Thrsu opened a new issue, #17488:
URL: https://github.com/apache/tvm/issues/17488
The code defines a relax function that allocates a tensor using
R.builtin.alloc_tensor inside a dataflow block. When applying the
StaticPlanBlockMemory transformation, it results in an internal error with the
message: `Check failed: (!block_stack_.empty()) is false`. However, if the
dataflow block is removed, the transformation completes without error.
### Expected behavior
The `StaticPlanBlockMemory` transformation should handle functions with
dataflow blocks correctly, allowing the memory planning without causing
internal errors.
### Actual behavior
```
File "/software/tvm/src/relax/transform/static_plan_block_memory.cc", line
597
InternalError: Check failed: (!block_stack_.empty()) is false:
```
### Steps to reproduce
```python
import tvm
from tvm import relax
from tvm.script import ir as I
from tvm.script import relax as R
@I.ir_module
class Module:
@R.function
def main() -> R.Tensor((10,), dtype="float32"):
with R.dataflow():
gv: R.Tensor((10,), dtype="float32") = R.builtin.alloc_tensor(
R.shape([10]), R.dtype("float32"), R.prim_value(0),
R.str("global")
)
R.output(gv)
return gv
mod = Module
mod_seq =
tvm.transform.Sequential([relax.transform.StaticPlanBlockMemory()])(mod)
```
This issue may indicate a problem with how dataflow blocks are processed
within the StaticPlanBlockMemory pass. Any guidance or fixes to resolve this
would be appreciated.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
