You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by "MPolaris (via GitHub)" <gi...@apache.org> on 2025/09/28 10:34:38 UTC
[I] [Bug] Did you forget to bind? for cuda backend. [tvm]
MPolaris opened a new issue, #18350:
URL: https://github.com/apache/tvm/issues/18350
I built a very simple model with only one Linear layer, and encountered the error 'Did you forget to bind?' when turning to TVM through ONNX.
May I ask if I have made any operational errors?
Building Python code:
```python
# import torch
# import torch.nn as nn
# class MyModule(nn.Module):
# def __init__(self):
# super().__init__()
# self.linear = nn.Linear(8, 16)
# def forward(self, x):
# return self.linear(x)
# model = MyModule()
# torch.onnx.export(model, torch.randn(1, 8), "model.onnx", opset=13, input_names=["input"], output_names=["output"])
import tvm
import onnx
import numpy as np
import tvm
from tvm import relax
from tvm.relax.frontend.onnx import from_onnx
print(tvm.cuda().exist)
onnx_model = onnx.load("/home/xiaohe/codes/run_tvm/model.onnx")
target = "llvm"
shape_dict = {"input": (1, 8)}
mod = from_onnx(onnx_model, shape_dict, keep_params_in_input=True)
del onnx_model
mod, params = relax.frontend.detach_params(mod)
mod = relax.transform.BundleModelParams()(mod)
target = tvm.target.Target("cuda") # llvm is ok
with target:
lib = relax.build(mod, target=target)
lib.export_library("/home/xiaohe/codes/run_tvm/model.so")
```
TVM compilation process:
```bash
cd tvm
rm -rf build && mkdir build && cd build
cp ../cmake/config.cmake .
echo "set(USE_LLVM \"llvm-config --ignore-libllvm --link-static\")" >> config.cmake
echo "set(HIDE_PRIVATE_SYMBOLS ON)" >> config.cmake
echo "set(USE_CUDA ON)" >> config.cmake
cmake .. && cmake --build . --parallel $(nproc)
```
Complete error message:
```txt
Traceback (most recent call last):
File "/home/codes/run_tvm/gen_onnx.py", line 36, in <module>
lib = relax.build(mod, target=target)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/codes/tvm/python/tvm/relax/vm_build.py", line 259, in build
return _vmlink(
^^^^^^^^
File "/home/codes/tvm/python/tvm/relax/vm_build.py", line 154, in _vmlink
lib = tvm.tir.build(tir_mod, target=target, pipeline=tir_pipeline)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/codes/tvm/python/tvm/tir/build.py", line 173, in build
mod = pipeline(mod)
^^^^^^^^^^^^^
File "/home/codes/tvm/python/tvm/ir/transform.py", line 238, in __call__
return _ffi_transform_api.RunPass(self, mod)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "tvm/_ffi/_cython/packed_func.pxi", line 339, in tvm._ffi._cy3.core.PackedFuncBase.__call__
File "tvm/_ffi/_cython/packed_func.pxi", line 270, in tvm._ffi._cy3.core.FuncCall
File "tvm/_ffi/_cython/packed_func.pxi", line 259, in tvm._ffi._cy3.core.FuncCall3
File "tvm/_ffi/_cython/base.pxi", line 185, in tvm._ffi._cy3.core.CHECK_CALL
File "/home/codes/tvm/python/tvm/_ffi/base.py", line 468, in raise_last_ffi_error
raise py_err
File "tvm/_ffi/_cython/packed_func.pxi", line 56, in tvm._ffi._cy3.core.tvm_callback
File "/home/codes/tvm/python/tvm/tir/pipeline.py", line 122, in _pipeline
mod = tvm.ir.transform.Sequential(passes)(mod)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/codes/tvm/python/tvm/ir/transform.py", line 238, in __call__
return _ffi_transform_api.RunPass(self, mod)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "tvm/_ffi/_cython/packed_func.pxi", line 339, in tvm._ffi._cy3.core.PackedFuncBase.__call__
File "tvm/_ffi/_cython/packed_func.pxi", line 270, in tvm._ffi._cy3.core.FuncCall
File "tvm/_ffi/_cython/packed_func.pxi", line 259, in tvm._ffi._cy3.core.FuncCall3
File "tvm/_ffi/_cython/base.pxi", line 185, in tvm._ffi._cy3.core.CHECK_CALL
File "/home/codes/tvm/src/tir/analysis/verify_memory.cc", line 203, in operator()
LOG(FATAL) << "RuntimeError: Memory verification failed with the following errors:\n"
^^^^^^^^^^^^^^^^^^^^^^^^^^^
tvm._ffi.base.TVMError: Traceback (most recent call last):
0: operator()
at /home/codes/tvm/src/tir/analysis/verify_memory.cc:203
Did you forget to bind?
Variable `lv` is directly accessed by host memory (it is not contained in a thread environment or in the function arguments.
Variable `input` is directly accessed by host memory (it is not contained in a thread environment or in the function arguments.
Variable `matmul` is directly accessed by host memory (it is not contained in a thread environment or in the function arguments.
Variable `matmul` is directly accessed by host memory (it is not contained in a thread environment or in the function arguments.
Variable `matmul` is directly accessed by host memory (it is not contained in a thread environment or in the function arguments.
File "/home/codes/tvm/src/tir/analysis/verify_memory.cc", line 203
RuntimeError: Memory verification failed with the following errors:
# from tvm.script import tir as T
@T.prim_func
def matmul(input: T.Buffer((T.int64(1), T.int64(8)), "float32"), lv: T.Buffer((T.int64(8), T.int64(16)), "float32"), matmul: T.Buffer((T.int64(1), T.int64(16)), "float32")):
T.func_attr({"target": T.target({"arch": "sm_61", "host": {"keys": ["cpu"], "kind": "llvm", "mtriple": "x86_64-pc-linux-gnu", "tag": ""}, "keys": ["cuda", "gpu"], "kind": "cuda", "max_num_threads": 1024, "tag": "", "thread_warp_size": 32}), "tir.noalias": T.bool(True)})
for i1, k in T.grid(16, 8):
matmul_1 = T.Buffer((T.int64(16),), data=matmul.data)
if k == 0:
matmul_1[i1] = T.float32(0.0)
input_1 = T.Buffer((T.int64(8),), data=input.data)
lv_1 = T.Buffer((T.int64(128),), data=lv.data)
matmul_1[i1] = matmul_1[i1] + input_1[k] * lv_1[k * 16 + i1]
```
--
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: commits-unsubscribe@tvm.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] [Bug] Did you forget to bind? for cuda backend. [tvm]
Posted by "MPolaris (via GitHub)" <gi...@apache.org>.
MPolaris closed issue #18350: [Bug] Did you forget to bind? for cuda backend.
URL: https://github.com/apache/tvm/issues/18350
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] [Bug] Did you forget to bind? for cuda backend. [tvm]
Posted by "MPolaris (via GitHub)" <gi...@apache.org>.
MPolaris commented on issue #18350:
URL: https://github.com/apache/tvm/issues/18350#issuecomment-3342844812
I have reviewed the relevant issues, and it seems none of them have been directly fixed.
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] [Bug] Did you forget to bind? for cuda backend. [tvm]
Posted by "MPolaris (via GitHub)" <gi...@apache.org>.
MPolaris commented on issue #18350:
URL: https://github.com/apache/tvm/issues/18350#issuecomment-3344691303
Solved, it needs passes as below at least if your target is cuda:
```python
tvm.transform.Sequential(
relax.transform.LegalizeOps(),
dlight.ApplyDefaultSchedule(
dlight.gpu.Matmul(),
dlight.gpu.GEMV(),
dlight.gpu.Reduction(),
dlight.gpu.GeneralReduction(),
dlight.gpu.Fallback(),
),
```
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [I] [Bug] Did you forget to bind? for cuda backend. [tvm]
Posted by "humberthumbert (via GitHub)" <gi...@apache.org>.
humberthumbert commented on issue #18350:
URL: https://github.com/apache/tvm/issues/18350#issuecomment-3489828288
Could you show the complete code to make the `build` run? I encountered the similar issues. Much thanks!
> Solved, it needs passes as below at least if your target is cuda:
>
> tvm.transform.Sequential(
> relax.transform.LegalizeOps(),
> dlight.ApplyDefaultSchedule(
> dlight.gpu.Matmul(),
> dlight.gpu.GEMV(),
> dlight.gpu.Reduction(),
> dlight.gpu.GeneralReduction(),
> dlight.gpu.Fallback(),
> ),
--
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: commits-unsubscribe@tvm.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org