SWCU195A December 2024 – May 2025 CC2744R7-Q1 , CC2745P10-Q1 , CC2745R10-Q1 , CC2745R7-Q1 , CC2755R10
The APU is a generic mathematical acceleration module that operates with single-precision floating-point numbers (IEEE 754 format) and is optimized to work with complex numbers.
The APU operates autonomously from the main CPU in the system and can be used to offload numerically intensive operations. This module handles efficient vector (and matrix) operations and sustains a complex Multiply-and-Add operation per clock cycle.
The APU has an 8kB local data memory (separate from the system RAM) and a core to handle advanced APIs developed for the APU hardware accelerator sub-modules. This core is controlled by the RAM-based local program memory (separate from the system RAM and APU data RAM) where the APU APIs are loaded. TI provides the APU APIs that shall be executed by the APU programmable core.
The APU supports algorithms including: FFT, Eigenvalue decomposition, Matrix factorization, Sorting, Matrix Frobenius Normalization. These are also useful for the upcoming Bluetooth Channel Sounding post-processing operations to perform precise ranging over RF.