<div dir="ltr">Hi Balaram,<div><br></div><div>I didn't do anything special about libcxx. All I did was "configure + make".</div><div>Maybe someone else can help you with this error.</div><div><br></div><div>I will try to very the patch today, it is going to lock up my computer for 2 hours.</div><div>If I can't, I will try it after the Thanksgiving break.</div><div><br></div><div>Manman<br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Nov 21, 2014 at 11:58 AM, Balaram Makam <span dir="ltr"><<a href="mailto:bmakam@codeaurora.org" target="_blank">bmakam@codeaurora.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Manman,<br>
<br>
Could you please point me out how to build libcxx and libcxxabi on linux to build lto selfhost?<br>
<br>
I keep getting the following error when trying to build libcxx and libcxxabi:<br>
<br>
CMake Error at cmake/modules/HandleLLVMOptions.cmake:38 (message):<br>
Host Clang must be able to find libstdc++4.7 or newer!<br>
<br>
Thanks,<br>
Balaram<br>
<span class=""><br>
-----Original Message-----<br>
From: <a href="mailto:llvm-commits-bounces@cs.uiuc.edu">llvm-commits-bounces@cs.uiuc.edu</a> [mailto:<a href="mailto:llvm-commits-bounces@cs.uiuc.edu">llvm-commits-bounces@cs.uiuc.edu</a>] On Behalf Of Hao Liu<br>
Sent: Friday, November 21, 2014 2:25 AM<br>
To: Manman Ren<br>
Cc: <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
Subject: Re: [llvm] r222039 - Reapply r221924: "[GVN] Perform Scalar PRE on gep indices that feed loads before<br>
<br>
</span><div><div class="h5">Hi Manman,<br>
<br>
Unfortunately, I still have such build failures. It seems not related to the source. Something must be wrong with my process.<br>
<br>
As you can easily reproduce it, maybe you can help Balaram to verify his new patch,<br>
<br>
Thanks,<br>
-Hao<br>
<br>
2014-11-21 0:42 GMT+08:00 Manman Ren <<a href="mailto:manman.ren@gmail.com">manman.ren@gmail.com</a>>:<br>
> Hi Hao,<br>
><br>
> Sorry about missing the step of acquiring source. What I did was<br>
> trying to replicate what the build bot is doing.<br>
><br>
> Yes, libcxx is part of the source acquiring. See<br>
><br>
> <a href="http://lab.llvm.org:8080/green/job/clang-Rlto_master_acquire/514/conso" target="_blank">http://lab.llvm.org:8080/green/job/clang-Rlto_master_acquire/514/conso</a><br>
> le<br>
><br>
> for a list of projects.<br>
><br>
> If you still have problem reproducing, let me know.<br>
><br>
> Manman<br>
><br>
> On Thu, Nov 20, 2014 at 2:49 AM, Hao Liu <<a href="mailto:Hao.Liu@arm.com">Hao.Liu@arm.com</a>> wrote:<br>
>><br>
>> Hi Manman,<br>
>><br>
>><br>
>><br>
>> I tried your steps on the same revision as yours: r222239@trunk. But<br>
>> failed to reproduce it. Is it related to libc++ version? If so, what<br>
>> revision is used in your test.<br>
>><br>
>><br>
>><br>
>> I tired libc++ reversion of r222425 and r222425 and followed your steps.<br>
>><br>
>> But it had build failures on 1st step:<br>
>><br>
>> llvm[2]: Linking Release+Asserts executable FileCheck (without<br>
>> symbols)<br>
>><br>
>><br>
>> build/utils/FileCheck/Release+Asserts/FileCheck.o:llvm/utils/FileChec<br>
>> k/FileCheck.cpp:function llvm::cl::list<std::__1::basic_string<char,<br>
>> std::__1::char_traits<char>, std::__1::allocator<char> >, bool,<br>
>> llvm::cl::parser<std::__1::basic_string<char,<br>
>> std::__1::char_traits<char>, std::__1::allocator<char> > ><br>
>> >::~list(): error: undefined reference to 'operator delete(void*)'<br>
>><br>
>> …<br>
>><br>
>><br>
>><br>
>> Also I tried to configure without “--enable-libcpp”, both steps were<br>
>> finished. But can not reproduce the failures on fixxfti.c and fixunsxfti.c.<br>
>> So I think libc++ is necessary.<br>
>><br>
>><br>
>><br>
>> Thanks,<br>
>><br>
>> -Hao<br>
>><br>
>><br>
>><br>
>> From: <a href="mailto:llvm-commits-bounces@cs.uiuc.edu">llvm-commits-bounces@cs.uiuc.edu</a><br>
>> [mailto:<a href="mailto:llvm-commits-bounces@cs.uiuc.edu">llvm-commits-bounces@cs.uiuc.edu</a>] On Behalf Of Manman Ren<br>
>> Sent: Thursday, November 20, 2014 8:29 AM<br>
>> To: Balaram Makam<br>
>> Cc: <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>><br>
>><br>
>> Subject: Re: [llvm] r222039 - Reapply r221924: "[GVN] Perform Scalar<br>
>> PRE on gep indices that feed loads before<br>
>><br>
>><br>
>><br>
>><br>
>><br>
>> On Nov 19, 2014, at 3:46 PM, Balaram Makam <<a href="mailto:bmakam@codeaurora.org">bmakam@codeaurora.org</a>> wrote:<br>
>><br>
>><br>
>><br>
>> Hi Manman,<br>
>><br>
>> I have a patch with a potential fix to the problem, but I cannot<br>
>> reproduce the original failure with the preprocessed source files you provided.<br>
>><br>
>><br>
>><br>
>> Hi Balaram,<br>
>><br>
>><br>
>><br>
>> You need to build a lto selfhost to reproduce.<br>
>><br>
>><br>
>><br>
>> 1> make a release+assert build with your patch (install the compiler<br>
</div></div>>> 1> to a<br>
<div><div class="h5">>> directory clang-install for example).<br>
>><br>
>><br>
>><br>
>> create directory clang-build and clang-install under llvm<br>
>><br>
>> under clang-build, do:<br>
>> ./llvm/configure --enable-assertions --enable-optimized<br>
>> --disable-bindings --enable-targets=x86,x86_64 --enable-libcpp<br>
>> --prefix=Users/manmanren/llvm/clang-install<br>
>><br>
>> make -j 12 VERBOSE=1<br>
>> CLANG_REPOSITORY_STRING=clang-stage1-configure-RA_build SVN_REVISION=222059<br>
>> make install-clang -j 4<br>
>><br>
>><br>
>><br>
>> 2> use clang-install binary and library to build the same repo with lto.<br>
>><br>
>><br>
>> create directory selfhost, under selfhost, do:<br>
>><br>
>> ../llvm/configure --disable-assertions --with-extra-options="-flto<br>
>> -gline-tables-only” --enable-optimized --disable-bindings<br>
>> --enable-targets=x86,x86_64 --enable-libcpp<br>
>> CC=/Users/manmanren/llvm/clang-install/bin/clang<br>
>> CXX=/Users/manmanren/gmail/clang-install/llvm/clang++<br>
>><br>
>><br>
>><br>
>> make -j 4 VERBOSE=1<br>
>> CLANG_REPOSITORY_STRING=clang-Rlto_master_build<br>
>> SVN_REVISION=222059<br>
>> DYLD_LIBRARY_PATH=/Users/manmanren/llvm_gmail/clang-install/lib/<br>
>><br>
>><br>
>><br>
>> The whole process takes a few hours to run.You can also commit the<br>
>> patch if you are confident enough and then watch the bot (phase 2 lto).<br>
>><br>
>><br>
>><br>
>> Manman<br>
>><br>
>><br>
>><br>
>> Could you please test this patch (attached) and let me know if it works?<br>
>><br>
>> Thanks,<br>
>> Balaram<br>
>><br>
>><br>
>> -----Original Message-----<br>
>> From: Chad Rosier [mailto:<a href="mailto:mcrosier@codeaurora.org">mcrosier@codeaurora.org</a>]<br>
>> Sent: Tuesday, November 18, 2014 9:09 PM<br>
>> To: Manman Ren<br>
>> Cc: <a href="mailto:mcrosier@codeaurora.org">mcrosier@codeaurora.org</a>; <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a>; Balaram Makam;<br>
>> James Molloy<br>
>> Subject: Re: [llvm] r222039 - Reapply r221924: "[GVN] Perform Scalar<br>
>> PRE on gep indices that feed loads before<br>
>><br>
>> Thanks, Manman. Balaram or another one of our team members will<br>
>> investigate.<br>
>><br>
>> Chad<br>
>><br>
>><br>
>><br>
>><br>
>> Hi Chad,<br>
>><br>
>> I reverted the commit in r222287.<br>
>><br>
>> The lto self host of r222239 without this commit succeeded.<br>
>> And I have finished the lto self host of r222239 including this<br>
>> commit, it failed. If you need any more files, let me know.<br>
>><br>
>> Thanks,<br>
>> Manman<br>
>><br>
>> clang(59539,0x7fff7c035300) malloc: *** error for object 0x7fa438648450:<br>
>> pointer being freed was not allocated<br>
>> *** set a breakpoint in malloc_error_break to debug<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixxfti.c<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/<br>
</div></div>>> f loatdidf.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/fixxfti.o<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixxfti.c<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/<br>
</span>>> f loatdisf.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins__x86_64/floatdisf.o<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/<br>
</span>>> f loatdisf.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</span>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins__x86_64/floatdidf.o<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/<br>
<div><div class="h5">>> f<br>
>> loatdidf.c<br>
>> clang(59547,0x7fff7c035300) malloc: *** error for object 0x7f9410f1c0d0:<br>
>> pointer being freed was not allocated<br>
>> *** set a breakpoint in malloc_error_break to debug<br>
>> 0 clang 0x000000010c15d259<br>
>> llvm::sys::PrintStackTrace(__sFILE*) + 57<br>
>> 1 clang 0x000000010c15ddab SignalHandler(int) + 347<br>
>> 2 libsystem_platform.dylib 0x00007fff980faf1a _sigtramp + 26<br>
>> 3 libsystem_platform.dylib 0xffffffffffffffff _sigtramp + 1743802623<br>
>> 4 libsystem_c.dylib 0x00007fff982ccb73 abort + 129<br>
>> 5 libsystem_malloc.dylib 0x00007fff98e7b937 nano_size + 0<br>
>> 6 clang 0x000000010b7b77b9 (anonymous<br>
>> namespace)::X86TTI::getIntImmCost(llvm::APInt const&, llvm::Type*)<br>
>> const +<br>
>> 601<br>
>> 7 clang 0x000000010b7b6cd8 non-virtual thunk to<br>
>> (anonymous namespace)::X86TTI::getIntImmCost(unsigned int, unsigned<br>
>> int, llvm::APInt const&, llvm::Type*) const + 216<br>
>> 8 clang 0x000000010bb73333 (anonymous<br>
>> namespace)::ConstantHoisting::collectConstantCandidates(llvm::DenseMa<br>
>> p<br>
>> <llvm::ConstantInt*, unsigned int,<br>
>> llvm::DenseMapInfo<llvm::ConstantInt*> >&, llvm::Instruction*,<br>
>> unsigned int, llvm::ConstantInt*) + 211<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>><br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/floatdixf.c<br>
>> 9 clang 0x000000010bb6ecf4 (anonymous<br>
>> namespace)::ConstantHoisting::runOnFunction(llvm::Function&) + 516<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/floatti<br>
</div></div>>> d f.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</span>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins__x86_64/floatdixf.o<br>
<div><div class="h5">>><br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/x86_64/floatdixf.c<br>
>> 10 clang 0x000000010c0e3a25<br>
>> llvm::FPPassManager::runOnFunction(llvm::Function&) + 549<br>
>> 11 clang 0x000000010c0e3c9b<br>
>> llvm::FPPassManager::runOnModule(llvm::Module&) + 43<br>
>> 12 clang 0x000000010c0e412a<br>
>> llvm::legacy::PassManagerImpl::run(llvm::Module&) + 922<br>
>> 13 clang 0x000000010a445430<br>
>> clang::EmitBackendOutput(clang::DiagnosticsEngine&,<br>
>> clang::CodeGenOptions const&, clang::TargetOptions const&,<br>
>> clang::LangOptions const&, llvm::StringRef, llvm::Module*,<br>
>> clang::BackendAction, llvm::raw_ostream*)<br>
>> + 12208<br>
>> 0 clang 0x00000001059d5259<br>
>> llvm::sys::PrintStackTrace(__sFILE*) + 57<br>
>> 14 clang 0x000000010a5cda44<br>
>> clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) +<br>
>> 516 /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</div></div>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/floattidf.o<br>
<div><div class="h5">>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/floattidf.c<br>
>> 1 clang 0x00000001059d5dab SignalHandler(int) + 347<br>
>> 2 libsystem_platform.dylib 0x00007fff980faf1a _sigtramp + 26<br>
>> 3 libsystem_platform.dylib 0xffffffffffffffff _sigtramp + 1743802623<br>
>> 15 clang 0x000000010a6864c5 4 libsystem_c.dylib<br>
>> 0x00007fff982ccb73 clang::ParseAST(clang::Sema&, bool, bool)abort +<br>
>> 661 +<br>
>> 129<br>
>><br>
>> 5 libsystem_malloc.dylib 0x00007fff98e7b937 nano_size + 0<br>
>> 16 clang 0x000000010a1f2b7e<br>
>> clang::FrontendAction::Execute() + 62<br>
>> 6 clang 0x000000010502f7b9 (anonymous<br>
>> namespace)::X86TTI::getIntImmCost(llvm::APInt const&, llvm::Type*)<br>
>> const +<br>
>> 601<br>
>> 17 clang 0x000000010a1ba8d3<br>
>> clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) + 915<br>
>> 7 clang 0x000000010502ecd8 non-virtual thunk to<br>
>> (anonymous namespace)::X86TTI::getIntImmCost(unsigned int, unsigned<br>
>> int, llvm::APInt const&, llvm::Type*) const + 216<br>
>> 18 clang 0x000000010a16a3fa<br>
>> clang::ExecuteCompilerInvocation(clang::CompilerInstance*) + 2890<br>
>> 8 clang 0x00000001053eb333 (anonymous<br>
>> namespace)::ConstantHoisting::collectConstantCandidates(llvm::DenseMa<br>
>> p<br>
>> <llvm::ConstantInt*, unsigned int,<br>
>> llvm::DenseMapInfo<llvm::ConstantInt*> >&, llvm::Instruction*,<br>
>> unsigned int, llvm::ConstantInt*) + 211<br>
>> 19 clang 0x000000010a16000c<br>
>> cc1_main(llvm::ArrayRef<char const*>, char const*, void*) + 1836<br>
>> 9 clang 0x00000001053e6cf4 (anonymous<br>
>> namespace)::ConstantHoisting::runOnFunction(llvm::Function&) + 516<br>
>> 20 clang 0x000000010a168b21 main + 13745<br>
>> 21 libdyld.dylib 0x00007fff8b4215c9 start + 1<br>
>> 22 libdyld.dylib 0x0000000000000045 start + 1958603389<br>
>> 10 clang 0x000000010595ba25<br>
>> llvm::FPPassManager::runOnFunction(llvm::Function&) + 549 Stack dump:<br>
>> 0. Program arguments:<br>
>> /Users/manmanren/gmail/selfhost/Release/bin/clang<br>
>> -cc1 -triple x86_64-apple-macosx10.4.0 -emit-obj -disable-free<br>
>> -disable-llvm-verifier -main-file-name fixunsxfti.c -static-define<br>
>> -mrelocation-model static -mthread-model posix -masm-verbose<br>
>> -fforbid-guard-variables -target-cpu core2 -disable-red-zone<br>
>> -no-implicit-float -target-linker-version 242 -dwarf-column-info<br>
>> -coverage-file<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/fixunsxfti.o<br>
>> -resource-dir<br>
>> /Users/manmanren/gmail/selfhost/Release/bin/../lib/clang/3.6.0<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -D VISIBILITY_HIDDEN -D KERNEL_USE -I<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib -I<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/include -O3 -Wall<br>
>> -Werror -fdebug-compilation-dir<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt<br>
>> -ferror-limit 19 -fmessage-length 164 -fvisibility hidden<br>
>> -ffreestanding -mstackrealign -fno-builtin -fblocks<br>
>> -fblocks-runtime-optional -fno-rtti -fno-use-cxa-atexit<br>
>> -fobjc-runtime=macosx-10.4.0 -fobjc-dispatch-method=non-legacy<br>
>> -fencode-extended-block-signature<br>
>> -fmax-type-align=16 -fno-common -fdiagnostics-show-option<br>
>> -fcolor-diagnostics -vectorize-loops -vectorize-slp -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</div></div>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/fixunsxfti.o<br>
<div><div class="h5">>> -x c<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixunsxfti.c<br>
>> 1. <eof> parser at end of file<br>
>> 2. Code generation<br>
>> 3. Running pass 'Function Pass Manager' on module<br>
>><br>
>> '/Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixunsxfti.c'.<br>
>> 4. Running pass 'Constant Hoisting' on function '@__fixunsxfti'<br>
>> 11 clang 0x000000010595bc9b<br>
>> llvm::FPPassManager::runOnModule(llvm::Module&) + 43<br>
>> 12 clang 0x000000010595c12a<br>
>> llvm::legacy::PassManagerImpl::run(llvm::Module&) + 922<br>
>> 13 clang 0x0000000103cbd430<br>
>> clang::EmitBackendOutput(clang::DiagnosticsEngine&,<br>
>> clang::CodeGenOptions const&, clang::TargetOptions const&,<br>
>> clang::LangOptions const&, llvm::StringRef, llvm::Module*,<br>
>> clang::BackendAction, llvm::raw_ostream*)<br>
>> + 12208<br>
>> 14 clang 0x0000000103e45a44<br>
>> clang::BackendConsumer::HandleTranslationUnit(clang::ASTContext&) + 516<br>
>> 15 clang 0x0000000103efe4c5<br>
>> clang::ParseAST(clang::Sema&, bool, bool) + 661<br>
>> 16 clang 0x0000000103a6ab7e<br>
>> clang::FrontendAction::Execute() + 62<br>
>> 17 clang 0x0000000103a328d3<br>
>> clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) + 915<br>
>> 18 clang 0x00000001039e23fa<br>
>> clang::ExecuteCompilerInvocation(clang::CompilerInstance*) + 2890<br>
>> 19 clang 0x00000001039d800c<br>
>> cc1_main(llvm::ArrayRef<char const*>, char const*, void*) + 1836<br>
>> 20 clang 0x00000001039e0b21 main + 13745<br>
>> 21 libdyld.dylib 0x00007fff8b4215c9 start + 1<br>
>> 22 libdyld.dylib 0x0000000000000045 start + 1958603389<br>
>> Stack dump:<br>
>> 0. Program arguments:<br>
>> /Users/manmanren/gmail/selfhost/Release/bin/clang<br>
>> -cc1 -triple x86_64-apple-macosx10.4.0 -emit-obj -disable-free<br>
>> -disable-llvm-verifier -main-file-name fixxfti.c -static-define<br>
>> -mrelocation-model static -mthread-model posix -masm-verbose<br>
>> -fforbid-guard-variables -target-cpu core2 -disable-red-zone<br>
>> -no-implicit-float -target-linker-version 242 -dwarf-column-info<br>
>> -coverage-file<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/fixxfti.o<br>
>> -resource-dir<br>
>> /Users/manmanren/gmail/selfhost/Release/bin/../lib/clang/3.6.0<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -D VISIBILITY_HIDDEN -D KERNEL_USE -I<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib -I<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/include -O3 -Wall<br>
>> -Werror -fdebug-compilation-dir<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt<br>
>> -ferror-limit 19 -fmessage-length 164 -fvisibility hidden<br>
>> -ffreestanding -mstackrealign -fno-builtin -fblocks<br>
>> -fblocks-runtime-optional -fno-rtti -fno-use-cxa-atexit<br>
>> -fobjc-runtime=macosx-10.4.0 -fobjc-dispatch-method=non-legacy<br>
>> -fencode-extended-block-signature<br>
>> -fmax-type-align=16 -fno-common -fdiagnostics-show-option<br>
>> -fcolor-diagnostics -vectorize-loops -vectorize-slp -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</div></div>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/fixxfti.o<br>
<span class="">>> -x c<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixxfti.c<br>
>> 1. <eof> parser at end of file<br>
>> 2. Code generation<br>
>> 3. Running pass 'Function Pass Manager' on module<br>
>> '/Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/fixxfti.c'.<br>
>> 4. Running pass 'Constant Hoisting' on function '@__fixxfti'<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/floatti<br>
</span>>> s f.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</span>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/floattisf.o<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/floatti<br>
<span class="">>> s<br>
>> f.c<br>
>><br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powidf2.<br>
>> c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/popcountti2.o<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/popcoun<br>
</span>>> t ti2.c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
<span class="">>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</span>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/powidf2.o<br>
<div><div class="h5">>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powidf2.<br>
>> c<br>
>> clang: error: unable to execute command: Abort trap: 6<br>
>> clang: error: clang frontend command failed due to signal (use -v to<br>
>> see<br>
>> invocation)<br>
>> clang version 3.6.0 (clang-Rlto_master_build 222059) (llvm/trunk<br>
>> 222239)<br>
>> Target: x86_64-apple-darwin14.0.0<br>
>> Thread model: posix<br>
>> clang: note: diagnostic msg: PLEASE submit a bug report to<br>
>> <a href="http://llvm.org/bugs/" target="_blank">http://llvm.org/bugs/</a> and include the crash backtrace, preprocessed<br>
>> source, and associated run script.<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powisf2.<br>
>> c /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</div></div>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/powisf2.o<br>
<span class="">>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powisf2.<br>
>> c<br>
>> clang: note: diagnostic msg:<br>
>> ********************<br>
>><br>
>> PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:<br>
>> Preprocessed source(s) and associated run script(s) are located at:<br>
>> clang: note: diagnostic msg:<br>
>> /var/folders/8d/mksp0c9x6yngd578zj_s72ym0000gn/T/fixxfti-868934.c<br>
>> clang: note: diagnostic msg:<br>
>> /var/folders/8d/mksp0c9x6yngd578zj_s72ym0000gn/T/fixxfti-868934.sh<br>
>> clang: note: diagnostic msg:<br>
>><br>
>> ********************<br>
>> COMPILE: clang_darwin/cc_kext/x86_64:<br>
>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powixf2.<br>
>> c<br>
>> make[5]: ***<br>
>> [/Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clan<br>
</span>>> g _darwin/cc_kext/x86_64/SubDir.lib__builtins/fixxfti.o]<br>
<span class="">>> Error 254<br>
>> make[5]: *** Waiting for unfinished jobs....<br>
>> /Users/manmanren/gmail/selfhost/Release/bin/clang -fPIC<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/lib<br>
>> -I/Users/manmanren/gmail/llvm/projects/compiler-rt/include -arch<br>
>> x86_64 -fvisibility=hidden -DVISIBILITY_HIDDEN -mkernel -DKERNEL_USE<br>
>> -Wall -Werror -O3 -fomit-frame-pointer -mmacosx-version-min=10.4<br>
>> -isysroot<br>
>> /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/<br>
>> D<br>
>> eveloper/SDKs/MacOSX10.10.sdk<br>
>> -c -o<br>
>> /Users/manmanren/gmail/selfhost/tools/clang/runtime/compiler-rt/clang<br>
</span>>> _ darwin/cc_kext/x86_64/SubDir.lib__builtins/powixf2.o<br>
<div><div class="h5">>> /Users/manmanren/gmail/llvm/projects/compiler-rt/lib/builtins/powixf2.<br>
>> c<br>
>> clang: error: unable to execute command: Abort trap: 6<br>
>> clang: error: clang frontend command failed due to signal (use -v to<br>
>> see<br>
>> invocation)<br>
>> clang version 3.6.0 (clang-Rlto_master_build 222059) (llvm/trunk<br>
>> 222239)<br>
>> Target: x86_64-apple-darwin14.0.0<br>
>> Thread model: posix<br>
>> clang: note: diagnostic msg: PLEASE submit a bug report to<br>
>> <a href="http://llvm.org/bugs/" target="_blank">http://llvm.org/bugs/</a> and include the crash backtrace, preprocessed<br>
>> source, and associated run script.<br>
>> clang: note: diagnostic msg:<br>
>> ********************<br>
>><br>
>> PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:<br>
>> Preprocessed source(s) and associated run script(s) are located at:<br>
>> clang: note: diagnostic msg:<br>
>> /var/folders/8d/mksp0c9x6yngd578zj_s72ym0000gn/T/fixunsxfti-33f776.c<br>
>> clang: note: diagnostic msg:<br>
>> /var/folders/8d/mksp0c9x6yngd578zj_s72ym0000gn/T/fixunsxfti-33f776.sh<br>
>> clang: note: diagnostic msg:<br>
>><br>
>> ********************<br>
>><br>
>><br>
>><br>
>> On Nov 18, 2014, at 2:53 PM, Chad Rosier <<a href="mailto:mcrosier@codeaurora.org">mcrosier@codeaurora.org</a>><br>
>> wrote:<br>
>><br>
>> Hi Manman,<br>
>> Feel free to revert the patch if you're convinced this is the problem.<br>
>> Please keep Balaram, James, and I updated on the progress. This is a<br>
>> performance critical fix, so we need to make sure it doesn't get lost<br>
>> in the mix.<br>
>><br>
>> Chad<br>
>><br>
>><br>
>> Hi Chad,<br>
>><br>
>> About the buildbot phase 2 failure, I now think this commit is the<br>
>> root cause.<br>
>><br>
>> What I did was:<br>
</div></div>>> 1> update the repo to r222239, then "svn merge -c -r222039 .†, made<br>
>> 1> a<br>
<div><div class="h5">>> release+assert build (installed to clang-install).<br>
>> 2> use clang-install binary and library to build the same repo with<br>
>> lto.<br>
>> make -j 4 VERBOSE=1<br>
>> CLANG_REPOSITORY_STRING=clang-Rlto_master_build<br>
>> SVN_REVISION=222059<br>
>> DYLD_LIBRARY_PATH=/Users/manmanren/llvm_gmail/clang-install/lib/<br>
>> 3> the build completed<br>
>> make[1]: Nothing to be done for `all'.<br>
>> make[1]: Nothing to be done for `all'.<br>
>> llvm[0]: ***** Completed Release Build<br>
>><br>
>> Is it okay for us to revert this commit and watch the bot? If the bot<br>
>> is still red, we can re-submit it.<br>
>><br>
>> If you want me to do more testing (maybe a lto self host with<br>
>> r22239), let me know,<br>
>><br>
>> Thanks,<br>
>> Manman (The build czar)<br>
>><br>
>><br>
>> On Nov 14, 2014, at 1:09 PM, Chad Rosier <<a href="mailto:mcrosier@codeaurora.org">mcrosier@codeaurora.org</a>><br>
>> wrote:<br>
>><br>
>> Author: mcrosier<br>
>> Date: Fri Nov 14 15:09:13 2014<br>
>> New Revision: 222039<br>
>><br>
>> URL: <a href="http://llvm.org/viewvc/llvm-project?rev=222039&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=222039&view=rev</a><br>
>> Log:<br>
>> Reapply r221924: "[GVN] Perform Scalar PRE on gep indices that feed<br>
>> loads before doing Load PRE"<br>
>><br>
>> This commit updates the failing test in<br>
>> Analysis/TypeBasedAliasAnalysis/gvn-nonlocal-type-mismatch.ll<br>
>><br>
>> The failing test is sensitive to the order in which we process loads.<br>
>> This<br>
>> version turns on the RPO traversal instead of the while DT traversal<br>
>> in GVN.<br>
>> The new test code is functionally same just the order of loads that<br>
>> are eliminated is swapped.<br>
>><br>
>> This new version also fixes an issue where GVN splits a critical edge<br>
>> and potentially invalidate the RPO/DT iterator.<br>
>><br>
>> Added:<br>
>> llvm/trunk/test/Transforms/GVN/pre-gep-load.ll<br>
>> Modified:<br>
>> llvm/trunk/lib/Transforms/Scalar/GVN.cpp<br>
>><br>
>> llvm/trunk/test/Analysis/TypeBasedAliasAnalysis/gvn-nonlocal-type-m<br>
>> ismatch.ll<br>
>><br>
>> Modified: llvm/trunk/lib/Transforms/Scalar/GVN.cpp<br>
>> URL:<br>
>> <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scala" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scala</a><br>
>> r/GVN.cpp?rev=222039&r1=222038&r2=222039&view=diff<br>
>> ===================================================================<br>
>> ===========<br>
>> --- llvm/trunk/lib/Transforms/Scalar/GVN.cpp (original)<br>
>> +++ llvm/trunk/lib/Transforms/Scalar/GVN.cpp Fri Nov 14 15:09:13<br>
>> +++ 2014<br>
>> @@ -20,6 +20,7 @@<br>
>> #include "llvm/ADT/DepthFirstIterator.h"<br>
>> #include "llvm/ADT/Hashing.h"<br>
>> #include "llvm/ADT/MapVector.h"<br>
>> +#include "llvm/ADT/PostOrderIterator.h"<br>
>> #include "llvm/ADT/SetVector.h"<br>
>> #include "llvm/ADT/SmallPtrSet.h"<br>
>> #include "llvm/ADT/Statistic.h"<br>
>> @@ -709,6 +710,7 @@ namespace {<br>
>> void dump(DenseMap<uint32_t, Value*> &d);<br>
>> bool iterateOnFunction(Function &F);<br>
>> bool performPRE(Function &F);<br>
>> + bool performScalarPRE(Instruction *I);<br>
>> Value *findLeader(const BasicBlock *BB, uint32_t num);<br>
>> void cleanupGlobalSets();<br>
>> void verifyRemoved(const Instruction *I) const; @@ -1729,6<br>
>> +1731,15 @@ bool GVN::processNonLocalLoad(LoadInst *<br>
>> return false;<br>
>> }<br>
>><br>
>> + // If this load follows a GEP, see if we can PRE the indices<br>
>> + before<br>
>> analyzing.<br>
>> + if (GetElementPtrInst *GEP =<br>
>> dyn_cast<GetElementPtrInst>(LI->getOperand(0))) {<br>
>> + for (GetElementPtrInst::op_iterator OI = GEP->idx_begin(),<br>
>> + OE = GEP->idx_end();<br>
>> + OI != OE; ++OI)<br>
>> + if (Instruction *I = dyn_cast<Instruction>(OI->get()))<br>
>> + performScalarPRE(I);<br>
>> + }<br>
>> +<br>
>> // Step 2: Analyze the availability of the load AvailValInBlkVect<br>
>> ValuesPerBlock; UnavailBlkVect UnavailableBlocks; @@ -2431,175<br>
>> +2442,182 @@ bool GVN::processBlock(BasicBlock *BB) { return<br>
>> ChangedFunction; }<br>
>><br>
>> -/// performPRE - Perform a purely local form of PRE that looks for<br>
>> diamond -/// control flow patterns and attempts to perform simple PRE<br>
>> at the join point.<br>
>> -bool GVN::performPRE(Function &F) {<br>
>> - bool Changed = false;<br>
>> +bool GVN::performScalarPRE(Instruction *CurInst) {<br>
>> SmallVector<std::pair<Value*, BasicBlock*>, 8> predMap;<br>
>> - for (BasicBlock *CurrentBlock : depth_first(&F.getEntryBlock())) {<br>
>> - // Nothing to PRE in the entry block.<br>
>> - if (CurrentBlock == &F.getEntryBlock()) continue;<br>
>><br>
>> - // Don't perform PRE on a landing pad.<br>
>> - if (CurrentBlock->isLandingPad()) continue;<br>
>> + if (isa<AllocaInst>(CurInst) || isa<TerminatorInst>(CurInst) ||<br>
>> + isa<PHINode>(CurInst) || CurInst->getType()->isVoidTy() ||<br>
>> + CurInst->mayReadFromMemory() ||<br>
>> + CurInst->mayHaveSideEffects()<br>
>> ||<br>
>> + isa<DbgInfoIntrinsic>(CurInst))<br>
>> + return false;<br>
>><br>
>> - for (BasicBlock::iterator BI = CurrentBlock->begin(),<br>
>> - BE = CurrentBlock->end(); BI != BE; ) {<br>
>> - Instruction *CurInst = BI++;<br>
</div></div>>> + // Don't do PRE on compares. The PHI would prevent CodeGenPrepare<br>
<span class="">>> from<br>
>> + // sinking the compare again, and it would force the code<br>
>> + generator<br>
>> to<br>
</span>>> + // move the i1 from processor flags or predicate registers into a<br>
<div><div class="h5">>> general<br>
>> + // purpose register.<br>
>> + if (isa<CmpInst>(CurInst))<br>
>> + return false;<br>
>><br>
>> - if (isa<AllocaInst>(CurInst) ||<br>
>> - isa<TerminatorInst>(CurInst) || isa<PHINode>(CurInst) ||<br>
>> - CurInst->getType()->isVoidTy() ||<br>
>> - CurInst->mayReadFromMemory() ||<br>
>> CurInst->mayHaveSideEffects()<br>
>> ||<br>
>> - isa<DbgInfoIntrinsic>(CurInst))<br>
>> - continue;<br>
>> + // We don't currently value number ANY inline asm calls.<br>
>> + if (CallInst *CallI = dyn_cast<CallInst>(CurInst))<br>
>> + if (CallI->isInlineAsm())<br>
>> + return false;<br>
>><br>
>> - // Don't do PRE on compares. The PHI would prevent<br>
>> CodeGenPrepare<br>
>> from<br>
>> - // sinking the compare again, and it would force the code<br>
>> generator to<br>
>> - // move the i1 from processor flags or predicate registers into<br>
>> a<br>
>> general<br>
>> - // purpose register.<br>
>> - if (isa<CmpInst>(CurInst))<br>
>> - continue;<br>
>> + uint32_t ValNo = VN.lookup(CurInst);<br>
>><br>
>> - // We don't currently value number ANY inline asm calls.<br>
>> - if (CallInst *CallI = dyn_cast<CallInst>(CurInst))<br>
>> - if (CallI->isInlineAsm())<br>
>> - continue;<br>
>> + // Look for the predecessors for PRE opportunities. We're //<br>
>> + only trying to solve the basic diamond case, where // a value is<br>
>> + computed in the successor and one predecessor, // but not the<br>
>> + other. We also explicitly disallow cases // where the successor<br>
</div></div>>> + is its own predecessor, because they're // more complicated to get<br>
<span class="">>> + right.<br>
>> + unsigned NumWith = 0;<br>
>> + unsigned NumWithout = 0;<br>
>> + BasicBlock *PREPred = nullptr;<br>
</span>>> + BasicBlock *CurrentBlock = CurInst->getParent(); predMap.clear();<br>
<div><div class="h5">>> +<br>
>> + for (pred_iterator PI = pred_begin(CurrentBlock), PE =<br>
>> pred_end(CurrentBlock);<br>
>> + PI != PE; ++PI) {<br>
>> + BasicBlock *P = *PI;<br>
>> + // We're not interested in PRE where the block is its<br>
>> + // own predecessor, or in blocks with predecessors<br>
>> + // that are not reachable.<br>
>> + if (P == CurrentBlock) {<br>
>> + NumWithout = 2;<br>
>> + break;<br>
>> + } else if (!DT->isReachableFromEntry(P)) {<br>
>> + NumWithout = 2;<br>
>> + break;<br>
>> + }<br>
>><br>
>> - uint32_t ValNo = VN.lookup(CurInst);<br>
>> + Value *predV = findLeader(P, ValNo);<br>
>> + if (!predV) {<br>
>> + predMap.push_back(std::make_pair(static_cast<Value<br>
>> + *>(nullptr),<br>
>> P));<br>
>> + PREPred = P;<br>
>> + ++NumWithout;<br>
>> + } else if (predV == CurInst) {<br>
>> + /* CurInst dominates this predecessor. */<br>
>> + NumWithout = 2;<br>
>> + break;<br>
>> + } else {<br>
>> + predMap.push_back(std::make_pair(predV, P));<br>
>> + ++NumWith;<br>
>> + }<br>
>> + }<br>
>><br>
>> - // Look for the predecessors for PRE opportunities. We're<br>
>> - // only trying to solve the basic diamond case, where<br>
>> - // a value is computed in the successor and one predecessor,<br>
>> - // but not the other. We also explicitly disallow cases<br>
>> - // where the successor is its own predecessor, because they're<br>
>> - // more complicated to get right.<br>
>> - unsigned NumWith = 0;<br>
>> - unsigned NumWithout = 0;<br>
>> - BasicBlock *PREPred = nullptr;<br>
>> - predMap.clear();<br>
>> -<br>
>> - for (pred_iterator PI = pred_begin(CurrentBlock),<br>
>> - PE = pred_end(CurrentBlock); PI != PE; ++PI) {<br>
>> - BasicBlock *P = *PI;<br>
>> - // We're not interested in PRE where the block is its<br>
>> - // own predecessor, or in blocks with predecessors<br>
>> - // that are not reachable.<br>
>> - if (P == CurrentBlock) {<br>
>> - NumWithout = 2;<br>
>> - break;<br>
>> - } else if (!DT->isReachableFromEntry(P)) {<br>
>> - NumWithout = 2;<br>
>> - break;<br>
>> - }<br>
</div></div>>> + // Don't do PRE when it might increase code size, i.e. when // we<br>
<div><div class="h5">>> + would need to insert instructions in more than one pred.<br>
>> + if (NumWithout != 1 || NumWith == 0)<br>
>> + return false;<br>
>><br>
>> - Value* predV = findLeader(P, ValNo);<br>
>> - if (!predV) {<br>
>> - predMap.push_back(std::make_pair(static_cast<Value<br>
>> *>(nullptr), P));<br>
>> - PREPred = P;<br>
>> - ++NumWithout;<br>
>> - } else if (predV == CurInst) {<br>
>> - /* CurInst dominates this predecessor. */<br>
>> - NumWithout = 2;<br>
>> - break;<br>
>> - } else {<br>
>> - predMap.push_back(std::make_pair(predV, P));<br>
>> - ++NumWith;<br>
>> - }<br>
>> - }<br>
>> + // Don't do PRE across indirect branch.<br>
>> + if (isa<IndirectBrInst>(PREPred->getTerminator()))<br>
>> + return false;<br>
>><br>
>> - // Don't do PRE when it might increase code size, i.e. when<br>
>> - // we would need to insert instructions in more than one pred.<br>
>> - if (NumWithout != 1 || NumWith == 0)<br>
>> - continue;<br>
>> + // We can't do PRE safely on a critical edge, so instead we<br>
>> schedule<br>
>> + // the edge to be split and perform the PRE the next time we<br>
>> iterate<br>
>> + // on the function.<br>
</div></div>>> + unsigned SuccNum = GetSuccessorNumber(PREPred, CurrentBlock); if<br>
<div><div class="h5">>> + (isCriticalEdge(PREPred->getTerminator(), SuccNum)) {<br>
>> + toSplit.push_back(std::make_pair(PREPred->getTerminator(),<br>
>> SuccNum));<br>
>> + return false;<br>
>> + }<br>
>><br>
>> - // Don't do PRE across indirect branch.<br>
>> - if (isa<IndirectBrInst>(PREPred->getTerminator()))<br>
>> - continue;<br>
>> + // Instantiate the expression in the predecessor that lacked it.<br>
>> + // Because we are going top-down through the block, all value<br>
>> numbers<br>
>> + // will be available in the predecessor by the time we need them.<br>
>> Any<br>
>> + // that weren't originally present will have been instantiated<br>
>> earlier<br>
>> + // in this loop.<br>
>> + Instruction *PREInstr = CurInst->clone(); bool success = true;<br>
>> + for (unsigned i = 0, e = CurInst->getNumOperands(); i != e; ++i) {<br>
>> + Value *Op = PREInstr->getOperand(i);<br>
>> + if (isa<Argument>(Op) || isa<Constant>(Op) ||<br>
>> isa<GlobalValue>(Op))<br>
>> + continue;<br>
>><br>
>> - // We can't do PRE safely on a critical edge, so instead we<br>
>> schedule<br>
>> - // the edge to be split and perform the PRE the next time we<br>
>> iterate<br>
>> - // on the function.<br>
>> - unsigned SuccNum = GetSuccessorNumber(PREPred, CurrentBlock);<br>
>> - if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) {<br>
>> - toSplit.push_back(std::make_pair(PREPred->getTerminator(),<br>
>> SuccNum));<br>
>> - continue;<br>
>> - }<br>
>> + if (Value *V = findLeader(PREPred, VN.lookup(Op))) {<br>
>> + PREInstr->setOperand(i, V);<br>
>> + } else {<br>
>> + success = false;<br>
>> + break;<br>
>> + }<br>
>> + }<br>
>><br>
>> - // Instantiate the expression in the predecessor that lacked<br>
>> it.<br>
>> - // Because we are going top-down through the block, all value<br>
>> numbers<br>
>> - // will be available in the predecessor by the time we need<br>
>> them.<br>
>> Any<br>
>> - // that weren't originally present will have been instantiated<br>
>> earlier<br>
>> - // in this loop.<br>
>> - Instruction *PREInstr = CurInst->clone();<br>
>> - bool success = true;<br>
>> - for (unsigned i = 0, e = CurInst->getNumOperands(); i != e;<br>
>> ++i)<br>
>> {<br>
>> - Value *Op = PREInstr->getOperand(i);<br>
>> - if (isa<Argument>(Op) || isa<Constant>(Op) ||<br>
>> isa<GlobalValue>(Op))<br>
>> - continue;<br>
</div></div>>> + // Fail out if we encounter an operand that is not available in //<br>
>> + the PRE predecessor. This is typically because of loads which //<br>
<div><div class="h5">>> + are not value numbered precisely.<br>
>> + if (!success) {<br>
>> + DEBUG(verifyRemoved(PREInstr));<br>
>> + delete PREInstr;<br>
>> + return false;<br>
>> + }<br>
>><br>
>> - if (Value *V = findLeader(PREPred, VN.lookup(Op))) {<br>
>> - PREInstr->setOperand(i, V);<br>
>> - } else {<br>
>> - success = false;<br>
>> - break;<br>
>> - }<br>
>> - }<br>
>> + PREInstr->insertBefore(PREPred->getTerminator());<br>
>> + PREInstr->setName(CurInst->getName() + ".pre");<br>
>> + PREInstr->setDebugLoc(CurInst->getDebugLoc());<br>
>> + VN.add(PREInstr, ValNo);<br>
>> + ++NumGVNPRE;<br>
>> +<br>
>> + // Update the availability map to include the new instruction.<br>
>> + addToLeaderTable(ValNo, PREInstr, PREPred);<br>
>> +<br>
>> + // Create a PHI to make the value available in this block.<br>
>> + PHINode *Phi =<br>
>> + PHINode::Create(CurInst->getType(), predMap.size(),<br>
>> + CurInst->getName() + ".pre-phi",<br>
>> CurrentBlock->begin());<br>
>> + for (unsigned i = 0, e = predMap.size(); i != e; ++i) {<br>
>> + if (Value *V = predMap[i].first)<br>
>> + Phi->addIncoming(V, predMap[i].second);<br>
>> + else<br>
>> + Phi->addIncoming(PREInstr, PREPred); }<br>
>><br>
>> - // Fail out if we encounter an operand that is not available in<br>
>> - // the PRE predecessor. This is typically because of loads<br>
>> which<br>
>> - // are not value numbered precisely.<br>
>> - if (!success) {<br>
>> - DEBUG(verifyRemoved(PREInstr));<br>
>> - delete PREInstr;<br>
>> - continue;<br>
>> - }<br>
>> + VN.add(Phi, ValNo);<br>
>> + addToLeaderTable(ValNo, Phi, CurrentBlock);<br>
>> + Phi->setDebugLoc(CurInst->getDebugLoc());<br>
>> + CurInst->replaceAllUsesWith(Phi); if<br>
>> + (Phi->getType()->getScalarType()->isPointerTy()) {<br>
</div></div>>> + // Because we have added a PHI-use of the pointer value, it has<br>
<div><div class="h5">>> now<br>
>> + // "escaped" from alias analysis' perspective. We need to inform<br>
>> + // AA of this.<br>
>> + for (unsigned ii = 0, ee = Phi->getNumIncomingValues(); ii !=<br>
>> + ee;<br>
>> ++ii) {<br>
>> + unsigned jj = PHINode::getOperandNumForIncomingValue(ii);<br>
>> + VN.getAliasAnalysis()->addEscapingUse(Phi->getOperandUse(jj));<br>
>> + }<br>
>><br>
>> - PREInstr->insertBefore(PREPred->getTerminator());<br>
>> - PREInstr->setName(CurInst->getName() + ".pre");<br>
>> - PREInstr->setDebugLoc(CurInst->getDebugLoc());<br>
>> - VN.add(PREInstr, ValNo);<br>
>> - ++NumGVNPRE;<br>
>> -<br>
>> - // Update the availability map to include the new instruction.<br>
>> - addToLeaderTable(ValNo, PREInstr, PREPred);<br>
>> -<br>
>> - // Create a PHI to make the value available in this block.<br>
>> - PHINode* Phi = PHINode::Create(CurInst->getType(),<br>
>> predMap.size(),<br>
>> - CurInst->getName() + ".pre-phi",<br>
>> - CurrentBlock->begin());<br>
>> - for (unsigned i = 0, e = predMap.size(); i != e; ++i) {<br>
>> - if (Value *V = predMap[i].first)<br>
>> - Phi->addIncoming(V, predMap[i].second);<br>
>> - else<br>
>> - Phi->addIncoming(PREInstr, PREPred);<br>
>> - }<br>
>> -<br>
>> - VN.add(Phi, ValNo);<br>
>> - addToLeaderTable(ValNo, Phi, CurrentBlock);<br>
>> - Phi->setDebugLoc(CurInst->getDebugLoc());<br>
>> - CurInst->replaceAllUsesWith(Phi);<br>
>> - if (Phi->getType()->getScalarType()->isPointerTy()) {<br>
>> - // Because we have added a PHI-use of the pointer value, it<br>
>> has<br>
>> now<br>
>> - // "escaped" from alias analysis' perspective. We need to<br>
>> inform<br>
>> - // AA of this.<br>
>> - for (unsigned ii = 0, ee = Phi->getNumIncomingValues(); ii !=<br>
>> ee;<br>
>> - ++ii) {<br>
>> - unsigned jj = PHINode::getOperandNumForIncomingValue(ii);<br>
>> -<br>
>> VN.getAliasAnalysis()->addEscapingUse(Phi->getOperandUse(jj));<br>
>> - }<br>
>> + if (MD)<br>
>> + MD->invalidateCachedPointerInfo(Phi);<br>
>> + }<br>
>> + VN.erase(CurInst);<br>
>> + removeFromLeaderTable(ValNo, CurInst, CurrentBlock);<br>
>><br>
>> - if (MD)<br>
>> - MD->invalidateCachedPointerInfo(Phi);<br>
>> - }<br>
>> - VN.erase(CurInst);<br>
>> - removeFromLeaderTable(ValNo, CurInst, CurrentBlock);<br>
>> + DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n');<br>
>> + if (MD)<br>
>> + MD->removeInstruction(CurInst);<br>
>> + DEBUG(verifyRemoved(CurInst));<br>
>> + CurInst->eraseFromParent();<br>
>> + return true;<br>
>> +}<br>
>> +<br>
>> +/// performPRE - Perform a purely local form of PRE that looks for<br>
>> diamond<br>
</div></div>>> +/// control flow patterns and attempts to perform simple PRE at the<br>
<div class="HOEnZb"><div class="h5">>> join point.<br>
>> +bool GVN::performPRE(Function &F) {<br>
>> + bool Changed = false;<br>
>> + for (BasicBlock *CurrentBlock : depth_first(&F.getEntryBlock())) {<br>
>> + // Nothing to PRE in the entry block.<br>
>> + if (CurrentBlock == &F.getEntryBlock())<br>
>> + continue;<br>
>><br>
>> - DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n');<br>
>> - if (MD) MD->removeInstruction(CurInst);<br>
>> - DEBUG(verifyRemoved(CurInst));<br>
>> - CurInst->eraseFromParent();<br>
>> - Changed = true;<br>
>> + // Don't perform PRE on a landing pad.<br>
>> + if (CurrentBlock->isLandingPad())<br>
>> + continue;<br>
>> +<br>
>> + for (BasicBlock::iterator BI = CurrentBlock->begin(),<br>
>> + BE = CurrentBlock->end();<br>
>> + BI != BE;) {<br>
>> + Instruction *CurInst = BI++;<br>
>> + Changed = performScalarPRE(CurInst);<br>
>> }<br>
>> }<br>
>><br>
>> @@ -2637,25 +2655,21 @@ bool GVN::iterateOnFunction(Function &F)<br>
>><br>
>> // Top-down walk of the dominator tree bool Changed = false; -#if<br>
>> 0<br>
>> - // Needed for value numbering with phi construction to work.<br>
>> - ReversePostOrderTraversal<Function*> RPOT(&F);<br>
>> - for (ReversePostOrderTraversal<Function*>::rpo_iterator RI =<br>
>> RPOT.begin(),<br>
>> - RE = RPOT.end(); RI != RE; ++RI)<br>
>> - Changed |= processBlock(*RI);<br>
>> -#else<br>
>> // Save the blocks this function have before transformation begins.<br>
>> GVN may<br>
>> // split critical edge, and hence may invalidate the RPO/DT iterator.<br>
>> //<br>
>> std::vector<BasicBlock *> BBVect;<br>
>> BBVect.reserve(256);<br>
>> - for (DomTreeNode *X : depth_first(DT->getRootNode()))<br>
>> - BBVect.push_back(X->getBlock());<br>
>> + // Needed for value numbering with phi construction to work.<br>
>> + ReversePostOrderTraversal<Function *> RPOT(&F); for<br>
>> + (ReversePostOrderTraversal<Function *>::rpo_iterator RI =<br>
>> RPOT.begin(),<br>
>> + RE =<br>
>> RPOT.end();<br>
>> + RI != RE; ++RI)<br>
>> + BBVect.push_back(*RI);<br>
>><br>
>> for (std::vector<BasicBlock *>::iterator I = BBVect.begin(), E =<br>
>> BBVect.end();<br>
>> I != E; I++)<br>
>> Changed |= processBlock(*I);<br>
>> -#endif<br>
>><br>
>> return Changed;<br>
>> }<br>
>><br>
>> Modified:<br>
>> llvm/trunk/test/Analysis/TypeBasedAliasAnalysis/gvn-nonlocal-type-m<br>
>> ismatch.ll<br>
>> URL:<br>
>> <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/TypeBa" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/TypeBa</a><br>
>> sedAliasAnalysis/gvn-nonlocal-type-mismatch.ll?rev=222039&r1=222038<br>
>> &r2=222039&view=diff<br>
>> ===================================================================<br>
>> ===========<br>
>> ---<br>
>> llvm/trunk/test/Analysis/TypeBasedAliasAnalysis/gvn-nonlocal-type-m<br>
>> ismatch.ll<br>
>> (original)<br>
>> +++<br>
>> llvm/trunk/test/Analysis/TypeBasedAliasAnalysis/gvn-nonlocal-type-m<br>
>> ismatch.ll<br>
>> Fri Nov 14 15:09:13 2014<br>
>> @@ -46,12 +46,12 @@ entry:<br>
>> br i1 %c, label %if.else, label %if.then<br>
>><br>
>> if.then:<br>
>> - %t = load i32* %p, !tbaa !4<br>
>> + %t = load i32* %p, !tbaa !3<br>
>> store i32 %t, i32* %q<br>
>> ret void<br>
>><br>
>> if.else:<br>
>> - %u = load i32* %p, !tbaa !3<br>
>> + %u = load i32* %p, !tbaa !4<br>
>> store i32 %u, i32* %q<br>
>> ret void<br>
>> }<br>
>> @@ -61,11 +61,11 @@ if.else:<br>
>><br>
>> ; CHECK: @watch_out_for_another_type_change<br>
>> ; CHECK: if.then:<br>
>> -; CHECK: %t = load i32* %p<br>
>> -; CHECK: store i32 %t, i32* %q<br>
>> +; CHECK: store i32 0, i32* %q<br>
>> ; CHECK: ret void<br>
>> ; CHECK: if.else:<br>
>> -; CHECK: store i32 0, i32* %q<br>
>> +; CHECK: %u = load i32* %p<br>
>> +; CHECK: store i32 %u, i32* %q<br>
>><br>
>> define void @watch_out_for_another_type_change(i1 %c, i32* %p, i32*<br>
>> %p1,<br>
>> i32* %q) nounwind {<br>
>> entry:<br>
>> @@ -74,12 +74,12 @@ entry:<br>
>> br i1 %c, label %if.else, label %if.then<br>
>><br>
>> if.then:<br>
>> - %t = load i32* %p, !tbaa !3<br>
>> + %t = load i32* %p, !tbaa !4<br>
>> store i32 %t, i32* %q<br>
>> ret void<br>
>><br>
>> if.else:<br>
>> - %u = load i32* %p, !tbaa !4<br>
>> + %u = load i32* %p, !tbaa !3<br>
>> store i32 %u, i32* %q<br>
>> ret void<br>
>> }<br>
>><br>
>> Added: llvm/trunk/test/Transforms/GVN/pre-gep-load.ll<br>
>> URL:<br>
>> <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GVN/" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GVN/</a><br>
>> pre-gep-load.ll?rev=222039&view=auto<br>
>> ===================================================================<br>
>> ===========<br>
>> --- llvm/trunk/test/Transforms/GVN/pre-gep-load.ll (added)<br>
>> +++ llvm/trunk/test/Transforms/GVN/pre-gep-load.ll Fri Nov 14<br>
>> +++ 15:09:13<br>
>> 2014<br>
>> @@ -0,0 +1,49 @@<br>
>> +; RUN: opt < %s -basicaa -gvn -enable-load-pre -S | FileCheck %s<br>
>> +target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"<br>
>> +target triple = "aarch64--linux-gnu"<br>
>> +<br>
>> +define double @foo(i32 %stat, i32 %i, double** %p) { ;<br>
>> +CHECK-LABEL: @foo(<br>
>> +entry:<br>
>> + switch i32 %stat, label %sw.default [<br>
>> + i32 0, label %<a href="http://sw.bb" target="_blank">sw.bb</a><br>
>> + i32 1, label %<a href="http://sw.bb" target="_blank">sw.bb</a><br>
>> + i32 2, label %sw.bb2<br>
>> + ]<br>
>> +<br>
>> +<a href="http://sw.bb" target="_blank">sw.bb</a>: ; preds = %entry,<br>
>> %entry<br>
>> + %idxprom = sext i32 %i to i64<br>
>> + %arrayidx = getelementptr inbounds double** %p, i64 0<br>
>> + %0 = load double** %arrayidx, align 8<br>
>> + %arrayidx1 = getelementptr inbounds double* %0, i64 %idxprom<br>
>> + %1 = load double* %arrayidx1, align 8 %sub = fsub double %1,<br>
>> + 1.000000e+00 %cmp = fcmp olt double %sub, 0.000000e+00 br i1<br>
>> + %cmp, label %if.then, label %if.end<br>
>> +<br>
>> +if.then: ; preds = %<a href="http://sw.bb" target="_blank">sw.bb</a><br>
>> + br label %return<br>
>> +<br>
>> +if.end: ; preds = %<a href="http://sw.bb" target="_blank">sw.bb</a><br>
>> + br label %sw.bb2<br>
>> +<br>
>> +sw.bb2: ; preds = %if.end,<br>
>> %entry<br>
>> + %idxprom3 = sext i32 %i to i64<br>
>> + %arrayidx4 = getelementptr inbounds double** %p, i64 0<br>
>> + %2 = load double** %arrayidx4, align 8<br>
>> + %arrayidx5 = getelementptr inbounds double* %2, i64 %idxprom3<br>
>> + %3 = load double* %arrayidx5, align 8 ; CHECK: sw.bb2:<br>
>> +; CHECK-NEXT-NOT: sext<br>
>> +; CHECK-NEXT: phi double [<br>
>> +; CHECK-NOT: load<br>
>> + %sub6 = fsub double 3.000000e+00, %3<br>
>> + br label %return<br>
>> +<br>
>> +sw.default: ; preds = %entry<br>
>> + br label %return<br>
>> +<br>
>> +return: ; preds =<br>
>> %sw.default, %sw.bb2, %if.then<br>
>> + %retval.0 = phi double [ 0.000000e+00, %sw.default ], [ %sub6,<br>
>> %sw.bb2 ], [ %sub, %if.then ]<br>
>> + ret double %retval.0<br>
>> +}<br>
>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>><br>
>><br>
>><br>
>><br>
>><br>
>><br>
>><br>
>><br>
>><br>
>> <0001-Reapply-r221924-GVN-Perform-Scalar-PRE-on-gep-indice.patch><br>
>><br>
>><br>
>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>><br>
><br>
><br>
> _______________________________________________<br>
> llvm-commits mailing list<br>
> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
><br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
<br>
</div></div></blockquote></div><br></div></div></div>