmx_simd_fraction_f
872 # define gmx_simd_fraction_r gmx_simd_fraction_f
874 /*! \brief Return the FP exponent of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
876 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_exponent_d,
877 * otherwise \ref gmx_simd_get_exponent_f.
879 * \copydetails gmx_simd_exponent_f
881 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_f
883 /*! \brief Return the FP mantissa of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
885 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_mantissa_d,
886 * otherwise \ref gmx_simd_get_mantissa_f.
888 * \copydetails gmx_simd_mantissa_f
890 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_f
892 /*! \brief Set the exponent of a SIMD \ref gmx_simd_real_t from a \ref gmx_simd_real_t.
894 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set_exponent_d,
895 * otherwise \ref gmx_simd_set_exponent_f.
897 * \copydetails gmx_simd_set_exponent_f
899 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_f
902 * \name SIMD comparison, boolean, and select operations for gmx_simd_real_t
906 /*! \brief SIMD a==b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
908 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_d,
909 * otherwise \ref gmx_simd_cmpeq_f.
911 * \copydetails gmx_simd_cmpeq_f
913 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_f
915 /*! \brief SIMD a<b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
917 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_d,
918 * otherwise \ref gmx_simd_cmplt_f.
920 * \copydetails gmx_simd_cmplt_f
922 # define gmx_simd_cmplt_r gmx_simd_cmplt_f
924 /*! \brief SIMD a<=b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
926 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmple_d,
927 * otherwise \ref gmx_simd_cmple_f.
929 * \copydetails gmx_simd_cmple_f
931 # define gmx_simd_cmple_r gmx_simd_cmple_f
933 /*! \brief For each element, the result boolean is true if both arguments are true
935 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_db,
936 * otherwise \ref gmx_simd_and_fb.
938 * \copydetails gmx_simd_and_fb
940 # define gmx_simd_and_b gmx_simd_and_fb
942 /*! \brief For each element, the result boolean is true if either argument is true
944 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_db,
945 * otherwise \ref gmx_simd_or_fb.
947 * \copydetails gmx_simd_or_fn
949 # define gmx_simd_or_b gmx_simd_or_fb
951 /*! \brief Return nonzero if any element in gmx_simd_bool_t is true, otherwise 0.
953 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_db,
954 * otherwise \ref gmx_simd_anytrue_fb.
956 * \copydetails gmx_simd_anytrue_fb
958 # define gmx_simd_anytrue_b gmx_simd_anytrue_fb
960 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is true, otherwise 0.
962 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_d,
963 * otherwise \ref gmx_simd_blendzero_f.
965 * \copydetails gmx_simd_blendzero_f
967 * \sa gmx_simd_blendzero_i
969 # define gmx_simd_blendzero_r gmx_simd_blendzero_f
971 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is false, otherwise 0.
973 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_d,
974 * otherwise \ref gmx_simd_blendnotzero_f.
976 * \copydetails gmx_simd_blendnotzero_f
978 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_f
980 /*! \brief Selects from 2nd real SIMD arg where boolean is true, otherwise 1st arg.
982 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_d,
983 * otherwise \ref gmx_simd_blendv_f.
985 * \copydetails gmx_simd_blendv_f
987 # define gmx_simd_blendv_r gmx_simd_blendv_f
989 /*! \brief Return sum of all elements in SIMD floating-point variable.
991 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_reduce_d,
992 * otherwise \ref gmx_simd_reduce_f.
994 * \copydetails gmx_simd_reduce_f
996 # define gmx_simd_reduce_r gmx_simd_reduce_f
999 * \name SIMD integer logical operations on gmx_simd_int32_t
1001 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_LOGICAL is defined.
1005 /*! \brief Shift each element in \ref gmx_simd_int32_t left by immediate
1007 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_slli_di,
1008 * otherwise \ref gmx_simd_slli_fi.
1010 * \copydetails gmx_simd_slli_fi
1012 # define gmx_simd_slli_i gmx_simd_slli_fi
1014 /*! \brief Shift each element in \ref gmx_simd_int32_t right by immediate
1016 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_srli_di,
1017 * otherwise \ref gmx_simd_srli_fi.
1019 * \copydetails gmx_simd_srli_fi
1021 # define gmx_simd_srli_i gmx_simd_srli_fi
1023 /*! \brief Bitwise \a and on two \ref gmx_simd_int32_t.
1025 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_di,
1026 * otherwise \ref gmx_simd_and_fi.
1028 * \copydetails gmx_simd_and_fi
1030 # define gmx_simd_and_i gmx_simd_and_fi
1032 /*! \brief Bitwise \a and-not on two \ref gmx_simd_int32_t; 1st arg is complemented.
1034 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_di,
1035 * otherwise \ref gmx_simd_andnot_fi.
1037 * \copydetails gmx_simd_andnot_fi
1039 # define gmx_simd_andnot_i gmx_simd_andnot_fi
1041 /*! \brief Bitwise \a or on two \ref gmx_simd_int32_t.
1043 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_di,
1044 * otherwise \ref gmx_simd_or_fi.
1046 * \copydetails gmx_simd_or_fi
1048 # define gmx_simd_or_i gmx_simd_or_fi
1050 /*! \brief Bitwise \a xor on two \ref gmx_simd_int32_t.
1052 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_di,
1053 * otherwise \ref gmx_simd_xor_fi.
1055 * \copydetails gmx_simd_xor_fi
1057 # define gmx_simd_xor_i gmx_simd_xor_fi
1060 * \name SIMD integer arithmetic operations on gmx_simd_int32_t
1062 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1066 /*! \brief SIMD a+b for two \ref gmx_simd_int32_t.
1068 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_di,
1069 * otherwise \ref gmx_simd_add_fi.
1071 * \copydetails gmx_simd_add_fi
1073 # define gmx_simd_add_i gmx_simd_add_fi
1075 /*! \brief SIMD a-b for two \ref gmx_simd_int32_t.
1077 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_di,
1078 * otherwise \ref gmx_simd_sub_fi.
1080 * \copydetails gmx_simd_sub_fi
1082 # define gmx_simd_sub_i gmx_simd_sub_fi
1084 /*! \brief SIMD a*b for two \ref gmx_simd_int32_t.
1086 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_di,
1087 * otherwise \ref gmx_simd_mul_fi.
1089 * \copydetails gmx_simd_mul_fi
1091 # define gmx_simd_mul_i gmx_simd_mul_fi
1094 * \name SIMD integer comparison, booleans, and selection on gmx_simd_int32_t
1096 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1100 /*! \brief Returns boolean describing whether a==b, for \ref gmx_simd_int32_t
1102 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_di,
1103 * otherwise \ref gmx_simd_cmpeq_fi.
1105 * \copydetails gmx_simd_cmpeq_fi
1107 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_fi
1109 /*! \brief Returns boolean describing whether a<b, for \ref gmx_simd_int32_t
1111 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_di,
1112 * otherwise \ref gmx_simd_cmplt_fi.
1114 * \copydetails gmx_simd_cmplt_fi
1116 # define gmx_simd_cmplt_i gmx_simd_cmplt_fi
1118 /*! \brief For each element, the result boolean is true if both arguments are true
1120 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_dib,
1121 * otherwise \ref gmx_simd_and_fib.
1123 * \copydetails gmx_simd_and_fib
1125 # define gmx_simd_and_ib gmx_simd_and_fib
1127 /*! \brief For each element, the result boolean is true if either argument is true.
1129 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_dib,
1130 * otherwise \ref gmx_simd_or_fib.
1132 * \copydetails gmx_simd_or_fib
1134 # define gmx_simd_or_ib gmx_simd_or_fib
1136 /*! \brief Return nonzero if any element in gmx_simd_ibool_t is true, otherwise 0.
1138 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_dib,
1139 * otherwise \ref gmx_simd_anytrue_fib.
1141 * \copydetails gmx_simd_anytrue_fib
1143 # define gmx_simd_anytrue_ib gmx_simd_anytrue_fib
1145 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is true, otherwise 0.
1147 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_di,
1148 * otherwise \ref gmx_simd_blendzero_fi.
1150 * \copydetails gmx_simd_blendzero_fi
1152 # define gmx_simd_blendzero_i gmx_simd_blendzero_fi
1154 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is false, otherwise 0.
1156 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_di,
1157 * otherwise \ref gmx_simd_blendnotzero_fi.
1159 * \copydetails gmx_simd_blendnotzero_fi
1161 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_fi
1163 /*! \brief Selects from 2nd int SIMD arg where boolean is true, otherwise 1st arg.
1165 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_di,
1166 * otherwise \ref gmx_simd_blendv_fi.
1168 * \copydetails gmx_simd_blendv_fi
1170 # define gmx_simd_blendv_i gmx_simd_blendv_fi
1173 * \name SIMD conversion operations
1175 * These instructions are available when both types involved in the conversion
1176 * are defined, e.g. \ref GMX_SIMD_HAVE_REAL and \ref GMX_SIMD_HAVE_INT32
1177 * for real-to-integer conversion.
1181 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, round to nearest integer.
1183 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_d2i,
1184 * otherwise \ref gmx_simd_cvt_f2i.
1186 * \copydetails gmx_simd_cvt_f2i
1188 # define gmx_simd_cvt_r2i gmx_simd_cvt_f2i
1190 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, truncate towards zero
1192 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvtt_d2i,
1193 * otherwise \ref gmx_simd_cvtt_f2i.
1195 * \copydetails gmx_simd_cvtt_f2i
1197 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_f2i
1199 /*! \brief Convert gmx_simd_int32_t to gmx_simd_real_t
1201 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_i2d,
1202 * otherwise \ref gmx_simd_cvt_i2f.
1204 * \copydetails gmx_simd_cvt_i2f
1206 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2f
1208 /*! \brief Convert from gmx_simd_bool_t to gmx_simd_ibool_t
1210 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_db2dib,
1211 * otherwise \ref gmx_simd_cvt_fb2fib.
1213 * \copydetails gmx_simd_cvt_fb2fib
1215 # define gmx_simd_cvt_b2ib gmx_simd_cvt_fb2fib
1217 /*! \brief Convert from gmx_simd_ibool_t to gmx_simd_bool_t
1219 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_dib2db,
1220 * otherwise \ref gmx_simd_cvt_fib2fb.
1222 * \copydetails gmx_simd_cvt_fib2fb
1224 # define gmx_simd_cvt_ib2b gmx_simd_cvt_fib2fb
1228 * \name SIMD memory alignment operations
1232 /*! \brief Align real memory for SIMD usage.
1234 * This routine will only align memory if \ref GMX_SIMD_HAVE_REAL is defined.
1235 * Otherwise the original pointer will be returned.
1237 * Start by allocating an extra \ref GMX_SIMD_REAL_WIDTH float elements of memory,
1238 * and then call this function. The returned pointer will be greater or equal
1239 * to the one you provided, and point to an address inside your provided memory
1240 * that is aligned to the SIMD width.
1242 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_d,
1243 * otherwise \ref gmx_simd_align_f. For detailed documentation, see the
1244 * precision-specific implementation routines.
1246 # define gmx_simd_align_r gmx_simd_align_f
1248 /*! \brief Align integer memory for SIMD usage.
1250 * This routine will only align memory if \ref GMX_SIMD_HAVE_INT32 is defined.
1251 * Otherwise the original pointer will be returned.
1253 * Start by allocating an extra \ref GMX_SIMD_INT32_WIDTH elements of memory,
1254 * and then call this function. The returned pointer will be greater or equal
1255 * to the one you provided, and point to an address inside your provided memory
1256 * that is aligned to the SIMD width.
1258 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_di,
1259 * otherwise \ref gmx_simd_align_fi. For detailed documentation, see the
1260 * precision-specific implementation routines.
1262 # define gmx_simd_align_i gmx_simd_align_fi
1266 /*! \name SIMD4 - constant width-four SIMD datatypes
1268 * These operations are only meant to be used for a few coordinate
1269 * manipulation and grid interpolation routines, so we only support a subset
1270 * of operations for SIMD4. To avoid repeating all the documentation from
1271 * the generic width SIMD routines, we only provide brief documentation for
1272 * these operations. Follow the link to the implementation documentation or the
1273 * reference to the corresponding generic SIMD routine. The format will be
1274 * exactly the same, but they have SIMD replaced with SIMD4.
1278 /*! \brief SIMD real datatype guaranteed to be 4 elements wide, if available.
1280 * All the SIMD4 datatypes and operations behave like their counterparts for
1281 * the generic SIMD implementation, but they might be implemented with different
1282 * registers, or not supported at all. It is important that you check the
1283 * define \ref GMX_SIMD4_HAVE_REAL before using it.
1285 * Just as the normal SIMD operations, all SIMD4 types and routines will
1286 * be aliased to either single or double precision ones based on whether
1287 * GMX_DOUBLE is defined.
1289 * \note There is no support for integer or math operations in SIMD4.
1291 # define gmx_simd4_real_t gmx_simd4_float_t
1293 /*! \brief Boolean for \ref gmx_simd4_real_t comparision/selection */
1294 # define gmx_simd4_bool_t gmx_simd4_fbool_t
1296 /*! \brief Load aligned data to gmx_simd4_real_t.
1298 * \copydetails gmx_simd4_load_f
1300 # define gmx_simd4_load_r gmx_simd4_load_f
1302 /*! \brief Load single element to gmx_simd4_real_t
1304 * \copydetails gmx_simd4_load1_f
1306 # define gmx_simd4_load1_r gmx_simd4_load1_f
1308 /*! \brief Set gmx_simd4_real_t from scalar value
1310 * \copydetails gmx_simd4_set1_f
1312 # define gmx_simd4_set1_r gmx_simd4_set1_f
1314 /*! \brief store aligned data from gmx_simd4_real_t
1316 * \copydetails gmx_simd4_store_f
1318 # define gmx_simd4_store_r gmx_simd4_store_f
1320 /*! \brief Load unaligned data to gmx_simd4_real_t
1322 * \copydetails gmx_simd4_loadu_f
1324 # define gmx_simd4_loadu_r gmx_simd4_loadu_f
1326 /*! \brief Store unaligned data from gmx_simd4_real_t
1328 * \copydetails gmx_simd4_storeu_f
1330 # define gmx_simd4_storeu_r gmx_simd4_storeu_f
1332 /*! \brief Set all elements in gmx_simd4_real_t to 0.0
1334 * \copydetails gmx_simd4_setzero_f
1336 # define gmx_simd4_setzero_r gmx_simd4_setzero_f
1338 /*! \brief Bitwise and for two gmx_simd4_real_t
1340 * \copydetails gmx_simd4_and_f
1342 # define gmx_simd4_and_r gmx_simd4_and_f
1344 /*! \brief Bitwise and-not for two gmx_simd4_real_t. 1st arg is complemented.
1346 * \copydetails gmx_simd4_andnot_f
1348 # define gmx_simd4_andnot_r gmx_simd4_andnot_f
1350 /*! \brief Bitwise or for two gmx_simd4_real_t
1352 * \copydetails gmx_simd4_or_f
1354 # define gmx_simd4_or_r gmx_simd4_or_f
1356 /*! \brief Bitwise xor for two gmx_simd4_real_t
1358 * \copydetails gmx_simd4_xor_f
1360 # define gmx_simd4_xor_r gmx_simd4_xor_f
1362 /*! \brief a+b for \ref gmx_simd4_real_t
1364 * \copydetails gmx_simd4_add_f
1366 # define gmx_simd4_add_r gmx_simd4_add_f
1368 /*! \brief a-b for \ref gmx_simd4_real_t
1370 * \copydetails gmx_simd4_sub_f
1372 # define gmx_simd4_sub_r gmx_simd4_sub_f
1374 /*! \brief a*b for \ref gmx_simd4_real_t
1376 * \copydetails gmx_simd4_mul_f
1378 # define gmx_simd4_mul_r gmx_simd4_mul_f
1380 /*! \brief a*b+c for \ref gmx_simd4_real_t
1382 * \copydetails gmx_simd4_fmadd_f
1384 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_f
1386 /*! \brief a*b-c for \ref gmx_simd4_real_t
1388 * \copydetails gmx_simd4_fmsub_f
1390 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_f
1392 /*! \brief -a*b+c for \ref gmx_simd4_real_t
1394 * \copydetails gmx_simd4_fnmadd_f
1396 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_f
1398 /*! \brief -a*b-c for \ref gmx_simd4_real_t
1400 * \copydetails gmx_simd4_fnmsub_f
1402 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_f
1404 /*! \brief 1/sqrt(x) approximate lookup for \ref gmx_simd4_real_t
1406 * \copydetails gmx_simd4_rsqrt_f
1408 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_f
1410 /*! \brief fabs(x) for \ref gmx_simd4_real_t
1412 * \copydetails gmx_simd4_fabs_f
1414 # define gmx_simd4_fabs_r gmx_simd4_fabs_f
1416 /*! \brief Change sign (-x) for \ref gmx_simd4_real_t
1418 * \copydetails gmx_simd4_fneg_f
1420 # define gmx_simd4_fneg_r gmx_simd4_fneg_f
1422 /*! \brief Select maximum of each pair of elements from args for \ref gmx_simd4_real_t
1424 * \copydetails gmx_simd4_max_f
1426 # define gmx_simd4_max_r gmx_simd4_max_f
1428 /*! \brief Select minimum of each pair of elements from args for \ref gmx_simd4_real_t
1430 * \copydetails gmx_simd4_min_f
1432 # define gmx_simd4_min_r gmx_simd4_min_f
1434 /*! \brief Round \ref gmx_simd4_real_t to nearest integer, return \ref gmx_simd4_real_t
1436 * \copydetails gmx_simd4_round_f
1438 # define gmx_simd4_round_r gmx_simd4_round_f
1440 /*! \brief Truncate \ref gmx_simd4_real_t towards zero, return \ref gmx_simd4_real_t
1442 * \copydetails gmx_simd4_trunc_f
1444 # define gmx_simd4_trunc_r gmx_simd4_trunc_f
1446 /*! \brief Scalar product of first three elements of two \ref gmx_simd4_real_t *
1448 * \copydetails gmx_simd4_dotproduct3_f
1450 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_f
1452 /*! \brief Return booleans whether a==b for each element two \ref gmx_simd4_real_t
1454 * \copydetails gmx_simd4_cmpeq_f
1456 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_f
1457 /*! \brief Return booleans whether a<b for each element two \ref gmx_simd4_real_t
1459 * \copydetails gmx_simd4_cmplt_f
1461 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_f
1462 /*! \brief Return booleans whether a<=b for each element two \ref gmx_simd4_real_t
1464 * \copydetails gmx_simd4_cmple_f
1466 # define gmx_simd4_cmple_r gmx_simd4_cmple_f
1468 /*! \brief Logical and for two \ref gmx_simd4_bool_t
1470 * \copydetails gmx_simd4_and_fb
1472 # define gmx_simd4_and_b gmx_simd4_and_fb
1473 /*! \brief Logical or for two \ref gmx_simd4_bool_t
1475 * \copydetails gmx_simd4_or_fb
1477 # define gmx_simd4_or_b gmx_simd4_or_fb
1479 /*! \brief Return nonzero if any element in \ref gmx_simd4_bool_t is true, otherwise 0
1481 * \copydetails gmx_simd4_anytrue_fb
1483 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_fb
1485 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1487 * \copydetails gmx_simd4_blendzero_f
1489 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_f
1491 /*! \brief Selects from 2nd real SIMD4 arg where boolean is false, otherwise 1st arg
1493 * \copydetails gmx_simd4_blendnotzero_f
1495 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_f
1497 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1499 * \copydetails gmx_simd4_blendv_f
1501 # define gmx_simd4_blendv_r gmx_simd4_blendv_f
1503 /*! \brief Return sum of all elements in SIMD4 floating-point variable.
1505 * \copydetails gmx_simd4_reduce_f
1507 # define gmx_simd4_reduce_r gmx_simd4_reduce_f
1509 /*! \brief Align real memory for SIMD4 usage.
1511 * \copydetails gmx_simd4_align_f
1513 # define gmx_simd4_align_r gmx_simd4_align_f
1517 /*! \name SIMD predefined macros to describe high-level capabilities
1521 # if (defined GMX_SIMD_HAVE_FLOAT) || (defined DOXYGEN)
1522 /*! \brief Defined if gmx_simd_real_t is available.
1524 * if GMX_DOUBLE is defined, this will be aliased to
1525 * \ref GMX_SIMD_HAVE_DOUBLE, otherwise GMX_SIMD_HAVE_FLOAT.
1527 # define GMX_SIMD_HAVE_REAL
1528 /*! \brief Width of gmx_simd_real_t.
1530 * if GMX_DOUBLE is defined, this will be aliased to
1531 * \ref GMX_SIMD_DOUBLE_WIDTH, otherwise GMX_SIMD_FLOAT_WIDTH.
1533 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_FLOAT_WIDTH
1535 # if (defined GMX_SIMD_HAVE_FINT32) || (defined DOXYGEN)
1536 /*! \brief Defined if gmx_simd_int32_t is available.
1538 * if GMX_DOUBLE is defined, this will be aliased to
1539 * \ref GMX_SIMD_HAVE_DINT32, otherwise GMX_SIMD_HAVE_FINT32.
1541 # define GMX_SIMD_HAVE_INT32
1542 /*! \brief Width of gmx_simd_int32_t.
1544 * if GMX_DOUBLE is defined, this will be aliased to
1545 * \ref GMX_SIMD_DINT32_WIDTH, otherwise GMX_SIMD_FINT32_WIDTH.
1547 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_FINT32_WIDTH
1549 # if (defined GMX_SIMD_HAVE_FINT32_EXTRACT) || (defined DOXYGEN)
1550 /*! \brief Defined if gmx_simd_extract_i() is available.
1552 * if GMX_DOUBLE is defined, this will be aliased to
1553 * \ref GMX_SIMD_HAVE_DINT32_EXTRACT, otherwise GMX_SIMD_HAVE_FINT32_EXTRACT.
1555 # define GMX_SIMD_HAVE_INT32_EXTRACT
1557 # if (defined GMX_SIMD_HAVE_FINT32_LOGICAL) || (defined DOXYGEN)
1558 /*! \brief Defined if logical ops are supported on gmx_simd_int32_t.
1560 * if GMX_DOUBLE is defined, this will be aliased to
1561 * \ref GMX_SIMD_HAVE_DINT32_LOGICAL, otherwise GMX_SIMD_HAVE_FINT32_LOGICAL.
1563 # define GMX_SIMD_HAVE_INT32_LOGICAL
1565 # if (defined GMX_SIMD_HAVE_FINT32_ARITHMETICS) || (defined DOXYGEN)
1566 /*! \brief Defined if arithmetic ops are supported on gmx_simd_int32_t.
1568 * if GMX_DOUBLE is defined, this will be aliased to
1569 * \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS, otherwise GMX_SIMD_HAVE_FINT32_ARITHMETICS.
1571 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
1573 # if (defined GMX_SIMD4_HAVE_FLOAT) || (defined DOXYGEN)
1574 /*! \brief Defined if gmx_simd4_real_t is available.
1576 * if GMX_DOUBLE is defined, this will be aliased to
1577 * \ref GMX_SIMD4_HAVE_DOUBLE, otherwise GMX_SIMD4_HAVE_FLOAT.
1579 # define GMX_SIMD4_HAVE_REAL
1584 #endif /* GMX_DOUBLE */
1589 #endif /* GMX_SIMD_SIMD_H */