OpenJPH
Open-source implementation of JPEG2000 Part-15
ojph_arch.h
Go to the documentation of this file.
1//***************************************************************************/
2// This software is released under the 2-Clause BSD license, included
3// below.
4//
5// Copyright (c) 2019, Aous Naman
6// Copyright (c) 2019, Kakadu Software Pty Ltd, Australia
7// Copyright (c) 2019, The University of New South Wales, Australia
8//
9// Redistribution and use in source and binary forms, with or without
10// modification, are permitted provided that the following conditions are
11// met:
12//
13// 1. Redistributions of source code must retain the above copyright
14// notice, this list of conditions and the following disclaimer.
15//
16// 2. Redistributions in binary form must reproduce the above copyright
17// notice, this list of conditions and the following disclaimer in the
18// documentation and/or other materials provided with the distribution.
19//
20// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
21// IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22// TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
23// PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
26// TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
27// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
28// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
29// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
30// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31//***************************************************************************/
32// This file is part of the OpenJPH software implementation.
33// File: ojph_arch.h
34// Author: Aous Naman
35// Date: 28 August 2019
36//***************************************************************************/
37
38
39#ifndef OJPH_ARCH_H
40#define OJPH_ARCH_H
41
42#include <cstdio>
43#include <cstdint>
44#include <cmath>
45
46#include "ojph_defs.h"
47
48
50// preprocessor directives for compiler
52#ifdef _MSC_VER
53#define OJPH_COMPILER_MSVC
54#elif (defined __GNUC__)
55#define OJPH_COMPILER_GNUC
56#endif
57
58#ifdef __EMSCRIPTEN__
59#define OJPH_EMSCRIPTEN
60#endif
61
62#ifdef OJPH_COMPILER_MSVC
63#include <intrin.h>
64#endif
65
66namespace ojph {
67
69 // cpu features
72
73 enum : int {
86 };
87
89 static inline ui32 population_count(ui32 val)
90 {
91 #ifdef OJPH_COMPILER_MSVC
92 return (ui32)__popcnt(val);
93 #elif (defined OJPH_COMPILER_GNUC)
94 return (ui32)__builtin_popcount(val);
95 #else
96 val -= ((val >> 1) & 0x55555555);
97 val = (((val >> 2) & 0x33333333) + (val & 0x33333333));
98 val = (((val >> 4) + val) & 0x0f0f0f0f);
99 val += (val >> 8);
100 val += (val >> 16);
101 return (int)(val & 0x0000003f);
102 #endif
103 }
104
106#ifdef OJPH_COMPILER_MSVC
107 #pragma intrinsic(_BitScanReverse)
108#endif
109 static inline ui32 count_leading_zeros(ui32 val)
110 {
111 #ifdef OJPH_COMPILER_MSVC
112 unsigned long result = 0;
113 _BitScanReverse(&result, val);
114 return 31 ^ (ui32)result;
115 #elif (defined OJPH_COMPILER_GNUC)
116 return (ui32)__builtin_clz(val);
117 #else
118 val |= (val >> 1);
119 val |= (val >> 2);
120 val |= (val >> 4);
121 val |= (val >> 8);
122 val |= (val >> 16);
123 return 32 - population_count(val);
124 #endif
125 }
126
128#ifdef OJPH_COMPILER_MSVC
129 #pragma intrinsic(_BitScanForward)
130#endif
131 static inline ui32 count_trailing_zeros(ui32 val)
132 {
133 #ifdef OJPH_COMPILER_MSVC
134 unsigned long result = 0;
135 _BitScanForward(&result, val);
136 return (ui32)result;
137 #elif (defined OJPH_COMPILER_GNUC)
138 return (ui32)__builtin_ctz(val);
139 #else
140 val |= (val << 1);
141 val |= (val << 2);
142 val |= (val << 4);
143 val |= (val << 8);
144 val |= (val << 16);
145 return 32 - population_count(val);
146 #endif
147 }
148
150 static inline si32 ojph_round(float val)
151 {
152 #ifdef OJPH_COMPILER_MSVC
153 return (si32)(val + (val >= 0.0f ? 0.5f : -0.5f));
154 #elif (defined OJPH_COMPILER_GNUC)
155 return (si32)(val + (val >= 0.0f ? 0.5f : -0.5f));
156 #else
157 return (si32)round(val);
158 #endif
159 }
160
162 static inline si32 ojph_trunc(float val)
163 {
164 #ifdef OJPH_COMPILER_MSVC
165 return (si32)(val);
166 #elif (defined OJPH_COMPILER_GNUC)
167 return (si32)(val);
168 #else
169 return (si32)trunc(val);
170 #endif
171 }
172
174 // constants
176 const ui32 byte_alignment = 32; //32 bytes == 256 bits
179
181 // templates for alignment
183
185 // finds the size such that it is a multiple of byte_alignment
186 template <typename T, int N>
187 size_t calc_aligned_size(size_t size) {
188 size = size * sizeof(T) + N - 1;
189 size &= ~((1ULL << (31 - count_leading_zeros(N))) - 1);
190 size >>= (31 - count_leading_zeros(sizeof(T)));
191 return size;
192 }
193
195 // moves the pointer to first address that is a multiple of byte_alignment
196 template <typename T, int N>
197 inline T *align_ptr(T *ptr) {
198 intptr_t p = reinterpret_cast<intptr_t>(ptr);
199 p += N - 1;
200 p &= ~((1ULL << (31 - count_leading_zeros(N))) - 1);
201 return reinterpret_cast<T *>(p);
202 }
203
205 // OS detection definitions
207#if (defined WIN32) || (defined _WIN32) || (defined _WIN64)
208 #define OJPH_OS_WINDOWS
209#elif (defined __APPLE__)
210 #define OJPH_OS_APPLE
211#elif (defined __linux)
212 #define OJPH_OS_LINUX
213#endif
214
216 // defines for dll
218#if defined(OJPH_OS_WINDOWS) && defined(OJPH_BUILD_SHARED_LIBRARY)
219 #define OJPH_EXPORT __declspec(dllexport)
220#else
221 #define OJPH_EXPORT
222#endif
223
224}
225
226#endif // !OJPH_ARCH_H
const ui32 object_alignment
Definition: ojph_arch.h:178
const ui32 byte_alignment
Definition: ojph_arch.h:176
static si32 ojph_round(float val)
Definition: ojph_arch.h:150
size_t calc_aligned_size(size_t size)
Definition: ojph_arch.h:187
@ X86_CPU_EXT_LEVEL_AVX2
Definition: ojph_arch.h:83
@ X86_CPU_EXT_LEVEL_AVX
Definition: ojph_arch.h:82
@ X86_CPU_EXT_LEVEL_AVX512
Definition: ojph_arch.h:85
@ X86_CPU_EXT_LEVEL_GENERIC
Definition: ojph_arch.h:74
@ X86_CPU_EXT_LEVEL_SSE2
Definition: ojph_arch.h:77
@ X86_CPU_EXT_LEVEL_SSE41
Definition: ojph_arch.h:80
@ X86_CPU_EXT_LEVEL_SSE
Definition: ojph_arch.h:76
@ X86_CPU_EXT_LEVEL_MMX
Definition: ojph_arch.h:75
@ X86_CPU_EXT_LEVEL_SSE42
Definition: ojph_arch.h:81
@ X86_CPU_EXT_LEVEL_SSSE3
Definition: ojph_arch.h:79
@ X86_CPU_EXT_LEVEL_SSE3
Definition: ojph_arch.h:78
@ X86_CPU_EXT_LEVEL_AVX2FMA
Definition: ojph_arch.h:84
T * align_ptr(T *ptr)
Definition: ojph_arch.h:197
static ui32 population_count(ui32 val)
Definition: ojph_arch.h:89
int get_cpu_ext_level()
Definition: ojph_arch.cpp:182
static si32 ojph_trunc(float val)
Definition: ojph_arch.h:162
static ui32 count_trailing_zeros(ui32 val)
Definition: ojph_arch.h:131
static ui32 count_leading_zeros(ui32 val)
Definition: ojph_arch.h:109
int32_t si32
Definition: ojph_defs.h:55
const ui32 log_byte_alignment
Definition: ojph_arch.h:177
uint32_t ui32
Definition: ojph_defs.h:54