[CPU] Setting parallel to tbb static partitioner or tbb auto partitoner by thread pool #30167
Conversation
github-actions
bot
added
category: Core
github-actions
bot
removed
category: Core
src/plugins/intel_cpu/src/nodes/executors/dnnl/dnnl_matmul_primitive.cpp
Outdated
Show resolved
Hide resolved
| constexpr int default_multiplier = 32; | ||
| class ThreadPool; | ||
|
|
||
| class CpuParallel { |
There was a problem hiding this comment.
NIT. Could you please move it inside the class as a static field?
class CpuParallel {
public:
static constexpr int default_multiplier = 32;
public:
CpuParallel() = delete;
....
};There was a problem hiding this comment.
Ok moved it inside CpuParallel class.
| template <typename T0, typename R, typename F> | ||
| [[nodiscard]] R cpu_parallel_sum(const T0& D0, const R& input, const F& func) const { | ||
| #if OV_THREAD == OV_THREAD_TBB_ADAPTIVE |
There was a problem hiding this comment.
Can we move all the cpu_parallel_... methods into the private section? Because with this design, the cpu_parallel_... methods and the parallel_... methods of the same class may be used interchangeably, which would introduce a mess into the code base. Better to avoid such a situation and use the uniform name convention.
There was a problem hiding this comment.
Yes, you are right, moved these functions into private section.
| // cpu stream executor for current graph | ||
| ov::threading::CPUStreamsExecutor::Ptr m_cpuStreamExecutor; | ||
| std::shared_ptr<CpuParallel> m_cpuParallel = nullptr; | ||
| // std::shared_ptr<ThreadPool> m_threadPool = nullptr; |
|
@sunxiaoxia2022 , could you please fix the CI errors? |
maxnick
changed the title
…xiaoxia2022/openvino into xiaoxia/auto_tbb_thread_pool
|
@sunxiaoxia2022 , please resolve the merge conflicts. |
8 participants
Details:
Tickets: