Skip to content

Conversation

@vtjnash
Copy link
Member

@vtjnash vtjnash commented Apr 26, 2019

A couple have not yet been updated, but this is a start towards updating the patch list Only one is missing now for LLVM 8. LLVM 7 is complete.

Also, this currently fails an offsetarray test. tracked this down to bad TBAA information and fixed it

Needs nanosoldier, and Keno suggested it should have a PkgEval run too.

@vtjnash vtjnash requested a review from vchuravy April 26, 2019 22:00
Copy link
Member

@vchuravy vchuravy left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Needs the version bump for windows

@vtjnash
Copy link
Member Author

vtjnash commented Apr 28, 2019

Almost working now (doesn't need to block merging though).

@vtjnash
Copy link
Member Author

vtjnash commented Apr 28, 2019

Ah, it wasn't LICM's fault (sorry LICM), someone (emit_select) was creating bad TBAA data for it. We should probably backport that commit fix.

@vchuravy
Copy link
Member

Probably separate f826bba out into it's own PR? It currently fails tests:

      From worker 6:	signal (6): Aborted
      From worker 6:	in expression starting at /buildworker/worker/tester_linux32/build/share/julia/test/bitarray.jl:1097
      From worker 6:	gsignal at /lib/i386-linux-gnu/libc.so.6 (unknown line)
      From worker 6:	abort at /lib/i386-linux-gnu/libc.so.6 (unknown line)
      From worker 6:	unknown function (ip: 0xf72780f5)
      From worker 6:	__assert_fail at /lib/i386-linux-gnu/libc.so.6 (unknown line)
      From worker 6:	mark_julia_slot at /buildworker/worker/package_linux32/build/src/codegen.cpp:624
      From worker 6:	mark_julia_slot at /buildworker/worker/package_linux32/build/src/intrinsics.cpp:854 [inlined]
      From worker 6:	emit_ifelse at /buildworker/worker/package_linux32/build/src/intrinsics.cpp:851 [inlined]

Would be good to have the backport for the scev patch for LLVM 8, otherwise LGTM.

cc: @banex19 since this might impact you.

@vtjnash vtjnash force-pushed the jn/llvm-up branch 2 times, most recently from bdc5256 to 39dff7f Compare April 29, 2019 15:54
@Keno
Copy link
Member

Keno commented May 3, 2019

https://gist.github.com/Keno/0107cea84de827845d204fc398d877ff for a llvm 8.0 version of the scev patch. It's a backport of the current state of the review rather than a forward port of the current patch, FYI.

vtjnash added 2 commits May 3, 2019 09:45
also disabled a new X86 pass (x86-disable-avoid-SFB) to avoid running
into an assertion error in that pass (see gist)
Copy link
Member

@vchuravy vchuravy left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

}
#else
FunctionPass *createNVVMReflectPass(unsigned int SmVersion);
extern "C" JL_DLLEXPORT void LLVMExtraAddNVVMReflectFunctionPass(LLVMPassManagerRef PM, unsigned int SmVersion)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

cc: @maleadt

@vchuravy vchuravy merged commit b96a591 into master May 3, 2019
@vchuravy vchuravy deleted the jn/llvm-up branch May 3, 2019 20:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants