Benchmarks on this later, when they become available. On Wed, Jan 14, 2026 at 1:56 PM John Sarabacha <[email protected]> wrote:
> A true JIT compiler is feasible for amForth (riscv & arm). This will give > amForth a performance boost comparable to C compiled code, maybe even > better. C compiled code needs to reside in flash memory unless it is > copied and relocated to sram. The JIT compiler resides in flash but can > compile directly to sram. The amForth core words (forth assembler > instructions) can serve as the DNA words for the compiled programs in > sram. The JIT compiler can be invoked in interpreter mode to compile new > words that use existing DNA words (by copying the existing previously > assembled machine instructions for the core word) directly to sram for each > word involved (eg. shared words - that have a sequence of words). The > advantage here is having an on board JIT compiler that can generate > (sequence) code without needing the ITC to mediate execution and running in > sram. C implementations of forth would have more trouble doing this than > amForth (being based on assembler code). > This would be a major performance boost since the majority of amForth's > core words have only a few machine instructions (a lot only have one), > most of the time the ITC is consuming the majority of the instruction > cycles for word execution. > > Regards, > John S > _______________________________________________ Amforth-devel mailing list for http://amforth.sf.net/ [email protected] https://lists.sourceforge.net/lists/listinfo/amforth-devel
