bnjbvr commented on Issue #911:
Noted, thanks for the pings! I'll make sure to revisit these test cases once we have isel and regalloc in a position where we can test useful benchmarks.
fitzgen commented on Issue #911:
Here is another test case that hangs in register allocation: slow-regalloc.wat.gz
fitzgen commented on Issue #911:
Here is another that hangs in register allocation: testcase1.wasm.gz
Wondering if some of these might be the same as https://github.com/bytecodealliance/wasmtime/issues/1306...
alexcrichton commented on Issue #911:
Another fuzz bug that came up over the weekend, again looks like it's largely register allocation.
julian-seward1 commented on Issue #911:
Note that https://github.com/bytecodealliance/regalloc.rs/pull/89 fixes extreme slowness in the new regalloc's coalescing analysis, that was reported for some of the examples listed here.
julian-seward1 commented on Issue #911:
With the abovementioned fix (regalloc.rs PR89) in place, having looked at the profiles on one of these cases, I reckon we could probably chop another 10% off the allocation costs by rewriting the allocator's
avl_tree.rs
fn insert_wrk
to be non-recursive. Not sure if this is really worth it though; these test cases are pretty artificial -- a function with 2 basic blocks containing in total 400000+ instructions, including several tens of thousands of calls.
alexcrichton commented on Issue #911:
FWIW these are all fuzz bugs so it's not like they're representative of actual code. If they're generating ridiculously large functions it may be that we just need to tweak the fuzzer limits, so once your patch lands seems reasonable to basically close this out!
alexcrichton commented on Issue #911:
Another fuzz bug:
foo.wasm.gz
alexcrichton commented on Issue #911:
Another fuzz bug that cropped up today - https://gist.github.com/alexcrichton/fc234498ac40b718ed2d1b0aead6bc79
Last updated: Dec 23 2024 at 13:07 UTC