Optimized RandomX initialization and switching.
This commit is contained in:
XMRig
parent
ab0d3b8919
commit
84ff8af4bd
9 changed files with 156 additions and 157 deletions
|
|
@ -43,6 +43,7 @@
|
|||
#include "base/tools/Buffer.h"
|
||||
#include "base/tools/Chrono.h"
|
||||
#include "crypto/rx/Rx.h"
|
||||
#include "crypto/rx/RxAlgo.h"
|
||||
#include "crypto/rx/RxCache.h"
|
||||
#include "crypto/rx/RxDataset.h"
|
||||
|
||||
|
|
@ -57,11 +58,50 @@ static const char *tag = BLUE_BG(WHITE_BOLD_S " rx ") " ";
|
|||
static RxPrivate *d_ptr = nullptr;
|
||||
|
||||
|
||||
#ifdef XMRIG_FEATURE_HWLOC
|
||||
static void bindToNUMANode(uint32_t nodeId)
|
||||
{
|
||||
hwloc_topology_t topology;
|
||||
hwloc_topology_init(&topology);
|
||||
hwloc_topology_load(topology);
|
||||
|
||||
hwloc_obj_t node = hwloc_get_numanode_obj_by_os_index(topology, nodeId);
|
||||
if (node) {
|
||||
if (HwlocCpuInfo::has(HwlocCpuInfo::SET_THISTHREAD_MEMBIND)) {
|
||||
# if HWLOC_API_VERSION >= 0x20000
|
||||
hwloc_set_membind(topology, node->nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD | HWLOC_MEMBIND_BYNODESET);
|
||||
# else
|
||||
hwloc_set_membind_nodeset(topology, node->nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD);
|
||||
# endif
|
||||
}
|
||||
|
||||
Platform::setThreadAffinity(static_cast<uint64_t>(hwloc_bitmap_first(node->cpuset)));
|
||||
}
|
||||
|
||||
hwloc_topology_destroy(topology);
|
||||
}
|
||||
#else
|
||||
inline static void bindToNUMANode(uint32_t) {}
|
||||
#endif
|
||||
|
||||
|
||||
class RxPrivate
|
||||
{
|
||||
public:
|
||||
inline RxPrivate()
|
||||
inline RxPrivate() :
|
||||
m_seed()
|
||||
{
|
||||
# ifdef XMRIG_FEATURE_HWLOC
|
||||
if (Cpu::info()->nodes() > 1) {
|
||||
for (uint32_t nodeId : HwlocCpuInfo::nodeIndexes()) {
|
||||
datasets.insert({ nodeId, nullptr });
|
||||
}
|
||||
}
|
||||
else
|
||||
# endif
|
||||
{
|
||||
datasets.insert({ 0, nullptr });
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -75,32 +115,19 @@ public:
|
|||
}
|
||||
|
||||
|
||||
inline bool isNUMA() const { return m_numa; }
|
||||
inline const Algorithm &algorithm() const { return m_algorithm; }
|
||||
inline const uint8_t *seed() const { return m_seed; }
|
||||
inline size_t count() const { return isNUMA() ? datasets.size() : 1; }
|
||||
|
||||
|
||||
static void allocate(uint32_t nodeId)
|
||||
{
|
||||
const uint64_t ts = Chrono::steadyMSecs();
|
||||
|
||||
# ifdef XMRIG_FEATURE_HWLOC
|
||||
if (d_ptr->numa) {
|
||||
hwloc_topology_t topology;
|
||||
hwloc_topology_init(&topology);
|
||||
hwloc_topology_load(topology);
|
||||
|
||||
hwloc_obj_t node = hwloc_get_numanode_obj_by_os_index(topology, nodeId);
|
||||
if (node) {
|
||||
if (HwlocCpuInfo::has(HwlocCpuInfo::SET_THISTHREAD_MEMBIND)) {
|
||||
# if HWLOC_API_VERSION >= 0x20000
|
||||
hwloc_set_membind(topology, node->nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD | HWLOC_MEMBIND_BYNODESET);
|
||||
# else
|
||||
hwloc_set_membind_nodeset(topology, node->nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD);
|
||||
# endif
|
||||
}
|
||||
|
||||
Platform::setThreadAffinity(static_cast<uint64_t>(hwloc_bitmap_first(node->cpuset)));
|
||||
}
|
||||
|
||||
hwloc_topology_destroy(topology);
|
||||
if (d_ptr->isNUMA()) {
|
||||
bindToNUMANode(nodeId);
|
||||
}
|
||||
# endif
|
||||
|
||||
LOG_INFO("%s" CYAN_BOLD("#%u") MAGENTA_BOLD(" allocate") CYAN_BOLD(" %zu MB") BLACK_BOLD(" (%zu+%zu) for RandomX dataset & cache"),
|
||||
tag,
|
||||
|
|
@ -110,7 +137,7 @@ public:
|
|||
RxCache::size() / 1024 / 1024
|
||||
);
|
||||
|
||||
RxDataset *dataset = new RxDataset(d_ptr->hugePages);
|
||||
RxDataset *dataset = new RxDataset(d_ptr->m_hugePages);
|
||||
d_ptr->datasets[nodeId] = dataset;
|
||||
|
||||
if (dataset->get() != nullptr) {
|
||||
|
|
@ -134,29 +161,84 @@ public:
|
|||
}
|
||||
|
||||
|
||||
bool hugePages = true;
|
||||
bool numa = true;
|
||||
static void initDataset(uint32_t nodeId, uint32_t threads)
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(d_ptr->mutex);
|
||||
|
||||
const uint64_t ts = Chrono::steadyMSecs();
|
||||
|
||||
d_ptr->getOrAllocate(nodeId)->init(d_ptr->seed(), threads);
|
||||
d_ptr->m_ready++;
|
||||
|
||||
LOG_INFO("%s" CYAN_BOLD("#%u") GREEN(" init done") BLACK_BOLD(" (%" PRIu64 " ms)"), tag, nodeId, Chrono::steadyMSecs() - ts);
|
||||
}
|
||||
|
||||
|
||||
inline RxDataset *getOrAllocate(uint32_t nodeId)
|
||||
{
|
||||
RxDataset *dataset = datasets.at(nodeId);
|
||||
|
||||
if (dataset == nullptr) {
|
||||
# ifdef XMRIG_FEATURE_HWLOC
|
||||
if (d_ptr->isNUMA()) {
|
||||
std::thread thread(allocate, nodeId);
|
||||
thread.join();
|
||||
} else
|
||||
# endif
|
||||
{
|
||||
allocate(nodeId);
|
||||
}
|
||||
|
||||
dataset = datasets.at(nodeId);
|
||||
}
|
||||
|
||||
return dataset;
|
||||
}
|
||||
|
||||
|
||||
inline void setState(const Job &job, bool hugePages, bool numa)
|
||||
{
|
||||
if (m_algorithm != job.algorithm()) {
|
||||
m_algorithm = RxAlgo::apply(job.algorithm());
|
||||
}
|
||||
|
||||
m_ready = 0;
|
||||
m_numa = numa && Cpu::info()->nodes() > 1;
|
||||
m_hugePages = hugePages;
|
||||
|
||||
memcpy(m_seed, job.seedHash(), sizeof(m_seed));
|
||||
}
|
||||
|
||||
|
||||
inline bool isReady(const Job &job)
|
||||
{
|
||||
return m_ready == count() && m_algorithm == job.algorithm() && memcmp(m_seed, job.seedHash(), sizeof(m_seed)) == 0;
|
||||
}
|
||||
|
||||
|
||||
std::map<uint32_t, RxDataset *> datasets;
|
||||
std::mutex mutex;
|
||||
|
||||
private:
|
||||
bool m_hugePages = true;
|
||||
bool m_numa = true;
|
||||
Algorithm m_algorithm;
|
||||
size_t m_ready = 0;
|
||||
uint8_t m_seed[32];
|
||||
};
|
||||
|
||||
|
||||
} // namespace xmrig
|
||||
|
||||
|
||||
bool xmrig::Rx::isReady(const Job &job, uint32_t nodeId)
|
||||
xmrig::RxDataset *xmrig::Rx::dataset(const Job &job, uint32_t nodeId)
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(d_ptr->mutex);
|
||||
if (!d_ptr->isReady(job)) {
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
return isReady(job.seedHash(), job.algorithm(), d_ptr->numa ? nodeId : 0);
|
||||
}
|
||||
|
||||
|
||||
xmrig::RxDataset *xmrig::Rx::dataset(uint32_t nodeId)
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(d_ptr->mutex);
|
||||
|
||||
return d_ptr->datasets[d_ptr->numa ? nodeId : 0];
|
||||
return d_ptr->datasets.at(d_ptr->isNUMA() ? nodeId : 0);
|
||||
}
|
||||
|
||||
|
||||
|
|
@ -201,88 +283,33 @@ void xmrig::Rx::init(const Job &job, int initThreads, bool hugePages, bool numa)
|
|||
|
||||
std::lock_guard<std::mutex> lock(d_ptr->mutex);
|
||||
|
||||
size_t ready = 0;
|
||||
|
||||
for (auto const &item : d_ptr->datasets) {
|
||||
if (isReady(job.seedHash(), job.algorithm(), item.first)) {
|
||||
ready++;
|
||||
}
|
||||
}
|
||||
|
||||
if (!d_ptr->datasets.empty() && ready == d_ptr->datasets.size()) {
|
||||
if (d_ptr->isReady(job)) {
|
||||
return;
|
||||
}
|
||||
|
||||
d_ptr->hugePages = hugePages;
|
||||
d_ptr->numa = numa && Cpu::info()->nodes() > 1;
|
||||
const uint32_t threads = initThreads < 1 ? static_cast<uint32_t>(Cpu::info()->threads())
|
||||
: static_cast<uint32_t>(initThreads);
|
||||
d_ptr->setState(job, hugePages, numa);
|
||||
const uint32_t threads = initThreads < 1 ? static_cast<uint32_t>(Cpu::info()->threads()) : static_cast<uint32_t>(initThreads);
|
||||
const String buf = Buffer::toHex(job.seedHash(), 8);
|
||||
|
||||
LOG_INFO("%s" MAGENTA_BOLD("init dataset%s") " algo " WHITE_BOLD("%s (") CYAN_BOLD("%u") WHITE_BOLD(" threads)") BLACK_BOLD(" seed %s..."),
|
||||
tag,
|
||||
d_ptr->count() > 1 ? "s" : "",
|
||||
job.algorithm().shortName(),
|
||||
threads,
|
||||
buf.data()
|
||||
);
|
||||
|
||||
# ifdef XMRIG_FEATURE_HWLOC
|
||||
if (d_ptr->numa) {
|
||||
for (uint32_t nodeId : HwlocCpuInfo::nodeIndexes()) {
|
||||
std::thread thread(initDataset, nodeId, job.seedHash(), job.algorithm(), threads);
|
||||
if (d_ptr->isNUMA()) {
|
||||
for (auto const &item : d_ptr->datasets) {
|
||||
std::thread thread(RxPrivate::initDataset, item.first, threads);
|
||||
thread.detach();
|
||||
}
|
||||
}
|
||||
else
|
||||
# endif
|
||||
{
|
||||
std::thread thread(initDataset, 0, job.seedHash(), job.algorithm(), threads);
|
||||
std::thread thread(RxPrivate::initDataset, 0, threads);
|
||||
thread.detach();
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
bool xmrig::Rx::isReady(const uint8_t *seed, const Algorithm &algorithm, uint32_t nodeId)
|
||||
{
|
||||
return !d_ptr->datasets.empty() && d_ptr->datasets[nodeId] != nullptr && d_ptr->datasets[nodeId]->isReady(seed, algorithm);
|
||||
}
|
||||
|
||||
|
||||
void xmrig::Rx::initDataset(uint32_t nodeId, const uint8_t *seed, const Algorithm &algorithm, uint32_t threads)
|
||||
{
|
||||
std::lock_guard<std::mutex> lock(d_ptr->mutex);
|
||||
|
||||
RxDataset *dataset = d_ptr->datasets[nodeId];
|
||||
|
||||
if (!dataset) {
|
||||
# ifdef XMRIG_FEATURE_HWLOC
|
||||
if (d_ptr->numa) {
|
||||
std::thread thread(RxPrivate::allocate, nodeId);
|
||||
thread.join();
|
||||
} else
|
||||
# endif
|
||||
{
|
||||
RxPrivate::allocate(nodeId);
|
||||
}
|
||||
|
||||
dataset = d_ptr->datasets[nodeId];
|
||||
}
|
||||
|
||||
if (!dataset->isReady(seed, algorithm)) {
|
||||
const uint64_t ts = Chrono::steadyMSecs();
|
||||
|
||||
if (dataset->get() != nullptr) {
|
||||
LOG_INFO("%s" CYAN_BOLD("#%u") MAGENTA_BOLD(" init dataset") " algo " WHITE_BOLD("%s (") CYAN_BOLD("%u") WHITE_BOLD(" threads)") BLACK_BOLD(" seed %s..."),
|
||||
tag,
|
||||
nodeId,
|
||||
algorithm.shortName(),
|
||||
threads,
|
||||
Buffer::toHex(seed, 8).data()
|
||||
);
|
||||
}
|
||||
else {
|
||||
LOG_INFO("%s" CYAN_BOLD("#%u") MAGENTA_BOLD(" init cache") " algo " WHITE_BOLD("%s") BLACK_BOLD(" seed %s..."),
|
||||
tag,
|
||||
nodeId,
|
||||
algorithm.shortName(),
|
||||
Buffer::toHex(seed, 8).data()
|
||||
);
|
||||
}
|
||||
|
||||
dataset->init(seed, algorithm, threads);
|
||||
|
||||
LOG_INFO("%s" CYAN_BOLD("#%u") GREEN(" init done") BLACK_BOLD(" (%" PRIu64 " ms)"), tag, nodeId, Chrono::steadyMSecs() - ts);
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue