[PATCH] D113107: Support of expression granularity for _Float16.

[Steve Canon via Phabricator via cfe-commits]

scanon added a comment.

In D113107#3138671 <https://reviews.llvm.org/D113107#3138671>, @rjmccall wrote:

>> Basically agree with everything John said, with a note that #3 is not quite FP_CONTRACT, which allows evaluating an expression as if intermediate steps were infinitely-precise, but rather FLT_EVAL_METHOD == 32 as defined in ISO/IEC TS 18661-3: "evaluate operations and constants, whose semantic type has at most the range and precision of the _Float32 type, to the range and precision of the _Float32 type; evaluate all other operations and constants to the range and precision of the semantic type".
>
> Ah, thank you, I wasn't aware that the intent was for FP_CONTRACT to only allow evaluation as if infinitely precise.  That's much more limited; wouldn't it preclude things like doing `float` arithmetic with `double`-only hardware, or with the x87 operations without intermediate rounding?

Yes; the C definition of "contracted" is "A floating expression may be contracted, that is, evaluated as though it were a single operation, thereby omitting rounding errors implied by the source code and the expression evaluation method."

All the things you mention are covered by FLT_EVAL_METHOD, except that it can't actually represent all the combinations that one might choose to implement because it's a single scalar value. E.g. when targeting no-sse2, one might want to evaluate _Float16 and float using SSE, but double and long double (80-bit) using x87. There's no way to express those semantics. Fortunately, that's the only combination that I can think of someone wanting to use that's not expressible, and it's rather niche.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113107/new/

https://reviews.llvm.org/D113107