Add Hexagon VTCM and discontiguous allocation support #9525
+352
−119
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
adstraw
commented
Nov 17, 2021
| } | ||
|
|
||
| void* HexagonBuffer::GetPointer() { | ||
| void** HexagonBuffer::GetPointer() { |
There was a problem hiding this comment.
Returning a void ** here disambiguates between the cases where there is a single allocation vs. multiple. We always return a pointer to a pointer. This may be controversial.
There was a problem hiding this comment.
Did we decide to revert this change so that tests pass and revisit once codegen supports pointer array indexing?
There was a problem hiding this comment.
I decided to keep HexagonBuffer as-is returning a void** which means that it (HexagonBuffer) supports discontiguous allocation. And, I noted the portions of the device API that force / assume contiguous allocation awaiting RFC 39 with "TODO" for myself.
6589f51 to
acc78e5
Compare
csullivan
reviewed
Nov 17, 2021
adstraw
commented
Nov 17, 2021
e099787 to
e01dc93
Compare
98a413f to
61db96e
Compare
c70256c to
7122d2b
Compare
csullivan
requested changes
Nov 24, 2021
csullivan
reviewed
Nov 29, 2021
csullivan
reviewed
Nov 29, 2021
csullivan
reviewed
Nov 29, 2021
csullivan
reviewed
Dec 10, 2021
|
|
||
| mod["add"](A_data, B_data, C_data) | ||
| result = C_data.numpy() | ||
| assert (result == np.array([6, 7])).all() |
There was a problem hiding this comment.
Safer to use all_close, and probably want to make another test where with larger tensors.
There was a problem hiding this comment.
Let's consider adding some more testing in a follow up.
98f152c to
643e9ad
Compare
tmoreau89
approved these changes
Dec 11, 2021
|
Thank you for the great work @adstraw and for the review @csullivan ; the PR has been merged! |
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 7, 2022
* WIP Allocation abstraction for VTCM and DDR. * Add Hexagon VTCM and discontiguous allocation support * differentiate between dimensions and allocations * remove change to llvm codegen * add integration test_add_vtcm to demo vtcm alloc * remove cmake change * forcing contiguous allocation in device API, for now Co-authored-by: Chris Sullivan <csullivan@octoml.ai>
yangulei
pushed a commit
to yangulei/tvm
that referenced
this pull request
Jan 11, 2022
* WIP Allocation abstraction for VTCM and DDR. * Add Hexagon VTCM and discontiguous allocation support * differentiate between dimensions and allocations * remove change to llvm codegen * add integration test_add_vtcm to demo vtcm alloc * remove cmake change * forcing contiguous allocation in device API, for now Co-authored-by: Chris Sullivan <csullivan@octoml.ai>
yangulei
pushed a commit
to yangulei/tvm
that referenced
this pull request
Jan 12, 2022
* WIP Allocation abstraction for VTCM and DDR. * Add Hexagon VTCM and discontiguous allocation support * differentiate between dimensions and allocations * remove change to llvm codegen * add integration test_add_vtcm to demo vtcm alloc * remove cmake change * forcing contiguous allocation in device API, for now Co-authored-by: Chris Sullivan <csullivan@octoml.ai>
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 13, 2022
* WIP Allocation abstraction for VTCM and DDR. * Add Hexagon VTCM and discontiguous allocation support * differentiate between dimensions and allocations * remove change to llvm codegen * add integration test_add_vtcm to demo vtcm alloc * remove cmake change * forcing contiguous allocation in device API, for now Co-authored-by: Chris Sullivan <csullivan@octoml.ai>
qsqqsqqsq-intellif
pushed a commit
to qsqqsqqsq-intellif/tvm
that referenced
this pull request
Apr 29, 2022
* WIP Allocation abstraction for VTCM and DDR. * Add Hexagon VTCM and discontiguous allocation support * differentiate between dimensions and allocations * remove change to llvm codegen * add integration test_add_vtcm to demo vtcm alloc * remove cmake change * forcing contiguous allocation in device API, for now Co-authored-by: Chris Sullivan <csullivan@octoml.ai>
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
memcpy. DMA support still pending. Code for VTCM allocation is present in this PR. VTCM allocation code is stubbed out when not building on Hexagon in favor or “mock” VTCM allocations usingmallocwhich allows for unit testing of theHexagonBufferclass on CPU. The Hexagon Device API is modified to use theHexagonBufferclass including implementation of theCopyDataFromToAPI.