#0: Add borrowed storage support for the aggregate_as_tensor function #18555
+63
−0
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
jjiangTT
requested review from
cfjchu,
ayerofieiev-tt,
dmakoviichuk-tt and
omilyutin-tt
as code owners
omilyutin-tt
reviewed
Mar 3, 2025
ttnn/cpp/ttnn/distributed/api.cpp
Outdated
Comment on lines
103
to
110
| auto visitor = tt::stl::overloaded{[&shard, &host_owned_buffers](const auto& buffer) -> OwnedBuffer { | ||
| using BufferType = std::decay_t<decltype(buffer)>; | ||
| using ValueType = typename BufferType::value_type; | ||
|
|
||
| std::vector<ValueType> physical_data(buffer.begin(), buffer.end()); | ||
|
|
||
| std::vector<ValueType> logical_data = | ||
| tensor_impl::decode_tensor_data(std::move(physical_data), shard.get_tensor_spec()); | ||
|
|
||
| return owned_buffer::create(std::move(logical_data)); | ||
| }}; |
There was a problem hiding this comment.
I don't think we need this? Does this work:
auto borrowed_buffer = std::get<BorrowedStorage>(shard.get_storage()).buffer;
auto owned_buffer = std::visit(
[](auto&& buffer) {
using BorrowedStorageType = std::vector<std::decay_t<decltype(*(buffer.begin()))>>;
return owned_buffer::create(BorrowedStorageType(buffer.begin(), buffer.end()));
},
borrowed_buffer);
omilyutin-tt
requested changes
Mar 3, 2025
| @@ -93,7 +93,31 @@ Tensor aggregate_as_tensor( | |||
| } | |||
| auto storage = MultiDeviceHostStorage{config, std::move(host_owned_buffers), specs}; | |||
| return Tensor(std::move(storage), reference_shard.get_tensor_spec()); | |||
| } else if (storage_type == StorageType::BORROWED) { | |||
There was a problem hiding this comment.
I don't think this is formatted?
jjiangTT
force-pushed
the
jjiang/support_borrowed_aggregate
branch
2 times, most recently
from
eb73a5e to
84c65ae
Compare
jjiangTT
force-pushed
the
jjiang/support_borrowed_aggregate
branch
5 times, most recently
from
cf2f060 to
e2a46b1
Compare
jjiangTT
requested review from
rfurko-tt,
TT-BrianLiu,
razorback3,
dongjin-na and
bbradelTT
as code owners
jjiangTT
force-pushed
the
jjiang/support_borrowed_aggregate
branch
from
4e5f8a9 to
0f8151f
Compare
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.
Ticket
Link to Github Issue
Problem description
The aggregate_as_tensor function currently goes straight to assuming anything that isn't an owned tensor is a device tensor and then failing. Borrowed storage is currently the "preferred default" so newly created non-bfp_b types would require a conversion before they could be aggregated otherwise.
What's changed
Added support for borrowed storage tensors by turning them into owned tensors and then aggregating them into a multidevicehost storage backed tensor.
Added TT_FATAL checking to prevent any other types from crashing on the storage variant fetch and give some more visibility into failures.
Checklist