| 1 |
#[macro_use] |
| 2 |
extern crate bencher; |
| 3 |
|
| 4 |
extern crate exr; |
| 5 |
use exr::prelude::*; |
| 6 |
|
| 7 |
use bencher::Bencher; |
| 8 |
use std::fs; |
| 9 |
use std::io::Cursor; |
| 10 |
use exr::image::pixel_vec::PixelVec; |
| 11 |
use exr::block::samples::FromNativeSample; |
| 12 |
|
| 13 |
const F32_ZIPS_PATH: &'static str = "tests/images/valid/custom/crowskull/crow_zips.exr"; |
| 14 |
const F32_UNCOMPRESSED_PATH: &'static str = "tests/images/valid/custom/crowskull/crow_uncompressed.exr"; |
| 15 |
const F16_UNCOMPRESSED_PATH: &'static str = "tests/images/valid/custom/crowskull/crow_uncompressed_half.exr"; |
| 16 |
const F16_ZIP_PATH: &'static str = "tests/images/valid/custom/crowskull/crow_zip_half.exr"; |
| 17 |
|
| 18 |
/// Read an image from an in-memory buffer into its native f32 format |
| 19 |
fn read_f32_as_f32_uncompressed_1thread(bench: &mut Bencher) { |
| 20 |
bench_read_image_rgba_as::<f32>(bench, F32_UNCOMPRESSED_PATH, false); |
| 21 |
} |
| 22 |
|
| 23 |
/// Read image and convert the samples to u32 (from native f32) |
| 24 |
fn read_f32_as_u32_uncompressed_1thread(bench: &mut Bencher) { |
| 25 |
bench_read_image_rgba_as::<u32>(bench, F32_UNCOMPRESSED_PATH, false); |
| 26 |
} |
| 27 |
|
| 28 |
/// f16 is not natively supported by CPUs, which introduces unique performance pitfalls |
| 29 |
fn read_f32_as_f16_uncompressed_1thread(bench: &mut Bencher) { |
| 30 |
bench_read_image_rgba_as::<f16>(bench, F32_UNCOMPRESSED_PATH, false); |
| 31 |
} |
| 32 |
|
| 33 |
fn read_f16_as_f16_uncompressed_1thread(bench: &mut Bencher) { |
| 34 |
bench_read_image_rgba_as::<f16>(bench, F16_UNCOMPRESSED_PATH, false); |
| 35 |
} |
| 36 |
|
| 37 |
fn read_f16_as_f32_uncompressed_1thread(bench: &mut Bencher) { |
| 38 |
bench_read_image_rgba_as::<f32>(bench, F16_UNCOMPRESSED_PATH, false); |
| 39 |
} |
| 40 |
|
| 41 |
fn read_f16_as_u32_uncompressed_1thread(bench: &mut Bencher) { |
| 42 |
bench_read_image_rgba_as::<u32>(bench, F16_UNCOMPRESSED_PATH, false); |
| 43 |
} |
| 44 |
|
| 45 |
|
| 46 |
fn read_f32_as_f16_zips_1thread(bench: &mut Bencher) { |
| 47 |
bench_read_image_rgba_as::<f16>(bench, F32_ZIPS_PATH, false); |
| 48 |
} |
| 49 |
|
| 50 |
fn read_f16_as_f32_zip_1thread(bench: &mut Bencher) { |
| 51 |
bench_read_image_rgba_as::<f32>(bench, F16_ZIP_PATH, false); |
| 52 |
} |
| 53 |
|
| 54 |
fn read_f32_as_f16_zips_nthreads(bench: &mut Bencher) { |
| 55 |
bench_read_image_rgba_as::<f16>(bench, F32_ZIPS_PATH, true); |
| 56 |
} |
| 57 |
|
| 58 |
fn read_f16_as_f32_zip_nthreads(bench: &mut Bencher) { |
| 59 |
bench_read_image_rgba_as::<f32>(bench, F16_ZIP_PATH, true); |
| 60 |
} |
| 61 |
|
| 62 |
fn read_f32_as_f32_zips_nthreads(bench: &mut Bencher) { |
| 63 |
bench_read_image_rgba_as::<f32>(bench, F32_ZIPS_PATH, true); |
| 64 |
} |
| 65 |
|
| 66 |
fn read_f16_as_f16_zip_nthreads(bench: &mut Bencher) { |
| 67 |
bench_read_image_rgba_as::<f16>(bench, F16_ZIP_PATH, true); |
| 68 |
} |
| 69 |
|
| 70 |
fn read_f32_as_f32_zips_1thread(bench: &mut Bencher) { |
| 71 |
bench_read_image_rgba_as::<f32>(bench, F32_ZIPS_PATH, false); |
| 72 |
} |
| 73 |
|
| 74 |
fn read_f16_as_f16_zip_1thread(bench: &mut Bencher) { |
| 75 |
bench_read_image_rgba_as::<f16>(bench, F16_ZIP_PATH, false); |
| 76 |
} |
| 77 |
|
| 78 |
fn bench_read_image_rgba_as<T>(bench: &mut Bencher, path: &str, parallel: bool) { |
| 79 |
let mut file = fs::read(path).unwrap(); |
| 80 |
bencher::black_box(&mut file); |
| 81 |
|
| 82 |
bench.iter(||{ |
| 83 |
let image = read_file_from_memory_as::<f16>(file.as_slice(), parallel); |
| 84 |
bencher::black_box(image); |
| 85 |
}) |
| 86 |
} |
| 87 |
|
| 88 |
fn read_file_from_memory_as<T>(file: &[u8], parallel: bool) -> RgbaImage<PixelVec<(T, T, T, T)>> |
| 89 |
where T: FromNativeSample |
| 90 |
{ |
| 91 |
let read = exr::prelude::read() |
| 92 |
.no_deep_data().largest_resolution_level() |
| 93 |
.rgba_channels(PixelVec::<(T, T, T, T)>::constructor, PixelVec::set_pixel) |
| 94 |
.first_valid_layer().all_attributes(); |
| 95 |
|
| 96 |
let read = if parallel { read } else { read.non_parallel() }; |
| 97 |
read.from_buffered(Cursor::new(file)).unwrap() |
| 98 |
} |
| 99 |
|
| 100 |
benchmark_group!(pixel_format_conversion, |
| 101 |
read_f32_as_f32_uncompressed_1thread, |
| 102 |
read_f32_as_u32_uncompressed_1thread, |
| 103 |
read_f32_as_f16_uncompressed_1thread, |
| 104 |
read_f32_as_f16_zips_1thread, |
| 105 |
read_f32_as_f16_zips_nthreads, |
| 106 |
read_f32_as_f32_zips_nthreads, |
| 107 |
read_f32_as_f32_zips_1thread, |
| 108 |
|
| 109 |
read_f16_as_f16_uncompressed_1thread, |
| 110 |
read_f16_as_u32_uncompressed_1thread, |
| 111 |
read_f16_as_f32_uncompressed_1thread, |
| 112 |
read_f16_as_f32_zip_1thread, |
| 113 |
read_f16_as_f32_zip_nthreads, |
| 114 |
read_f16_as_f16_zip_nthreads, |
| 115 |
read_f16_as_f16_zip_1thread, |
| 116 |
); |
| 117 |
|
| 118 |
benchmark_main!(pixel_format_conversion); |