| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2012, 2014 Free Software Foundation, Inc. | ||
| 4 | * | ||
| 5 | * This file is part of VOLK | ||
| 6 | * | ||
| 7 | * SPDX-License-Identifier: LGPL-3.0-or-later | ||
| 8 | */ | ||
| 9 | |||
| 10 | /*! | ||
| 11 | * \page volk_32u_popcnt | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Computes the population count (popcnt), or Hamming distance of a | ||
| 16 | * binary string. This kernel takes in a single unsigned 32-bit value | ||
| 17 | * and returns the count of 1's that the value contains. | ||
| 18 | * | ||
| 19 | * <b>Dispatcher Prototype</b> | ||
| 20 | * \code | ||
| 21 | * void volk_32u_popcnt(uint32_t* ret, const uint32_t value) | ||
| 22 | * \endcode | ||
| 23 | * | ||
| 24 | * \b Inputs | ||
| 25 | * \li value: The input value. | ||
| 26 | * | ||
| 27 | * \b Outputs | ||
| 28 | * \li ret: The return value containing the popcnt. | ||
| 29 | * | ||
| 30 | * \b Example | ||
| 31 | * \code | ||
| 32 | int N = 10; | ||
| 33 | unsigned int alignment = volk_get_alignment(); | ||
| 34 | |||
| 35 | uint32_t bitstring = 0x55555555; | ||
| 36 | uint32_t hamming_distance = 0; | ||
| 37 | |||
| 38 | volk_32u_popcnt(&hamming_distance, bitstring); | ||
| 39 | printf("hamming distance of %x = %i\n", bitstring, hamming_distance); | ||
| 40 | * \endcode | ||
| 41 | */ | ||
| 42 | |||
| 43 | #ifndef INCLUDED_VOLK_32u_POPCNT_A16_H | ||
| 44 | #define INCLUDED_VOLK_32u_POPCNT_A16_H | ||
| 45 | |||
| 46 | #include <inttypes.h> | ||
| 47 | #include <stdio.h> | ||
| 48 | |||
| 49 | #ifdef LV_HAVE_GENERIC | ||
| 50 | |||
| 51 | 262142 | static inline void volk_32u_popcnt_generic(uint32_t* ret, const uint32_t value) | |
| 52 | { | ||
| 53 | // This is faster than a lookup table | ||
| 54 | 262142 | uint32_t retVal = value; | |
| 55 | |||
| 56 | 262142 | retVal = (retVal & 0x55555555) + (retVal >> 1 & 0x55555555); | |
| 57 | 262142 | retVal = (retVal & 0x33333333) + (retVal >> 2 & 0x33333333); | |
| 58 | 262142 | retVal = (retVal + (retVal >> 4)) & 0x0F0F0F0F; | |
| 59 | 262142 | retVal = (retVal + (retVal >> 8)); | |
| 60 | 262142 | retVal = (retVal + (retVal >> 16)) & 0x0000003F; | |
| 61 | |||
| 62 | 262142 | *ret = retVal; | |
| 63 | 262142 | } | |
| 64 | |||
| 65 | #endif /*LV_HAVE_GENERIC*/ | ||
| 66 | |||
| 67 | |||
| 68 | #ifdef LV_HAVE_SSE4_2 | ||
| 69 | |||
| 70 | #include <nmmintrin.h> | ||
| 71 | |||
| 72 | 262142 | static inline void volk_32u_popcnt_a_sse4_2(uint32_t* ret, const uint32_t value) | |
| 73 | { | ||
| 74 | 262142 | *ret = _mm_popcnt_u32(value); | |
| 75 | 262142 | } | |
| 76 | |||
| 77 | #endif /*LV_HAVE_SSE4_2*/ | ||
| 78 | |||
| 79 | #endif /*INCLUDED_VOLK_32u_POPCNT_A16_H*/ | ||
| 80 |