[LLVMdev] [cfe-dev] computing a conservatively rounded square of a double (original) (raw)

Geoffrey Irving irving at naml.us
Thu Mar 27 09:42:37 PDT 2014

It would be wonderful to get this in! What would the front-end interface be (-frounding-math, #pragma, special builtins, etc.)?

Geoffrey

On Thu, Mar 27, 2014 at 9:18 AM, Philip Reames <listmail at philipreames.com> wrote:

Paul,

If any of those llvm changes are still pending, let me know. I'd be happy to review. Philip

On 03/27/2014 08:46 AM, Robinson, Paul wrote: While I did propose a patch to add this attribute to Clang, and it was approved, I never committed it because the LLVM work to fully support it has been dangling in nobody-wants-to-review-it limbo forever. And it seemed like I shouldn't put the attribute into Clang if LLVM didn't fully honor the contract.

I'll redo the Clang part (Aaron Ballman has refactored the attribute-handling stuff about a hundred times since I did the original patch so it no longer applies) and hopefully get that into Clang soon. Thanks, --paulr Geoffrey - In the short term, if you marked this function as "attribute((optnone))", that might have the effect you desire. See http://clang-developers.42468.n3.nabble.com/RFC-add-Function-Attribute-to-disable-optimization-td4032641.html

More information about the llvm-dev mailing list