OpenJPH
Open-source implementation of JPEG2000 Part-15
ojph_colour_avx.cpp
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_colour_avx.cpp
34 // Author: Aous Naman
35 // Date: 11 October 2019
36 //***************************************************************************/
37 
38 #include <cmath>
39 
40 #include "ojph_defs.h"
41 #include "ojph_arch.h"
42 #include "ojph_colour.h"
43 #include "ojph_colour_local.h"
44 
45 #include <immintrin.h>
46 
47 namespace ojph {
48  namespace local {
49 
51  void avx_cnvrt_si32_to_float_shftd(const si32 *sp, float *dp, float mul,
52  ui32 width)
53  {
54  __m256 shift = _mm256_set1_ps(0.5f);
55  __m256 m = _mm256_set1_ps(mul);
56  for (int i = (width + 7) >> 3; i > 0; --i, sp+=8, dp+=8)
57  {
58  __m256i t = _mm256_loadu_si256((__m256i*)sp);
59  __m256 s = _mm256_cvtepi32_ps(t);
60  s = _mm256_mul_ps(s, m);
61  s = _mm256_sub_ps(s, shift);
62  _mm256_store_ps(dp, s);
63  }
64  }
65 
67  void avx_cnvrt_si32_to_float(const si32 *sp, float *dp, float mul,
68  ui32 width)
69  {
70  __m256 m = _mm256_set1_ps(mul);
71  for (int i = (width + 7) >> 3; i > 0; --i, sp+=8, dp+=8)
72  {
73  __m256i t = _mm256_loadu_si256((__m256i*)sp);
74  __m256 s = _mm256_cvtepi32_ps(t);
75  s = _mm256_mul_ps(s, m);
76  _mm256_store_ps(dp, s);
77  }
78  }
79 
81  void avx_cnvrt_float_to_si32_shftd(const float *sp, si32 *dp, float mul,
82  ui32 width)
83  {
84  __m256 shift = _mm256_set1_ps(0.5f);
85  __m256 m = _mm256_set1_ps(mul);
86  for (int i = (width + 7) >> 3; i > 0; --i, sp+=8, dp+=8)
87  {
88  __m256 t = _mm256_load_ps(sp);
89  __m256 s = _mm256_add_ps(t, shift);
90  s = _mm256_mul_ps(s, m);
91  s = _mm256_round_ps(s, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
92  _mm256_storeu_si256((__m256i*)dp, _mm256_cvtps_epi32(s));
93  }
94  }
95 
97  void avx_cnvrt_float_to_si32(const float *sp, si32 *dp, float mul,
98  ui32 width)
99  {
100  __m256 m = _mm256_set1_ps(mul);
101  for (int i = (width + 7) >> 3; i > 0; --i, sp+=8, dp+=8)
102  {
103  __m256 t = _mm256_load_ps(sp);
104  __m256 s = _mm256_mul_ps(t, m);
105  s = _mm256_round_ps(s, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
106  _mm256_storeu_si256((__m256i*)dp, _mm256_cvtps_epi32(s));
107  }
108  }
109 
111  void avx_ict_forward(const float *r, const float *g, const float *b,
112  float *y, float *cb, float *cr, ui32 repeat)
113  {
114  __m256 alpha_rf = _mm256_set1_ps(CT_CNST::ALPHA_RF);
115  __m256 alpha_gf = _mm256_set1_ps(CT_CNST::ALPHA_GF);
116  __m256 alpha_bf = _mm256_set1_ps(CT_CNST::ALPHA_BF);
117  __m256 beta_cbf = _mm256_set1_ps(CT_CNST::BETA_CbF);
118  __m256 beta_crf = _mm256_set1_ps(CT_CNST::BETA_CrF);
119  for (int i = (repeat + 7) >> 3; i > 0; --i)
120  {
121  __m256 mr = _mm256_load_ps(r);
122  __m256 mb = _mm256_load_ps(b);
123  __m256 my = _mm256_mul_ps(alpha_rf, mr);
124  my = _mm256_add_ps(my, _mm256_mul_ps(alpha_gf, _mm256_load_ps(g)));
125  my = _mm256_add_ps(my, _mm256_mul_ps(alpha_bf, mb));
126  _mm256_store_ps(y, my);
127  _mm256_store_ps(cb, _mm256_mul_ps(beta_cbf, _mm256_sub_ps(mb, my)));
128  _mm256_store_ps(cr, _mm256_mul_ps(beta_crf, _mm256_sub_ps(mr, my)));
129 
130  r += 8; g += 8; b += 8;
131  y += 8; cb += 8; cr += 8;
132  }
133  }
134 
136  void avx_ict_backward(const float *y, const float *cb, const float *cr,
137  float *r, float *g, float *b, ui32 repeat)
138  {
139  __m256 gamma_cr2g = _mm256_set1_ps(CT_CNST::GAMMA_CR2G);
140  __m256 gamma_cb2g = _mm256_set1_ps(CT_CNST::GAMMA_CB2G);
141  __m256 gamma_cr2r = _mm256_set1_ps(CT_CNST::GAMMA_CR2R);
142  __m256 gamma_cb2b = _mm256_set1_ps(CT_CNST::GAMMA_CB2B);
143  for (int i = (repeat + 7) >> 3; i > 0; --i)
144  {
145  __m256 my = _mm256_load_ps(y);
146  __m256 mcr = _mm256_load_ps(cr);
147  __m256 mcb = _mm256_load_ps(cb);
148  __m256 mg = _mm256_sub_ps(my, _mm256_mul_ps(gamma_cr2g, mcr));
149  _mm256_store_ps(g, _mm256_sub_ps(mg, _mm256_mul_ps(gamma_cb2g, mcb)));
150  _mm256_store_ps(r, _mm256_add_ps(my, _mm256_mul_ps(gamma_cr2r, mcr)));
151  _mm256_store_ps(b, _mm256_add_ps(my, _mm256_mul_ps(gamma_cb2b, mcb)));
152 
153  y += 8; cb += 8; cr += 8;
154  r += 8; g += 8; b += 8;
155  }
156  }
157 
158  }
159 }
void avx_cnvrt_float_to_si32(const float *sp, si32 *dp, float mul, ui32 width)
void avx_ict_forward(const float *r, const float *g, const float *b, float *y, float *cb, float *cr, ui32 repeat)
void avx_cnvrt_si32_to_float_shftd(const si32 *sp, float *dp, float mul, ui32 width)
void avx_cnvrt_si32_to_float(const si32 *sp, float *dp, float mul, ui32 width)
void avx_cnvrt_float_to_si32_shftd(const float *sp, si32 *dp, float mul, ui32 width)
void avx_ict_backward(const float *y, const float *cb, const float *cr, float *r, float *g, float *b, ui32 repeat)
int32_t si32
Definition: ojph_defs.h:55
uint32_t ui32
Definition: ojph_defs.h:54
static const float GAMMA_CR2R
static const float BETA_CbF
static const float GAMMA_CB2B
static const float ALPHA_RF
static const float GAMMA_CB2G
static const float GAMMA_CR2G
static const float ALPHA_BF
static const float BETA_CrF
static const float ALPHA_GF