Revert "Revert "[LTO] Support LLVM level link time optimization on Darwin, Linux and Windows"" #32237

kateinoigakukun · 2020-06-08T00:10:36Z

Resolve CI failure on android and other hosts addressed at #31146 (comment) and restore #32235

Android test cases try to use gold linker -use-ld=gold but we don't support gold and other linkers except for lld for LTO now. So I changed to force to use lld when LTO on Unix like toolchain

: 'RUN: at line 13';   /home/ubuntu/jenkins/workspace/oss-swift-RA-linux-ubuntu-16.04-android/buildbot_linux/swift-linux-x86_64/bin/swiftc -target armv7-none-linux-android -toolchain-stdlib-rpath -Xcc --sysroot=/home/ubuntu/android-ndk-r17/platforms/android-21/arch-arm -Xclang-linker --sysroot=/home/ubuntu/android-ndk-r17/platforms/android-21/arch-arm -tools-directory /home/ubuntu/android-ndk-r17/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/arm-linux-androideabi/bin -L /home/ubuntu/android-ndk-r17/sources/cxx-stl/llvm-libc++/libs/armeabi-v7a -L /home/ubuntu/android-ndk-r17/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x -L /home/ubuntu/android-ndk-r17/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/arm-linux-androideabi/lib  -module-cache-path '/home/ubuntu/jenkins/workspace/oss-swift-RA-linux-ubuntu-16.04-android/buildbot_linux/swift-linux-x86_64/swift-test-results/armv7-none-linux-androideabi/clang-module-cache' -use-ld=gold /home/ubuntu/jenkins/workspace/oss-swift-RA-linux-ubuntu-16.04-android/swift/test/Driver/Inputs/lto/lib.swift -static -lto=llvm -emit-library -emit-module -module-name A -working-directory /home/ubuntu/jenkins/workspace/oss-swift-RA-linux-ubuntu-16.04-android/buildbot_linux/swift-linux-x86_64/test-android-armv7/Driver/Output/link-time-opt-staticlib-thin.swift.tmp/thin-static

CC: @compnerd

…rwin, Linux and Windows"" This reverts commit 71309a8.

At this time, we don't support gold and other linkers for LTO.

compnerd · 2020-06-08T02:37:46Z

@swift-ci please test

kateinoigakukun · 2020-06-09T10:47:14Z

@compnerd Could you trigger android and other platforms CI jobs?

compnerd · 2020-06-09T15:25:01Z

I dont think its possible to trigger a run for android sadly. I can trigger it for AWS Linux and CentOS 8 though, so lets just ensure that those will stay green.

compnerd · 2020-06-09T15:25:05Z

@swift-ci Please test Amazon Linux 2 platform

compnerd · 2020-06-09T15:25:20Z

@swift-ci Please test CentOS 8 platform

swift-ci · 2020-06-09T15:33:28Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

swift-ci · 2020-06-09T15:38:48Z

Build failed
Swift Test CentOS 8 Platform
Git Sha - 97eb7dc

kateinoigakukun · 2020-06-09T17:31:12Z

Hmm, it seems these CI jobs are not working well...

kateinoigakukun · 2020-06-11T00:54:54Z

I'm not sure who maintains these platforms CI, but it seems something wrong in CI system. CC: @compnerd

compnerd · 2020-06-11T03:17:53Z

CC: @shahmishal - are CentOS and AWS Linux CI builders supposed to work?

shahmishal · 2020-06-11T03:24:01Z

They should be working, I will look into this tomorrow.

swift-ci · 2020-06-12T00:13:35Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

swift-ci · 2020-06-12T01:34:32Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

swift-ci · 2020-06-12T05:56:28Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

compnerd · 2020-06-14T18:44:11Z

@swift-ci Please test Amazon Linux 2 platform

swift-ci · 2020-06-14T20:15:28Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

compnerd · 2020-06-15T15:48:01Z

FAILED: src/swift/libswiftDispatch.a src/swift/CMakeFiles/swiftDispatch.dir/Block.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Data.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Dispatch.swift.o src/swift/CMakeFiles/swiftDispatch.dir/IO.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Private.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Queue.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Source.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Time.swift.o src/swift/CMakeFiles/swiftDispatch.dir/Wrapper.swift.o src/swift/swift/Dispatch.swiftmodule 
: && /home/build-user/build/buildbot_linux/swift-linux-x86_64/bin/swiftc -output-file-map src/swift/CMakeFiles/swiftDispatch.dir/output-file-map.json -incremental -j 32 -emit-library -static -o src/swift/libswiftDispatch.a -module-name Dispatch -module-link-name swiftDispatch -emit-module -emit-module-path src/swift/swift/Dispatch.swiftmodule -emit-dependencies  -O   -Xcc -fblocks -Xcc -fmodule-map-file=/home/build-user/swift-corelibs-libdispatch/dispatch/module.modulemap -Xcc -I/home/build-user/swift-corelibs-libdispatch -I /home/build-user/swift-corelibs-libdispatch/src/BlocksRuntime -I . -I /home/build-user/swift-corelibs-libdispatch -I /home/build-user/swift-corelibs-libdispatch/src -I src /home/build-user/swift-corelibs-libdispatch/src/swift/Block.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Data.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Dispatch.swift /home/build-user/swift-corelibs-libdispatch/src/swift/IO.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Private.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Queue.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Source.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Time.swift /home/build-user/swift-corelibs-libdispatch/src/swift/Wrapper.swift    && :
/home/build-user/swift-corelibs-libdispatch/src/swift/Dispatch.swift:99:4: warning: default will never be executed
                        default: return nil
                        ^
<unknown>:0: error: unable to execute command: <unknown>

Thats not really helpful :/, lets try this once more. If it fails (due to infra), lets go ahead and merge and deal with the fallout.

compnerd · 2020-06-15T15:48:21Z

@swift-ci please test Amazon Linux 2 platform

shahmishal · 2020-06-15T15:54:00Z

Thats not really helpful :/, lets try this once more. If it fails (due to infra), lets go ahead and merge and deal with the fallout.

Why merge it with known failure? Are you not able to reproduce this locally?

compnerd · 2020-06-15T15:58:42Z

@shahmishal hmm, are there instructions for reproducing the builds locally? But, I suspect that its a disk size thing.

compnerd · 2020-06-15T15:59:15Z

@swift-ci please test CentOS 8 platform

swift-ci · 2020-06-15T16:12:54Z

Build failed
Swift Test CentOS 8 Platform
Git Sha - 97eb7dc

compnerd · 2020-06-15T16:25:32Z

@shahmishal CentOS still is failing without building:

 ---> Running in 0432b5ffb4f0
CentOS-8 - AppStream                            0.0  B/s |   0  B     00:20    
�[91mFailed to download metadata for repo 'AppStream'
�[0m�[91mError: Failed to download metadata for repo 'AppStream'
�[0mThe command '/bin/sh -c yum install -y https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm' returned a non-zero code: 1
Build step 'Execute shell' marked build as failure

shahmishal · 2020-06-15T16:31:09Z

@compnerd I am looking into the CentOS 8 issues, however Amazon Linux 2 issue looks real.

swift-ci · 2020-06-15T16:31:46Z

Build failed
Swift Test CentOS 8 Platform
Git Sha - 97eb7dc

swift-ci · 2020-06-15T16:40:53Z

Build failed
Swift Test CentOS 8 Platform
Git Sha - 97eb7dc

gottesmm · 2020-06-15T17:38:41Z

include/swift/Driver/Action.h

    assert(Kind != LinkKind::None && Kind != LinkKind::StaticLibrary);
  }

  LinkKind getKind() const { return Kind; }

+  bool PerformLTO() const { return LTO; }


Please follow naming guidelines!

-> shouldPerformLTO().

Can you rename the flag to ShouldPerformLTO? What does bool LTO even mean? It is ambiguous.

gottesmm · 2020-06-15T17:39:11Z

include/swift/AST/IRGenOptions.h

@@ -217,6 +223,8 @@ class IRGenOptions {
  /// Whether we should embed the bitcode file.
  IRGenEmbedMode EmbedMode : 2;

+  IRGenLLVMLTOKind LLVMLTOKind: 2;


This code has not been run through git-clang-format. Please do so before this lands!

gottesmm · 2020-06-15T17:41:25Z

lib/Driver/Driver.cpp

@@ -1779,6 +1782,18 @@ void Driver::buildOutputInfo(const ToolChain &TC, const DerivedArgList &Args,

  }

+  if (const Arg *A = Args.getLastArg(options::OPT_lto)) {
+    auto LTOVariant = llvm::StringSwitch<Optional<OutputInfo::LTOKind>>(A->getValue())
+      .Case("llvm", OutputInfo::LTOKind::LLVMThin)


Lets be explicit here and be clear that it is llvm-thin.

gottesmm · 2020-06-15T17:43:27Z

lib/Driver/WindowsToolChains.cpp

+    }
+    break;
+  }
+  case OutputInfo::LTOKind::None: break;


This is another example where you haven't used git-clang-format. Please do so before this lands!

gottesmm

I want to look at this again before it goes back in. All of the changes that have been made to IRGen need IRGen FileCheck test cases.

swift-ci · 2020-06-15T17:45:55Z

Build failed
Swift Test Amazon Linux 2 Platform
Git Sha - 97eb7dc

gottesmm

At a higher level, you probably want to split up this PR into a PR that handles the driver part and the part that handles the pipeline part. The first PR would have a bunch of IR tests that validate that all of the proper LTO bigs work together ignoring the driver. This PR smooshes the two together and doesn't provide any .ll tests. That needs to be fixed.

gottesmm · 2020-06-15T17:45:45Z

lib/IRGen/IRGenModule.cpp

+  // When performing LTO, we always use lld that supports auto linking mechanism with ELF.
+  // So embed dependent libraries names in "llvm.dependent-libraries" instead of "llvm.linker.options".
+  const StringRef AutolinkSectionName =
+      TargetInfo.OutputObjectFormat == llvm::Triple::ELF && IRGen.Opts.LLVMLTOKind != IRGenLLVMLTOKind::None


gottesmm · 2020-06-15T17:45:53Z

lib/IRGen/IRGenModule.cpp

-        llvm::MDNode::get(ctx, llvm::MDString::get(ctx, opt)));
+    if (TargetInfo.OutputObjectFormat == llvm::Triple::ELF && IRGen.Opts.LLVMLTOKind != IRGenLLVMLTOKind::None) {
+      // When performing LTO, we always use lld that supports auto linking mechanism with ELF.
+      // So embed dependent libraries names in "llvm.dependent-libraries" instead of options


compnerd · 2020-06-15T17:58:09Z

CC: @atrick @eeckstein

kateinoigakukun added 2 commits June 8, 2020 09:08

Revert "Revert "[LTO] Support LLVM level link time optimization on Da…

ac96df6

…rwin, Linux and Windows"" This reverts commit 71309a8.

[LTO] Force to use lld when LTO on Unix like toolchain

97eb7dc

At this time, we don't support gold and other linkers for LTO.

kateinoigakukun mentioned this pull request Jun 14, 2020

[LTO] Setup LTO pipeline and swift-lto tool #32233

Closed

gottesmm reviewed Jun 15, 2020

View reviewed changes

This was referenced Jun 17, 2020

[LTO] Build lld on Windows CI #32427

Merged

[LTO] Support LLVM LTO for IRGen and frontend #32429

Merged

[LTO] Support LLVM LTO for driver #32430

Merged

kateinoigakukun closed this Jun 27, 2020

Revert "Revert "[LTO] Support LLVM level link time optimization on Darwin, Linux and Windows"" #32237

Revert "Revert "[LTO] Support LLVM level link time optimization on Darwin, Linux and Windows"" #32237

Uh oh!

Conversation

kateinoigakukun commented Jun 8, 2020

Uh oh!

compnerd commented Jun 8, 2020

Uh oh!

kateinoigakukun commented Jun 9, 2020

Uh oh!

compnerd commented Jun 9, 2020

Uh oh!

compnerd commented Jun 9, 2020

Uh oh!

compnerd commented Jun 9, 2020

Uh oh!

swift-ci commented Jun 9, 2020

Uh oh!

swift-ci commented Jun 9, 2020

Uh oh!

kateinoigakukun commented Jun 9, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kateinoigakukun commented Jun 11, 2020

Uh oh!

compnerd commented Jun 11, 2020

Uh oh!

shahmishal commented Jun 11, 2020

Uh oh!

swift-ci commented Jun 12, 2020

Uh oh!

swift-ci commented Jun 12, 2020

Uh oh!

swift-ci commented Jun 12, 2020

Uh oh!

compnerd commented Jun 14, 2020

Uh oh!

swift-ci commented Jun 14, 2020

Uh oh!

compnerd commented Jun 15, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

compnerd commented Jun 15, 2020

Uh oh!

shahmishal commented Jun 15, 2020

Uh oh!

compnerd commented Jun 15, 2020

Uh oh!

compnerd commented Jun 15, 2020

Uh oh!

swift-ci commented Jun 15, 2020

Uh oh!

compnerd commented Jun 15, 2020

Uh oh!

shahmishal commented Jun 15, 2020

Uh oh!

swift-ci commented Jun 15, 2020

Uh oh!

swift-ci commented Jun 15, 2020

Uh oh!

gottesmm Jun 15, 2020

Choose a reason for hiding this comment

Uh oh!

gottesmm Jun 15, 2020

Choose a reason for hiding this comment

Uh oh!

gottesmm Jun 15, 2020

Choose a reason for hiding this comment

Uh oh!

gottesmm Jun 15, 2020

Choose a reason for hiding this comment

Uh oh!

gottesmm left a comment

Choose a reason for hiding this comment

Uh oh!

swift-ci commented Jun 15, 2020

Uh oh!

gottesmm left a comment

Choose a reason for hiding this comment

Uh oh!

kateinoigakukun commented Jun 9, 2020 •

edited

Loading

compnerd commented Jun 15, 2020 •

edited

Loading