Code running slow on M4 CPU of a STM32H747 DISCO board

[Ss.9]

I'm using a STM32H747 DISCO boad and I just want to use the M4 CPU. When start, the M7 configs the Clock and does nothing since then. There is only a simple case running on the M4 CPU. The case takes 200ms on M4 while it took only 20ms on a Risc-v@300Mhz CPU.

I use the "-O2" and hardfloat during compiling. But I have no idea what should I do next.

Any Ideas?

The codes are arrays multiply operation with data of float type.

[KnarfB]

Are you sure that the compiler uses hard fp and issues floating point instructions? M4 is single precision only? What is the M4 clock?

[Ss.9 · Accepted solution]

Turns out I was using the AXISRAM implicitly allocated to M7. Now It is much faster. But The Dhrystone is 0.37 DMIPS/MHz, which is still not a good number for me. The CPU clock is 200Mhz and ahb is the same.

[Piranha]

For M4 core to run at decent speeds from the FLASH, one has to enable the ART.