* Port elementwise metrics to GPU.
* All elementwise metrics are converted to static polymorphic.
* Create a reducer for metrics reduction.
* Remove const of Metric::Eval to accommodate CubMemory.
- Improved GPU performance logging
- Only use one execute shards function
- Revert performance regression on multi-GPU
- Use threads to launch NCCL AllReduce
* Enable running objectives with 0 GPU.
* Enable 0 GPU for objectives.
* Add doc for GPU objectives.
* Fix some objectives defaulted to running on all GPUs.
* Clean up logic for converting tree_method to updater sequence
* Use C++11 enum class for extra safety
Compiler will give warnings if switch statements don't handle all
possible values of C++11 enum class.
Also allow enum class to be used as DMLC parameter.
* Fix compiler error + lint
* Address reviewer comment
* Better docstring for DECLARE_FIELD_ENUM_CLASS
* Fix lint
* Add C++ test to see if tree_method is recognized
* Fix clang-tidy error
* Add test_learner.h to R package
* Update comments
* Fix lint error
* Multi-GPU support in GPUPredictor.
- GPUPredictor is multi-GPU
- removed DeviceMatrix, as it has been made obsolete by using HostDeviceVector in DMatrix
* Replaced pointers with spans in GPUPredictor.
* Added a multi-GPU predictor test.
* Fix multi-gpu test.
* Fix n_rows < n_gpus.
* Reinitialize shards when GPUSet is changed.
* Tests range of data.
* Remove commented code.
* Remove commented code.
**Symptom** Apple Clang's implementation of `std::shuffle` expects doesn't work
correctly when it is run with the random bit generator for R package:
```cpp
CustomGlobalRandomEngine::result_type
CustomGlobalRandomEngine::operator()() {
return static_cast<result_type>(
std::floor(unif_rand() * CustomGlobalRandomEngine::max()));
}
```
Minimial reproduction of failure (compile using Apple Clang 10.0):
```cpp
std::vector<int> feature_set(100);
std::iota(feature_set.begin(), feature_set.end(), 0);
// initialize with 0, 1, 2, 3, ..., 99
std::shuffle(feature_set.begin(), feature_set.end(), common::GlobalRandom());
// This returns 0, 1, 2, ..., 99, so content didn't get shuffled at all!!!
```
Note that this bug is platform-dependent; it does not appear when GCC or
upstream LLVM Clang is used.
**Diagnosis** Apple Clang's `std::shuffle` expects 32-bit integer
inputs, whereas `CustomGlobalRandomEngine::operator()` produces 64-bit
integers.
**Fix** Have `CustomGlobalRandomEngine::operator()` produce 32-bit integers.
Closes#3523.
* Split building histogram into separated class.
* Extract `InitCompressedRow` definition.
* Basic tests for gpu-hist.
* Document the code more verbosely.
* Removed `HistCutUnit`.
* Removed some duplicated copies in `GPUHistMaker`.
* Implement LCG and use it in tests.
* Implement Transform class.
* Add tests for softmax.
* Use Transform in regression, softmax and hinge objectives, except for Cox.
* Mark old gpu objective functions deprecated.
* static_assert for softmax.
* Split up multi-gpu tests.
* DMatrix refactor 2
* Remove buffered rowset usage where possible
* Transition to c++11 style iterators for row access
* Transition column iterators to C++ 11
- previously, vec_ in DeviceShard wasn't updated on copy; as a result,
the shards continued to refer to the old HostDeviceVectorImpl object,
which resulted in a dangling pointer once that object was deallocated
* Replaced std::vector with HostDeviceVector in MetaInfo and SparsePage.
- added distributions to HostDeviceVector
- using HostDeviceVector for labels, weights and base margings in MetaInfo
- using HostDeviceVector for offset and data in SparsePage
- other necessary refactoring
* Added const version of HostDeviceVector API calls.
- const versions added to calls that can trigger data transfers, e.g. DevicePointer()
- updated the code that uses HostDeviceVector
- objective functions now accept const HostDeviceVector<bst_float>& for predictions
* Updated src/linear/updater_gpu_coordinate.cu.
* Added read-only state for HostDeviceVector sync.
- this means no copies are performed if both host and devices access
the HostDeviceVector read-only
* Fixed linter and test errors.
- updated the lz4 plugin
- added ConstDeviceSpan to HostDeviceVector
- using device % dh::NVisibleDevices() for the physical device number,
e.g. in calls to cudaSetDevice()
* Fixed explicit template instantiation errors for HostDeviceVector.
- replaced HostDeviceVector<unsigned int> with HostDeviceVector<int>
* Fixed HostDeviceVector tests that require multiple GPUs.
- added a mock set device handler; when set, it is called instead of cudaSetDevice()
* Add basic Span class based on ISO++20.
* Use Span<Entry const> instead of Inst in SparsePage.
* Add DeviceSpan in HostDeviceVector, use it in regression obj.
* Expand histogram memory dynamically to prevent large allocations for large tree depths (e.g. > 15)
* Remove GPU memory allocation messages. These are misleading as a large number of allocations are now dynamic.
* Fix appveyor R test
* Added finding quantiles on GPU.
- this includes datasets where weights are assigned to data rows
- as the quantiles found by the new algorithm are not the same
as those found by the old one, test thresholds in
tests/python-gpu/test_gpu_updaters.py have been adjusted.
* Adjustments and improved testing for finding quantiles on the GPU.
- added C++ tests for the DeviceSketch() function
- reduced one of the thresholds in test_gpu_updaters.py
- adjusted the cuts found by the find_cuts_k kernel
* Fail GPU CI after test failure
* Fix GPU linear tests
* Reduced number of GPU tests to speed up CI
* Remove static allocations of device memory
* Resolve illegal memory access for updater_fast_hist.cc
* Fix broken r tests dependency
* Update python install documentation for GPU
* Upgrading to NCCL2
* Part - II of NCCL2 upgradation
- Doc updates to build with nccl2
- Dockerfile.gpu update for a correct CI build with nccl2
- Updated FindNccl package to have env-var NCCL_ROOT to take precedence
* Upgrading to v9.2 for CI workflow, since it has the nccl2 binaries available
* Added NCCL2 license + copy the nccl binaries into /usr location for the FindNccl module to find
* Set LD_LIBRARY_PATH variable to pick nccl2 binary at runtime
* Need the nccl2 library download instructions inside Dockerfile.release as well
* Use NCCL2 as a static library
* Use sparse page as singular CSR matrix representation
* Simplify dmatrix methods
* Reduce statefullness of batch iterators
* BREAKING CHANGE: Remove prob_buffer_row parameter. Users are instead recommended to sample their dataset as a preprocessing step before using XGBoost.
* GPU binning and compression.
- binning and index compression are done inside the DeviceShard constructor
- in case of a DMatrix with multiple row batches, it is first converted into a single row batch
* Multi-GPU HostDeviceVector.
- HostDeviceVector instances can now span multiple devices, defined by GPUSet struct
- the interface of HostDeviceVector has been modified accordingly
- GPU objective functions are now multi-GPU
- GPU predicting from cache is now multi-GPU
- avoiding omp_set_num_threads() calls
- other minor changes
* Replaced std::vector-based interfaces with HostDeviceVector-based interfaces.
- replacement was performed in the learner, boosters, predictors,
updaters, and objective functions
- only interfaces used in training were replaced;
interfaces like PredictInstance() still use std::vector
- refactoring necessary for replacement of interfaces was also performed,
such as using HostDeviceVector in prediction cache
* HostDeviceVector-based interfaces for custom objective function example plugin.
- thrust::copy() called from dvec::copy() for gpairs invoked a GPU kernel instead of
cudaMemcpy()
- this resulted in illegal memory access if the GPU running the kernel could not access
the data being copied
- new version of dvec::copy() for thrust::device_ptr iterators calls cudaMemcpy(),
avoiding the problem.
* Added GPU objective function and no-copy interface.
- xgboost::HostDeviceVector<T> syncs automatically between host and device
- no-copy interfaces have been added
- default implementations just sync the data to host
and call the implementations with std::vector
- GPU objective function, predictor, histogram updater process data
directly on GPU
