Skip to content

Latest commit

聽

History

History
88 lines (77 loc) 路 6.38 KB

CHANGELOG.md

File metadata and controls

88 lines (77 loc) 路 6.38 KB
Raw

Compiler Changelog

This changelog summarizes newly introduced optimizations and other compiler related changes.

GraalVM for JDK 23 (Internal Version 24.1.0)

  • (GR-50352): Added -Djdk.graal.PrintPropertiesAll to make -XX:+JVMCIPrintProperties show all Graal options.
  • (GR-25968): New optimization for reducing code size on AMD64, by emitting smaller jump instructions if the displacement fits in one byte. Enabled for Native Image O1-O3 per default; disabled elsewhere. Use -Djdk.graal.OptimizeLongJumps=true to enable.
  • (GR-45919): Added support for Generational ZGC (JEP 439).

GraalVM for JDK 22 (Internal Version 24.0.0)

  • (GR-49876): Added -Dgraal.PrintIntrinsics=true to log the intrinsics used by Graal in the current runtime.
  • (GR-49960): The Graal options now use the jdk.graal. prefix (e.g. -Djdk.graal.PrintCompilation=true). The legacy graal. prefix is deprecated but still supported (e.g. -Dgraal.PrintCompilation=true).
  • (GR-49610): The Graal module has been renamed from jdk.internal.vm.compiler to jdk.graal.compiler. Likewise, the compiler packages moved into the jdk.graal.compiler namespace. These renamings were done in preparation for Project Galahad.
  • (GR-20827): Extend endbranch support: Add endbranch CFI landing pad markers to exception targets. Ensure that LIR insertion buffers do not move existing endbranches on basic blocks. Extend the AMD64MacroAssembler with a PostCallAction that is performed after a call is emitted.

GraalVM for JDK 21 (Internal Version 23.1.0)

  • (GR-43228): Enforce backward-edge control-flow integrity (CFI) on aarch64 based on the UseBranchProtection JVM flag.

GraalVM for JDK 17 and GraalVM for JDK 20 (Internal Version 23.0.0)

  • (GR-42212): Remove support for all JDKs earlier than JDK 17.
  • (GR-42044): Improved output of -Dgraal.ShowConfiguration=info. For example: Using "Graal Community compiler" loaded from a Native Image shared library instead of: Using compiler configuration 'community' provided by org.graalvm.compiler.hotspot.CommunityCompilerConfigurationFactory loaded from JVMCI native library.
  • (GR-42145): The periodic dumping of benchmark counters enabled by -Dgraal.TimedDynamicCounters is now limited to jargraal.
  • (GR-31578): Novel Optimization Log: Unified interface to log and dump (e.g. via JSON) optimization decisions. Optimization phases should use the OptimizationLog to log transformations. Read more in OptimizationLog.md and read Profdiff.md to learn how to compare performed optimizations in hot compilations of 2 experiments.
  • The -Dgraal.PrintCompilation=true output now includes stub compilations. For example: StubCompilation-57 <stub> exceptionHandler (Object,Word)void | 166us 0B bytecodes 88B codesize 137kB allocated
  • (GR-27475) Add support for the ZGC collector on HotSpot.

Version 22.3.0

  • (GR-19840): An image produced by GraalVM's jlink now includes and uses libgraal by default and its java -version output includes GraalVM branding.
  • (GR-32382): Added a dedicated Native Image GC policy for libgraal that will adjust the eden space aggressively to minimize RSS memory usage.
  • (GR-38950): Removed deprecated JMX HotSpotGraalRuntime management bean from both libgraal and jargraal.

Version 22.2.0

  • (GR-23737): New global value numbering optimization for fixed nodes early in the compilation pipeline. Early global value numbering and loop invariant code motion is enabled per default. Disable early global value numbering with -Dgraal.EarlyGVN=false. Disable early loop invariant code motion with -Dgraal.EarlyLICM=false.
  • (GR-16452): Compute the code emission basic block order after backend control flow optimizations.
  • (GR-35033): Enable floating and global value numbering of division nodes early on in the compilation pipeline if it is known they will not trap.
  • (GR-38405): Compute all unswitchable invariant then pick the most frequent one.
  • (GR-38857): Deprecated and disabled the JMX HotSpotGraalRuntime management bean. Re-enable the HotSpotGraalRuntime management bean with -Dgraal.LibGraalManagementDelay=0.

Version 22.1.0

  • (GR-36751): Removed the DuplicateIrreducibleLoops option. To disable irreducible loop handling, set -Dgraal.MaxDuplicationFactor to a value less than or equal to 1. For AOT compilations, the effort spent to handle irreducible loops is boosted to let Native Image support more programs with irreducible loops.

Version 22.0.0

  • (GR-22707) (GR-30838): New, inner loops first, reverse post order and loop frequency calculations for the compiler.

Version 21.2.0

  • (GR-29770) Loop safepoint elimination: Not only consider 32bit loops for safepoint removal but also 64bit ones that iterate in 32bit ranges.
  • (GR-29341) AVX-512 support: Fix EVEX encoding and feature checks for existing instructions and add AVX-512 alternatives.
  • (GR-31162) Do not de-duplicate ValueAnchorNode. As part of this change, there is a new marker interface NodeWithIdentity to mark nodes that have identity.
  • (GR-8974) Speculative guard movement: An optimization that tries to move a loop invariant guard (e.g., an array bounds check) inside a loop to outside of the loop. Disable with -Dgraal.SpeculativeGuardMovement=false. Included in this change is enhanced output for -Dgraal.ShowConfiguration=verbose in terms of showing the compilation phases of a compiler configuration.
  • (GR-31031) Add intrinsic for Reference.refersTo and PhantomReference.refersTo.
  • (GR-31059) Permit unusually structured locking in Kotlin coroutines.

Version 21.1.0

  • (GR-29126) Unify box optimizations in the compiler. Remove -Dgraal.ReuseOutOfCacheBoxedValues=false.
  • (GR-28523) Optimize Box nodes: Optimizes box operations by re-using boxed representations if the value of the boxed primitive is outside of the cache range of the Int/Long/Short/Char caches. Box node optimization is enabled per default. Disable it with -Dgraal.ReuseOutOfCacheBoxedValues=false.
  • (GR-29373) Eliminate unneeded STORE_LOAD barriers on sequential volatile writes on x86. This improves ConcurrentHashMap performance.
  • (GR-29337) Volatile loads were losing type information about the underlying field, resulting in unneeded casts. This improves ConcurrentHashMap performance.
  • (GR-28956) Use more informative ProfileData objects instead of raw branch probabilities and loop frequencies.