Compilation error for ParameterClient2.cpp - Narrowing conversion

[paddlepaddle]

Compilation error for ParameterClient2.cpp - Narrowing conversion

[emailweixu]

Please provide the compiler version and error message. Thx.

[brucekissinger]

The gcc version is 5.4.0.20160609

Unfortunately, I don't have a good way to copy/paste the errors from the build. I'm manually typing the messages below:

pserver/ParameterClient2.cpp:281:58: error: narrowing conversion of '(((long long unsigned int)sizeof (paddle::real)) * ((long long unsigned int)blockSize))' from 'long long unsigned int' to 'size_t {aka unsigned int}' inside { } [-Werror=narrowing]
{sendMat->getLocalRow(row), sizeof(real) * blockSize});

the second error is:

pserver/ParameterCllent2.cpp:306:51: error: narrowing conversion of '(((uint64_t)sizeof (paddle::real)) * (endDim - beginDim))' from 'uint64_t {aka long long unsigned int}] to 'size_t {aka unsigned int}' inside { } [-Werror=narrowing]
{buf + beginDim, sizeof(real) * (endDim - beginDim)});

[reyoung]

It seems to be the size_t in your compiler is unsigned int, not unsigned long long. We are fixing this issue.

By the way, Is gcc 5.4 supported by CUDA? Or you are just building CPU version of PaddlePaddle?

[brucekissinger]

Thank you for the extremely fast response!

Yes - I was just building the CPU version. I'm running it on a VirtualBox instance and was not sure if access to the GPU of the underlying host machine would be possible.

[reyoung]

Current PaddlePaddle is not tested in 32bit environment. I will do the test and fix this ASAP.

[reyoung]

Sorry, I forgot to reply this issue. In fdd40e5, we fix 32 bit compile issues in PaddlePaddle.