Skip to content

graph: backend: dnnl: fix nthr in decomp kernels under threadpool rt#5337

Open
TaoLv wants to merge 1 commit into
mainfrom
lvtao/main/fix-thrpool-decomp-kernel
Open

graph: backend: dnnl: fix nthr in decomp kernels under threadpool rt#5337
TaoLv wants to merge 1 commit into
mainfrom
lvtao/main/fix-thrpool-decomp-kernel

Conversation

@TaoLv

@TaoLv TaoLv commented Jun 16, 2026

Copy link
Copy Markdown
Contributor

Properly get the number of thread (nthr) under threadpool runtime, which is needed to estimate the scratchpad size for decomposition kernels.

Previously, nthr was queried during execution time and used for internal scratchpad allocation and parallel-for.
Now with the fix, nthr is queried during compilation time and used for scratchpad size estimation. It's stored and used for parallel-for during execution time.

@TaoLv TaoLv requested a review from a team as a code owner June 16, 2026 13:37
@github-actions github-actions Bot added the component:graph-api Codeowner: @oneapi-src/onednn-graph label Jun 16, 2026
@TaoLv TaoLv force-pushed the lvtao/main/fix-thrpool-decomp-kernel branch from f48e9b4 to b440c27 Compare June 17, 2026 02:47
Need proper nthr for scratchpad size estimation.
@TaoLv

TaoLv commented Jun 17, 2026

Copy link
Copy Markdown
Contributor Author

make test
set test_scope=NIGHTLY
disable benchdnn_all
enable benchdnn_graph
enable compiler_clang19
enable build_cpu_runtime_thrpool_eigen
enable build_cpu_runtime_thrpool_eigen_async

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

component:graph-api Codeowner: @oneapi-src/onednn-graph

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants