Remove internal use of gpu_id. #9568
Conversation
|
Why are we dropping gtest from the CI? |
|
@hcho3 windows build is failing with cmake find gtest. (same error on master branch). Since the test is building the Python package, I figure we don't need to build c++ tests as we are not running it. |
|
Are we not running gtest on Windows CI? |
|
only on buildkite. |
|
Ah, got it |
trivialfis
force-pushed
the
device-ordinal
branch
from
786b4aa
to
ae18374
Compare
src/common/host_device_vector.cu
Outdated
| @@ -138,7 +138,7 @@ class HostDeviceVectorImpl { | |||
| } else { | |||
| auto ptr = other->ConstDevicePointer(); | |||
| SetDevice(); | |||
| CHECK_EQ(this->DeviceIdx(), other->DeviceIdx()); | |||
| CHECK_EQ(this->Device().ordinal, other->Device().ordinal); | |||
There was a problem hiding this comment.
| CHECK_EQ(this->Device().ordinal, other->Device().ordinal); | |
| CHECK(this->Device() == other->Device()); |
| if (device_.IsCUDA() && device.IsCUDA()) { | ||
| CHECK_EQ(device_.ordinal, device.ordinal) |
There was a problem hiding this comment.
Is it possible for two DeviceOrd objects to be different when they have the same positive value in the ordinal field ?
There was a problem hiding this comment.
Not possible at the moment. Unless there's new device type other than CUDA.
src/predictor/gpu_predictor.cu
Outdated
| CHECK(!p_fmat->Info().IsColumnSplit()) | ||
| << "Predict contribution support for column-wise data split is not yet implemented."; |
There was a problem hiding this comment.
Is SHAP support available for column-wise split data now?
There was a problem hiding this comment.
The check was removed between PR rebases. I will bring it back.
Device communicator is still using integer ordinal, which will be changed by upcoming PRs.