Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add ROCm support #756

Closed
wants to merge 8 commits into from
Closed

Add ROCm support #756

wants to merge 8 commits into from
+159 −32

Commits on Jan 22, 2024

  1. Add ROCm make target 

    ROCM_TARGET=gfx1030 make hip
Uses define BITS_AND_BYTES_USE_ROCM to redefine cuda functions
to ROCm equivalent

credit to previous ports:

Co-authored-by: broncotc <i@broncotc.net>
Co-authored-by: agrocylo <130291676+agrocylo@users.noreply.github.com>
    @arlo-phoenix @broncotc @agrocylo
    3 people committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    5ee2ec1 View commit details
    Browse the repository at this point in the history

  2. Add ROCm support to python library 

    disables igemm for now and adds path to compiled library
libbitsandbytes_hip_nohipblaslt
    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    f9e2a84 View commit details
    Browse the repository at this point in the history

  3. Add ROCm information to README

    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    78eecb3 View commit details
    Browse the repository at this point in the history

  4. Makefile: Add fallback to /opt/rocm home

    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    940c52e View commit details
    Browse the repository at this point in the history

  5. make cut one less

    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    a485a02 View commit details
    Browse the repository at this point in the history

  6. Rename BITS_AND_BYTES_USE_ROCM to BNB_USE_HIP

    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    32cd5e0 View commit details
    Browse the repository at this point in the history

  7. Adjust kQuantizeBlockwise to work with WARP size 64

    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    e03a8bd View commit details
    Browse the repository at this point in the history

  8. Make sure DATA_NUM_PER_TH <= CUB_NUM_PER_TH 

    the unrolls already somehow worked correctly before, but they shouldn't have.
    @arlo-phoenix
    arlo-phoenix committed Jan 22, 2024
    Configuration menu
    Copy the full SHA
    ac20c05 View commit details
    Browse the repository at this point in the history