I recently got a break with normal work and was able to check back and was thrilled to see a v0.4 update in some scripts.
The new v0.4 scripts were setup with x64 OpenCL AMD SDK, but there were still some (solved!) problems.
It was found that some of the environmental options in the "env-parameters.sh" were not picked up in certain places of "build-accull.sh" and "accull" and caused linking problems.
Two notable edits were made:
One which replaced a environmental variable with hard-coded values for the particular OpenCL system (line 44 in "accull"),
It used {AMDOCLDIR}/lib/ when it could have used LD_LIBRARY_PATH created in the "env-parameters.sh" file.
(yes it might need some manipulation to properly insert -L after each directory... but importing a variable of some-sort might be nice here.)
The other edited the final compilation by gcc to read the .c files instead of the .o files.
I don't understand why the last edit (.o => .c) would make a difference under OpenCL vs CUDA, but it was the solution that was found after some googling:
http://stackoverflow.com/questions/20323842/compile-build-opencl-program-using-amd-app-sdk-on-linuxThe details of the step by step OpenCL setup and configuration used, along with the necessary script edits is in the pastie:
http://pastie.org/9344816It is also noted that while the compiled "example" folder code worked, it compiled with warnings.
However when compiling directly with the command:
gcc -I/opt/AMDAPP/include -L/opt/AMDAPP/lib/x86_64 pi.acc.c -lOpenCL
there was no such problem (and the program ran just the same).
(I will post a reply with the warnings because I forgot to copy them.. and I am on different machine at the moment.)
Other usage notes on the v0.4 update:
The interactive improvements in the v0.4 build script were nice, thank you:)
The rest of the test programs were not completed yet... It seems other edits may be needed since they didn't even though the "example" folder's pi generator program was working.
Thanks for the great effort!
Brett