[Libclc-dev] [PATCH 2/8] Implement distance builtin

Aaron Watry awatry at gmail.com
Mon Mar 9 08:44:02 PDT 2015

I can attempt to handle that now...  It'll just take a bit of research on
my part. The AMD implementation of float erf assumes that you have support
for doubles when the device doesn't support subnormals.  I'm not sure that
this is a safe assumption to make, so we'll probably want/need to find some
other way to deal with that situation.

I can at least surround the existing implementation with an #if
defined(cl_khr_fp64) && !defined(SUBNORMALS_SUPPORTED)), and we'll just
have to have an else if for !SUBNORMALS && !DOUBLES.


On Mon, Mar 9, 2015 at 10:08 AM, Matt Arsenault <arsenm2 at gmail.com> wrote:

> On Mar 9, 2015, at 7:30 AM, Tom Stellard <tom at stellard.net> wrote:
> Supporting subnormals for float types is not required by OpenCL 1.1/1.2.
> Does this means we can ignore them in the library?
> I think it would be a good idea to deal with it while you’re already
> working on the function for later
> -Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/libclc-dev/attachments/20150309/6605def8/attachment.html>

More information about the Libclc-dev mailing list