Jit the perl5 runops loop in proper execution order
WARNING: It does only work yet for simple functions! No branches, no non-local jumps. Only intel CPU's 32 and 64bit (i386 and amd64).
This perl5 jitter is super-simple.
The original compiled optree from the perl5 parser is a linked list in memory in non-execution order, with wide-spread jumps, almost in reverse order. Additionally the calls are indirect, and with a shared libperl even far, which is stops the CPU prefetching.
This Jit module properly aligns the run-time calls in linear "exec" order, so that the CPU can prefetch the next (and other) instructions. The old indirect far call within a shared libperl costs about 70 cycles, the new direct call near costs 3-5 cycles and enables CPU prefetching.
Additional memory costs: 2-10 byte per op
email@example.com written from scratch.
Copyright (c) 2010 Reini Urban
You may distribute under the terms of either the GNU General Public License or the Artistic License, as specified in the README file.