| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 3/2025. All rights reserved. |
| 3 |
* // |
| 4 |
* // Redistribution and use in source and binary forms, with or without modification, |
| 5 |
* // are permitted provided that the following conditions are met: |
| 6 |
* // |
| 7 |
* // 1. Redistributions of source code must retain the above copyright notice, this |
| 8 |
* // list of conditions and the following disclaimer. |
| 9 |
* // |
| 10 |
* // 2. Redistributions in binary form must reproduce the above copyright notice, |
| 11 |
* // this list of conditions and the following disclaimer in the documentation |
| 12 |
* // and/or other materials provided with the distribution. |
| 13 |
* // |
| 14 |
* // 3. Neither the name of the copyright holder nor the names of its |
| 15 |
* // contributors may be used to endorse or promote products derived from |
| 16 |
* // this software without specific prior written permission. |
| 17 |
* // |
| 18 |
* // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 19 |
* // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 20 |
* // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 21 |
* // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE |
| 22 |
* // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 23 |
* // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 24 |
* // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 25 |
* // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 26 |
* // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 27 |
* // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 28 |
*/ |
| 29 |
use crate::conversions::interpolator::BarycentricWeight; |
| 30 |
use crate::math::{FusedMultiplyAdd, FusedMultiplyNegAdd}; |
| 31 |
use std::arch::x86_64::*; |
| 32 |
use std::ops::{Add, Mul, Sub}; |
| 33 |
|
| 34 |
#[repr(align(16), C)] |
| 35 |
pub(crate) struct SseAlignedF32(pub(crate) [f32; 4]); |
| 36 |
|
| 37 |
#[cfg(feature = "options")] |
| 38 |
pub(crate) struct TetrahedralAvxFma<const GRID_SIZE: usize> {} |
| 39 |
|
| 40 |
#[cfg(feature = "options")] |
| 41 |
pub(crate) struct PyramidalAvxFma<const GRID_SIZE: usize> {} |
| 42 |
|
| 43 |
#[cfg(feature = "options")] |
| 44 |
pub(crate) struct PrismaticAvxFma<const GRID_SIZE: usize> {} |
| 45 |
|
| 46 |
pub(crate) struct TrilinearAvxFma<const GRID_SIZE: usize> {} |
| 47 |
|
| 48 |
#[cfg(feature = "options")] |
| 49 |
pub(crate) struct PrismaticAvxFmaDouble<const GRID_SIZE: usize> {} |
| 50 |
|
| 51 |
pub(crate) struct TrilinearAvxFmaDouble<const GRID_SIZE: usize> {} |
| 52 |
|
| 53 |
#[cfg(feature = "options")] |
| 54 |
pub(crate) struct PyramidAvxFmaDouble<const GRID_SIZE: usize> {} |
| 55 |
|
| 56 |
#[cfg(feature = "options")] |
| 57 |
pub(crate) struct TetrahedralAvxFmaDouble<const GRID_SIZE: usize> {} |
| 58 |
|
| 59 |
pub(crate) trait AvxMdInterpolationDouble { |
| 60 |
fn inter3_sse( |
| 61 |
&self, |
| 62 |
table0: &[SseAlignedF32], |
| 63 |
table1: &[SseAlignedF32], |
| 64 |
in_r: usize, |
| 65 |
in_g: usize, |
| 66 |
in_b: usize, |
| 67 |
lut: &[BarycentricWeight<f32>], |
| 68 |
) -> (AvxVectorSse, AvxVectorSse); |
| 69 |
} |
| 70 |
|
| 71 |
pub(crate) trait AvxMdInterpolation { |
| 72 |
fn inter3_sse( |
| 73 |
&self, |
| 74 |
table: &[SseAlignedF32], |
| 75 |
in_r: usize, |
| 76 |
in_g: usize, |
| 77 |
in_b: usize, |
| 78 |
lut: &[BarycentricWeight<f32>], |
| 79 |
) -> AvxVectorSse; |
| 80 |
} |
| 81 |
|
| 82 |
trait Fetcher<T> { |
| 83 |
fn fetch(&self, x: i32, y: i32, z: i32) -> T; |
| 84 |
} |
| 85 |
|
| 86 |
#[derive(Copy, Clone)] |
| 87 |
#[repr(transparent)] |
| 88 |
pub(crate) struct AvxVectorSse { |
| 89 |
pub(crate) v: __m128, |
| 90 |
} |
| 91 |
|
| 92 |
#[derive(Copy, Clone)] |
| 93 |
#[repr(transparent)] |
| 94 |
pub(crate) struct AvxVector { |
| 95 |
pub(crate) v: __m256, |
| 96 |
} |
| 97 |
|
| 98 |
impl AvxVector { |
| 99 |
#[inline(always)] |
| 100 |
pub(crate) fn from_sse(lo: AvxVectorSse, hi: AvxVectorSse) -> AvxVector { |
| 101 |
unsafe { |
| 102 |
AvxVector { |
| 103 |
v: _mm256_insertf128_ps::<1>(_mm256_castps128_ps256(lo.v), hi.v), |
| 104 |
} |
| 105 |
} |
| 106 |
} |
| 107 |
|
| 108 |
#[inline(always)] |
| 109 |
pub(crate) fn split(self) -> (AvxVectorSse, AvxVectorSse) { |
| 110 |
unsafe { |
| 111 |
( |
| 112 |
AvxVectorSse { |
| 113 |
v: _mm256_castps256_ps128(self.v), |
| 114 |
}, |
| 115 |
AvxVectorSse { |
| 116 |
v: _mm256_extractf128_ps::<1>(self.v), |
| 117 |
}, |
| 118 |
) |
| 119 |
} |
| 120 |
} |
| 121 |
} |
| 122 |
|
| 123 |
impl From<f32> for AvxVectorSse { |
| 124 |
#[inline(always)] |
| 125 |
fn from(v: f32) -> Self { |
| 126 |
AvxVectorSse { |
| 127 |
v: unsafe { _mm_set1_ps(v) }, |
| 128 |
} |
| 129 |
} |
| 130 |
} |
| 131 |
|
| 132 |
impl From<f32> for AvxVector { |
| 133 |
#[inline(always)] |
| 134 |
fn from(v: f32) -> Self { |
| 135 |
AvxVector { |
| 136 |
v: unsafe { _mm256_set1_ps(v) }, |
| 137 |
} |
| 138 |
} |
| 139 |
} |
| 140 |
|
| 141 |
impl Sub<AvxVectorSse> for AvxVectorSse { |
| 142 |
type Output = Self; |
| 143 |
#[inline(always)] |
| 144 |
fn sub(self, rhs: AvxVectorSse) -> Self::Output { |
| 145 |
AvxVectorSse { |
| 146 |
v: unsafe { _mm_sub_ps(self.v, rhs.v) }, |
| 147 |
} |
| 148 |
} |
| 149 |
} |
| 150 |
|
| 151 |
impl Sub<AvxVector> for AvxVector { |
| 152 |
type Output = Self; |
| 153 |
#[inline(always)] |
| 154 |
fn sub(self, rhs: AvxVector) -> Self::Output { |
| 155 |
AvxVector { |
| 156 |
v: unsafe { _mm256_sub_ps(self.v, rhs.v) }, |
| 157 |
} |
| 158 |
} |
| 159 |
} |
| 160 |
|
| 161 |
impl Add<AvxVectorSse> for AvxVectorSse { |
| 162 |
type Output = Self; |
| 163 |
#[inline(always)] |
| 164 |
fn add(self, rhs: AvxVectorSse) -> Self::Output { |
| 165 |
AvxVectorSse { |
| 166 |
v: unsafe { _mm_add_ps(self.v, rhs.v) }, |
| 167 |
} |
| 168 |
} |
| 169 |
} |
| 170 |
|
| 171 |
impl Mul<AvxVectorSse> for AvxVectorSse { |
| 172 |
type Output = Self; |
| 173 |
#[inline(always)] |
| 174 |
fn mul(self, rhs: AvxVectorSse) -> Self::Output { |
| 175 |
AvxVectorSse { |
| 176 |
v: unsafe { _mm_mul_ps(self.v, rhs.v) }, |
| 177 |
} |
| 178 |
} |
| 179 |
} |
| 180 |
|
| 181 |
impl AvxVector { |
| 182 |
#[inline(always)] |
| 183 |
pub(crate) fn neg_mla(self, b: AvxVector, c: AvxVector) -> Self { |
| 184 |
Self { |
| 185 |
v: unsafe { _mm256_fnmadd_ps(b.v, c.v, self.v) }, |
| 186 |
} |
| 187 |
} |
| 188 |
} |
| 189 |
|
| 190 |
impl FusedMultiplyNegAdd<AvxVectorSse> for AvxVectorSse { |
| 191 |
#[inline(always)] |
| 192 |
fn neg_mla(&self, b: AvxVectorSse, c: AvxVectorSse) -> Self { |
| 193 |
Self { |
| 194 |
v: unsafe { _mm_fnmadd_ps(b.v, c.v, self.v) }, |
| 195 |
} |
| 196 |
} |
| 197 |
} |
| 198 |
|
| 199 |
impl Add<AvxVector> for AvxVector { |
| 200 |
type Output = Self; |
| 201 |
#[inline(always)] |
| 202 |
fn add(self, rhs: AvxVector) -> Self::Output { |
| 203 |
AvxVector { |
| 204 |
v: unsafe { _mm256_add_ps(self.v, rhs.v) }, |
| 205 |
} |
| 206 |
} |
| 207 |
} |
| 208 |
|
| 209 |
impl Mul<AvxVector> for AvxVector { |
| 210 |
type Output = Self; |
| 211 |
#[inline(always)] |
| 212 |
fn mul(self, rhs: AvxVector) -> Self::Output { |
| 213 |
AvxVector { |
| 214 |
v: unsafe { _mm256_mul_ps(self.v, rhs.v) }, |
| 215 |
} |
| 216 |
} |
| 217 |
} |
| 218 |
|
| 219 |
impl FusedMultiplyAdd<AvxVectorSse> for AvxVectorSse { |
| 220 |
#[inline(always)] |
| 221 |
fn mla(&self, b: AvxVectorSse, c: AvxVectorSse) -> AvxVectorSse { |
| 222 |
AvxVectorSse { |
| 223 |
v: unsafe { _mm_fmadd_ps(b.v, c.v, self.v) }, |
| 224 |
} |
| 225 |
} |
| 226 |
} |
| 227 |
|
| 228 |
impl FusedMultiplyAdd<AvxVector> for AvxVector { |
| 229 |
#[inline(always)] |
| 230 |
fn mla(&self, b: AvxVector, c: AvxVector) -> AvxVector { |
| 231 |
AvxVector { |
| 232 |
v: unsafe { _mm256_fmadd_ps(b.v, c.v, self.v) }, |
| 233 |
} |
| 234 |
} |
| 235 |
} |
| 236 |
|
| 237 |
struct TetrahedralAvxSseFetchVector<'a, const GRID_SIZE: usize> { |
| 238 |
cube: &'a [SseAlignedF32], |
| 239 |
} |
| 240 |
|
| 241 |
struct TetrahedralAvxFetchVector<'a, const GRID_SIZE: usize> { |
| 242 |
cube0: &'a [SseAlignedF32], |
| 243 |
cube1: &'a [SseAlignedF32], |
| 244 |
} |
| 245 |
|
| 246 |
impl<const GRID_SIZE: usize> Fetcher<AvxVector> for TetrahedralAvxFetchVector<'_, GRID_SIZE> { |
| 247 |
#[inline(always)] |
| 248 |
fn fetch(&self, x: i32, y: i32, z: i32) -> AvxVector { |
| 249 |
let offset = (x as u32 * (GRID_SIZE as u32 * GRID_SIZE as u32) |
| 250 |
+ y as u32 * GRID_SIZE as u32 |
| 251 |
+ z as u32) as usize; |
| 252 |
let jx0 = unsafe { self.cube0.get_unchecked(offset..) }; |
| 253 |
let jx1 = unsafe { self.cube1.get_unchecked(offset..) }; |
| 254 |
AvxVector { |
| 255 |
v: unsafe { |
| 256 |
_mm256_insertf128_ps::<1>( |
| 257 |
_mm256_castps128_ps256(_mm_load_ps(jx0.as_ptr() as *const f32)), |
| 258 |
_mm_load_ps(jx1.as_ptr() as *const f32), |
| 259 |
) |
| 260 |
}, |
| 261 |
} |
| 262 |
} |
| 263 |
} |
| 264 |
|
| 265 |
impl<const GRID_SIZE: usize> Fetcher<AvxVectorSse> for TetrahedralAvxSseFetchVector<'_, GRID_SIZE> { |
| 266 |
#[inline(always)] |
| 267 |
fn fetch(&self, x: i32, y: i32, z: i32) -> AvxVectorSse { |
| 268 |
let offset = (x as u32 * (GRID_SIZE as u32 * GRID_SIZE as u32) |
| 269 |
+ y as u32 * GRID_SIZE as u32 |
| 270 |
+ z as u32) as usize; |
| 271 |
let jx = unsafe { self.cube.get_unchecked(offset..) }; |
| 272 |
AvxVectorSse { |
| 273 |
v: unsafe { _mm_load_ps(jx.as_ptr() as *const f32) }, |
| 274 |
} |
| 275 |
} |
| 276 |
} |
| 277 |
|
| 278 |
#[cfg(feature = "options")] |
| 279 |
impl<const GRID_SIZE: usize> TetrahedralAvxFma<GRID_SIZE> { |
| 280 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 281 |
unsafe fn interpolate( |
| 282 |
&self, |
| 283 |
in_r: usize, |
| 284 |
in_g: usize, |
| 285 |
in_b: usize, |
| 286 |
lut: &[BarycentricWeight<f32>], |
| 287 |
r: impl Fetcher<AvxVectorSse>, |
| 288 |
) -> AvxVectorSse { |
| 289 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 290 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 291 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 292 |
|
| 293 |
let x: i32 = lut_r.x; |
| 294 |
let y: i32 = lut_g.x; |
| 295 |
let z: i32 = lut_b.x; |
| 296 |
|
| 297 |
let x_n: i32 = lut_r.x_n; |
| 298 |
let y_n: i32 = lut_g.x_n; |
| 299 |
let z_n: i32 = lut_b.x_n; |
| 300 |
|
| 301 |
let rx = lut_r.w; |
| 302 |
let ry = lut_g.w; |
| 303 |
let rz = lut_b.w; |
| 304 |
|
| 305 |
let c0 = r.fetch(x, y, z); |
| 306 |
|
| 307 |
let c2; |
| 308 |
let c1; |
| 309 |
let c3; |
| 310 |
if rx >= ry { |
| 311 |
if ry >= rz { |
| 312 |
//rx >= ry && ry >= rz |
| 313 |
c1 = r.fetch(x_n, y, z) - c0; |
| 314 |
c2 = r.fetch(x_n, y_n, z) - r.fetch(x_n, y, z); |
| 315 |
c3 = r.fetch(x_n, y_n, z_n) - r.fetch(x_n, y_n, z); |
| 316 |
} else if rx >= rz { |
| 317 |
//rx >= rz && rz >= ry |
| 318 |
c1 = r.fetch(x_n, y, z) - c0; |
| 319 |
c2 = r.fetch(x_n, y_n, z_n) - r.fetch(x_n, y, z_n); |
| 320 |
c3 = r.fetch(x_n, y, z_n) - r.fetch(x_n, y, z); |
| 321 |
} else { |
| 322 |
//rz > rx && rx >= ry |
| 323 |
c1 = r.fetch(x_n, y, z_n) - r.fetch(x, y, z_n); |
| 324 |
c2 = r.fetch(x_n, y_n, z_n) - r.fetch(x_n, y, z_n); |
| 325 |
c3 = r.fetch(x, y, z_n) - c0; |
| 326 |
} |
| 327 |
} else if rx >= rz { |
| 328 |
//ry > rx && rx >= rz |
| 329 |
c1 = r.fetch(x_n, y_n, z) - r.fetch(x, y_n, z); |
| 330 |
c2 = r.fetch(x, y_n, z) - c0; |
| 331 |
c3 = r.fetch(x_n, y_n, z_n) - r.fetch(x_n, y_n, z); |
| 332 |
} else if ry >= rz { |
| 333 |
//ry >= rz && rz > rx |
| 334 |
c1 = r.fetch(x_n, y_n, z_n) - r.fetch(x, y_n, z_n); |
| 335 |
c2 = r.fetch(x, y_n, z) - c0; |
| 336 |
c3 = r.fetch(x, y_n, z_n) - r.fetch(x, y_n, z); |
| 337 |
} else { |
| 338 |
//rz > ry && ry > rx |
| 339 |
c1 = r.fetch(x_n, y_n, z_n) - r.fetch(x, y_n, z_n); |
| 340 |
c2 = r.fetch(x, y_n, z_n) - r.fetch(x, y, z_n); |
| 341 |
c3 = r.fetch(x, y, z_n) - c0; |
| 342 |
} |
| 343 |
let s0 = c0.mla(c1, AvxVectorSse::from(rx)); |
| 344 |
let s1 = s0.mla(c2, AvxVectorSse::from(ry)); |
| 345 |
s1.mla(c3, AvxVectorSse::from(rz)) |
| 346 |
} |
| 347 |
} |
| 348 |
|
| 349 |
macro_rules! define_interp_avx { |
| 350 |
($interpolator: ident) => { |
| 351 |
impl<const GRID_SIZE: usize> AvxMdInterpolation for $interpolator<GRID_SIZE> { |
| 352 |
fn inter3_sse( |
| 353 |
&self, |
| 354 |
table: &[SseAlignedF32], |
| 355 |
in_r: usize, |
| 356 |
in_g: usize, |
| 357 |
in_b: usize, |
| 358 |
lut: &[BarycentricWeight<f32>], |
| 359 |
) -> AvxVectorSse { |
| 360 |
unsafe { |
| 361 |
self.interpolate( |
| 362 |
in_r, |
| 363 |
in_g, |
| 364 |
in_b, |
| 365 |
lut, |
| 366 |
TetrahedralAvxSseFetchVector::<GRID_SIZE> { cube: table }, |
| 367 |
) |
| 368 |
} |
| 369 |
} |
| 370 |
} |
| 371 |
}; |
| 372 |
} |
| 373 |
|
| 374 |
#[cfg(feature = "options")] |
| 375 |
macro_rules! define_interp_avx_d { |
| 376 |
($interpolator: ident) => { |
| 377 |
impl<const GRID_SIZE: usize> AvxMdInterpolationDouble for $interpolator<GRID_SIZE> { |
| 378 |
fn inter3_sse( |
| 379 |
&self, |
| 380 |
table0: &[SseAlignedF32], |
| 381 |
table1: &[SseAlignedF32], |
| 382 |
in_r: usize, |
| 383 |
in_g: usize, |
| 384 |
in_b: usize, |
| 385 |
lut: &[BarycentricWeight<f32>], |
| 386 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 387 |
unsafe { |
| 388 |
self.interpolate( |
| 389 |
in_r, |
| 390 |
in_g, |
| 391 |
in_b, |
| 392 |
lut, |
| 393 |
TetrahedralAvxSseFetchVector::<GRID_SIZE> { cube: table0 }, |
| 394 |
TetrahedralAvxSseFetchVector::<GRID_SIZE> { cube: table1 }, |
| 395 |
) |
| 396 |
} |
| 397 |
} |
| 398 |
} |
| 399 |
}; |
| 400 |
} |
| 401 |
|
| 402 |
#[cfg(feature = "options")] |
| 403 |
define_interp_avx!(TetrahedralAvxFma); |
| 404 |
#[cfg(feature = "options")] |
| 405 |
define_interp_avx!(PyramidalAvxFma); |
| 406 |
#[cfg(feature = "options")] |
| 407 |
define_interp_avx!(PrismaticAvxFma); |
| 408 |
define_interp_avx!(TrilinearAvxFma); |
| 409 |
#[cfg(feature = "options")] |
| 410 |
define_interp_avx_d!(PrismaticAvxFmaDouble); |
| 411 |
#[cfg(feature = "options")] |
| 412 |
define_interp_avx_d!(PyramidAvxFmaDouble); |
| 413 |
|
| 414 |
#[cfg(feature = "options")] |
| 415 |
impl<const GRID_SIZE: usize> AvxMdInterpolationDouble for TetrahedralAvxFmaDouble<GRID_SIZE> { |
| 416 |
fn inter3_sse( |
| 417 |
&self, |
| 418 |
table0: &[SseAlignedF32], |
| 419 |
table1: &[SseAlignedF32], |
| 420 |
in_r: usize, |
| 421 |
in_g: usize, |
| 422 |
in_b: usize, |
| 423 |
lut: &[BarycentricWeight<f32>], |
| 424 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 425 |
unsafe { |
| 426 |
self.interpolate( |
| 427 |
in_r, |
| 428 |
in_g, |
| 429 |
in_b, |
| 430 |
lut, |
| 431 |
TetrahedralAvxFetchVector::<GRID_SIZE> { |
| 432 |
cube0: table0, |
| 433 |
cube1: table1, |
| 434 |
}, |
| 435 |
) |
| 436 |
} |
| 437 |
} |
| 438 |
} |
| 439 |
|
| 440 |
impl<const GRID_SIZE: usize> AvxMdInterpolationDouble for TrilinearAvxFmaDouble<GRID_SIZE> { |
| 441 |
fn inter3_sse( |
| 442 |
&self, |
| 443 |
table0: &[SseAlignedF32], |
| 444 |
table1: &[SseAlignedF32], |
| 445 |
in_r: usize, |
| 446 |
in_g: usize, |
| 447 |
in_b: usize, |
| 448 |
lut: &[BarycentricWeight<f32>], |
| 449 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 450 |
unsafe { |
| 451 |
self.interpolate( |
| 452 |
in_r, |
| 453 |
in_g, |
| 454 |
in_b, |
| 455 |
lut, |
| 456 |
TetrahedralAvxFetchVector::<GRID_SIZE> { |
| 457 |
cube0: table0, |
| 458 |
cube1: table1, |
| 459 |
}, |
| 460 |
) |
| 461 |
} |
| 462 |
} |
| 463 |
} |
| 464 |
|
| 465 |
#[cfg(feature = "options")] |
| 466 |
impl<const GRID_SIZE: usize> PyramidalAvxFma<GRID_SIZE> { |
| 467 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 468 |
unsafe fn interpolate( |
| 469 |
&self, |
| 470 |
in_r: usize, |
| 471 |
in_g: usize, |
| 472 |
in_b: usize, |
| 473 |
lut: &[BarycentricWeight<f32>], |
| 474 |
r: impl Fetcher<AvxVectorSse>, |
| 475 |
) -> AvxVectorSse { |
| 476 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 477 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 478 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 479 |
|
| 480 |
let x: i32 = lut_r.x; |
| 481 |
let y: i32 = lut_g.x; |
| 482 |
let z: i32 = lut_b.x; |
| 483 |
|
| 484 |
let x_n: i32 = lut_r.x_n; |
| 485 |
let y_n: i32 = lut_g.x_n; |
| 486 |
let z_n: i32 = lut_b.x_n; |
| 487 |
|
| 488 |
let dr = lut_r.w; |
| 489 |
let dg = lut_g.w; |
| 490 |
let db = lut_b.w; |
| 491 |
|
| 492 |
let c0 = r.fetch(x, y, z); |
| 493 |
|
| 494 |
let w0 = AvxVectorSse::from(db); |
| 495 |
let w1 = AvxVectorSse::from(dr); |
| 496 |
let w2 = AvxVectorSse::from(dg); |
| 497 |
|
| 498 |
if dr > db && dg > db { |
| 499 |
let w3 = AvxVectorSse::from(dr * dg); |
| 500 |
let x0 = r.fetch(x_n, y_n, z_n); |
| 501 |
let x1 = r.fetch(x_n, y_n, z); |
| 502 |
let x2 = r.fetch(x_n, y, z); |
| 503 |
let x3 = r.fetch(x, y_n, z); |
| 504 |
|
| 505 |
let c1 = x0 - x1; |
| 506 |
let c2 = x2 - c0; |
| 507 |
let c3 = x3 - c0; |
| 508 |
let c4 = c0 - x3 - x2 + x1; |
| 509 |
|
| 510 |
let s0 = c0.mla(c1, w0); |
| 511 |
let s1 = s0.mla(c2, w1); |
| 512 |
let s2 = s1.mla(c3, w2); |
| 513 |
s2.mla(c4, w3) |
| 514 |
} else if db > dr && dg > dr { |
| 515 |
let w3 = AvxVectorSse::from(dg * db); |
| 516 |
|
| 517 |
let x0 = r.fetch(x, y, z_n); |
| 518 |
let x1 = r.fetch(x_n, y_n, z_n); |
| 519 |
let x2 = r.fetch(x, y_n, z_n); |
| 520 |
let x3 = r.fetch(x, y_n, z); |
| 521 |
|
| 522 |
let c1 = x0 - c0; |
| 523 |
let c2 = x1 - x2; |
| 524 |
let c3 = x3 - c0; |
| 525 |
let c4 = c0 - x3 - x0 + x2; |
| 526 |
|
| 527 |
let s0 = c0.mla(c1, w0); |
| 528 |
let s1 = s0.mla(c2, w1); |
| 529 |
let s2 = s1.mla(c3, w2); |
| 530 |
s2.mla(c4, w3) |
| 531 |
} else { |
| 532 |
let w3 = AvxVectorSse::from(db * dr); |
| 533 |
|
| 534 |
let x0 = r.fetch(x, y, z_n); |
| 535 |
let x1 = r.fetch(x_n, y, z); |
| 536 |
let x2 = r.fetch(x_n, y, z_n); |
| 537 |
let x3 = r.fetch(x_n, y_n, z_n); |
| 538 |
|
| 539 |
let c1 = x0 - c0; |
| 540 |
let c2 = x1 - c0; |
| 541 |
let c3 = x3 - x2; |
| 542 |
let c4 = c0 - x1 - x0 + x2; |
| 543 |
|
| 544 |
let s0 = c0.mla(c1, w0); |
| 545 |
let s1 = s0.mla(c2, w1); |
| 546 |
let s2 = s1.mla(c3, w2); |
| 547 |
s2.mla(c4, w3) |
| 548 |
} |
| 549 |
} |
| 550 |
} |
| 551 |
|
| 552 |
#[cfg(feature = "options")] |
| 553 |
impl<const GRID_SIZE: usize> PrismaticAvxFma<GRID_SIZE> { |
| 554 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 555 |
unsafe fn interpolate( |
| 556 |
&self, |
| 557 |
in_r: usize, |
| 558 |
in_g: usize, |
| 559 |
in_b: usize, |
| 560 |
lut: &[BarycentricWeight<f32>], |
| 561 |
r: impl Fetcher<AvxVectorSse>, |
| 562 |
) -> AvxVectorSse { |
| 563 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 564 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 565 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 566 |
|
| 567 |
let x: i32 = lut_r.x; |
| 568 |
let y: i32 = lut_g.x; |
| 569 |
let z: i32 = lut_b.x; |
| 570 |
|
| 571 |
let x_n: i32 = lut_r.x_n; |
| 572 |
let y_n: i32 = lut_g.x_n; |
| 573 |
let z_n: i32 = lut_b.x_n; |
| 574 |
|
| 575 |
let dr = lut_r.w; |
| 576 |
let dg = lut_g.w; |
| 577 |
let db = lut_b.w; |
| 578 |
|
| 579 |
let c0 = r.fetch(x, y, z); |
| 580 |
|
| 581 |
let w0 = AvxVectorSse::from(db); |
| 582 |
let w1 = AvxVectorSse::from(dr); |
| 583 |
let w2 = AvxVectorSse::from(dg); |
| 584 |
let w3 = AvxVectorSse::from(dg * db); |
| 585 |
let w4 = AvxVectorSse::from(dr * dg); |
| 586 |
|
| 587 |
if db > dr { |
| 588 |
let x0 = r.fetch(x, y, z_n); |
| 589 |
let x1 = r.fetch(x_n, y, z_n); |
| 590 |
let x2 = r.fetch(x, y_n, z); |
| 591 |
let x3 = r.fetch(x, y_n, z_n); |
| 592 |
let x4 = r.fetch(x_n, y_n, z_n); |
| 593 |
|
| 594 |
let c1 = x0 - c0; |
| 595 |
let c2 = x1 - x0; |
| 596 |
let c3 = x2 - c0; |
| 597 |
let c4 = c0 - x2 - x0 + x3; |
| 598 |
let c5 = x0 - x3 - x1 + x4; |
| 599 |
|
| 600 |
let s0 = c0.mla(c1, w0); |
| 601 |
let s1 = s0.mla(c2, w1); |
| 602 |
let s2 = s1.mla(c3, w2); |
| 603 |
let s3 = s2.mla(c4, w3); |
| 604 |
s3.mla(c5, w4) |
| 605 |
} else { |
| 606 |
let x0 = r.fetch(x_n, y, z); |
| 607 |
let x1 = r.fetch(x_n, y, z_n); |
| 608 |
let x2 = r.fetch(x, y_n, z); |
| 609 |
let x3 = r.fetch(x_n, y_n, z); |
| 610 |
let x4 = r.fetch(x_n, y_n, z_n); |
| 611 |
|
| 612 |
let c1 = x1 - x0; |
| 613 |
let c2 = x0 - c0; |
| 614 |
let c3 = x2 - c0; |
| 615 |
let c4 = x0 - x3 - x1 + x4; |
| 616 |
let c5 = c0 - x2 - x0 + x3; |
| 617 |
|
| 618 |
let s0 = c0.mla(c1, w0); |
| 619 |
let s1 = s0.mla(c2, w1); |
| 620 |
let s2 = s1.mla(c3, w2); |
| 621 |
let s3 = s2.mla(c4, w3); |
| 622 |
s3.mla(c5, w4) |
| 623 |
} |
| 624 |
} |
| 625 |
} |
| 626 |
|
| 627 |
#[cfg(feature = "options")] |
| 628 |
impl<const GRID_SIZE: usize> PrismaticAvxFmaDouble<GRID_SIZE> { |
| 629 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 630 |
unsafe fn interpolate( |
| 631 |
&self, |
| 632 |
in_r: usize, |
| 633 |
in_g: usize, |
| 634 |
in_b: usize, |
| 635 |
lut: &[BarycentricWeight<f32>], |
| 636 |
r0: impl Fetcher<AvxVectorSse>, |
| 637 |
r1: impl Fetcher<AvxVectorSse>, |
| 638 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 639 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 640 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 641 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 642 |
|
| 643 |
let x: i32 = lut_r.x; |
| 644 |
let y: i32 = lut_g.x; |
| 645 |
let z: i32 = lut_b.x; |
| 646 |
|
| 647 |
let x_n: i32 = lut_r.x_n; |
| 648 |
let y_n: i32 = lut_g.x_n; |
| 649 |
let z_n: i32 = lut_b.x_n; |
| 650 |
|
| 651 |
let dr = lut_r.w; |
| 652 |
let dg = lut_g.w; |
| 653 |
let db = lut_b.w; |
| 654 |
|
| 655 |
let c0_0 = r0.fetch(x, y, z); |
| 656 |
let c0_1 = r0.fetch(x, y, z); |
| 657 |
|
| 658 |
let w0 = AvxVector::from(db); |
| 659 |
let w1 = AvxVector::from(dr); |
| 660 |
let w2 = AvxVector::from(dg); |
| 661 |
let w3 = AvxVector::from(dg * db); |
| 662 |
let w4 = AvxVector::from(dr * dg); |
| 663 |
|
| 664 |
let c0 = AvxVector::from_sse(c0_0, c0_1); |
| 665 |
|
| 666 |
if db > dr { |
| 667 |
let x0_0 = r0.fetch(x, y, z_n); |
| 668 |
let x1_0 = r0.fetch(x_n, y, z_n); |
| 669 |
let x2_0 = r0.fetch(x, y_n, z); |
| 670 |
let x3_0 = r0.fetch(x, y_n, z_n); |
| 671 |
let x4_0 = r0.fetch(x_n, y_n, z_n); |
| 672 |
|
| 673 |
let x0_1 = r1.fetch(x, y, z_n); |
| 674 |
let x1_1 = r1.fetch(x_n, y, z_n); |
| 675 |
let x2_1 = r1.fetch(x, y_n, z); |
| 676 |
let x3_1 = r1.fetch(x, y_n, z_n); |
| 677 |
let x4_1 = r1.fetch(x_n, y_n, z_n); |
| 678 |
|
| 679 |
let x0 = AvxVector::from_sse(x0_0, x0_1); |
| 680 |
let x1 = AvxVector::from_sse(x1_0, x1_1); |
| 681 |
let x2 = AvxVector::from_sse(x2_0, x2_1); |
| 682 |
let x3 = AvxVector::from_sse(x3_0, x3_1); |
| 683 |
let x4 = AvxVector::from_sse(x4_0, x4_1); |
| 684 |
|
| 685 |
let c1 = x0 - c0; |
| 686 |
let c2 = x1 - x0; |
| 687 |
let c3 = x2 - c0; |
| 688 |
let c4 = c0 - x2 - x0 + x3; |
| 689 |
let c5 = x0 - x3 - x1 + x4; |
| 690 |
|
| 691 |
let s0 = c0.mla(c1, w0); |
| 692 |
let s1 = s0.mla(c2, w1); |
| 693 |
let s2 = s1.mla(c3, w2); |
| 694 |
let s3 = s2.mla(c4, w3); |
| 695 |
s3.mla(c5, w4).split() |
| 696 |
} else { |
| 697 |
let x0_0 = r0.fetch(x_n, y, z); |
| 698 |
let x1_0 = r0.fetch(x_n, y, z_n); |
| 699 |
let x2_0 = r0.fetch(x, y_n, z); |
| 700 |
let x3_0 = r0.fetch(x_n, y_n, z); |
| 701 |
let x4_0 = r0.fetch(x_n, y_n, z_n); |
| 702 |
|
| 703 |
let x0_1 = r1.fetch(x_n, y, z); |
| 704 |
let x1_1 = r1.fetch(x_n, y, z_n); |
| 705 |
let x2_1 = r1.fetch(x, y_n, z); |
| 706 |
let x3_1 = r1.fetch(x_n, y_n, z); |
| 707 |
let x4_1 = r1.fetch(x_n, y_n, z_n); |
| 708 |
|
| 709 |
let x0 = AvxVector::from_sse(x0_0, x0_1); |
| 710 |
let x1 = AvxVector::from_sse(x1_0, x1_1); |
| 711 |
let x2 = AvxVector::from_sse(x2_0, x2_1); |
| 712 |
let x3 = AvxVector::from_sse(x3_0, x3_1); |
| 713 |
let x4 = AvxVector::from_sse(x4_0, x4_1); |
| 714 |
|
| 715 |
let c1 = x1 - x0; |
| 716 |
let c2 = x0 - c0; |
| 717 |
let c3 = x2 - c0; |
| 718 |
let c4 = x0 - x3 - x1 + x4; |
| 719 |
let c5 = c0 - x2 - x0 + x3; |
| 720 |
|
| 721 |
let s0 = c0.mla(c1, w0); |
| 722 |
let s1 = s0.mla(c2, w1); |
| 723 |
let s2 = s1.mla(c3, w2); |
| 724 |
let s3 = s2.mla(c4, w3); |
| 725 |
s3.mla(c5, w4).split() |
| 726 |
} |
| 727 |
} |
| 728 |
} |
| 729 |
|
| 730 |
#[cfg(feature = "options")] |
| 731 |
impl<const GRID_SIZE: usize> PyramidAvxFmaDouble<GRID_SIZE> { |
| 732 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 733 |
unsafe fn interpolate( |
| 734 |
&self, |
| 735 |
in_r: usize, |
| 736 |
in_g: usize, |
| 737 |
in_b: usize, |
| 738 |
lut: &[BarycentricWeight<f32>], |
| 739 |
r0: impl Fetcher<AvxVectorSse>, |
| 740 |
r1: impl Fetcher<AvxVectorSse>, |
| 741 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 742 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 743 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 744 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 745 |
|
| 746 |
let x: i32 = lut_r.x; |
| 747 |
let y: i32 = lut_g.x; |
| 748 |
let z: i32 = lut_b.x; |
| 749 |
|
| 750 |
let x_n: i32 = lut_r.x_n; |
| 751 |
let y_n: i32 = lut_g.x_n; |
| 752 |
let z_n: i32 = lut_b.x_n; |
| 753 |
|
| 754 |
let dr = lut_r.w; |
| 755 |
let dg = lut_g.w; |
| 756 |
let db = lut_b.w; |
| 757 |
|
| 758 |
let c0_0 = r0.fetch(x, y, z); |
| 759 |
let c0_1 = r1.fetch(x, y, z); |
| 760 |
|
| 761 |
let w0 = AvxVector::from(db); |
| 762 |
let w1 = AvxVector::from(dr); |
| 763 |
let w2 = AvxVector::from(dg); |
| 764 |
|
| 765 |
let c0 = AvxVector::from_sse(c0_0, c0_1); |
| 766 |
|
| 767 |
if dr > db && dg > db { |
| 768 |
let w3 = AvxVector::from(dr * dg); |
| 769 |
|
| 770 |
let x0_0 = r0.fetch(x_n, y_n, z_n); |
| 771 |
let x1_0 = r0.fetch(x_n, y_n, z); |
| 772 |
let x2_0 = r0.fetch(x_n, y, z); |
| 773 |
let x3_0 = r0.fetch(x, y_n, z); |
| 774 |
|
| 775 |
let x0_1 = r1.fetch(x_n, y_n, z_n); |
| 776 |
let x1_1 = r1.fetch(x_n, y_n, z); |
| 777 |
let x2_1 = r1.fetch(x_n, y, z); |
| 778 |
let x3_1 = r1.fetch(x, y_n, z); |
| 779 |
|
| 780 |
let x0 = AvxVector::from_sse(x0_0, x0_1); |
| 781 |
let x1 = AvxVector::from_sse(x1_0, x1_1); |
| 782 |
let x2 = AvxVector::from_sse(x2_0, x2_1); |
| 783 |
let x3 = AvxVector::from_sse(x3_0, x3_1); |
| 784 |
|
| 785 |
let c1 = x0 - x1; |
| 786 |
let c2 = x2 - c0; |
| 787 |
let c3 = x3 - c0; |
| 788 |
let c4 = c0 - x3 - x2 + x1; |
| 789 |
|
| 790 |
let s0 = c0.mla(c1, w0); |
| 791 |
let s1 = s0.mla(c2, w1); |
| 792 |
let s2 = s1.mla(c3, w2); |
| 793 |
s2.mla(c4, w3).split() |
| 794 |
} else if db > dr && dg > dr { |
| 795 |
let w3 = AvxVector::from(dg * db); |
| 796 |
|
| 797 |
let x0_0 = r0.fetch(x, y, z_n); |
| 798 |
let x1_0 = r0.fetch(x_n, y_n, z_n); |
| 799 |
let x2_0 = r0.fetch(x, y_n, z_n); |
| 800 |
let x3_0 = r0.fetch(x, y_n, z); |
| 801 |
|
| 802 |
let x0_1 = r1.fetch(x, y, z_n); |
| 803 |
let x1_1 = r1.fetch(x_n, y_n, z_n); |
| 804 |
let x2_1 = r1.fetch(x, y_n, z_n); |
| 805 |
let x3_1 = r1.fetch(x, y_n, z); |
| 806 |
|
| 807 |
let x0 = AvxVector::from_sse(x0_0, x0_1); |
| 808 |
let x1 = AvxVector::from_sse(x1_0, x1_1); |
| 809 |
let x2 = AvxVector::from_sse(x2_0, x2_1); |
| 810 |
let x3 = AvxVector::from_sse(x3_0, x3_1); |
| 811 |
|
| 812 |
let c1 = x0 - c0; |
| 813 |
let c2 = x1 - x2; |
| 814 |
let c3 = x3 - c0; |
| 815 |
let c4 = c0 - x3 - x0 + x2; |
| 816 |
|
| 817 |
let s0 = c0.mla(c1, w0); |
| 818 |
let s1 = s0.mla(c2, w1); |
| 819 |
let s2 = s1.mla(c3, w2); |
| 820 |
s2.mla(c4, w3).split() |
| 821 |
} else { |
| 822 |
let w3 = AvxVector::from(db * dr); |
| 823 |
|
| 824 |
let x0_0 = r0.fetch(x, y, z_n); |
| 825 |
let x1_0 = r0.fetch(x_n, y, z); |
| 826 |
let x2_0 = r0.fetch(x_n, y, z_n); |
| 827 |
let x3_0 = r0.fetch(x_n, y_n, z_n); |
| 828 |
|
| 829 |
let x0_1 = r1.fetch(x, y, z_n); |
| 830 |
let x1_1 = r1.fetch(x_n, y, z); |
| 831 |
let x2_1 = r1.fetch(x_n, y, z_n); |
| 832 |
let x3_1 = r1.fetch(x_n, y_n, z_n); |
| 833 |
|
| 834 |
let x0 = AvxVector::from_sse(x0_0, x0_1); |
| 835 |
let x1 = AvxVector::from_sse(x1_0, x1_1); |
| 836 |
let x2 = AvxVector::from_sse(x2_0, x2_1); |
| 837 |
let x3 = AvxVector::from_sse(x3_0, x3_1); |
| 838 |
|
| 839 |
let c1 = x0 - c0; |
| 840 |
let c2 = x1 - c0; |
| 841 |
let c3 = x3 - x2; |
| 842 |
let c4 = c0 - x1 - x0 + x2; |
| 843 |
|
| 844 |
let s0 = c0.mla(c1, w0); |
| 845 |
let s1 = s0.mla(c2, w1); |
| 846 |
let s2 = s1.mla(c3, w2); |
| 847 |
s2.mla(c4, w3).split() |
| 848 |
} |
| 849 |
} |
| 850 |
} |
| 851 |
|
| 852 |
#[cfg(feature = "options")] |
| 853 |
impl<const GRID_SIZE: usize> TetrahedralAvxFmaDouble<GRID_SIZE> { |
| 854 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 855 |
unsafe fn interpolate( |
| 856 |
&self, |
| 857 |
in_r: usize, |
| 858 |
in_g: usize, |
| 859 |
in_b: usize, |
| 860 |
lut: &[BarycentricWeight<f32>], |
| 861 |
rv: impl Fetcher<AvxVector>, |
| 862 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 863 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 864 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 865 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 866 |
|
| 867 |
let x: i32 = lut_r.x; |
| 868 |
let y: i32 = lut_g.x; |
| 869 |
let z: i32 = lut_b.x; |
| 870 |
|
| 871 |
let x_n: i32 = lut_r.x_n; |
| 872 |
let y_n: i32 = lut_g.x_n; |
| 873 |
let z_n: i32 = lut_b.x_n; |
| 874 |
|
| 875 |
let rx = lut_r.w; |
| 876 |
let ry = lut_g.w; |
| 877 |
let rz = lut_b.w; |
| 878 |
|
| 879 |
let c0 = rv.fetch(x, y, z); |
| 880 |
|
| 881 |
let w0 = AvxVector::from(rx); |
| 882 |
let w1 = AvxVector::from(ry); |
| 883 |
let w2 = AvxVector::from(rz); |
| 884 |
|
| 885 |
let c2; |
| 886 |
let c1; |
| 887 |
let c3; |
| 888 |
if rx >= ry { |
| 889 |
if ry >= rz { |
| 890 |
//rx >= ry && ry >= rz |
| 891 |
c1 = rv.fetch(x_n, y, z) - c0; |
| 892 |
c2 = rv.fetch(x_n, y_n, z) - rv.fetch(x_n, y, z); |
| 893 |
c3 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x_n, y_n, z); |
| 894 |
} else if rx >= rz { |
| 895 |
//rx >= rz && rz >= ry |
| 896 |
c1 = rv.fetch(x_n, y, z) - c0; |
| 897 |
c2 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x_n, y, z_n); |
| 898 |
c3 = rv.fetch(x_n, y, z_n) - rv.fetch(x_n, y, z); |
| 899 |
} else { |
| 900 |
//rz > rx && rx >= ry |
| 901 |
c1 = rv.fetch(x_n, y, z_n) - rv.fetch(x, y, z_n); |
| 902 |
c2 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x_n, y, z_n); |
| 903 |
c3 = rv.fetch(x, y, z_n) - c0; |
| 904 |
} |
| 905 |
} else if rx >= rz { |
| 906 |
//ry > rx && rx >= rz |
| 907 |
c1 = rv.fetch(x_n, y_n, z) - rv.fetch(x, y_n, z); |
| 908 |
c2 = rv.fetch(x, y_n, z) - c0; |
| 909 |
c3 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x_n, y_n, z); |
| 910 |
} else if ry >= rz { |
| 911 |
//ry >= rz && rz > rx |
| 912 |
c1 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x, y_n, z_n); |
| 913 |
c2 = rv.fetch(x, y_n, z) - c0; |
| 914 |
c3 = rv.fetch(x, y_n, z_n) - rv.fetch(x, y_n, z); |
| 915 |
} else { |
| 916 |
//rz > ry && ry > rx |
| 917 |
c1 = rv.fetch(x_n, y_n, z_n) - rv.fetch(x, y_n, z_n); |
| 918 |
c2 = rv.fetch(x, y_n, z_n) - rv.fetch(x, y, z_n); |
| 919 |
c3 = rv.fetch(x, y, z_n) - c0; |
| 920 |
} |
| 921 |
let s0 = c0.mla(c1, w0); |
| 922 |
let s1 = s0.mla(c2, w1); |
| 923 |
s1.mla(c3, w2).split() |
| 924 |
} |
| 925 |
} |
| 926 |
|
| 927 |
impl<const GRID_SIZE: usize> TrilinearAvxFmaDouble<GRID_SIZE> { |
| 928 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 929 |
unsafe fn interpolate( |
| 930 |
&self, |
| 931 |
in_r: usize, |
| 932 |
in_g: usize, |
| 933 |
in_b: usize, |
| 934 |
lut: &[BarycentricWeight<f32>], |
| 935 |
rv: impl Fetcher<AvxVector>, |
| 936 |
) -> (AvxVectorSse, AvxVectorSse) { |
| 937 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 938 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 939 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 940 |
|
| 941 |
let x: i32 = lut_r.x; |
| 942 |
let y: i32 = lut_g.x; |
| 943 |
let z: i32 = lut_b.x; |
| 944 |
|
| 945 |
let x_n: i32 = lut_r.x_n; |
| 946 |
let y_n: i32 = lut_g.x_n; |
| 947 |
let z_n: i32 = lut_b.x_n; |
| 948 |
|
| 949 |
let rx = lut_r.w; |
| 950 |
let ry = lut_g.w; |
| 951 |
let rz = lut_b.w; |
| 952 |
|
| 953 |
let w0 = AvxVector::from(rx); |
| 954 |
let w1 = AvxVector::from(ry); |
| 955 |
let w2 = AvxVector::from(rz); |
| 956 |
|
| 957 |
let c000 = rv.fetch(x, y, z); |
| 958 |
let c100 = rv.fetch(x_n, y, z); |
| 959 |
let c010 = rv.fetch(x, y_n, z); |
| 960 |
let c110 = rv.fetch(x_n, y_n, z); |
| 961 |
let c001 = rv.fetch(x, y, z_n); |
| 962 |
let c101 = rv.fetch(x_n, y, z_n); |
| 963 |
let c011 = rv.fetch(x, y_n, z_n); |
| 964 |
let c111 = rv.fetch(x_n, y_n, z_n); |
| 965 |
|
| 966 |
let dx = AvxVector::from(rx); |
| 967 |
|
| 968 |
let c00 = c000.neg_mla(c000, dx).mla(c100, w0); |
| 969 |
let c10 = c010.neg_mla(c010, dx).mla(c110, w0); |
| 970 |
let c01 = c001.neg_mla(c001, dx).mla(c101, w0); |
| 971 |
let c11 = c011.neg_mla(c011, dx).mla(c111, w0); |
| 972 |
|
| 973 |
let dy = AvxVector::from(ry); |
| 974 |
|
| 975 |
let c0 = c00.neg_mla(c00, dy).mla(c10, w1); |
| 976 |
let c1 = c01.neg_mla(c01, dy).mla(c11, w1); |
| 977 |
|
| 978 |
let dz = AvxVector::from(rz); |
| 979 |
|
| 980 |
c0.neg_mla(c0, dz).mla(c1, w2).split() |
| 981 |
} |
| 982 |
} |
| 983 |
|
| 984 |
impl<const GRID_SIZE: usize> TrilinearAvxFma<GRID_SIZE> { |
| 985 |
#[target_feature(enable = "avx2", enable = "fma")] |
| 986 |
unsafe fn interpolate( |
| 987 |
&self, |
| 988 |
in_r: usize, |
| 989 |
in_g: usize, |
| 990 |
in_b: usize, |
| 991 |
lut: &[BarycentricWeight<f32>], |
| 992 |
r: impl Fetcher<AvxVectorSse>, |
| 993 |
) -> AvxVectorSse { |
| 994 |
let lut_r = unsafe { lut.get_unchecked(in_r) }; |
| 995 |
let lut_g = unsafe { lut.get_unchecked(in_g) }; |
| 996 |
let lut_b = unsafe { lut.get_unchecked(in_b) }; |
| 997 |
|
| 998 |
let x: i32 = lut_r.x; |
| 999 |
let y: i32 = lut_g.x; |
| 1000 |
let z: i32 = lut_b.x; |
| 1001 |
|
| 1002 |
let x_n: i32 = lut_r.x_n; |
| 1003 |
let y_n: i32 = lut_g.x_n; |
| 1004 |
let z_n: i32 = lut_b.x_n; |
| 1005 |
|
| 1006 |
let dr = lut_r.w; |
| 1007 |
let dg = lut_g.w; |
| 1008 |
let db = lut_b.w; |
| 1009 |
|
| 1010 |
let w0 = AvxVector::from(dr); |
| 1011 |
let w1 = AvxVector::from(dg); |
| 1012 |
let w2 = AvxVectorSse::from(db); |
| 1013 |
|
| 1014 |
let c000 = r.fetch(x, y, z); |
| 1015 |
let c100 = r.fetch(x_n, y, z); |
| 1016 |
let c010 = r.fetch(x, y_n, z); |
| 1017 |
let c110 = r.fetch(x_n, y_n, z); |
| 1018 |
let c001 = r.fetch(x, y, z_n); |
| 1019 |
let c101 = r.fetch(x_n, y, z_n); |
| 1020 |
let c011 = r.fetch(x, y_n, z_n); |
| 1021 |
let c111 = r.fetch(x_n, y_n, z_n); |
| 1022 |
|
| 1023 |
let x000 = AvxVector::from_sse(c000, c001); |
| 1024 |
let x010 = AvxVector::from_sse(c010, c011); |
| 1025 |
let x011 = AvxVector::from_sse(c100, c101); |
| 1026 |
let x111 = AvxVector::from_sse(c110, c111); |
| 1027 |
|
| 1028 |
let c00 = x000.neg_mla(x000, w0).mla(x011, w0); |
| 1029 |
let c10 = x010.neg_mla(x010, w0).mla(x111, w0); |
| 1030 |
|
| 1031 |
let z0 = c00.neg_mla(c00, w1).mla(c10, w1); |
| 1032 |
|
| 1033 |
let (c0, c1) = z0.split(); |
| 1034 |
|
| 1035 |
c0.neg_mla(c0, w2).mla(c1, w2) |
| 1036 |
} |
| 1037 |
} |