tiandiao123 commented on a change in pull request #8808:
URL: https://github.com/apache/tvm/pull/8808#discussion_r697720446
##########
File path: src/runtime/contrib/tensorrt/tensorrt_runtime.cc
##########
@@ -267,13 +320,68 @@ class TensorRTRuntime : public JSONRuntimeBase {
}
// Build engine.
- trt_engine_cache_[std::make_pair(symbol_name_, batch_size)] =
builder.BuildEngine();
- DLOG(INFO) << "Finished building TensorRT engine for subgraph " <<
symbol_name_
+ // trt_engine_cache_[std::make_pair(symbol_name_, batch_size)] =
builder.BuildEngine();
+ const bool use_int8 = (dmlc::GetEnv("TVM_TENSORRT_USE_INT8", 0) != 0);
+ TensorRTEngineAndContext engine_and_context = builder.BuildEngine();
+ trt_engine_cache_[std::make_pair(symbol_name_, batch_size)] =
engine_and_context;
+ if(use_int8 == true){
+ if(calibrator_ == nullptr){
+ this->CreateCalibratorIfUsingInt8(engine_and_context);
+ }
+
+ if(num_calibration_batches_remaining_ == 0){
+ engine_and_context.context->destroy();
+ engine_and_context.engine->destroy();
+
+ LOG(INFO)<<"rebuild builder using int8 mode";
+ TensorRTBuilder builder2(&logger_, data_entry_, max_workspace_size_,
use_implicit_batch_,
Review comment:
Hi @trevor-m, if we don't build the engine from scratch, how can we
update original builder? I didn't see any functions in tensorrt_builder to
rebuild engine using calibrator. I am a little confused here. I saw you
originally re-build it using the same call, but re-load graph based on that
json file, but that function call (builder.BuildEngine(expr)) has been deleted
in the current tensorrt_builder.h
(https://github.com/trevor-m/tvm/blob/trevmorr-calibrate-int8/src/runtime/contrib/tensorrt/tensorrt_module.cc#L257).
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
