yzh119 opened a new issue #6157:
URL: https://github.com/apache/incubator-tvm/issues/6157
Currently, if I build tvm with cublas and CUDA 11, I'll encounter the
following error:
```
/home/expye/repo/incubator-tvm/src/runtime/contrib/cublas/cublas.cc: In
function ‘void tvm::contrib::CallLtIgemm(tvm::runtime::TVMArgs,
tvm::runtime::TVMRetValue*, cublasLtHandle_t)’:
/home/expye/repo/incubator-tvm/src/runtime/contrib/cublas/cublas.cc:175:63:
error: cannot convert ‘cudaDataType_t’ to ‘cublasComputeType_t’
175 | CHECK_CUBLAS_ERROR(cublasLtMatmulDescCreate(&operationDesc,
CUDA_R_32I));
|
^~~~~~~~~~
| |
|
cudaDataType_t
```
which is no longer compatible with CUDA 11 [cublasLtMatmulDescCreate
interface](https://docs.nvidia.com/cuda/cublas/index.html#cublasLtMatmulDescCreate),
which has three arguments:
```
cublasStatus_t cublasLtMatmulDescCreate(
cublasLtMatmulDesc_t *matmulDesc,
cudaDataType computeType);
```
which breaks the old api which only has two arguments.
I'm not sure if there are other breaking changes in CUDA 11 that would make
the compilation of tvm failed. I think a macro like `#if CUDART_VERSION >=
11000` could solve the problem mentioned above.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
