OpenJPH
Open-source implementation of JPEG2000 Part-15
ojph_colour_wasm.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) 2021, Aous Naman
6 // Copyright (c) 2021, Kakadu Software Pty Ltd, Australia
7 // Copyright (c) 2021, 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_wasm.cpp
34 // Author: Aous Naman
35 // Date: 9 February 2021
36 //***************************************************************************/
37 
38 #include <cmath>
39 #include <wasm_simd128.h>
40 
41 #include "ojph_defs.h"
42 #include "ojph_colour.h"
43 #include "ojph_colour_local.h"
44 
45 namespace ojph {
46  namespace local {
47 
49  void wasm_cnvrt_si32_to_float_shftd(const si32 *sp, float *dp, float mul,
50  ui32 width)
51  {
52  v128_t shift = wasm_f32x4_splat(0.5f);
53  v128_t m = wasm_f32x4_splat(mul);
54  for (ui32 i = (width + 3) >> 2; i > 0; --i, sp+=4, dp+=4)
55  {
56  v128_t t = wasm_v128_load(sp);
57  v128_t s = wasm_f32x4_convert_i32x4(t);
58  s = wasm_f32x4_mul(s, m);
59  s = wasm_f32x4_sub(s, shift);
60  wasm_v128_store(dp, s);
61  }
62  }
63 
65  void wasm_cnvrt_si32_to_float(const si32 *sp, float *dp, float mul,
66  ui32 width)
67  {
68  v128_t m = wasm_f32x4_splat(mul);
69  for (ui32 i = (width + 3) >> 2; i > 0; --i, sp+=4, dp+=4)
70  {
71  v128_t t = wasm_v128_load(sp);
72  v128_t s = wasm_f32x4_convert_i32x4(t);
73  s = wasm_f32x4_mul(s, m);
74  wasm_v128_store(dp, s);
75  }
76  }
77 
79  void wasm_cnvrt_float_to_si32_shftd(const float *sp, si32 *dp, float mul,
80  ui32 width)
81  {
82  // rounding mode is always set to _MM_ROUND_NEAREST
83  v128_t shift = wasm_f32x4_splat(0.5f);
84  v128_t m = wasm_f32x4_splat(mul);
85  for (int i = (width + 3) >> 2; i > 0; --i, sp+=4, dp+=4)
86  {
87  v128_t t = wasm_v128_load(sp);
88  v128_t s = wasm_f32x4_add(t, shift);
89  s = wasm_f32x4_mul(s, m);
90  s = wasm_f32x4_add(s, shift); // + 0.5 and followed by floor next
91  wasm_v128_store(dp, wasm_i32x4_trunc_sat_f32x4(s));
92  }
93  }
94 
96  void wasm_cnvrt_float_to_si32(const float *sp, si32 *dp, float mul,
97  ui32 width)
98  {
99  // rounding mode is always set to _MM_ROUND_NEAREST
100  v128_t shift = wasm_f32x4_splat(0.5f);
101  v128_t m = wasm_f32x4_splat(mul);
102  for (int i = (width + 3) >> 2; i > 0; --i, sp+=4, dp+=4)
103  {
104  v128_t t = wasm_v128_load(sp);
105  v128_t s = wasm_f32x4_mul(t, m);
106  s = wasm_f32x4_add(s, shift); // + 0.5 and followed by floor next
107  wasm_v128_store(dp, wasm_i32x4_trunc_sat_f32x4(s));
108  }
109  }
110 
111 
113  void wasm_cnvrt_si32_to_si32_shftd(const si32 *sp, si32 *dp, int shift,
114  ui32 width)
115  {
116  v128_t sh = wasm_i32x4_splat(shift);
117  for (int i = (width + 3) >> 2; i > 0; --i, sp+=4, dp+=4)
118  {
119  v128_t s = wasm_v128_load(sp);
120  s = wasm_i32x4_add(s, sh);
121  wasm_v128_store(dp, s);
122  }
123  }
124 
126  void wasm_rct_forward(const si32 *r, const si32 *g, const si32 *b,
127  si32 *y, si32 *cb, si32 *cr, ui32 repeat)
128  {
129  for (int i = (repeat + 3) >> 2; i > 0; --i)
130  {
131  v128_t mr = wasm_v128_load(r);
132  v128_t mg = wasm_v128_load(g);
133  v128_t mb = wasm_v128_load(b);
134  v128_t t = wasm_i32x4_add(mr, mb);
135  t = wasm_i32x4_add(t, wasm_i32x4_shl(mg, 1));
136  wasm_v128_store(y, wasm_i32x4_shr(t, 2));
137  t = wasm_i32x4_sub(mb, mg);
138  wasm_v128_store(cb, t);
139  t = wasm_i32x4_sub(mr, mg);
140  wasm_v128_store(cr, t);
141 
142  r += 4; g += 4; b += 4;
143  y += 4; cb += 4; cr += 4;
144  }
145  }
146 
148  void wasm_rct_backward(const si32 *y, const si32 *cb, const si32 *cr,
149  si32 *r, si32 *g, si32 *b, ui32 repeat)
150  {
151  for (int i = (repeat + 3) >> 2; i > 0; --i)
152  {
153  v128_t my = wasm_v128_load(y);
154  v128_t mcb = wasm_v128_load(cb);
155  v128_t mcr = wasm_v128_load(cr);
156 
157  v128_t t = wasm_i32x4_add(mcb, mcr);
158  t = wasm_i32x4_sub(my, wasm_i32x4_shr(t, 2));
159  wasm_v128_store(g, t);
160  v128_t u = wasm_i32x4_add(mcb, t);
161  wasm_v128_store(b, u);
162  u = wasm_i32x4_add(mcr, t);
163  wasm_v128_store(r, u);
164 
165  y += 4; cb += 4; cr += 4;
166  r += 4; g += 4; b += 4;
167  }
168  }
169 
171  void wasm_ict_forward(const float *r, const float *g, const float *b,
172  float *y, float *cb, float *cr, ui32 repeat)
173  {
174  v128_t alpha_rf = wasm_f32x4_splat(CT_CNST::ALPHA_RF);
175  v128_t alpha_gf = wasm_f32x4_splat(CT_CNST::ALPHA_GF);
176  v128_t alpha_bf = wasm_f32x4_splat(CT_CNST::ALPHA_BF);
177  v128_t beta_cbf = wasm_f32x4_splat(CT_CNST::BETA_CbF);
178  v128_t beta_crf = wasm_f32x4_splat(CT_CNST::BETA_CrF);
179  for (ui32 i = (repeat + 3) >> 2; i > 0; --i)
180  {
181  v128_t mr = wasm_v128_load(r);
182  v128_t mb = wasm_v128_load(b);
183  v128_t my = wasm_f32x4_mul(alpha_rf, mr);
184  my = wasm_f32x4_add(my, wasm_f32x4_mul(alpha_gf, wasm_v128_load(g)));
185  my = wasm_f32x4_add(my, wasm_f32x4_mul(alpha_bf, mb));
186  wasm_v128_store(y, my);
187  wasm_v128_store(cb, wasm_f32x4_mul(beta_cbf, wasm_f32x4_sub(mb, my)));
188  wasm_v128_store(cr, wasm_f32x4_mul(beta_crf, wasm_f32x4_sub(mr, my)));
189 
190  r += 4; g += 4; b += 4;
191  y += 4; cb += 4; cr += 4;
192  }
193  }
194 
196  void wasm_ict_backward(const float *y, const float *cb, const float *cr,
197  float *r, float *g, float *b, ui32 repeat)
198  {
199  v128_t gamma_cr2g = wasm_f32x4_splat(CT_CNST::GAMMA_CR2G);
200  v128_t gamma_cb2g = wasm_f32x4_splat(CT_CNST::GAMMA_CB2G);
201  v128_t gamma_cr2r = wasm_f32x4_splat(CT_CNST::GAMMA_CR2R);
202  v128_t gamma_cb2b = wasm_f32x4_splat(CT_CNST::GAMMA_CB2B);
203  for (ui32 i = (repeat + 3) >> 2; i > 0; --i)
204  {
205  v128_t my = wasm_v128_load(y);
206  v128_t mcr = wasm_v128_load(cr);
207  v128_t mcb = wasm_v128_load(cb);
208  v128_t mg = wasm_f32x4_sub(my, wasm_f32x4_mul(gamma_cr2g, mcr));
209  wasm_v128_store(g, wasm_f32x4_sub(mg, wasm_f32x4_mul(gamma_cb2g, mcb)));
210  wasm_v128_store(r, wasm_f32x4_add(my, wasm_f32x4_mul(gamma_cr2r, mcr)));
211  wasm_v128_store(b, wasm_f32x4_add(my, wasm_f32x4_mul(gamma_cb2b, mcb)));
212 
213  y += 4; cb += 4; cr += 4;
214  r += 4; g += 4; b += 4;
215  }
216  }
217 
218  }
219 }
void wasm_cnvrt_float_to_si32_shftd(const float *sp, si32 *dp, float mul, ui32 width)
void wasm_ict_backward(const float *y, const float *cb, const float *cr, float *r, float *g, float *b, ui32 repeat)
void wasm_cnvrt_si32_to_float_shftd(const si32 *sp, float *dp, float mul, ui32 width)
void wasm_cnvrt_float_to_si32(const float *sp, si32 *dp, float mul, ui32 width)
void wasm_ict_forward(const float *r, const float *g, const float *b, float *y, float *cb, float *cr, ui32 repeat)
void wasm_rct_forward(const si32 *r, const si32 *g, const si32 *b, si32 *y, si32 *cb, si32 *cr, ui32 repeat)
void wasm_cnvrt_si32_to_si32_shftd(const si32 *sp, si32 *dp, int shift, ui32 width)
void wasm_cnvrt_si32_to_float(const si32 *sp, float *dp, float mul, ui32 width)
void wasm_rct_backward(const si32 *y, const si32 *cb, const si32 *cr, si32 *r, si32 *g, si32 *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