Packages

class GpuTransitionOverrides extends Rule[SparkPlan]

Rules that run after the row to columnar and columnar to row transitions have been inserted. These rules insert transitions to and from the GPU, and then optimize various transitions.

Linear Supertypes
Rule[SparkPlan], Logging, SQLConfHelper, AnyRef, Any
Ordering
  1. Alphabetic
  2. By Inheritance
Inherited
  1. GpuTransitionOverrides
  2. Rule
  3. Logging
  4. SQLConfHelper
  5. AnyRef
  6. Any
  1. Hide All
  2. Show All
Visibility
  1. Public
  2. All

Instance Constructors

  1. new GpuTransitionOverrides()

Value Members

  1. final def !=(arg0: Any): Boolean
    Definition Classes
    AnyRef → Any
  2. final def ##(): Int
    Definition Classes
    AnyRef → Any
  3. final def ==(arg0: Any): Boolean
    Definition Classes
    AnyRef → Any
  4. def apply(sparkPlan: SparkPlan): SparkPlan
    Definition Classes
    GpuTransitionOverrides → Rule
  5. final def asInstanceOf[T0]: T0
    Definition Classes
    Any
  6. def assertIsOnTheGpu(plan: SparkPlan, conf: RapidsConf): Unit
  7. def assertIsOnTheGpu(exp: Expression, conf: RapidsConf): Unit
  8. def clone(): AnyRef
    Attributes
    protected[lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( ... ) @native()
  9. def conf: SQLConf
    Definition Classes
    SQLConfHelper
  10. def detectAndTagFinalColumnarOutput(plan: SparkPlan): SparkPlan
  11. final def eq(arg0: AnyRef): Boolean
    Definition Classes
    AnyRef
  12. def equals(arg0: Any): Boolean
    Definition Classes
    AnyRef → Any
  13. def extractAttrReferences[A <: Expression](expression: A): Seq[AttributeReference]
  14. def finalize(): Unit
    Attributes
    protected[lang]
    Definition Classes
    AnyRef
    Annotations
    @throws( classOf[java.lang.Throwable] )
  15. def fixupHostColumnarTransitions(plan: SparkPlan): SparkPlan

    Fixes up instances of HostColumnarToGpu that are operating on nested types.

    Fixes up instances of HostColumnarToGpu that are operating on nested types. There are no batch methods to access nested types in Spark's ColumnVector, and as such HostColumnarToGpu does not support nested types due to the performance problem. If there's nested types involved, use a CPU columnar to row transition followed by a GPU row to columnar transition which is a more optimized code path for these types. This is done as a fixup pass since there are earlier transition optimizations that are looking for HostColumnarToGpu when optimizing transitions.

  16. final def getClass(): Class[_]
    Definition Classes
    AnyRef → Any
    Annotations
    @native()
  17. def hashCode(): Int
    Definition Classes
    AnyRef → Any
    Annotations
    @native()
  18. def initializeLogIfNecessary(isInterpreter: Boolean, silent: Boolean): Boolean
    Attributes
    protected
    Definition Classes
    Logging
  19. def initializeLogIfNecessary(isInterpreter: Boolean): Unit
    Attributes
    protected
    Definition Classes
    Logging
  20. final def isInstanceOf[T0]: Boolean
    Definition Classes
    Any
  21. def isTraceEnabled(): Boolean
    Attributes
    protected
    Definition Classes
    Logging
  22. def log: Logger
    Attributes
    protected
    Definition Classes
    Logging
  23. def logDebug(msg: ⇒ String, throwable: Throwable): Unit
    Attributes
    protected
    Definition Classes
    Logging
  24. def logDebug(msg: ⇒ String): Unit
    Attributes
    protected
    Definition Classes
    Logging
  25. def logError(msg: ⇒ String, throwable: Throwable): Unit
    Attributes
    protected
    Definition Classes
    Logging
  26. def logError(msg: ⇒ String): Unit
    Attributes
    protected
    Definition Classes
    Logging
  27. def logInfo(msg: ⇒ String, throwable: Throwable): Unit
    Attributes
    protected
    Definition Classes
    Logging
  28. def logInfo(msg: ⇒ String): Unit
    Attributes
    protected
    Definition Classes
    Logging
  29. def logName: String
    Attributes
    protected
    Definition Classes
    Logging
  30. def logTrace(msg: ⇒ String, throwable: Throwable): Unit
    Attributes
    protected
    Definition Classes
    Logging
  31. def logTrace(msg: ⇒ String): Unit
    Attributes
    protected
    Definition Classes
    Logging
  32. def logWarning(msg: ⇒ String, throwable: Throwable): Unit
    Attributes
    protected
    Definition Classes
    Logging
  33. def logWarning(msg: ⇒ String): Unit
    Attributes
    protected
    Definition Classes
    Logging
  34. final def ne(arg0: AnyRef): Boolean
    Definition Classes
    AnyRef
  35. final def notify(): Unit
    Definition Classes
    AnyRef
    Annotations
    @native()
  36. final def notifyAll(): Unit
    Definition Classes
    AnyRef
    Annotations
    @native()
  37. def optimizeAdaptiveTransitions(plan: SparkPlan, parent: Option[SparkPlan]): SparkPlan
  38. def optimizeCoalesce(plan: SparkPlan): SparkPlan

    This optimizes the plan to remove GpuCoalesceBatches nodes that are unnecessary or undesired in some situations.

    This optimizes the plan to remove GpuCoalesceBatches nodes that are unnecessary or undesired in some situations.

    Note

    This does not examine GpuShuffleCoalesceExec nodes in the plan, as they are always required after GPU columnar exchanges during normal shuffle to place the data after shuffle on the GPU. Those nodes also do not coalesce to the same goal as used by GpuCoalesceBatches, so a GpuShuffleCoalesceExec immediately followed by a GpuCoalesceBatches is not unusual.

  39. def optimizeGpuPlanTransitions(plan: SparkPlan): SparkPlan
  40. var rapidsConf: RapidsConf
  41. lazy val ruleId: RuleId
    Attributes
    protected
    Definition Classes
    Rule
  42. val ruleName: String
    Definition Classes
    Rule
  43. def shuffledHashJoinOptimizeShuffle(plan: SparkPlan): SparkPlan

    Removes GpuCoalesceBatches(GpuShuffleCoalesceExec(build side)) from either side of a GpuShuffledInnerHashJoinExec since that node handles shuffled data directly.

    Removes GpuCoalesceBatches(GpuShuffleCoalesceExec(build side)) from either side of a GpuShuffledInnerHashJoinExec since that node handles shuffled data directly. For other joins, it removes it for for the build side. The coalesce logic has been moved to the GpuShuffleCoalesceExec class, and is handled differently to prevent holding onto the GPU semaphore for stream IO.

  44. final def synchronized[T0](arg0: ⇒ T0): T0
    Definition Classes
    AnyRef
  45. def toString(): String
    Definition Classes
    AnyRef → Any
  46. final def wait(): Unit
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )
  47. final def wait(arg0: Long, arg1: Int): Unit
    Definition Classes
    AnyRef
    Annotations
    @throws( ... )
  48. final def wait(arg0: Long): Unit
    Definition Classes
    AnyRef
    Annotations
    @throws( ... ) @native()

Inherited from Rule[SparkPlan]

Inherited from Logging

Inherited from SQLConfHelper

Inherited from AnyRef

Inherited from Any

Ungrouped