[LLVMdev] RFC: Missing canonicalization in LLVM

[Smith, Kevin B]

OK.

Then I think the clear choice ought to be to canonicalize to the same sized int representation to account for architectures
whose FP may not exactly match LLVMs assumptions.  Even when fp exceptions are masked, X87 load changes bit pattern of SNan to
QNan, so it is going to be generally incorrect to convert an integer load to an FP load, as that won’t properly preserve the all possible bit patterns
of the integer, unless great pains are taken not to use fld instruction to implement FP load, which just seems wrong as well.

Kevin Smith

From: chandlerc at google.com [mailto:chandlerc at google.com] On Behalf Of Chandler Carruth
Sent: Wednesday, January 21, 2015 3:34 PM
To: Smith, Kevin B
Cc: Chandler Carruth; LLVM Developers Mailing List
Subject: Re: [LLVMdev] RFC: Missing canonicalization in LLVM


On Wed, Jan 21, 2015 at 3:06 PM, Smith, Kevin B <kevin.b.smith at intel.com<mailto:kevin.b.smith at intel.com>> wrote:
I don’t think these are equivalent representations.  The one with the float loads and stores has the potential of FP exceptions
both during the load and during the store

LLVM explicitly doesn't support FP exceptions on loads and stores. And it would break the world to add it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150121/c3c6351c/attachment.html>