GitLab

* Add helper method for temporary affinitization of compute threads

* Rework DL affinitization as single helper

* Add example usage in benchmarks

* Fix python linter warnings

* Fix affinity helper params

* Use NUMA node 0 cores only by default

* Fix benchmarks

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

* Update nccl communicator for when NCCL is missing

* Use static_cast

* Add doc string

* Fix whitespace

* Resrtict unit test to GPU runs
Co-authored-by: Xin Yao <xiny@nvidia.com>

* * Added missing specializations for `__half` of `DLDataTypeTraits`, `IndexSelect`, `Full`, `Scatter_`, `CSRGetData`, `CSRMM`, `CSRSum`, `IndexSelectCPUFromGPU`
* Fixed casting issue in `_LinearSearchKernel` that was preventing it from supporting `__half`
* Added `#if`'d out specializations of `CSRGEMM`, `CSRGEAM`, and `Xgeam`, which would require functions that aren't currently provided by cublas

* * Added more specific error messages for unimplemented FP16 specializations of Xgeam, CSRGEMM, and CSRGEAM

* * Added missing instantiation of DLDataTypeTraits<__half>::dtype

* * Fixed linter error
* Added clearer comment explaining why the cast to long long is necessary

* * Worked around a compile error in some particular setup, where __half can't be constructed on the host side

* * Fixed linter formatting errors

* * Changes to comments as recommended

* * Made recommended changes to logging errors in FP16 specializations
* Also changed the existing Xgeam function for unsupported data types from LOG(INFO) to LOG(FATAL)

* [BugFix] fix rpc-related build issue on mac OS

* add warning message

* add warning message

* Explicitly unpin tensoradapter allocated arrays

* Undo unrelated change

* Add unit test

* update unit test

* add pinned_by_dgl flag to NDArray::Container

* use dgl.ndarray for holding the pinning status

* update multi-gpu uva inference

* reinterpret cast NDArray::Container* to DLTensor* in MoveAsDLTensor

* update unpin column and examples

* add unit test for unpin column
Co-authored-by: Dominique LaSalle <dlasalle@nvidia.com>
Co-authored-by: nv-dlasalle <63612878+nv-dlasalle@users.noreply.github.com>

* Wrap all CUDA runtime API/CUB calls with macro

* remove the usage of explicit cudaMalloc in favor of AllocWorkspace

* fix typo
Co-authored-by: Israt Nisa <neesha295@gmail.com>

* [ist] enable time out when fetching msg

* fix lint error

* minor refinements

* improve minor log

* fix dist test

* fix timeout issue in tensorpipe

Co-authored-by: nv-dlasalle <63612878+nv-dlasalle@users.noreply.github.com>
Co-authored-by: Israt Nisa <neesha295@gmail.com>

* change warning message

* Update tensordispatch.cc

This reverts commit fdd1fe19.

* Explicitly unpin tensoradapter allocated arrays

* Undo unrelated change

* Add unit test

* update unit test

[Examples] Update graphsage multi-gpu example to use mutliple GPUs for validation and testing. (#3827)

* Update graphsage multi-gpu example to use mutliple GPUs for validation and
testing.

* Remove argmax

* Fix rebase error

* Add more documentation to example and simplify

* Switch to name shared memory

* Add comment about how training is distributed

* Restore iteration count

* fix munmap error reporting for better error messages
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

* fixes

* fix

* more fixes

* update

* oops

* lint?

* temporarily revert - will fix in another PR

* more fixes

* skipping mxnet test

* address comments

* fix DDP

* fix edge dataloader exclusion problems

* stupid bug

* fix

* use_uvm option

* fix

* fixes

* fixes

* fixes

* fixes

* add evaluation for cluster gcn and ddp

* stupid bug again

* fixes

* move sanity checks to only support DGLGraphs

* pytorch lightning compatibility fixes

* remove

* poke

* more fixes

* fix

* fix

* disable test

* docstrings

* why is it getting a memory leak?

* fix

* update

* updates and temporarily disable forkingpickler

* update

* fix?

* fix?

* oops

* oops

* fix

* lint

* huh

* uh

* update

* fix

* made it memory efficient

* refine exclude interface

* fix tutorial

* fix tutorial

* fix graph duplication in CPU dataloader workers

* lint

* lint

* Revert "lint"

This reverts commit 805484dd553695111b5fb37f2125214a6b7276e9.

* Revert "lint"

This reverts commit 0bce411b2b415c2ab770343949404498436dc8b2.

* Revert "fix graph duplication in CPU dataloader workers"

This reverts commit 9e3a8cf34c175d3093c773f6bb023b155f2bd27f.
Co-authored-by: xiny <xiny@nvidia.com>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>

* add

* fix

* fix

* fix

* fix

* add

* add

* fix

* fix

* fix

* new loader

* fix

* fix

* fix for 3.6

* fix

* add

* add receipes and also some bug fixes

* fix

* fix

* fix

* fix receipies

* allow AsNodeDataset to work on ogb

* add ut

* many fixes for nodepred-ns pipeline

* receipe for nodepred-ns

* Update enter/README.md
Co-authored-by: Zihao Ye <zihaoye.cs@gmail.com>

* fix layers

* fix

* fix

* fix

* fix

* fix multiple issues

* fix for citation2

* fix comment

* fix

* fix

* clean up

* fix
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>
Co-authored-by: Minjie Wang <minjie.wang@nyu.edu>
Co-authored-by: Zihao Ye <zihaoye.cs@gmail.com>

* implement pin_memory/unpin_memory/is_pinned for dgl.graph

* update python docstring

* update c++ docstring

* add test

* fix the broken UnifiedTensor

* XPU_SWITCH for kDLCPUPinned

* a rough version ready for testing

* eliminate extra context parameter for pin/unpin

* update train_sampling

* fix linting

* fix typo

* multi-gpu uva sampling case

* disable new format materialization for pinned graphs

* update python doc for pin_memory_

* fix unit test

* UVA sampling for link prediction

* dispatch most csr ops

* update graphsage example to combine uva sampling and UnifiedTensor

* update graphsage example to combine uva sampling and UnifiedTensor

* update graphsage example to combine uva sampling and UnifiedTensor

* update doc

* update examples

* change unitgraph and heterograph's PinMemory to in-place

* update examples for multi-gpu uva sampling

* update doc

* fix linting

* fix cpu build

* fix is_pinned for DistGraph

* fix is_pinned for DistGraph

* update graphsage unsupervised example

* update doc for gpu sampling

* update some check for sampling device switching

* fix linting

* adapt for new dataloader

* fix linting

* fix

* fix some name issue

* adjust device check

* add unit test for uva sampling & fix some zero_copy bug

* fix linting

* update num_threads in graphsage examples
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>

* implement pin_memory/unpin_memory/is_pinned for dgl.graph

* update python docstring

* update c++ docstring

* add test

* fix the broken UnifiedTensor

* eliminate extra context parameter for pin/unpin

* fix linting

* fix typo

* disable new format materialization for pinned graphs

* update python doc for pin_memory_

* fix unit test

* update doc

* change unitgraph and heterograph's PinMemory to in-place

* update comments for NDArray's PinMemory_ and PinData

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

* first commit

* a bunch of fixes

* add unique

* lint

* lint

* lint

* address comments

* Update negative_sampler.py

* fix

* description

* address comments and fix

* fix

* replace unique with replace

* test pylint

* Update negative_sampler.py

* support shared memory on windows

* Update shared_mem.cc

* Add pytorch-direct version

* remove

* add documentation for UnifiedTensor

* Revert "add documentation for UnifiedTensor"

This reverts commit 63ba42644d4aba197c1cb4ea4b85fa1bc43b8849.

* add boundary check for UVM IndexSelect

* relocate boundary check index kernels to cuda

* fix function name

* fix indexkernel in nccl api

* fix argument ordering

* simplify code

* Add a comment for the uvm version
Co-authored-by: shhssdm <shhssdm@gmail.com>
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>

* [Bugfix] fix a compile error for Debug-BuildType on Windows Platform

When using CMakeLists.txt to build the "Debug" BuildType on the Windows Platform, it has three compile errors (C4716) in the file "dgl\src\runtime\shared_mem.cc":

'dgl::runtime::SharedMemory::CreateNew': must return a value
'dgl::runtime::SharedMemory::Open': must return a value
'dgl::runtime::SharedMemory::Exist': must return a value

* [Bugfix] cmake error "cannot find load file" when DGL as a sub_directory on Linux

When using DGL as a subdirectory in a CMake Project, the "CMAKE_SOURCE_DIR" here will return the parent cmake scope dir, which is not a expected dir.
Maybe it is better to use "CMAKE_CURRENT_SOURCE_DIR" to set "GKLIB_PATH".

* [Bugfix] cmd cmake error when DGL as a subdirectory

When DGL as a subdirectory of another project, the WORKING_DIRECTORY of "add_custom_command" will be incorrect at the line 255 of "CMakeLists.txt", such that making a cmake "setlocal" error.

* [Feature] enable create/set/free cuda stream for internal use

* add unit test

* fix unit test failure on mxnet and tf

* refactor stream wrapper

* fix lint error

* fix lint error

Co-authored-by: JingchengYu94 <jingchengyu94@gmail.com>

* remove

* remove

* fix

* remove

* remove

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

* Update filter code

* Add unit tests

* Fixes

* Switch to indices

* Rename functions

* Fix linting

* Fix whitespace

* Add doc

* Fix heterograph

* Change workspace allocation

* Fix linting

* Fix docs in filter.py

* Add todo
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

* Only link tensordispatcher against pytorch

* Only modify libraries when not using MSVC

[Feature][Performance][GPU] Introducing UnifiedTensor for efficient zero-copy host memory access from GPU (#3086)

* Add pytorch-direct version

* Initial commit of unified tensor

* Merge branch 'master' of https://github.com/davidmin7/dgl



* Remove unnecessary things

* Fix error message

* Fix/Add descriptions

* whitespace fix

* add unpin

* disable IndexSelectCPUFromGPU with no CUDA

* add a newline for unified_tensor.py

* Apply changes based on feedback

* add 'os' module

* skip unified tensor unit test for cpu only

* Update tests/pytorch/test_unified_tensor.py
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

* reflect feedback
Co-authored-by: shhssdm <shhssdm@gmail.com>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

* Add dgl.utils.is_sorted_srcdst

* Fix linting issues

* delete blank line

* Specify datatype to index tensor in test

* Force integer conversion
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>
Co-authored-by: Quan (Andy) Gan <coin2028@hotmail.com>

* fix

* remove nvidiasmi

* fix

* fix docs

* fix

* fix

* 1

* fix

* remove

* skip deprecated kernel

* fix

* Revert "skip deprecated kernel"

This reverts commit c5ceb7f60dbbaf065b81cc3680757fd611d90ad3.

* fix

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

* Split from NCCL PR

* Fix type in comment

* Expand documentation for sparse_all_to_all_push

* Restore previous behavior in example

* Re-work optimizer to use NCCL based on gradient location

* Allow for running with embedding on CPU but using NCCL for gradient exchange

* Optimize single partition case

* Fix pylint errors

* Add missing include

* fix gradient indexing

* Fix line continuation

* Migrate 'first_step'

* Skip tests without enough GPUs to run NCCL

* Improve empty tensor handling for pytorch 1.5

* Fix indentation

* Allow multiple NCCL communicator to coexist

* Improve handling of empty message

* Update python/dgl/nn/pytorch/sparse_emb.py
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

* Update python/dgl/nn/pytorch/sparse_emb.py
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

* Keepy empty tensor dimensionaless

* th.empty -> th.tensor

* Preserve shape for empty non-zero dimension tensors

* Use shared state, when embedding is shared

* Add support for gathering an embedding

* Fix typo

* Fix more typos

* Fix backend call

* Use NodeDataLoader to take advantage of ddp

* Update training script to share memory

* Only squeeze last dimension

* Better handle empty message

* Keep embedding on the target device GPU if dgl_sparse if false in RGCN example

* Fix typo in comment

* Add asserts

* Improve documentation in example
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

[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>