Comment #2 on issue 2510 by [email protected]: OOM on sqlite http://code.google.com/p/v8/issues/detail?id=2510
The analysis so far: The emscripten code results in a rather large and dense Hydrogen graph: At the point where OOM is reached, we e.g. have 20654 blocks and many of them have 877 phis. When we try to eliminate redundant phis, we naively try to build a work list of all these phis, but run OOM while doing that (2.7GB doesn't seem to be enough :-).
If there is no non-fixed point algorithm for this phase (have to check that), we have to do a space-time tradeoff here: Don't build the whole work list at once, but limit its size, and re-do all the stuff until we find no more work.
-- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
