Crate apple_mps 

mps-rs

Safe Rust bindings for Apple’s MetalPerformanceShaders framework on macOS.

The GitHub repository is mps-rs; the published crates.io package is apple-mps because the mps-rs package name is already taken.

Install

cargo add apple-mps apple-metal

Quick start

use apple_metal::MetalDevice;
use apple_mps::{feature_channel_format, Image, ImageDescriptor, ImageGaussianBlur};

let device = MetalDevice::system_default().expect("no Metal device");
let queue = device.new_command_queue().expect("queue");
let descriptor = ImageDescriptor::new(256, 256, 1, feature_channel_format::FLOAT32);
let src = Image::new(&device, descriptor).expect("source image");
let dst = Image::new(&device, descriptor).expect("destination image");
let blur = ImageGaussianBlur::new(&device, 2.0).expect("gaussian blur");
let command_buffer = queue.new_command_buffer().expect("command buffer");
blur.encode_image(&command_buffer, &src, &dst);

v0.2.1 surface

• Core helpers:
  • supports_mtl_device, preferred_device, hint_temporary_memory_high_water_mark, set_heap_cache_duration
  • Predicate and MpsCommandBuffer
• Images:
  • ImageDescriptor + Image for lazily allocated MPS images or texture-backed images
  • Float32 image read/write helpers plus raw byte transfer with MPSDataLayout
  • Unary image filters:
    • ImageGaussianBlur
    • ImageBox
    • ImageSobel
    • ImageMedian
    • ImageConvolution
    • ImageBilinearScale
    • ImageLanczosScale
    • ImageThresholdBinary
    • ImageStatisticsMinAndMax
    • ImageStatisticsMean
    • ImageReduceRowMin, ImageReduceRowMax, ImageReduceRowMean, ImageReduceRowSum
  • ImageHistogram, ImageAdd, and ImageScaleAndAdd
• Matrix/vector:
  • MatrixDescriptor, VectorDescriptor, MatrixMultiplicationDescriptor, Matrix, Vector, and MatrixMultiplication
• NDArray:
  • NDArrayDescriptor, NDArray, NDArrayIdentity, and NDArrayMatrixMultiplication
• State:
  • State, StateResourceList, StateTextureInfo
  • state_batch_increment_read_count, state_batch_resource_size, state_batch_synchronize
• Ray tracing / denoising:
  • PolygonAccelerationStructure, RayIntersector, and SVGF
• Neural / optimizer / RNN:
  • NNImageNode, NNGraph
  • CnnNeuronReluNode, CnnPoolingMaxNode, CnnSoftMaxNode, CnnUpsamplingNearestNode
  • CnnConvolutionDescriptor, CnnConvolution, CnnConvolutionWeightsAndBiasesState
  • NNOptimizerDescriptor, NNOptimizer, NNOptimizerStochasticGradientDescent, NNOptimizerRmsProp, NNOptimizerAdam
  • RnnDescriptor, RnnSingleGateDescriptor, GruDescriptor, LstmDescriptor, RnnImageInferenceLayer, RnnRecurrentImageState
• Shared constants for MPSKernelOptions, MPSImageEdgeMode, MPSImageFeatureChannelFormat, MPSDataType, MPSDataLayout, plus convolution / optimizer / RNN / state enums

See COVERAGE.md for the Wave-C audit and the implemented/partial matrix.

Validation

cargo test
cargo run --example 01_blur_image
cargo run --example 02_matrix_multiply
cargo run --example 03_ndarray_identity
cargo run --example 04_ray_intersection
cargo run --example 05_nn_graph_relu
cargo run --example 06_ndarray_matrix_multiplication
cargo run --example 07_optimizer_and_state
cargo run --example 08_cnn_convolution
cargo run --example 09_rnn_image_inference

Re-exports

pub use crate::core::device_options;

pub use crate::core::hint_temporary_memory_high_water_mark;

pub use crate::core::preferred_device;

pub use crate::core::set_heap_cache_duration;

pub use crate::core::supports_mtl_device;

pub use crate::core::CommandBuffer as MpsCommandBuffer;

pub use crate::core::Predicate;

pub use crate::core::PreferredDevice;

pub use crate::error::Error;

pub use crate::error::Result;

pub use crate::filters::HistogramInfo;

pub use crate::filters::ImageAdd;

pub use crate::filters::ImageBilinearScale;

pub use crate::filters::ImageBox;

pub use crate::filters::ImageConvolution;

pub use crate::filters::ImageGaussianBlur;

pub use crate::filters::ImageHistogram;

pub use crate::filters::ImageLanczosScale;

pub use crate::filters::ImageMedian;

pub use crate::filters::ImageReduceRowMax;

pub use crate::filters::ImageReduceRowMean;

pub use crate::filters::ImageReduceRowMin;

pub use crate::filters::ImageReduceRowSum;

pub use crate::filters::ImageScaleAndAdd;

pub use crate::filters::ImageSobel;

pub use crate::filters::ImageStatisticsMean;

pub use crate::filters::ImageStatisticsMinAndMax;

pub use crate::filters::ImageThresholdBinary;

pub use crate::filters::ScaleTransform;

pub use crate::image::feature_channel_format;

pub use crate::image::image_edge_mode;

pub use crate::image::image_layout;

pub use crate::image::kernel_options;

pub use crate::image::Image;

pub use crate::image::ImageDescriptor;

pub use crate::image::ImageReadWriteParams;

pub use crate::image::ImageRegion;

pub use crate::matrix::data_type;

pub use crate::matrix::data_type_size;

pub use crate::matrix::Matrix;

pub use crate::matrix::MatrixDescriptor;

pub use crate::matrix::MatrixMultiplication;

pub use crate::matrix::MatrixMultiplicationDescriptor;

pub use crate::matrix::Vector;

pub use crate::matrix::VectorDescriptor;

pub use crate::ndarray::NDArray;

pub use crate::ndarray::NDArrayDescriptor;

pub use crate::ndarray::NDArrayIdentity;

pub use crate::ndarray::NDArrayMatrixMultiplication;

pub use crate::neural::cnn_accumulator_precision_option;

pub use crate::neural::cnn_convolution_flags;

pub use crate::neural::cnn_convolution_weights_layout;

pub use crate::neural::nn_regularization_type;

pub use crate::neural::rnn_bidirectional_combine_mode;

pub use crate::neural::rnn_sequence_direction;

pub use crate::neural::CnnConvolution;

pub use crate::neural::CnnConvolutionDescriptor;

pub use crate::neural::CnnConvolutionWeightsAndBiasesState;

pub use crate::neural::CnnNeuronReluNode;

pub use crate::neural::CnnPoolingMaxNode;

pub use crate::neural::CnnSoftMaxNode;

pub use crate::neural::CnnUpsamplingNearestNode;

pub use crate::neural::GruDescriptor;

pub use crate::neural::LstmDescriptor;

pub use crate::neural::NNGraph;

pub use crate::neural::NNImageNode;

pub use crate::neural::NNOptimizer;

pub use crate::neural::NNOptimizerAdam;

pub use crate::neural::NNOptimizerDescriptor;

pub use crate::neural::NNOptimizerRmsProp;

pub use crate::neural::NNOptimizerStochasticGradientDescent;

pub use crate::neural::RnnDescriptor;

pub use crate::neural::RnnImageInferenceLayer;

pub use crate::neural::RnnRecurrentImageState;

pub use crate::neural::RnnSingleGateDescriptor;

pub use crate::ray::acceleration_structure_status;

pub use crate::ray::acceleration_structure_usage;

pub use crate::ray::cull_mode;

pub use crate::ray::intersection_data_type;

pub use crate::ray::intersection_type;

pub use crate::ray::polygon_type;

pub use crate::ray::ray_data_type;

pub use crate::ray::winding;

pub use crate::ray::PolygonAccelerationStructure;

pub use crate::ray::RayIntersector;

pub use crate::ray::SVGF;

pub use crate::state::state_batch_increment_read_count;

pub use crate::state::state_batch_resource_size;

pub use crate::state::state_batch_synchronize;

pub use crate::state::state_resource_type;

pub use crate::state::State;

pub use crate::state::StateResourceList;

pub use crate::state::StateTextureInfo;

Modules

core

error

ffi

Raw FFI declarations matching the Swift bridge.

filters

image

matrix

ndarray

neural

ray

state