Dnn ops #734
Draft
Dnn ops #734
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
corepointer
force-pushed
the
dnn-ops
branch
2 times, most recently
from
315d0db
to
3e02d16
Compare
corepointer
force-pushed
the
dnn-ops
branch
2 times, most recently
from
66df10a
to
40025a8
Compare
|
This PR is now based on #758 |
Merged
* This commit introduces the meta data object to the CSR data type * Memory pinning To prevent excessive allocation ID lookups in the hot path when using --vec, this change "pins" memory by allocation type of previous accesses.
Due to the use of ptr to local var the distributed (GRPC_SYNC) mode crashed in test cases. This patch fixes this by using std::unique_ptr appropriately.
batch_norm2d shape & type inference affine shape/type inference softmax namespace, shape & type inference "fix" shape inference of some dnn ops by returning -1 instead of trying to calculate proper dimensions based on data that is sometimes just not there :-/ * getShape(getInput()) returns dimensions when input comes from readMatrix() but not if it's a rand() ?! * getInputHeight(), getInputWidth() does not return proper dimensions
This commit adds the necessary code changes to call CUDNN's activation backward function with ReLU activation type. No tests yet.
Contains ports from SystemDS * script level alternatives for pooling, convolution, etc * wrapper scripts for DAPHNE builtins (conv2d() -> conv2d.forward(), etc) * script path in default UserConfig.json
Example ported from SystemDS
currently supported: relu & conv2d tests are failing atm
This change makes the DaphneContext object global to avoid creation/desctruction in every UDF. The global context is passed by int64 casted pointer through the UserConfig.
agg all cuda launch config bugfix fix allagg cuda launch configs (now looping) aggall log
Added more operators to apply elementwise
The handling of the 1x1 matrix case should not be needed anymore once this is fixed in the compiler to call EwBinaryObjSca
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This is a dirty and buggy snapshot of the progress on integrating all the needed ops for DNN.
Also contains fixes for some of the bugs :)