CudaMiner
CudaMiner copied to clipboard
cbuchner1
building on osx
I've tried to build cudaminer on mac osx. problems i solved:
- in scrypt.cpp
#include <malloc.h>
have to be#include <malloc/malloc.h>
or simply deleted - nvcc needs "-m 64" argument to build for 64bit
now i get segmentation fault :(
./cudaminer --url http://... --userpass ...:... --no-autotune
*** CudaMiner for nVidia GPUs by Christian Buchner ***
This is version 2013-04-22 (alpha)
based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
Cuda additions Copyright 2013 Christian Buchner
My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm
[2013-05-03 02:24:46] 1 miner threads started, using 'scrypt' algorithm.
[2013-05-03 02:24:47] Long-polling activated for http://notroll.in:6332/LP
[2013-05-03 02:24:47] GPU #0: GeForce 9400 with compute capability 1.1
[2013-05-03 02:24:47] GPU #0: interactive: 1, tex-cache: 0 , single-alloc: 0
[2013-05-03 02:24:50] GPU #0: using launch configuration 2x4
Segmentation fault: 11
ok i know that this is s nearly useless bugreport but i hope you can help :)
I also have attempted to port this software to OSX, and got about as far as you did. I did create a crashlog and the gdb to see where the seg fault was though, not sure if it really helps.
$ gdb ~/Dev/CudaMiner/cudaminer core.35143
GNU gdb 6.3.50-20050815 (Apple version gdb-1824) (Wed Feb 6 22:51:23 UTC 2013)
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for details.
This GDB was configured as "x86_64-apple-darwin"...Reading symbols for shared libraries ..... done
Reading symbols for shared libraries . done
Reading symbols for shared libraries ........................................................................................................................... done
#0 0x00007fff883d6386 in __semwait_signal ()
(gdb) bt
#0 0x00007fff883d6386 in __semwait_signal ()
#1 0x00007fff8541bcbd in pthread_join ()
#2 0x00000001022427fe in restart_threads [inlined] () at /Users/deweydb/Dev/CudaMiner/cuda-miner.cpp:1298
(gdb) thread apply all bt
Thread 5 (core thread 4):
#0 0x00007fff883d4686 in mach_msg_trap ()
#1 0x00007fff883d3c42 in mach_msg ()
#2 0x00000001027c56e8 in cudbgGetAPIVersion ()
#3 0x00000001027136c8 in cuGraphicsGLRegisterImage ()
#4 0x00000001027c6972 in cudbgGetAPIVersion ()
#5 0x00007fff853917a2 in _pthread_start ()
#6 0x00007fff8537e1e1 in thread_start ()
Thread 4 (core thread 3):
#0 0x000000010224113d in GOMP_parallel_start ()
#1 0x0000000102251711 in scanhash_scrypt (thr_id=54189520, pdata=0x1033addd0, ptarget=0x1033addd0, max_nonce=54189520, hashes_done=0x1033addd0) at scrypt.cpp:705
#2 0x00000001022436d0 in miner_thread (userdata=0x7f9511c17cd0) at cuda-miner.cpp:688
#3 0x00007fff853917a2 in _pthread_start ()
#4 0x00007fff8537e1e1 in thread_start ()
Thread 3 (core thread 2):
#0 0x00007fff883d6322 in select$DARWIN_EXTSN ()
#1 0x00007fff8a1e5e2d in Curl_socket_check ()
#2 0x00007fff8a1da6f3 in Curl_do_perform ()
#3 0x0000000102244acc in json_rpc_call (curl=0x7f9514005600, url=0x10332ad08 "?", userpass=******* "******.*******:******", rpc_req=0x7f95128005e0 "?\005?\022?", curl_err=0x10332aee4, longpoll=true, longpoll_scan=40) at util.cpp:365
#4 0x0000000102243149 in longpoll_thread (userdata=0x10332af10) at cuda-miner.cpp:790
#5 0x00007fff853917a2 in _pthread_start ()
#6 0x00007fff8537e1e1 in thread_start ()
Thread 2 (core thread 1):
#0 0x00007fff883d60fa in __psynch_cvwait ()
#1 0x00007fff85395fe9 in _pthread_cond_wait ()
#2 0x0000000102243e95 in tq_pop (tq=0x7f9511c17d20, abstime=0x0) at util.cpp:643
#3 0x0000000102242982 in workio_thread (userdata=0x1032a7f10) at cuda-miner.cpp:506
#4 0x00007fff853917a2 in _pthread_start ()
#5 0x00007fff8537e1e1 in thread_start ()
Thread 1 (core thread 0):
#0 0x00007fff883d6386 in __semwait_signal ()
#1 0x00007fff8541bcbd in pthread_join ()
#2 0x00000001022427fe in restart_threads [inlined] () at /Users/deweydb/Dev/CudaMiner/cuda-miner.cpp:1298
I've been able to compile it on OSX, and it does run for me, but - with sha256d only, scrypt gives me segfaults.
gdb output for default run
kuba$ gdb ./cudaminer
GNU gdb 6.3.50-20050815 (Apple version gdb-1824) (Wed Feb 6 22:51:23 UTC 2013)
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for details.
This GDB was configured as "x86_64-apple-darwin"...Reading symbols for shared libraries ........ done
(gdb) set args -o http://192.168.4.3:8332 -O kubatyszko.1:x -D --no-autotune -l4x8 -C 1
(gdb) run
Starting program: /Users/kuba/CudaMiner/cudaminer -o http://192.168.4.3:8332 -O user:pass -D --no-autotune -l4x8 -C 1
Reading symbols for shared libraries ++++++.+............................................................................................................. done
*** CudaMiner for nVidia GPUs by Christian Buchner ***
This is version 2013-07-13 (alpha)
based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler
Cuda additions Copyright 2013 Christian Buchner
My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm
Reading symbols for shared libraries .. done
Reading symbols for shared libraries ... done
[2013-07-21 08:26:48] 1 miner threads started, using 'scrypt' algorithm.
[2013-07-21 08:26:49] Long-polling activated for http://192.168.4.3:8332/lp
[2013-07-21 08:26:49] DEBUG: got new work in 646 ms
[2013-07-21 08:26:49] GPU #0: GeForce GT 650M with compute capability 3.0
[2013-07-21 08:26:49] GPU #0: interactive: 1, tex-cache: 1D, single-alloc: 1
[2013-07-21 08:26:49] GPU #0: using launch configuration 4x8
Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_INVALID_ADDRESS at address: 0x0000000000000010
[Switching to process 84746 thread 0x2803]
0x000000010000184d in GOMP_parallel_start ()
(gdb) bt
#0 0x000000010000184d in GOMP_parallel_start ()
#1 0x0000000100014e69 in scanhash_scrypt (thr_id=35884336, pdata=0x102238d30, ptarget=0x102238d30, max_nonce=35884336, hashes_done=0x102238d30) at scrypt.cpp:694
#2 0x0000000100004226 in miner_thread (userdata=0x102238f10) at cpu-miner.c:814
#3 0x00007fff913e88bf in _pthread_start ()
#4 0x00007fff913ebb75 in thread_start ()
(gdb) quit
The program is running. Exit anyway? (y or n) y
output for sha256d (bitcoin) on bit minter:
kuba$ ./cudaminer -o stratum+tcp://mint.bitminter.com:3333 -O user:pass -D --no-autotune -l4x8 -C 1 -a sha256d
*** CudaMiner for nVidia GPUs by Christian Buchner ***
This is version 2013-07-13 (alpha)
based on pooler-cpuminer 2.3.2 (c) 2010 Jeff Garzik, 2012 pooler
Cuda additions Copyright 2013 Christian Buchner
My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm
[2013-07-21 08:32:29] 1 miner threads started, using 'sha256d' algorithm.
[2013-07-21 08:32:29] Starting Stratum on stratum+tcp://mint.bitminter.com:3333
[2013-07-21 08:32:30] Failed to get Stratum session id
[2013-07-21 08:32:30] Stratum difficulty set to 1
[2013-07-21 08:32:34] DEBUG: job_id='8781-d5' extranonce2=000000000000 ntime=51eb8e97
[2013-07-21 08:32:34] Stratum detected new block
[2013-07-21 08:32:36] GPU #0: (null), 2097152 hashes, 951.80 khash/s
[2013-07-21 08:32:48] DEBUG: job_id='8782-3' extranonce2=000000000000 ntime=51eb8ea9
[2013-07-21 08:33:19] DEBUG: job_id='8784-2' extranonce2=000000000000 ntime=51eb8ec9
[2013-07-21 08:33:36] GPU #0: (null), 57107881 hashes, 952.14 khash/s
[2013-07-21 08:33:50] DEBUG: job_id='8786-11' extranonce2=000000000000 ntime=51eb8ee8
[2013-07-21 08:34:21] DEBUG: job_id='8788-5' extranonce2=000000000000 ntime=51eb8f07
[2013-07-21 08:34:36] GPU #0: (null), 57128461 hashes, 950.92 khash/s
[2013-07-21 08:34:52] DEBUG: job_id='878a-c8' extranonce2=000000000000 ntime=51eb8f26
[2013-07-21 08:35:23] DEBUG: job_id='878c-4' extranonce2=000000000000 ntime=51eb8f45
Speed is extremely low, but at least it does run. Afair, OpenCL (cgminer) gives me some 20mhash/sec.
This would mean the issue is somewhere in the scrypt implementation throwing segfaults - at least now we know where to look :)
http://atomton.roon.io/compiling-cudaminer-on-os-x-10-9
Those darn CUDA Toolkits with their not including lib64 for OS X.
On a GTX 760 I'm seeing around 160kh/s to 130kh/s which is 10kh/s less than what I get on windows but twice as much as cgminer on OS X.
Make sure to do "xcode-select --install" to ensure you have all the command line tools needed. Also the article doesn't make it very clear on how to install gcc 4.9, but the site to download it from explains that you need to do gunzip on the .tar.gz to get a .tar file. Also be sure to remove any previous CUDA toolkits installed (just drag them to the trash) so that there is no conflict. Finally, make sure to change line 6433 in the configure file even though there is a else statement with the same code you're editing right underneath it, that's just a shortcut I made the mistake of making. I stumbled upon this article just tonight and I'm tired.
I've only tested this on Mavericks, also I'm on a hackintosh (but shouldn't make a difference as the article uses a real mac). I hope someone could test this with older versions of OS X, gcc, and/or CUDA Toolkit. It seems like they will work but I cannot guarantee it.
But all the credit goes to the article for figuring out how to build cudaminer on OS X.
I followed the blog article's instructions, except I'm using Mountain Lion and I installed gcc-4.9 from the homebrew-versions tap and exported it as CC.
I'm getting the following error when running make:
make all-recursive
Making all in compat
Making all in jansson
Makefile:254: .deps/dump.Po: No such file or directory
Makefile:255: .deps/hashtable.Po: No such file or directory
Makefile:256: .deps/load.Po: No such file or directory
Makefile:257: .deps/strbuffer.Po: No such file or directory
Makefile:258: .deps/utf.Po: No such file or directory
Makefile:259: .deps/value.Po: No such file or directory
make[3]: *** No rule to make target `.deps/value.Po'. Stop.
make[2]: *** [all-recursive] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
@Nitrouspony I will update my article with your suggestions, thanks!
@Kudu Unfortunately I didn't have a chance to try and compile it on 10.8 yet since all my Macs are currently running 10.9, but I'm going to setup 10.8 on a spare HDD and see what I can do.
@Kudu I have built cudaminer for 10.8.5 based on the directions from @atomton. I came up against a different error than you, but I mention it here in case it could help you (or someone else).
The error I came up against was earlier in the build than you are seeing:
make all-recursive Making all in compat make[3]: Nothing to be done for 'all-am'. /usr/local/cuda/bin/nvcc -O3 -arch=compute_10 --maxrregcount=64 --ptxas-options=-v -o salsa_kernel.o -c salsa_kernel.cu /usr/local/include/c++/4.8.1/cstdlib(178): error: identifier "__int128" is undefined
A stackoverflow search indicates that this is a CUDA/gcc issue, and that using an older version of gcc might help. Based on that suggestion I was able to build successfully with gcc 4.6.4
Here is the compilation result: https://mega.co.nz/#!htABAIzC!ACgOmgDqabV-mF8aOqgzZ2iSNC90B1WVcOjrriVBfyc You can escape the "try hard hour" ;)
@atomton I am building on mac os x 10.9 now but it fails here
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 4 (Illegal Instruction) make[2]: *** [spinlock_kernel.o] Error 4 make[1]: *** [all-recursive] Error 1 make: *** [all] Error 2
Tried following the OS X 10.9 instructions with CUDA 5.5 toolkit and GCC 4.9, then installed CUDA 5.0 toolkit and result is the same. Seems basic. Here is my error msg:
/Applications/Xcode.app/Contents/Developer/usr/bin/make all-recursive
Making all in compat
make[3]: Nothing to be done for all-am'. g++ -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -g -O2 -MT cudaminer-scrypt.o -MD -MP -MF .deps/cudaminer-scrypt.Tpo -c -o cudaminer-scrypt.otest -f 'scrypt.cpp' || echo './'`scrypt.cpp
scrypt.cpp:47:15: fatal error: new: No such file or directory
#include
except carat indicates a position after
/Applications/Xcode.app/Contents/Developer/usr/bin/make all-recursive
Making all in compat
make[3]: Nothing to be done for `all-am'.
g++ -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -g -O2 -MT cudaminer-scrypt.o -MD -MP -MF .deps/cudaminer-scrypt.Tpo -c -o cudaminer-scrypt.o `test -f 'scrypt.cpp' || echo './'`scrypt.cpp
scrypt.cpp:47:15: fatal error: new: No such file or directory
#include <new>
^
compilation terminated.
make[2]: *** [cudaminer-scrypt.o] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
@tzbauknight Well I'd say for some reason, whatever compiler you're using, it's not using libstdc++. Would you mind posting the output of gcc -v?
atomton HELP! I get this error on the last command.
/Applications/Xcode.app/Contents/Developer/usr/bin/make all-recursive Making all in compat make[3]: Nothing to be done for `all-am'. /usr/local/cuda/bin/nvcc -O3 -Xptxas "-abi=no -v" -arch=compute_30 --maxrregcount=63 -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 4 (Illegal Instruction) make[2]: *** [spinlock_kernel.o] Error 4 make[1]: *** [all-recursive] Error 1 make: *** [all] Error 2
Okay...seem to have gotten past the error above, but now errors out with
/Applications/Xcode.app/Contents/Developer/usr/bin/make all-recursive Making all in compat make[3]: Nothing to be done for all-am'.
/usr/local/bin/gcc-4.9 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -MT cudaminer-cpu-miner.o -MD -MP -MF .deps/cudaminer-cpu-miner.Tpo -c -o cudaminer-cpu-miner.o test -f 'cpu-miner.c' || echo './'cpu-miner.c
mv -f .deps/cudaminer-cpu-miner.Tpo .deps/cudaminer-cpu-miner.Po
/usr/local/bin/gcc-4.9 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -MT cudaminer-util.o -MD -MP -MF .deps/cudaminer-util.Tpo -c -o cudaminer-util.o test -f 'util.c' || echo './'util.c
mv -f .deps/cudaminer-util.Tpo .deps/cudaminer-util.Po
/usr/local/bin/gcc-4.9 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -MT cudaminer-sha2.o -MD -MP -MF .deps/cudaminer-sha2.Tpo -c -o cudaminer-sha2.o test -f 'sha2.c' || echo './'sha2.c
mv -f .deps/cudaminer-sha2.Tpo .deps/cudaminer-sha2.Po
/usr/local/bin/g++-4.9 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -MT cudaminer-scrypt.o -MD -MP -MF .deps/cudaminer-scrypt.Tpo -c -o cudaminer-scrypt.o test -f 'scrypt.cpp' || echo './'scrypt.cpp
mv -f .deps/cudaminer-scrypt.Tpo .deps/cudaminer-scrypt.Po
/usr/local/cuda/bin/nvcc -O3 -arch=compute_10 --maxrregcount=64 --ptxas-options=-v -o salsa_kernel.o -c salsa_kernel.cu
clang: error: unsupported option '-dumpspecs'
clang: error: no input files
make[2]: *** [salsa_kernel.o] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2`
using CUDA Toolkit 5.0, homebrew installed gcc-4.9 (with CC, CPP and CXX variables pointing to it), although the output of gcc -v is:
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1 Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn) Target: x86_64-apple-darwin13.0.0 Thread model: posix
@tzbauknight I think CUDA 5.0 only supports GCC versions below 4.7 – I'm using 4.6.4 when compiling with CUDA 5.0.
All right...switched CC, CPP and CXX variables to point at gcc46, also installed via homebrew. Same error.
@tzbauknight Did you try altering the $PATH environment variable yet, so gcc actually points to gcc and not clang? I personally use MacPorts, so I'm not sure how brew handles this.
In addition, I just updated my original post about compiling on OS X with new instructions for CUDA 5.0 and GCC 4.6.4.
I started over from scratch, following the new instructions. I have environment variables CC, CPP and CXX all pointing to gcc46 as installed by homebrew. ./configure with those arguments seems to use gcc-4.6 and creates the necessary files. "make" results in the following:
/Applications/Xcode.app/Contents/Developer/usr/bin/make all-recursive Making all in compat make[3]: Nothing to be done for all-am'.
/usr/local/bin/gcc-4.6 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -m64 -MT cudaminer-cpu-miner.o -MD -MP -MF .deps/cudaminer-cpu-miner.Tpo -c -o cudaminer-cpu-miner.o test -f 'cpu-miner.c' || echo './'cpu-miner.c
mv -f .deps/cudaminer-cpu-miner.Tpo .deps/cudaminer-cpu-miner.Po
/usr/local/bin/gcc-4.6 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -m64 -MT cudaminer-util.o -MD -MP -MF .deps/cudaminer-util.Tpo -c -o cudaminer-util.o test -f 'util.c' || echo './'util.c
mv -f .deps/cudaminer-util.Tpo .deps/cudaminer-util.Po
/usr/local/bin/gcc-4.6 -std=gnu99 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -m64 -MT cudaminer-sha2.o -MD -MP -MF .deps/cudaminer-sha2.Tpo -c -o cudaminer-sha2.o test -f 'sha2.c' || echo './'sha2.c
mv -f .deps/cudaminer-sha2.Tpo .deps/cudaminer-sha2.Po
/usr/local/bin/g++-4.6 -DHAVE_CONFIG_H -I. -msse2 -fopenmp -pthread -fno-strict-aliasing -O3 -m64 -MT cudaminer-scrypt.o -MD -MP -MF .deps/cudaminer-scrypt.Tpo -c -o cudaminer-scrypt.o test -f 'scrypt.cpp' || echo './'scrypt.cpp
mv -f .deps/cudaminer-scrypt.Tpo .deps/cudaminer-scrypt.Po
/usr/local/cuda/bin/nvcc -O3 -m64 -arch=compute_10 --maxrregcount=124 --ptxas-options=-v -o salsa_kernel.o -c salsa_kernel.cu
clang: error: unsupported option '-dumpspecs'
clang: error: no input files
make[2]: *** [salsa_kernel.o] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2`
"which gcc" returns:
`````` /usr/bin/gcc```
My $PATH looks like this:
/usr/local/cuda/bin:/usr/local/bin:/usr/local/sbin:/usr/bin:/bin:/usr/sbin:/sbin:/usr/local/bin:/opt/X11/bin
Thanks for all your help. Looking forward to getting this running! 10.9 Hackintosh lo-speed mining!
sorry..."which gcc" points to /usr/bin/gcc which I'm guessing is some sort of front end for the wrong thing. I'm not a coder, but I don't mind tinkering with things like this. All help appreciated on setting up my dev environment. Learning.
@tzbauknight Problem with Mavericks is that /usr/bin/gcc maps to clang – hence the dumpspecs error. Now, your $PATH is set up to look in /usr/local/bin before /usr/bin – like I said before, no idea how homebrew handles this, but you might give the following command a try, to see if there is something like gcc-4.6:
ls -l /usr/local/bin | grep gcc
This will list all files named like gcc in /usr/local/bin – supposing homebrew puts it in there.
Progress! Got past the above using gcc-4.8 and CUDA 5.5 Toolkit. Here is the current error:
/usr/local/cuda/bin/nvcc -O3 -m64 -arch=compute_10 --maxrregcount=124 --ptxas-options=-v -o salsa_kernel.o -c salsa_kernel.cu /usr/local/cuda/bin/nvcc -O3 -m64 -Xptxas "-abi=no -v" -arch=compute_30 --maxrregcount=63 -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 4 (Illegal Instruction) make[2]: *** [spinlock_kernel.o] Error 4 make[1]: *** [all-recursive] Error 1 make: *** [all] Error 2
all the gcc-4.X versions are in /usr/local/bin where homebrew puts them, just not sure how to make sure they're referenced before the Apple clang other than using CC=gcc-4.6 in .profile.
@tzbauknight Well, that's the reason you should use CUDA 5.0. However, if you want to try it with GCC 4.6 and CUDA 5.0 again, you could try using an alias like:
alias gcc='gcc-4.6'
i am up to the same place as tzbauknight. I understand some other people have gotten further, could be the exact version of the cuds drivers that are at fault but I do know people with 5.5 who have managed to compile cudaminer!
