Skip to content

AVX-512 Support for Vector API #6268

Open

Description

@BradleyWood

This issue will track the AVX-512 implementation work for the vector API. AVX-512 is needed to support 512-bit vector lengths on x86 architecture.

Status & Todo

  • Add support for AVX512 #6463
    • ✓ Extended register file to include xmm/ymm/zmm 0 - 31 (Not included in Add support for AVX512 #6463)
    • ✓ EVEX instruction encoding,
    • ✓ Mask register support (TR_VMR) (Not included in Add support for AVX512 #6463)
      • Register Assignment
      • Mask tracing aware of new registers
    • AVX-512 Instructions
      • ✓ vmovdqu/vmovdqu8/vmovdqu16/vmovdqu32/vmovdqu64, vmovdqa/vmovdqa8/vmovdqa16/vmovdqa32/vmovdqa64
      • ✓ kmovw/kmovq/kmovd/kmovb (Not included in Add support for AVX512 #6463)
      • Encoding tests (to be included after Commonize binary encoding tests and add support to x86  #6216 is merged)
      • Other instructions to be added as needed by vector IL evaluators
        • broadcast
        • masking register operations (and, or, not, add),
        • Additive, multiplicative, logical, instructions
  • Todo
    • Add new instruction kinds for masked operations,
      • Needed for EVEX encoded instructions with vector write mask
      • Make traces aware of new instruction kinds
    • New vector IL opcodes
      • Need to be implemented with 256, 512 bit vector lengths, depending on target CPU
      • Add appropriate AVX instructions where required

Activity

  1. BradleyWood commented on Dec 8, 2021

    @BradleyWood
    ContributorAuthor
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions