Skip to main content

Crate aprender_gemm_codegen

Crate aprender_gemm_codegen 

Source
Expand description

trueno-gemm-codegen: Compile-time GEMM microkernel code generation.

Contract: cgp-gemm-codegen-v1.yaml (C-CODEGEN-001 through C-CODEGEN-004)

Generates shape-specialized AVX-512 microkernels at compile time via proc macros. Sovereign implementation — no external BLAS dependencies.

§Usage

use trueno_gemm_codegen::avx512_microkernel;

avx512_microkernel!(mr = 8, nr = 32);
// Generates: pub unsafe fn microkernel_8x32_avx512_gen(k, a, b, c, ldc)

Macros§

avx512_microkernel
Generate an AVX-512 row-major C microkernel.
avx512_microkernel_broadcast_b
Generate an AVX-512 broadcast-B microkernel (faer-style).