Skip to content
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.

Sign up for GitHub

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

Jump to bottom

kernarg buffer allocation strategy improvement #1395

Open
wants to merge 14 commits into
base: clang_tot_upgrade
Choose a base branch
from
Open

kernarg buffer allocation strategy improvement #1395

wants to merge 14 commits into from
+327 −46

Conversation

scchan
Copy link
Collaborator

@scchan scchan commented Feb 23, 2020

  • add the capability to allocate kernarg buffer in coarse grained GPU memory to leverage caching (off by default)
  • implemented a new algorithm to manage kernarg buffers
    • provides a control point when to recycle old kernel buffers (for coarse grained buffers)
    • more efficient free kernarg buffer lookup
    • supports multiple pools with kernarg buffer sizes

@scchan scchan requested a review from jeffdaily February 23, 2020 23:47
@scchan scchan force-pushed the kernarg_coarse_grain_vm branch 2 times, most recently from c622acf to 646c749 Compare March 2, 2020 21:30
@scchan
Copy link
Collaborator Author

scchan commented Mar 2, 2020

branch rebased and resolved conflicts

scchan added 13 commits March 6, 2020 15:23
@scchan
first commit to revamp kernarg pool
347e568
@scchan
Support multiple kernarg pools of different buffer sizes, minor cleanup
c5d20d5
@scchan
Add a knob to switch to the legacy kernarg pool management
87da59f
@scchan
Put the released kernarg buffers back to the free pool
4bf79a3
@scchan
Add an env var to force kernarg to allocate from coarse grained memory
59d0fd4
@scchan
Several updates to new kernarg allocation strategy
3a4dddc 
- fix check for host access to gpu coarse grained memory
- add counter for signaling kernarg buffers recycling
- rename env var to control kernarg memory allocation
@scchan
Flush HDP after writing to kernarg if the memory is gpu coarse grained
3e0d6bc
@scchan
If kernarg is in coarse grained GPU memory, add an sys scope acq fence
207935b 
after recycling old kernarg buffers such that the GPU cache is flushed
@scchan
Proper synchronization sequence for kernarg write, HDP and AQL
4b71137
@scchan
Update the kernarg pool growth heuristic, add a few knobs for debugging
e4d74b4
@scchan
Minor simplification to kernarg buffer recycling, add debug messages
d565751
@scchan
add a few more debug messages for coarse grained VM kernarg
3cba197
@scchan
Kernarg - enable coarse grained memory support by default
d4badc6
@scchan scchan force-pushed the kernarg_coarse_grain_vm branch from 646c749 to d4badc6 Compare March 6, 2020 20:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jeffdaily jeffdaily Awaiting requested review from jeffdaily

At least 1 approving review is required to merge this pull request.

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@scchan