RandomX dataset specific code moved into OclSharedData class.

src/backend/opencl/OclBackend.cpp

@@ -165,7 +165,7 @@ public:
     }
 
 
-    inline void start()
+    inline void start(const Job &job)
     {
         LOG_INFO("%s use profile " BLUE_BG(WHITE_BOLD_S " %s ") WHITE_BOLD_S " (" CYAN_BOLD("%zu") WHITE_BOLD(" threads)") " scratchpad " CYAN_BOLD("%zu KB"),
                  tag,
@@ -195,7 +195,7 @@ public:
                     i++;
         }
 
-        OclSharedState::start(threads);
+        OclSharedState::start(threads, job);
 
         status.start(threads.size());
         workers.start(threads);
@@ -332,7 +332,7 @@ void xmrig::OclBackend::setJob(const Job &job)
         return stop();
     }
 
-    if (!d_ptr->context.init(d_ptr->devices, threads, job)) {
+    if (!d_ptr->context.init(d_ptr->devices, threads)) {
         LOG_WARN("%s " RED_BOLD("disabled") YELLOW(" (OpenCL context unavailable)"), tag);
 
         return stop();
@@ -341,7 +341,7 @@ void xmrig::OclBackend::setJob(const Job &job)
     stop();
 
     d_ptr->threads = std::move(threads);
-    d_ptr->start();
+    d_ptr->start(job);
 }
 
 

src/backend/opencl/OclConfig.cpp

@@ -193,29 +193,13 @@ std::vector<xmrig::OclLaunchData> xmrig::OclConfig::get(const Miner *miner, cons
             continue;
         }
 
-#       ifdef XMRIG_ALGO_RANDOMX
-        auto dataset = algorithm.family() == Algorithm::RANDOM_X ? std::make_shared<OclRxDataset>() : nullptr;
-#       endif
-
         if (thread.threads().size() > 1) {
             for (int64_t affinity : thread.threads()) {
-                OclLaunchData data(miner, algorithm, *this, platform, thread, devices[thread.index()], affinity);
+                out.emplace_back(miner, algorithm, *this, platform, thread, devices[thread.index()], affinity);
-
-#               ifdef XMRIG_ALGO_RANDOMX
-                data.dataset = dataset;
-#               endif
-
-                out.emplace_back(std::move(data));
             }
         }
         else {
-            OclLaunchData data(miner, algorithm, *this, platform, thread, devices[thread.index()], thread.threads().front());
+            out.emplace_back(miner, algorithm, *this, platform, thread, devices[thread.index()], thread.threads().front());
-
-#           ifdef XMRIG_ALGO_RANDOMX
-            data.dataset = dataset;
-#           endif
-
-            out.emplace_back(std::move(data));
         }
     }
 

src/backend/opencl/OclLaunchData.h

@@ -35,11 +35,6 @@
 #include "crypto/common/Nonce.h"
 
 
-#ifdef XMRIG_ALGO_RANDOMX
-#   include "backend/opencl/runners/tools/OclRxDataset.h"
-#endif
-
-
 using cl_context = struct _cl_context *;
 
 
@@ -72,10 +67,6 @@ public:
     const OclDevice device;
     const OclPlatform platform;
     const OclThread thread;
-
-#   ifdef XMRIG_ALGO_RANDOMX
-    OclRxDatasetPtr dataset;
-#   endif
 };
 
 

src/backend/opencl/opencl.cmake

@@ -80,7 +80,6 @@ if (WITH_OPENCL)
            src/backend/opencl/runners/OclRxBaseRunner.h
            src/backend/opencl/runners/OclRxJitRunner.h
            src/backend/opencl/runners/OclRxVmRunner.h
-           src/backend/opencl/runners/tools/OclRxDataset.h
            )
 
        list(APPEND SOURCES_BACKEND_OPENCL
@@ -97,7 +96,6 @@ if (WITH_OPENCL)
            src/backend/opencl/runners/OclRxBaseRunner.cpp
            src/backend/opencl/runners/OclRxJitRunner.cpp
            src/backend/opencl/runners/OclRxVmRunner.cpp
-           src/backend/opencl/runners/tools/OclRxDataset.cpp
            )
    endif()
 

src/backend/opencl/runners/OclRxBaseRunner.cpp

@@ -30,6 +30,7 @@
 #include "backend/opencl/kernels/rx/FindSharesKernel.h"
 #include "backend/opencl/kernels/rx/HashAesKernel.h"
 #include "backend/opencl/OclLaunchData.h"
+#include "backend/opencl/runners/tools/OclSharedState.h"
 #include "backend/opencl/wrappers/OclLib.h"
 #include "base/net/stratum/Job.h"
 #include "crypto/rx/Rx.h"
@@ -120,7 +121,7 @@ void xmrig::OclRxBaseRunner::set(const Job &job, uint8_t *blob)
         m_seed = job.seed();
 
         auto dataset = Rx::dataset(job, 0);
-        enqueueWriteBuffer(data().dataset->get(), CL_TRUE, 0, dataset->size(), dataset->raw());
+        enqueueWriteBuffer(m_dataset, CL_TRUE, 0, dataset->size(), dataset->raw());
     }
 
     if (job.size() < Job::kMaxBlobSize) {
@@ -177,4 +178,5 @@ void xmrig::OclRxBaseRunner::init()
     m_hashes      = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, 64 * m_intensity);
     m_entropy     = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, (128 + 2560) * m_intensity);
     m_rounding    = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, sizeof(uint32_t) * m_intensity);
+    m_dataset     = OclSharedState::get(data().device.index()).dataset();
 }

src/backend/opencl/runners/OclRxBaseRunner.h

@@ -62,6 +62,7 @@ protected:
     Blake2bHashRegistersKernel *m_blake2b_hash_registers_64 = nullptr;
     Blake2bInitialHashKernel *m_blake2b_initial_hash        = nullptr;
     Buffer m_seed;
+    cl_mem m_dataset                                        = nullptr;
     cl_mem m_entropy                                        = nullptr;
     cl_mem m_hashes                                         = nullptr;
     cl_mem m_rounding                                       = nullptr;

src/backend/opencl/runners/OclRxJitRunner.cpp

@@ -74,7 +74,7 @@ void xmrig::OclRxJitRunner::build()
     }
 
     m_randomx_run = new RxRunKernel(m_asmProgram);
-    m_randomx_run->setArgs(data().dataset->get(), m_scratchpads, m_registers, m_rounding, m_programs, m_intensity, m_algorithm);
+    m_randomx_run->setArgs(m_dataset, m_scratchpads, m_registers, m_rounding, m_programs, m_intensity, m_algorithm);
 }
 
 

src/backend/opencl/runners/OclRxVmRunner.cpp

@@ -70,7 +70,7 @@ void xmrig::OclRxVmRunner::build()
     m_init_vm->setArgs(m_entropy, m_vm_states, m_rounding);
 
     m_execute_vm = new ExecuteVmKernel(m_program);
-    m_execute_vm->setArgs(m_vm_states, m_rounding, m_scratchpads, data().dataset->get(), m_intensity);
+    m_execute_vm->setArgs(m_vm_states, m_rounding, m_scratchpads, m_dataset, m_intensity);
 }
 
 

src/backend/opencl/runners/tools/OclRxDataset.cpp

@@ -1,54 +0,0 @@
-/* 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 2017-2018 XMR-Stak    <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
- * 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
- *   the Free Software Foundation, either version 3 of the License, or
- *   (at your option) any later version.
- *
- *   This program is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- *   GNU General Public License for more details.
- *
- *   You should have received a copy of the GNU General Public License
- *   along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-
-#include "backend/opencl/runners/tools/OclRxDataset.h"
-#include "backend/opencl/wrappers/OclLib.h"
-#include "crypto/rx/Rx.h"
-#include "crypto/rx/RxDataset.h"
-
-
-void xmrig::OclRxDataset::createBuffer(cl_context ctx, const Job &job, bool host)
-{
-    if (m_dataset) {
-        return;
-    }
-
-    cl_int ret;
-
-    if (host) {
-        auto dataset = Rx::dataset(job, 0);
-
-        m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, RxDataset::maxSize(), dataset->raw(), &ret);
-    }
-    else {
-        m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY, RxDataset::maxSize(), nullptr, &ret);
-    }
-}
-
-
-xmrig::OclRxDataset::~OclRxDataset()
-{
-    OclLib::release(m_dataset);
-}

src/backend/opencl/runners/tools/OclRxDataset.h

@@ -1,68 +0,0 @@
-/* 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 2017-2018 XMR-Stak    <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
- * 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
- *   the Free Software Foundation, either version 3 of the License, or
- *   (at your option) any later version.
- *
- *   This program is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- *   GNU General Public License for more details.
- *
- *   You should have received a copy of the GNU General Public License
- *   along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef XMRIG_OCLRXDATASET_H
-#define XMRIG_OCLRXDATASET_H
-
-
-#include "base/tools/Object.h"
-
-
-#include <memory>
-
-
-using cl_context = struct _cl_context *;
-using cl_mem     = struct _cl_mem *;
-
-
-namespace xmrig {
-
-
-class Job;
-
-
-class OclRxDataset
-{
-public:
-    XMRIG_DISABLE_COPY_MOVE(OclRxDataset)
-
-    OclRxDataset() = default;
-    ~OclRxDataset();
-
-    inline cl_mem get() const { return m_dataset; }
-
-    void createBuffer(cl_context ctx,const Job &job, bool host);
-
-private:
-    cl_mem m_dataset = nullptr;
-};
-
-
-using OclRxDatasetPtr = std::shared_ptr<OclRxDataset>;
-
-
-} /* namespace xmrig */
-
-
-#endif /* XMRIG_OCLINTERLEAVE_H */

src/backend/opencl/runners/tools/OclSharedData.cpp

@@ -24,11 +24,15 @@
 
 
 #include "backend/opencl/runners/tools/OclSharedData.h"
+#include "backend/opencl/wrappers/OclLib.h"
 #include "base/io/log/Log.h"
 #include "base/tools/Chrono.h"
+#include "crypto/rx/Rx.h"
+#include "crypto/rx/RxDataset.h"
 
 
 #include <cinttypes>
+#include <stdexcept>
 #include <thread>
 
 
@@ -107,6 +111,14 @@ uint64_t xmrig::OclSharedData::resumeDelay(size_t id)
 }
 
 
+void xmrig::OclSharedData::release()
+{
+#   ifdef XMRIG_ALGO_RANDOMX
+    OclLib::release(m_dataset);
+#   endif
+}
+
+
 void xmrig::OclSharedData::setResumeCounter(uint32_t value)
 {
     if (m_threads < 2) {
@@ -128,3 +140,34 @@ void xmrig::OclSharedData::setRunTime(uint64_t time)
     std::lock_guard<std::mutex> lock(m_mutex);
     m_averageRunTime = m_averageRunTime * (1.0 - averagingBias) + time * averagingBias;
 }
+
+
+#ifdef XMRIG_ALGO_RANDOMX
+cl_mem xmrig::OclSharedData::dataset() const
+{
+    if (!m_dataset) {
+        throw std::runtime_error("RandomX dataset is not available");
+    }
+
+    return m_dataset;
+}
+
+
+void xmrig::OclSharedData::createDataset(cl_context ctx, const Job &job, bool host)
+{
+    if (m_dataset) {
+        return;
+    }
+
+    cl_int ret;
+
+    if (host) {
+        auto dataset = Rx::dataset(job, 0);
+
+        m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, RxDataset::maxSize(), dataset->raw(), &ret);
+    }
+    else {
+        m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY, RxDataset::maxSize(), nullptr, &ret);
+    }
+}
+#endif

src/backend/opencl/runners/tools/OclSharedData.h

@@ -30,9 +30,16 @@
 #include <mutex>
 
 
+using cl_context = struct _cl_context *;
+using cl_mem     = struct _cl_mem *;
+
+
 namespace xmrig {
 
 
+class Job;
+
+
 class OclSharedData
 {
 public:
@@ -40,11 +47,17 @@ public:
 
     uint64_t adjustDelay(size_t id);
     uint64_t resumeDelay(size_t id);
+    void release();
     void setResumeCounter(uint32_t value);
     void setRunTime(uint64_t time);
 
     inline OclSharedData &operator++() { ++m_threads; return *this; }
 
+#   ifdef XMRIG_ALGO_RANDOMX
+    cl_mem dataset() const;
+    void createDataset(cl_context ctx, const Job &job, bool host);
+#   endif
+
 private:
     double m_averageRunTime   = 0.0;
     double m_threshold        = 0.95;
@@ -52,12 +65,13 @@ private:
     std::mutex m_mutex;
     uint32_t m_resumeCounter  = 0;
     uint64_t m_timestamp      = 0;
+
+#   ifdef XMRIG_ALGO_RANDOMX
+    cl_mem m_dataset          = nullptr;
+#   endif
 };
 
 
-using OclSharedDataPtr = std::shared_ptr<OclSharedData>;
-
-
 } /* namespace xmrig */
 
 

src/backend/opencl/runners/tools/OclSharedState.cpp

@@ -48,15 +48,27 @@ xmrig::OclSharedData &xmrig::OclSharedState::get(uint32_t index)
 
 void xmrig::OclSharedState::release()
 {
+    for (auto &kv : map) {
+        kv.second.release();
+    }
+
     map.clear();
 }
 
 
-void xmrig::OclSharedState::start(const std::vector<OclLaunchData> &threads)
+void xmrig::OclSharedState::start(const std::vector<OclLaunchData> &threads, const Job &job)
 {
     assert(map.empty());
 
     for (const auto &data : threads) {
-        ++map[data.device.index()];
+        auto &sharedData = map[data.device.index()];
+
+        ++sharedData;
+
+#       ifdef XMRIG_ALGO_RANDOMX
+        if (data.algorithm.family() == Algorithm::RANDOM_X) {
+            sharedData.createDataset(data.ctx, job, data.thread.isDatasetHost());
+        }
+#       endif
     }
 }

src/backend/opencl/runners/tools/OclSharedState.h

@@ -37,8 +37,7 @@ class OclSharedState
 public:
     static OclSharedData &get(uint32_t index);
     static void release();
-    static void start(const std::vector<OclLaunchData> &threads);
+    static void start(const std::vector<OclLaunchData> &threads, const Job &job);
-
 };
 
 

src/backend/opencl/wrappers/OclContext.cpp

@@ -24,6 +24,7 @@
 
 
 #include "backend/opencl/wrappers/OclContext.h"
+#include "backend/opencl/runners/tools/OclSharedState.h"
 #include "backend/opencl/wrappers/OclLib.h"
 
 
@@ -42,7 +43,7 @@ xmrig::OclContext::~OclContext()
 }
 
 
-bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads, const Job &job)
+bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads)
 {
     if (!m_ctx) {
         std::vector<cl_device_id> ids(devices.size());
@@ -59,12 +60,6 @@ bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<
 
     for (OclLaunchData &data : threads) {
         data.ctx = m_ctx;
-
-#       ifdef XMRIG_ALGO_RANDOMX
-        if (data.algorithm.family() == Algorithm::RANDOM_X) {
-            data.dataset->createBuffer(m_ctx, job, data.thread.isDatasetHost());
-        }
-#       endif
     }
 
     return true;

src/backend/opencl/wrappers/OclContext.h

@@ -49,7 +49,7 @@ public:
     OclContext(const OclDevice &device);
     ~OclContext();
 
-    bool init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads, const Job &job);
+    bool init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads);
 
     inline bool isValid() const     { return m_ctx != nullptr; }
     inline cl_context ctx() const   { return m_ctx; }