Opened 7 years ago

Closed 6 years ago

#11852 closed Bugs (fixed)

GCC 4.8.5 & CUDA: float128 missing

Reported by: a.huebl@… Owned by: John Maddock
Milestone: Boost 1.61.0 Component: config
Version: Boost 1.60.0 Severity: Showstopper
Keywords: cuda nvcc gcc float128 Cc:

Description

Similar to

​https://svn.boost.org/trac/boost/ticket/8048

float128 is still unsupported when compiling with nvcc (tested with CUDA 7.5.18). First noticed with the latest release (1.60.0) and GCC 4.8.5 but should affect all previous releases depending on used modules.

Attached is a patch for config/compiler/gcc.hpp adding the missing && !defined(__CUDACC__).

Attachments (1)

float128GccNvcc.patch​ (552 bytes ) - added by a.huebl@… 7 years ago.
Patch to avoid float128 with nvcc

Download all attachments as: .zip

Change History (8)

by a.huebl@…, 7 years ago

Attachment: float128GccNvcc.patch​ added

Patch to avoid float128 with nvcc

comment:2 by a.huebl@…, 7 years ago

Update: when using nvcc with -std=c++11 the 128 variants are available.

Patch updated here:

​https://patch-diff.githubusercontent.com/raw/boostorg/config/pull/83.patch

comment:3 by a.huebl@…, 7 years ago

updated the intel and clang checks, too now.

note: of course, 128 variants will with current hardware only work on the host side, afaik.

comment:4 by a.huebl@…, 7 years ago

the reason this is triggered already with header includes with 1.60.0 and not with 1.59.0 is that 

type_traits/is_floating_point.hpp

implements new __float128 type traits in the current release.

comment:5 by a.huebl@…, 7 years ago

The PR was accepted and the issue should be fixed with the next release.

comment:6 by anonymous, 7 years ago

Milestone: To Be Determined → Boost 1.61.0

comment:7 by John Maddock, 6 years ago

Resolution: → fixed
Status: new → closed
Note: See TracTickets for help on using tickets.