[PATCH] D87835: [APFloat] prevent NaN morphing into Inf on conversion (PR43907)

[John McCall via Phabricator via llvm-commits]

rjmccall added inline comments.


================
Comment at: llvm/lib/Support/APFloat.cpp:2248
+    // bit in the significand must be set (otherwise it is Inf).
+    // Set the 1st bit after the quiet bit to preserve SNaN.
+    // TODO: We should check if the shifting loses any bits above here and set
----------------
Probably best to clarify:

> This can only happen with sNaN, so set the 1st bit after the quiet bit so that we still have an sNaN.


================
Comment at: llvm/lib/Support/APFloat.cpp:2251
+    //       losesInfo.
+    // TODO: Would it be better to set quiet and return opInvalidOp?
+    if (APInt::tcIsZero(significandParts(), newPartCount)) {
----------------
That might be more reasonable than manufacturing an artificial sNaN, yeah.  But if we're going to do that, it should happen unconditionally, not just when we ended up with a zero significand.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D87835/new/

https://reviews.llvm.org/D87835