Hi, I've been looking through this project quite a bit and I think the concept is super interesting and promising.
I just wanted to point out that most microcontroller targets that this could potentially run on can do double precision calculations, but they only have single precision FPUs (if they have an FPU at all). This effectively means that double precision calculations are going to be anywhere from 2x-32x slower than single precision calculations
With that in mind, it would be cool if there was a way to select via compilation arguments whether to use single vs double precision calculations, and that way we could get some performance enhancement on microcontrollers.
Hi, I've been looking through this project quite a bit and I think the concept is super interesting and promising.
I just wanted to point out that most microcontroller targets that this could potentially run on can do double precision calculations, but they only have single precision FPUs (if they have an FPU at all). This effectively means that double precision calculations are going to be anywhere from 2x-32x slower than single precision calculations
With that in mind, it would be cool if there was a way to select via compilation arguments whether to use single vs double precision calculations, and that way we could get some performance enhancement on microcontrollers.