search

pytorch / TensorRT

PyTorch/TorchScript/FX compiler for NVIDIA GPUs using TensorRT
https://pytorch.org/TensorRT
BSD 3-Clause "New" or "Revised" License
2.53k stars 349 forks source link

🐛 [Bug] Engine cache does not take shape change into account #3148

Closed HolyWu closed 2 weeks ago

HolyWu commented 3 weeks ago

To Reproduce

import os
from tempfile import gettempdir

import torch
import torch_tensorrt

class MyModule(torch.nn.Module):
    def forward(self, x):
        return torch.relu(x)

with torch.inference_mode():
    device = torch.device("cuda", 0)
    model = MyModule().eval().to(device)

    inputs1 = [torch.randn(1, 3, 224, 224, device=device)]
    trt_model1 = torch_tensorrt.compile(
        model,
        ir="dynamo",
        inputs=inputs1,
        enabled_precisions={torch.float},
        device=device,
        make_refitable=True,
        debug=True,
        min_block_size=1,
        engine_cache_dir=os.path.join(gettempdir(), "torchtrt_issue3148"),
    )
    trt_model1(*inputs1)

    print("\n========================================\n")

    inputs2 = [torch.randn(2, 3, 224, 224, device=device)]
    trt_model2 = torch_tensorrt.compile(
        model,
        ir="dynamo",
        inputs=inputs2,
        enabled_precisions={torch.float},
        device=device,
        make_refitable=True,
        debug=True,
        min_block_size=1,
        engine_cache_dir=os.path.join(gettempdir(), "torchtrt_issue3148"),
    )
    trt_model2(*inputs2)
DEBUG:torch_tensorrt.dynamo.lowering.passes.remove_detach:Removed 0 detach nodes:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo._compiler:Input graph: graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo.lowering.passes.constant_folding:Graph after constant folding:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo.lowering.passes.remove_assert_scalar:Removed 0 assert_scalar nodes:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo._compiler:Lowered Input graph: graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
INFO:torch_tensorrt.dynamo._engine_cache:Disk engine cache initialized (cache directory:/tmp/torchtrt_issue3148, max size: 1073741824)
INFO:torch_tensorrt.dynamo._compiler:Compilation Settings: CompilationSettings(enabled_precisions={<dtype.f32: 7>}, debug=True, workspace_size=0, min_block_size=1, torch_executed_ops=set(), pass_through_build_failures=False, max_aux_streams=None, version_compatible=False, optimization_level=None, use_python_runtime=False, truncate_double=False, use_fast_partitioner=True, enable_experimental_decompositions=False, device=Device(type=DeviceType.GPU, gpu_id=0), require_full_compilation=False, disable_tf32=False, assume_dynamic_shape_support=False, sparse_weights=False, make_refitable=True, engine_capability=<EngineCapability.STANDARD: 1>, num_avg_timing_iters=1, dla_sram_size=1048576, dla_local_dram_size=1073741824, dla_global_dram_size=536870912, dryrun=False, hardware_compatible=False, timing_cache_path='/tmp/torch_tensorrt_engine_cache/timing_cache.bin', lazy_engine_init=False, cache_built_engines=True, reuse_cached_engines=True)

DEBUG:torch_tensorrt.dynamo.partitioning._global_partitioner:
Supported Nodes:
- torch.ops.aten.relu.default + Operator Count: 1

DEBUG:torch_tensorrt.dynamo.partitioning._global_partitioner:
All Nodes Supported

DEBUG:torch_tensorrt.dynamo._compiler:Detected support for 1 operators out of 1 in subgraph.
INFO:torch_tensorrt.dynamo._compiler:Partitioning the graph via the fast partitioner
DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
Number of TensorRT-Accelerated Engines Generated: 1
DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
Supported Nodes:
- torch.ops.aten.relu.default + Operator Count: 1

DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
All Nodes Supported

DEBUG:torch_tensorrt.dynamo._compiler:Converting submodule: _run_on_acc_0
 Input shapes: [(1, 3, 224, 224)]
 graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return relu
WARNING:torch_tensorrt.dynamo.utils:Could not detect the device on which the model exists. Assuming the model is on CPU
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node x (kind: x, args: ())
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Adding input to in-progress INetwork: x [shape=[1, 3, 224, 224], dtype=DataType.FLOAT]
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node x [x] (Inputs: () | Outputs: (x: (1, 3, 224, 224)@torch.float32))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node /relu (kind: aten.relu.default, args: ('x <Node>',))
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node /relu [aten.relu.default] (Inputs: (x: (1, 3, 224, 224)@torch.float32) | Outputs: (relu: (1, 3, 224, 224)@torch.float32))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node output (kind: output, args: ('relu <Node>',))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Marking output output0 [shape=(1, 3, 224, 224), dtype=DataType.FLOAT]
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node output [output] (Inputs: (relu: (1, 3, 224, 224)@torch.float32) | Outputs: (output: ))
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:TRT INetwork construction elapsed time: 0:00:00.004144
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Building weight name mapping...
WARNING:torch_tensorrt.dynamo.utils:Could not detect the device on which the model exists. Assuming the model is on CPU
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Not found cached TRT engines. Start building engine.
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Build TRT engine elapsed time: 0:00:03.345116
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:TRT Engine uses: 12932 bytes of Memory
DEBUG:torch_tensorrt.dynamo._engine_cache:The engine added to cache, saved to /tmp/torchtrt_issue3148/qcp2nbn7adw2zbhxzqql4er37brkw33awbzf4jqx33pbhuvino3/blob.bin

DEBUG:torch_tensorrt.dynamo._DryRunTracker:
++++++++++++++++++++++++++++++++++++++++++++++++++ Dry-Run Results for Graph ++++++++++++++++++++++++++++++++++++++++++++++++++

The graph consists of 1 Total Operators, of which 1 operators are supported, 100.0% coverage

Compiled with: CompilationSettings(enabled_precisions={<dtype.f32: 7>}, debug=True, workspace_size=0, min_block_size=1, torch_executed_ops=set(), pass_through_build_failures=False, max_aux_streams=None, version_compatible=False, optimization_level=None, use_python_runtime=False, truncate_double=False, use_fast_partitioner=True, enable_experimental_decompositions=False, device=Device(type=DeviceType.GPU, gpu_id=0), require_full_compilation=False, disable_tf32=False, assume_dynamic_shape_support=False, sparse_weights=False, make_refitable=True, engine_capability=<EngineCapability.STANDARD: 1>, num_avg_timing_iters=1, dla_sram_size=1048576, dla_local_dram_size=1073741824, dla_global_dram_size=536870912, dryrun=False, hardware_compatible=False, timing_cache_path='/tmp/torch_tensorrt_engine_cache/timing_cache.bin', lazy_engine_init=False, cache_built_engines=True, reuse_cached_engines=True)

  Graph Structure:

   Inputs: List[Tensor: (1, 3, 224, 224)@float32]
    ...
    TRT Engine #1 - Submodule name: _run_on_acc_0
     Engine Inputs: List[Tensor: (1, 3, 224, 224)@float32]
     Number of Operators in Engine: 1
     Engine Outputs: List[Tensor: (1, 3, 224, 224)@float32]
    ...
   Outputs: List[Tensor: (1, 3, 224, 224)@float32]

  ------------------------- Aggregate Stats -------------------------

   Average Number of Operators per TRT Engine: 1.0
   Most Operators in a TRT Engine: 1

  ********** Recommendations **********

   - For minimal graph segmentation, select min_block_size=1 which would generate 1 TRT engine(s)
   - The current level of graph segmentation is equivalent to selecting min_block_size=1 which generates 1 TRT engine(s)

========================================

DEBUG:torch_tensorrt.dynamo.lowering.passes.remove_detach:Removed 0 detach nodes:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo._compiler:Input graph: graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo.lowering.passes.constant_folding:Graph after constant folding:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo.lowering.passes.remove_assert_scalar:Removed 0 assert_scalar nodes:
graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
DEBUG:torch_tensorrt.dynamo._compiler:Lowered Input graph: graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return (relu,)
INFO:torch_tensorrt.dynamo._engine_cache:Disk engine cache initialized (cache directory:/tmp/torchtrt_issue3148, max size: 1073741824)
INFO:torch_tensorrt.dynamo._compiler:Compilation Settings: CompilationSettings(enabled_precisions={<dtype.f32: 7>}, debug=True, workspace_size=0, min_block_size=1, torch_executed_ops=set(), pass_through_build_failures=False, max_aux_streams=None, version_compatible=False, optimization_level=None, use_python_runtime=False, truncate_double=False, use_fast_partitioner=True, enable_experimental_decompositions=False, device=Device(type=DeviceType.GPU, gpu_id=0), require_full_compilation=False, disable_tf32=False, assume_dynamic_shape_support=False, sparse_weights=False, make_refitable=True, engine_capability=<EngineCapability.STANDARD: 1>, num_avg_timing_iters=1, dla_sram_size=1048576, dla_local_dram_size=1073741824, dla_global_dram_size=536870912, dryrun=False, hardware_compatible=False, timing_cache_path='/tmp/torch_tensorrt_engine_cache/timing_cache.bin', lazy_engine_init=False, cache_built_engines=True, reuse_cached_engines=True)

DEBUG:torch_tensorrt.dynamo.partitioning._global_partitioner:
Supported Nodes:
- torch.ops.aten.relu.default + Operator Count: 1

DEBUG:torch_tensorrt.dynamo.partitioning._global_partitioner:
All Nodes Supported

DEBUG:torch_tensorrt.dynamo._compiler:Detected support for 1 operators out of 1 in subgraph.
INFO:torch_tensorrt.dynamo._compiler:Partitioning the graph via the fast partitioner
DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
Number of TensorRT-Accelerated Engines Generated: 1
DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
Supported Nodes:
- torch.ops.aten.relu.default + Operator Count: 1

DEBUG:torch_tensorrt.dynamo.partitioning._adjacency_partitioner:
All Nodes Supported

DEBUG:torch_tensorrt.dynamo._compiler:Converting submodule: _run_on_acc_0
 Input shapes: [(2, 3, 224, 224)]
 graph():
    %x : [num_users=1] = placeholder[target=x]
    %relu : [num_users=1] = call_function[target=torch.ops.aten.relu.default](args = (%x,), kwargs = {})
    return relu
WARNING:torch_tensorrt.dynamo.utils:Could not detect the device on which the model exists. Assuming the model is on CPU
DEBUG:torch_tensorrt.dynamo._engine_cache:Engine found in cache, loaded from /tmp/torchtrt_issue3148/qcp2nbn7adw2zbhxzqql4er37brkw33awbzf4jqx33pbhuvino3/blob.bin
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Found the cached engine that corresponds to this graph. It is directly loaded.
WARNING:torch_tensorrt.dynamo.utils:Could not detect the device on which the model exists. Assuming the model is on CPU
WARNING:torch_tensorrt.dynamo.utils:Could not detect the device on which the model exists. Assuming the model is on CPU
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node x (kind: x, args: ())
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Adding input to in-progress INetwork: x [shape=[2, 3, 224, 224], dtype=DataType.FLOAT]
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node x [x] (Inputs: () | Outputs: (x: (2, 3, 224, 224)@torch.float32))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node /relu (kind: aten.relu.default, args: ('x <Node>',))
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node /relu [aten.relu.default] (Inputs: (x: (2, 3, 224, 224)@torch.float32) | Outputs: (relu: (2, 3, 224, 224)@torch.float32))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converting node output (kind: output, args: ('relu <Node>',))
DEBUG:torch_tensorrt.dynamo.conversion._TRTInterpreter:Marking output output0 [shape=(2, 3, 224, 224), dtype=DataType.FLOAT]
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:Converted node output [output] (Inputs: (relu: (2, 3, 224, 224)@torch.float32) | Outputs: (output: ))
INFO:torch_tensorrt.dynamo.conversion._TRTInterpreter:TRT INetwork construction elapsed time: 0:00:00.001975
DEBUG:torch_tensorrt.dynamo._DryRunTracker:
++++++++++++++++++++++++++++++++++++++++++++++++++ Dry-Run Results for Graph ++++++++++++++++++++++++++++++++++++++++++++++++++

The graph consists of 1 Total Operators, of which 1 operators are supported, 100.0% coverage

Compiled with: CompilationSettings(enabled_precisions={<dtype.f32: 7>}, debug=True, workspace_size=0, min_block_size=1, torch_executed_ops=set(), pass_through_build_failures=False, max_aux_streams=None, version_compatible=False, optimization_level=None, use_python_runtime=False, truncate_double=False, use_fast_partitioner=True, enable_experimental_decompositions=False, device=Device(type=DeviceType.GPU, gpu_id=0), require_full_compilation=False, disable_tf32=False, assume_dynamic_shape_support=False, sparse_weights=False, make_refitable=True, engine_capability=<EngineCapability.STANDARD: 1>, num_avg_timing_iters=1, dla_sram_size=1048576, dla_local_dram_size=1073741824, dla_global_dram_size=536870912, dryrun=False, hardware_compatible=False, timing_cache_path='/tmp/torch_tensorrt_engine_cache/timing_cache.bin', lazy_engine_init=False, cache_built_engines=True, reuse_cached_engines=True)

  Graph Structure:

   Inputs: List[Tensor: (2, 3, 224, 224)@float32]
    ...
    TRT Engine #1 - Submodule name: _run_on_acc_0
     Engine Inputs: List[Tensor: (2, 3, 224, 224)@float32]
     Number of Operators in Engine: 1
     Engine Outputs: List[Tensor: (2, 3, 224, 224)@float32]
    ...
   Outputs: List[Tensor: (2, 3, 224, 224)@float32]

  ------------------------- Aggregate Stats -------------------------

   Average Number of Operators per TRT Engine: 1.0
   Most Operators in a TRT Engine: 1

  ********** Recommendations **********

   - For minimal graph segmentation, select min_block_size=1 which would generate 1 TRT engine(s)
   - The current level of graph segmentation is equivalent to selecting min_block_size=1 which generates 1 TRT engine(s)
ERROR: [Torch-TensorRT] - IExecutionContext::setInputShape: Error Code 3: API Usage Error (Parameter check failed, condition: engineDims.d[i] == dims.d[i]. Static dimension mismatch while setting input shape for x. Set dimensions are [2,3,224,224]. Expected dimensions are [1,3,224,224].)
Traceback (most recent call last):
  File "/home/holywu/test.py", line 45, in <module>
    trt_model2(*inputs2)
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/fx/graph_module.py", line 784, in call_wrapped
    return self._wrapped_call(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/fx/graph_module.py", line 361, in __call__
    raise e
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/fx/graph_module.py", line 348, in __call__
    return super(self.cls, obj).__call__(*args, **kwargs)  # type: ignore[misc]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<eval_with_key>.85", line 6, in forward
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1736, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1747, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch_tensorrt/_features.py", line 56, in wrapper
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.py", line 279, in forward
    outputs: List[torch.Tensor] = torch.ops.tensorrt.execute_engine(
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/holywu/.local/lib/python3.12/site-packages/torch/_ops.py", line 1116, in __call__
    return self._op(*args, **(kwargs or {}))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: [Error thrown at core/runtime/execute_engine.cpp:231] Expected compiled_engine->exec_ctx->setInputShape(name.c_str(), dims) to be true but got false
Error while setting the input shape

Environment

zewenli98 commented 2 weeks ago

Can you try specifying a dim as dynamic? like https://github.com/pytorch/TensorRT/blob/main/examples/dynamo/engine_caching_example.py#L132-L136

HolyWu commented 2 weeks ago

No issue if the model was compiled with dynamic input spec in the first place.