The CMSIS DSP library is written completely in C, builds and runs on the Cortex™-M4, Cortex-M3 and Cortex-M0 processors. Maximum use of Cortex-M4 intrinsics allows for high performance in Cortex M4-based devices. DSP intrinsics have been emulated through software in Cortex-M3 and M0-based devices. The DSP library performs about 2x faster on Cortex-M4 in comparison with Cortex-M3. It also performs about 10x faster on Cortex-M4 with floating point in comparison with Cortex-M3. The diagram on this slide shows a comparison between the CMSIS DSP Library performance on Cortex-M4 and -M3 platforms for different operations. For example, using DSP library on Cortex-M4, with fixed point, consumes 70% less CPU clock cycles as compared to Cortex-M3 for a matrix multiplication operation. Likewise, using DSP library on Cortex-M4 with floating point consumes 91% less CPU clock cycles when compared with Cortex-M3 for a correlation operation.