[PATCH] D128955: [WPD] Use new llvm.public.type.test intrinsic for potentially publicly visible classes

Wed Jul 20 14:41:48 PDT 2022

aeubanks added a comment.

the assert that there are no public.type.tests in LTT fails on `CodeGenCXX/thinlto-distributed-type-metadata.cpp`. for some reason clang doesn't go through the LTO API at [1], it just ends up calling the normal optimization pipeline. any ideas why?

[1] https://github.com/llvm/llvm-project/blob/0c1b32717bcffcf8edf95294e98933bd4c1e76ed/clang/lib/CodeGen/BackendUtil.cpp#L1173

================
Comment at: clang/test/CodeGenCXX/thinlto-distributed-type-metadata.cpp:22
 // OPT-NOT: @llvm.type.test
-// OPT-NOT: call void @llvm.assume
 // We should have only one @llvm.assume call, the one that was expanded
----------------
tejohnson wrote:
> aeubanks wrote:
> > tejohnson wrote:
> > > Why remove this one here and below?
> > we end up RAUWing the `public.type.test` with true and end up with `assume(true)` which is harmless and gets removed by something like instcombine
> Sure, but it should still be gone and can confirm that here? If it isn't useful to check for, then the comment should probably be changed too since it mentions assumes.
There's an `-O0` RUN line below where the `assume(true)` won't be removed.
Updated comments.

================
Comment at: clang/test/CodeGenCXX/thinlto_public_type_test_distributed.ll:23
+; RUN:   -o %t.native.o -x ir %t.o --save-temps=obj
+; RUN: llvm-dis %t.native.o.1.promote.bc -o - | FileCheck %s --check-prefix=HIDDEN
+
----------------
tejohnson wrote:
> Is there a reason why the hidden version checks after promote but the public version above checks after importing?
unintentional, fixed. I've changed everything to check 0.preopt since 1.promote wasn't appearing in some cases, not sure why

================
Comment at: lld/test/ELF/lto/update_public_type_test.ll:1
+; REQUIRES: x86
+
----------------
tejohnson wrote:
> aeubanks wrote:
> > tejohnson wrote:
> > > Add comments about what is being tested. Also good to test via llvm-lto2 which has a similar -whole-program-visibility option, rather than just via lld here. See for example llvm/test/ThinLTO/X86/devirt_vcall_vis_public.ll.
> > added comments
> > 
> > I extended llvm/test/ThinLTO/X86/devirt_vcall_vis_public.ll, is that ok?
> Yeah sure that's fine. But maybe also add a direct check there in both cases after promote that the public.type.test/assume were transformed as expected as you do in this test.
done

================
Comment at: llvm/lib/LTO/LTOBackend.cpp:595

+  updatePublicTypeTestCalls(Mod, Conf.HasWholeProgramVisibility);
+
----------------
aeubanks wrote:
> tejohnson wrote:
> > aeubanks wrote:
> > > tejohnson wrote:
> > > > aeubanks wrote:
> > > > > I'm not sure where the best place to put these is
> > > > I don't think that this will work for distributed ThinLTO, where the ThinLTO Backends are run via clang, which isn't going to have this config variable set (since it is set only during LTO linking). I think something may need to be recorded in the index as a flag there at thin link / indexing time that can be checked here.
> > > > 
> > > > It would then be nice to consolidate this handling into WPD itself, e.g. at the start of DevirtModule::run, but unfortunately we won't have a summary index for pure regular LTO WPD so I don't think that would work. So in here is ok but I would do it early to handle some of the early return cases.
> > > > 
> > > > (Please add a distributed ThinLTO test too)
> > > Added a distributed ThinLTO test: clang/test/CodeGenCXX/thinlto_public_type_test_distributed.ll.
> > > 
> > > I added `ModuleSummaryIndex::WithWholeProgramVisibility` but I'm not sure where I'd set it, and that's causing the test to fail.
> > I see from your most recent update that you added code to set/check this new index flag. But I think you would want to set it around the same place where we update the vcall visibility during the thin link (see calls to updateVCallVisibilityInIndex). And I would do it via a new method in the WPD source file that uses the existing hasWholeProgramVisibility() that checks the OR of the config flag or the internal option. Then you don't need to add a new flag to llvm-lto2 for this purpose.
> > 
> > Also, as I noted for the regular LTO handling elsewhere, I think you need to add similar handling for the legacy LTO implementation. See other callers to updateVCallVisibilityInIndex.
> ah, I needed a separate WPV flag in llvm-lto2 to set `llvm::lto::Config`, now everything passes
exposed the existing `hasWholeProgramVisibility`

================
Comment at: llvm/lib/Transforms/IPO/LowerTypeTests.cpp:1849
+      dropTypeTests(M, *TypeTestFunc);
+    Function *PublicTypeTestFunc =
+        M.getFunction(Intrinsic::getName(Intrinsic::public_type_test));
----------------
tejohnson wrote:
> aeubanks wrote:
> > tejohnson wrote:
> > > Shouldn't we have converted all of these by now?
> > Hmm, I vaguely remember adding this because some clang test failed (something to do with the extra LTT we add in BackendUtil.cpp), but reverting this doesn't cause any failures.
> Related to some of my other comments about ensuring we do the conversion for the legacy LTO interfaces, perhaps this we should assert here if there are any remaining public.type.test intrinsics? Not sure what the failure mode is if any of those stick around otherwise.
added a check

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D128955/new/

https://reviews.llvm.org/D128955