Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8262356

Optimize existing masked operation support for AVX-512.

    XMLWordPrintable

Details

    • Enhancement
    • Resolution: Unresolved
    • P4
    • tbd
    • 17
    • hotspot
    • None

    Description

      - Currently a vector masked operation performs an operation over all the vector lanes followed by a blend operation which selectively updates the result vector under the influence of mask vector.

      - Prior to AVX-512 blending newly computed result with older value was the only way to facilitate masked/predicated vector operations.

      - A non-AVX-512 vector blend instruction probes the MSB bit for each mask vector lane in order to selectively choose between two source vector lanes.
       
      - With AVX-512 there are two ways in which masked operation can be performed as follows
      Method 1:
             vmask = vector_cmp(mask, ALL_ONES)
             vres = vector_operation vsrc1, vsrc2
             vector_blend(vdst, vres, vmask)
       
      Method 2:
            opmask = vector_cmp(mask, ALL_ONES)
            ves = vector_operation vsrc1, vsrc2, opmask

      Clearly emitting a predicated vector operation is much more optimal in terms of emitted code size and is energy efficient since a vector operation conditionally operates over portion of vectors.

      - VectorAPI has significantly extended to scope of masked operations, additionally it offer APIs to perform direct mask manipulation e.g. VectorMask.or/and/not. Thus a direct operation over an Opmask register will enable generating efficient code.

      - Using opmask register we can further optimized existing implementation for VectorMask querying operation like VectorMask.firstTrue/lastTrue/anyTrue/allTrue/trueCount.

      Attachments

        Issue Links

          relates to

          Sub-task - The sub-task of the issue JDK-8272359 X86: Backend support for optimizing vector masked operations over AVX512 target.

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Closed

          Bug - A problem which impairs or prevents the functions of the product. JDK-8274569 X86 backend related incorrectness issues in legacy store mask patterns

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8262982 [vector API] add IR and API points for synthetic multi-vectors

          • P3 - Major loss of function.
          • Open

          Enhancement - null JDK-8262983 [vector API] add IR and API points for synthetic partial vectors

          • P3 - Major loss of function.
          • Open

          Enhancement - null JDK-8262355 Support for AVX-512 opmask register allocation.

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8264563 Add masked vector intrinsics for binary/store operations

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8266621 Add masking support for unary/ternary vector intrinsics

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8270264 Add the masking support for vector lanewiseShift

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8273406 Optimize various masked vector operations for AVX512 target.

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8273949 Intrinsic creation for VectorMask.toLong operation.

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8266287 Basic mask IR implementation for the Vector API masking feature support

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8267368 Add masking support for reduction vector intrinsics

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8271273 Java API and IR changes for masked compare operation

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8271539 Add masking support for load/store from/into byte array/buffer

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8272100 VectorAPI: modify existing implementation of masked neg and not operation.

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8272479 Java API and IR changes for masked rearrange operation

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8272971 Intrinsification of VectorMask.cast operation for all compatible vector species

          • P4 - Minor loss of function, or other problem where easy workaround is present.
          • Resolved

          Enhancement - null JDK-8264954 unified handling for VectorMask object re-materialization during de-optimization

          • P3 - Major loss of function.
          • Closed

          Activity

            People

              jbhateja Jatin Bhateja
              jbhateja Jatin Bhateja
              Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

                Created:
                Updated: