Pillar 9: Hardware-Accelerated Math

Concept

Hardware-Accelerated Math exposes high-performance computing capabilities directly. It allows developers to define blocks of code intended for GPU execution (kernels) and high-dimensional data types (tensors, matrices).

Syntax

GPU Block

gpu kernel MatMul {
    // Code to run on GPU
}

Math Types

• tensor: N-dimensional array.
• matrix: 2D array specialized for linear algebra.

Usage

The compiler identifies gpu blocks and compiles them to SPIR-V, CUDA, or OpenCL depending on the backend, abstracting the low-level details of graphics API interoperability.