Merge branch 'pr1713' into evo

src/backend/cuda/CudaBackend.cpp

@@ -238,7 +238,7 @@ public:
 #           ifdef XMRIG_ALGO_KAWPOW
             if (algo.family() == Algorithm::KAWPOW) {
                 const uint32_t epoch = job.height() / KPHash::EPOCH_LENGTH;
-                mem_used = (KPCache::cache_size(epoch) + KPCache::dag_size(epoch)) / oneMiB;
+                mem_used = (KPCache::dag_size(epoch) + oneMiB - 1) / oneMiB;
             }
 #           endif
 

src/backend/cuda/runners/CudaKawPowRunner.cpp

@@ -66,15 +66,16 @@ bool xmrig::CudaKawPowRunner::set(const Job &job, uint8_t *blob)
 
     const uint64_t start_ms = Chrono::steadyMSecs();
 
-    const bool result = CudaLib::kawPowPrepare(m_ctx, cache.data(), cache.size(), cache.dag_size(epoch), height, dag_sizes);
+    const bool result = CudaLib::kawPowPrepare(m_ctx, cache.data(), cache.size(), cache.l1_cache(), cache.dag_size(epoch), height, dag_sizes);
     if (!result) {
         LOG_ERR("%s " YELLOW("KawPow") RED(" failed to initialize DAG: ") RED_BOLD("%s"), Tags::nvidia(), CudaLib::lastError(m_ctx));
     }
-
+    else {
         const int64_t dt = Chrono::steadyMSecs() - start_ms;
         if (dt > 1000) {
             LOG_INFO("%s " YELLOW("KawPow") " DAG for epoch " WHITE_BOLD("%u") " calculated " BLACK_BOLD("(%" PRIu64 "ms)"), Tags::nvidia(), epoch, dt);
         }
+    }
 
     return result;
 }

src/backend/cuda/wrappers/CudaLib.cpp

@@ -66,7 +66,7 @@ static const char *kRelease                             = "release";
 static const char *kRxHash                              = "rxHash";
 static const char *kRxPrepare                           = "rxPrepare";
 static const char *kKawPowHash                          = "kawPowHash";
-static const char *kKawPowPrepare                       = "kawPowPrepare";
+static const char *kKawPowPrepare_v2                    = "kawPowPrepare_v2";
 static const char *kKawPowStopHash                      = "kawPowStopHash";
 static const char *kSetJob                              = "setJob";
 static const char *kSetJob_v2                           = "setJob_v2";
@@ -92,7 +92,7 @@ using release_t                                         = void (*)(nvid_ctx *);
 using rxHash_t                                          = bool (*)(nvid_ctx *, uint32_t, uint64_t, uint32_t *, uint32_t *);
 using rxPrepare_t                                       = bool (*)(nvid_ctx *, const void *, size_t, bool, uint32_t);
 using kawPowHash_t                                      = bool (*)(nvid_ctx *, uint8_t*, uint64_t, uint32_t *, uint32_t *, uint32_t *);
-using kawPowPrepare_t                                   = bool (*)(nvid_ctx *, const void *, size_t, size_t, uint32_t, const uint64_t*);
+using kawPowPrepare_v2_t                                = bool (*)(nvid_ctx *, const void *, size_t, const void *, size_t, uint32_t, const uint64_t*);
 using kawPowStopHash_t                                  = bool (*)(nvid_ctx *);
 using setJob_t                                          = bool (*)(nvid_ctx *, const void *, size_t, int32_t);
 using setJob_v2_t                                       = bool (*)(nvid_ctx *, const void *, size_t, const char *);
@@ -118,7 +118,7 @@ static release_t pRelease                               = nullptr;
 static rxHash_t pRxHash                                 = nullptr;
 static rxPrepare_t pRxPrepare                           = nullptr;
 static kawPowHash_t pKawPowHash                         = nullptr;
-static kawPowPrepare_t pKawPowPrepare                   = nullptr;
+static kawPowPrepare_v2_t pKawPowPrepare_v2             = nullptr;
 static kawPowStopHash_t pKawPowStopHash                 = nullptr;
 static setJob_t pSetJob                                 = nullptr;
 static setJob_v2_t pSetJob_v2                           = nullptr;
@@ -214,9 +214,9 @@ bool xmrig::CudaLib::kawPowHash(nvid_ctx *ctx, uint8_t* job_blob, uint64_t targe
 }
 
 
-bool xmrig::CudaLib::kawPowPrepare(nvid_ctx *ctx, const void* cache, size_t cache_size, size_t dag_size, uint32_t height, const uint64_t* dag_sizes) noexcept
+bool xmrig::CudaLib::kawPowPrepare(nvid_ctx *ctx, const void* cache, size_t cache_size, const void* dag_precalc, size_t dag_size, uint32_t height, const uint64_t* dag_sizes) noexcept
 {
-    return pKawPowPrepare(ctx, cache, cache_size, dag_size, height, dag_sizes);
+    return pKawPowPrepare_v2(ctx, cache, cache_size, dag_precalc, dag_size, height, dag_sizes);
 }
 
 
@@ -375,7 +375,7 @@ bool xmrig::CudaLib::load()
         DLSYM(AstroBWTHash);
         DLSYM(AstroBWTPrepare);
         DLSYM(KawPowHash);
-        DLSYM(KawPowPrepare);
+        DLSYM(KawPowPrepare_v2);
         DLSYM(KawPowStopHash);
         DLSYM(Version);
 

src/backend/cuda/wrappers/CudaLib.h

@@ -81,7 +81,7 @@ public:
     static bool rxHash(nvid_ctx *ctx, uint32_t startNonce, uint64_t target, uint32_t *rescount, uint32_t *resnonce) noexcept;
     static bool rxPrepare(nvid_ctx *ctx, const void *dataset, size_t datasetSize, bool dataset_host, uint32_t batchSize) noexcept;
     static bool kawPowHash(nvid_ctx *ctx, uint8_t* job_blob, uint64_t target, uint32_t *rescount, uint32_t *resnonce, uint32_t *skipped_hashes) noexcept;
-    static bool kawPowPrepare(nvid_ctx *ctx, const void* cache, size_t cache_size, size_t dag_size, uint32_t height, const uint64_t* dag_sizes) noexcept;
+    static bool kawPowPrepare(nvid_ctx *ctx, const void* cache, size_t cache_size, const void* dag_precalc, size_t dag_size, uint32_t height, const uint64_t* dag_sizes) noexcept;
     static bool kawPowStopHash(nvid_ctx *ctx) noexcept;
     static bool setJob(nvid_ctx *ctx, const void *data, size_t size, const Algorithm &algorithm) noexcept;
     static const char *deviceName(nvid_ctx *ctx) noexcept;

src/crypto/kawpow/KPCache.cpp

@@ -1,14 +1,6 @@
 /* XMRig
- * Copyright 2010      Jeff Garzik <jgarzik@pobox.com>
+ * Copyright 2018-2020 SChernykh   <https://github.com/SChernykh>
- * Copyright 2012-2014 pooler      <pooler@litecoinpool.org>
+ * Copyright 2016-2020 XMRig       <https://github.com/xmrig>, <support@xmrig.com>
- * Copyright 2014      Lucas Jones <https://github.com/lucasjones>
- * Copyright 2014-2016 Wolf9466    <https://github.com/OhGodAPet>
- * Copyright 2016      Jay D Dee   <jayddee246@gmail.com>
- * Copyright 2017-2019 XMR-Stak    <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
- * Copyright 2018      Lee Clagett <https://github.com/vtnerd>
- * Copyright 2018-2019 tevador     <tevador@gmail.com>
- * Copyright 2018-2019 SChernykh   <https://github.com/SChernykh>
- * Copyright 2016-2019 XMRig       <https://github.com/xmrig>, <support@xmrig.com>
  *
  *   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
@@ -26,6 +18,8 @@
 
 
 #include <cinttypes>
+#include <algorithm>
+#include <thread>
 
 #include "crypto/kawpow/KPCache.h"
 #include "3rdparty/libethash/data_sizes.h"
@@ -83,8 +77,30 @@ bool KPCache::init(uint32_t epoch)
     cache.num_parent_nodes = cache.cache_size / sizeof(node);
     calculate_fast_mod_data(cache.num_parent_nodes, cache.reciprocal, cache.increment, cache.shift);
 
-    for (uint32_t i = 0; i < sizeof(m_l1Cache) / sizeof(node); ++i) {
+    const uint64_t cache_nodes = (size + sizeof(node) * 4 - 1) / sizeof(node);
-        ethash_calculate_dag_item_opt(((node*)m_l1Cache) + i, i, num_dataset_parents, &cache);
+    m_DAGCache.resize(cache_nodes * (sizeof(node) / sizeof(uint32_t)));
+
+    // Init DAG cache
+    {
+        const uint64_t n = std::max(std::thread::hardware_concurrency(), 1U);
+
+        std::vector<std::thread> threads;
+        threads.reserve(n);
+
+        for (uint64_t i = 0; i < n; ++i) {
+            const uint32_t a = (cache_nodes * i) / n;
+            const uint32_t b = (cache_nodes * (i + 1)) / n;
+
+            threads.emplace_back([this, a, b, cache_nodes, &cache]() {
+                for (uint32_t j = a; j < b; ++j) {
+                    ethash_calculate_dag_item_opt(((node*)m_DAGCache.data()) + j, j, num_dataset_parents, &cache);
+                }
+            });
+        }
+
+        for (auto& t : threads) {
+            t.join();
+        }
     }
 
     m_size = size;

src/crypto/kawpow/KPCache.h

@@ -1,14 +1,6 @@
 /* XMRig
- * Copyright 2010      Jeff Garzik <jgarzik@pobox.com>
- * Copyright 2012-2014 pooler      <pooler@litecoinpool.org>
- * Copyright 2014      Lucas Jones <https://github.com/lucasjones>
- * Copyright 2014-2016 Wolf9466    <https://github.com/OhGodAPet>
- * Copyright 2016      Jay D Dee   <jayddee246@gmail.com>
- * Copyright 2017-2019 XMR-Stak    <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
- * Copyright 2018      Lee Clagett <https://github.com/vtnerd>
- * Copyright 2018-2019 tevador     <tevador@gmail.com>
  * Copyright 2018-2020 SChernykh   <https://github.com/SChernykh>
- * Copyright 2016-2019 XMRig       <https://github.com/xmrig>, <support@xmrig.com>
+ * Copyright 2016-2020 XMRig       <https://github.com/xmrig>, <support@xmrig.com>
  *
  *   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
@@ -30,6 +22,7 @@
 
 #include "base/tools/Object.h"
 #include <mutex>
+#include <vector>
 
 
 namespace xmrig
@@ -57,7 +50,7 @@ public:
     size_t size() const { return m_size; }
     uint32_t epoch() const { return m_epoch; }
 
-    const uint32_t* l1_cache() const { return m_l1Cache; }
+    const uint32_t* l1_cache() const { return m_DAGCache.data(); }
 
     static uint64_t cache_size(uint32_t epoch);
     static uint64_t dag_size(uint32_t epoch);
@@ -71,7 +64,7 @@ private:
     VirtualMemory* m_memory = nullptr;
     size_t m_size = 0;
     uint32_t m_epoch = 0xFFFFFFFFUL;
-    uint32_t m_l1Cache[l1_cache_num_items] = {};
+    std::vector<uint32_t> m_DAGCache;
 };