From a0bd7fac4efc601d9560980ce725c79d087d429f Mon Sep 17 00:00:00 2001 From: pytorchbot Date: Tue, 23 Apr 2024 17:01:42 -0700 Subject: [PATCH] Update Profiling Section in XNNPACK Delegate Docs (#3237) (#3261) Summary: Pull Request resolved: https://github.com/pytorch/executorch/pull/3237 Updating Profiling Section of the docs Main point is pointing the the SDK Profiling Tutorial on how to get XNNPACK profiling information Reviewed By: metascroy, cccclai Differential Revision: D56439491 fbshipit-source-id: 1d724ffae6d89e8769ea427cb37b4ec85fe3452f (cherry picked from commit 329184ac28f3fea41dc05afb342ad934403a8dbc) Co-authored-by: Max Ren --- .../native-delegates-executorch-xnnpack-delegate.md | 10 +--------- 1 file changed, 1 insertion(+), 9 deletions(-) diff --git a/docs/source/native-delegates-executorch-xnnpack-delegate.md b/docs/source/native-delegates-executorch-xnnpack-delegate.md index 12b2e9c2ba..1d12daef9d 100644 --- a/docs/source/native-delegates-executorch-xnnpack-delegate.md +++ b/docs/source/native-delegates-executorch-xnnpack-delegate.md @@ -74,16 +74,8 @@ Since weight packing creates an extra copy of the weights inside XNNPACK, We fre When executing the XNNPACK subgraphs, we prepare the tensor inputs and outputs and feed them to the XNNPACK runtime graph. After executing the runtime graph, the output pointers are filled with the computed tensors. #### **Profiling** -We have enabled basic profiling for XNNPACK delegate that can be enabled with the following compiler flag `-DENABLE_XNNPACK_PROFILING`. After running the model it will produce basic per-op and total timings. We provide an example of the profiling below. The timings listed are the average across runs, and the units are in microseconds. +We have enabled basic profiling for XNNPACK delegate that can be enabled with the following compiler flag `-DENABLE_XNNPACK_PROFILING`. With ExecuTorch's SDK integration, you can also now use the SDK tools to profile the model. You can follow the steps in [Using the ExecuTorch SDK to Profile a Model](./tutorials/sdk-integration-tutorial) on how to profile ExecuTorch models and use SDK's Inspector API to view XNNPACK's internal profiling information. -``` -Fully Connected (NC, F32) GEMM: 109.510002 -Total Time: 109.510002 -``` - -::::{note} -Profiling is a work in progress, and is planned to be integrated with [SDK Tools](sdk-delegate-integration.md) and Tensorboard. -:::: [comment]: <> (TODO: Refactor quantizer to a more official quantization doc) ## Quantization