Peter Goodman

@sschriner can you take this one? You can find a suitable alternative in `include/drlojekyll/Util/BitManipulation.h` and just copy it into this file.

I don't remember why :-( I think the basic issue is that we don't have a good way of managing precision-related issues across the semantics, and this comes up in...

You are right about mxcsr. Though one thing to consider is that once we lift up to bitcode, and say, compile back down to machine code, we have no real...

The `_IF_32BIT(REG_SS_BASE)` expands to `, REG_SS_BASE` [1] when `ADDRESS_SIZE_BITS == 32`, and expands to nothing otherwise, meaning that the two-argument version of `WritePtr` is called [2]: `WritePtr(next_sp, REG_SS_BASE)`, and so...

Does the return address get written in the "destination" address space? If so, it may make sense to have a sync hyper call, somewhat like... ```c++ template DEF_SEM(CALL_FAR_PTR, S1 target_pc,...

Ping @tathanhdinh.

ping @tathanhdinh

My understanding of the major complexity of this issue is that it's more than just "support TCG": it's about supporting the individual CPU implementations of QEMU which generate TCG.

That's for x87 precision differences, not sse/avx ones.

I am curious about the case of spurious NaNs in the round-to-odd algorithm, and whether or not the current non-fused operations produce NaNs in those cases. If the unfused multiple+add...