Note: This is a public test instance of Red Hat Bugzilla. The data contained within is a snapshot of the live data so any changes you make will not be reflected in the production Bugzilla. Email is disabled so feel free to test any aspect of the site that you want. File any problems you find or give feedback at bugzilla.redhat.com.
Bug 1803351 - mesa builds fail with llvm 10.0.0 rc1/rc2 on ppc64le
Summary: mesa builds fail with llvm 10.0.0 rc1/rc2 on ppc64le
Keywords:
Status: CLOSED RAWHIDE
Alias: None
Product: Fedora
Classification: Fedora
Component: llvm
Version: 32
Hardware: Unspecified
OS: Unspecified
unspecified
unspecified
Target Milestone: ---
Assignee: Tom Stellard
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks: PPCTracker
TreeView+ depends on / blocked
 
Reported: 2020-02-15 12:26 UTC by Pete Walter
Modified: 2020-02-26 17:05 UTC (History)
11 users (show)

Fixed In Version: mesa-20.0.0-2.fc32
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
Environment:
Last Closed: 2020-02-26 17:05:44 UTC
Type: Bug
Embargoed:

Attachments (Terms of Use)

Description Pete Walter 2020-02-15 12:26:57 UTC 
mesa doesn't build with new llvm 10.0.0 rc1/rc2 on ppc64le. Lots of compiler errors about llvm headers. Could one of the llvm maintainers investigate this, please?

e.g. "/usr/include/llvm/ADT/PointerIntPair.h:150:8: error: enumerator value for 'IntShift' is not an integer constant"

https://koji.fedoraproject.org/koji/taskinfo?taskID=41514452
Comment 1 Dan Horák 2020-02-26 14:25:52 UTC 
the errors start with

In file included from /usr/include/llvm/IR/Module.h:19,
                 from ../src/gallium/state_trackers/clover/llvm/codegen.hpp:35,
                 from ../src/gallium/state_trackers/clover/llvm/codegen/bitcode.cpp:34:
/usr/include/llvm/ADT/STLExtras.h:559:49: error: 'std::index_sequence' has not been declared
  559 |   template <size_t... Ns> value_type deref(std::index_sequence<Ns...>) const {
      |                                                 ^~~~~~~~~~~~~~
/usr/include/llvm/ADT/STLExtras.h:559:63: error: expected ',' or '...' before '<' token
  559 |   template <size_t... Ns> value_type deref(std::index_sequence<Ns...>) const {
      |                                                               ^
/usr/include/llvm/ADT/STLExtras.h:564:36: error: 'std::index_sequence' has not been declared
  564 |   decltype(iterators) tup_inc(std::index_sequence<Ns...>) const {
      |                                    ^~~~~~~~~~~~~~
/usr/include/llvm/ADT/STLExtras.h:564:50: error: expected ',' or '...' before '<' token
  564 |   decltype(iterators) tup_inc(std::index_sequence<Ns...>) const {
      |                                                  ^
/usr/include/llvm/ADT/STLExtras.h:569:36: error: 'std::index_sequence' has not been declared

so my first guess is a missing #include ...
Comment 2 Tom Stellard 2020-02-26 14:51:12 UTC 
This looks like a bug in the mesa build.  On ppc64le it's compiling with -std=gnu++11, but on other arches it is compiling with -std=gnu++14.  LLVM uses c++14 features, so compiling with -std=gnu++11 won't work.
Comment 3 Kalev Lember 2020-02-26 16:03:43 UTC 
Ahh, that makes sense. Indeed, mesa has this in meson.build:

  if host_machine.cpu_family().startswith('ppc') and cpp.compiles('''
      #if !defined(__VEC__) || !defined(__ALTIVEC__)
      #error "AltiVec not enabled"
      #endif''',
      name : 'Altivec')
    clover_cpp_std += ['cpp_std=gnu++11']
  endif

... which ends up forcing gnu++11 on ppc arches.
Comment 4 Kalev Lember 2020-02-26 16:15:47 UTC 
I went ahead and hacked around this in https://src.fedoraproject.org/rpms/mesa/c/bd392f431cc481ea754581d3052959bb7454ca5b
Comment 5 Dan Horák 2020-02-26 16:52:30 UTC 
(In reply to Kalev Lember from comment #3)
> Ahh, that makes sense. Indeed, mesa has this in meson.build:
> 
>   if host_machine.cpu_family().startswith('ppc') and cpp.compiles('''
>       #if !defined(__VEC__) || !defined(__ALTIVEC__)
>       #error "AltiVec not enabled"
>       #endif''',
>       name : 'Altivec')
>     clover_cpp_std += ['cpp_std=gnu++11']
>   endif
> 
> ... which ends up forcing gnu++11 on ppc arches.

which is probably from times, when the default was c++11 and using the altivec required gnu++11, because of some gnu extensions
Note You need to log in before you can comment on or make changes to this bug.