I updated my cntk and I compiled "Makefile.gpu" file again.
( I already compiled Makefile.gpu before successfully)
And it showed the following error:
creating .build/x86_64.gpu.release.acml/Common/BestGpu.o for x86_64 with build type release
g++ -c Common/BestGpu.cpp -o .build/x86_64.gpu.release.acml/Common/BestGpu.o -O4 -msse3 -std=c++0x -std=c++11 -D_POSIX_SOURCE -D_XOPEN_SOURCE=600 -D__USE_XOPEN2K -DUSE_ACML -fopenmp -fpermissive -I Common/Include -I Math/Math -I MachineLearning/CNTK -I /usr/local/cuda-7.0/include -I /home/dong/CNTK/cntk/acml/gfortran64/include -I usr/include/hwloc/ -fPIC -MD -MP -MF .build/x86_64.gpu.release.acml/Common/BestGpu.d
Common/BestGpu.cpp:20:145: fatal error: nvml.h: No such file or directory
#include <nvml.h> // note: expected at "c:\Program Files\NVIDIA Corporation\GDK\gdk_win7_amd64_release\nvml\include" (Windows)
make: *** [.build/x86_64.gpu.release.acml/Common/BestGpu.o] 1 error
So I checked Makefile.gpu file and I found NVML_INCLUDE didn't set.
For setting this, I searched nvml.h on my server and I checked this file is in usr/include/hwloc/
so I modified NVML_INCLUDE to "NVML_INCLUDE=usr/include/hwloc/"
However, the above error occurred again.
If I set something wrong, please tell me what I do wrong.