GitLab

[Feature][Performance] Implement NCCL wrapper for communicating NodeEmbeddings and sparse gradients. (#2825)

* Split NCCL wrapper from sparse optimizer and sparse embedding

* Add more unit tests for single node nccl

* Fix unit test for tf

* Switch to device histogram

* Fix histgram issues

* Finish migration to histogram

* Handle cases with zero send/recieve data

* Start on partition object

* Get compiling

* Updates

* Add unit tests

* Switch to partition object

* Fix linting issues

* Rename partition file

* Add python doc

* Fix python assert and finish doxygen comments

* Remove stubs for range based partition to satisfy pylint

* Wrap unit test in GPU only

* Wrap explicit cuda call in ifdef

* Merge with partition.py

* update docstrings

* Cleanup partition_op

* Add Workspace object

* Switch to using workspace object

* Move last remainder based function out of nccl_api

* Add error messages

* Update docs with examples

* Fix linting erros
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

* test commit

* fixes

* oops

* add docs

* lint

* why does it say I have a trailing whitespace

* oh ok

* fixes

* why there's an invalid argument error

* address comments

* fix

* address comments

Co-authored-by: Ubuntu <ubuntu@ip-172-31-1-191.ec2.internal>
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

* init cuda support

* cuSPARSE err

* passed unittest for csr_mm/SpGEMM. int64 not supported

* Debugging cuSPARSE error 3

* csrgeam only supports int32?

* disabling int64 for cuda

* refactor and add CSRMask

* lint

* oops

* remove todo

* rewrite CSRMask with CSRGetData

* lint

* fix test

* address comments

* lint

* fix

* addresses comments and rename BUG_ON
Co-authored-by: Israt Nisa <nisisrat@amazon.com>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-30-71.ec2.internal>
Co-authored-by: Quan Gan <coin2028@hotmail.com>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* Add row/col sorted flags

* improve sorting paths

* Remove print statement

* Keep track of sorted matrices

* Remove sort check in to_block

* Improve CPU sorted COO->CSR

* Handle the zero edge case

* Remove omp default clause to work with MSVC

* Update comments on sorted COO->CSR cpu implementatoin

* Expose sorted to python interface

* Make check_sorted default to false for dgl.graph()

* remove check sorted; add utests

* remove check_sorted flag
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* Start on uniform GPU sampling

* Save more work

* Get cu file compiling

* Update sampling

* More changes

* Get GPU sampling for uniform probabilities solved

* Fix batch tensor migration

* Fix

* update kernels

* expand blocking

* Undo testing change

* Cut down on sampling overhead

* Fix replacement

* Update unit tests

* Add option to gpu sample in graphsage

* Copy only csc to gpu

* Add ogbn support

* Fix linting

* Remove nvtx from sample

* Improve documentation and error checking

* Expand documentation

* Update assert checking

* delete extra space

* Use standard dataloader when dataset is a dictionary

* ogb -> ogbn

* Fix edge selection determinism

* Fix typos

* Remove nvtx

* Add comment for self.fanout_arrays and assert

* Fix linting

* Migrate to scalarbatcher

* Fix indentation

* Fix batcher

* Fix indexing

* Only use databatcher for GPU

* Convert to DGL NDArray to PyTorch Tensor

* Add optimization for PyTorch's F.tensor() for list of GPU tensors
Co-authored-by: Da Zheng <zhengda1936@gmail.com>

* disable cpu fp16

* spell mistakes

* test

* more stuff

* add test

* fixes

* optimize algo

* replace unordered_map with arrays

* lint

* lint x2

* oops

* disable gpu csrmm tests

* remove gpu invocation

* optimize with openmp

* remove python functions

* add back with docstrings

* lint

* lint

* update python interface

* functionize

* functionize

* lint

* lint

[Bugfix] Wrap cub with CUB_NS_PREFIX and remove dependency on Thrust to linking issues with Torch 1.8 (#2758)

* Wrap cub with prefixes and remove thrust

* Using counting iterator
Co-authored-by: Zihao Ye <expye@outlook.com>
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* finish graph matching gpu version

* use C++ shuffle

* finish graph matching

* fix bug

* fix bug

* change name and use swap

* upt

* fix format problem

* fix format problem

* stronger test

* upt

* upt

* change python api

* upt

* upt

* format check

* upt

* upt

* fix bug
Co-authored-by: Tong He <hetong007@gmail.com>

* upd

* fix

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* fix

* refactor

* upd test

* large feat_len or n in segment reduce

* lint

* add tvm as submodule

* compilation is ok but calling fails

* can call now

* pack multiple modules, change names

* upd

* upd

* upd

* fix cmake

* upd

* upd

* upd

* upd

* fix

* relative path

* upd

* upd

* upd

* singleton

* upd

* trigger

* fix

* upd

* count reducible

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* only keep related files

* upd

* upd

* upd

* upd

* lint

* lint

* lint

* lint

* pylint

* upd

* upd

* compilation

* fix

* upd

* upd

* upd

* upd

* upd

* upd

* upd doc

* refactor

* fix

* upd number
Co-authored-by: Zhi Lin <linzhilynn@gmail.com>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-42-78.us-east-2.compute.internal>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-21-156.us-east-2.compute.internal>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>

* Remove double-checking sorted

* Remove sorting of CSR by default

* Update unit test to use unsorted matix

* delete whitespace

* Expand unit tests

* Replace cusparse sort

* Fix row column sorting

* Explicitly don't sort columns

* Fix linting errors

* Fix bit-width calculation

* Fix sorting assertion and unit test

* Fix linting

* Improve CPU COO2CSR

* Remove references

* Rename and add documentation to edge encoding/decoding funcionts

* Fix sorting keys as 64 bit

* Revert cosmetic changes to unit tests

* Update documentation

* Update complexity documentation for coo to csr conversion

* Remove COOIsSorted check in CPU implementation too

* upd

* upd

* upd

* upd

* fix

* upd

* upd
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>

Co-authored-by: Zihao Ye <expye@outlook.com>

* first commit

* some thoughts

* move around

* more commit

* more fixes

* now it uses torch allocator

* fix symbol export error

* fix

* fixes

* test fix

* add script

* building separate library per version

* fix for vs2019

* more fixes

* fix on windows build

* update jenkinsfile

* auto copy built dlls for windows

* lint and installation guide update

* fix

* specify conda environment

* set environment for ci

* fix

* fix

* fix

* fix again

* revert

* fix cmake

* fix

* switch to using python interpreter path

* remove scripts

* debug

* oops sorry

* Update index.rst

* Update index.rst

* copies automatically, no need for this

* do not print message if library not found

* tiny fixes

* debug on nightly

* replace add_compile_definitions to make CMake 3.5 happy

* fix linking to wrong lib for multiple pytorch envs

* changed building strategy

* fix nightly

* fix windows

* fix windows again

* setup bugfix

* address comments

* change README

* upd cmake

* upd

* format

* disable openmp in arithmetic operation

* lint

* Update array_op_impl.cc
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* upd

* upd

* update

* upd

* upd

* upd

* fix

* lint

* lint

* pylint

* doc

* support AVX512

* env DGL_CPU_INTEL_KERNEL_ENABLED=1

* env DGL_CPU_INTEL_KERNEL_LOG=1

* Add unittest test_spmm.cc
Co-authored-by: Izabela Mazur <izabela.mazur@intel.com>
Co-authored-by: Michal Szarmach <michal.szarmach@intel.com>

Review patch

* multiple fixes

* fix CI

* fiddle

* revert stubs

* remove stubs

* poke

* remove linking of driver library

* minor
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* multiple fixes

* fix CI

* fiddle

* revert stubs

* upd

* upd

* unmerge

* unmerge
Co-authored-by: Quan Gan <coin2028@hotmail.com>

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* fix

* upd

* upd

* upd

* upd

* fix cumsum

* udp
Co-authored-by: Zihao <expye@outlook.com>

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* fix

* upd

* upd

* upd

* upd

* fix

* upd
Co-authored-by: VoVAllen <jz1749@nyu.edu>

* upd

* upd

* upd

* upd

* lint

* upd

* upd

* fmt
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* upd

* test idea

* cuda kernels

* lint and fixes

* lint

* change to another strategy

* use infinity

* fix
Co-authored-by: Zihao Ye <expye@outlook.com>

* up

* pylint

* upd

[bugfix] Fix the memory leak issue of Cluster GAT under 0.5 kernel and simplify the bipartite GAT. (#1908)

* uipd

* upd

* upd

* upd

* upd

* COO sort

* COOToCSR

* CSR2COO

* CSRSort; CSRTranspose

* pass all CSR tests

* lint

* remove int32 conversion

* fix tensorflow nn tests

* turn on CI

* fix

* addreess comments