[llvm] r212638 - AArch64: Better codegen for storing to __fp16.
scanon at apple.com
Thu Jul 10 12:31:51 PDT 2014
> On Jul 10, 2014, at 3:27 PM, Jim Grosbach <grosbach at apple.com> wrote:
>> On Jul 10, 2014, at 10:02 AM, Tim Northover <t.p.northover at gmail.com> wrote:
>>> Sort-of. It’s not wrong to go through float, but doing so does require care about rounding (which we weren’t doing).
>> And vice-versa, surely? If we were asked in the IR to perform a double
>> -> float -> fp16 conversion don't we have to do that unless we know
>> the result isn't changed?
> If we had good support for rounding modes, yes, definitely. As is it’s a bit muddled. Still best to avoid, you’re right. I think what we have now is at least marginally better than before, but we should definitely do more.
Absolutely. Greedily fusing double -> float -> __fp16 is correct with higher frequency than the old state of affairs, given the current front-end.
More information about the llvm-commits