wzh99 opened a new issue, #11664:
URL: https://github.com/apache/tvm/issues/11664
### Expected behavior
The following Relay source program should be successfully parsed:
```
#[version = "0.0.5"]
def @main(%x: Tensor[(1, 4, 8, 8, 8), float32], %w: Tensor[(2, 2, 1, 1, 1),
float32]) {
nn.conv3d(%x, %w, kernel_size=[1, 1, 1], channels=2, groups=2)
}
```
Here is the explanation: 4 input channels are divided into 2 groups, with
each group taking 2 channels as input. Each group produces 1 output channel.
Therefore, there are 2 * 1 = 2 output channels in total.
### Actual behavior
The Relay type checker reports a type mismatch:
```
error: The Relay type checker is unable to show the following types match:
Tensor[(4, 0, 1, 1, 1), float32]
Tensor[(2, 2, 1, 1, 1), float32]
In particular:
dimension 0 conflicts: 4 does not match 2. dimension 1 conflicts: 0 does
not match 2.
--> from_string:4:5
|
4 | nn.conv3d(%x, %w, kernel_size=[1, 1, 1], channels=2, groups=2)
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
error: The Relay type checker is unable to show the following types match.
In particular `Tensor[(2, 2, 1, 1, 1), float32]` does not match `Tensor[(4,
0, 1, 1, 1), float32]`
--> from_string:4:5
|
4 | nn.conv3d(%x, %w, kernel_size=[1, 1, 1], channels=2, groups=2)
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Traceback (most recent call last):
File "/Users/wzh/tvm-bug/bug_conv3d_kernel_size.py", line 12, in <module>
mod = parser.parse(src)
File "/Users/wzh/tvm-dev/python/tvm/parser/__init__.py", line 32, in parse
return _ffi_api.ParseModuleInContext(source_name, source, init_module,
init_meta_table)
File "/Users/wzh/tvm-dev/python/tvm/_ffi/_ctypes/packed_func.py", line
237, in __call__
raise get_last_ffi_error()
tvm.error.DiagnosticError: Traceback (most recent call last):
[bt] (8) 9 libtvm.dylib 0x0000000118057102
tvm::runtime::PackedFuncObj::Extractor<tvm::runtime::PackedFuncSubObj<void
tvm::runtime::TypedPackedFunc<tvm::IRModule (std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::basic_string<char, std::__1::char_traits<char>,
std::__1::allocator<char> > const&, tvm::runtime::Optional<tvm::IRModule>
const&, tvm::runtime::Map<tvm::runtime::String,
tvm::runtime::Array<tvm::runtime::ObjectRef, void>, void, void>
const&)>::AssignTypedLambda<tvm::parser::$_2>(tvm::parser::$_2,
std::__1::basic_string<char, std::__1::char_traits<char>,
std::__1::allocator<char> >)::'lambda'(tvm::runtime::TVMArgs const&,
tvm::runtime::TVMRetValue*)> >::Call(tvm::runtime::PackedFuncObj const*,
tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) + 1090
[bt] (7) 8 libtvm.dylib 0x0000000118023d79
tvm::parser::ParseModule(std::__1::basic_string<char,
std::__1::char_traits<char>, std::__1::allocator<char> > const&,
std::__1::basic_string<char, std::__1::char_traits<char>,
std::__1::allocator<char> > const&, tvm::runtime::Optional<tvm::IRModule>
const&, tvm::runtime::Map<tvm::runtime::String,
tvm::runtime::Array<tvm::runtime::ObjectRef, void>, void, void> const&) + 425
[bt] (6) 7 libtvm.dylib 0x0000000117e892f4
tvm::transform::Pass::operator()(tvm::IRModule) const + 148
[bt] (5) 6 libtvm.dylib 0x0000000117e89701
tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext
const&) const + 753
[bt] (4) 5 libtvm.dylib 0x0000000117e8a203
tvm::transform::ModulePassNode::operator()(tvm::IRModule,
tvm::transform::PassContext const&) const + 819
[bt] (3) 4 libtvm.dylib 0x00000001191aa0d5
tvm::runtime::PackedFuncObj::Extractor<tvm::runtime::PackedFuncSubObj<void
tvm::runtime::TypedPackedFunc<tvm::IRModule (tvm::IRModule,
tvm::transform::PassContext)>::AssignTypedLambda<tvm::relay::transform::InferType()::$_2>(tvm::relay::transform::InferType()::$_2)::'lambda'(tvm::runtime::TVMArgs
const&, tvm::runtime::TVMRetValue*)> >::Call(tvm::runtime::PackedFuncObj
const*, tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) + 2053
[bt] (2) 3 libtvm.dylib 0x0000000117e0da1b
tvm::DiagnosticContext::Render() + 459
[bt] (1) 2 libtvm.dylib 0x0000000117b2bd29
tvm::runtime::detail::LogFatal::Entry::Finalize() + 89
[bt] (0) 1 libtvm.dylib 0x000000011952da28
tvm::runtime::Backtrace() + 24
File "/Users/wzh/tvm-dev/src/ir/diagnostic.cc", line 105
DiagnosticError: one or more error diagnostics were emitted, please check
diagnostic render for output.
```
### Environment
macOS 12.4. Compiled using Clang 13.1.6 with LLVM support. TVM commit
[`df4f4c0b4`](https://github.com/apache/tvm/commit/df4f4c0b4bccd775af25967fdf057392c1a2826e).
### Steps to reproduce
```python
from tvm import parser
src = """
#[version = "0.0.5"]
def @main(%x: Tensor[(1, 4, 8, 8, 8), float32], %w: Tensor[(2, 2, 1, 1, 1),
float32]) {
nn.conv3d(%x, %w, kernel_size=[1, 1, 1], channels=2, groups=2)
}
"""
mod = parser.parse(src)
```
### Possible fix
https://github.com/apache/tvm/blob/df4f4c0b4bccd775af25967fdf057392c1a2826e/src/relay/op/nn/convolution.cc#L444-L451
The condition of this if statement seems to be incorrect. `channels ==
groups && channels != 1` does not necessarily mean that the convolution is
depthwise. The test case shown above is a counterexample.
A possible fix is to remove this if statement and only keeps the assignment
to `wshape` in else branch, since this assignment already covers the case of
depthwise convolution.
--
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-unsubscr...@tvm.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
