Repository navigation
refactor!: support modern InfiniLM execution - #50
Open
gongchensu wants to merge 3 commits into
Open
gongchensu wants to merge 3 commits into
gongchensu wants to merge 3 commits into
Conversation
gongchensu
changed the title
* fix: guard pool capture and validate runtime device selection Reject unavailable devices before dispatch and prevent CUDA pool growth during native capture. Complete event not-ready mappings and cover the failure paths with regression tests. * fix: restore device after same-device guard scopes * style: satisfy Ruff and clang-format checks
This branch has not been deployed
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
DeviceGuardininclude/infini/rt/device_guard.hfor scoped runtime device selection.ReusableMemoryPoolininclude/infini/rt/reusable_memory_pool.handsrc/reusable_memory_pool.cc, with block reuse, logical ownership tracking, pin transactions, graph retention leases, and trimming of unused blocks.scripts/generate_public_headers.pywithGetDeviceCount(Device::Type, int*)andIsEventNotReady(Error), including NVIDIA's not-ready status mapping.Float16::FromFloatinsrc/data_type.hto round to nearest, ties to even, including subnormal boundary handling.SetDeviceandGetDeviceCount; change Cambricon FP16/BF16 type mappings to InfiniRT storage types, removing the BANG half-header dependency from that RT header.Compared revisions:
master@5175a69→refactor/modern-infini-stack-review@b413816(2 commits, 8 files).Motivation
Provide the shared runtime primitives required by the modern InfiniLM stack: scoped device selection, reusable memory, and retained storage for recorded graphs. Keep these resource-management facilities in InfiniRT while supporting the coordinated Ascend and Cambricon integration in the matching stack branches.
The FP16 conversion change also aligns host conversion with round-to-nearest-even behavior used by CUDA.
Type of Change
feat- new feature / new backend capability / new public APIfix- bug fixperf- performance improvement without behavior changerefactor- code restructuring without behavior changetest- adding or fixing tests onlydocs- documentation onlybuild/ci- build system or CI configurationchore- tooling, formatting, or other non-code changes!in the Conventional Commits prefix or aBREAKING CHANGE:footer)Platforms Affected
WITH_CPU)WITH_NVIDIA)WITH_ILUVATAR)WITH_HYGON)WITH_METAX)WITH_MOORE)WITH_CAMBRICON)WITH_ASCEND)Smoke Build and Test Result
Test Results on Supported Platforms
Full `ctest` output (optional)
Benchmark / Performance Impact
Notes for Reviewers