CUDA 2.21 SDK example projects were successful make. However, emulation was not. Make complains as below:
dct8x8.cu: In function ‘main’:
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
dct8x8.cu:498: sorry, unimplemented: called from here
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
dct8x8.cu:499: sorry, unimplemented: called from here
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
dct8x8.cu:500: sorry, unimplemented: called from here
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
dct8x8.cu:501: sorry, unimplemented: called from here
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
dct8x8.cu:502: sorry, unimplemented: called from here
/usr/include/bits/string3.h:49: sorry, unimplemented: inlining failed in call to ‘memcpy’: function body not available
It seems like the problem is originated from a single source, which I don’t have any clue. Please share your wisdom. Thanks.
I just did ‘make emu=1’ in the SDK installed directory to compile projects for emulation.
Since dct8x8 project failed for now, I looked into the content of Makefile of the project and below is the content. The errors occured for .cu files, it may be related to cudacc compile. (of course, it may not but we need some starting point anyway)
By reviewing the below Makefile, I found that the compiler options are specified in common.mk file. I am looking for the options for now. But I don’t know what to look for.
I see. Nvidia installer issued a warning about inconsistency of kernerl compiled version (gcc-4.2). Maybe it might be related. I will try to do that with gcc-4.2 and let you know the result. Thank you, N. Your help is awesomly good :)