Skip to content

Commit

Permalink
add discussion of known issues in release notes.
Browse files Browse the repository at this point in the history
  • Loading branch information
jmellorcrummey committed Jun 29, 2017
1 parent be881d9 commit 1d908a3
Showing 1 changed file with 32 additions and 0 deletions.
32 changes: 32 additions & 0 deletions README.ReleaseNotes
Original file line number Diff line number Diff line change
Expand Up @@ -239,6 +239,38 @@ bug prevented using hpcrun to profile programs launched with shell scripts.
-- fixed bug in hpcstruct in getRealPath() that caused hpcstruct
to sometimes report incorrect file names.


Known issues:

When profiling optimized code with HPCToolkit, one may find that a program
generates a significant number of "partial unwinds" where the call stack
can't be unwound all the way up to main. This more commonly happens on
x86-64 architectures than on PowerPC and ARM. A large number partial unwinds
may make it harder to use the top-down calling context view in hpcviewer,
which works best when call stacks unwind all the way up to main. Even
with significant numbers of partial unwinds, the bottom-up caller's view
and the flat view in hpcviewer can be used effectively for analyzing
performance. Ongoing work aims to improve call stack unwinding of
optimized code by employing compiler-generated unwinding information
where available in addition to using binary analysis to discover unwinding
recipes.

On x86-64, hpcfnbounds occasionally is too aggressive about inferring the
presence of stripped functions in optimized programs. We have noticed
this particularly for optimized Fortran. This can cause "partial unwinds",
where a call stack can't be unwound fully up to main. Improving this
analysis is the subject of ongoing work.

When using with the LLVM OpenMP runtime's OMPT support, measurements
of programs compiled with GCC using HPCToolkit's ompt branch sometimes
reveal implementation-level stack frames that belong to the OpenMP
runtime system. This will improve with the transition of HPCToolkit
and the LLVM OpenMP runtime to the new OMPT ABI designed for OpenMP
5.0. This transition should occur over the next 6 months. In the
meantime, there is nothing wrong with the quality of the information
collected. The only problem is HPCToolkit's measurements reveal more of
an implementation-level view of OpenMP than intended.

----------------------------------------
HPCToolkit Version 2016.12, Dec 2016
----------------------------------------
Expand Down

0 comments on commit 1d908a3

Please sign in to comment.