Is x86 supposed to be slower than x64?

[wangg]

First of all, thank you very much for sharing the x86 building. It was literally killing me...

I tried to compile the same C++ code with r1.8 x86(libtensorflow-cpu-windows-x86_32-1.8.0-sse2) and x64(libtensorflow-cpu-windows-x86_64-1.8.0-sse2). The exactly same inference takes about 200ms in the x64 version, but as long as 400ms in the x86 version. Is it normal?

[fo40225]

I have no idea about this.

[wangg]

Thanks for the quick response. Could you please give some instructions for build x86 on Windows then? I would like to compile on my own machine to see if it is the same. I tried to compile several versions of tensorflow with cmake and MSVC 2015 (x86) but they all ended up with "error C2719: '_Func': formal parameter with requested alignment of 16 won't be aligned"...

[fo40225]

tensorflow/tensorflow#18295

#6 (comment)

[ghost]

      Thanks for the quick response. Could you please give some instructions for build x86 on Windows then? I would like to compile on my own machine to see if it is the same. I tried to compile several versions of tensorflow with cmake and MSVC 2015 (x86) but they all ended up with "error C2719: '_Func': formal parameter with requested alignment of 16 won't be  aligned"...

Have you solved this problem? How to do？