You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Since arms processors supports both arm and thumb operation modes, we shall try both mode. Just a quick rembember :
Thumb mode is a smaller set, witch gives a smaller program memory footprint and has a small decoding machine ( less power ), but it also has limited access to registers.
Arm mode ( normal ) is bigger and uses all the processor can give, thus, spending more power.
The mode can be toogle through a instruction, so code can be splited in thumb and normal, but the processor must accept that.
Now the elua is being compiling with -mthumb. witch makes the processor use only thumb instructions. We shall test both with a cpu-intensive benchmark ( some taylor series maybe do the job ) and choose the best cost-benefit relation. We shall also seek for a way to compile with both instructions.
The text was updated successfully, but these errors were encountered:
According to jvpportal : the flag "-march=armv7-m " could be added to compile with thumb2 instructions.
I've used it, and I've also used the '-mtune=cortex-m4','-O3' and '-Os' flags. The memory footprint just got bigger or equal.
I think the options '-mcpu=cortex-m4' and '-mthumb' are fine enough to this port. We have to reduce
footprint in other things.
Since arms processors supports both arm and thumb operation modes, we shall try both mode. Just a quick rembember :
Thumb mode is a smaller set, witch gives a smaller program memory footprint and has a small decoding machine ( less power ), but it also has limited access to registers.
Arm mode ( normal ) is bigger and uses all the processor can give, thus, spending more power.
The mode can be toogle through a instruction, so code can be splited in thumb and normal, but the processor must accept that.
Now the elua is being compiling with -mthumb. witch makes the processor use only thumb instructions. We shall test both with a cpu-intensive benchmark ( some taylor series maybe do the job ) and choose the best cost-benefit relation. We shall also seek for a way to compile with both instructions.
The text was updated successfully, but these errors were encountered: