search

accel-sim / gpu-app-collection

A repository where GPU applications are aggregated using a common build flow that supports multiple CUDA versions.
44 stars 32 forks source link

Recent CUDA versions and unsupported SM numbers in build process #1

Closed mahmoodn closed 3 years ago

mahmoodn commented 4 years ago

Hi and thanks for sharing this repo. It seems that the setup_environment script uses sdk-4.2. My CUDA_INSTALL_PATH is ~/cuda_10.1.168 and the script fails with some unsupported sm architectures which are old and unsupported by cuda_10.1.

make[2]: Entering directory `/home/mahmood/gpu-app-collection/4.2/C/src/inlinePTX'
make[2]: Entering directory `/home/mahmood/gpu-app-collection/4.2/C/src/fastWalshTransform'
nvcc fatal   : Unsupported gpu architecture 'compute_10'
cc1plus: warning: command line option ‘-Wimplicit’ is valid for C/ObjC but not for C++ [enabled by default]
make[2]: [obj/x86_64/release/inlinePTX.cu.o] Error 1 (ignored)
make[2]: Entering directory `/home/mahmood/gpu-app-collection/4.2/C/src/vectorAddDrv'
cc1plus: warning: command line option ‘-Wimplicit’ is valid for C/ObjC but not for C++ [enabled by default]
make[2]: Entering directory `/home/mahmood/gpu-app-collection/4.2/C/src/threadFenceReduction'
nvcc fatal   : Unsupported gpu architecture 'compute_10'
make[2]: Entering directory `/home/mahmood/gpu-app-collection/4.2/C/src/threadMigration'
nvcc fatal   : Unsupported gpu architecture 'compute_11'
g++: error: obj/x86_64/release/inlinePTX.cu.o: No such file or directory

Although I can manually remove such old sm numbers, I want to know what is the purpose of having them while this repo (and accel-sim) emphasis on new cuda versions? Thanks.