Skip to content

Feature Request: Proper Support for Multiple Metrics in @nsight.analyze.kernel #8

Description

@ConvolutedDog

Hi, I'd like to request proper support for specifying multiple metrics in the @nsight.analyze.kernel decorator. Currently, when providing a comma-separated list of metrics, the data aggregation doesn't work correctly.

Being able to collect multiple related metrics in a single profiling run is essential for performance analysis. For example, comparing both load and store operations (smsp__sass_inst_executed_op_shared_ld.sum and smsp__sass_inst_executed_op_shared_st.sum) in one run is much more convenient than running separate profiling sessions for each metric.

Users currently need to run separate profiling sessions for each metric, which is cumbersome and time-consuming:

@nsight.analyze.kernel(runs=1, metric="smsp__sass_inst_executed_op_shared_ld.sum")
def profile1(...): ...

@nsight.analyze.kernel(runs=1, metric="smsp__sass_inst_executed_op_shared_st.sum")
def profile2(...): ...

It'll be convenient if all the metrics could be separated by commas:

@nsight.analyze.kernel(runs=1, metric="smsp__sass_inst_executed_op_shared_ld.sum,smsp__sass_inst_executed_op_shared_st.sum")
def profile1(...): ...

Activity

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

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions