[PATCH] D99815: [CSSPGO][Test] XFAIL profile-context-tracker-debug.ll on AIX

[Hongtao Yu via Phabricator via llvm-commits]

hoy added a comment.

In D99815#2684221 <https://reviews.llvm.org/D99815#2684221>, @jsji wrote:

> In D99815#2684210 <https://reviews.llvm.org/D99815#2684210>, @hoy wrote:
>
>> In D99815#2683075 <https://reviews.llvm.org/D99815#2683075>, @jsji wrote:
>>
>>> So this is not only failing for AIX, it is actually depends on the libc++ implementation.  It may fail on Linux if using non default libc++ .
>>>
>>> Any plan of fixing the code to not using std::hash<std::string> ? Thanks. @hoy
>>
>> Thanks for pointing it out. Yes, we think that may be related to libstdc++ function `_Hash_bytes`. My plan is to tweak the test with Check-DAG since the promoting order of the children should not cause codegen diffs. We could also sort the children by names instead of hash value for a more stable order across platforms, but that''ll likely incur more overhead. What do you think?
>>
>>   namespace std
>>   {
>>   _GLIBCXX_BEGIN_NAMESPACE_VERSION
>>   
>>     // Hash function implementation for the nontrivial specialization.
>>     // All of them are based on a primitive that hashes a pointer to a
>>     // byte array. The actual hash algorithm is not guaranteed to stay
>>     // the same from release to release -- it may be updated or tuned to
>>     // improve hash quality or speed.
>>     size_t
>>     _Hash_bytes(const void* __ptr, size_t __len, size_t __seed);
>
> Thanks, I am OK with tweaking test with check-DAG as long as you can confirm that this is safe in codegen ( I am not familiar with these code enough to judge).

Thanks. Do you still have the output of the failing test? That'll help to tweak the test.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D99815/new/

https://reviews.llvm.org/D99815