Optimise bounds checking for memory and tables #38
Labels
optimisation
something that needs doing for performance
Comments
SimonJF
added a commit
that referenced
this issue
Jul 27, 2018
Implements #38 for memory (but not tables). Also cleans up a lot of spaghetti code for division, which I rewrote when debugging. This PR enables implicit memory bounds checks by using mmap instead of malloc for linear memory. The key idea is to rely on hardware protection, as described in the original WASM paper, instead of explicit checks (although these can be re-enabled via the --explicit-bounds-checks flag. Doing so yields significant speedups -- we are on par with (and sometimes beat!) V8 in the benchmarks I've tried.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Currently, memory (and soon tables) require explicit bounds checks to ensure safety. This incurs an additional read, and certainly additional branches.
One technique to reduce this would be to use
mmapto ensure that out-of-bounds accesses cause a segmentation fault, and then handle the segmentation fault by trapping. Thus, the assumption is that the memory access will succeed, but that there's a guarantee that a segfault will be raised on out-of-bounds accesses.The text was updated successfully, but these errors were encountered: