--- Begin Message ---
|
Subject: |
Compiling 3.8 with nvhpc 22.1 : malloca.c 64 bits operations with signed and unsigned operands are currently unsupported by __builtin_add_overflow |
|
Date: |
Fri, 14 Jan 2022 12:33:40 +0100 |
|
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.14.0 |
Hello,
I'm trying to compile diffutils 3.8 with nvc 22.1.
I get an error:
<<
"diffutils-3.8/lib/malloca.c", line 52: error: 64 bits operations with
signed and unsigned operands are currently unsupported by
__builtin_add_overflow
if (!INT_ADD_WRAPV (n, plus, &nplus) && !xalloc_oversized (nplus, 1))
^
1 error detected in the compilation of "diffutils-3.8/lib/malloca.c".
>>
It was fine with 3.7.
Thanks for your help.
Olivier Cessenat
Ubuntu 20.04 - compilation still fine with gcc 9.3.0
--- End Message ---
--- Begin Message ---
|
Subject: |
Re: [bug-diffutils] bug#53256: Compiling 3.8 with nvhpc 22.1 : malloca.c 64 bits operations with signed and unsigned operands are currently unsupported by __builtin_add_overflow |
|
Date: |
Fri, 14 Jan 2022 19:53:39 -0800 |
|
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.3.1 |
On 1/14/22 03:33, Olivier Cessenat wrote:
"diffutils-3.8/lib/malloca.c", line 52: error: 64 bits operations with
signed and unsigned operands are currently unsupported by
__builtin_add_overflow
Thanks for the bug report. Although nvc 22.1 defines __GNUC__ to be 11
which means it supports GNU C as of GCC 11, that's obviously untrue for
__builtin_add_overflow. That's unfortunate. Perhaps you could report
this as a bug to NVIDIA, as their compiler shouldn't claim to support
GNU C features that it lacks. nvc can set __GNUC__ to be the version of
GNU C that it actually does support.
To work around the bug I installed the attached into Gnulib and this
should appear in the next diffutils release. In the meantime you can
compile with gcc.
0001-intprops-work-around-nvc-22.1-bug.patch
Description: Text Data
--- End Message ---