Merge pull request #3605 from SChernykh/dev

CUDA backend: update RandomX dataset when it changes

src/backend/cuda/runners/CudaRxRunner.cpp

@@ -55,6 +55,15 @@ bool xmrig::CudaRxRunner::run(uint32_t startNonce, uint32_t *rescount, uint32_t
 
 bool xmrig::CudaRxRunner::set(const Job &job, uint8_t *blob)
 {
+    if (!m_datasetHost && (m_seed != job.seed())) {
+        m_seed = job.seed();
+
+        if (m_ready) {
+            auto dataset = Rx::dataset(job, 0);
+            callWrapper(CudaLib::rxUpdateDataset(m_ctx, dataset->raw(), dataset->size(false)));
+        }
+    }
+
     const bool rc = CudaBaseRunner::set(job, blob);
     if (!rc || m_ready) {
         return rc;

src/backend/cuda/runners/CudaRxRunner.h

@@ -27,6 +27,7 @@
 
 
 #include "backend/cuda/runners/CudaBaseRunner.h"
+#include "base/tools/Buffer.h"
 
 
 namespace xmrig {
@@ -46,6 +47,7 @@ protected:
 private:
     bool m_ready             = false;
     const bool m_datasetHost = false;
+    Buffer m_seed;
     size_t m_intensity       = 0;
 };
 

src/backend/cuda/wrappers/CudaLib.cpp

@@ -68,6 +68,7 @@ static const char *kPluginVersion                       = "pluginVersion";
 static const char *kRelease                             = "release";
 static const char *kRxHash                              = "rxHash";
 static const char *kRxPrepare                           = "rxPrepare";
+static const char *kRxUpdateDataset                     = "rxUpdateDataset";
 static const char *kSetJob                              = "setJob";
 static const char *kSetJob_v2                           = "setJob_v2";
 static const char *kVersion                             = "version";
@@ -92,6 +93,7 @@ using pluginVersion_t                                   = const char * (*)();
 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 rxUpdateDataset_t                                 = bool (*)(nvid_ctx *, const void *, size_t);
 using setJob_t                                          = bool (*)(nvid_ctx *, const void *, size_t, uint32_t);
 using setJob_v2_t                                       = bool (*)(nvid_ctx *, const void *, size_t, const char *);
 using version_t                                         = uint32_t (*)(Version);
@@ -116,6 +118,7 @@ static pluginVersion_t pPluginVersion                   = nullptr;
 static release_t pRelease                               = nullptr;
 static rxHash_t pRxHash                                 = nullptr;
 static rxPrepare_t pRxPrepare                           = nullptr;
+static rxUpdateDataset_t pRxUpdateDataset               = nullptr;
 static setJob_t pSetJob                                 = nullptr;
 static setJob_v2_t pSetJob_v2                           = nullptr;
 static version_t pVersion                               = nullptr;
@@ -206,6 +209,12 @@ bool xmrig::CudaLib::rxPrepare(nvid_ctx *ctx, const void *dataset, size_t datase
 }
 
 
+bool xmrig::CudaLib::rxUpdateDataset(nvid_ctx* ctx, const void* dataset, size_t datasetSize) noexcept
+{
+    return pRxUpdateDataset(ctx, dataset, datasetSize);
+}
+
+
 bool xmrig::CudaLib::kawPowHash(nvid_ctx *ctx, uint8_t* job_blob, uint64_t target, uint32_t *rescount, uint32_t *resnonce, uint32_t *skipped_hashes) noexcept
 {
     return pKawPowHash(ctx, job_blob, target, rescount, resnonce, skipped_hashes);
@@ -388,6 +397,7 @@ void xmrig::CudaLib::load()
     DLSYM(Release);
     DLSYM(RxHash);
     DLSYM(RxPrepare);
+    DLSYM(RxUpdateDataset);
     DLSYM(KawPowHash);
     DLSYM(KawPowPrepare_v2);
     DLSYM(KawPowStopHash);

src/backend/cuda/wrappers/CudaLib.h

@@ -71,6 +71,7 @@ public:
     static bool deviceInit(nvid_ctx *ctx) noexcept;
     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 rxUpdateDataset(nvid_ctx* ctx, const void* dataset, size_t datasetSize) 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, const void* dag_precalc, size_t dag_size, uint32_t height, const uint64_t* dag_sizes) noexcept;
     static bool kawPowStopHash(nvid_ctx *ctx) noexcept;