Open
Conversation
.. fix some of the memory leaks in the forward .. make note about optimization in the forward .. add tensor_make_transpose_1230 (not working) and test (not working) .. moves tests for tensor_make_transpose<....> into test_tensor_op.cpp Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
This second function which does the same thing as the first is only here to show an example of a more direct mapping between the original memory and a single index. This will allow a direct conversion to cuda code without much issue. Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Tpose 3012 dev
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
…aterials into test-speed-1-32
…izes Also creates a separate file for testing the device utils and for their harnesses. Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
…aterials into test-speed-1-32
moves some things around, but creates harness functions as an example of what all our call structures will need to be in the future in order to gain control over the block and thread sizes. Note that this should be further extended to actually accept dim3, but for now it will not be. Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu> Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
This function was really designed to do a parameter search for the forward_and backward device convolution operations. However, for the submission, I have reduced it to performing a single benchmark. Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
…ce utils Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
…puting-materials into test-speed-1-32
Also creates a separate file for testing the device utils and for their harnesses. Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
Also creates a separate file for testing the device utils and for their harnesses. Signed-off-by: Christopher Goebel <cmgoebel@iupui.edu>
…mputing-materials into gpu-computing-submission
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Changes the device allocation and deallocation to take in real tensors which allow me to do a better job forcing everything to notify on its allocation and deallocation counter. Should help figure out whats up with the memory leaks in convolution forward and backward. Also further changes toward unifying on in in the whole project. Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
The problem was a simple missing deallocation on d_dx_cols, and in this case, it was missing because d_dx_cols and d_x_cols look almost the same. It was a very easy mistake to make, but silly none-the-less. Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
Signed-off-by: Christopher Goebel <cmgoebel@iu.edu>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
PR for both forward-device and backward-device setup.
This PR incorporates a number of significant refactors
which enable CUDA functionality in both the primary
forward and backward functions.
Note that these functions are not optimized, nor are they
well written. However, they work.
Additionally, in several locations, there is leftover code
from where I was testing my mappings from the large
deeply nested loops in the col2im_inner and im2col_inner
functions to the 1D mappings needed for the device code.
Those sections should be moved out of the code and
into documentation somewhere.