FIX: CU_SAFE_CALL used with cublasXXX func

[selaselah]

all CuBLAS func return a cublasStatus_t, which is not same as cudaError_t and the value mean is not compatible

[danpovey]

thanks for the PR.
So there wasn't an official version of this function, that we could have used?

[danpovey]

I see that there wasn't an official function like this, but please use Google style w.r.t. open-braces: put them on the same line as the previous one, after a space, e.g.
const char* cublasGetStatusString(cublasStatus_t status) {

And please put a comment before the declaration of this function, saying:
// This is analogous to the CUDA function cudaGetErrorString().

[selaselah]

Reformat done.
Howerver, the CI raise some reasonless error :(

[selaselah]

the CI failed randomly. I have reopened this request several times but all are failed. Here is some error log:

ln: failed to create symbolic link ‘/home/travis/build/kaldi-asr/kaldi/src/lib/libkaldi-matrix.so’: File exists 

[danpovey]

Thanks. Maybe @jtrmal or @kkm can look into the CI errors-- probably an issue with the latest scripts there.

…

On Fri, Sep 22, 2017 at 10:00 AM, selaselah ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/cudamatrix/cu-common.cc <#1900 (comment)>: > @@ -67,6 +67,25 @@ void GetBlockSizesForSimpleMatrixOperation(int32 num_rows, "Matrix has too many rows to process"); dimGrid->z = 1; } + +const char* cublasGetStatusString(cublasStatus_t status) Reformat done. Howerver, the CI raise some reasonless error :( — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1900 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADJVu9zVE46SKfJG1bA1_7lBEM8jdf1Sks5sk72JgaJpZM4PgLcu> .

[jtrmal]

The compilation went fine, only the scripts weren't able to finish storing the ccache on time -- not sure what to do with that. y.

…

On Fri, Sep 22, 2017 at 12:43 PM, Daniel Povey ***@***.***> wrote: Merged #1900 <#1900>. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1900 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKisX4ntCQXLhw8Fm1PHXSnSoZCKxojNks5sk-OugaJpZM4PgLcu> .

[danpovey]

We have some CuSparse library calls in the 'rnnlm' branch and I think the same issue exists there. Do you have time to address that via another PR?

[selaselah]

I check the CuSparse's doc,. The status value also conflict with cudaError_t. I'll take a PR tomorrow.