[llvm-commits] [PATCH] review for fix of profiling test failures on ARM

[NAKAMURA Takumi]

2012/9/13 David Tweed <david.tweed at arm.com>:
> on the Linux on ARM there are three test failures in profiling
> (Analysis/Profiling/load-branch-weights-ifs.ll,
> Analysis/Profiling/load-branch-weights-loops.ll,
> Analysis/Profiling/load-branch-weights-switches.ll). Looking at what's
> happening it's actually from failure to run some JITted code, which is
> happening because it's attempting to use the default "old" JIT. Using the
> MCJIT the JITing then works and the test passes. I gather pretty much every
> platform MCJIT is at least as good as the old JIT, so the attached patch
> makes these tests explicitly use MCJIT. The tests have been checked that
> they pass on both x86-64 Linux and ARM Linux, and the important tests (that
> profiling is doing the right things, rather than jit-ing) are now being run
> on ARM.
>
> Could this patch be reviewed with a view to being committed please?

David,

Some targets don't support MCJIT. Please consider Win32 PECOFF.
(3 of them don't work on Win32 due to atexit issue, though)

What would happen if -use-mcjit were set by default for arm-elf?

...Takumi