Expand description
GhostFlow Core - High-performance tensor operations
This crate provides the foundational tensor type and operations for the GhostFlow ML framework.
§Phase 4 Optimizations (Beat JAX!)
- Operation fusion engine
- JIT compilation
- Memory layout optimization
- Custom optimized kernels
Re-exports§
pub use dtype::DType;pub use shape::Shape;pub use shape::Strides;pub use storage::Storage;pub use tensor::Tensor;pub use device::Device;pub use device::Cpu;pub use error::GhostError;pub use error::Result;pub use serialize::StateDict;pub use serialize::save_state_dict;pub use serialize::load_state_dict;pub use serialize::Serializable;pub use sparse::SparseTensorCOO;pub use sparse::SparseTensorCSR;pub use sparse::SparseTensorCSC;pub use hardware::HardwareBackend;pub use hardware::HardwareDevice;pub use hardware::HardwareOps;pub use hardware::ElementwiseOp;pub use hardware::list_devices;pub use fusion::FusionEngine;pub use fusion::ComputeGraph;pub use fusion::FusionPattern;pub use layout::LayoutOptimizer;pub use layout::MemoryLayout;pub use layout::DeviceInfo;pub use simd_ops::simd_add_f32;pub use simd_ops::simd_mul_f32;pub use simd_ops::simd_dot_f32;pub use simd_ops::simd_relu_f32;pub use memory::MemoryPool;pub use memory::MemoryStats;pub use memory::MemoryLayoutOptimizer;pub use memory::TrackedAllocator;pub use profiler::Profiler;pub use profiler::ProfileScope;pub use profiler::Benchmark;pub use profiler::BenchmarkResult;pub use profiler::global_profiler;
Modules§
- device
- Device abstraction for CPU/GPU execution
- dtype
- Data types supported by GhostFlow tensors
- error
- Error types for GhostFlow
- fusion
- Kernel fusion engine for optimizing computation graphs
- hardware
- Hardware abstraction layer
- layout
- Memory Layout Optimizer
- memory
- Memory optimization utilities
- metal
- Metal backend for Apple Silicon
- neon
- ARM NEON SIMD optimizations
- ops
- Tensor operations
- prelude
- Prelude for convenient imports
- profiler
- Profiling tools for performance analysis
- rocm
- ROCm (AMD GPU) backend
- serialize
- Model serialization and deserialization
- shape
- Shape and stride handling for tensors
- simd_
ops - Advanced SIMD optimizations for tensor operations
- sparse
- Sparse tensor operations
- storage
- Storage backend for tensor data
- tensor
- Core Tensor type - the foundation of GhostFlow
- tensor_
ops - Tensor operations trait extensions
- tpu
- TPU (Tensor Processing Unit) backend
Macros§
- profile
- Profile a code block