NVIDIA / NVIDIA/nvbench

Throughput statistics are not calculated when reads/writes are declared after `state.exec()`

Open
#175 0 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Cuda
Stars
927
Forks
123
Avg merge
2d 10h
Merged PRs (30d)
2

Description

The current implementation computes the throughput statistics in measure_cold, which is invoked during state.exec. This has the undesirable effect that throughput statistics are not generated when reads/writes are declared after state.exec.

The statistics are added here.

The only piece of information that is needed from measure_cold is the average CUDA time. However, this information is added to a summary here, so it should be possible to add a post-processing step to retrieve this information and compute throughput statistics after the KernelGenerator is finished.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start in nvbench/detail/measure_cold.cu, checking the throughput-statistics code at lines 199-224 and the average CUDA-time summary at lines 164-170. Trace how state.exec invokes measure_cold and how the KernelGenerator finishes. Done means throughput statistics are also generated when reads or writes are declared after state.exec().

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
performance
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.