I’ve made a kernel that writes 10 in a matrix. If I use it in EmuDebug o EmuRelease mode runs correctly, but in debug or release doesn’t write anything in the matrix.
I have a warning: 1>LINK : warning LNK4098: defaultlib ‘LIBCMT’ conflicts with use of other libs; use /NODEFAULTLIB:library
what about your compiler commands?
nvcc.exe -ccbin “C:\Archivos de programa\Microsoft Visual Studio 8\VC\bin” -I"C:\CUDA\include" -I"C:\Archivos de programa\NVIDIA Corporation\NVIDIA CUDA SDK\common\inc" -O0 -D_DEBUG -D_CONSOLE -arch sm_11 -code sm_11 --host-compilation C++ -g -c -m 32 -o “Debug\CUDAWinApp1.obj” -odir “Debug” -ext none -int real CUDAWinApp1.vcproj
the commands are correct.~
can you debug it step by step?
add check mac like CUDA_SAFE_CALL in your program.
No, I cant debug the kernel. The program said that there isn’t source code for the actual location.
It trys to find tmpxft_00008928_00000000-3_kerneltest.cudafe1.stub.c.
I can’t use the macro with the kernel: Too parameters.
I’ve used it with the other cuda functions and work perfectly
Thanks
can you send the project to me, if the source codes are not necessary? I want to challenge it…
I’ll send you tomorrow, now I’m in my home and I have the project in the university.
I had sent back to you, check your email:)
BTW,
1.Right key on project → properties → cuda → advanced → gpu architecture compile name → 1.1(virtual)Arch
2.change double to float. Only device version 1.3 can support double.
kyzhao, could you tell how did you fix this ? I didn’t see this project, but I have similar problem.
Thanks in advance.
Thank you.
Can you show any error message to me?