Skip to content

Latest commit

 

History

History
429 lines (288 loc) · 28.7 KB

HACKING.md

File metadata and controls

429 lines (288 loc) · 28.7 KB

Profiling the compiler

Profiling with Async profiler

IDEA Ultimate contains an Async sampling profiler. As of IDEA 2018.3 Async sampling profiler is still an experimental feature, so use Ctrl-Alt-Shift-/ on Linux, Cmd-Alt-Shift-/ on macOS to activate it. Then start compilation in CLI with --no-daemon and -Porg.gradle.workers.max=1 flags (running Gradle task with the profiler doesn't seem to work properly) and attach to the running process using "Run/Attach Profiler to Local Process" menu item.

Select "K2NativeKt" or "org.jetbrains.kotlin.cli.utilities.MainKt" process. On completion profiler will produce flame diagram which could be navigated with the mouse (click-drag moves, wheel scales). More RAM in IDE (>4G) could be helpful when analyzing longer runs. As Async is a sampling profiler, to get sensible coverage longer runs are important.

Profiling with YourKit

Unlike Async profiler in IDEA, YourKit can work as an exact profiler and provide complete coverage of all methods along with exact invocation counters.

Install the YourKit profiler for your platform from https://www.yourkit.com/java/profiler. Set AGENT variable to the JVMTI agent provided by YourKit, like

    export AGENT=/Applications/YourKit-Java-Profiler-2018.04.app/Contents/Resources/bin/mac/libyjpagent.jnilib

To profile standard library compilation:

    ./gradlew -PstdLibJvmArgs="-agentpath:$AGENT=probe_disable=*,listen=all,tracing"  :kotlin-native:dist

To profile platform libraries start build of proper target like this:

    ./gradlew -PplatformLibsJvmArgs="-agentpath:$AGENT=probe_disable=*,listen=all,tracing"  :kotlin-native:ios_arm64PlatformLibs

To profile standalone code compilation use:

    JAVA_OPTS="-agentpath:$AGENT=probe_disable=*,listen=all,tracing" ./kotlin-native/dist/bin/konanc file.kt

Then attach to the desired application in YourKit GUI and use CPU tab to inspect CPU consuming methods. Saving the trace may be needed for more analysis. Adjusting -Xmx in $HOME/.yjp/ui.ini could help with the big traces.

To perform memory profiling follow the steps above, and after attachment to the running process use "Start Object Allocation Recording" button. See https://www.yourkit.com/docs/java/help/allocations.jsp for more details.

Compiler Gradle options

There are several gradle flags one can use for Konan build.

  • -Pbuild_flags passes flags to the compiler used to build stdlib

      ./gradlew -Pbuild_flags="--disable lower_inline --print_ir" :kotlin-native:stdlib
    
  • -Pshims compiles LLVM interface with tracing "shims". Allowing one to trace the LLVM calls from the compiler. Make sure to rebuild the project.

      ./gradlew -Pshims=true :kotlin-native:dist
    

Compiler environment variables

  • KONAN_DATA_DIR changes .konan local data directory location ($HOME/.konan by default). Works both with cli compiler and gradle plugin

Testing

Compiler blackbox tests

There are blackbox tests that check the correctness of Kotlin language features support in the Kotlin/Native compiler. The same tests are also used in other Kotlin backends (JVM, JS) for the same purpose, and they generally do not depend on a particular Kotlin/Native target.

To run blackbox compiler tests use:

./gradlew :native:native.tests:codegenBoxTest
  • --tests allows one to choose test suite(s) or test case(s) to run.

    ./gradlew :native:native.tests:codegenBoxTest --tests "org.jetbrains.kotlin.konan.test.blackbox.NativeCodegenBoxTestGenerated\$Box\$*"
    
  • There are also Gradle project properties that can be used to control various aspects of blackbox tests. Example:

    ./gradlew :native:native.tests:codegenBoxTest \
        -Pkotlin.internal.native.test.<property1Name>=<property1Value> \
        -Pkotlin.internal.native.test.<property2Name>=<property2Value>
    
Property Description
nativeHome The full path to the Kotlin/Native distribution that will be used to run tests on. If not specified, then the distribution will be built by the corresponding Gradle task as a precondition before running tests: :kotlin-native:dist or :kotlin-native:${target}CrossDist.

Typically, this parameter is used to run tests against a distribution that was already built and cached somewhere. For example, to reproduce a test failure on a certain Kotlin/Native build.
compilerClasspath The full path to the Kotlin/Native compiler classpath. If not specified, then the classpath is deduced as ${nativeHome}/konan/lib/kotlin-native-compiler-embeddable.jar

This property allows to override the compiler itself preserving the rest of the distribution, and this way to test various backward compatibility cases.
target The name of the Kotlin/Native target under test
mode * ONE_STAGE_MULTI_MODULE : Compile each test file as one or many modules (depending on MODULE directives declared in the file). Produce a KLIB per each module except the last one. Finally, produce an executable file by compiling the latest module with all other KLIBs passed as -library
* TWO_STAGE_MULTI_MODULE (default): Compile each test file as one or many modules (depending on MODULE directives declared in the file). Produce a KLIB per each module. Finally, produce an executable file by passing the latest KLIB as -Xinclude and all other KLIBs as -library.
forceStandalone If true then all tests with // KIND: REGULAR inside test data file are executed as if they were be with // KIND: STANDALONE. The default is false.
compileOnly If true then tests are fully compiled to the executable binary, but not executed afterwards. The default is false.
optimizationMode Compiler optimization mode: DEBUG (default), OPT, NO
memoryModel The memory model: LEGACY or EXPERIMENTAL (default)
useThreadStateChecker If true the thread state checker is enabled. The default is false.

Note: Thread state checker can be enabled only in combination with optimizationMode=DEBUG, memoryModel=EXPERIMENTAL and cacheMode=NO.
gcType The type of GC: UNSPECIFIED (default), NOOP, STMS, CMS

Note: The GC type can be specified only in combination with memoryModel=EXPERIMENTAL.
gcScheduler The type of GC scheduler: UNSPECIFIED (default), DISABLED, WITH_TIMER, ON_SAFE_POINTS, AGGRESSIVE

Note: The GC scheduler type can be specified only in combination with memoryModel=EXPERIMENTAL.
cacheMode * NO: no caches
* STATIC_ONLY_DIST (default): use only caches for libs from the distribution
* STATIC_EVERYWHERE: use caches for libs from the distribution and generate caches for all produced KLIBs

Note: Any cache mode that permits using caches can be enabled only when thread state checker is disabled.
executionTimeout Max permitted duration of each individual test execution in milliseconds
sanitizer Run tests with sanitizer: NONE (default), THREAD.

A test can be ignored for certain property values with the help of test directives within test source files:

  • // IGNORE_NATIVE: <name>=<value> to ignore test for both K1 and K2 frontends
  • // IGNORE_NATIVE_K1: <name>=<value> to ignore test for K1 frontend only
  • // IGNORE_NATIVE_K2: <name>=<value> to ignore test for K2 frontend only

Good examples are:

  • // IGNORE_NATIVE: cacheMode=STATIC_EVERYWHERE
  • // IGNORE_NATIVE_K1: mode=ONE_STAGE_MULTI_MODULE
  • // IGNORE_NATIVE_K2: optimizationMode=OPT
  • // IGNORE_NATIVE: cacheMode=STATIC_EVERYWHERE && target=linux_x64

Test will be ignored in case value of any // IGNORE_NATIVE* directive would match to an actual test run setting.

Target-specific tests

There are also tests that are very Native-backend specific: tests for Kotlin/Native-specific function, C-interop tests, linkage tests, etc. In common, they are called "target-specific tests".

Note: on MacOS aarch64, JDK aarch64 is required

To run Kotlin/Native target-specific tests use (takes time):

./gradlew :kotlin-native:backend.native:tests:sanity  2>&1 | tee log                             # quick one
./gradlew :kotlin-native:backend.native:tests:run  2>&1 | tee log                                # sanity tests + platform tests
  • -Pprefix allows one to choose tests by prefix to run.

      ./gradlew -Pprefix=array :kotlin-native:backend.native:tests:run
    
  • -Ptest_flags passes flags to the compiler used to compile tests

      ./gradlew -Ptest_flags="--time" :kotlin-native:backend.native:tests:array0
    
  • -Ptest_target specifies cross target for a test run.

      ./gradlew -Ptest_target=raspberrypi :kotlin-native:backend.native:tests:array0
    
  • -Premote=user@host sets remote test execution login/hostname. Good for cross compiled tests.

      ./gradlew [email protected] :kotlin-native:backend.native:tests:run
    
  • -Ptest_verbose enables printing compiler args and other helpful information during a test execution.

      ./gradlew -Ptest_verbose :kotlin-native:backend.native:tests:mpp_optional_expectation
    
  • -Ptest_two_stage enables two-stage compilation of tests. If two-stage compilation is enabled, test sources are compiled into a klibrary and then a final native binary is produced from this klibrary using the -Xinclude compiler flag.

      ./gradlew -Ptest_two_stage :kotlin-native:backend.native:tests:array0
    
  • -Ptest_with_cache_kind=static|dynamic enables using caches during testing.

  • -Ptest_compile_only allows one to only compile tests, without actually running them. It is useful for testing compilation pipeline in case of targets that are tricky to execute tests on.

Runtime unit tests

To run all runtime unit tests on the host machine:

./gradlew :kotlin-native:runtime:hostRuntimeTests

Use -Pgtest_filter= to filter which tests to run (uses Google Test filter syntax). Use -Pgtest_timeout= to limit how much time each test executable can take (accepts values like 30s, 1h15m20s, and so on).

We use Google Test to execute the runtime unit tests. The build automatically fetches the specified Google Test revision to kotlin-native/runtime/googletest. It is possible to manually modify the downloaded GTest sources for debug purposes; the build will not overwrite them by default.

To forcibly redownload Google Test when running tests, use the corresponding project property:

 ./gradlew :kotlin-native:runtime:hostRuntimeTests -Prefresh-gtest

or run the downloadGoogleTest task directly with the --refresh CLI key:

./gradlew :kotlin-native:downloadGoogleTest --refresh

To use a local GTest copy instead of the downloaded one, add the following line to kotlin-native/runtime/build.gradle.kts:

googletest.useLocalSources("<path to local GTest sources>")

Debugging Kotlin/Native compiler

To debug Kotlin/Native compiler with a debugger (e.g. in IntelliJ IDEA), you should run the compiler in a special way to make it wait for a debugger connection. There are different ways to achieve that.

Making the compiler wait for a debugger when running Gradle build

If you use Kotlin/Native compiler as a part of your Gradle build (which is usually the case), you can simply debug the entire Gradle process by adding -Dorg.gradle.debug=true to Gradle arguments. I.e., run the build like

./gradlew $task -Dorg.gradle.debug=true

This will make the Gradle process wait for a debugger to connect right after start, and you will be able to connect with IntelliJ IDEA (see below) or other debuggers.

In this case you will debug the execution of all triggered tasks. In particular, Kotlin/Native tasks.

Note: this won't work if you have kotlin.native.disableCompilerDaemon=true in your Gradle properties.

Making the command-line compiler wait for a debugger

Kotlin/Native compiler is also available in command line. To make it wait for a debugger, you have to add certain JVM options when launching it. These flags could be set via environment variable JAVA_OPTS. The following bash script (debug.sh) can be used to debug:

#!/bin/bash
set -e
JAVA_OPTS="-agentlib:jdwp=transport=dt_socket,server=y,suspend=y,address=5005" "$@"

So running Kotlin/Native compiler with this script

~/debug.sh <path_to_kotlin>/kotlin-native/dist/bin/kotlinc-native ...

makes the compiler wait for a debugger to connect.

Additionally, even if you build with Gradle, you can extract command-line compiler arguments from the detailed Gradle output of your project's build process. This will allow you to run the command-line compiler instead of Gradle, which might be helpful when debugging. To get the detailed Gradle output, run the Gradle command with -i flag. See also degrade tool -- it automates extracting Kotlin/Native command-line tools invocations from Gradle builds.

Attaching with IntelliJ IDEA

After you instructed a Gradle process or a command-line compiler invocation to wait for a debugger connection, now it's time to connect. It is possible to do this with different JVM debuggers. Here is the explanation for IntelliJ IDEA.

First, set up your debug configuration in IntelliJ IDEA. After opening Kotlin project in IDEA (more information can be found on https://github.com/JetBrains/kotlin#build-environment-requirements and https://github.com/JetBrains/kotlin#-working-with-the-project-in-intellij-idea), create a remote debugger config.

Use Edit Configurations in dropdown menu with your run/debug configurations and add a new Remote JVM Debug. Ensure the port is 5005. Now just run this configuration. It’ll attach and let the compiler run.

Developing Kotlin/Native runtime in CLion

It's possible to use CLion to develop C++ runtime code efficiently and to have navigation in C++ code. To open runtime code in CLion as project and use all provided features of CLion, use a compilation database. It lets CLion detect project files and extract all the necessary compiler information, such as include paths and compilation flags. To generate a compilation database for the Kotlin/Native runtime, run the Gradle task ./gradlew :kotlin-native:compdb. This task generates <path_to_kotlin>/kotlin-native/compile_commands.json file that should be opened in CLion as project. Other developer tools also can use generated compilation database, but then clangd tool should be installed manually.

Also, it's possible to build Kotlin/Native runtime with dwarf debug information, which can be useful for debugging. To do this you should add kotlin.native.isNativeRuntimeDebugInfoEnabled=true line to local.properties file. Note, that changing this property requires clean compiler rebuild with gradle daemon restart.

Unfortunately, this feature works quite unstable because of using several llvm versions simultaneously, so it's need to be additionally enabled while compiling application with -Xbinary=stripDebugInfoFromNativeLibs=false compiler flag or corresponding setting in gradle build script. After doing this, Kotlin/Native runtime in application is debuggable in CLion, with Attach to process tool.

Performance measurement

Pre-requisite

konanRun task needs built compiler and platform POSIX libs. To test against working tree make sure to run

./gradlew :kotlin-native:dist :kotlin-native:distPlatformLibs

Run tests

To measure performance of Kotlin/Native compiler on existing benchmarks:

cd kotlin-native/performance
../../gradlew :konanRun

konanRun task can be run separately for one/several benchmark applications:

cd kotlin-native/performance
../../gradlew :cinterop:konanRun

konanRun task has parameter filter which allows to run only some subset of benchmarks:

cd kotlin-native/performance
../../gradlew :cinterop:konanRun --filter=struct,macros
../../gradlew :ring:konanRun --filter=Euler.problem9,ForLoops.charArrayIndicesLoop

Or you can use filterRegex if you want to specify the filter as regexes:

cd kotlin-native/performance
../../gradlew :ring:konanRun --filterRegex=String.*,Loop.*

There us also verbose mode to follow progress of running benchmarks

cd kotlin-native/performance
../../gradlew :cinterop:konanRun --verbose

> Task :performance:cinterop:konanRun
[DEBUG] Warm up iterations for benchmark macros
[DEBUG] Running benchmark macros
...

There are also tasks for running benchmarks on JVM (pay attention, some benchmarks e.g. cinterop benchmarks can't be run on JVM)

cd kotlin-native/performance
../../gradlew :jvmRun

You can use the compilerArgs property to pass flags to the compiler used to compile the benchmarks:

cd kotlin-native/performance
../../gradlew :konanRun -PcompilerArgs="--time -g"

Analyze the results

Files with results of benchmarks run are saved in kotlin-native/performance/build folder: nativeReport.json for konanRun and jvmReport.json for jvmRun. You can change the output filename by setting the nativeJson property for konanRun and jvmJson for jvmRun:

cd kotlin-native/performance
../../gradlew :ring:konanRun --filter=String.*,Loop.* -PnativeJson=stringsAndLoops.json

To compare different results use benchmarksAnalyzer tool:

./gradlew macos_arm64PlatformLibs  # use target of your laptop here instead
cd kotlin-native/tools/benchmarksAnalyzer
../../../gradlew build
./build/bin/<target>/benchmarksAnalyzerReleaseExecutable/benchmarksAnalyzer.kexe <file1> <file2>

Tool has several renders which allow produce output report in different forms (text, html, etc.). To set up render use flag --render/-r. Output can be redirected to file with flag --output/-o. To get detailed information about supported options, please use --help/-h.

Analyzer tool can compare both local files and files placed on Artifactory/TeamCity.

File description stored on Artifactory

artifactory:<build number>:<target (Linux|Windows10|MacOSX)>:<filename>

Example

artifactory:1.2-dev-7942:Windows10:nativeReport.json

File description stored on TeamCity

 teamcity:<build locator>:<filename>

Example

 teamcity:id:42491947:nativeReport.json

Pay attention, user and password information(with flag -u <username>:<password>) should be provided to get data from TeamCity.

By default analyzing tool splits benchmarks into stable and unstable taking information from database. If you have no connection to inner network please use -f flag.

./benchmarksAnalyzer.kexe -f <file1> <file2>

LLVM

See BUILDING_LLVM.md if you want to build and use your own LLVM distribution instead of provided one.

Using different LLVM distributions as part of Kotlin/Native compilation pipeline.

-Xllvm-variant compiler option allows to choose which LLVM distribution should be used during compilation. The following values are supported:

  • user — The compiler downloads (if necessary) and uses small LLVM distribution that contains only necessary tools. This is what compiler does by default.
  • dev — The compiler downloads (if necessary) and uses large LLVM distribution that contains additional development tools like llvm-nm, opt, etc.
  • <absolute path> — Use local distribution of LLVM.

Playing with compilation pipeline.

The following compiler phases control different parts of LLVM pipeline:

  1. LinkBitcodeDependencies. Linkage of produced bitcode with runtime and some other dependencies.
  2. Running different parts of LLVM optimization pipeline:
    1. MandatoryBitcodeLLVMPostprocessingPhase: important postprocessing. Disabling can break generated code.
    2. ModuleBitcodeOptimization: Basic optimization pipeline. Something close to clang -O3
    3. LTOBitcodeOptimization: LTO pipeline. Slower, but better optimizations, assuming whole program knowledge.
  3. ObjectFiles. Compilation of bitcode with Clang.

For example, pass -Xdisable-phases=LTOBitcodeOptimization to skip this part of optimization pipeline for faster compilation with slower code. Note that disabling LinkBitcodeDependencies or ObjectFiles will break compilation pipeline.

Compiler takes options for Clang from konan.properties file by combining clangFlags.<TARGET> and clang<Noopt/Opt/Debug>Flags.<TARGET> properties. Use -Xoverride-konan-properties=<key_1=value_1; ...;key_n=value_n> flag to override default values.

Please note:

  1. Kotlin Native passes bitcode files to Clang instead of C or C++, so many flags won't work.
  2. -cc1 -emit-obj should be passed because Kotlin/Native calls linker by itself.
  3. Use clang -cc1 -help to see a list of available options.

Example: replace predefined LLVM pipeline with Clang options.

CLANG_FLAGS="clangFlags.macos_x64=-cc1 -emit-obj;clangNooptFlags.macos_x64=-O2"
kotlinc-native main.kt -Xdisable-phases=MandatoryBitcodeLLVMPostprocessingPhase,ModuleBitcodeOptimization,LTOBitcodeOptimization -Xoverride-konan-properties="$CLANG_FLAGS"

Dumping LLVM IR

It is possible to dump LLVM IR after a particular compiler phase.

kotlinc-native main.kt -Xsave-llvm-ir-after=<PhaseName> -Xsave-llvm-ir-directory=<PATH>

<PATH>/out.<PhaseName>.ll will contain LLVM IR after given phase.

Passing Codegen phase allows to get LLVM IR right after translation from Kotlin Backend IR, and BitcodeOptimization phase allows to see the result of LLVM optimization pipeline. The list of phases that support LLVM IR dumping is constantly changing, so check out compiler sources if you want to get the full list of such phases.

Running Clang the same way Kotlin/Native compiler does

Kotlin/Native compiler (including cinterop tool) has machinery that manages LLVM, Clang and native SDKs for supported targets and runs bundled Clang with proper arguments. To utilize this machinery, use $dist/bin/run_konan clang $tool $target $arguments, e.g.

$dist/bin/run_konan clang clang ios_arm64 1.c

will print and run the following command:

~/.konan/dependencies/clang-llvm-apple-8.0.0-darwin-macos/bin/clang \
    -B/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin \
    -fno-stack-protector -stdlib=libc++ -arch arm64 \
    -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS13.5.sdk \
    -miphoneos-version-min=9.0 1.c

The similar helper is available for LLVM tools, $dist/bin/run_konan llvm $tool $arguments.