Adjust cpp torch trt logging level with compiler option (#3181)

Author: keehyuna

py/torch_tensorrt/dynamo/utils.py

@@ -11,6 +11,7 @@
 from torch._subclasses.fake_tensor import FakeTensor
 from torch_tensorrt._Device import Device
 from torch_tensorrt._enums import dtype
+from torch_tensorrt._features import ENABLED_FEATURES
 from torch_tensorrt._Input import Input
 from torch_tensorrt.dynamo import _defaults
 from torch_tensorrt.dynamo._defaults import default_device
@@ -204,10 +205,27 @@ def set_log_level(parent_logger: Any, level: Any) -> None:
     Sets the log level to the user provided level.
     This is used to set debug logging at a global level
     at entry points of tracing, dynamo and torch_compile compilation.
+    And set log level for c++ torch trt logger if runtime is available.
     """
     if parent_logger:
         parent_logger.setLevel(level)
 
+    if ENABLED_FEATURES.torch_tensorrt_runtime:
+        if level == logging.DEBUG:
+            log_level = trt.ILogger.Severity.VERBOSE
+        elif level == logging.INFO:
+            log_level = trt.ILogger.Severity.INFO
+        elif level == logging.WARNING:
+            log_level = trt.ILogger.Severity.WARNING
+        elif level == logging.ERROR:
+            log_level = trt.ILogger.Severity.ERROR
+        elif level == logging.CRITICAL:
+            log_level = trt.ILogger.Severity.INTERNAL_ERROR
+        else:
+            raise AssertionError(f"{level} is not valid log level")
+
+        torch.ops.tensorrt.set_logging_level(int(log_level))
+
 
 def prepare_inputs(
     inputs: Input | torch.Tensor | Sequence[Any] | Dict[Any, Any],