Joe Ranweiler

> Actually, I think @ranweiler already wrote the documentation :) https://github.com/microsoft/onefuzz/blob/main/src/agent/LibFuzzerDotnetLoader/README.md ☺️ Note for end users: these docs are mostly for an _auxiliary_ project that is part of OneFuzz, and...

Hi! This bug made me curious, but I'm having trouble reproducing it locally. Based on [this comment](https://github.com/trailofbits/remill/blob/master/tests/X86/FMA/VFMADDSD.S#L43-L44), my expectation is that adding the inputs `1.2, 2.3, 3.4` to the constant...

@mike-myers-tob, I think this issue might really be about the precision of C++ `double`. I came up with the following conceptual repro, based upon your known failing test case of...

Okay, so the algorithm in the above paper seems to work well. Partly to satisfy my own curiosity, I now have a fairly thoroughly fuzz-tested implementation (whose test harness I...

> the year 2099 (?) Welcome to the future! :rocket: > Briefly looking at the paper though, it seems like the root problem is a rounding behavior called "round to...

Okay! Back from the holidays, with some conclusions. My experiments in FMA emulation can be found [here](https://github.com/ranweiler/fmaemu). I think the round-to-odd based emulation is a no-go, for both theoretical and...

> [...] whether or not the current non-fused operations produce NaNs in those cases. In my property tests, whenever the FMA instruction produced a non-NaN and RTO emulation produced a...

I'd like to wrap up the FBLD/FBSTP impls in #219, but this is a bit of cross-cutting issue, and it might make sense to address first. Otherwise I'll have to...

This is mostly done in #224, but it needs one more pass to add NaN test cases to _all_ the instructions that consume the X87 stack. Some of the existing...

I'd like to volunteer as an additional maintainer. I depend on this library for various projects, and would rather contribute to its success than have to fork it or fall...