[TensorRT EP] Switch to enqueueV3 with support DDS output #17751
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
chilo-ms
changed the title
jywu-msft
previously approved these changes
Dec 4, 2023
souptc
reviewed
Dec 4, 2023
| * | ||
| * \since Version 1.17. | ||
| */ | ||
| ORT_API2_STATUS(KernelContext_SetOutput, _Inout_ OrtKernelContext* context, _In_ size_t index, |
There was a problem hiding this comment.
KernelContext_SetOutput
why we want to expose it to CAPI? we have TRT based custom op?
compile api based EP's need to implement compute_func which only has access to public OrtKernelContext api , not the internal OpKernelContext api.
we need to use SetOutputMLValue() so that's why it's plumbed thru to the public api in this PR
|
this will be replaced by version which copies output rather than binds output to kernel context since we don't want to expose that api publicly. #18714 |
jywu-msft
added a commit
that referenced
this pull request
Dec 14, 2023
…on) (#18714) It's branched off from #17751 but removes KernelContext_SetOutput() API. It copies output allocation buffer to kernel context. --------- Co-authored-by: George Wu <[email protected]>
jywu-msft
pushed a commit
that referenced
this pull request
Jan 12, 2024
When the TRT engine cache (precompiled engine) is present, it doesn't make sense to go over the processes of model verification, model optimization, TRT EP's GetCapability(), TRT EP's model proto reconstruction, calling TRT parser and engine compilation. This PR makes TRT EP skip those processes and directly load the engine to perform inference. The feature request: #18072 Features: - Replace original model with TRT engine wrapped ONNX model. It can save a lot of time as mentioned above. - How to get TRT engine wrapped ONNX model? 1. Set `trt_dump_ep_context_model` provider option to "true" and run the inference. You will find the "xxx_wrapper.onnx" at the engine cache path. (The same logic of generating engine cache) 2. Use gen_trt_engine_wrapper_onnx_model.py - Three provider options are added, `trt_dump_ep_context_model`: Enable dump wrapped onnx model by TRT EP `trt_ep_context_embed_mode`: Add embed_mode as attribute. 0 means engine cache path, 1 means engine binary data. `trt_ep_context_compute_capability_enable`: Add hardware_arch as attribute. When running the model, TRT EP will check consistency between model's hardware_arch and GPU's compute capability. - When the engine cache path is given in the wrapped model, TRT EP will first search for the engine file using the path (relative to model path), if it can't find it, it will change to use the path as it is (depends on user, could be relative to working dir or absolute path) Note: 1. This PR includes the change of #17751 Constraints: 1. The whole model should be fully supported by TRT. 4. Users need to make sure the engine is built with min/max/opt optimization profiles that large enough to cover the range of all inputs. TRT EP will simply fail and won't rebuild the engine if the input shape is out of range during runtime.
mszhanyi
pushed a commit
that referenced
this pull request
Jan 15, 2024
When the TRT engine cache (precompiled engine) is present, it doesn't make sense to go over the processes of model verification, model optimization, TRT EP's GetCapability(), TRT EP's model proto reconstruction, calling TRT parser and engine compilation. This PR makes TRT EP skip those processes and directly load the engine to perform inference. The feature request: #18072 Features: - Replace original model with TRT engine wrapped ONNX model. It can save a lot of time as mentioned above. - How to get TRT engine wrapped ONNX model? 1. Set `trt_dump_ep_context_model` provider option to "true" and run the inference. You will find the "xxx_wrapper.onnx" at the engine cache path. (The same logic of generating engine cache) 2. Use gen_trt_engine_wrapper_onnx_model.py - Three provider options are added, `trt_dump_ep_context_model`: Enable dump wrapped onnx model by TRT EP `trt_ep_context_embed_mode`: Add embed_mode as attribute. 0 means engine cache path, 1 means engine binary data. `trt_ep_context_compute_capability_enable`: Add hardware_arch as attribute. When running the model, TRT EP will check consistency between model's hardware_arch and GPU's compute capability. - When the engine cache path is given in the wrapped model, TRT EP will first search for the engine file using the path (relative to model path), if it can't find it, it will change to use the path as it is (depends on user, could be relative to working dir or absolute path) Note: 1. This PR includes the change of #17751 Constraints: 1. The whole model should be fully supported by TRT. 4. Users need to make sure the engine is built with min/max/opt optimization profiles that large enough to cover the range of all inputs. TRT EP will simply fail and won't rebuild the engine if the input shape is out of range during runtime.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There are 2 phases to switch to enqueueV3. This PR is the 2nd phases. (The 1st phase PR is here)
One of the ways TRT handles data-dependent shape (DDS) output is relying on user to provide an allocator as a callback. TRT calls this allocator when it knows the shape of the tensor during runtime to allocate output memory. So, here, we need a way to bind the allocation output to the kernel context output.
"If the output tensor has data-dependent shape, TRT EP will provide an IOutputAllocator for enqueueV3 to dynamically allocate memory buffer.
Once enqueueV3 returns, TRT EP will then bind the output allocation to ORT kernel context output.
(Please note that we take strategy A mentioned in https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#dynamic-shaped-output,
which we defer allocation until the size is known and don't call IExecution::setTensorAddress)
Otherwise, if the shape of the output tensor is known prior to the runtime, ORT will pre-allocate memory buffer for the output tensor for enqueueV3."
Add new ORT
KernelContext_SetOutput()API where it calls the existedSetOutputMLValue()which only being used by training before.This is because Compiled based EP's can only use the public OrtKernelContext api's and not the internal OpKernelContext api.