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_IBM_VMX
133 # include "impl_ibm_vmx/impl_ibm_vmx.h"
134 #elif defined GMX_SIMD_IBM_VSX
135 # include "impl_ibm_vsx/impl_ibm_vsx.h"
136 #elif defined GMX_SIMD_SPARC64_HPC_ACE
137 # include "impl_sparc64_hpc_ace/impl_sparc64_hpc_ace.h"
138 #elif (defined GMX_SIMD_REFERENCE) || (defined DOXYGEN)
139 /* Plain C SIMD reference implementation, also serves as documentation.
140 * For now this code path will also be taken for Sparc64_HPC_ACE since we have
141 * not yet added the verlet kernel extensions there. The group kernels do not
142 * depend on this file, so they will still be accelerated with SIMD.
144 # include "impl_reference/impl_reference.h"
146 /* Turn off the GMX_SIMD flag if we do not even have reference support */
151 * SIMD4 width is always 4, but use this for clarity in definitions.
153 * It improves code readability to allocate e.g. 2*GMX_SIMD4_WIDTH instead of 8.
155 #define GMX_SIMD4_WIDTH 4
159 /*! \name SIMD memory alignment operations
164 * Align a float pointer for usage with SIMD instructions.
166 * You should typically \a not call this function directly (unless you explicitly
167 * want single precision even when GMX_DOUBLE is set), but use the
168 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
170 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FLOAT_WIDTH extra elements.
172 * \return Aligned pointer (>=p) suitable for loading/storing float fp SIMD.
173 * If \ref GMX_SIMD_HAVE_FLOAT is not set, p will be returned unchanged.
175 * Start by allocating an extra \ref GMX_SIMD_FLOAT_WIDTH float elements of memory,
176 * and then call this function. The returned pointer will be greater or equal
177 * to the one you provided, and point to an address inside your provided memory
178 * that is aligned to the SIMD width.
180 static gmx_inline float *
181 gmx_simd_align_f(float *p)
183 # ifdef GMX_SIMD_HAVE_FLOAT
184 return (float *)(((size_t)((p)+GMX_SIMD_FLOAT_WIDTH-1)) & (~((size_t)(GMX_SIMD_FLOAT_WIDTH*sizeof(float)-1))));
191 * Align a double pointer for usage with SIMD instructions.
193 * You should typically \a not call this function directly (unless you explicitly
194 * want double precision even when GMX_DOUBLE is not set), but use the
195 * \ref gmx_simd_align_r macro to align memory in default Gromacs real precision.
197 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DOUBLE_WIDTH extra elements.
199 * \return Aligned pointer (>=p) suitable for loading/storing double fp SIMD.
200 * If \ref GMX_SIMD_HAVE_DOUBLE is not set, p will be returned unchanged.
202 * Start by allocating an extra \ref GMX_SIMD_DOUBLE_WIDTH double elements of memory,
203 * and then call this function. The returned pointer will be greater or equal
204 * to the one you provided, and point to an address inside your provided memory
205 * that is aligned to the SIMD width.
207 static gmx_inline double *
208 gmx_simd_align_d(double *p)
210 # ifdef GMX_SIMD_HAVE_DOUBLE
211 return (double *)(((size_t)((p)+GMX_SIMD_DOUBLE_WIDTH-1)) & (~((size_t)(GMX_SIMD_DOUBLE_WIDTH*sizeof(double)-1))));
218 * Align a (float) integer pointer for usage with SIMD instructions.
220 * You should typically \a not call this function directly (unless you explicitly
221 * want integers corresponding to single precision even when GMX_DOUBLE is
222 * set), but use the \ref gmx_simd_align_i macro to align integer memory
223 * corresponding to Gromacs default floating-point precision.
225 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_FINT32_WIDTH extra elements.
227 * \return Aligned pointer (>=p) suitable for loading/storing float-integer SIMD.
228 * If \ref GMX_SIMD_HAVE_FINT32 is not set, p will be returned unchanged.
230 * This routine provides aligned memory for usage with \ref gmx_simd_fint32_t. You
231 * should have allocated an extra \ref GMX_SIMD_FINT32_WIDTH * sizeof(int) bytes. The
232 * reason why we need to separate float-integer vs. double-integer is that the
233 * width of registers after conversions from the floating-point types might not
234 * be identical, or even supported, in both cases.
236 static gmx_inline int *
237 gmx_simd_align_fi(int *p)
239 # ifdef GMX_SIMD_HAVE_FINT32
240 return (int *)(((size_t)((p)+GMX_SIMD_FINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_FINT32_WIDTH*sizeof(int)-1))));
247 * Align a (double) integer pointer for usage with SIMD instructions.
249 * You should typically \a not call this function directly (unless you explicitly
250 * want integers corresponding to doublele precision even when GMX_DOUBLE is
251 * not set), but use the \ref gmx_simd_align_i macro to align integer memory
252 * corresponding to Gromacs default floating-point precision.
254 * \param p Pointer to memory, allocate at least \ref GMX_SIMD_DINT32_WIDTH extra elements.
256 * \return Aligned pointer (>=p) suitable for loading/storing double-integer SIMD.
257 * If \ref GMX_SIMD_HAVE_DINT32 is not set, p will be returned unchanged.
259 * This routine provides aligned memory for usage with \ref gmx_simd_dint32_t. You
260 * should have allocated an extra \ref GMX_SIMD_DINT32_WIDTH*sizeof(int) bytes. The
261 * reason why we need to separate float-integer vs. double-integer is that the
262 * width of registers after conversions from the floating-point types might not
263 * be identical, or even supported, in both cases.
265 static gmx_inline int *
266 gmx_simd_align_di(int *p)
268 # ifdef GMX_SIMD_HAVE_DINT32
269 return (int *)(((size_t)((p)+GMX_SIMD_DINT32_WIDTH-1)) & (~((size_t)(GMX_SIMD_DINT32_WIDTH*sizeof(int)-1))));
276 * Align a float pointer for usage with SIMD4 instructions.
278 * You should typically \a not call this function directly (unless you explicitly
279 * want single precision even when GMX_DOUBLE is set), but use the
280 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
282 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
284 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
285 * If \ref GMX_SIMD4_HAVE_FLOAT is not set, p will be returned unchanged.
287 * This routine provides aligned memory for usage with \ref gmx_simd4_float_t.
288 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(float) bytes.
290 static gmx_inline float *
291 gmx_simd4_align_f(float *p)
293 # ifdef GMX_SIMD4_HAVE_FLOAT
294 return (float *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(float)-1))));
301 * Align a double pointer for usage with SIMD4 instructions.
303 * You should typically \a not call this function directly (unless you explicitly
304 * want double precision even when GMX_DOUBLE is not set), but use the
305 * \ref gmx_simd4_align_r macro to align memory in default Gromacs real precision.
307 * \param p Pointer to memory, allocate at least \ref GMX_SIMD4_WIDTH extra elements.
309 * \return Aligned pointer (>=p) suitable for loading/storing float SIMD.
310 * If \ref GMX_SIMD4_HAVE_DOUBLE is not set, p will be returned unchanged.
312 * This routine provides aligned memory for usage with \ref gmx_simd4_double_t.
313 * should have allocated an extra \ref GMX_SIMD4_WIDTH * sizeof(double) bytes.
315 static gmx_inline double *
316 gmx_simd4_align_d(double *p)
318 # ifdef GMX_SIMD4_HAVE_DOUBLE
319 return (double *)(((size_t)((p)+GMX_SIMD4_WIDTH-1)) & (~((size_t)(GMX_SIMD4_WIDTH*sizeof(double)-1))));
328 /* Define Gromacs "real" precision macros depending on Gromacs config. Note
329 * that conversions float-to-double and v.v. are not included here since they
330 * are not precision-dependent - find them in the implementation files.
333 /* Double floating-point. The documentation is in the float part below */
334 # define gmx_simd_real_t gmx_simd_double_t
335 # define gmx_simd_load_r gmx_simd_load_d
336 # define gmx_simd_load1_r gmx_simd_load1_d
337 # define gmx_simd_set1_r gmx_simd_set1_d
338 # define gmx_simd_store_r gmx_simd_store_d
339 # define gmx_simd_loadu_r gmx_simd_loadu_d
340 # define gmx_simd_storeu_r gmx_simd_storeu_d
341 # define gmx_simd_setzero_r gmx_simd_setzero_d
342 # define gmx_simd_add_r gmx_simd_add_d
343 # define gmx_simd_sub_r gmx_simd_sub_d
344 # define gmx_simd_mul_r gmx_simd_mul_d
345 # define gmx_simd_fmadd_r gmx_simd_fmadd_d
346 # define gmx_simd_fmsub_r gmx_simd_fmsub_d
347 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_d
348 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_d
349 # define gmx_simd_and_r gmx_simd_and_d
350 # define gmx_simd_andnot_r gmx_simd_andnot_d
351 # define gmx_simd_or_r gmx_simd_or_d
352 # define gmx_simd_xor_r gmx_simd_xor_d
353 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_d
354 # define gmx_simd_rcp_r gmx_simd_rcp_d
355 # define gmx_simd_fabs_r gmx_simd_fabs_d
356 # define gmx_simd_fneg_r gmx_simd_fneg_d
357 # define gmx_simd_max_r gmx_simd_max_d
358 # define gmx_simd_min_r gmx_simd_min_d
359 # define gmx_simd_round_r gmx_simd_round_d
360 # define gmx_simd_trunc_r gmx_simd_trunc_d
361 # define gmx_simd_fraction_r gmx_simd_fraction_d
362 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_d
363 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_d
364 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_d
365 /* Double integer and conversions */
366 # define gmx_simd_int32_t gmx_simd_dint32_t
367 # define gmx_simd_load_i gmx_simd_load_di
368 # define gmx_simd_set1_i gmx_simd_set1_di
369 # define gmx_simd_store_i gmx_simd_store_di
370 # define gmx_simd_loadu_i gmx_simd_loadu_di
371 # define gmx_simd_storeu_i gmx_simd_storeu_di
372 # define gmx_simd_setzero_i gmx_simd_setzero_di
373 # define gmx_simd_cvt_r2i gmx_simd_cvt_d2i
374 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_d2i
375 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2d
376 # define gmx_simd_extract_i gmx_simd_extract_di
377 # define gmx_simd_slli_i gmx_simd_slli_di
378 # define gmx_simd_srli_i gmx_simd_srli_di
379 # define gmx_simd_and_i gmx_simd_and_di
380 # define gmx_simd_andnot_i gmx_simd_andnot_di
381 # define gmx_simd_or_i gmx_simd_or_di
382 # define gmx_simd_xor_i gmx_simd_xor_di
383 # define gmx_simd_add_i gmx_simd_add_di
384 # define gmx_simd_sub_i gmx_simd_sub_di
385 # define gmx_simd_mul_i gmx_simd_mul_di
386 /* Double booleans and selection */
387 # define gmx_simd_bool_t gmx_simd_dbool_t
388 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_d
389 # define gmx_simd_cmplt_r gmx_simd_cmplt_d
390 # define gmx_simd_cmple_r gmx_simd_cmple_d
391 # define gmx_simd_and_b gmx_simd_and_db
392 # define gmx_simd_or_b gmx_simd_or_db
393 # define gmx_simd_anytrue_b gmx_simd_anytrue_db
394 # define gmx_simd_blendzero_r gmx_simd_blendzero_d
395 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_d
396 # define gmx_simd_blendv_r gmx_simd_blendv_d
397 # define gmx_simd_reduce_r gmx_simd_reduce_d
398 # define gmx_simd_ibool_t gmx_simd_dibool_t
399 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_di
400 # define gmx_simd_cmplt_i gmx_simd_cmplt_di
401 # define gmx_simd_and_ib gmx_simd_and_dib
402 # define gmx_simd_or_ib gmx_simd_or_dib
403 # define gmx_simd_anytrue_ib gmx_simd_anytrue_dib
404 # define gmx_simd_blendzero_i gmx_simd_blendzero_di
405 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_di
406 # define gmx_simd_blendv_i gmx_simd_blendv_di
407 /* Conversions between integer and double floating-point booleans */
408 # define gmx_simd_cvt_b2ib gmx_simd_cvt_db2dib
409 # define gmx_simd_cvt_ib2b gmx_simd_cvt_dib2db
411 /* SIMD4 double fp - we only support a subset of SIMD instructions for SIMD4 */
412 # define gmx_simd4_real_t gmx_simd4_double_t
413 # define gmx_simd4_load_r gmx_simd4_load_d
414 # define gmx_simd4_load1_r gmx_simd4_load1_d
415 # define gmx_simd4_set1_r gmx_simd4_set1_d
416 # define gmx_simd4_store_r gmx_simd4_store_d
417 # define gmx_simd4_loadu_r gmx_simd4_loadu_d
418 # define gmx_simd4_storeu_r gmx_simd4_storeu_d
419 # define gmx_simd4_setzero_r gmx_simd4_setzero_d
420 # define gmx_simd4_add_r gmx_simd4_add_d
421 # define gmx_simd4_sub_r gmx_simd4_sub_d
422 # define gmx_simd4_mul_r gmx_simd4_mul_d
423 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_d
424 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_d
425 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_d
426 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_d
427 # define gmx_simd4_and_r gmx_simd4_and_d
428 # define gmx_simd4_andnot_r gmx_simd4_andnot_d
429 # define gmx_simd4_or_r gmx_simd4_or_d
430 # define gmx_simd4_xor_r gmx_simd4_xor_d
431 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_d
432 # define gmx_simd4_fabs_r gmx_simd4_fabs_d
433 # define gmx_simd4_fneg_r gmx_simd4_fneg_d
434 # define gmx_simd4_max_r gmx_simd4_max_d
435 # define gmx_simd4_min_r gmx_simd4_min_d
436 # define gmx_simd4_round_r gmx_simd4_round_d
437 # define gmx_simd4_trunc_r gmx_simd4_trunc_d
438 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_d
439 # define gmx_simd4_bool_t gmx_simd4_dbool_t
440 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_d
441 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_d
442 # define gmx_simd4_cmple_r gmx_simd4_cmple_d
443 # define gmx_simd4_and_b gmx_simd4_and_db
444 # define gmx_simd4_or_b gmx_simd4_or_db
445 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_db
446 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_d
447 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_d
448 # define gmx_simd4_blendv_r gmx_simd4_blendv_d
449 # define gmx_simd4_reduce_r gmx_simd4_reduce_d
451 /* Memory allocation */
452 # define gmx_simd_align_r gmx_simd_align_d
453 # define gmx_simd_align_i gmx_simd_align_di
454 # define gmx_simd4_align_r gmx_simd4_align_d
456 # ifdef GMX_SIMD_HAVE_DOUBLE
457 # define GMX_SIMD_HAVE_REAL
458 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_DOUBLE_WIDTH
460 # ifdef GMX_SIMD_HAVE_DINT32
461 # define GMX_SIMD_HAVE_INT32
462 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_DINT32_WIDTH
464 # ifdef GMX_SIMD_HAVE_DINT32_EXTRACT
465 # define GMX_SIMD_HAVE_INT32_EXTRACT
467 # ifdef GMX_SIMD_HAVE_DINT32_LOGICAL
468 # define GMX_SIMD_HAVE_INT32_LOGICAL
470 # ifdef GMX_SIMD_HAVE_DINT32_ARITHMETICS
471 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
473 # ifdef GMX_SIMD4_HAVE_DOUBLE
474 # define GMX_SIMD4_HAVE_REAL
477 #else /* GMX_DOUBLE */
479 /*! \name SIMD data types
481 * The actual storage of these types is implementation dependent. The
482 * documentation is generated from the reference implementation, but for
483 * normal usage this will likely not be what you are using.
486 /*! \brief Real precision floating-point SIMD datatype.
488 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
490 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_double_t
491 * internally, otherwise \ref gmx_simd_float_t.
493 # define gmx_simd_real_t gmx_simd_float_t
495 /*! \brief 32-bit integer SIMD type.
497 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
499 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dint32_t
500 * internally, otherwise \ref gmx_simd_fint32_t. This might seem a strange
501 * implementation detail, but it is because some SIMD implementations use
502 * different types/widths of integers registers when converting from
503 * double vs. single precision floating point. As long as you just use
504 * this type you will not have to worry about precision.
506 # define gmx_simd_int32_t gmx_simd_fint32_t
508 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_real_t.
510 * This type is only available if \ref GMX_SIMD_HAVE_REAL is defined.
512 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dbool_t
513 * internally, otherwise \ref gmx_simd_fbool_t. This is necessary since some
514 * SIMD implementations use bitpatterns for marking truth, so single-
515 * vs. double precision booleans are not necessarily exchangable.
516 * As long as you just use this type you will not have to worry about precision.
518 * See \ref gmx_simd_ibool_t for an explanation of real vs. integer booleans.
520 # define gmx_simd_bool_t gmx_simd_fbool_t
522 /*! \brief Boolean SIMD type for usage with \ref gmx_simd_int32_t.
524 * This type is only available if \ref GMX_SIMD_HAVE_INT32 is defined.
526 * If GMX_DOUBLE is defined, this will be set to \ref gmx_simd_dibool_t
527 * internally, otherwise \ref gmx_simd_fibool_t. This is necessary since some
528 * SIMD implementations use bitpatterns for marking truth, so single-
529 * vs. double precision booleans are not necessarily exchangable, and while
530 * a double-precision boolean might be represented with a 64-bit mask, the
531 * corresponding integer might only use a 32-bit mask.
533 * We provide conversion routines for these cases, so the only thing you need to
534 * keep in mind is to use \ref gmx_simd_bool_t when working with
535 * \ref gmx_simd_real_t while you pick \ref gmx_simd_ibool_t when working with
536 * \ref gmx_simd_int32_t.
538 * To convert between them, use \ref gmx_simd_cvt_b2ib and \ref gmx_simd_cvt_ib2b.
540 # define gmx_simd_ibool_t gmx_simd_fibool_t
544 * \name SIMD load/store operations on gmx_simd_real_t
546 * \note Unaligned load/stores are only available when
547 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
551 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from aligned memory to \ref gmx_simd_real_t
553 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_d,
554 * otherwise \ref gmx_simd_load_f.
556 * \copydetails gmx_simd_load_f
558 # define gmx_simd_load_r gmx_simd_load_f
560 /*! \brief Set all elements in \ref gmx_simd_real_t from single value in memory.
562 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load1_d,
563 * otherwise \ref gmx_simd_load1_f.
565 * \copydetails gmx_simd_load1_f
567 # define gmx_simd_load1_r gmx_simd_load1_f
569 /*! \brief Set all elements in \ref gmx_simd_real_t from a scalar.
571 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_d,
572 * otherwise \ref gmx_simd_set1_f.
574 * \copydetails gmx_simd_set1_f
576 # define gmx_simd_set1_r gmx_simd_set1_f
578 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to aligned memory.
580 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_d,
581 * otherwise \ref gmx_simd_store_f.
583 * \copydetails gmx_simd_store_f
585 # define gmx_simd_store_r gmx_simd_store_f
587 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_real_t.
589 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_d,
590 * otherwise \ref gmx_simd_loadu_f.
592 * \copydetails gmx_simd_loadu_f
594 # define gmx_simd_loadu_r gmx_simd_loadu_f
596 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_real_t to unaligned memory.
598 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_d,
599 * otherwise \ref gmx_simd_storeu_f.
601 * \copydetails gmx_simd_storeu_f
603 # define gmx_simd_storeu_r gmx_simd_storeu_f
605 /*! \brief Set all elements in \ref gmx_simd_real_t to 0.0.
607 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_setzero_d,
608 * otherwise \ref gmx_simd_setzero_f.
610 * \copydetails gmx_simd_setzero_f
612 # define gmx_simd_setzero_r gmx_simd_setzero_f
615 * \name SIMD load/store operations on gmx_simd_int32_t
617 * \note Unaligned load/stores are only available when
618 * \ref GMX_SIMD_HAVE_LOADU and \ref GMX_SIMD_HAVE_STOREU are set, respectively.
622 /*! \brief Load \ref GMX_SIMD_INT32_WIDTH values from aligned memory to \ref gmx_simd_int32_t .
624 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_load_di ,
625 * otherwise \ref gmx_simd_load_fi .
627 * \copydetails gmx_simd_load_fi
629 # define gmx_simd_load_i gmx_simd_load_fi
631 /*! \brief Set all elements in \ref gmx_simd_int32_t from a single integer.
633 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set1_di ,
634 * otherwise \ref gmx_simd_set1_fi .
636 * \copydetails gmx_simd_set1_fi
638 # define gmx_simd_set1_i gmx_simd_set1_fi
640 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to aligned memory.
642 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_store_di ,
643 * otherwise \ref gmx_simd_store_fi .
645 * \copydetails gmx_simd_store_fi
647 # define gmx_simd_store_i gmx_simd_store_fi
649 /*! \brief Load \ref GMX_SIMD_REAL_WIDTH values from unaligned memory to \ref gmx_simd_int32_t.
651 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_loadu_di ,
652 * otherwise \ref gmx_simd_loadu_fi .
654 * \copydetails gmx_simd_loadu_fi
656 # define gmx_simd_loadu_i gmx_simd_loadu_fi
658 /*! \brief Store \ref GMX_SIMD_REAL_WIDTH values from \ref gmx_simd_int32_t to unaligned memory.
660 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_storeu_di ,
661 * otherwise \ref gmx_simd_storeu_fi .
663 * \copydetails gmx_simd_storeu_fi
665 # define gmx_simd_storeu_i gmx_simd_storeu_fi
667 /*! \brief Extract single integer from \ref gmx_simd_int32_t element.
669 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_extract_di ,
670 * otherwise \ref gmx_simd_extract_fi .
672 * \copydetails gmx_simd_extract_fi
674 # define gmx_simd_extract_i gmx_simd_extract_fi
676 /*! \brief Set all elements in \ref gmx_simd_int32_t to 0.
678 * If GMX_DOUBLE is defined, it will be aliased to \ref gmx_simd_setzero_di ,
679 * otherwise \ref gmx_simd_setzero_fi .
681 * \copydetails gmx_simd_setzero_fi
683 # define gmx_simd_setzero_i gmx_simd_setzero_fi
687 * \name SIMD floating-point logical operations on gmx_simd_real_t
689 * These instructions are available if \ref GMX_SIMD_HAVE_LOGICAL is defined.
693 /*! \brief Bitwise \a and on two \ref gmx_simd_real_t.
695 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_d,
696 * otherwise \ref gmx_simd_and_f.
698 * \copydetails gmx_simd_and_f
700 # define gmx_simd_and_r gmx_simd_and_f
702 /*! \brief Bitwise \a and-not on two \ref gmx_simd_real_t; 1st arg is complemented.
704 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_d,
705 * otherwise \ref gmx_simd_andnot_f.
707 * \copydetails gmx_simd_andnot_f
709 # define gmx_simd_andnot_r gmx_simd_andnot_f
711 /*! \brief Bitwise \a or on two \ref gmx_simd_real_t.
713 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_d,
714 * otherwise \ref gmx_simd_or_f.
716 * \copydetails gmx_simd_or_f
718 # define gmx_simd_or_r gmx_simd_or_f
720 /*! \brief Bitwise \a exclusive-or on two \ref gmx_simd_real_t.
722 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_d,
723 * otherwise \ref gmx_simd_xor_f.
725 * \copydetails gmx_simd_xor_f
727 # define gmx_simd_xor_r gmx_simd_xor_f
730 * \name SIMD floating-point arithmetic operations on gmx_simd_real_t
734 /*! \brief SIMD a+b for two \ref gmx_simd_real_t.
736 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_d,
737 * otherwise \ref gmx_simd_add_f.
739 * \copydetails gmx_simd_add_f
741 # define gmx_simd_add_r gmx_simd_add_f
743 /*! \brief SIMD a-b for two \ref gmx_simd_real_t.
745 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_d,
746 * otherwise \ref gmx_simd_sub_f.
748 * \copydetails gmx_simd_sub_f
750 # define gmx_simd_sub_r gmx_simd_sub_f
752 /*! \brief SIMD a*b for two \ref gmx_simd_real_t.
754 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_d,
755 * otherwise \ref gmx_simd_mul_f.
757 * \copydetails gmx_simd_mul_f
759 # define gmx_simd_mul_r gmx_simd_mul_f
761 /*! \brief SIMD a*b+c for three \ref gmx_simd_real_t.
763 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmadd_d,
764 * otherwise \ref gmx_simd_fmadd_f.
766 * \copydetails gmx_simd_fmadd_f
768 # define gmx_simd_fmadd_r gmx_simd_fmadd_f
770 /*! \brief SIMD a*b-c for three \ref gmx_simd_real_t.
772 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fmsub_d,
773 * otherwise \ref gmx_simd_fmsub_f.
775 * \copydetails gmx_simd_fmsub_f
777 # define gmx_simd_fmsub_r gmx_simd_fmsub_f
779 /*! \brief SIMD -a*b+c for three \ref gmx_simd_real_t.
781 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmadd_d,
782 * otherwise \ref gmx_simd_fnmadd_f.
784 * \copydetails gmx_simd_fnmadd_f
786 # define gmx_simd_fnmadd_r gmx_simd_fnmadd_f
788 /*! \brief SIMD -a*b-c for three \ref gmx_simd_real_t.
790 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fnmsub_d,
791 * otherwise \ref gmx_simd_fnmsub_f.
793 * \copydetails gmx_simd_fnmsub_f
795 # define gmx_simd_fnmsub_r gmx_simd_fnmsub_f
797 /*! \brief SIMD table lookup for 1/sqrt(x) approximation.
799 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rsqrt_d,
800 * otherwise \ref gmx_simd_rsqrt_f.
802 * \copydetails gmx_simd_rsqrt_f
804 # define gmx_simd_rsqrt_r gmx_simd_rsqrt_f
806 /*! \brief SIMD table lookup for 1/x approximation.
808 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_rcp_d,
809 * otherwise \ref gmx_simd_rcp_f.
811 * \copydetails gmx_simd_rcp_f
813 # define gmx_simd_rcp_r gmx_simd_rcp_f
815 /*! \brief SIMD fabs(x) for \ref gmx_simd_real_t.
817 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fabs_d,
818 * otherwise \ref gmx_simd_fabs_f.
820 * \copydetails gmx_simd_fabs_f
822 # define gmx_simd_fabs_r gmx_simd_fabs_f
824 /*! \brief SIMD -x for \ref gmx_simd_real_t.
826 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fneg_d,
827 * otherwise \ref gmx_simd_fneg_f.
829 * \copydetails gmx_simd_fneg_f
831 # define gmx_simd_fneg_r gmx_simd_fneg_f
833 /*! \brief SIMD max(a,b) for each element in \ref gmx_simd_real_t.
835 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_max_d,
836 * otherwise \ref gmx_simd_max_f.
838 * \copydetails gmx_simd_max_f
840 # define gmx_simd_max_r gmx_simd_max_f
842 /*! \brief SIMD min(a,b) for each element in \ref gmx_simd_real_t.
844 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_min_d,
845 * otherwise \ref gmx_simd_min_f.
847 * \copydetails gmx_simd_min_f
849 # define gmx_simd_min_r gmx_simd_min_f
851 /*! \brief Round \ref gmx_simd_real_t to nearest int, return \ref gmx_simd_real_t.
853 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_round_d,
854 * otherwise \ref gmx_simd_round_f.
856 * \copydetails gmx_simd_round_f
858 # define gmx_simd_round_r gmx_simd_round_f
860 /*! \brief Truncate \ref gmx_simd_real_t towards 0, return \ref gmx_simd_real_t.
862 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_trunc_d,
863 * otherwise \ref gmx_simd_trunc_f.
865 * \copydetails gmx_simd_trunc_f
867 # define gmx_simd_trunc_r gmx_simd_trunc_f
869 /*! \brief SIMD Fraction, i.e. x-trunc(x) for \ref gmx_simd_real_t.
871 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_fraction_d,
872 * otherwise \ref gmx_simd_fraction_f.
874 * \copydetails gmx_simd_fraction_f
876 # define gmx_simd_fraction_r gmx_simd_fraction_f
878 /*! \brief Return the FP exponent of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
880 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_exponent_d,
881 * otherwise \ref gmx_simd_get_exponent_f.
883 * \copydetails gmx_simd_exponent_f
885 # define gmx_simd_get_exponent_r gmx_simd_get_exponent_f
887 /*! \brief Return the FP mantissa of a SIMD \ref gmx_simd_real_t as a \ref gmx_simd_real_t.
889 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_get_mantissa_d,
890 * otherwise \ref gmx_simd_get_mantissa_f.
892 * \copydetails gmx_simd_mantissa_f
894 # define gmx_simd_get_mantissa_r gmx_simd_get_mantissa_f
896 /*! \brief Set the exponent of a SIMD \ref gmx_simd_real_t from a \ref gmx_simd_real_t.
898 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_set_exponent_d,
899 * otherwise \ref gmx_simd_set_exponent_f.
901 * \copydetails gmx_simd_set_exponent_f
903 # define gmx_simd_set_exponent_r gmx_simd_set_exponent_f
906 * \name SIMD comparison, boolean, and select operations for gmx_simd_real_t
910 /*! \brief SIMD a==b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
912 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_d,
913 * otherwise \ref gmx_simd_cmpeq_f.
915 * \copydetails gmx_simd_cmpeq_f
917 # define gmx_simd_cmpeq_r gmx_simd_cmpeq_f
919 /*! \brief SIMD a<b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
921 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_d,
922 * otherwise \ref gmx_simd_cmplt_f.
924 * \copydetails gmx_simd_cmplt_f
926 # define gmx_simd_cmplt_r gmx_simd_cmplt_f
928 /*! \brief SIMD a<=b for \ref gmx_simd_real_t. Returns a \ref gmx_simd_bool_t.
930 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmple_d,
931 * otherwise \ref gmx_simd_cmple_f.
933 * \copydetails gmx_simd_cmple_f
935 # define gmx_simd_cmple_r gmx_simd_cmple_f
937 /*! \brief For each element, the result boolean is true if both arguments are true
939 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_db,
940 * otherwise \ref gmx_simd_and_fb.
942 * \copydetails gmx_simd_and_fb
944 # define gmx_simd_and_b gmx_simd_and_fb
946 /*! \brief For each element, the result boolean is true if either argument is true
948 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_db,
949 * otherwise \ref gmx_simd_or_fb.
951 * \copydetails gmx_simd_or_fn
953 # define gmx_simd_or_b gmx_simd_or_fb
955 /*! \brief Return nonzero if any element in gmx_simd_bool_t is true, otherwise 0.
957 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_db,
958 * otherwise \ref gmx_simd_anytrue_fb.
960 * \copydetails gmx_simd_anytrue_fb
962 # define gmx_simd_anytrue_b gmx_simd_anytrue_fb
964 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is true, otherwise 0.
966 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_d,
967 * otherwise \ref gmx_simd_blendzero_f.
969 * \copydetails gmx_simd_blendzero_f
971 * \sa gmx_simd_blendzero_i
973 # define gmx_simd_blendzero_r gmx_simd_blendzero_f
975 /*! \brief Selects elements from \ref gmx_simd_real_t where boolean is false, otherwise 0.
977 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_d,
978 * otherwise \ref gmx_simd_blendnotzero_f.
980 * \copydetails gmx_simd_blendnotzero_f
982 # define gmx_simd_blendnotzero_r gmx_simd_blendnotzero_f
984 /*! \brief Selects from 2nd real SIMD arg where boolean is true, otherwise 1st arg.
986 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_d,
987 * otherwise \ref gmx_simd_blendv_f.
989 * \copydetails gmx_simd_blendv_f
991 # define gmx_simd_blendv_r gmx_simd_blendv_f
993 /*! \brief Return sum of all elements in SIMD floating-point variable.
995 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_reduce_d,
996 * otherwise \ref gmx_simd_reduce_f.
998 * \copydetails gmx_simd_reduce_f
1000 # define gmx_simd_reduce_r gmx_simd_reduce_f
1003 * \name SIMD integer logical operations on gmx_simd_int32_t
1005 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_LOGICAL is defined.
1009 /*! \brief Shift each element in \ref gmx_simd_int32_t left by immediate
1011 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_slli_di,
1012 * otherwise \ref gmx_simd_slli_fi.
1014 * \copydetails gmx_simd_slli_fi
1016 # define gmx_simd_slli_i gmx_simd_slli_fi
1018 /*! \brief Shift each element in \ref gmx_simd_int32_t right by immediate
1020 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_srli_di,
1021 * otherwise \ref gmx_simd_srli_fi.
1023 * \copydetails gmx_simd_srli_fi
1025 # define gmx_simd_srli_i gmx_simd_srli_fi
1027 /*! \brief Bitwise \a and on two \ref gmx_simd_int32_t.
1029 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_di,
1030 * otherwise \ref gmx_simd_and_fi.
1032 * \copydetails gmx_simd_and_fi
1034 # define gmx_simd_and_i gmx_simd_and_fi
1036 /*! \brief Bitwise \a and-not on two \ref gmx_simd_int32_t; 1st arg is complemented.
1038 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_andnot_di,
1039 * otherwise \ref gmx_simd_andnot_fi.
1041 * \copydetails gmx_simd_andnot_fi
1043 # define gmx_simd_andnot_i gmx_simd_andnot_fi
1045 /*! \brief Bitwise \a or on two \ref gmx_simd_int32_t.
1047 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_di,
1048 * otherwise \ref gmx_simd_or_fi.
1050 * \copydetails gmx_simd_or_fi
1052 # define gmx_simd_or_i gmx_simd_or_fi
1054 /*! \brief Bitwise \a xor on two \ref gmx_simd_int32_t.
1056 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_xor_di,
1057 * otherwise \ref gmx_simd_xor_fi.
1059 * \copydetails gmx_simd_xor_fi
1061 # define gmx_simd_xor_i gmx_simd_xor_fi
1064 * \name SIMD integer arithmetic operations on gmx_simd_int32_t
1066 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1070 /*! \brief SIMD a+b for two \ref gmx_simd_int32_t.
1072 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_add_di,
1073 * otherwise \ref gmx_simd_add_fi.
1075 * \copydetails gmx_simd_add_fi
1077 # define gmx_simd_add_i gmx_simd_add_fi
1079 /*! \brief SIMD a-b for two \ref gmx_simd_int32_t.
1081 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_sub_di,
1082 * otherwise \ref gmx_simd_sub_fi.
1084 * \copydetails gmx_simd_sub_fi
1086 # define gmx_simd_sub_i gmx_simd_sub_fi
1088 /*! \brief SIMD a*b for two \ref gmx_simd_int32_t.
1090 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_mul_di,
1091 * otherwise \ref gmx_simd_mul_fi.
1093 * \copydetails gmx_simd_mul_fi
1095 # define gmx_simd_mul_i gmx_simd_mul_fi
1098 * \name SIMD integer comparison, booleans, and selection on gmx_simd_int32_t
1100 * These instructions are available if \ref GMX_SIMD_HAVE_INT32_ARITHMETICS is defined.
1104 /*! \brief Returns boolean describing whether a==b, for \ref gmx_simd_int32_t
1106 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmpeq_di,
1107 * otherwise \ref gmx_simd_cmpeq_fi.
1109 * \copydetails gmx_simd_cmpeq_fi
1111 # define gmx_simd_cmpeq_i gmx_simd_cmpeq_fi
1113 /*! \brief Returns boolean describing whether a<b, for \ref gmx_simd_int32_t
1115 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cmplt_di,
1116 * otherwise \ref gmx_simd_cmplt_fi.
1118 * \copydetails gmx_simd_cmplt_fi
1120 # define gmx_simd_cmplt_i gmx_simd_cmplt_fi
1122 /*! \brief For each element, the result boolean is true if both arguments are true
1124 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_and_dib,
1125 * otherwise \ref gmx_simd_and_fib.
1127 * \copydetails gmx_simd_and_fib
1129 # define gmx_simd_and_ib gmx_simd_and_fib
1131 /*! \brief For each element, the result boolean is true if either argument is true.
1133 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_or_dib,
1134 * otherwise \ref gmx_simd_or_fib.
1136 * \copydetails gmx_simd_or_fib
1138 # define gmx_simd_or_ib gmx_simd_or_fib
1140 /*! \brief Return nonzero if any element in gmx_simd_ibool_t is true, otherwise 0.
1142 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_anytrue_dib,
1143 * otherwise \ref gmx_simd_anytrue_fib.
1145 * \copydetails gmx_simd_anytrue_fib
1147 # define gmx_simd_anytrue_ib gmx_simd_anytrue_fib
1149 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is true, otherwise 0.
1151 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendzero_di,
1152 * otherwise \ref gmx_simd_blendzero_fi.
1154 * \copydetails gmx_simd_blendzero_fi
1156 # define gmx_simd_blendzero_i gmx_simd_blendzero_fi
1158 /*! \brief Selects elements from \ref gmx_simd_int32_t where boolean is false, otherwise 0.
1160 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendnotzero_di,
1161 * otherwise \ref gmx_simd_blendnotzero_fi.
1163 * \copydetails gmx_simd_blendnotzero_fi
1165 # define gmx_simd_blendnotzero_i gmx_simd_blendnotzero_fi
1167 /*! \brief Selects from 2nd int SIMD arg where boolean is true, otherwise 1st arg.
1169 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_blendv_di,
1170 * otherwise \ref gmx_simd_blendv_fi.
1172 * \copydetails gmx_simd_blendv_fi
1174 # define gmx_simd_blendv_i gmx_simd_blendv_fi
1177 * \name SIMD conversion operations
1179 * These instructions are available when both types involved in the conversion
1180 * are defined, e.g. \ref GMX_SIMD_HAVE_REAL and \ref GMX_SIMD_HAVE_INT32
1181 * for real-to-integer conversion.
1185 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, round to nearest integer.
1187 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_d2i,
1188 * otherwise \ref gmx_simd_cvt_f2i.
1190 * \copydetails gmx_simd_cvt_f2i
1192 # define gmx_simd_cvt_r2i gmx_simd_cvt_f2i
1194 /*! \brief Convert gmx_simd_real_t to gmx_simd_int32_t, truncate towards zero
1196 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvtt_d2i,
1197 * otherwise \ref gmx_simd_cvtt_f2i.
1199 * \copydetails gmx_simd_cvtt_f2i
1201 # define gmx_simd_cvtt_r2i gmx_simd_cvtt_f2i
1203 /*! \brief Convert gmx_simd_int32_t to gmx_simd_real_t
1205 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_i2d,
1206 * otherwise \ref gmx_simd_cvt_i2f.
1208 * \copydetails gmx_simd_cvt_i2f
1210 # define gmx_simd_cvt_i2r gmx_simd_cvt_i2f
1212 /*! \brief Convert from gmx_simd_bool_t to gmx_simd_ibool_t
1214 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_db2dib,
1215 * otherwise \ref gmx_simd_cvt_fb2fib.
1217 * \copydetails gmx_simd_cvt_fb2fib
1219 # define gmx_simd_cvt_b2ib gmx_simd_cvt_fb2fib
1221 /*! \brief Convert from gmx_simd_ibool_t to gmx_simd_bool_t
1223 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_cvt_dib2db,
1224 * otherwise \ref gmx_simd_cvt_fib2fb.
1226 * \copydetails gmx_simd_cvt_fib2fb
1228 # define gmx_simd_cvt_ib2b gmx_simd_cvt_fib2fb
1232 * \name SIMD memory alignment operations
1236 /*! \brief Align real memory for SIMD usage.
1238 * This routine will only align memory if \ref GMX_SIMD_HAVE_REAL is defined.
1239 * Otherwise the original pointer will be returned.
1241 * Start by allocating an extra \ref GMX_SIMD_REAL_WIDTH float elements of memory,
1242 * and then call this function. The returned pointer will be greater or equal
1243 * to the one you provided, and point to an address inside your provided memory
1244 * that is aligned to the SIMD width.
1246 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_d,
1247 * otherwise \ref gmx_simd_align_f. For detailed documentation, see the
1248 * precision-specific implementation routines.
1250 # define gmx_simd_align_r gmx_simd_align_f
1252 /*! \brief Align integer memory for SIMD usage.
1254 * This routine will only align memory if \ref GMX_SIMD_HAVE_INT32 is defined.
1255 * Otherwise the original pointer will be returned.
1257 * Start by allocating an extra \ref GMX_SIMD_INT32_WIDTH elements of memory,
1258 * and then call this function. The returned pointer will be greater or equal
1259 * to the one you provided, and point to an address inside your provided memory
1260 * that is aligned to the SIMD width.
1262 * If GMX_DOUBLE is defined, this will be aliased to \ref gmx_simd_align_di,
1263 * otherwise \ref gmx_simd_align_fi. For detailed documentation, see the
1264 * precision-specific implementation routines.
1266 # define gmx_simd_align_i gmx_simd_align_fi
1270 /*! \name SIMD4 - constant width-four SIMD datatypes
1272 * These operations are only meant to be used for a few coordinate
1273 * manipulation and grid interpolation routines, so we only support a subset
1274 * of operations for SIMD4. To avoid repeating all the documentation from
1275 * the generic width SIMD routines, we only provide brief documentation for
1276 * these operations. Follow the link to the implementation documentation or the
1277 * reference to the corresponding generic SIMD routine. The format will be
1278 * exactly the same, but they have SIMD replaced with SIMD4.
1282 /*! \brief SIMD real datatype guaranteed to be 4 elements wide, if available.
1284 * All the SIMD4 datatypes and operations behave like their counterparts for
1285 * the generic SIMD implementation, but they might be implemented with different
1286 * registers, or not supported at all. It is important that you check the
1287 * define \ref GMX_SIMD4_HAVE_REAL before using it.
1289 * Just as the normal SIMD operations, all SIMD4 types and routines will
1290 * be aliased to either single or double precision ones based on whether
1291 * GMX_DOUBLE is defined.
1293 * \note There is no support for integer or math operations in SIMD4.
1295 # define gmx_simd4_real_t gmx_simd4_float_t
1297 /*! \brief Boolean for \ref gmx_simd4_real_t comparision/selection */
1298 # define gmx_simd4_bool_t gmx_simd4_fbool_t
1300 /*! \brief Load aligned data to gmx_simd4_real_t.
1302 * \copydetails gmx_simd4_load_f
1304 # define gmx_simd4_load_r gmx_simd4_load_f
1306 /*! \brief Load single element to gmx_simd4_real_t
1308 * \copydetails gmx_simd4_load1_f
1310 # define gmx_simd4_load1_r gmx_simd4_load1_f
1312 /*! \brief Set gmx_simd4_real_t from scalar value
1314 * \copydetails gmx_simd4_set1_f
1316 # define gmx_simd4_set1_r gmx_simd4_set1_f
1318 /*! \brief store aligned data from gmx_simd4_real_t
1320 * \copydetails gmx_simd4_store_f
1322 # define gmx_simd4_store_r gmx_simd4_store_f
1324 /*! \brief Load unaligned data to gmx_simd4_real_t
1326 * \copydetails gmx_simd4_loadu_f
1328 # define gmx_simd4_loadu_r gmx_simd4_loadu_f
1330 /*! \brief Store unaligned data from gmx_simd4_real_t
1332 * \copydetails gmx_simd4_storeu_f
1334 # define gmx_simd4_storeu_r gmx_simd4_storeu_f
1336 /*! \brief Set all elements in gmx_simd4_real_t to 0.0
1338 * \copydetails gmx_simd4_setzero_f
1340 # define gmx_simd4_setzero_r gmx_simd4_setzero_f
1342 /*! \brief Bitwise and for two gmx_simd4_real_t
1344 * \copydetails gmx_simd4_and_f
1346 # define gmx_simd4_and_r gmx_simd4_and_f
1348 /*! \brief Bitwise and-not for two gmx_simd4_real_t. 1st arg is complemented.
1350 * \copydetails gmx_simd4_andnot_f
1352 # define gmx_simd4_andnot_r gmx_simd4_andnot_f
1354 /*! \brief Bitwise or for two gmx_simd4_real_t
1356 * \copydetails gmx_simd4_or_f
1358 # define gmx_simd4_or_r gmx_simd4_or_f
1360 /*! \brief Bitwise xor for two gmx_simd4_real_t
1362 * \copydetails gmx_simd4_xor_f
1364 # define gmx_simd4_xor_r gmx_simd4_xor_f
1366 /*! \brief a+b for \ref gmx_simd4_real_t
1368 * \copydetails gmx_simd4_add_f
1370 # define gmx_simd4_add_r gmx_simd4_add_f
1372 /*! \brief a-b for \ref gmx_simd4_real_t
1374 * \copydetails gmx_simd4_sub_f
1376 # define gmx_simd4_sub_r gmx_simd4_sub_f
1378 /*! \brief a*b for \ref gmx_simd4_real_t
1380 * \copydetails gmx_simd4_mul_f
1382 # define gmx_simd4_mul_r gmx_simd4_mul_f
1384 /*! \brief a*b+c for \ref gmx_simd4_real_t
1386 * \copydetails gmx_simd4_fmadd_f
1388 # define gmx_simd4_fmadd_r gmx_simd4_fmadd_f
1390 /*! \brief a*b-c for \ref gmx_simd4_real_t
1392 * \copydetails gmx_simd4_fmsub_f
1394 # define gmx_simd4_fmsub_r gmx_simd4_fmsub_f
1396 /*! \brief -a*b+c for \ref gmx_simd4_real_t
1398 * \copydetails gmx_simd4_fnmadd_f
1400 # define gmx_simd4_fnmadd_r gmx_simd4_fnmadd_f
1402 /*! \brief -a*b-c for \ref gmx_simd4_real_t
1404 * \copydetails gmx_simd4_fnmsub_f
1406 # define gmx_simd4_fnmsub_r gmx_simd4_fnmsub_f
1408 /*! \brief 1/sqrt(x) approximate lookup for \ref gmx_simd4_real_t
1410 * \copydetails gmx_simd4_rsqrt_f
1412 # define gmx_simd4_rsqrt_r gmx_simd4_rsqrt_f
1414 /*! \brief fabs(x) for \ref gmx_simd4_real_t
1416 * \copydetails gmx_simd4_fabs_f
1418 # define gmx_simd4_fabs_r gmx_simd4_fabs_f
1420 /*! \brief Change sign (-x) for \ref gmx_simd4_real_t
1422 * \copydetails gmx_simd4_fneg_f
1424 # define gmx_simd4_fneg_r gmx_simd4_fneg_f
1426 /*! \brief Select maximum of each pair of elements from args for \ref gmx_simd4_real_t
1428 * \copydetails gmx_simd4_max_f
1430 # define gmx_simd4_max_r gmx_simd4_max_f
1432 /*! \brief Select minimum of each pair of elements from args for \ref gmx_simd4_real_t
1434 * \copydetails gmx_simd4_min_f
1436 # define gmx_simd4_min_r gmx_simd4_min_f
1438 /*! \brief Round \ref gmx_simd4_real_t to nearest integer, return \ref gmx_simd4_real_t
1440 * \copydetails gmx_simd4_round_f
1442 # define gmx_simd4_round_r gmx_simd4_round_f
1444 /*! \brief Truncate \ref gmx_simd4_real_t towards zero, return \ref gmx_simd4_real_t
1446 * \copydetails gmx_simd4_trunc_f
1448 # define gmx_simd4_trunc_r gmx_simd4_trunc_f
1450 /*! \brief Scalar product of first three elements of two \ref gmx_simd4_real_t *
1452 * \copydetails gmx_simd4_dotproduct3_f
1454 # define gmx_simd4_dotproduct3_r gmx_simd4_dotproduct3_f
1456 /*! \brief Return booleans whether a==b for each element two \ref gmx_simd4_real_t
1458 * \copydetails gmx_simd4_cmpeq_f
1460 # define gmx_simd4_cmpeq_r gmx_simd4_cmpeq_f
1461 /*! \brief Return booleans whether a<b for each element two \ref gmx_simd4_real_t
1463 * \copydetails gmx_simd4_cmplt_f
1465 # define gmx_simd4_cmplt_r gmx_simd4_cmplt_f
1466 /*! \brief Return booleans whether a<=b for each element two \ref gmx_simd4_real_t
1468 * \copydetails gmx_simd4_cmple_f
1470 # define gmx_simd4_cmple_r gmx_simd4_cmple_f
1472 /*! \brief Logical and for two \ref gmx_simd4_bool_t
1474 * \copydetails gmx_simd4_and_fb
1476 # define gmx_simd4_and_b gmx_simd4_and_fb
1477 /*! \brief Logical or for two \ref gmx_simd4_bool_t
1479 * \copydetails gmx_simd4_or_fb
1481 # define gmx_simd4_or_b gmx_simd4_or_fb
1483 /*! \brief Return nonzero if any element in \ref gmx_simd4_bool_t is true, otherwise 0
1485 * \copydetails gmx_simd4_anytrue_fb
1487 # define gmx_simd4_anytrue_b gmx_simd4_anytrue_fb
1489 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1491 * \copydetails gmx_simd4_blendzero_f
1493 # define gmx_simd4_blendzero_r gmx_simd4_blendzero_f
1495 /*! \brief Selects from 2nd real SIMD4 arg where boolean is false, otherwise 1st arg
1497 * \copydetails gmx_simd4_blendnotzero_f
1499 # define gmx_simd4_blendnotzero_r gmx_simd4_blendnotzero_f
1501 /*! \brief Selects from 2nd real SIMD4 arg where boolean is true, otherwise 1st arg
1503 * \copydetails gmx_simd4_blendv_f
1505 # define gmx_simd4_blendv_r gmx_simd4_blendv_f
1507 /*! \brief Return sum of all elements in SIMD4 floating-point variable.
1509 * \copydetails gmx_simd4_reduce_f
1511 # define gmx_simd4_reduce_r gmx_simd4_reduce_f
1513 /*! \brief Align real memory for SIMD4 usage.
1515 * \copydetails gmx_simd4_align_f
1517 # define gmx_simd4_align_r gmx_simd4_align_f
1521 /*! \name SIMD predefined macros to describe high-level capabilities
1525 # if (defined GMX_SIMD_HAVE_FLOAT) || (defined DOXYGEN)
1526 /*! \brief Defined if gmx_simd_real_t is available.
1528 * if GMX_DOUBLE is defined, this will be aliased to
1529 * \ref GMX_SIMD_HAVE_DOUBLE, otherwise GMX_SIMD_HAVE_FLOAT.
1531 # define GMX_SIMD_HAVE_REAL
1532 /*! \brief Width of gmx_simd_real_t.
1534 * if GMX_DOUBLE is defined, this will be aliased to
1535 * \ref GMX_SIMD_DOUBLE_WIDTH, otherwise GMX_SIMD_FLOAT_WIDTH.
1537 # define GMX_SIMD_REAL_WIDTH GMX_SIMD_FLOAT_WIDTH
1539 # if (defined GMX_SIMD_HAVE_FINT32) || (defined DOXYGEN)
1540 /*! \brief Defined if gmx_simd_int32_t is available.
1542 * if GMX_DOUBLE is defined, this will be aliased to
1543 * \ref GMX_SIMD_HAVE_DINT32, otherwise GMX_SIMD_HAVE_FINT32.
1545 # define GMX_SIMD_HAVE_INT32
1546 /*! \brief Width of gmx_simd_int32_t.
1548 * if GMX_DOUBLE is defined, this will be aliased to
1549 * \ref GMX_SIMD_DINT32_WIDTH, otherwise GMX_SIMD_FINT32_WIDTH.
1551 # define GMX_SIMD_INT32_WIDTH GMX_SIMD_FINT32_WIDTH
1553 # if (defined GMX_SIMD_HAVE_FINT32_EXTRACT) || (defined DOXYGEN)
1554 /*! \brief Defined if gmx_simd_extract_i() is available.
1556 * if GMX_DOUBLE is defined, this will be aliased to
1557 * \ref GMX_SIMD_HAVE_DINT32_EXTRACT, otherwise GMX_SIMD_HAVE_FINT32_EXTRACT.
1559 # define GMX_SIMD_HAVE_INT32_EXTRACT
1561 # if (defined GMX_SIMD_HAVE_FINT32_LOGICAL) || (defined DOXYGEN)
1562 /*! \brief Defined if logical ops are supported on gmx_simd_int32_t.
1564 * if GMX_DOUBLE is defined, this will be aliased to
1565 * \ref GMX_SIMD_HAVE_DINT32_LOGICAL, otherwise GMX_SIMD_HAVE_FINT32_LOGICAL.
1567 # define GMX_SIMD_HAVE_INT32_LOGICAL
1569 # if (defined GMX_SIMD_HAVE_FINT32_ARITHMETICS) || (defined DOXYGEN)
1570 /*! \brief Defined if arithmetic ops are supported on gmx_simd_int32_t.
1572 * if GMX_DOUBLE is defined, this will be aliased to
1573 * \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS, otherwise GMX_SIMD_HAVE_FINT32_ARITHMETICS.
1575 # define GMX_SIMD_HAVE_INT32_ARITHMETICS
1577 # if (defined GMX_SIMD4_HAVE_FLOAT) || (defined DOXYGEN)
1578 /*! \brief Defined if gmx_simd4_real_t is available.
1580 * if GMX_DOUBLE is defined, this will be aliased to
1581 * \ref GMX_SIMD4_HAVE_DOUBLE, otherwise GMX_SIMD4_HAVE_FLOAT.
1583 # define GMX_SIMD4_HAVE_REAL
1588 #endif /* GMX_DOUBLE */
1593 #endif /* GMX_SIMD_SIMD_H */