ninesheep opened a new pull request, #13551:
URL: https://github.com/apache/tvm/pull/13551
construct a graph like below:
``` python
from tvm import relay
p0 = relay.var("p0", shape = [384,144,32], dtype="int8")
p1 = relay.var("p1", shape = [384,144,32], dtype="int8")
x1 = relay.nn.batch_matmul(p0, p1, transpose_b=True, out_dtype="int32")
x2 = relay.cast(x1, "int64")
func = relay.Function(relay.analysis.free_vars(x2), x2)
mod = tvm.IRModule.from_expr(func)
with tvm.transform.PassContext(
opt_level=3,
):
graph, lib, params = tvm.relay.build_module.build(
new_mod, target="cuda --host=llvm", params=None
)
````
and build it, it will report a error like this:
`
Check failed: (!repl_op.same_as(s->op)) is false: Cannot find
Tensor(shape=[384, 144, 32], op.name=compute) in the inputs of compute(T_cast,
body=[int8(compute[ax0, ax1, ax2])], axis=[iter_var(ax0, range(min=0,
ext=384)), iter_var(ax1, range(min=0, ext=144)), iter_var(ax2, range(min=0,
ext=144))], reduce_axis=[], tag=elemwise, attrs={})
`
--
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]
