OpenJPH
Open-source implementation of JPEG2000 Part-15
ojph_codestream_avx2.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) 2022, Aous Naman
6// Copyright (c) 2022, Kakadu Software Pty Ltd, Australia
7// Copyright (c) 2022, 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_codestream_avx2.cpp
34// Author: Aous Naman
35// Date: 15 May 2022
36//***************************************************************************/
37
38#include <immintrin.h>
39#include "ojph_defs.h"
40
41namespace ojph {
42 namespace local {
43
46 {
47 __m128i x0 = _mm_loadu_si128((__m128i*)address);
48 __m128i x1 = _mm_loadu_si128((__m128i*)address + 1);
49 x0 = _mm_or_si128(x0, x1);
50 x1 = _mm_shuffle_epi32(x0, 0xEE); // x1 = x0[2,3,2,3]
51 x0 = _mm_or_si128(x0, x1);
52 x1 = _mm_shuffle_epi32(x0, 0x55); // x1 = x0[1,1,1,1]
53 x0 = _mm_or_si128(x0, x1);
54 ui32 t = (ui32)_mm_extract_epi32(x0, 0);
55 return t;
56 }
57
59 void avx2_rev_tx_to_cb(const void *sp, ui32 *dp, ui32 K_max,
60 float delta_inv, ui32 count, ui32* max_val)
61 {
62 ojph_unused(delta_inv);
63
64 // convert to sign and magnitude and keep max_val
65 ui32 shift = 31 - K_max;
66 __m256i m0 = _mm256_set1_epi32((int)0x80000000);
67 __m256i tmax = _mm256_loadu_si256((__m256i*)max_val);
68 __m256i *p = (__m256i*)sp;
69 for (ui32 i = 0; i < count; i += 8, p += 1, dp += 8)
70 {
71 __m256i v = _mm256_loadu_si256(p);
72 __m256i sign = _mm256_and_si256(v, m0);
73 __m256i val = _mm256_abs_epi32(v);
74 val = _mm256_slli_epi32(val, (int)shift);
75 tmax = _mm256_or_si256(tmax, val);
76 val = _mm256_or_si256(val, sign);
77 _mm256_storeu_si256((__m256i*)dp, val);
78 }
79 _mm256_storeu_si256((__m256i*)max_val, tmax);
80 }
81
83 void avx2_irv_tx_to_cb(const void *sp, ui32 *dp, ui32 K_max,
84 float delta_inv, ui32 count, ui32* max_val)
85 {
86 ojph_unused(K_max);
87
88 //quantize and convert to sign and magnitude and keep max_val
89 __m256 d = _mm256_set1_ps(delta_inv);
90 __m256i m0 = _mm256_set1_epi32((int)0x80000000);
91 __m256i tmax = _mm256_loadu_si256((__m256i*)max_val);
92 float *p = (float*)sp;
93
94 for (ui32 i = 0; i < count; i += 8, p += 8, dp += 8)
95 {
96 __m256 vf = _mm256_loadu_ps(p);
97 vf = _mm256_mul_ps(vf, d); // multiply
98 __m256i val = _mm256_cvtps_epi32(vf); // convert to int
99 __m256i sign = _mm256_and_si256(val, m0); // get sign
100 val = _mm256_abs_epi32(val);
101 tmax = _mm256_or_si256(tmax, val);
102 val = _mm256_or_si256(val, sign);
103 _mm256_storeu_si256((__m256i*)dp, val);
104 }
105 _mm256_storeu_si256((__m256i*)max_val, tmax);
106 }
107
109 void avx2_rev_tx_from_cb(const ui32 *sp, void *dp, ui32 K_max,
110 float delta, ui32 count)
111 {
112 ojph_unused(delta);
113 ui32 shift = 31 - K_max;
114 __m256i m1 = _mm256_set1_epi32(0x7FFFFFFF);
115 si32 *p = (si32*)dp;
116 for (ui32 i = 0; i < count; i += 8, sp += 8, p += 8)
117 {
118 __m256i v = _mm256_load_si256((__m256i*)sp);
119 __m256i val = _mm256_and_si256(v, m1);
120 val = _mm256_srli_epi32(val, (int)shift);
121 val = _mm256_sign_epi32(val, v);
122 _mm256_storeu_si256((__m256i*)p, val);
123 }
124 }
125
127 void avx2_irv_tx_from_cb(const ui32 *sp, void *dp, ui32 K_max,
128 float delta, ui32 count)
129 {
130 ojph_unused(K_max);
131 __m256i m1 = _mm256_set1_epi32(0x7FFFFFFF);
132 __m256 d = _mm256_set1_ps(delta);
133 float *p = (float*)dp;
134 for (ui32 i = 0; i < count; i += 8, sp += 8, p += 8)
135 {
136 __m256i v = _mm256_load_si256((__m256i*)sp);
137 __m256i vali = _mm256_and_si256(v, m1);
138 __m256 valf = _mm256_cvtepi32_ps(vali);
139 valf = _mm256_mul_ps(valf, d);
140 __m256i sign = _mm256_andnot_si256(m1, v);
141 valf = _mm256_or_ps(valf, _mm256_castsi256_ps(sign));
142 _mm256_storeu_ps(p, valf);
143 }
144 }
145 }
146}
ui32 avx2_find_max_val(ui32 *address)
void avx2_irv_tx_from_cb(const ui32 *sp, void *dp, ui32 K_max, float delta, ui32 count)
void avx2_rev_tx_to_cb(const void *sp, ui32 *dp, ui32 K_max, float delta_inv, ui32 count, ui32 *max_val)
void avx2_rev_tx_from_cb(const ui32 *sp, void *dp, ui32 K_max, float delta, ui32 count)
void avx2_irv_tx_to_cb(const void *sp, ui32 *dp, ui32 K_max, float delta_inv, ui32 count, ui32 *max_val)
int32_t si32
Definition: ojph_defs.h:55
uint32_t ui32
Definition: ojph_defs.h:54
#define ojph_unused(x)
Definition: ojph_defs.h:78