comaniac opened a new pull request, #12912:
URL: https://github.com/apache/tvm/pull/12912
#10516 used the Relay parameter name when lowering to TE. However, this
creates an issue when the parameter name is empty. This is legal in Relay, but
results in errors during code generation. For example, this is the generated
CUDA kernel for bias add:
```
extern "C" __global__ void __launch_bounds__(1024)
fused_raf_op_tvm_add_kernel0(
float* __restrict__ T_add,
float* __restrict__ , /* Name is missing and it results in compile
errors. */
float* __restrict__ _1) {
T_add[((((int)blockIdx.x) * 1024) + ((int)threadIdx.x))] =
([((((int)blockIdx.x) * 1024) + ((int)threadIdx.x))] + _1[((((((int)blockIdx.x)
* 16) + (((int)threadIdx.x) >> 6)) % 54) / 9)]);
}
```
This PR adds "placeholder" back as a default to make sure no empty string
will be passed when lowering to TE.
cc @Lunderberg @tkonolige
--
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]
