Thrsu opened a new issue, #17392:
URL: https://github.com/apache/tvm/issues/17392
I encountered an issue when trying to build a TIR-based module. The
following error occurs when using `T.match_buffer` for a scalar element:
```
File "/software/tvm/src/tir/transforms/lower_match_buffer.cc", line 222
TVMError: Check failed: arg.dtype() == value.dtype() (int64 vs. int32) : The
data type mismatched: int64 vs. int32
```
If I replace the line `C1 = T.match_buffer(C0[jj], ())` with direct
assignment using `C0[jj] = T.float32(0)`, the error goes away, and the code
builds successfully without any issues.
### Steps to reproduce
```python
import tvm
from tvm import tir, relax
from tvm.script import ir as I
from tvm.script import tir as T
@I.ir_module
class Module:
@T.prim_func
def main():
# with T.block("root"):
C = T.alloc_buffer((128, 128))
for i in range(128):
with T.block(""):
vi = T.axis.spatial(128, i)
T.reads()
T.writes(C[vi, 0:128])
C0 = T.match_buffer(C[vi, 0:128], (128))
for j in range(128):
with T.block(""):
jj = T.axis.spatial(128, j)
T.reads()
T.writes(C0[jj])
C1 = T.match_buffer(C0[jj], ())
C1[()] = T.float32(0)
mod = Module
ex = relax.build(mod, target='llvm')
```
CC @Lunderberg @tqchen @junrushao
--
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]
