Reduce the number of copy call in residual norm and allow pinned memory for criterion #1897
Conversation
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
yhmtsai
added
1:ST:ready-for-review
ginkgo-bot
added
reg:testing
upsj
approved these changes
Jul 21, 2025
There was a problem hiding this comment.
LGTM except for small nits! I would like to suggest that it would probably be safer if we don't have to leave a comment at every place where we set the flags, and instead put the two bool values into a struct. If that's too much work, we should put it into an issue somewhere 😄
| @@ -41,12 +41,12 @@ __global__ __launch_bounds__(default_block_size) void residual_norm_kernel( | |||
| if (tidx < num_cols) { | |||
| if (tau[tidx] <= rel_residual_goal * orig_tau[tidx]) { | |||
| stop_status[tidx].converge(stoppingId, setFinalized); | |||
| device_storage[1] = true; | |||
| device_storage[0] = true; | |||
There was a problem hiding this comment.
should we put this into a struct or behind a wrapper to avoid mixing up the two values?
Comment on lines
227
to
+228
| exec->run(ir::make_initialize(&stop_status)); | ||
| exec->run(ir::make_initialize(&stop_status)); |
| // use indicators in each critirion such that we can reduce the | ||
| // #copy_call | ||
| indicators[0] = false; | ||
| one_converged |= |
There was a problem hiding this comment.
This is a bitwise or, maybe it should be a logical one?
| gko::array<gko::stopping_status> stop_status(this->exec_, 1); | ||
| stop_status.get_data()[0].reset(); | ||
|
|
||
| EXPECT_TRUE(criterion->update().residual_norm(res_norm).check( | ||
| RelativeStoppingId, true, &stop_status, &one_changed)); | ||
| RelativeStoppingId, true, &stop_status, |
There was a problem hiding this comment.
General thought: This boolean parameter doesn't tell us what it does. Maybe it could be an enum?
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 uses master executor to allocate the one_changed + one additional workspace (for all_converged).
When the master executor uses device-specific host allocator like
cudaMallocHostto allocate the pinned memory, we can get better bandwidth than default pageable memory. Also, it reduces the number of copy call in residual norm.