Skip to content

[NVIDIA][DYNAMIC SHAPES] Dynamic allocation#1038

Open
redradist wants to merge 2 commits intoopenvinotoolkit:masterfrom
redradist:featurue/dynamic-shapes-dynamic-allocation
Open

[NVIDIA][DYNAMIC SHAPES] Dynamic allocation#1038
redradist wants to merge 2 commits intoopenvinotoolkit:masterfrom
redradist:featurue/dynamic-shapes-dynamic-allocation

Conversation

@redradist
Copy link
Contributor

Description

This is the first pull-request from series of PRs to support dynamic shapes

@redradist redradist requested a review from a team as a code owner February 18, 2026 20:24
@github-actions github-actions bot added the category: NVIDIA plugin OpenVINO NVIDIA plugin label Feb 18, 2026
@redradist
Copy link
Contributor Author

@ilya-lavrenov @p-durandin Take a look

redradist and others added 2 commits February 19, 2026 12:51
Remove hard-coded dynamic model rejection from CompiledModel and Plugin.
Extend MemoryPool with DynamicChunk, DynamicHandle and pending request
queue to support GPU memory allocation with sizes unknown at compile time.
Proxy exposes AllocateDynamic() as the only entry point for dynamic
allocation, keeping MemoryPool internals private.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cover fast-path allocation, RAII handle, move semantics, alignment,
interrupt, FIFO fairness, sub-allocation, chunk recycling via cudaMalloc
retry, and multi-threaded concurrent allocation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@redradist redradist force-pushed the featurue/dynamic-shapes-dynamic-allocation branch from b3f2786 to b3b906f Compare February 19, 2026 10:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

category: NVIDIA plugin OpenVINO NVIDIA plugin

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant

Comments