In commit 1496f4c1d507dab5521309520de5c77ba3b5d82d, I changed the execute cache mechanism in SVM1 to be simpler, probably faster, and endian-independent, a few hours after struggling with a kludgey version of execute caches that would work on big-endian systems.
However, I goofed: I said in the commit message that it's compatible so that existing unlinked .com files will continue to work, but that's not true. While I didn't change the meaning of any SVM1 byte code instructions, the way you invoke the new-style execute caches with an IJUMP instruction is different, so the old compiled code that jumps to them with JUMP won't work. I can undo this, but it'll be a bit of work and either we'd have to break big-endian SVM1 again or go back to the larger execute caches I briefly implemented in e44423bd982a31c6b7db77eaa7f96bbfbc503935 or something. Is anyone likely to have been using SVM1 extensively enough to care? (I used it on my ARM64 laptop, but now that the aarch64 compiler back end is nearly complete on the riastradh-20190113-aarch64 branch, I don't need that any more, not even as an intermediate stage now that cross-compilation works better. I used it on a PowerPC system, but, well, the old execute caches never worked on big-endian, so it was only with the new execute cache layout anyway!) _______________________________________________ MIT-Scheme-devel mailing list MIT-Scheme-devel@gnu.org https://lists.gnu.org/mailman/listinfo/mit-scheme-devel
