I got an “Internal error” ( no more output information ) compiling a kernel for GPU usage ( with -deviceemu compiles ok ) in both Debug and Release mode.
My command line in debug mode is:
"$(CUDA_BIN_PATH)\nvcc.exe" -ccbin "$(VCInstallDir)bin" -c -keep -use_fast_math -maxrregcount 16 -D_DEBUG -DWIN32 -Xcompiler /EHa,/W3,/nologo,/Wp64,/Od,/Zi,/RTC1,/MTd -I"$(CUDA_INC_PATH)" -o "$(PlatformName)\$(ConfigurationName)\CUDA.obj" CUDA.cu
in Release mode is:
“$(CUDA_BIN_PATH)\nvcc.exe” -ccbin “$(VCInstallDir)bin” -c -keep -use_fast_math -maxrregcount 16 -DNDEBUG -DWIN32 -Xcompiler /EHa,/W3,/nologo,/Wp64,/Ox,/Ob2,/Oi,/Ot,/Oy,/GT,/GL,/MTd -I"$(CUDA_INC_PATH)" -o “$(PlatformName)$(ConfigurationName)\CUDA.obj” CUDA.cu
Yep, i’m using cpp integration btw.
If I remove the “-maxrregcount 16” compiles ok. With maxreg 32 fails too with “Internal error”. So I bet is some error related to the number of registers… but it doesn’t give more error information… .just a simple “Internal error”.
I would wish to keep my kernel code private unless a NVIDIA worker asks it for debugging purposes. thx.