SWCU195A December 2024 – May 2025 CC2744R7-Q1 , CC2745P10-Q1 , CC2745R10-Q1 , CC2745R7-Q1 , CC2755R10
All mathematical operations that can be performed on the APU are accessed by the system CPU through SIMPLELINK-LOWPOWER-F3-SDK APU Driver APIs. This allows higher-level algorithms implemented on the system CPU to use the APU as an accelerator for these mathematical operations.
Examples of the mathematical operations provided by the APIs are listed below:
Generic linear algebra:
Vector operations; for example, scalar product, vector summation, vector sorting
Matrix operations; for example, scalar multiplication, multiplication of two matrices
Cartesian and polar coordinate functions:
Polar to Cartesian conversion and vice versa
Advanced algebraic operations:
Covariance matrix generation with spatial smoothing
Gauss-Jordan elimination method
Jacobi eigenvalue decomposition
Frobenius norm (L2 norm) of a matrix
Fourier-related transforms
Fast Fourier Transform (FFT) and Inverse Fast Fourier Transform (IFFT)
Discrete Cosine Transform (DCT)
The APU consists of a number of individual hardware accelerators that are used in conjunction with each other by the various APIs in order to implement a specific mathematical operation.
The internal pipeline organization of the hardware accelerators allows a peak performance of 0.672 GFLOPS, where FLOPS performance is measured from a complex number multiply-and-accumulate operation consisting of 7 FLOPs operating at a 96MHz clock frequency.
The APU includes data memory for local storage of input and output data. The data memory consists of 1024 64-bit words, that is, 8KB of storage. The APU will access two complex numbers; that is, a total of 64 bits, from the data memory during each clock cycle. The data memory access width gives a memory read bandwidth of 12.288Gbps.