search

adammeghji / ansible-ltc-mining-on-ec2

Automates the setup of LTC miners on fresh EC2 GPU instances
MIT License
45 stars 9 forks source link

"make cudaminer" install phase bombs on UNREACHABLE executed! #5

Open adammeghji opened 10 years ago

adammeghji commented 10 years ago

It appears the "make cudaminer" phase of the installation is now bombing on an obscure error:

nvcc -g -O2 -Xptxas "-abi=no -v" -arch=compute_30 --maxrregcount=63 -I./compat/jansson -o spinlock_kernel.o -c spinlock_kernel.cu

UNREACHABLE executed!
Stack dump:
0.  Running pass 'NVPTX DAG->DAG Pattern Instruction Selection' on function '@_Z28spinlock_scrypt_core_kernelBILi3EEvPj'
nvcc error   : 'cicc' died due to signal 6
make[2]: *** [spinlock_kernel.o] Error 6
make[2]: Leaving directory `/home/ec2-user/CudaMiner'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/ec2-user/CudaMiner'
make: *** [all] Error 2

This started happening once we upgraded to the latest linux kernel bundled via the Amazon Linux HVM AMI. Investigating ..

adammeghji commented 10 years ago

it seems this is a known issue with the CudaMiner project on HEAD: https://github.com/cbuchner1/CudaMiner/issues/42

adammeghji commented 10 years ago

Temporarily reverted to the last known good version of CudaMiner via 8b9c20e224c64df3c8fd5b4466cd1e0beb9b9dbe. Leaving this issue open as a reminder to revert once cbuchner1/CudaMiner#42 is fixed.

OnkelTem commented 10 years ago

adammeghji, can you explain how to use this "fix"?

adammeghji commented 10 years ago

The workaround has been committed to HEAD, so it should be present in the latest version of this repo.

Are you experiencing an issue with this?  If so, feel free to paste a gist with your output and I'll try to help!

OnkelTem commented 10 years ago

This is my output with latest CUDA 5.5 and git master CudaMiner

/usr/local/cuda-5.5/bin/nvcc -O3 -Xptxas "-abi=no -v" -arch=compute_30 --maxrregcount=63 -I./compat/jansson -o spinlock_kernel.o -c spinlock_kernel.cu
UNREACHABLE executed!
Stack dump:
0.  Running pass 'NVPTX DAG->DAG Pattern Instruction Selection' on function '@_Z28spinlock_scrypt_core_kernelBILi3EEvPj'
Aborted (core dumped)
make[2]: *** [spinlock_kernel.o] Error 134
adammeghji commented 10 years ago

@aneganov if you're using our script here, this should be working. If you're trying to manually get CudaMiner working, the workaround we used here can be applied via:

git clone https://github.com/cbuchner1/CudaMiner
git checkout 88c6da6d5c2b798d1de7031e8dbcc2678f635e4b