If you can wait a bit, I will commit the changes to github.
Thank You! I can wait.
Topic: CCminer M7 (XCN) by djm34, fixed + optimized for cuda 8 and new cards by PALLAS - page 10. (Read 52620 times)
October 15, 2016, 04:25:52 PM
Thank You! I can wait.
October 15, 2016, 04:18:01 PM
If you can wait a bit, I will commit the changes to github.
October 15, 2016, 04:15:01 PM
nvcc is not in the PATH.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
OK, GOOD, NVCC IS THERE--
But, then "make" failed on not finding "gcc-4.9". An attempt to install v4.9 revealed that gcc-4.9 was up-to-date. I then created a symlink to the "/usr/bin/gcc-4.8" directory with "sudo ln -s /usr/bin/gcc-4.8 /usr/bin/gcc-4.9".
On returning to the "pallas" directory, I issued the command "make", and the compilation continued from where it had errored out. The compilation is running while I type this.
VOILA' It compiled!!!! I ran the "--help" command. Right now, the machine is mining Ethereum (ETH), and I will check out your work sometime today. Thank you, Pallas!
--scryptr
Evening scryptr - I tried your method above - in full disclosure - I am a Ubuntu noob. I keep running into this error and have tried now for 24 hours with no luck to get past this point. I am running Ubuntu 14.04 LTS.
The error:
nvcc -g -O2 -I . -Xptxas "-v" --compiler-bindir /usr/bin/gcc-4.9 -gencode=arch=compute_50,code=\"sm_50,compute_50\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" --maxrregcount=80 --ptxas-options=-v -I./compat/jansson -o heavy/cuda_blake512.o -c heavy/cuda_blake512.cu
heavy/cuda_blake512.cu(11): error: invalid redeclaration of type name "uint64_t"
/usr/include/stdint.h(55): here
1 error detected in the compilation of "/tmp/tmpxft_00003be0_00000000-9_cuda_blake512.compute_52.cpp1.ii".
make[2]: *** [heavy/cuda_blake512.o] Error 2
make[2]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make: *** [all] Error 2
Any help is appreciated.
Thanks - pokeytex
your system headers have those typedefs already defined. probably gcc 5.4, the issue wasn't there with older compilers.
you need to remove those definitions at the beginning of the related files.
there are a bunch.
or use gcc 4.9 (the default for this fork).
@pallas I am not sure what you mean by typedefs. I am running Ubuntu 14.04 - clean install on an intel box. I believe the 64 bit version. I installed all basic stuff and have followed along on all of the tutorials so far. The error I am getting is based in GCC - I downloaded gcc 4.9 but get the same error still. I refuse to give up! :-) Any more advice please? With a cherry on top!
October 15, 2016, 01:38:04 PM
nvcc is not in the PATH.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
OK, GOOD, NVCC IS THERE--
But, then "make" failed on not finding "gcc-4.9". An attempt to install v4.9 revealed that gcc-4.9 was up-to-date. I then created a symlink to the "/usr/bin/gcc-4.8" directory with "sudo ln -s /usr/bin/gcc-4.8 /usr/bin/gcc-4.9".
On returning to the "pallas" directory, I issued the command "make", and the compilation continued from where it had errored out. The compilation is running while I type this.
VOILA' It compiled!!!! I ran the "--help" command. Right now, the machine is mining Ethereum (ETH), and I will check out your work sometime today. Thank you, Pallas!
--scryptr
Evening scryptr - I tried your method above - in full disclosure - I am a Ubuntu noob. I keep running into this error and have tried now for 24 hours with no luck to get past this point. I am running Ubuntu 14.04 LTS.
The error:
nvcc -g -O2 -I . -Xptxas "-v" --compiler-bindir /usr/bin/gcc-4.9 -gencode=arch=compute_50,code=\"sm_50,compute_50\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" --maxrregcount=80 --ptxas-options=-v -I./compat/jansson -o heavy/cuda_blake512.o -c heavy/cuda_blake512.cu
heavy/cuda_blake512.cu(11): error: invalid redeclaration of type name "uint64_t"
/usr/include/stdint.h(55): here
1 error detected in the compilation of "/tmp/tmpxft_00003be0_00000000-9_cuda_blake512.compute_52.cpp1.ii".
make[2]: *** [heavy/cuda_blake512.o] Error 2
make[2]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make: *** [all] Error 2
Any help is appreciated.
Thanks - pokeytex
your system headers have those typedefs already defined. probably gcc 5.4, the issue wasn't there with older compilers.
you need to remove those definitions at the beginning of the related files.
there are a bunch.
or use gcc 4.9 (the default for this fork).
October 15, 2016, 01:31:53 PM
nvcc is not in the PATH.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
find it and then:
export PATH=***nvcc directory***:$PATH
if you install cuda by .run file, it will be into a subdir of /usr/local, don't know about the deb method.
you can run "find / -xdev -name nvcc" if unsure.
OK, GOOD, NVCC IS THERE--
But, then "make" failed on not finding "gcc-4.9". An attempt to install v4.9 revealed that gcc-4.9 was up-to-date. I then created a symlink to the "/usr/bin/gcc-4.8" directory with "sudo ln -s /usr/bin/gcc-4.8 /usr/bin/gcc-4.9".
On returning to the "pallas" directory, I issued the command "make", and the compilation continued from where it had errored out. The compilation is running while I type this.
VOILA' It compiled!!!! I ran the "--help" command. Right now, the machine is mining Ethereum (ETH), and I will check out your work sometime today. Thank you, Pallas!
--scryptr
Evening scryptr - I tried your method above - in full disclosure - I am a Ubuntu noob. I keep running into this error and have tried now for 24 hours with no luck to get past this point. I am running Ubuntu 14.04 LTS.
The error:
nvcc -g -O2 -I . -Xptxas "-v" --compiler-bindir /usr/bin/gcc-4.9 -gencode=arch=compute_50,code=\"sm_50,compute_50\" -gencode=arch=compute_52,code=\"sm_52,compute_52\" --maxrregcount=80 --ptxas-options=-v -I./compat/jansson -o heavy/cuda_blake512.o -c heavy/cuda_blake512.cu
heavy/cuda_blake512.cu(11): error: invalid redeclaration of type name "uint64_t"
/usr/include/stdint.h(55): here
1 error detected in the compilation of "/tmp/tmpxft_00003be0_00000000-9_cuda_blake512.compute_52.cpp1.ii".
make[2]: *** [heavy/cuda_blake512.o] Error 2
make[2]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/pokeytex/Downloads/ccminer-m7-branch-master'
make: *** [all] Error 2
Any help is appreciated.
Thanks - pokeytex
October 14, 2016, 09:27:43 PM
Honestly this topic is ...
Still no windows compile after few weeks.
We better donate to tpruvot, he's the man !!
Peace
Still no windows compile after few weeks.
We better donate to tpruvot, he's the man !!
Peace
October 13, 2016, 04:01:00 AM
Would there be much involved in changing the cuda toolkit used to 8.0 so it compiles with VS2015?
It does use cuda 8 already and compiles fine on linux and, I assume, Windows with mingw. For visual studio, I think you need to update the project file.
That's what I thought but when using the master branch from github it tries to use 6.5.
line 54: "The imported project "C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V140\BuildCustomizations\CUDA 6.5.props" was not found" I only have 8.0 in that folder.
October 13, 2016, 02:06:08 AM
Would there be much involved in changing the cuda toolkit used to 8.0 so it compiles with VS2015?
It does use cuda 8 already and compiles fine on linux and, I assume, Windows with mingw. For visual studio, I think you need to update the project file.
October 12, 2016, 08:56:55 PM
Would there be much involved in changing the cuda toolkit used to 8.0 so it compiles with VS2015?
October 12, 2016, 10:37:30 AM
Code:
m7_keccak512_cpu_hash(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_sha512_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], d_prod1[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 8, 8, d_prod1[thr_id], KeccakH[thr_id], d_prod0[thr_id],order); //64
// MyStreamSynchronize(0,order++,thr_id);
m7_whirlpool512_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id],8, 16, KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order); //128
// MyStreamSynchronize(0,order++,thr_id);
m7_sha256_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 4, 24, KeccakH[thr_id], d_prod1[thr_id], d_prod0[thr_id],order); //96
// MyStreamSynchronize(0,order++,thr_id);
m7_haval256_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 4, 28, KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order); //112
// MyStreamSynchronize(0,order++,thr_id);
m7_tiger192_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_bigmul_unroll1_cpu(thr_id, throughput*tp_coef_f[thr_id], KeccakH[thr_id], d_prod1[thr_id], d_prod0[thr_id],order);
// MyStreamSynchronize(0,order++,thr_id);
m7_ripemd160_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_bigmul_unroll2_cpu(thr_id, throughput*tp_coef_f[thr_id], KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order);
// MyStreamSynchronize(0,order++,thr_id);
uint32_t foundNonce = m7_sha256_cpu_hash_300(thr_id, throughput*tp_coef_f[thr_id], pdata[29], NULL, d_prod1[thr_id], order);
someone must rewrite this(join in one function) to prevent to many memory writes... like Tanguy Pruvot did with lbry miner.
no, it will be much slower.
you can do it with lbry because they are a few little hashes, those are 7 (+ all the muls) and some are much heavier (like whirlpool).
October 12, 2016, 10:22:02 AM
Code:
m7_keccak512_cpu_hash(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_sha512_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], d_prod1[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 8, 8, d_prod1[thr_id], KeccakH[thr_id], d_prod0[thr_id],order); //64
// MyStreamSynchronize(0,order++,thr_id);
m7_whirlpool512_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id],8, 16, KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order); //128
// MyStreamSynchronize(0,order++,thr_id);
m7_sha256_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 4, 24, KeccakH[thr_id], d_prod1[thr_id], d_prod0[thr_id],order); //96
// MyStreamSynchronize(0,order++,thr_id);
m7_haval256_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
cpu_mulT4(0, throughput*tp_coef_f[thr_id], 4, 28, KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order); //112
// MyStreamSynchronize(0,order++,thr_id);
m7_tiger192_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_bigmul_unroll1_cpu(thr_id, throughput*tp_coef_f[thr_id], KeccakH[thr_id], d_prod1[thr_id], d_prod0[thr_id],order);
// MyStreamSynchronize(0,order++,thr_id);
m7_ripemd160_cpu_hash_120(thr_id, throughput*tp_coef_f[thr_id], pdata[29], KeccakH[thr_id], order++);
m7_bigmul_unroll2_cpu(thr_id, throughput*tp_coef_f[thr_id], KeccakH[thr_id], d_prod0[thr_id], d_prod1[thr_id],order);
// MyStreamSynchronize(0,order++,thr_id);
uint32_t foundNonce = m7_sha256_cpu_hash_300(thr_id, throughput*tp_coef_f[thr_id], pdata[29], NULL, d_prod1[thr_id], order);
someone must rewrite this(join in one function) to prevent to many memory writes... like Tanguy Pruvot did with lbry miner.
October 12, 2016, 06:16:31 AM
hm, is there source code? just require compile?
yes it's on git, see first post.
what version there is there on git, the one that does 20MH?
the one that works on cuda8 and pascal and is about 10% faster than the code base (djm version).
October 12, 2016, 05:17:44 AM
hm, is there source code? just require compile?
yes it's on git, see first post.
what version there is there on git, the one that does 20MH?
October 12, 2016, 04:54:34 AM
hm, is there source code? just require compile?
yes it's on git, see first post.
October 12, 2016, 03:52:59 AM
hm, is there source code? just require compile?
October 11, 2016, 11:26:06 PM
Weird there isn't anyone around willing to compile this. I even offered some devs BTC to compile it and fix the bugs, no one has responded.
October 11, 2016, 09:35:29 AM
I still have a 2y/o copy of djm34's Cryptonite miner m7_v5 for windows using cuda 6.5 if anyone wants.
Was the best I could find for gtx 650/750/750ti back then.
Gets around 5MH/s on the 750ti and works well with Suprnova.
Was the best I could find for gtx 650/750/750ti back then.
Gets around 5MH/s on the 750ti and works well with Suprnova.
October 10, 2016, 07:58:35 PM
Someone can link me his compile windows version please?
Thanks
Thanks
October 10, 2016, 01:27:27 PM
Cuda 8 final is out, no need to register and get the rc.
GOOD! WE CAN USE WGET--
The packages can be downloaded to a headless rig with wget, lynx, or curl under Linux. Then simply read the online Install manual at nVidia, and stepwise upgrade to CUDA 8.0. My 750ti rig is now CUDA 8.0 compliant. I was able to compile KlausT's CCminer-CryptoNight and it will mine on NiceHash. He updated the extranonce functions required there from tsiv's 2014 version.
--scryptr
October 09, 2016, 12:07:54 PM
Cuda 8 final is out, no need to register and get the rc.
Jump to: