Added config option "cpu/max-threads-hint" and command line option "--cpu-max-threads-hint".
This commit is contained in:
parent
daed23422e
commit
7c463849cc
14 changed files with 97 additions and 57 deletions
|
@ -23,10 +23,10 @@
|
|||
*/
|
||||
|
||||
#include <algorithm>
|
||||
#include <assert.h>
|
||||
#include <math.h>
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <cassert>
|
||||
#include <cmath>
|
||||
#include <cstdio>
|
||||
#include <cstring>
|
||||
|
||||
|
||||
#include "3rdparty/libcpuid/libcpuid.h"
|
||||
|
@ -109,7 +109,7 @@ xmrig::AdvancedCpuInfo::AdvancedCpuInfo() :
|
|||
}
|
||||
|
||||
|
||||
xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm) const
|
||||
xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
|
||||
{
|
||||
if (threads() == 1) {
|
||||
return 1;
|
||||
|
@ -153,5 +153,12 @@ xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm) co
|
|||
}
|
||||
# endif
|
||||
|
||||
return CpuThreads(std::max<size_t>(std::min<size_t>(count, threads()), 1), intensity);
|
||||
if (limit > 0 && limit < 100) {
|
||||
count = std::min(count, static_cast<size_t>(round(threads() * (limit / 100.0))));
|
||||
}
|
||||
else {
|
||||
count = std::min(count, threads());
|
||||
}
|
||||
|
||||
return CpuThreads(std::max<size_t>(count, 1), intensity);
|
||||
}
|
||||
|
|
|
@ -38,7 +38,7 @@ public:
|
|||
AdvancedCpuInfo();
|
||||
|
||||
protected:
|
||||
CpuThreads threads(const Algorithm &algorithm) const override;
|
||||
CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
|
||||
|
||||
inline Assembly::Id assembly() const override { return m_assembly; }
|
||||
inline bool hasAES() const override { return m_aes; }
|
||||
|
|
|
@ -179,7 +179,7 @@ const char *xmrig::BasicCpuInfo::backend() const
|
|||
}
|
||||
|
||||
|
||||
xmrig::CpuThreads xmrig::BasicCpuInfo::threads(const Algorithm &algorithm) const
|
||||
xmrig::CpuThreads xmrig::BasicCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
|
||||
{
|
||||
const size_t count = std::thread::hardware_concurrency();
|
||||
|
||||
|
|
|
@ -39,7 +39,7 @@ public:
|
|||
|
||||
protected:
|
||||
const char *backend() const override;
|
||||
CpuThreads threads(const Algorithm &algorithm) const override;
|
||||
CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
|
||||
|
||||
inline Assembly::Id assembly() const override { return m_assembly; }
|
||||
inline bool hasAES() const override { return m_aes; }
|
||||
|
|
|
@ -29,6 +29,7 @@
|
|||
|
||||
|
||||
#include <algorithm>
|
||||
#include <cmath>
|
||||
#include <hwloc.h>
|
||||
|
||||
|
||||
|
@ -127,9 +128,7 @@ static inline bool isCacheExclusive(hwloc_obj_t obj)
|
|||
} // namespace xmrig
|
||||
|
||||
|
||||
xmrig::HwlocCpuInfo::HwlocCpuInfo() : BasicCpuInfo(),
|
||||
m_backend(),
|
||||
m_cache()
|
||||
xmrig::HwlocCpuInfo::HwlocCpuInfo()
|
||||
{
|
||||
m_threads = 0;
|
||||
|
||||
|
@ -149,7 +148,7 @@ xmrig::HwlocCpuInfo::HwlocCpuInfo() : BasicCpuInfo(),
|
|||
# endif
|
||||
|
||||
const std::vector<hwloc_obj_t> packages = findByType(hwloc_get_root_obj(m_topology), HWLOC_OBJ_PACKAGE);
|
||||
if (packages.size()) {
|
||||
if (!packages.empty()) {
|
||||
const char *value = hwloc_obj_get_info_by_name(packages[0], "CPUModel");
|
||||
if (value) {
|
||||
strncpy(m_brand, value, 64);
|
||||
|
@ -202,10 +201,10 @@ xmrig::HwlocCpuInfo::~HwlocCpuInfo()
|
|||
}
|
||||
|
||||
|
||||
xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm) const
|
||||
xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
|
||||
{
|
||||
if (L2() == 0 && L3() == 0) {
|
||||
return BasicCpuInfo::threads(algorithm);
|
||||
return BasicCpuInfo::threads(algorithm, limit);
|
||||
}
|
||||
|
||||
const unsigned depth = L3() > 0 ? 3 : 2;
|
||||
|
@ -218,21 +217,37 @@ xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm) const
|
|||
|
||||
findCache(hwloc_get_root_obj(m_topology), depth, depth, [&caches](hwloc_obj_t found) { caches.emplace_back(found); });
|
||||
|
||||
for (hwloc_obj_t cache : caches) {
|
||||
processTopLevelCache(cache, algorithm, threads);
|
||||
if (limit > 0 && limit < 100 && !caches.empty()) {
|
||||
const double maxTotalThreads = round(m_threads * (limit / 100.0));
|
||||
const auto maxPerCache = std::max(static_cast<int>(round(maxTotalThreads / caches.size())), 1);
|
||||
int remaining = std::max(static_cast<int>(maxTotalThreads), 1);
|
||||
|
||||
for (hwloc_obj_t cache : caches) {
|
||||
processTopLevelCache(cache, algorithm, threads, std::min(maxPerCache, remaining));
|
||||
|
||||
remaining -= maxPerCache;
|
||||
if (remaining <= 0) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
for (hwloc_obj_t cache : caches) {
|
||||
processTopLevelCache(cache, algorithm, threads, 0);
|
||||
}
|
||||
}
|
||||
|
||||
if (threads.isEmpty()) {
|
||||
LOG_WARN("hwloc auto configuration for algorithm \"%s\" failed.", algorithm.shortName());
|
||||
|
||||
return BasicCpuInfo::threads(algorithm);
|
||||
return BasicCpuInfo::threads(algorithm, limit);
|
||||
}
|
||||
|
||||
return threads;
|
||||
}
|
||||
|
||||
|
||||
void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorithm &algorithm, CpuThreads &threads) const
|
||||
void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorithm &algorithm, CpuThreads &threads, size_t limit) const
|
||||
{
|
||||
constexpr size_t oneMiB = 1024u * 1024u;
|
||||
|
||||
|
@ -296,6 +311,10 @@ void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorith
|
|||
}
|
||||
# endif
|
||||
|
||||
if (limit > 0) {
|
||||
cacheHashes = std::min(cacheHashes, limit);
|
||||
}
|
||||
|
||||
if (cacheHashes >= PUs) {
|
||||
for (hwloc_obj_t core : cores) {
|
||||
const std::vector<hwloc_obj_t> units = findByType(core, HWLOC_OBJ_PU);
|
||||
|
|
|
@ -27,10 +27,11 @@
|
|||
|
||||
|
||||
#include "backend/cpu/platform/BasicCpuInfo.h"
|
||||
#include "base/tools/Object.h"
|
||||
|
||||
|
||||
typedef struct hwloc_obj *hwloc_obj_t;
|
||||
typedef struct hwloc_topology *hwloc_topology_t;
|
||||
using hwloc_obj_t = struct hwloc_obj *;
|
||||
using hwloc_topology_t = struct hwloc_topology *;
|
||||
|
||||
|
||||
namespace xmrig {
|
||||
|
@ -39,6 +40,9 @@ namespace xmrig {
|
|||
class HwlocCpuInfo : public BasicCpuInfo
|
||||
{
|
||||
public:
|
||||
XMRIG_DISABLE_COPY_MOVE(HwlocCpuInfo)
|
||||
|
||||
|
||||
enum Feature : uint32_t {
|
||||
SET_THISTHREAD_MEMBIND = 1
|
||||
};
|
||||
|
@ -51,7 +55,7 @@ public:
|
|||
static inline const std::vector<uint32_t> &nodeIndexes() { return m_nodeIndexes; }
|
||||
|
||||
protected:
|
||||
CpuThreads threads(const Algorithm &algorithm) const override;
|
||||
CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
|
||||
|
||||
inline const char *backend() const override { return m_backend; }
|
||||
inline size_t cores() const override { return m_cores; }
|
||||
|
@ -61,17 +65,17 @@ protected:
|
|||
inline size_t packages() const override { return m_packages; }
|
||||
|
||||
private:
|
||||
void processTopLevelCache(hwloc_obj_t obj, const Algorithm &algorithm, CpuThreads &threads) const;
|
||||
void processTopLevelCache(hwloc_obj_t obj, const Algorithm &algorithm, CpuThreads &threads, size_t limit) const;
|
||||
|
||||
static std::vector<uint32_t> m_nodeIndexes;
|
||||
static uint32_t m_features;
|
||||
|
||||
char m_backend[20];
|
||||
hwloc_topology_t m_topology;
|
||||
size_t m_cache[5];
|
||||
size_t m_cores = 0;
|
||||
size_t m_nodes = 0;
|
||||
size_t m_packages = 0;
|
||||
char m_backend[20] = { 0 };
|
||||
hwloc_topology_t m_topology = nullptr;
|
||||
size_t m_cache[5] = { 0 };
|
||||
size_t m_cores = 0;
|
||||
size_t m_nodes = 0;
|
||||
size_t m_packages = 0;
|
||||
};
|
||||
|
||||
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue