Special case for Intel CPUs with 1 MB L2 cache per core.

This commit is contained in:
XMRig 2019-07-26 16:22:44 +07:00
parent ffa6bda106
commit 35d9c755e0
2 changed files with 33 additions and 14 deletions

View file

@ -189,28 +189,47 @@ xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm) const
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) const
{ {
constexpr size_t oneMiB = 1024u * 1024u;
size_t PUs = countByType(cache, HWLOC_OBJ_PU); size_t PUs = countByType(cache, HWLOC_OBJ_PU);
if (PUs == 0) { if (PUs == 0) {
return; return;
} }
size_t size = cache->attr->cache.size;
const size_t scratchpad = algorithm.memory();
if (cache->attr->cache.depth == 3 && isCacheExclusive(cache)) {
for (size_t i = 0; i < cache->arity; ++i) {
hwloc_obj_t l2 = cache->children[i];
if (isCacheObject(l2) && l2->attr != nullptr && l2->attr->cache.size >= scratchpad) {
size += scratchpad;
}
}
}
std::vector<hwloc_obj_t> cores; std::vector<hwloc_obj_t> cores;
cores.reserve(m_cores); cores.reserve(m_cores);
findByType(cache, HWLOC_OBJ_CORE, [&cores](hwloc_obj_t found) { cores.emplace_back(found); }); findByType(cache, HWLOC_OBJ_CORE, [&cores](hwloc_obj_t found) { cores.emplace_back(found); });
size_t cacheHashes = (size + (scratchpad / 2)) / scratchpad; size_t L3 = cache->attr->cache.size;
size_t L2 = 0;
int L2_associativity = 0;
size_t extra = 0;
const size_t scratchpad = algorithm.memory();
if (cache->attr->cache.depth == 3 && isCacheExclusive(cache)) {
for (size_t i = 0; i < cache->arity; ++i) {
hwloc_obj_t l2 = cache->children[i];
if (!isCacheObject(l2) || l2->attr == nullptr) {
continue;
}
L2 += l2->attr->cache.size;
L2_associativity = l2->attr->cache.associativity;
if (l2->attr->cache.size >= scratchpad) {
extra += scratchpad;
}
}
}
if (scratchpad == 2 * oneMiB) {
if (L2 && (cores.size() * oneMiB) == L2 && L2_associativity == 16 && L3 >= L2) {
L3 = L2;
extra = L2;
}
}
size_t cacheHashes = ((L3 + extra) + (scratchpad / 2)) / scratchpad;
# ifdef XMRIG_ALGO_CN_GPU # ifdef XMRIG_ALGO_CN_GPU
if (algorithm == Algorithm::CN_GPU) { if (algorithm == Algorithm::CN_GPU) {

View file

@ -147,7 +147,7 @@ void xmrig::Rx::initDataset(size_t index, const uint8_t *seed, const Algorithm &
if (!d_ptr->datasets[index]) { if (!d_ptr->datasets[index]) {
const uint64_t ts = Chrono::steadyMSecs(); const uint64_t ts = Chrono::steadyMSecs();
LOG_INFO("%s" MAGENTA_BOLD(" allocate") CYAN_BOLD(" %zu MiB") BLACK_BOLD(" (%zu+%zu) for RandomX dataset & cache"), LOG_INFO("%s" MAGENTA_BOLD(" allocate") CYAN_BOLD(" %zu MB") BLACK_BOLD(" (%zu+%zu) for RandomX dataset & cache"),
tag, tag,
(RxDataset::size() + RxCache::size()) / 1024 / 1024, (RxDataset::size() + RxCache::size()) / 1024 / 1024,
RxDataset::size() / 1024 / 1024, RxDataset::size() / 1024 / 1024,