Implementation:Ggml org Ggml Blas backend api

Metadata

Field	Value
Page Type	Implementation (API Header)
Knowledge Sources	GGML
Domains	ML_Infrastructure, Tensor_Computing, Linear_Algebra
Last Updated	2026-02-10 12:00 GMT

Overview

Public header declaring the BLAS backend interface for accelerating matrix operations via libraries like OpenBLAS, MKL, Accelerate, BLIS, or NVPL.

Description

ggml-blas.h (25 lines) provides the minimal public API for the BLAS backend. It exposes four functions:

ggml_backend_blas_init() -- creates and returns a new BLAS backend instance
ggml_backend_is_blas() -- checks whether a given backend is a BLAS backend
ggml_backend_blas_set_n_threads() -- configures thread count for float conversion and BLAS operations (for OpenBLAS and BLIS, this also sets the internal BLAS thread count)
ggml_backend_blas_reg() -- returns the backend registration handle for dynamic discovery

All functions are marked with GGML_BACKEND_API for proper DLL export/import on Windows.

Usage

Include this header in application code to initialize and configure the BLAS backend for CPU-side matrix multiplication acceleration.

Code Reference

Source Location

GGML repo, file: include/ggml-blas.h, 25 lines.

Signature

GGML_BACKEND_API ggml_backend_t ggml_backend_blas_init(void);
GGML_BACKEND_API bool ggml_backend_is_blas(ggml_backend_t backend);
GGML_BACKEND_API void ggml_backend_blas_set_n_threads(ggml_backend_t backend_blas,
                                                       int n_threads);
GGML_BACKEND_API ggml_backend_reg_t ggml_backend_blas_reg(void);

Import

#include "ggml-blas.h"

Dependencies

ggml.h -- core GGML types
ggml-backend.h -- backend abstraction types (ggml_backend_t, ggml_backend_reg_t)

I/O Contract

Inputs

Parameter	Type	Required	Description
`backend`	`ggml_backend_t`	Yes (for is_blas)	Backend instance to check.
`backend_blas`	`ggml_backend_t`	Yes (for set_n_threads)	BLAS backend instance to configure.
`n_threads`	`int`	Yes (for set_n_threads)	Number of threads for conversion and BLAS operations.

Outputs

Output	Type	Description
Backend handle	`ggml_backend_t`	From `ggml_backend_blas_init()`: initialized BLAS backend.
Is BLAS	`bool`	From `ggml_backend_is_blas()`: true if the backend is BLAS-based.
Registration	`ggml_backend_reg_t`	From `ggml_backend_blas_reg()`: registration handle for backend discovery.

Usage Examples

Basic BLAS Backend Setup

#include "ggml-blas.h"

ggml_backend_t blas = ggml_backend_blas_init();
ggml_backend_blas_set_n_threads(blas, 8);

// ... use blas backend with scheduler ...

ggml_backend_free(blas);

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment