YuanchengJiang opened a new issue, #20467:
URL: https://github.com/apache/tvm/issues/20467
**Repo:** apache/tvm · **Commit:** `5b4b753` · LLVM 18.1.8
## What happens
An IRModule holding a `tirx.call_extern` to a symbol that is not linked
compiles and jits
without complaint; calling an *unrelated* function of the same module prints
```
JIT session error: Symbols not found: [ dummy_function_name ]
!!!!!!! Segfault encountered !!!!!!!
```
and the process dies on SIGSEGV (2/2 runs, exit 139).
## Reproducer
`jit_missing_symbol.py`:
```python
import numpy as np
import tvm
from tvm.script import ir as I
from tvm.script import tirx as T
@I.ir_module(s_tir=True)
class Module:
@T.prim_func(s_tir=True)
def main():
T.Call(tvm.ir.Op.get("tirx.call_extern"), ["dummy_function_name"],
ret_ty="void")
@T.prim_func(s_tir=True)
def rowsum(a: T.handle, b: T.handle) -> None:
A = T.match_buffer(a, [128])
B = T.match_buffer(b, [])
B_rf = T.sblock_alloc_buffer([128], elem_offset=T.int64(0))
for i in range(128):
with T.sblock("B_rf"):
vi0 = T.axis.S(128, i)
B_rf[vi0] = A[vi0]
for i in range(128):
with T.sblock("B"):
vi0_1 = T.axis.R(128, i)
with T.init():
B[()] = 0.0
B[()] = B[()] + B_rf[vi0_1]
lib = tvm.compile(Module, target="llvm")
rt = lib.jit()
a = tvm.runtime.tensor(np.zeros(128, dtype="float32"))
b = tvm.runtime.tensor(np.zeros((), dtype="float32"))
rt["rowsum"](a, b) # dies here
print("called, b =", b.numpy())
```
```bash
python3 jit_missing_symbol.py
```
## Expected
An unresolved symbol should surface as an exception from `jit()` or from the
call, not a
crash. `jit()` on the same module *without* the second function returns
normally, so the
failure is deferred until an unrelated call runs.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
