This repository has been archived by the owner on Nov 17, 2023.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) #18095
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hey @ptrendx , Thanks for submitting the PR
CI supported jobs: [centos-gpu, unix-cpu, website, edge, centos-cpu, miscellaneous, sanity, unix-gpu, clang, windows-gpu, windows-cpu] Note: |
DickJC123
approved these changes
Apr 18, 2020
* Vectorized loads for binary elemwise kernel * More generalization * Add backwardusenone * Remove the unused _backward_add op * Add vectorized backwardusein * Extending vectorization to more binary ops, binary ops with scalar and unary ops * Handling ElementwiseSum * Get rid of half2 in mshadow * Remove backward_elemwiseaddex * Revert "Remove the unused _backward_add op" This reverts commit f86da86. * Revert "Remove backward_elemwiseaddex" This reverts commit 7729114. * Add back the backward_add since C++ test relies on it * Test bcast implementations * First version of vecotrized bcast * Adding single side vectorized bcast kernel * Removing debug prints * Actually run the single side kernel * Move the default implementation of bcast to the vectorized one * Limit the new implementation to GPU only * Enabling vectorization when broadcast does not actually do broadcast * Cleaning * Cleaning part 2 * Fix for numpy ops using stuff from broadcast * Fix * Fix lint * Try to debug pinv numpy test * Fix * Fix the vectorized broadcast implementation for misaligned input pointers * Added tests * Added docs to cuda_vectorization.cuh * Another fix for broadcast and fix INT64 compilation * Optimize for aligned=true * 1 more addition to test * Reverting the change to Numpy op test * Trying mcmodel=medium to fix the failure in CMake static build * Revert "Trying mcmodel=medium to fix the failure in CMake static build" This reverts commit 1af684c. * Limiting the PR to just elementwise ops
ptrendx
force-pushed
the
pr_vectorized_loads_1.x
branch
from
ad91c46
to
c5893e3
Compare
|
Adding this PR to 1.7.0 roadmap #16864. |
|
@mxnet-bot run ci [windows-gpu] |
|
Jenkins CI successfully triggered : [windows-gpu] |
|
@ciyongch This PR is now merged to v1.x branch, please cherry pick it to v1.7. |
|
@ptrendx as the v1.7.x is already rebased, could you please submit it to v1.7.x, then we can get it merged once the CI pass? Thanks. |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Vectorized loads for binary elemwise kernel
More generalization
Add backwardusenone
Remove the unused _backward_add op
Add vectorized backwardusein
Extending vectorization to more binary ops, binary ops with scalar and
unary ops
Handling ElementwiseSum
Get rid of half2 in mshadow
Remove backward_elemwiseaddex
Revert "Remove the unused _backward_add op"
This reverts commit f86da86.
This reverts commit 7729114.
Add back the backward_add since C++ test relies on it
Test bcast implementations
First version of vecotrized bcast
Adding single side vectorized bcast kernel
Removing debug prints
Actually run the single side kernel
Move the default implementation of bcast to the vectorized one
Limit the new implementation to GPU only
Enabling vectorization when broadcast does not actually do broadcast
Cleaning
Cleaning part 2
Fix for numpy ops using stuff from broadcast
Fix
Fix lint
Try to debug pinv numpy test
Fix
Fix the vectorized broadcast implementation for misaligned input
pointers
Added tests
Added docs to cuda_vectorization.cuh
Another fix for broadcast and fix INT64 compilation
Optimize for aligned=true
1 more addition to test
Reverting the change to Numpy op test
Trying mcmodel=medium to fix the failure in CMake static build
Revert "Trying mcmodel=medium to fix the failure in CMake static build"
This reverts commit 1af684c.
Description
Cherry pick #17767 to v1.x branch. @ciyongch