Definitely in quite a few places. We've had to insert inline assembler in a couple of places as it was using blocks of code to implement what would be one M4 instruction, but not optimising it across all that it had created.Branches (and related) take 1 cycle less.Also interesting that the M0+ scores better than the M0 in those benchmarks
It looks like the usual compilers are not really optimized for M0+; the code looks more like scheduled for M3/M4 (or even M7) with the instruction set limited to M0. So in general, generated code might be better for M33.
Statistics: Posted by MikeDB — Thu Feb 27, 2025 7:09 am