| 1 |
/* |
| 2 |
* Copyright (c) 2023. |
| 3 |
* |
| 4 |
* This software is free software; |
| 5 |
* |
| 6 |
* You can redistribute it or modify it under terms of the MIT, Apache License or Zlib license |
| 7 |
*/ |
| 8 |
|
| 9 |
//! Platform independent IDCT algorithm |
| 10 |
//! |
| 11 |
//! Not as fast as AVX one. |
| 12 |
|
| 13 |
const SCALE_BITS: i32 = 512 + 65536 + (128 << 17); |
| 14 |
|
| 15 |
#[inline(always)] |
| 16 |
fn wa(a: i32, b: i32) -> i32 { |
| 17 |
a.wrapping_add(b) |
| 18 |
} |
| 19 |
|
| 20 |
#[inline(always)] |
| 21 |
fn ws(a: i32, b: i32) -> i32 { |
| 22 |
a.wrapping_sub(b) |
| 23 |
} |
| 24 |
|
| 25 |
#[inline(always)] |
| 26 |
fn wm(a: i32, b: i32) -> i32 { |
| 27 |
a.wrapping_mul(b) |
| 28 |
} |
| 29 |
|
| 30 |
#[inline] |
| 31 |
pub fn idct_int_1x1(in_vector: &mut [i32; 64], mut out_vector: &mut [i16], stride: usize) { |
| 32 |
let coeff = ((wa(wa(in_vector[0], 4), 1024) >> 3).clamp(0, 255)) as i16; |
| 33 |
|
| 34 |
out_vector[..8].fill(coeff); |
| 35 |
for _ in 0..7 { |
| 36 |
out_vector = &mut out_vector[stride..]; |
| 37 |
out_vector[..8].fill(coeff); |
| 38 |
} |
| 39 |
} |
| 40 |
|
| 41 |
#[allow(unused_assignments)] |
| 42 |
#[allow( |
| 43 |
clippy::too_many_lines, |
| 44 |
clippy::op_ref, |
| 45 |
clippy::cast_possible_truncation |
| 46 |
)] |
| 47 |
pub fn idct_int(in_vector: &mut [i32; 64], out_vector: &mut [i16], stride: usize) { |
| 48 |
let mut pos = 0; |
| 49 |
let mut i = 0; |
| 50 |
|
| 51 |
if &in_vector[1..] == &[0_i32; 63] { |
| 52 |
return idct_int_1x1(in_vector, out_vector, stride); |
| 53 |
} |
| 54 |
|
| 55 |
// vertical pass |
| 56 |
for ptr in 0..8 { |
| 57 |
let p2 = in_vector[ptr + 16]; |
| 58 |
let p3 = in_vector[ptr + 48]; |
| 59 |
|
| 60 |
let p1 = wm(wa(p2, p3), 2217); |
| 61 |
|
| 62 |
let t2 = wa(p1, wm(p3, -7567)); |
| 63 |
let t3 = wa(p1, wm(p2, 3135)); |
| 64 |
|
| 65 |
let p2 = in_vector[ptr]; |
| 66 |
let p3 = in_vector[32 + ptr]; |
| 67 |
|
| 68 |
let t0 = fsh(wa(p2, p3)); |
| 69 |
let t1 = fsh(ws(p2, p3)); |
| 70 |
|
| 71 |
let x0 = wa(wa(t0, t3), 512); |
| 72 |
let x3 = wa(ws(t0, t3), 512); |
| 73 |
let x1 = wa(wa(t1, t2), 512); |
| 74 |
let x2 = wa(ws(t1, t2), 512); |
| 75 |
|
| 76 |
let mut t0 = in_vector[ptr + 56]; |
| 77 |
let mut t1 = in_vector[ptr + 40]; |
| 78 |
let mut t2 = in_vector[ptr + 24]; |
| 79 |
let mut t3 = in_vector[ptr + 8]; |
| 80 |
|
| 81 |
let p3 = wa(t0, t2); |
| 82 |
let p4 = wa(t1, t3); |
| 83 |
let p1 = wa(t0, t3); |
| 84 |
let p2 = wa(t1, t2); |
| 85 |
let p5 = wm(wa(p3, p4), 4816); |
| 86 |
|
| 87 |
t0 = wm(t0, 1223); |
| 88 |
t1 = wm(t1, 8410); |
| 89 |
t2 = wm(t2, 12586); |
| 90 |
t3 = wm(t3, 6149); |
| 91 |
|
| 92 |
let p1 = wa(p5, wm(p1, -3685)); |
| 93 |
let p2 = wa(p5, wm(p2, -10497)); |
| 94 |
let p3 = wm(p3, -8034); |
| 95 |
let p4 = wm(p4, -1597); |
| 96 |
|
| 97 |
t3 = wa(t3, wa(p1, p4)); |
| 98 |
t2 = wa(t2, wa(p2, p3)); |
| 99 |
t1 = wa(t1, wa(p2, p4)); |
| 100 |
t0 = wa(t0, wa(p1, p3)); |
| 101 |
|
| 102 |
in_vector[ptr] = ws(wa(x0, t3), 0) >> 10; |
| 103 |
in_vector[ptr + 8] = ws(wa(x1, t2), 0) >> 10; |
| 104 |
in_vector[ptr + 16] = ws(wa(x2, t1), 0) >> 10; |
| 105 |
in_vector[ptr + 24] = ws(wa(x3, t0), 0) >> 10; |
| 106 |
in_vector[ptr + 32] = ws(ws(x3, t0), 0) >> 10; |
| 107 |
in_vector[ptr + 40] = ws(ws(x2, t1), 0) >> 10; |
| 108 |
in_vector[ptr + 48] = ws(ws(x1, t2), 0) >> 10; |
| 109 |
in_vector[ptr + 56] = ws(ws(x0, t3), 0) >> 10; |
| 110 |
} |
| 111 |
|
| 112 |
// horizontal pass |
| 113 |
while i < 64 { |
| 114 |
let p2 = in_vector[i + 2]; |
| 115 |
let p3 = in_vector[i + 6]; |
| 116 |
|
| 117 |
let p1 = wm(wa(p2, p3), 2217); |
| 118 |
let t2 = wa(p1, wm(p3, -7567)); |
| 119 |
let t3 = wa(p1, wm(p2, 3135)); |
| 120 |
|
| 121 |
let p2 = in_vector[i]; |
| 122 |
let p3 = in_vector[i + 4]; |
| 123 |
|
| 124 |
let t0 = fsh(wa(p2, p3)); |
| 125 |
let t1 = fsh(ws(p2, p3)); |
| 126 |
|
| 127 |
let x0 = wa(wa(t0, t3), SCALE_BITS); |
| 128 |
let x3 = wa(ws(t0, t3), SCALE_BITS); |
| 129 |
let x1 = wa(wa(t1, t2), SCALE_BITS); |
| 130 |
let x2 = wa(ws(t1, t2), SCALE_BITS); |
| 131 |
|
| 132 |
let mut t0 = in_vector[i + 7]; |
| 133 |
let mut t1 = in_vector[i + 5]; |
| 134 |
let mut t2 = in_vector[i + 3]; |
| 135 |
let mut t3 = in_vector[i + 1]; |
| 136 |
|
| 137 |
let p3 = wa(t0, t2); |
| 138 |
let p4 = wa(t1, t3); |
| 139 |
let p1 = wa(t0, t3); |
| 140 |
let p2 = wa(t1, t2); |
| 141 |
let p5 = wm(wa(p3, p4), f2f(1.175875602)); |
| 142 |
|
| 143 |
t0 = wm(t0, 1223); |
| 144 |
t1 = wm(t1, 8410); |
| 145 |
t2 = wm(t2, 12586); |
| 146 |
t3 = wm(t3, 6149); |
| 147 |
|
| 148 |
let p1 = wa(p5, wm(p1, -3685)); |
| 149 |
let p2 = wa(p5, wm(p2, -10497)); |
| 150 |
let p3 = wm(p3, -8034); |
| 151 |
let p4 = wm(p4, -1597); |
| 152 |
|
| 153 |
t3 = wa(t3, wa(p1, p4)); |
| 154 |
t2 = wa(t2, wa(p2, p3)); |
| 155 |
t1 = wa(t1, wa(p2, p4)); |
| 156 |
t0 = wa(t0, wa(p1, p3)); |
| 157 |
|
| 158 |
let out: &mut [i16; 8] = out_vector |
| 159 |
.get_mut(pos..pos + 8) |
| 160 |
.unwrap() |
| 161 |
.try_into() |
| 162 |
.unwrap(); |
| 163 |
|
| 164 |
out[0] = clamp(wa(x0, t3) >> 17); |
| 165 |
out[1] = clamp(wa(x1, t2) >> 17); |
| 166 |
out[2] = clamp(wa(x2, t1) >> 17); |
| 167 |
out[3] = clamp(wa(x3, t0) >> 17); |
| 168 |
out[4] = clamp(ws(x3, t0) >> 17); |
| 169 |
out[5] = clamp(ws(x2, t1) >> 17); |
| 170 |
out[6] = clamp(ws(x1, t2) >> 17); |
| 171 |
out[7] = clamp(ws(x0, t3) >> 17); |
| 172 |
|
| 173 |
i += 8; |
| 174 |
pos += stride; |
| 175 |
} |
| 176 |
} |
| 177 |
|
| 178 |
#[inline] |
| 179 |
#[allow(clippy::cast_possible_truncation)] |
| 180 |
/// Multiply a number by 4096 |
| 181 |
fn f2f(x: f32) -> i32 { |
| 182 |
(x * 4096.0 + 0.5) as i32 |
| 183 |
} |
| 184 |
|
| 185 |
#[inline] |
| 186 |
/// Multiply a number by 4096 |
| 187 |
fn fsh(x: i32) -> i32 { |
| 188 |
x << 12 |
| 189 |
} |
| 190 |
|
| 191 |
/// Clamp values between 0 and 255 |
| 192 |
#[inline] |
| 193 |
#[allow(clippy::cast_possible_truncation)] |
| 194 |
fn clamp(a: i32) -> i16 { |
| 195 |
a.clamp(0, 255) as i16 |
| 196 |
} |
| 197 |
|
| 198 |
/// IDCT assuming only the upper 4x4 is filled. |
| 199 |
pub fn idct4x4(in_vector: &mut [i32; 64], out_vector: &mut [i16], stride: usize) { |
| 200 |
let mut pos = 0; |
| 201 |
|
| 202 |
// vertical pass |
| 203 |
for ptr in 0..4 { |
| 204 |
let i0 = wa(fsh(in_vector[ptr]), 512); |
| 205 |
let i2 = in_vector[ptr + 16]; |
| 206 |
|
| 207 |
let p1 = wm(i2, 2217); |
| 208 |
let p3 = wm(i2, 5352); |
| 209 |
|
| 210 |
let x0 = wa(i0, p3); |
| 211 |
let x1 = wa(i0, p1); |
| 212 |
let x2 = ws(i0, p1); |
| 213 |
let x3 = ws(i0, p3); |
| 214 |
|
| 215 |
// odd part |
| 216 |
let i4 = in_vector[ptr + 24]; |
| 217 |
let i3 = in_vector[ptr + 8]; |
| 218 |
|
| 219 |
let p5 = wm(wa(i4, i3), 4816); |
| 220 |
|
| 221 |
let p1 = wa(p5, wm(i3, -3685)); |
| 222 |
let p2 = wa(p5, wm(i4, -10497)); |
| 223 |
|
| 224 |
let t3 = wa(p5, wm(i3, 867)); |
| 225 |
let t2 = wa(p5, wm(i4, -5945)); |
| 226 |
|
| 227 |
let t1 = wa(p2, wm(i3, -1597)); |
| 228 |
let t0 = wa(p1, wm(i4, -8034)); |
| 229 |
|
| 230 |
in_vector[ptr] = wa(x0, t3) >> 10; |
| 231 |
in_vector[ptr + 8] = wa(x1, t2) >> 10; |
| 232 |
in_vector[ptr + 16] = wa(x2, t1) >> 10; |
| 233 |
in_vector[ptr + 24] = wa(x3, t0) >> 10; |
| 234 |
in_vector[ptr + 32] = ws(x3, t0) >> 10; |
| 235 |
in_vector[ptr + 40] = ws(x2, t1) >> 10; |
| 236 |
in_vector[ptr + 48] = ws(x1, t2) >> 10; |
| 237 |
in_vector[ptr + 56] = ws(x0, t3) >> 10; |
| 238 |
} |
| 239 |
|
| 240 |
// horizontal pass |
| 241 |
for i in (0..8).map(|i| 8 * i) { |
| 242 |
let i2 = in_vector[i + 2]; |
| 243 |
let i0 = in_vector[i]; |
| 244 |
|
| 245 |
let t0 = wa(fsh(i0), SCALE_BITS); |
| 246 |
let t2 = wm(i2, 2217); |
| 247 |
let t3 = wm(i2, 5352); |
| 248 |
|
| 249 |
let x0 = wa(t0, t3); |
| 250 |
let x3 = ws(t0, t3); |
| 251 |
let x1 = wa(t0, t2); |
| 252 |
let x2 = ws(t0, t2); |
| 253 |
|
| 254 |
// odd part |
| 255 |
let i3 = in_vector[i + 3]; |
| 256 |
let i1 = in_vector[i + 1]; |
| 257 |
|
| 258 |
let p5 = wm(wa(i3, i1), f2f(1.175875602)); |
| 259 |
|
| 260 |
let p1 = wa(p5, wm(i1, -3685)); |
| 261 |
let p2 = wa(p5, wm(i3, -10497)); |
| 262 |
|
| 263 |
let t3 = wa(p5, wm(i1, 867)); |
| 264 |
let t2 = wa(p5, wm(i3, -5945)); |
| 265 |
|
| 266 |
let t1 = wa(p2, wm(i1, -1597)); |
| 267 |
let t0 = wa(p1, wm(i3, -8034)); |
| 268 |
|
| 269 |
let out: &mut [i16; 8] = out_vector |
| 270 |
.get_mut(pos..pos + 8) |
| 271 |
.unwrap() |
| 272 |
.try_into() |
| 273 |
.unwrap(); |
| 274 |
|
| 275 |
out.copy_from_slice(&[ |
| 276 |
clamp(wa(x0, t3) >> 17), |
| 277 |
clamp(wa(x1, t2) >> 17), |
| 278 |
clamp(wa(x2, t1) >> 17), |
| 279 |
clamp(wa(x3, t0) >> 17), |
| 280 |
clamp(ws(x3, t0) >> 17), |
| 281 |
clamp(ws(x2, t1) >> 17), |
| 282 |
clamp(ws(x1, t2) >> 17), |
| 283 |
clamp(ws(x0, t3) >> 17), |
| 284 |
]); |
| 285 |
|
| 286 |
pos += stride; |
| 287 |
} |
| 288 |
|
| 289 |
in_vector[32..36].fill(0); |
| 290 |
in_vector[40..44].fill(0); |
| 291 |
in_vector[48..52].fill(0); |
| 292 |
in_vector[56..60].fill(0); |
| 293 |
} |