Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

Fix and optimize handling of vectorized memory accesses #17767

Merged
merged 41 commits into from
Apr 17, 2020
Merged

Fix and optimize handling of vectorized memory accesses #17767

merged 41 commits into from
Apr 17, 2020

Commits on Mar 5, 2020

  1. Vectorized loads for binary elemwise kernel

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    2fe0eaf View commit details
    Browse the repository at this point in the history

  2. More generalization

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    6b89506 View commit details
    Browse the repository at this point in the history

  3. Add backwardusenone

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    37d81c8 View commit details
    Browse the repository at this point in the history

  4. Remove the unused _backward_add op

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    f86da86 View commit details
    Browse the repository at this point in the history

  5. Add vectorized backwardusein

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    ea56552 View commit details
    Browse the repository at this point in the history

  6. Extending vectorization to more binary ops, binary ops with scalar and 

    unary ops
    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    ec08749 View commit details
    Browse the repository at this point in the history

  7. Handling ElementwiseSum

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    28e5877 View commit details
    Browse the repository at this point in the history

  8. Get rid of half2 in mshadow

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    541aebb View commit details
    Browse the repository at this point in the history

  9. Remove backward_elemwiseaddex

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    7729114 View commit details
    Browse the repository at this point in the history

  10. Revert "Remove the unused _backward_add op" 

    This reverts commit f86da86.
    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    8455c0d View commit details
    Browse the repository at this point in the history

  11. Revert "Remove backward_elemwiseaddex" 

    This reverts commit 7729114.
    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    402bb59 View commit details
    Browse the repository at this point in the history

  12. Add back the backward_add since C++ test relies on it

    @ptrendx
    ptrendx committed Mar 5, 2020
    Configuration menu
    Copy the full SHA
    716aa1a View commit details
    Browse the repository at this point in the history

Commits on Mar 17, 2020

  1. Test bcast implementations

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    948cea1 View commit details
    Browse the repository at this point in the history

  2. First version of vecotrized bcast

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    f326f7e View commit details
    Browse the repository at this point in the history

  3. Adding single side vectorized bcast kernel

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    85f6070 View commit details
    Browse the repository at this point in the history

  4. Removing debug prints

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    ed8d745 View commit details
    Browse the repository at this point in the history

  5. Actually run the single side kernel

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    3d84675 View commit details
    Browse the repository at this point in the history

  6. Move the default implementation of bcast to the vectorized one

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    3227476 View commit details
    Browse the repository at this point in the history

  7. Limit the new implementation to GPU only

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    2017f75 View commit details
    Browse the repository at this point in the history

  8. Enabling vectorization when broadcast does not actually do broadcast

    @ptrendx
    ptrendx committed Mar 17, 2020
    Configuration menu
    Copy the full SHA
    320e91a View commit details
    Browse the repository at this point in the history

Commits on Mar 18, 2020

  1. Cleaning

    @ptrendx
    ptrendx committed Mar 18, 2020
    Configuration menu
    Copy the full SHA
    4decacd View commit details
    Browse the repository at this point in the history

  2. Cleaning part 2

    @ptrendx
    ptrendx committed Mar 18, 2020
    Configuration menu
    Copy the full SHA
    a16cec0 View commit details
    Browse the repository at this point in the history

  3. Fix for numpy ops using stuff from broadcast

    @ptrendx
    ptrendx committed Mar 18, 2020
    Configuration menu
    Copy the full SHA
    ff2243d View commit details
    Browse the repository at this point in the history

  4. Fix

    @ptrendx
    ptrendx committed Mar 18, 2020
    Configuration menu
    Copy the full SHA
    ecbdc6d View commit details
    Browse the repository at this point in the history

  5. Fix lint

    @ptrendx
    ptrendx committed Mar 18, 2020
    Configuration menu
    Copy the full SHA
    2592e53 View commit details
    Browse the repository at this point in the history

Commits on Apr 3, 2020

  1. Merge branch 'upstream' into pr_vectorized_loads

    @ptrendx
    ptrendx committed Apr 3, 2020
    Configuration menu
    Copy the full SHA
    0136fdd View commit details
    Browse the repository at this point in the history

Commits on Apr 10, 2020

  1. Merge branch 'upstream' into pr_vectorized_loads

    @ptrendx
    ptrendx committed Apr 10, 2020
    Configuration menu
    Copy the full SHA
    51ed0bb View commit details
    Browse the repository at this point in the history

  2. Try to debug pinv numpy test

    @ptrendx
    ptrendx committed Apr 10, 2020
    Configuration menu
    Copy the full SHA
    78d25a9 View commit details
    Browse the repository at this point in the history

Commits on Apr 13, 2020

  1. Merge branch 'upstream' into pr_vectorized_loads

    @ptrendx
    ptrendx committed Apr 13, 2020
    Configuration menu
    Copy the full SHA
    29d8cb2 View commit details
    Browse the repository at this point in the history

Commits on Apr 14, 2020

  1. Fix

    @ptrendx
    ptrendx committed Apr 14, 2020
    Configuration menu
    Copy the full SHA
    a801f8b View commit details
    Browse the repository at this point in the history

Commits on Apr 17, 2020

  1. Fix the vectorized broadcast implementation for misaligned input 

    pointers
    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    810f8c8 View commit details
    Browse the repository at this point in the history

  2. Added tests

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    f5f5d3e View commit details
    Browse the repository at this point in the history

  3. Added docs to cuda_vectorization.cuh

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    4120fe8 View commit details
    Browse the repository at this point in the history

  4. Merge branch 'upstream' into pr_vectorized_loads

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    0e38e81 View commit details
    Browse the repository at this point in the history

  5. Another fix for broadcast and fix INT64 compilation

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    c1a734a View commit details
    Browse the repository at this point in the history

  6. Optimize for aligned=true

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    de6125e View commit details
    Browse the repository at this point in the history

  7. 1 more addition to test

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    9129ba2 View commit details
    Browse the repository at this point in the history

  8. Reverting the change to Numpy op test

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    43f4b4e View commit details
    Browse the repository at this point in the history

  9. Trying mcmodel=medium to fix the failure in CMake static build

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    1af684c View commit details
    Browse the repository at this point in the history

  10. Revert "Trying mcmodel=medium to fix the failure in CMake static build" 

    This reverts commit 1af684c.
    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    31c84b1 View commit details
    Browse the repository at this point in the history

  11. Limiting the PR to just elementwise ops

    @ptrendx
    ptrendx committed Apr 17, 2020
    Configuration menu
    Copy the full SHA
    b9d1760 View commit details
    Browse the repository at this point in the history