Use DynamicScratchpad in KernelManager. #3670
Conversation
Signed-off-by: Michał Zientkiewicz <[email protected]>
|
CI MESSAGE: [3934559]: BUILD STARTED |
|
CI MESSAGE: [3934559]: BUILD FAILED |
|
CI MESSAGE: [3935973]: BUILD STARTED |
mzient
force-pushed
the
UseDynamicScratchpad
branch
from
e32b71d to
900733a
Compare
|
CI MESSAGE: [3936330]: BUILD STARTED |
|
CI MESSAGE: [3936330]: BUILD FAILED |
Signed-off-by: Michał Zientkiewicz <[email protected]>
|
CI MESSAGE: [3943354]: BUILD STARTED |
|
CI MESSAGE: [3943354]: BUILD FAILED |
Signed-off-by: Michał Zientkiewicz <[email protected]>
…allocating 0 bytes. Signed-off-by: Michał Zientkiewicz <[email protected]>
|
CI MESSAGE: [3957968]: BUILD STARTED |
|
CI MESSAGE: [3959429]: BUILD STARTED |
mzient
force-pushed
the
UseDynamicScratchpad
branch
from
8252a74 to
1ef772c
Compare
…d, Convolution and Laplacian tests. Signed-off-by: Michał Zientkiewicz <[email protected]>
mzient
force-pushed
the
UseDynamicScratchpad
branch
from
1ef772c to
4c7e414
Compare
|
CI MESSAGE: [3959606]: BUILD STARTED |
|
CI MESSAGE: [3959606]: BUILD PASSED |
| "Kernel instance index (instance_idx) out of range"); | ||
| auto &inst = instances[instance_idx]; | ||
| DynamicScratchpad scratchpad({}, AccessOrder(context.gpu.stream)); | ||
| auto *old_scratchpad = context.scratchpad; |
There was a problem hiding this comment.
Why do we need context.scratchpad for if we are not using it anyway here?
There was a problem hiding this comment.
It's used by the Run method invoked afterwards.
There was a problem hiding this comment.
I was rather asking about the ideas of replacing the old one with one created here.
There was a problem hiding this comment.
Well, now the old one is always NULL, so we don't need to store it. I'll simplify it (again, possibly in a follow-up).
There was a problem hiding this comment.
Ok, let do that in a follow-up.
There was a problem hiding this comment.
So kernel manager scratchpads are also to be removed in a follow-up, right?
There was a problem hiding this comment.
Yes, that's what I brought up in open floor. Since this involves an API change (we no longer need thread_idx), the change affects a large number of files - hence the decision to make the more important change that potentially affects performance (this one) first and then clean the rest.
Signed-off-by: Michal Zientkiewicz <[email protected]>
|
CI MESSAGE: [3968027]: BUILD STARTED |
|
CI MESSAGE: [3968027]: BUILD PASSED |
* Use DynamicScratchpad in KernelManager. * Use invalid stream as a default for KernelContext. * Lazily initialize upstream resources for dynamic scratchpad. * Do not initialize the underlying resource for DynamicScratchpad when allocating 0 bytes. Signed-off-by: Michal Zientkiewicz <[email protected]>
* Use DynamicScratchpad in KernelManager. * Use invalid stream as a default for KernelContext. * Lazily initialize upstream resources for dynamic scratchpad. * Do not initialize the underlying resource for DynamicScratchpad when allocating 0 bytes. Signed-off-by: Michal Zientkiewicz <[email protected]>
* Use DynamicScratchpad in KernelManager. * Use invalid stream as a default for KernelContext. * Lazily initialize upstream resources for dynamic scratchpad. * Do not initialize the underlying resource for DynamicScratchpad when allocating 0 bytes. Signed-off-by: Michal Zientkiewicz <[email protected]>
Signed-off-by: Michał Zientkiewicz [email protected]
Category:
New feature (non-breaking change which adds functionality)
Description:
The dynamic scratchpad is now the primary scratchpad implementation.
Additional information:
Affected modules and functionalities:
Key points relevant for the review:
Checklist
Tests
Documentation
DALI team only
Requirements
REQ IDs: N/A
JIRA TASK: DALI-2449