Fence on sample only #209
Merged
Fence on sample only #209
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
Putting in tool_invoked_fence code.
Fixing tool induced fences to always fence on device with DevID 0. Fencing with DevID will be a done in subsequent patch (where Pair object will be used in the hash table to capture the begin sample's information. Note that the pair/tuple object can capture other state information to store between the beginning of sampling event and ending of it.
…tools into fenceOnSampleOnly
|
Output for stream with Kokkos CUDA backend on Perlmutter with sampler, having fences on, when Kernel logger is sampled. The output shows the change gives correct behavior of the sampler. |
crtrott
requested changes
Oct 5, 2023
Passing devID to invoke_ktools_febce() instead of 0 is in a separate PR. Checking fence is done only on devID hasn't been tested in this PR and isn't directly related to this PR.
|
Here is another test with the globFence check at runtime taken out, i.e., the latest change as advised by @crtrott. The sampler skip rate is set to 7. This shows output of Kokkos stream with the CUDA backend, using the kokkos sampler applied to the kernel logger, on Perlmutter. Note that the device being printed out is not the physical device ID (on, e.g., a node of supercomputer) but a Kokkos execution space identifier. |
|
This is the same run as the previous post, but with the KOKKOS_TOOLS_SAMPLER_VERBOSE set to 2 instead of 1. This shows the invocation of the Kokkos Tools tool-induced fence via the print, insta. The print statement for this fence shows the physical device ID (converted from the execution space ID). We see from the output that the device ID is 0. This is correct, given each begin/end tools callback invokes a tool-induced fence using the parameter 0. |
crtrott
approved these changes
Oct 12, 2023
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 PR fences only when a sample event is taken, i.e., at the beginning of the sample in kokkosp_begin_xyz( mykID) and at the end of the corresponding sample in kokkosp_end_xyz(mykID). This improves efficiency of Kokkos Tools, when sampling is done.
Note that provide tools programming interface must be exposed in profiling/all/kp_core.hpp. This wasn't done previously, i.e., it is not in the develop branch of Kokkos Tools, and it is useful for other tools needing the tools programming interface.
Notes from PR #194 are relevant to this PR.