diff options
Diffstat (limited to 'xmrstak/backend/cpu')
| -rw-r--r-- | xmrstak/backend/cpu/autoAdjust.hpp | 33 | ||||
| -rw-r--r-- | xmrstak/backend/cpu/config.tpl | 3 | ||||
| -rw-r--r-- | xmrstak/backend/cpu/crypto/cryptonight_aesni.h | 13 | ||||
| -rw-r--r-- | xmrstak/backend/cpu/minethd.cpp | 13 |
4 files changed, 35 insertions, 27 deletions
diff --git a/xmrstak/backend/cpu/autoAdjust.hpp b/xmrstak/backend/cpu/autoAdjust.hpp index 7bdb14e..db805ec 100644 --- a/xmrstak/backend/cpu/autoAdjust.hpp +++ b/xmrstak/backend/cpu/autoAdjust.hpp @@ -33,25 +33,21 @@ class autoAdjust { public: - size_t hashMemSize; - size_t halfHashMemSize; - - autoAdjust() + bool printConfig() { + size_t hashMemSizeKB; + size_t halfHashMemSizeKB; + if(::jconf::inst()->IsCurrencyMonero()) { - hashMemSize = MONERO_MEMORY; - halfHashMemSize = hashMemSize / 2u; + hashMemSizeKB = MONERO_MEMORY / 1024u; + halfHashMemSizeKB = hashMemSizeKB / 2u; } else { - hashMemSize = AEON_MEMORY; - halfHashMemSize = hashMemSize / 2u; + hashMemSizeKB = AEON_MEMORY / 1024u; + halfHashMemSizeKB = hashMemSizeKB / 2u; } - } - - bool printConfig() - { configEditor configTpl{}; @@ -63,9 +59,10 @@ public: std::string conf; - if(!detectL3Size() || L3KB_size < halfHashMemSize || L3KB_size > (halfHashMemSize * 100u)) + + if(!detectL3Size() || L3KB_size < halfHashMemSizeKB || L3KB_size > (halfHashMemSizeKB * 2048u)) { - if(L3KB_size < halfHashMemSize || L3KB_size > (halfHashMemSize * 100)) + if(L3KB_size < halfHashMemSizeKB || L3KB_size > (halfHashMemSizeKB * 2048)) printer::inst()->print_msg(L0, "Autoconf failed: L3 size sanity check failed - %u KB.", L3KB_size); conf += std::string(" { \"low_power_mode\" : false, \"no_prefetch\" : true, \"affine_to_cpu\" : false },\n"); @@ -88,7 +85,7 @@ public: if(L3KB_size <= 0) break; - double_mode = L3KB_size / hashMemSize > (int32_t)(corecnt-i); + double_mode = L3KB_size / hashMemSizeKB > (int32_t)(corecnt-i); conf += std::string(" { \"low_power_mode\" : "); conf += std::string(double_mode ? "true" : "false"); @@ -107,9 +104,9 @@ public: aff_id++; if(double_mode) - L3KB_size -= hashMemSize * 2u; + L3KB_size -= hashMemSizeKB * 2u; else - L3KB_size -= hashMemSize; + L3KB_size -= hashMemSizeKB; } } @@ -142,7 +139,7 @@ private: } L3KB_size = ((get_masked(cpu_info[1], 31, 22) + 1) * (get_masked(cpu_info[1], 21, 12) + 1) * - (get_masked(cpu_info[1], 11, 0) + 1) * (cpu_info[2] + 1)) / halfHashMemSize; + (get_masked(cpu_info[1], 11, 0) + 1) * (cpu_info[2] + 1)) / 1024; return true; } diff --git a/xmrstak/backend/cpu/config.tpl b/xmrstak/backend/cpu/config.tpl index b21a22d..fc4acb9 100644 --- a/xmrstak/backend/cpu/config.tpl +++ b/xmrstak/backend/cpu/config.tpl @@ -24,6 +24,9 @@ R"===( * { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 0 }, * { "low_power_mode" : false, "no_prefetch" : true, "affine_to_cpu" : 1 }, * ], + * If you do not wish to mine with your CPU(s) then use: + * "cpu_threads_conf" : + * null, */ "cpu_threads_conf" : diff --git a/xmrstak/backend/cpu/crypto/cryptonight_aesni.h b/xmrstak/backend/cpu/crypto/cryptonight_aesni.h index 9b6e1dc..e4ccbc3 100644 --- a/xmrstak/backend/cpu/crypto/cryptonight_aesni.h +++ b/xmrstak/backend/cpu/crypto/cryptonight_aesni.h @@ -317,10 +317,9 @@ void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_c _mm_store_si128((__m128i *)&l0[idx0 & MASK], _mm_xor_si128(bx0, cx)); idx0 = _mm_cvtsi128_si64(cx); - bx0 = cx; - if(PREFETCH) _mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0); + bx0 = cx; uint64_t hi, lo, cl, ch; cl = ((uint64_t*)&l0[idx0 & MASK])[0]; @@ -329,15 +328,15 @@ void cryptonight_hash(const void* input, size_t len, void* output, cryptonight_c lo = _umul128(idx0, cl, &hi); al0 += hi; - ah0 += lo; ((uint64_t*)&l0[idx0 & MASK])[0] = al0; + al0 ^= cl; + if(PREFETCH) + _mm_prefetch((const char*)&l0[al0 & MASK], _MM_HINT_T0); + ah0 += lo; ((uint64_t*)&l0[idx0 & MASK])[1] = ah0; ah0 ^= ch; - al0 ^= cl; - idx0 = al0; - if(PREFETCH) - _mm_prefetch((const char*)&l0[idx0 & MASK], _MM_HINT_T0); + idx0 = al0; } // Optim - 90% time boundary diff --git a/xmrstak/backend/cpu/minethd.cpp b/xmrstak/backend/cpu/minethd.cpp index b41e4a9..1d9165e 100644 --- a/xmrstak/backend/cpu/minethd.cpp +++ b/xmrstak/backend/cpu/minethd.cpp @@ -73,7 +73,16 @@ namespace cpu bool minethd::thd_setaffinity(std::thread::native_handle_type h, uint64_t cpu_id) { #if defined(_WIN32) - return SetThreadAffinityMask(h, 1ULL << cpu_id) != 0; + // we can only pin up to 64 threads + if(cpu_id < 64) + { + return SetThreadAffinityMask(h, 1ULL << cpu_id) != 0; + } + else + { + printer::inst()->print_msg(L0, "WARNING: Windows supports only affinity up to 63."); + return false; + } #elif defined(__APPLE__) thread_port_t mach_thread; thread_affinity_policy_data_t policy = { static_cast<integer_t>(cpu_id) }; @@ -309,7 +318,7 @@ std::vector<iBackend*> minethd::thread_starter(uint32_t threadOffset, miner_work if(cfg.iCpuAff >= 0) { #if defined(__APPLE__) - printer::inst()->print_msg(L1, "WARNING on MacOS thread affinity is only advisory."); + printer::inst()->print_msg(L1, "WARNING on macOS thread affinity is only advisory."); #endif printer::inst()->print_msg(L1, "Starting %dx thread, affinity: %d.", cfg.iMultiway, (int)cfg.iCpuAff); |
