Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add FP16 support #257

Merged
merged 6 commits into from
Feb 5, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions .github/workflows/rust.yml
Original file line number Diff line number Diff line change
Expand Up @@ -9,14 +9,14 @@ jobs:
runs-on: ubuntu-latest
strategy:
matrix:
rust: ["1.61.0", stable, beta, nightly]
rust: ["1.70.0", stable, beta, nightly]
steps:
- uses: actions/checkout@v2

- uses: dtolnay/rust-toolchain@nightly
if: ${{ matrix.rust == '1.61.0' }}
if: ${{ matrix.rust == '1.70.0' }}
- name: Generate Cargo.lock with minimal-version dependencies
if: ${{ matrix.rust == '1.61.0' }}
if: ${{ matrix.rust == '1.70.0' }}
run: cargo -Zminimal-versions generate-lockfile

- uses: dtolnay/rust-toolchain@v1
Expand All @@ -29,7 +29,7 @@ jobs:
- name: build
run: cargo build -v
- name: test
if: ${{ matrix.rust != '1.61.0' }}
if: ${{ matrix.rust != '1.70.0' }}
run: cargo test -v && cargo doc -v

rustfmt:
Expand Down
1 change: 1 addition & 0 deletions Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@ categories = ["multimedia::images", "multimedia::encoding"]
exclude = ["tests/images/*", "tests/fuzz_images/*"]

[dependencies]
half = { version = "2.4.1" }
weezl = "0.1.0"
jpeg = { package = "jpeg-decoder", version = "0.3.0", default-features = false }
flate2 = "1.0.20"
Expand Down
3 changes: 3 additions & 0 deletions src/bytecast.rs
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,8 @@
//! TODO: Would like to use std-lib here.
use std::{mem, slice};

use half::f16;

macro_rules! integral_slice_as_bytes{($int:ty, $const:ident $(,$mut:ident)*) => {
pub(crate) fn $const(slice: &[$int]) -> &[u8] {
assert!(mem::align_of::<$int>() <= mem::size_of::<$int>());
Expand All @@ -31,4 +33,5 @@ integral_slice_as_bytes!(i32, i32_as_ne_bytes, i32_as_ne_mut_bytes);
integral_slice_as_bytes!(u64, u64_as_ne_bytes, u64_as_ne_mut_bytes);
integral_slice_as_bytes!(i64, i64_as_ne_bytes, i64_as_ne_mut_bytes);
integral_slice_as_bytes!(f32, f32_as_ne_bytes, f32_as_ne_mut_bytes);
integral_slice_as_bytes!(f16, f16_as_ne_bytes, f16_as_ne_mut_bytes);
integral_slice_as_bytes!(f64, f64_as_ne_bytes, f64_as_ne_mut_bytes);
14 changes: 8 additions & 6 deletions src/decoder/image.rs
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
use super::ifd::{Directory, Value};
use super::stream::{ByteOrder, DeflateReader, LZWReader, PackBitsReader};
use super::tag_reader::TagReader;
use super::{predict_f32, predict_f64, Limits};
use super::{predict_f16, predict_f32, predict_f64, Limits};
use super::{stream::SmartReader, ChunkType};
use crate::tags::{
CompressionMethod, PhotometricInterpretation, PlanarConfiguration, Predictor, SampleFormat, Tag,
Expand Down Expand Up @@ -378,11 +378,9 @@ impl Image {
CompressionMethod::None => Box::new(reader),
CompressionMethod::LZW => {
Box::new(LZWReader::new(reader, usize::try_from(compressed_length)?))
},
}
#[cfg(feature = "zstd")]
CompressionMethod::ZSTD => {
Box::new(zstd::Decoder::new(reader)?)
},
CompressionMethod::ZSTD => Box::new(zstd::Decoder::new(reader)?),
CompressionMethod::PackBits => Box::new(PackBitsReader::new(reader, compressed_length)),
CompressionMethod::Deflate | CompressionMethod::OldDeflate => {
Box::new(DeflateReader::new(reader))
Expand Down Expand Up @@ -594,7 +592,10 @@ impl Image {

// Validate that the predictor is supported for the sample type.
match (self.predictor, self.sample_format) {
(Predictor::Horizontal, SampleFormat::Int | SampleFormat::Uint) => {}
(
Predictor::Horizontal,
SampleFormat::Int | SampleFormat::Uint | SampleFormat::IEEEFP,
) => {}
(Predictor::Horizontal, _) => {
return Err(TiffError::UnsupportedError(
TiffUnsupportedError::HorizontalPredictor(color_type),
Expand Down Expand Up @@ -674,6 +675,7 @@ impl Image {

let row = &mut row[..data_row_bytes];
match color_type.bit_depth() {
16 => predict_f16(&mut encoded, row, samples),
32 => predict_f32(&mut encoded, row, samples),
64 => predict_f64(&mut encoded, row, samples),
_ => unreachable!(),
Expand Down
30 changes: 30 additions & 0 deletions src/decoder/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@ use crate::tags::{
use crate::{
bytecast, ColorType, TiffError, TiffFormatError, TiffResult, TiffUnsupportedError, UsageError,
};
use half::f16;

use self::ifd::Directory;
use self::image::Image;
Expand All @@ -29,6 +30,8 @@ pub enum DecodingResult {
U32(Vec<u32>),
/// A vector of 64 bit unsigned ints
U64(Vec<u64>),
/// A vector of 16 bit IEEE floats (held in u16)
F16(Vec<f16>),
/// A vector of 32 bit IEEE floats
F32(Vec<f32>),
/// A vector of 64 bit IEEE floats
Expand Down Expand Up @@ -92,6 +95,14 @@ impl DecodingResult {
}
}

fn new_f16(size: usize, limits: &Limits) -> TiffResult<DecodingResult> {
if size > limits.decoding_buffer_size / std::mem::size_of::<u16>() {
Err(TiffError::LimitsExceeded)
} else {
Ok(DecodingResult::F16(vec![f16::ZERO; size]))
}
}

fn new_i8(size: usize, limits: &Limits) -> TiffResult<DecodingResult> {
if size > limits.decoding_buffer_size / std::mem::size_of::<i8>() {
Err(TiffError::LimitsExceeded)
Expand Down Expand Up @@ -130,6 +141,7 @@ impl DecodingResult {
DecodingResult::U16(ref mut buf) => DecodingBuffer::U16(&mut buf[start..]),
DecodingResult::U32(ref mut buf) => DecodingBuffer::U32(&mut buf[start..]),
DecodingResult::U64(ref mut buf) => DecodingBuffer::U64(&mut buf[start..]),
DecodingResult::F16(ref mut buf) => DecodingBuffer::F16(&mut buf[start..]),
DecodingResult::F32(ref mut buf) => DecodingBuffer::F32(&mut buf[start..]),
DecodingResult::F64(ref mut buf) => DecodingBuffer::F64(&mut buf[start..]),
DecodingResult::I8(ref mut buf) => DecodingBuffer::I8(&mut buf[start..]),
Expand All @@ -150,6 +162,8 @@ pub enum DecodingBuffer<'a> {
U32(&'a mut [u32]),
/// A slice of 64 bit unsigned ints
U64(&'a mut [u64]),
/// A slice of 16 bit IEEE floats
F16(&'a mut [f16]),
/// A slice of 32 bit IEEE floats
F32(&'a mut [f32]),
/// A slice of 64 bit IEEE floats
Expand All @@ -175,6 +189,7 @@ impl<'a> DecodingBuffer<'a> {
DecodingBuffer::I32(buf) => bytecast::i32_as_ne_mut_bytes(buf),
DecodingBuffer::U64(buf) => bytecast::u64_as_ne_mut_bytes(buf),
DecodingBuffer::I64(buf) => bytecast::i64_as_ne_mut_bytes(buf),
DecodingBuffer::F16(buf) => bytecast::f16_as_ne_mut_bytes(buf),
DecodingBuffer::F32(buf) => bytecast::f32_as_ne_mut_bytes(buf),
DecodingBuffer::F64(buf) => bytecast::f64_as_ne_mut_bytes(buf),
}
Expand Down Expand Up @@ -303,6 +318,19 @@ fn predict_f32(input: &mut [u8], output: &mut [u8], samples: usize) {
}
}

fn predict_f16(input: &mut [u8], output: &mut [u8], samples: usize) {
for i in samples..input.len() {
input[i] = input[i].wrapping_add(input[i - samples]);
}

for (i, chunk) in output.chunks_mut(2).enumerate() {
chunk.copy_from_slice(&u16::to_ne_bytes(u16::from_be_bytes([
input[i],
input[input.len() / 2 + i],
])));
}
}

fn predict_f64(input: &mut [u8], output: &mut [u8], samples: usize) {
for i in samples..input.len() {
input[i] = input[i].wrapping_add(input[i - samples]);
Expand Down Expand Up @@ -340,6 +368,7 @@ fn fix_endianness_and_predict(
Predictor::FloatingPoint => {
let mut buffer_copy = buf.to_vec();
match bit_depth {
16 => predict_f16(&mut buffer_copy, buf, samples),
32 => predict_f32(&mut buffer_copy, buf, samples),
64 => predict_f64(&mut buffer_copy, buf, samples),
_ => unreachable!("Caller should have validated arguments. Please file a bug."),
Expand Down Expand Up @@ -1004,6 +1033,7 @@ impl<R: Read + Seek> Decoder<R> {
)),
},
SampleFormat::IEEEFP => match max_sample_bits {
16 => DecodingResult::new_f16(buffer_size, &self.limits),
32 => DecodingResult::new_f32(buffer_size, &self.limits),
64 => DecodingResult::new_f64(buffer_size, &self.limits),
n => Err(TiffError::UnsupportedError(
Expand Down
195 changes: 195 additions & 0 deletions tests/decode_fp16_images.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,195 @@
extern crate tiff;

use tiff::decoder::{Decoder, DecodingResult};
use tiff::ColorType;

use std::fs::File;
use std::path::PathBuf;

const TEST_IMAGE_DIR: &str = "./tests/images/";

/// Test a basic all white image
#[test]
fn test_white_ieee_fp16() {
let filenames = ["white-fp16.tiff"];

for filename in filenames.iter() {
let path = PathBuf::from(TEST_IMAGE_DIR).join(filename);
let img_file = File::open(path).expect("Cannot find test image!");
let mut decoder = Decoder::new(img_file).expect("Cannot create decoder");
assert_eq!(
decoder.dimensions().expect("Cannot get dimensions"),
(256, 256)
);
assert_eq!(
decoder.colortype().expect("Cannot get colortype"),
ColorType::Gray(16)
);
if let DecodingResult::F16(img) = decoder.read_image().unwrap() {
for p in img {
assert!(p == half::f16::from_f32_const(1.0));
}
} else {
panic!("Wrong data type");
}
}
}

/// Test a single black pixel, to make sure scaling is ok
#[test]
fn test_one_black_pixel_ieee_fp16() {
let filenames = ["single-black-fp16.tiff"];

for filename in filenames.iter() {
let path = PathBuf::from(TEST_IMAGE_DIR).join(filename);
let img_file = File::open(path).expect("Cannot find test image!");
let mut decoder = Decoder::new(img_file).expect("Cannot create decoder");
assert_eq!(
decoder.dimensions().expect("Cannot get dimensions"),
(256, 256)
);
assert_eq!(
decoder.colortype().expect("Cannot get colortype"),
ColorType::Gray(16)
);
if let DecodingResult::F16(img) = decoder.read_image().unwrap() {
for (i, p) in img.iter().enumerate() {
if i == 0 {
assert!(p < &half::f16::from_f32_const(0.001));
} else {
assert!(p == &half::f16::from_f32_const(1.0));
}
}
} else {
panic!("Wrong data type");
}
}
}

/// Test white with horizontal differencing predictor
#[test]
fn test_pattern_horizontal_differencing_ieee_fp16() {
let filenames = ["white-fp16-pred2.tiff"];

for filename in filenames.iter() {
let path = PathBuf::from(TEST_IMAGE_DIR).join(filename);
let img_file = File::open(path).expect("Cannot find test image!");
let mut decoder = Decoder::new(img_file).expect("Cannot create decoder");
assert_eq!(
decoder.dimensions().expect("Cannot get dimensions"),
(256, 256)
);
assert_eq!(
decoder.colortype().expect("Cannot get colortype"),
ColorType::Gray(16)
);
if let DecodingResult::F16(img) = decoder.read_image().unwrap() {
// 0, 2, 5, 8, 12, 16, 255 are black
let black = [0, 2, 5, 8, 12, 16, 255];
for (i, p) in img.iter().enumerate() {
if black.contains(&i) {
assert!(p < &half::f16::from_f32_const(0.001));
} else {
assert!(p == &half::f16::from_f32_const(1.0));
}
}
} else {
panic!("Wrong data type");
}
}
}

/// Test white with floating point predictor
#[test]
fn test_pattern_predictor_ieee_fp16() {
let filenames = ["white-fp16-pred3.tiff"];

for filename in filenames.iter() {
let path = PathBuf::from(TEST_IMAGE_DIR).join(filename);
let img_file = File::open(path).expect("Cannot find test image!");
let mut decoder = Decoder::new(img_file).expect("Cannot create decoder");
assert_eq!(
decoder.dimensions().expect("Cannot get dimensions"),
(256, 256)
);
assert_eq!(
decoder.colortype().expect("Cannot get colortype"),
ColorType::Gray(16)
);
if let DecodingResult::F16(img) = decoder.read_image().unwrap() {
// 0, 2, 5, 8, 12, 16, 255 are black
let black = [0, 2, 5, 8, 12, 16, 255];
for (i, p) in img.iter().enumerate() {
if black.contains(&i) {
assert!(p < &half::f16::from_f32_const(0.001));
} else {
assert!(p == &half::f16::from_f32_const(1.0));
}
}
} else {
panic!("Wrong data type");
}
}
}

/// Test several random images
/// we'rell compare against a pnm file, that scales from 0 (for 0.0) to 65767 (for 1.0)
#[test]
fn test_predictor_ieee_fp16() {
// first parse pnm, skip the first 4 \n
let pnm_path = PathBuf::from(TEST_IMAGE_DIR).join("random-fp16.pgm");
let pnm_bytes = std::fs::read(pnm_path).expect("Failed to read expected PNM file");

// PGM looks like this:
// ---
// P5
// #Created with GIMP
// 16 16
// 65535
// ... <big-endian bytes>
// ---
// get index of 4th \n
let byte_start = pnm_bytes
.iter()
.enumerate()
.filter(|(_, &v)| v == b'\n')
.map(|(i, _)| i)
.nth(3)
.expect("Must be 4 \\n's");

let pnm_values: Vec<f32> = pnm_bytes[(byte_start + 1)..]
.chunks(2)
.map(|slice| {
let bts = [slice[0], slice[1]];
(u16::from_be_bytes(bts) as f32) / (u16::MAX as f32)
})
.collect();
assert!(pnm_values.len() == 256);

let filenames = [
"random-fp16-pred2.tiff",
"random-fp16-pred3.tiff",
"random-fp16.tiff",
];

for filename in filenames.iter() {
let path = PathBuf::from(TEST_IMAGE_DIR).join(filename);
let img_file = File::open(path).expect("Cannot find test image!");
let mut decoder = Decoder::new(img_file).expect("Cannot create decoder");
assert_eq!(
decoder.dimensions().expect("Cannot get dimensions"),
(16, 16)
);
assert_eq!(
decoder.colortype().expect("Cannot get colortype"),
ColorType::Gray(16)
);
if let DecodingResult::F16(img) = decoder.read_image().unwrap() {
for (exp, found) in std::iter::zip(pnm_values.iter(), img.iter()) {
assert!((exp - found.to_f32()).abs() < 0.0001);
}
} else {
panic!("Wrong data type");
}
}
}
Binary file added tests/images/random-fp16-pred2.tiff
Binary file not shown.
Binary file added tests/images/random-fp16-pred3.tiff
Binary file not shown.
Binary file added tests/images/random-fp16.pgm
Binary file not shown.
Binary file added tests/images/random-fp16.tiff
Binary file not shown.
Binary file added tests/images/single-black-fp16.tiff
Binary file not shown.
Binary file added tests/images/white-fp16-pred2.tiff
Binary file not shown.
Binary file added tests/images/white-fp16-pred3.tiff
Binary file not shown.
Binary file added tests/images/white-fp16.tiff
Binary file not shown.