apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 00:57:33 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Backend] support bechmark mode for runtime and backend (#1201)

Browse Source 
* [backend] support bechmark mode for runtime and backend

* [backend] support bechmark mode for runtime and backend

* [pybind11] add benchmark methods pybind

* [pybind11] add benchmark methods pybind

* [Other] Update build scripts

* [Other] Update cmake/summary.cmake

* [Other] update build scripts

* [Other] add ENABLE_BENCHMARK option -> setup.py

* optimize backend time recording

* optimize backend time recording

* optimize trt backend time record

* [backend] optimze backend_time recording for trt

* [benchmark] remove redundant logs

* fixed ov_backend confilct

* [benchmark] fixed paddle_backend conflicts

* [benchmark] fixed paddle_backend conflicts

* [benchmark] fixed paddle_backend conflicts

* [benchmark] remove use_gpu option from ort backend option

* [benchmark] update benchmark_ppdet.py

* [benchmark] update benchmark_ppcls.py

* fixed lite backend conflicts

* [Lite] fixed lite xpu

* add benchmark macro

* add RUNTIME_PROFILE_LOOP macros

* add comments for RUNTIME_PROFILE macros

* add comments for new apis

* add comments for new apis

* update benchmark_ppdet.py

* afixed bugs

* remove unused codes

* optimize RUNTIME_PROFILE_LOOP macros

* optimize RUNTIME_PROFILE_LOOP macros

* add comments for benchmark option and result

* add docs for benchmark namespace
This commit is contained in:
DefTruth
2023-02-06 14:29:35 +08:00
committed by GitHub
parent 42d14e7119
commit f73a538f61
34 changed files with 741 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
fastdeploy/runtime/backends/tensorrt/trt_backend.cc
View File

@@ -287,14 +287,18 @@ bool TrtBackend::Infer(std::vector<FDTensor>& inputs,
BuildTrtEngine();
}
RUNTIME_PROFILE_LOOP_H2D_D2H_BEGIN
cudaSetDevice(option_.gpu_id);
SetInputs(inputs);
AllocateOutputsBuffer(outputs, copy_to_fd);
RUNTIME_PROFILE_LOOP_BEGIN(1)
if (!context_->enqueueV2(bindings_.data(), stream_, nullptr)) {
FDERROR << "Failed to Infer with TensorRT." << std::endl;
return false;
}
RUNTIME_PROFILE_LOOP_END
for (size_t i = 0; i < outputs->size(); ++i) {
// if the final output tensor's dtype is different from the model output
// tensor's dtype, then we need cast the data to the final output's dtype
@@ -335,7 +339,7 @@ bool TrtBackend::Infer(std::vector<FDTensor>& inputs,
FDASSERT(cudaStreamSynchronize(stream_) == cudaSuccess,
"[ERROR] Error occurs while sync cuda stream.");
}
RUNTIME_PROFILE_LOOP_H2D_D2H_END
return true;
}