LangRef: Clarify behavior of llvm.is.fpclass with "denormal-fp-math"

[arsenm]

Summary

This does not read canonicalized values, which matches the behavior of
the basic DAG expansion using integer operations. There is a buggy
expansion using FP-operations if legal which needs to be adjusted to
account for this. We need to be aware of the denormal mode to switch
between is.fpclass calls and fcmp.

There's no real spec for denormal handling anywhere, but I believe
this is the most harmonious way to deal with the question considering
the requirement to not quiet input signaling nans.

This matches the behavior of MSVC's _fpclass and AMDGPU's
v_cmp_class_f32. fpclassify currently does not use this, and has
inconsistent behavior for denormals under DAZ on different platforms
(i.e. clang and gcc report FP_ZERO return FP_ZERO for a denormal under
DAZ, MSVC reports FP_SUBNORMAL).

[arsenm]

Attach correct patch

[arsenm]

ping

[jcranmer-intel]

llvm/docs/LangRef.rst

24885–24886

"The function does not read canonicalized values" is a phrase which took me a few tries to understand what was meant. Maybe something like "The function does not canonicalize its input value" would make more sense?

[arsenm]

Rephrase

[arsenm]

ping

[arsenm]

d4f38ef288c3a4cf2318182c8585a5c7e760877a

llvm/docs/LangRef.rst

 return any value and uses platform-independent representation of IEEE rounding
 modes.
 
 
 Floating-Point Test Intrinsics
 ------------------------------
 
 These functions get properties of floating-point values.
 
 
 '``llvm.is.fpclass``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Syntax:
 """""""
 
 ::
 

 
 The function checks if ``op`` belongs to any of the floating-point classes
 specified by ``test``. If ``op`` is a vector, then the check is made element by
 element. Each check yields an :ref:`i1 <t_integer>` result, which is ``true``,
 if the element value satisfies the specified test. The argument ``test`` is a
 bit mask where each bit specifies floating-point class to test. For example, the
 value 0x108 makes test for normal value, - bits 3 and 8 in it are set, which
 means that the function returns ``true`` if ``op`` is a positive or negative
-normal value. The function never raises floating-point exceptions.
+normal value. The function never raises floating-point exceptions. The
+function does not canonicalize its input value and does not depend
+on the floating-point environment. If the floating-point environment
+has a zeroing treatment of subnormal input values (such as indicated
+by the ``"denormal-fp-math"`` attribute), a subnormal value will be
+observed (will not be implicitly treated as zero).
 
 
 General Intrinsics
 ------------------
 
 This class of intrinsics is designed to be generic and has no specific
 purpose.