yscv_kernels/lib.rs
1//! Execution kernels and backend abstraction for yscv.
2//!
3//! ## GPU Inference (Cross-Platform via wgpu)
4//!
5//! The `gpu` feature enables compute shader acceleration via wgpu —
6//! Vulkan (Linux/Windows/Android), Metal (macOS/iOS), DX12 (Windows).
7//! No CUDA dependency. GPU-accelerated operations:
8//! - Matrix multiplication (tiled 16×16 workgroups)
9//! - Elementwise: add, sub, mul
10//! - Activations: relu, sigmoid
11//! - Normalization: batch_norm, layer_norm, group_norm, rms_norm, softmax
12//! - Convolution: conv2d, depthwise_conv2d, separable_conv2d, transpose_conv2d
13//! - Pooling: max_pool2d, avg_pool2d
14//!
15//! GPU training (backward passes) is on the roadmap.
16//! CPU backend is fully optimized with NEON/AVX/SSE SIMD on all platforms.
17#![deny(unsafe_code)]
18
19mod core;
20
21pub use core::*;