numr 0.5.1

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

//! SIMD-optimized matrix multiplication.
//!
//! See [`dispatch`] for the public API and microkernel dispatch functions.

#[cfg(target_arch = "x86_64")]
pub(crate) mod avx2;
#[cfg(target_arch = "x86_64")]
pub(crate) mod avx512;
pub(crate) mod dispatch;
pub(crate) mod gemv_bt;
pub(crate) mod int32;
pub(crate) mod int8;
pub(crate) mod macros;
pub(crate) mod packing;
pub(crate) mod scalar;
pub(crate) mod small;
pub(crate) mod small_kernels;
pub(crate) mod tiling;

#[cfg(target_arch = "aarch64")]
pub(crate) mod aarch64;

#[cfg(all(feature = "f16", target_arch = "x86_64"))]
pub(crate) mod half_convert;

pub use dispatch::{KC, MC, MR, NC, matmul_bias_f32, matmul_bias_f64, matmul_f32, matmul_f64};

pub use dispatch::{
    call_microkernel_2x_f32, call_microkernel_2x_f64, call_microkernel_f32, call_microkernel_f64,
};