@dexonsmith Thank you for your reply. Are you considering the case where -print-after-all is used along with -print-* pass? If so, shouldn't we have -print-module explicitly in the test case?

Hello, can you please explain a bit more about this change? What kind of redundant outputs avoided by this? Thanks!

@Hahnfeld Got it. Then it seems that there's not much we can do here :(

@Hahnfeld I'm not an OpenMP expert and don't have much strong argument here, but IMHO it might be confusing to have LLVM_ENABLE_FFI and LIBOMPTARGET_ENABLE_FFI have different default value from user's perspective. I wonder if we can set it off by default and explicitly turn it on for builds with test?

@grokos Thanks for the suggestion. What about this?

Addressed comments from @Hahnfeld. Thanks!

Use isReachableFromEntry. Thank you for the suggestion!

Thank you for your comments @efriedma! I addressed your comments and added a test case. The test segfaults clang without proposed change.

Friendly ping again. This is a fix for a compiler crash.

Friendly ping. Thanks!

Friendly ping for comments. Thanks!

Hello @rsmith, @wenlei and I took another look at this, and we couldn't find any use of AnonymousTagLocations outside of debug info. If that's actually the case, wouldn't it make sense to have DebugColumnInfo to control the field even if AnonymousTagLocations the part of generic printing policy?

@wmi Thank you for the concrete example! I think what we need for your example is context-sensitive profiling and function specialization, not inlining. Admittedly we don't have an infrastructure in LLVM to support context-sensitive profiling for non-inlined case and we don't perform context sensitive function specialization...

@wmi Thank you for the detailed explanation!

@wmi Thanks for taking a look! Performance numbers were neutral from our side.

Check for the new pass manager added. Thanks @wmi for the comment!

@wmi Thanks for the reply! I can totally understand that entry count is not as precise as total count, but still don't think current implementation is the right way to address the issue. As I mentioned in the summary it compares two different things (instruction level counter vs function level counter), opens up a possibility for optimizing against wrong function (e.g. long and cold function), and makes it hard to find the root cause of the performance issue.

Remove unnecessary metadata.

Addressing comments from @Hahnfeld. Thanks!

@kkwli0 Thanks for letting me know! But wouldn't https://reviews.llvm.org/D57848 always fall to "else" case added if the compiler is not GCC?

LGTM

Can you please provide a bit more context about when the client might want this function?

LGTM

Comments for the test updated. Thanks for the comment!

Friendly ping. Thanks!

@rsmith I see. Thank you for the clarification!

I think the context is Derived here. My understanding of http://wg21.link/p0968r0#2227 (in this patch's context) is that when Derived is aggregate initialized, the destructor for each element of Base is potentially invoked as well.

@rsmith @dblaikie Thank you for the comments! It seems that this is not the appropriate way to handle the issue. I'll find different way to resolve the problem.

@dblaikie I see. The problem we're experiencing is that with Clang's naming scheme we end up having different function name between the original source and the preprocessed source (as macro expansion changes the column number). This doesn't work well for me if I want to use distcc on top of our caching system, as the caching scheme expects the output to be same as far as the original source has not been changed (regardless of it's compiled directly or first preprocessed then compiled), but the distcc preprocesses the source locally then sends it to the remote build machine (and we do not turn distcc on for all workflow). I wonder if you have any suggestion to resolve this issue? Thanks!

@aprantl It is a debug info. If you compile test/CodeGenCXX/debug-info-anonymous.cpp file with clang -g2 -emit-llvm -S , you will find debug metadata something like distinct !DISubprogram(name: "foo<X::(anonymous enum at /home/twoh/llvms/llvm/tools/clang/test/CodeGenCXX/debug-info-anonymous.cpp:9:3)>", linkageName: "_Z3fooIN1XUt_EEiT_" ..., which will eventually be included in .debug_info section.

Remove conflict line.

@joerg Sorry but I'm not sure if I understand your question. This doesn't pretend to honor source code order, but makes linker to place "hot" functions under .text.hot section (There's no guarantee of ordering between functions inside .hot.text section) while "cold" functions under .text.unlikely section. This is purely for performance.

Rebase. Tests are provided in the clang counterpart (D34796).

Rebase. Sorry I somehow missed the recent comments. I addresses @davidxl's comment on documentation. Thanks!

ping!

Addressing comments from @echristo. Reverted option name change, and added a test case. Sorry I haven't work on this code for a while so it took time to invent a test case.

Hello, I observed a case where atomic builtin generates libcall when the corresponding sync builtin generates an atomic instruction (https://bugs.llvm.org/show_bug.cgi?id=38846). It seems that the alignment checking for __atomic builtins (line 759 of this patch) results the difference, and wonder if the check is actually necessary. Could anyone please shed some light on understanding this? Thanks!

Hello, I wonder if we need to keep linkonce_odr symbols live here as well. I observe a case that a vtable for template class initiated has linkonce_odr linkage and marked dead here, which results compiler crash at WholeProgramDevirt because the global variable for vtable doesn't have initializer (https://github.com/llvm-mirror/llvm/blob/master/lib/Transforms/IPO/WholeProgramDevirt.cpp#L676 assumes that GV has the initializer). Thanks!

Addressing comments from @Hahnfeld. Thanks!

Addressing comment from @mssimpso. Thanks!

Hello @jlpeyton @tlwilmar, is there a chance that this breaks runtime/test/ompt/misc/api_calls_from_other_thread.cpp test? ompt_get_num_places() is expected to return 0 from the test but with this patch it returns 1, and I think it is because kmp_affinity_num_masks is set to 1 from kmp_create_affinity_none_places().

Addressing comments from @fhahn and @junbuml's comments. Thank you!

Thanks @fhahn for the comments! I added comments about why we don't perform splitting for the callsites inside the landing pad. Please let me know if you think it is too verbose.

Addressing @junbuml's comments. Thanks!

Update test to check f2.llvm.0 as well.

@tejohnson Thanks for the clarification. Regarding hotness, I'm not sure if providing "some" hotness is better than leaving it as unknown if profile data is not provided (If profile data is given, as you said, VP metadata will be attached to the callsite). I'm afraid that synthesized hotness may confuse optimizers, but please let me know if you have different idea.

@tejohnson I think your right. What I meant was that when the metadata is imported to bar.o, it references f1 and f2 by their promoted names, which makes the declarations with the promoted names to be added. Did I get it right, or still miss something?

clang-format

@jkorous-apple Got it. I agree that it would be better to move the comments to the header. Will land it soon. Thanks!

Thanks @jkorous-apple for the comment. I think your suggestion is a more
precise description for the implementation, and adjusted the comments
accordingly.

@jkorous-apple Thanks for the comments! Yeah, I was thinking of O(lenght_of_string) approach, but considering the complicatedness of the implementation (I guess the real implementation would be a bit more complex than your pseudo implementation to handle quote and '\n\r' '\r\n' cases) I decided to stay with O(length_of_string * number_of_endlines_in_string) but optimizing the number of move operations.

Thanks @jkorous-apple for your comments. I modified the type for the variables and replaced unnecessary inserts and erases with updates.

Addressing @vsapsai's comments. Thank you for the suggestion! Added test case actually finds an off-by-one error in the original patch. I improved the comments as well.

Ping. Thanks!

Ping. Thanks!

clang-format

Ping. Thanks!