add req_full_compilation_arg

add arg require_full_compilation

Author: apbose

py/torch_tensorrt/dynamo/_compiler.py

@@ -376,6 +376,7 @@ def contains_metadata(gm: torch.fx.GraphModule) -> bool:
                 verbose=settings.debug,
                 min_block_size=settings.min_block_size,
                 torch_executed_ops=settings.torch_executed_ops,
+                require_full_compilation=settings.require_full_compilation,
             )
         except torch.fx.passes.splitter_base.FxNetSplitterInternalError:
             logger.error(
@@ -394,6 +395,7 @@ def contains_metadata(gm: torch.fx.GraphModule) -> bool:
             verbose=settings.debug,
             min_block_size=settings.min_block_size,
             torch_executed_ops=settings.torch_executed_ops,
+            require_full_compilation=settings.require_full_compilation,
         )
 
     dryrun_tracker.unsupported_ops = supported_ops.unsupported_operators

py/torch_tensorrt/dynamo/backend/backends.py

@@ -107,6 +107,10 @@ def _pretraced_backend(
             torchtrt_inputs = prepare_inputs(
                 torch_inputs, disable_memory_format_check=True
             )
+            if settings.require_full_compilation:
+                logger.warning(
+                    "require_full_compilation arg is not applicable for torch.compile with backend='torch_tensorrt"
+                )
             trt_compiled = compile_module(
                 gm,
                 torchtrt_inputs,