From e394a8e045cd2138914d965770baf9368d63acaf Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Maximilian=20M=C3=BCller?=
 <44298237+gedoensmax@users.noreply.github.com>
Date: Wed, 10 Jan 2024 02:11:48 +0100
Subject: [PATCH] [Docs] Adding CUDA Graph docs for TRT (#19064)

### Description

Adding docs for the flag `trt_cuda_graph_enable`.
---
 docs/execution-providers/TensorRT-ExecutionProvider.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/docs/execution-providers/TensorRT-ExecutionProvider.md b/docs/execution-providers/TensorRT-ExecutionProvider.md
index af6b673c78898..89fdb895a05d2 100644
--- a/docs/execution-providers/TensorRT-ExecutionProvider.md
+++ b/docs/execution-providers/TensorRT-ExecutionProvider.md
@@ -99,6 +99,7 @@ There are two ways to configure TensorRT settings, either by **TensorRT Executio
 | trt_profile_min_shapes                | ORT_TENSORRT_PROFILE_MIN_SHAPES                | string |
 | trt_profile_max_shapes                | ORT_TENSORRT_PROFILE_MAX_SHAPES                | string |
 | trt_profile_opt_shapes                | ORT_TENSORRT_PROFILE_OPT_SHAPES                | string |
+| trt_cuda_graph_enable                 | ORT_TENSORRT_CUDA_GRAPH_ENABLE                 | bool   |
 
 > Note: for bool type options, assign them with **True**/**False** in python, or **1**/**0** in C++.
 
@@ -179,6 +180,8 @@ TensorRT configurations can be set by execution provider options. It's useful wh
 
 * `trt_build_heuristics_enable`: Build engine using heuristics to reduce build time.
 
+* `trt_cuda_graph_enable`: This will capture a [CUDA graph](https://developer.nvidia.com/blog/cuda-graphs/) which can drastically help for a network with many small layers as it reduces launch overhead on the CPU.
+
 * `trt_sparsity_enable`: Control if sparsity can be used by TRT.
   * Check `--sparsity` in `trtexec` command-line flags for [details](https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#trtexec-flags).