[ANN] CPU Miner Advanced v5.0.2.0 Beta By Spoetnik

[Spoetnik]

i'm going to add a newer test build pretty soon prob after i eat lunch lol
i want to add a command-line option this time though so we can try out the test cheat mode in one single build.
i tried actually chopping down the hasing function to it's bare minimum then i moved the init out of the loop
and then i went even further and deleted the hashing function and just copied the hashing routines
straight into the scan-hash function.. it's working good

plus i figured out a better version of the cheat mode i think..
rather than trying to piss around in Groestl in rounds i had a better idea.
this time i checked right after hashing keccak to see if the first byte in the hash is 1
and if it is ? then i invalidate the hash and have it go to a label where it will deliberatly fail the hash check (and do another hash)
skipping ALL the rounds entirely lol

the new cheat mode test result i think is a bit quicker (after dividing the hash rate by two for the UI)
and you can see a LOT of shares are getting submitted really well and the hash rate reported on the pool is looking good
and the pool's var-diff has at times given me far higher difficulties than ever before.. so i think that is a good sign.

i'll post what code i am testing out right this minute if people want to see what i mean..

int scanhash_jackpot(int thr_id, uint32_t *pdata, const uint32_t *ptarget,
   uint32_t max_nonce, unsigned long *hashes_done)
{
   uint32_t n = pdata[19] - 1;
   const uint32_t first_nonce = pdata[19];

   uint32_t endiandata[22];
   uint32_t hash[8] __attribute__((aligned(32)));

   jackpothash_context_holder ctx;
   memcpy(&ctx, &base_contexts, sizeof(base_contexts));

   int kk=0;
   for (; kk < 22; kk++){be32enc(&endiandata[kk], ((uint32_t*)pdata)[kk]);};

   do {
      pdata[19] = ++n;
      be32enc(&endiandata[19], n);
      //jackpothash(hash, &endiandata);

      sph_keccak512 (&ctx.ctx_keccak, &endiandata, 80);
      sph_keccak512_close(&ctx.ctx_keccak, hash);
      if (hash[0]&0x01) {
         hash[7]=0x0;
         goto done;
      }
      int x=0;
      for (int r = 0; r < 3; r++) {
         //if ((hash[0]&0x01)&&(x!=1)) {
         if (hash[0]&0x01) {
            sph_groestl512 (&ctx.ctx_groestl, hash, 64);
            sph_groestl512_close(&ctx.ctx_groestl, hash);
            //x=1;
            //hash[7]=0x0;
            //goto done;
         } else {
            sph_skein512 (&ctx.ctx_skein, hash, 64);
            sph_skein512_close(&ctx.ctx_skein, hash);
         }
         if (hash[0]&0x01) {
            sph_blake512 (&ctx.ctx_blake, hash, 64);
            sph_blake512_close(&ctx.ctx_blake, hash);
         } else {
            sph_jh512 (&ctx.ctx_jh, hash, 64);
            sph_jh512_close(&ctx.ctx_jh, hash);
         }
      }
      done:
      if ((hash[7]<=ptarget[7])&&(hash[7]!=0x0)) {
         *hashes_done = (n-first_nonce+1)/2;
         return 1;
      }

   } while (n < max_nonce && !work_restart[thr_id].restart);

byebye:
   *hashes_done = (n-first_nonce+1)/2;
   pdata[19] = n;
   return 0;
}

[Spoetnik]

http://www.multiupload.nl/05HQV1EELM

that is CMA-5.0.1.0-Beta-WIN-x64-Haswell.rar (1.33 MB)

has 6 windows 64bit versions.. sse2,sse4,avx2 (cheat test and normal versions) plus source code.

edit:
if your building by source code you have to change some lines in jackpot_scanhash()
i commented out the cheat code for now until i add a command-line to turn it on or off.
if you can code you should see what i did in that one file (no other files need changing, only jackpot.c)

[goodluck0319]

it seems the new version is slower than the old one. I thougth it is going to be faster. it seems not.

[Spoetnik]

which one did you test to which one and HOW did you test ? benchmark ? timer benchmark ?
pool or solo ? i uploaded 10 builds eh..
also you didn't give it enough time if you judging by pool numbers they take time to average out and i just uploaded the files lol
did you compile it yourself ?

edit:
oh and to anyone reading.. try using the first miner the Jackpot coin dev posted then try mine
my AMD Athlon x4 9950 BE (overclocked a bit) went from 81kh/s to 266kh/s on JPC lol

[goodluck0319]

I test 5.0.0.8 w/ cheat vesion (minerd_AVX2_CHEAT_stripped). now I am running with this new version w/ avx2.

same mining sites but different version.

5.0.08 is 1000k and new version is about half of that. is that normal?

I use pool on verters.com

[Spoetnik]

yeah your right and i made a mistake somehow.. i just checked everything in the builds and source of both versions.
the code i put in before to divide the hash rate by 2 failed to work so it's showing twice as fast as it should on the old v5008 ver.
i put in some defines in the code that didn't seem to work.. the cheat mode is setup but not offsetting the Reported UI hash speed.
and i last uploaded a version not using a c code define and simply put a divider in specifically.

line 909 in cpuminer.c in the old version is where it should have changed the hash rate and then straight in jackpot_scanhash() on the new version.

so anyway you can look at it this way.. one is wrong and one is right but they are the same lol
i know confusing hhahaa

if you double the hash rate it will calculate like old build to compare or cut the old hash rate in half to compare
either way i would pay more attention to pool performance on cheat test builds.. that gauges real world performance depending on how accurate the pool is i guess lol

so sorry about the confusion,
but i intended to modify the older version before so the hash rate was cut in half, but it's not doing it :/
so it's showing double what i intended on v5.0.0.8.

does that make sense ?
the older one used a different approach/method to the cheat mode vs the new one,
but they are similar in reported hash rate overall but the newer one i think looks better on on pool stats.
maybe try both on some pools and see what it tells you your hashing it etc.

[goodluck0319]

right now I am using 4 threads on my i7 4770k cpu.

one 4 threads is using avx2 and other 4 threads is sse4 miner. it is strange for me that both had the same hash speed when I mine. is that normal?

[Spoetnik]

you mean using sse4 on on 4770k and avx2 on another 4770k ?

speeds can vary because of a lot of factors and the difference sometimes from on build to another can be pretty small.
and sometimes the pools i am on for a while report what you said eh
maybe double check what version(s) your using and scan-time setting too maybe..
having scan-time at 30 on a worker and 1 on another can make them different for example.

so sorry i am no to sure what else to say lol

thanks for the feedback though.. it's nice when people take the time.

edit:
i always run my 4770k with 6 threads so i have 2 free for various other stuff and if i don't bog my machine down i average around 500kh/s
but some versions seem to be better when i look at pool stats eh.. how much coins you make is all that matters

[Spoetnik]

last version i added should have been reporting a lot faster speeds.

i already added some new features and have some more good ones coming pretty soon.
as well as spotting and fixing bugs.

i am thinking i will port chubners cudaminer multi-pool support too
he added a feature a while back i didn't know about until today.. specifying multiple pools on the cmd-line.

so stay tuned i didn't get bored and wander off lol

also i did a lot more testing with compiler settings and benchmarking them,
and also added build version info when using version cmd-line showing instruction set info and libcurl etc and base cpu miner version used.
this shows what i mean..

C:\minerd --version

cpuminer 2.3.3
 built on May  9 2014
 features: x86_64 SSE2 AVX AVX2 XOP
libcurl/7.36.0

Press any key to continue . . .

i have gone through EVERY single line of code numerous times and i ported Cpuminer v2.3.3 changes to this build manually (was using 2.3.2)
i also just finished building & updating with the newest version of LibCurl.. the version in my quote.

[goodluck0319]

so your latest version support x32 and x64 bit , is that correct?

[SolidStateSurvivor]

For curiosity's sake, I tried building v5.0.1.0 on Linux, but I only get errors, yet this is posix type source code these are the results of the Debianesque jury:

Code:

gcc -std=gnu99 -DHAVE_CONFIG_H -I. -pthread -fno-strict-aliasing     -O3 -MT minerd-cpu-miner.o -MD -MP -MF .deps/minerd-cpu-miner.Tpo -c -o minerd-cpu-miner.o `test -f 'cpu-miner.c' || echo './'`cpu-miner.c
cpu-miner.c: In function ‘share_result’:
cpu-miner.c:366: error: ‘LOG_TEST’ undeclared (first use in this function)
cpu-miner.c:366: error: (Each undeclared identifier is reported only once
cpu-miner.c:366: error: for each function it appears in.)
cpu-miner.c: In function ‘stratum_thread’:
cpu-miner.c:1137: error: ‘LOG_SPECIAL’ undeclared (first use in this function)
cpu-miner.c: In function ‘main’:
cpu-miner.c:1482: warning: implicit declaration of function ‘SetConsoleTitleA’
cpu-miner.c:1515: warning: implicit declaration of function ‘SetConsoleTextAttribute’
cpu-miner.c:1515: warning: implicit declaration of function ‘GetStdHandle’
cpu-miner.c:1515: error: ‘STD_OUTPUT_HANDLE’ undeclared (first use in this function)
cpu-miner.c:1515: error: ‘FOREGROUND_BLUE’ undeclared (first use in this function)
cpu-miner.c:1516: error: ‘FOREGROUND_INTENSITY’ undeclared (first use in this function)
cpu-miner.c:1519: error: ‘FOREGROUND_GREEN’ undeclared (first use in this function)
cpu-miner.c:1527: error: ‘FOREGROUND_RED’ undeclared (first use in this function)
make[2]: *** [minerd-cpu-miner.o] Error 1
make[2]: Leaving directory `/mnt/src/jpc-cpuminer-master-v2'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/mnt/src/jpc-cpuminer-master-v2'
make: *** [all] Error 2

Maybe I could get rid of all the colours, but there are still LOG_TEST and LOG_SPECIAL issues to deal with.

[Spoetnik]

hmmm interesting and thanks for the feedback i wondered before if that would be a problem eh
i didn't think it would since i have been building on MingW and MingW64,
it would be a little bit of work to get rid of the colors but i'll look into it and see what i can do..
i have tried to keep it Linux + Windows eh even though i am mostly into windows coding and don't have any Linux distribs running for testing at the moment.

i was just trying to get it running on visual studio 2010 but got almost done and got some old errors and got tired of doing it.. some other time lol

also x86 / 32bit is working but all optimization flags i try with MingW seem to be a hell of a lot slower than the 64bit counterpart builds for some odd reason.

in the last year i have started over making a cpu miner mod using chunks of hashing algo's from other mods
but i am getting tired of starting over again and hope to just stick with this one and port new coins to this one and only this one now.

got some good features planned and coming and already did some..
right now i am testing out having the average hash rate shown in the window title every 5 seconds
Note: currently as *always the colored hash lines in the console show the fastest (submitted good share) hash rate.
So i was thinking it would be good to see what the average hash rate is going on in the background if using the quiet cmd-line..
i always run in quiet mode eh..
Also detected new block messages are normally suppressed as well as debug reject reason in quiet mode but i enabled them for a good reason..
If someone is getting a lot of bad shares and the reason they see is because it says "Job Not Found"
people really may want to consider tuning their scan time (probably lower)
although i have considered adding a cmd-line option to hide the "new block" msg's..

i also added some color and fixed up the benchmark code a bit.. it should be a lot better soon.
and i also added a cmd-line to enable cheat mode testing so no more multiple builds needed..
i think i did a ton of other crap too i forget and will have to compare the files before i do a changelog update again lol

it's a lot of screwing around cleaning up the source code folder and putting together a new packages of builds eh
but of someone wants the build i am currently running for haswell or SSE2 or SSE4 just say so and i can upload just that one .exe pretty damn fast lol

so far the only way i know to build for x86 / 32bit windows is to re-compile the LibCurl install and that is a big time consuming pain the ass
i want to see later if there is some way i can have both compiled and installed setup so i just have to configure again in the mod folder only..
it means i am coding with x64 so i have to delete everything and rebuild for x86 then i have to rebuild it back to x64 again..
and even on a fast machine LibCurl takes time
i have a sick fast SSD (Samsung 840 EVO 256GB SSD Sata III) and my 4770k cpu only uses like 10% when compiling but it still takes a while lol
edit: i also am using G.SKILL Trident X Series 16GB (2 x 8GB) DDR3 2400 - Memory i paid good money for running with xmp profile at 2400mhz

a couple things i want to add..
- Multi pool fall back support (i can pretty much copy and past it from cudaminer)
- Diff modifier via cmd-line (same i can get it from cudaminer or an sgminer mod) so far i have been hard coding values for testing in the last year.
- more hash stats in console window.. i was thinking of maybe something like this...  CMA v5.5.5.5 [current% kh/s] [avg% kh/s] [highest% kh/s]
currently i am seeing a 5 second polled avg but i wrote the code and never did that before and i am not sure what i am doing lol
i had gettimeofday() start in Main then in MinerThread at the end i had it check if time has been 5 seconds and if so update the console window
and then start a new timer basically..
but what i am not sure of is what i should use to average it out..
i have tried 2 or 3 variables combined..
avg = (lowest_hash + hashrate) /2;
or
avg = (lowest_hash + hashrate + highest_hashrate) /3;

edit:
i tried also,
avg = (lowest_hash + highest_hashrate) /2;
but that seemed the most out of whack out of the three averaging strategies i can think of..

[Spoetnik]

so your latest version support x32 and x64 bit , is that correct?

no i almost uploaded a 32bit version i was packing it and testing an noticed some bugs so i got coding again..
i can upload 32bit but it's going to be waaay slower than x64 but prob faster than the 32bit version the JPC coin dev posted.

[Spoetnik]

For curiosity's sake, I tried building v5.0.1.0 on Linux, but I only get errors, yet this is posix type source code these are the results of the Debianesque jury:

Code:

gcc -std=gnu99 -DHAVE_CONFIG_H -I. -pthread -fno-strict-aliasing     -O3 -MT minerd-cpu-miner.o -MD -MP -MF .deps/minerd-cpu-miner.Tpo -c -o minerd-cpu-miner.o `test -f 'cpu-miner.c' || echo './'`cpu-miner.c
cpu-miner.c: In function ‘share_result’:
cpu-miner.c:366: error: ‘LOG_TEST’ undeclared (first use in this function)
cpu-miner.c:366: error: (Each undeclared identifier is reported only once
cpu-miner.c:366: error: for each function it appears in.)
cpu-miner.c: In function ‘stratum_thread’:
cpu-miner.c:1137: error: ‘LOG_SPECIAL’ undeclared (first use in this function)
cpu-miner.c: In function ‘main’:
cpu-miner.c:1482: warning: implicit declaration of function ‘SetConsoleTitleA’
cpu-miner.c:1515: warning: implicit declaration of function ‘SetConsoleTextAttribute’
cpu-miner.c:1515: warning: implicit declaration of function ‘GetStdHandle’
cpu-miner.c:1515: error: ‘STD_OUTPUT_HANDLE’ undeclared (first use in this function)
cpu-miner.c:1515: error: ‘FOREGROUND_BLUE’ undeclared (first use in this function)
cpu-miner.c:1516: error: ‘FOREGROUND_INTENSITY’ undeclared (first use in this function)
cpu-miner.c:1519: error: ‘FOREGROUND_GREEN’ undeclared (first use in this function)
cpu-miner.c:1527: error: ‘FOREGROUND_RED’ undeclared (first use in this function)
make[2]: *** [minerd-cpu-miner.o] Error 1
make[2]: Leaving directory `/mnt/src/jpc-cpuminer-master-v2'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/mnt/src/jpc-cpuminer-master-v2'
make: *** [all] Error 2

Maybe I could get rid of all the colours, but there are still LOG_TEST and LOG_SPECIAL issues to deal with.

hey run clean.bat and try this and let me know if it works

the run mingW console.. then
cd /c/some/path/etc
./autogen.sh
...........if for windows x86
./configure --host=x86_64-w64-mingw32 CPPFLAGS="-DWIN32_LEAN_AND_MEAN" CFLAGS="-O3 -march=native -s"
----------else x64
./configure --host=x86_64-w64-mingw32 CFLAGS="-O3 -march=native -s"
make
  
if you get two errors about mstcip.h and another ms header use the define, WIN32_LEAN_AND_MEAN before windows.h in cpuminer.c
or you can copy the two files from the 64bit mingw64 install if you got it and copy them into the include folder you found them in the 64 bit version.
and if using LibCurl 7.36 you need to patch a file lol

patch tool_hugehelp.h like this..

#ifndef HEADER_CURL_TOOL_HUGEHELP_H
#define HEADER_CURL_TOOL_HUGEHELP_H
/* blah blha blah*/
 #include "tool_setup.h"
#ifdef USE_MANUAL
 void hugehelp(void);
#endif /* USE_MANUAL */
 #endif /* HEADER_CURL_TOOL_HUGEHELP_H */

then make sure to use the disable command for curl when setting it up..

x64 LibCurl configure cmd-line example.

./configure --host=x86_64-w64-mingw32 –-prefix=/c/mingw64 --disable-manual

anyways...
edit Makefile.am contents below (added color.h now)

minerd_SOURCES   = elist.h miner.h compat.h color.h \
        cpu-miner.c util.c \
        sha2.c sha2-arm.S sha2-x86.S sha2-x64.S \
        scrypt.c scrypt-arm.S scrypt-x86.S scrypt-x64.S \
        blake.c bmw.c   groestl.c jh.c keccak.c skein.c \
        quark.c jackpot.c

[Spoetnik]

curl-config needs to be re-copied if your flipping back and forth from x86 to x64 builds too

C:\MinGW\bin\curl-config

the contents of that file need to be updated if flipping from one to the other (x86/x64)
reconfiguring LibCurl and running make clean etc will not fix this.. you HAVE to copy it manually yourself or do a cp command etc..

the first line says for me this

prefix=/c/mingw64
in file,
C:\MinGW\bin\curl-config
i am building on x64 right now..

if i do x86 now i need to go cd to the Libcurl folder and run "make clean"
configure for x86 then build..
after that i have to make sure,
C:\MinGW\bin\curl-config
says,
prefix=/c/mingw

of course your paths may be different guys
i installed MingW to my c drive on windows.

note:
there is three lines that are changed when switching back and forth in curl-config so don't just change that one i mentioned earlier
back up the original file and copy it back when needed or edit all three lines manually..

[SolidStateSurvivor]

Maybe I could get rid of all the colours, but there are still LOG_TEST and LOG_SPECIAL issues to deal with.

hey run clean.bat and try this and let me know if it works

the run mingW console.. then
cd /c/some/path/etc
./autogen.sh
...........if for windows x86
./configure --host=x86_64-w64-mingw32 CPPFLAGS="-DWIN32_LEAN_AND_MEAN" CFLAGS="-O3 -march=native -s"
----------else x64
./configure --host=x86_64-w64-mingw32 CFLAGS="-O3 -march=native -s"
make
  
if you get two errors about mstcip.h and another ms header use the define, WIN32_LEAN_AND_MEAN before windows.h in cpuminer.c
or you can copy the two files from the 64bit mingw64 install if you got it and copy them into the include folder you found them in the 64 bit version.
and if using LibCurl 7.36 you need to patch a file lol

patch tool_hugehelp.h like this..

#ifndef HEADER_CURL_TOOL_HUGEHELP_H
#define HEADER_CURL_TOOL_HUGEHELP_H
/* blah blha blah*/
 #include "tool_setup.h"
#ifdef USE_MANUAL
 void hugehelp(void);
#endif /* USE_MANUAL */
 #endif /* HEADER_CURL_TOOL_HUGEHELP_H */

then make sure to use the disable command for curl when setting it up..

x64 LibCurl configure cmd-line example.

./configure --host=x86_64-w64-mingw32 –-prefix=/c/mingw64 --disable-manual

anyways...
edit Makefile.am contents below (added color.h now)

minerd_SOURCES   = elist.h miner.h compat.h color.h \
        cpu-miner.c util.c \
        sha2.c sha2-arm.S sha2-x86.S sha2-x64.S \
        scrypt.c scrypt-arm.S scrypt-x86.S scrypt-x64.S \
        blake.c bmw.c   groestl.c jh.c keccak.c skein.c \
        quark.c jackpot.c

As I only compile / build on Linux, these things are hard to do, but I will what I can do to adopt it and incorporate these fixes into the Linux environment I'm using.

[Spoetnik]

basically i was giving tips for MingW guys.

and i think your problem may be solved by adding color.h to the makefile.am file.

[Spoetnik]

it took me some work but i went through every line of code *again and made prob a thousand edits
and now i got it compiling with Visual Studio 2010 and i am running the 64bit build right now compiled with VS2010

i just have to complete the 32bit configurations for debug and release.. and make sure it still builds on MingW.

I normally would add cpuminer code bases to a Visual Studio solution so i can view and edit source with Visual Assist X all black style
or some quick work with IDM's UltraEdit.. and Beyond Compare is massively useful !

having an IDE is nice.. notepad coding is gay lol

[SolidStateSurvivor]

basically i was giving tips for MingW guys.

and i think your problem may be solved by adding color.h to the makefile.am file.

Like this:

Makefile.am

if WANT_JANSSON
JANSSON_INCLUDES= -I$(top_srcdir)/compat/jansson
else
JANSSON_INCLUDES=
endif

EXTRA_DIST   = example-cfg.json nomacro.pl

SUBDIRS      = compat

INCLUDES   = $(PTHREAD_FLAGS) -fno-strict-aliasing $(JANSSON_INCLUDES)

bin_PROGRAMS   = minerd

minerd_SOURCES   = elist.h miner.h compat.h color.h \
        cpu-miner.c util.c \
        sha2.c sha2-arm.S sha2-x86.S sha2-x64.S \
        scrypt.c scrypt-arm.S scrypt-x86.S scrypt-x64.S \
        blake.c bmw.c   groestl.c jh.c keccak.c skein.c \
        quark.c \
        jackpot.c
minerd_LDFLAGS   = $(PTHREAD_FLAGS)
minerd_LDADD   = @LIBCURL@ @JANSSON_LIBS@ @PTHREAD_LIBS@ @WS2_LIBS@
minerd_CPPFLAGS = @LIBCURL_CPPFLAGS@

Hmm, nope that didn't work.

[Spoetnik]

hmm i don't know what to suggest
i assumed if what ever code i use compiles fine with MingW that uses GCC
then it should be fine for linux guys.. that is what i always hoped anyway lol
unless i can get a distrib setup i don't know what else to say.. sad there is prob a 100 people who could help you here but i doubt any will..
coders aren't very helpful in this scene at all.

supporting both are a problem i tried to setup a timer and the code i want to use works for both
but on Linux it meant soemthing different so i had to wrap the Timer Benchmark feature in Windows defines.