-
Notifications
You must be signed in to change notification settings - Fork 5.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Some work with CUDNN #2797
base: master
Are you sure you want to change the base?
Some work with CUDNN #2797
Commits on Sep 15, 2018
-
WIP: Add CUDNN as an optional dependency to cu-device.{cc,h}
Currently, this assumes that cudnn is installed in /usr/local/cuda/lib64 and /usr/local/cuda/include, because that's what my current machine has it installed. This can be cleaned up later.
Configuration menu - View commit details
-
Copy full SHA for 2d76ce9 - Browse repository at this point
Copy the full SHA 2d76ce9View commit details -
Configuration menu - View commit details
-
Copy full SHA for 0a00e73 - Browse repository at this point
Copy the full SHA 0a00e73View commit details -
Initial draft of CUDNN 2d convolution implementation.
Double check that we correctly transpose height and width everywhere!
Configuration menu - View commit details
-
Copy full SHA for 224d4da - Browse repository at this point
Copy the full SHA 224d4daView commit details -
Configuration menu - View commit details
-
Copy full SHA for c9806e0 - Browse repository at this point
Copy the full SHA c9806e0View commit details -
Configuration menu - View commit details
-
Copy full SHA for 5b1855a - Browse repository at this point
Copy the full SHA 5b1855aView commit details -
[src] Fix Singleton implementation of CuDevice.
No longer publicly expose the default constructor. That could allow someone to do something like this: CuDevice device1; CuDevice device2; We're not sure what this would cause, but it probably wouldn't be good.
Configuration menu - View commit details
-
Copy full SHA for 8e00e4f - Browse repository at this point
Copy the full SHA 8e00e4fView commit details -
Configuration menu - View commit details
-
Copy full SHA for 5e4d270 - Browse repository at this point
Copy the full SHA 5e4d270View commit details
Commits on Sep 30, 2018
-
Fix implementation's height-width switching.
Document data formats expected of each member function.
Configuration menu - View commit details
-
Copy full SHA for 0387a0c - Browse repository at this point
Copy the full SHA 0387a0cView commit details
Commits on Oct 8, 2018
-
Make CUDNN mandatory if building with CUDA.
There may be edge cases in the configure script with this remaining. We still don't look up the version of CUDNN to download for your CUDA version. Also make sure that non-CUDA builds still compile. This is achieved by the cu-cudnn-helper.h file, which forward declares cudnn types, which fortunately are all pointer types (except for enums, which don't matter in this case).
Configuration menu - View commit details
-
Copy full SHA for be6fc2a - Browse repository at this point
Copy the full SHA be6fc2aView commit details
Commits on Oct 13, 2018
-
Configuration menu - View commit details
-
Copy full SHA for e729495 - Browse repository at this point
Copy the full SHA e729495View commit details -
Configuration menu - View commit details
-
Copy full SHA for 5bb72e5 - Browse repository at this point
Copy the full SHA 5bb72e5View commit details -
Automatically download CUDNN as part of configure.
Remove support for CUDA versions that don't support CUDNN v7. Remove 32-bit CUDA mode, since no recent version of CUDA supports it. I may not have removed all instances of it.
Configuration menu - View commit details
-
Copy full SHA for 9647494 - Browse repository at this point
Copy the full SHA 9647494View commit details
Commits on Oct 15, 2018
-
Merge pull request #52 from galv/cudnn
Automatically download CUDNN as part of configure.
Configuration menu - View commit details
-
Copy full SHA for 6427e79 - Browse repository at this point
Copy the full SHA 6427e79View commit details -
Configuration menu - View commit details
-
Copy full SHA for 53d62af - Browse repository at this point
Copy the full SHA 53d62afView commit details
Commits on Oct 16, 2018
-
Configuration menu - View commit details
-
Copy full SHA for 7000850 - Browse repository at this point
Copy the full SHA 7000850View commit details
Commits on Oct 23, 2018
-
Configuration menu - View commit details
-
Copy full SHA for b5d2022 - Browse repository at this point
Copy the full SHA b5d2022View commit details -
Change filter type back to NCHW, since it supports more algos.
Convert assertion to warning, although I am not sure this works, since if an algorithm fails to be found because of an out-of-memory error, it is likely to fail at training time for the same reason.
Configuration menu - View commit details
-
Copy full SHA for 22669f6 - Browse repository at this point
Copy the full SHA 22669f6View commit details
Commits on Oct 24, 2018
-
Workaround cudnnSetTensor4dDescriptor's striding bug.
Use cudnnSetTensor4dDescriptor with strides we calculate ourselves instead.
Configuration menu - View commit details
-
Copy full SHA for c958143 - Browse repository at this point
Copy the full SHA c958143View commit details -
Merge pull request #53 from galv/cudnn-povey
Change filter type back to NCHW, since it supports more algos.
Configuration menu - View commit details
-
Copy full SHA for e4d3383 - Browse repository at this point
Copy the full SHA e4d3383View commit details
Commits on Oct 25, 2018
-
Configuration menu - View commit details
-
Copy full SHA for 74114d0 - Browse repository at this point
Copy the full SHA 74114d0View commit details
Commits on Oct 26, 2018
-
Don't use cudnnConvolutionBiasActivationForward.
It supports only the precomputed implicit GEMM implementation of convolution.
Configuration menu - View commit details
-
Copy full SHA for 0bed8aa - Browse repository at this point
Copy the full SHA 0bed8aaView commit details -
Configuration menu - View commit details
-
Copy full SHA for c546716 - Browse repository at this point
Copy the full SHA c546716View commit details
Commits on Oct 27, 2018
-
The ConvolutionComputation class does not know whether or not the bias is optional at initialization time. It will simply avoid using the bias if it is a nullptr.
Configuration menu - View commit details
-
Copy full SHA for 450f491 - Browse repository at this point
Copy the full SHA 450f491View commit details -
Merge pull request #54 from galv/cudnn-povey-2
Don't use cudnnConvolutionBiasActivationForward.
Configuration menu - View commit details
-
Copy full SHA for 464db5c - Browse repository at this point
Copy the full SHA 464db5cView commit details
Commits on Oct 28, 2018
-
[egs] Remove unnecessary alignment from mini_librispeech run.sh, than…
…ks: johnjosephmorgan@gmail.com
Configuration menu - View commit details
-
Copy full SHA for bfea6c8 - Browse repository at this point
Copy the full SHA bfea6c8View commit details
Commits on Oct 31, 2018
-
Configuration menu - View commit details
-
Copy full SHA for a2de7b9 - Browse repository at this point
Copy the full SHA a2de7b9View commit details
Commits on Nov 20, 2018
-
Configuration menu - View commit details
-
Copy full SHA for 0457a61 - Browse repository at this point
Copy the full SHA 0457a61View commit details