Public bug reported:
I reported that glibc 2.26-0ubuntu1 ( adopted in Arful-proposed ) causes
compilation error by using NVIDIA's CUDA9.0 or Intel's ICC
Under glibc2.24, they work.
It maybe comes from glibc226's new feature, i.e. "128-bit floating point
as defined by ISO/IEC/IEEE 60559:2011 (IEEE 754-2008) and ISO/IEC TS
18661-3:2015" . (see below, pls)
So, if possible, would like to withdraw from adopting glibc2.26 (and
switch back to glibc2.24)?
evidence
----------
1. Here is part of message during compiing Tensorflow with CUDA on UBUNTU17.10
beta with proposed-components. .
--------------
typedef _Complex float __cfloat128 __attribute__ ((__mode__ (__TC__)));
^
INFO: From Compiling external/nccl_archive/src/broadcast.cu.cc:
/usr/include/x86_64-linux-gnu/bits/floatn.h(61): error: invalid argument to
attribute "__mode__"
/usr/include/x86_64-linux-gnu/bits/floatn.h(73): error: identifier "__float128"
is undefined
--------------
2. Forums in Intel and NVIDIA
where issues the problems around glibc.2.26
Intel
https://software.intel.com/en-us/forums/intel-c-compiler/topic/742701
NVIDIA
https://devtalk.nvidia.com/default/topic/1023776/cuda-programming-and-performance/-request-add-nvcc-compatibility-with-glibc-2-26/
** Affects: glibc (Ubuntu)
Importance: Undecided
Status: New
** Tags: cuda glibc icc
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1716816
Title:
ISSUES of GLIBC2.26
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/1716816/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
