/*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see .
*
* Additional permission under GNU GPL version 3 section 7
*
* If you modify this Program, or any covered work, by linking or combining
* it with OpenSSL (or a modified version of that library), containing parts
* covered by the terms of OpenSSL License and SSLeay License, the licensors
* of this Program grant you additional permission to convey the resulting work.
*
*/
#include
#include
#include
#include
#include
#include
#include "console.h"
#ifdef _WIN32
#include
void thd_setaffinity(std::thread::native_handle_type h, uint64_t cpu_id)
{
SetThreadAffinityMask(h, 1ULL << cpu_id);
}
#else
#include
#if defined(__APPLE__)
#include
#include
#define SYSCTL_CORE_COUNT "machdep.cpu.core_count"
#elif defined(__FreeBSD__)
#include
#endif
void thd_setaffinity(std::thread::native_handle_type h, uint64_t cpu_id)
{
#if defined(__APPLE__)
thread_port_t mach_thread;
thread_affinity_policy_data_t policy = { static_cast(cpu_id) };
mach_thread = pthread_mach_thread_np(h);
thread_policy_set(mach_thread, THREAD_AFFINITY_POLICY, (thread_policy_t)&policy, 1);
#elif defined(__FreeBSD__)
cpuset_t mn;
CPU_ZERO(&mn);
CPU_SET(cpu_id, &mn);
pthread_setaffinity_np(h, sizeof(cpuset_t), &mn);
#else
cpu_set_t mn;
CPU_ZERO(&mn);
CPU_SET(cpu_id, &mn);
pthread_setaffinity_np(h, sizeof(cpu_set_t), &mn);
#endif
}
#endif // _WIN32
#include "executor.h"
#include "minethd.h"
#include "jconf.h"
#include "crypto/cryptonight_aesni.h"
#include "hwlocMemory.hpp"
telemetry::telemetry(size_t iThd)
{
ppHashCounts = new uint64_t*[iThd];
ppTimestamps = new uint64_t*[iThd];
iBucketTop = new uint32_t[iThd];
for (size_t i = 0; i < iThd; i++)
{
ppHashCounts[i] = new uint64_t[iBucketSize];
ppTimestamps[i] = new uint64_t[iBucketSize];
iBucketTop[i] = 0;
memset(ppHashCounts[0], 0, sizeof(uint64_t) * iBucketSize);
memset(ppTimestamps[0], 0, sizeof(uint64_t) * iBucketSize);
}
}
double telemetry::calc_telemetry_data(size_t iLastMilisec, size_t iThread)
{
using namespace std::chrono;
uint64_t iTimeNow = time_point_cast(high_resolution_clock::now()).time_since_epoch().count();
uint64_t iEarliestHashCnt = 0;
uint64_t iEarliestStamp = 0;
uint64_t iLastestStamp = 0;
uint64_t iLastestHashCnt = 0;
bool bHaveFullSet = false;
//Start at 1, buckettop points to next empty
for (size_t i = 1; i < iBucketSize; i++)
{
size_t idx = (iBucketTop[iThread] - i) & iBucketMask; //overflow expected here
if (ppTimestamps[iThread][idx] == 0)
break; //That means we don't have the data yet
if (iLastestStamp == 0)
{
iLastestStamp = ppTimestamps[iThread][idx];
iLastestHashCnt = ppHashCounts[iThread][idx];
}
if (iTimeNow - ppTimestamps[iThread][idx] > iLastMilisec)
{
bHaveFullSet = true;
break; //We are out of the requested time period
}
iEarliestStamp = ppTimestamps[iThread][idx];
iEarliestHashCnt = ppHashCounts[iThread][idx];
}
if (!bHaveFullSet || iEarliestStamp == 0 || iLastestStamp == 0)
return nan("");
//Don't think that can happen, but just in case
if (iLastestStamp - iEarliestStamp == 0)
return nan("");
double fHashes, fTime;
fHashes = iLastestHashCnt - iEarliestHashCnt;
fTime = iLastestStamp - iEarliestStamp;
fTime /= 1000.0;
return fHashes / fTime;
}
void telemetry::push_perf_value(size_t iThd, uint64_t iHashCount, uint64_t iTimestamp)
{
size_t iTop = iBucketTop[iThd];
ppHashCounts[iThd][iTop] = iHashCount;
ppTimestamps[iThd][iTop] = iTimestamp;
iBucketTop[iThd] = (iTop + 1) & iBucketMask;
}
minethd::minethd(miner_work& pWork, size_t iNo, bool double_work, bool no_prefetch, int affinity)
{
oWork = pWork;
bQuit = 0;
iThreadNo = (uint8_t)iNo;
iJobNo = 0;
iHashCount = 0;
iTimestamp = 0;
bNoPrefetch = no_prefetch;
this->affinity = affinity;
if(double_work)
oWorkThd = std::thread(&minethd::double_work_main, this);
else
oWorkThd = std::thread(&minethd::work_main, this);
}
std::atomic minethd::iGlobalJobNo;
std::atomic minethd::iConsumeCnt; //Threads get jobs as they are initialized
minethd::miner_work minethd::oGlobalWork;
uint64_t minethd::iThreadCount = 0;
cryptonight_ctx* minethd_alloc_ctx()
{
cryptonight_ctx* ctx;
alloc_msg msg = { 0 };
switch (jconf::inst()->GetSlowMemSetting())
{
case jconf::never_use:
ctx = cryptonight_alloc_ctx(1, 1, &msg);
if (ctx == NULL)
printer::inst()->print_msg(L0, "MEMORY ALLOC FAILED: %s", msg.warning);
return ctx;
case jconf::no_mlck:
ctx = cryptonight_alloc_ctx(1, 0, &msg);
if (ctx == NULL)
printer::inst()->print_msg(L0, "MEMORY ALLOC FAILED: %s", msg.warning);
return ctx;
case jconf::print_warning:
ctx = cryptonight_alloc_ctx(1, 1, &msg);
if (msg.warning != NULL)
printer::inst()->print_msg(L0, "MEMORY ALLOC FAILED: %s", msg.warning);
if (ctx == NULL)
ctx = cryptonight_alloc_ctx(0, 0, NULL);
return ctx;
case jconf::always_use:
return cryptonight_alloc_ctx(0, 0, NULL);
case jconf::unknown_value:
return NULL; //Shut up compiler
}
return nullptr; //Should never happen
}
bool minethd::self_test()
{
alloc_msg msg = { 0 };
size_t res;
bool fatal = false;
switch (jconf::inst()->GetSlowMemSetting())
{
case jconf::never_use:
res = cryptonight_init(1, 1, &msg);
fatal = true;
break;
case jconf::no_mlck:
res = cryptonight_init(1, 0, &msg);
fatal = true;
break;
case jconf::print_warning:
res = cryptonight_init(1, 1, &msg);
break;
case jconf::always_use:
res = cryptonight_init(0, 0, &msg);
break;
case jconf::unknown_value:
default:
return false; //Shut up compiler
}
if(msg.warning != nullptr)
printer::inst()->print_msg(L0, "MEMORY INIT ERROR: %s", msg.warning);
if(res == 0 && fatal)
return false;
cryptonight_ctx *ctx0, *ctx1;
if((ctx0 = minethd_alloc_ctx()) == nullptr)
return false;
if((ctx1 = minethd_alloc_ctx()) == nullptr)
{
cryptonight_free_ctx(ctx0);
return false;
}
unsigned char out[64];
bool bResult;
cn_hash_fun hashf;
cn_hash_fun_dbl hashdf;
hashf = func_selector(jconf::inst()->HaveHardwareAes(), false);
hashf("This is a test", 14, out, ctx0);
bResult = memcmp(out, "\xa0\x84\xf0\x1d\x14\x37\xa0\x9c\x69\x85\x40\x1b\x60\xd4\x35\x54\xae\x10\x58\x02\xc5\xf5\xd8\xa9\xb3\x25\x36\x49\xc0\xbe\x66\x05", 32) == 0;
hashf = func_selector(jconf::inst()->HaveHardwareAes(), true);
hashf("This is a test", 14, out, ctx0);
bResult &= memcmp(out, "\xa0\x84\xf0\x1d\x14\x37\xa0\x9c\x69\x85\x40\x1b\x60\xd4\x35\x54\xae\x10\x58\x02\xc5\xf5\xd8\xa9\xb3\x25\x36\x49\xc0\xbe\x66\x05", 32) == 0;
hashdf = func_dbl_selector(jconf::inst()->HaveHardwareAes(), false);
hashdf("The quick brown fox jumps over the lazy dogThe quick brown fox jumps over the lazy log", 43, out, ctx0, ctx1);
bResult &= memcmp(out, "\x3e\xbb\x7f\x9f\x7d\x27\x3d\x7c\x31\x8d\x86\x94\x77\x55\x0c\xc8\x00\xcf\xb1\x1b\x0c\xad\xb7\xff\xbd\xf6\xf8\x9f\x3a\x47\x1c\x59"
"\xb4\x77\xd5\x02\xe4\xd8\x48\x7f\x42\xdf\xe3\x8e\xed\x73\x81\x7a\xda\x91\xb7\xe2\x63\xd2\x91\x71\xb6\x5c\x44\x3a\x01\x2a\x41\x22", 64) == 0;
hashdf = func_dbl_selector(jconf::inst()->HaveHardwareAes(), true);
hashdf("The quick brown fox jumps over the lazy dogThe quick brown fox jumps over the lazy log", 43, out, ctx0, ctx1);
bResult &= memcmp(out, "\x3e\xbb\x7f\x9f\x7d\x27\x3d\x7c\x31\x8d\x86\x94\x77\x55\x0c\xc8\x00\xcf\xb1\x1b\x0c\xad\xb7\xff\xbd\xf6\xf8\x9f\x3a\x47\x1c\x59"
"\xb4\x77\xd5\x02\xe4\xd8\x48\x7f\x42\xdf\xe3\x8e\xed\x73\x81\x7a\xda\x91\xb7\xe2\x63\xd2\x91\x71\xb6\x5c\x44\x3a\x01\x2a\x41\x22", 64) == 0;
cryptonight_free_ctx(ctx0);
cryptonight_free_ctx(ctx1);
if(!bResult)
printer::inst()->print_msg(L0,
"Cryptonight hash self-test failed. This might be caused by bad compiler optimizations.");
return bResult;
}
std::vector* minethd::thread_starter(miner_work& pWork)
{
iGlobalJobNo = 0;
iConsumeCnt = 0;
std::vector* pvThreads = new std::vector;
//Launch the requested number of single and double threads, to distribute
//load evenly we need to alternate single and double threads
size_t i, n = jconf::inst()->GetThreadCount();
pvThreads->reserve(n);
jconf::thd_cfg cfg;
for (i = 0; i < n; i++)
{
jconf::inst()->GetThreadConfig(i, cfg);
minethd* thd = new minethd(pWork, i, cfg.bDoubleMode, cfg.bNoPrefetch, cfg.iCpuAff);
if(cfg.iCpuAff >= 0)
{
#if defined(__APPLE__)
printer::inst()->print_msg(L1, "WARNING on MacOS thread affinity is only advisory.");
#endif
thd_setaffinity(thd->oWorkThd.native_handle(), cfg.iCpuAff);
}
pvThreads->push_back(thd);
if(cfg.iCpuAff >= 0)
printer::inst()->print_msg(L1, "Starting %s thread, affinity: %d.", cfg.bDoubleMode ? "double" : "single", (int)cfg.iCpuAff);
else
printer::inst()->print_msg(L1, "Starting %s thread, no affinity.", cfg.bDoubleMode ? "double" : "single");
}
iThreadCount = n;
return pvThreads;
}
void minethd::switch_work(miner_work& pWork)
{
// iConsumeCnt is a basic lock-like polling mechanism just in case we happen to push work
// faster than threads can consume them. This should never happen in real life.
// Pool cant physically send jobs faster than every 250ms or so due to net latency.
while (iConsumeCnt.load(std::memory_order_seq_cst) < iThreadCount)
std::this_thread::sleep_for(std::chrono::milliseconds(100));
oGlobalWork = pWork;
iConsumeCnt.store(0, std::memory_order_seq_cst);
iGlobalJobNo++;
}
void minethd::consume_work()
{
memcpy(&oWork, &oGlobalWork, sizeof(miner_work));
iJobNo++;
iConsumeCnt++;
}
minethd::cn_hash_fun minethd::func_selector(bool bHaveAes, bool bNoPrefetch)
{
// We have two independent flag bits in the functions
// therefore we will build a binary digit and select the
// function as a two digit binary
// Digit order SOFT_AES, NO_PREFETCH
static const cn_hash_fun func_table[4] = {
cryptonight_hash<0x80000, MEMORY, false, false>,
cryptonight_hash<0x80000, MEMORY, false, true>,
cryptonight_hash<0x80000, MEMORY, true, false>,
cryptonight_hash<0x80000, MEMORY, true, true>
};
std::bitset<2> digit;
digit.set(0, !bNoPrefetch);
digit.set(1, !bHaveAes);
return func_table[digit.to_ulong()];
}
void minethd::work_main()
{
// pin memory to NUMA node
bindMemoryToNUMANode(affinity);
cn_hash_fun hash_fun;
cryptonight_ctx* ctx;
uint64_t iCount = 0;
uint64_t* piHashVal;
uint32_t* piNonce;
job_result result;
hash_fun = func_selector(jconf::inst()->HaveHardwareAes(), bNoPrefetch);
ctx = minethd_alloc_ctx();
piHashVal = (uint64_t*)(result.bResult + 24);
piNonce = (uint32_t*)(oWork.bWorkBlob + 39);
iConsumeCnt++;
while (bQuit == 0)
{
if (oWork.bStall)
{
/* We are stalled here because the executor didn't find a job for us yet,
either because of network latency, or a socket problem. Since we are
raison d'etre of this software it us sensible to just wait until we have something*/
while (iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
std::this_thread::sleep_for(std::chrono::milliseconds(100));
consume_work();
continue;
}
if(oWork.bNiceHash)
result.iNonce = calc_nicehash_nonce(*piNonce, oWork.iResumeCnt);
else
result.iNonce = calc_start_nonce(oWork.iResumeCnt);
assert(sizeof(job_result::sJobID) == sizeof(pool_job::sJobID));
memcpy(result.sJobID, oWork.sJobID, sizeof(job_result::sJobID));
while(iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
{
if ((iCount & 0xF) == 0) //Store stats every 16 hashes
{
using namespace std::chrono;
uint64_t iStamp = time_point_cast(high_resolution_clock::now()).time_since_epoch().count();
iHashCount.store(iCount, std::memory_order_relaxed);
iTimestamp.store(iStamp, std::memory_order_relaxed);
}
iCount++;
*piNonce = ++result.iNonce;
hash_fun(oWork.bWorkBlob, oWork.iWorkSize, result.bResult, ctx);
if (*piHashVal < oWork.iTarget)
executor::inst()->push_event(ex_event(result, oWork.iPoolId));
std::this_thread::yield();
}
consume_work();
}
cryptonight_free_ctx(ctx);
}
minethd::cn_hash_fun_dbl minethd::func_dbl_selector(bool bHaveAes, bool bNoPrefetch)
{
// We have two independent flag bits in the functions
// therefore we will build a binary digit and select the
// function as a two digit binary
// Digit order SOFT_AES, NO_PREFETCH
static const cn_hash_fun_dbl func_table[4] = {
cryptonight_double_hash<0x80000, MEMORY, false, false>,
cryptonight_double_hash<0x80000, MEMORY, false, true>,
cryptonight_double_hash<0x80000, MEMORY, true, false>,
cryptonight_double_hash<0x80000, MEMORY, true, true>
};
std::bitset<2> digit;
digit.set(0, !bNoPrefetch);
digit.set(1, !bHaveAes);
return func_table[digit.to_ulong()];
}
void minethd::double_work_main()
{
// pin memory to NUMA node
bindMemoryToNUMANode(affinity);
cn_hash_fun_dbl hash_fun;
cryptonight_ctx* ctx0;
cryptonight_ctx* ctx1;
uint64_t iCount = 0;
uint64_t *piHashVal0, *piHashVal1;
uint32_t *piNonce0, *piNonce1;
uint8_t bDoubleHashOut[64];
uint8_t bDoubleWorkBlob[sizeof(miner_work::bWorkBlob) * 2];
uint32_t iNonce;
job_result res;
hash_fun = func_dbl_selector(jconf::inst()->HaveHardwareAes(), bNoPrefetch);
ctx0 = minethd_alloc_ctx();
ctx1 = minethd_alloc_ctx();
piHashVal0 = (uint64_t*)(bDoubleHashOut + 24);
piHashVal1 = (uint64_t*)(bDoubleHashOut + 32 + 24);
piNonce0 = (uint32_t*)(bDoubleWorkBlob + 39);
piNonce1 = nullptr;
iConsumeCnt++;
while (bQuit == 0)
{
if (oWork.bStall)
{
/* We are stalled here because the executor didn't find a job for us yet,
either because of network latency, or a socket problem. Since we are
raison d'etre of this software it us sensible to just wait until we have something*/
while (iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
std::this_thread::sleep_for(std::chrono::milliseconds(100));
consume_work();
memcpy(bDoubleWorkBlob, oWork.bWorkBlob, oWork.iWorkSize);
memcpy(bDoubleWorkBlob + oWork.iWorkSize, oWork.bWorkBlob, oWork.iWorkSize);
piNonce1 = (uint32_t*)(bDoubleWorkBlob + oWork.iWorkSize + 39);
continue;
}
if(oWork.bNiceHash)
iNonce = calc_nicehash_nonce(*piNonce0, oWork.iResumeCnt);
else
iNonce = calc_start_nonce(oWork.iResumeCnt);
assert(sizeof(job_result::sJobID) == sizeof(pool_job::sJobID));
while (iGlobalJobNo.load(std::memory_order_relaxed) == iJobNo)
{
if ((iCount & 0x7) == 0) //Store stats every 16 hashes
{
using namespace std::chrono;
uint64_t iStamp = time_point_cast(high_resolution_clock::now()).time_since_epoch().count();
iHashCount.store(iCount, std::memory_order_relaxed);
iTimestamp.store(iStamp, std::memory_order_relaxed);
}
iCount += 2;
*piNonce0 = ++iNonce;
*piNonce1 = ++iNonce;
hash_fun(bDoubleWorkBlob, oWork.iWorkSize, bDoubleHashOut, ctx0, ctx1);
if (*piHashVal0 < oWork.iTarget)
executor::inst()->push_event(ex_event(job_result(oWork.sJobID, iNonce-1, bDoubleHashOut), oWork.iPoolId));
if (*piHashVal1 < oWork.iTarget)
executor::inst()->push_event(ex_event(job_result(oWork.sJobID, iNonce, bDoubleHashOut + 32), oWork.iPoolId));
std::this_thread::yield();
}
consume_work();
memcpy(bDoubleWorkBlob, oWork.bWorkBlob, oWork.iWorkSize);
memcpy(bDoubleWorkBlob + oWork.iWorkSize, oWork.bWorkBlob, oWork.iWorkSize);
piNonce1 = (uint32_t*)(bDoubleWorkBlob + oWork.iWorkSize + 39);
}
cryptonight_free_ctx(ctx0);
cryptonight_free_ctx(ctx1);
}