ptrendx commented on a change in pull request #19011:
URL: https://github.com/apache/incubator-mxnet/pull/19011#discussion_r486609772
##########
File path: src/operator/subgraph/tensorrt/onnx_to_tensorrt.cc
##########
@@ -112,18 +117,81 @@ std::tuple<unique_ptr<nvinfer1::ICudaEngine>,
}
throw dmlc::Error("Cannot parse ONNX into TensorRT Engine");
}
- if (dmlc::GetEnv("MXNET_TENSORRT_USE_FP16", true)) {
+ trt_builder->setMaxBatchSize(max_batch_size);
+ std::future<onnx_to_tensorrt::unique_ptr<nvinfer1::ICudaEngine>>
future_int8_engine;
+#if NV_TENSORRT_MAJOR > 6
+ auto builder_config = InferObject(trt_builder->createBuilderConfig());
+
+ if (fp16_mode) {
+ if (trt_builder->platformHasFastFp16()) {
+ builder_config->setFlag(nvinfer1::BuilderFlag::kFP16);
+ } else {
+ LOG(WARNING) << "TensorRT can't use fp16 on this platform";
+ }
+ }
+
+ builder_config->setMaxWorkspaceSize(max_workspace_size);
+ builder_config->setFlag(nvinfer1::BuilderFlag::kDEBUG);
Review comment:
Is this intentional?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
