[PATCH] D99160: [X86][FastISEL] Support DW_TAG_call_site_parameter with FastISEL

Alok Kumar Sharma via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Thu May 13 01:00:14 PDT 2021


alok added a comment.

In D99160#2700326 <https://reviews.llvm.org/D99160#2700326>, @alok wrote:

> In D99160#2699038 <https://reviews.llvm.org/D99160#2699038>, @dblaikie wrote:
>
>> In D99160#2698980 <https://reviews.llvm.org/D99160#2698980>, @alok wrote:
>>
>>> In D99160#2671899 <https://reviews.llvm.org/D99160#2671899>, @dblaikie wrote:
>>>
>>>> In D99160#2670460 <https://reviews.llvm.org/D99160#2670460>, @djtodoro wrote:
>>>>
>>>>> In D99160#2669576 <https://reviews.llvm.org/D99160#2669576>, @dblaikie wrote:
>>>>>
>>>>>> In D99160#2668977 <https://reviews.llvm.org/D99160#2668977>, @djtodoro wrote:
>>>>>>
>>>>>>> I think that the Debug Entry Values feature should not be enabled by default for non optimized code, so the `TargetOptions::ShouldEmitDebugEntryValues()` should be patched with checking of optimization level (it should be > 0).
>>>>>>
>>>>>> That's currently intended to be already handled by the frontend, right? (clang only sets `EnableDebugEntryValues` (which `ShouldEmitDebugEntryValues` checks (hmm, it checks under 'or', not 'and', so I'm not sure where the "only above -O0" is implemented, but it is implemented somewhere?) if optimizations are enabled, yeah?)
>>>>>>
>>>>>> Oh, is entry_values actually not conditionalized? It's only the call_site support that's currently conditionalized on "above -O0"?
>>>>>
>>>>> Looks like there is no explicit check of optimization level (above "-O0"), neither on frontend nor backend for entry-values generation. I think it is the situation since there should not be any optimization (at least that I am aware of, in the case of C/C++) that would cause the entry-values generation...
>>>>>
>>>>>> Hmm - If that's the case, and we currently have some cases where entry_values are emitted at -O0, I'm not sure /not/ emitting those is the right call either. If we believe/have data to show that there are so few useful uses of entry_value at -O0 that it's not worth the DWARF size growth to put call_sites in at -O0, then I think it might still be worth leaving the entry_values in (unless they take up a bunch of extra space) for the cases of mixed optimization compilation (-O0 some code you're debugging, but building the rest with optimizations).
>>>>>
>>>>> Yeah... That is valuable example... I am thinking in that direction as well, and I am closer to the enabling it for -O0 case if that is useful (and there is no dramatic cost in terms of DWARF size).
>>>>
>>>> Does anyone have this example (where DW_OP_entry_value is used at -O0)? It'd be great to look at it & see if it's a case of unnecessarily losing the location, or legitimately losing it and using entry_value for best-effort recovery (& then a question of whether the loss is appropriate at -O0, or if we want to pessimize -O0 further to avoid the loss).
>>
>> I think this ^ still needs understanding/investigation. Do you have an example with OP_entry_value at -O0?
>>
>>>> I'd worry about turning on call_sites at -O0 - there'd be a lot more calls (especially for C++ code with lots of implicit operations), but numbers will be needed in any case, so not worth much speculation.
>>>
>>> Sorry for late response.
>>> I tried building  https://github.com/flang-compiler/classic-flang-llvm-project.git (branch release_11x) with compiler (current patch (https://reviews.llvm.org/D99160) and https://reviews.llvm.org/D99238) with -O0 -g . 
>>> Interestingly there was no difference.
>>> Reason: https://reviews.llvm.org/D99238 is not sufficient for clang/clang++ to enable call-site generation with FastISel though it is sufficient for Flang compiler.
>>> Below additional patch is needed to generate call-sites
>>>
>>>   `````````````````````````````````````
>>>
>>> diff --git a/clang/lib/CodeGen/CGDebugInfo.cpp b/clang/lib/CodeGen/CGDebugInfo.cpp
>>> index a77f52bd235b..8d4e11faa018 100644
>>>
>>> - a/clang/lib/CodeGen/CGDebugInfo.cpp
>>>
>>> +++ b/clang/lib/CodeGen/CGDebugInfo.cpp
>>> @@ -5149,9 +5149,9 @@ llvm::DebugLoc CGDebugInfo::SourceLocToDebugLoc(SourceLocation Loc) {
>>>  }
>>>
>>> llvm::DINode::DIFlags CGDebugInfo::getCallSiteRelatedAttrs() const {
>>>
>>> - // Call site-related attributes are only useful in optimized programs, and
>>> - // when there's a possibility of debugging backtraces.
>>> - if (!CGM.getLangOpts().Optimize || DebugKind == codegenoptions::NoDebugInfo ||
>>>
>>> +  // Call site-related attributes are useful when there's a possibility of
>>> +  // debugging backtraces.
>>> +  if (DebugKind == codegenoptions::NoDebugInfo ||
>>>
>>>     DebugKind == codegenoptions::LocTrackingOnly)
>>>   return llvm::DINode::FlagZero;
>>>
>>>   `````````````````````````````````````
>>>
>>> With this patch Clang/Clang++ turn on LLVM IR flag "DIFlagAllCallsDescribed", in absence of this LLVM does not generate call-site.
>>>
>>> With the above patch applied below is the comparison of sizes of shared libraries.
>>>
>>> Name of shared library - Size (without callsite) - Size (with callsites) - % incresase in size
>>>
>>>   ```````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````````
>>>
>>> PipSqueak.so 73192 75048 2%
>>> SecondLib.so 73192 75048 2%
>>> TestPlugin.so 1694024 1700704 0%
>>> libLLVMDlltoolDriver.so 336568 347872 3%
>>> libLLVMDebugInfoPDB.so 14463832 15360784 6%
>>> libLLVMOrcError.so 108880 111184 2%
>>> libLLVMTarget.so 2645104 2677448 1%
>>> libLLVMFrontendOpenMP.so 2354728 2505232 6%
>>> libLLVMProfileData.so 7901688 8373632 5%
>>> libLLVMOrcJIT.so 28838432 30490640 5%
>>> libLLVMRemarks.so 3311680 3551856 7%
>>> libgtest.so 2374120 2523112 6%
>>> libLLVMDemangle.so 1350616 1490832 10%
>>> libLLVMAsmParser.so 6961216 7366040 5%
>>> LLVMHello.so 394624 396120 0%
>>> libLLVMGlobalISel.so 18886648 19698008 4%
>>> libLLVMDebugInfoMSF.so 1288376 1365040 5%
>>> libLLVMCoverage.so 4225224 4502104 6%
>>> libLLVMFuzzMutate.so 3859968 3973808 2%
>>> libRemarks.so 6696 6696 0%
>>> libLLVMDebugInfoDWARF.so 11914848 12750784 7%
>>> libLLVMMCParser.so 6419464 6873000 7%
>>> libLLVMTableGen.so 4855536 5180760 6%
>>> libLLVMDWARFLinker.so 5407528 5628576 4%
>>> Bye.so 1858848 1872672 0%
>>> libLLVMMCJIT.so 1470952 1526544 3%
>>> libLLVMMC.so 16931504 17741376 4%
>>> libLLVMipo.so 43554712 46019392 5%
>>> libLLVMLineEditor.so 208216 216360 3%
>>> libbenchmark_main.so 18904 19408 2%
>>> libbenchmark.so 3308304 3507632 6%
>>> libLTO.so 2240720 2277408 1%
>>> libLLVMInterpreter.so 2614696 2749128 5%
>>> libLLVMTransformUtils.so 47925248 50476512 5%
>>> libLLVMX86Desc.so 8047928 8213152 2%
>>> libLLVMCoroutines.so 6478080 6766880 4%
>>> libLLVMJITLink.so 5590936 6066736 8%
>>> libLLVMVectorize.so 19557808 20665544 5%
>>> libLLVMX86Disassembler.so 2820056 2849376 1%
>>> libLLVMBitReader.so 8282648 8823240 6%
>>> libLLVMMCA.so 3242016 3405624 5%
>>> libLLVMBitWriter.so 6544032 6867976 4%
>>> libLLVMMIRParser.so 5739688 5980104 4%
>>> libLLVMLTO.so 13272272 13786192 3%
>>> libLLVMCore.so 46109224 48840008 5%
>>> libLLVMBitstreamReader.so 561896 600624 6%
>>> libLLVMObjectYAML.so 23110160 24648160 6%
>>> libLLVMSupport.so 20349728 21953112 7%
>>> libLLVMIRReader.so 1215672 1237960 1%
>>> libLLVMX86Info.so 76488 76624 0%
>>> libLLVMSelectionDAG.so 34358968 36876128 7%
>>> libLLVMExecutionEngine.so 2962160 3073224 3%
>>> libLLVMSymbolize.so 1980760 2089728 5%
>>> libLLVMPasses.so 18574960 19459712 4%
>>> libLLVMOption.so 869784 920976 5%
>>> libLLVMObject.so 15138656 16383168 8%
>>> libLLVMTextAPI.so 3191272 3384560 6%
>>> libLLVMX86CodeGen.so 55202744 58068088 5%
>>> libLLVMAggressiveInstCombine.so 2354936 2419048 2%
>>> libLLVMExtensions.so 23904 23904 0%
>>> libLLVMWindowsManifest.so 351608 370168 5%
>>> libLLVMObjCARCOpts.so 4964488 5150120 3%
>>> libLLVMBinaryFormat.so 1325752 1431448 7%
>>> libLLVMDebugInfoGSYM.so 2909272 3094616 6%
>>> libLLVMTestingSupport.so 623608 656640 5%
>>> libgtest_main.so 42856 43608 1%
>>> libLLVMLinker.so 3070264 3210248 4%
>>> libLLVMCFGuard.so 1082488 1096104 1%
>>> libLLVMCodeGen.so 139859816 146452160 4%
>>> libLLVMDebugInfoCodeView.so 7742896 8497704 9%
>>> libLLVMX86AsmParser.so 2590816 2714008 4%
>>> libLLVMRuntimeDyld.so 6592016 7094048 7%
>>> libLLVMInstCombine.so 18194728 19705184 8%
>>> BugpointPasses.so 780112 788304 1%
>>> libLLVMScalarOpts.so 73110536 77534912 6%
>>> libLLVMXRay.so 3993696 4251280 6%
>>> libLLVMMCDisassembler.so 1292912 1313536 1%
>>> libLLVMFrontendOpenACC.so 96008 102656 6%
>>> libLLVMInstrumentation.so 21038808 22351384 6%
>>> libLLVMLibDriver.so 1595936 1638704 2%
>>> libLLVMAsmPrinter.so 25818000 26842816 3%
>>> libLLVMAnalysis.so 79615056 83995856 5%
>>>
>>>   `````````````````````````````````````````````````````````````````````
>>>
>>> sum 79615057 83995857 5%
>>>
>>> So the conclusion is,
>>>
>>> - We have a Flag which can help us if we want to enable callsite generation selectively (only for Flang)
>>
>> It doesn't seem to me, so far, like this is a place where Flang and Clang should diverge - they're both doing the same sort of thing for the same reasons/likely with the same sort of tradeoffs of location accuracy V size cost.
>>
>>> - If we are fine with 5% increase in size, we can enable call-site generation by default.
>>
>> I'd actually be somewhat more worried about object size, rather than/in addition to executable size, due to the increase in relocations (especially with DWARFv5 (& especially with the `-mllvm -minimize-addr-in-v5=Ranges` which further reduces debug_addr, but can't handle labels and call sites (unlike `-minimize-addr-in-v5=Expressions` or `-minimize-addr-in-v5=Form`)), which does a lot to reduce the number of relocations by using debug_addr and address sharing on debug_rnglists/loclists/etc) which can't be compressed, etc.
>
> Below is the comparison of total size of object files (*.o) for same build (default (O0) + -g).
>
> sum 3535044960 3680496672 4%
>
> I shall come back with numbers for options O0 -gdwarf-5 -mllvm -minimize-addr-in-v5=Ranges .

Below are the results.

PipSqueak.so 74376 75048 0%
SecondLib.so 74376 75048 0%
TestPlugin.so 1699776 1700704 0%
libLLVMDlltoolDriver.so 338224 347872 2%
libLLVMDebugInfoPDB.so 14276632 15360784 7%
libLLVMOrcError.so 111168 111184 0%
libLLVMTarget.so 2670776 2677448 0%
libLLVMFrontendOpenMP.so 2351736 2505232 6%
libLLVMProfileData.so 7782224 8373632 7%
libLLVMOrcJIT.so 28408664 30490640 7%
libLLVMRemarks.so 3253472 3551856 9%
libgtest.so 2338592 2523112 7%
libLLVMDemangle.so 1339416 1490832 11%
libLLVMAsmParser.so 6885320 7366040 6%
LLVMHello.so 403544 396120 -1%
libLLVMGlobalISel.so 18813752 19698008 4%
libLLVMDebugInfoMSF.so 1272328 1365040 7%
libLLVMCoverage.so 4167336 4502104 8%
libLLVMFuzzMutate.so 3860488 3973808 2%
libRemarks.so 7216 6696 -7%
libLLVMDebugInfoDWARF.so 11690048 12750784 9%
libLLVMMCParser.so 6378536 6873000 7%
libLLVMTableGen.so 4790168 5180760 8%
libLLVMDWARFLinker.so 5352344 5628576 5%
Bye.so 1862216 1872672 0%
libLLVMMCJIT.so 1461496 1526544 4%
libLLVMMC.so 16848920 17741376 5%
libLLVMipo.so 42955408 46019392 7%
libLLVMLineEditor.so 209704 216360 3%
libbenchmark_main.so 19880 19408 -2%
libbenchmark.so 3270440 3507632 7%
libLTO.so 2233464 2277408 1%
libLLVMInterpreter.so 2611408 2749128 5%
libLLVMTransformUtils.so 47500376 50476512 6%
libLLVMX86Desc.so 8234536 8213152 0%
libLLVMCoroutines.so 6424104 6766880 5%
libLLVMJITLink.so 5481928 6066736 10%
libLLVMVectorize.so 19318304 20665544 6%
libLLVMX86Disassembler.so 2859328 2849376 0%
libLLVMBitReader.so 8152552 8823240 8%
libLLVMMCA.so 3218504 3405624 5%
libLLVMBitWriter.so 6472160 6867976 6%
libLLVMMIRParser.so 5702232 5980104 4%
libLLVMLTO.so 13115600 13786192 5%
libLLVMCore.so 45551816 48840008 7%
libLLVMBitstreamReader.so 555824 600624 8%
libLLVMObjectYAML.so 22643568 24648160 8%
libLLVMSupport.so 20110440 21953112 9%
libLLVMIRReader.so 1217592 1237960 1%
libLLVMX86Info.so 80080 76624 -4%
libLLVMSelectionDAG.so 34103008 36876128 8%
libLLVMExecutionEngine.so 2956728 3073224 3%
libLLVMSymbolize.so 1959536 2089728 6%
libLLVMPasses.so 18367064 19459712 5%
libLLVMOption.so 866616 920976 6%
libLLVMObject.so 14939568 16383168 9%
libLLVMTextAPI.so 3158584 3384560 7%
libLLVMX86CodeGen.so 56063440 58068088 3%
libLLVMAggressiveInstCombine.so 2353832 2419048 2%
libLLVMExtensions.so 25280 23904 -5%
libLLVMWindowsManifest.so 350240 370168 5%
libLLVMObjCARCOpts.so 4946928 5150120 4%
libLLVMBinaryFormat.so 1323744 1431448 8%
libLLVMDebugInfoGSYM.so 2862936 3094616 8%
libLLVMTestingSupport.so 621208 656640 5%
libgtest_main.so 45168 43608 -3%
libLLVMLinker.so 3038736 3210248 5%
libLLVMCFGuard.so 1090016 1096104 0%
libLLVMCodeGen.so 138864176 146452160 5%
libLLVMDebugInfoCodeView.so 7612904 8497704 11%
libLLVMX86AsmParser.so 2620696 2714008 3%
libLLVMRuntimeDyld.so 6503104 7094048 9%
libLLVMInstCombine.so 17916992 19705184 9%
BugpointPasses.so 785784 788304 0%
libLLVMScalarOpts.so 72114960 77534912 7%
libLLVMXRay.so 3939008 4251280 7%
libLLVMMCDisassembler.so 1299632 1313536 1%
libLLVMFrontendOpenACC.so 97360 102656 5%
libLLVMInstrumentation.so 20774464 22351384 7%
libLLVMLibDriver.so 1589136 1638704 3%
libLLVMAsmPrinter.so 25641320 26842816 4%
libLLVMAnalysis.so 78794136 83995856 6%
sum 944078696 1005246112 6%


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D99160/new/

https://reviews.llvm.org/D99160



More information about the cfe-commits mailing list