RandomX integration (Wownero and Loki variants)

CMakeLists.txt

@@ -5,6 +5,7 @@ option(WITH_AEON      "CryptoNight-Lite support" ON)
 option(WITH_SUMO      "CryptoNight-Heavy support" ON)
 option(WITH_CN_PICO   "CryptoNight-Pico support" ON)
 option(WITH_CN_GPU    "CryptoNight-GPU support" ON)
+option(WITH_RANDOMX   "RandomX support" ON)
 option(WITH_HTTPD     "HTTP REST API" ON)
 option(WITH_TLS       "Enable OpenSSL support" ON)
 option(WITH_ASM       "Enable ASM PoW implementations" ON)
@@ -233,6 +234,52 @@ set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake")
 
 find_package(UV REQUIRED)
 
+if (WITH_RANDOMX)
+    include_directories(src/crypto/randomx)
+    add_definitions(/DXMRIG_ALGO_RANDOMX)
+    set(SOURCES_CRYPTO
+        "${SOURCES_CRYPTO}"
+        src/crypto/randomx/aes_hash.cpp
+        src/crypto/randomx/argon2_ref.c
+        src/crypto/randomx/bytecode_machine.cpp
+        src/crypto/randomx/dataset.cpp
+        src/crypto/randomx/soft_aes.cpp
+        src/crypto/randomx/virtual_memory.cpp
+        src/crypto/randomx/vm_interpreted.cpp
+        src/crypto/randomx/allocator.cpp
+        src/crypto/randomx/randomx.cpp
+        src/crypto/randomx/superscalar.cpp
+        src/crypto/randomx/vm_compiled.cpp
+        src/crypto/randomx/vm_interpreted_light.cpp
+        src/crypto/randomx/argon2_core.c
+        src/crypto/randomx/blake2_generator.cpp
+        src/crypto/randomx/instructions_portable.cpp
+        src/crypto/randomx/reciprocal.c
+        src/crypto/randomx/virtual_machine.cpp
+        src/crypto/randomx/vm_compiled_light.cpp
+        src/crypto/randomx/blake2/blake2b.c
+    )
+    if (NOT ARCH_ID)
+        set(ARCH_ID ${CMAKE_HOST_SYSTEM_PROCESSOR})
+    endif()
+    if (CMAKE_C_COMPILER_ID MATCHES MSVC)
+        enable_language(ASM_MASM)
+        list(APPEND SOURCES_CRYPTO
+             src/crypto/randomx/jit_compiler_x86_static.asm
+             src/crypto/randomx/jit_compiler_x86.cpp
+            )
+    elseif (NOT XMRIG_ARM AND CMAKE_SIZEOF_VOID_P EQUAL 8)
+        list(APPEND SOURCES_CRYPTO
+             src/crypto/randomx/jit_compiler_x86_static.S
+             src/crypto/randomx/jit_compiler_x86.cpp
+            )
+        # cheat because cmake and ccache hate each other
+        set_property(SOURCE src/crypto/randomx/jit_compiler_x86_static.S PROPERTY LANGUAGE C)
+    endif()
+else()
+    remove_definitions(/DXMRIG_ALGO_RANDOMX)
+endif()
+
 include(cmake/flags.cmake)
 
 add_definitions(/DCL_TARGET_OPENCL_VERSION=200)

src/amd/GpuContext.h

@@ -47,6 +47,7 @@ struct GpuContext
         deviceIdx(0),
         rawIntensity(0),
         workSize(0),
+        bfactor(6),
         threads(0),
         stridedIndex(2),
         memChunk(2),
@@ -69,15 +70,25 @@ struct GpuContext
         freeMem(0),
         globalMem(0),
         computeUnits(0),
-        Nonce(0)
+        Nonce(0),
+        rx_variant(xmrig::VARIANT_AUTO),
+        rx_dataset(nullptr),
+        rx_scratchpads(nullptr),
+        rx_hashes(nullptr),
+        rx_entropy(nullptr),
+        rx_vm_states(nullptr),
+        rx_rounding(nullptr)
     {
         memset(Kernels, 0, sizeof(Kernels));
+        memset(rx_dataset_seedhash, 0, sizeof(rx_dataset_seedhash));
+        memset(rx_kernels, 0, sizeof(rx_kernels));
     }
 
     /*Input vars*/
     size_t deviceIdx;
     size_t rawIntensity;
     size_t workSize;
+    size_t bfactor;
     size_t threads;
     int stridedIndex;
     int memChunk;
@@ -107,6 +118,16 @@ struct GpuContext
     xmrig::String name;
 
     uint32_t Nonce;
+
+    uint8_t rx_dataset_seedhash[32];
+    xmrig::Variant rx_variant;
+    cl_mem rx_dataset;
+    cl_mem rx_scratchpads;
+    cl_mem rx_hashes;
+    cl_mem rx_entropy;
+    cl_mem rx_vm_states;
+    cl_mem rx_rounding;
+    cl_kernel rx_kernels[32];
 };
 
 

src/amd/OclCache.cpp

@@ -68,20 +68,42 @@ cl_int OclCache::wait_build(cl_program program, cl_device_id device)
 
 void OclCache::getOptions(xmrig::Algo algo, xmrig::Variant, const GpuContext* ctx, char* options, size_t options_size)
 {
-    snprintf(options, options_size, "-DITERATIONS=%u -DMASK=%u -DWORKSIZE=%zu -DSTRIDED_INDEX=%d -DMEM_CHUNK_EXPONENT=%d -DCOMP_MODE=%d -DMEMORY=%zu "
-        "-DALGO=%d -DUNROLL_FACTOR=%d -DOPENCL_DRIVER_MAJOR=%d -DWORKSIZE_GPU=%zu -cl-fp32-correctly-rounded-divide-sqrt",
-        xmrig::cn_select_iter(algo, xmrig::VARIANT_AUTO),
-        xmrig::cn_select_mask(algo),
-        ctx->workSize,
-        ctx->stridedIndex,
-        static_cast<int>(1u << ctx->memChunk),
-        ctx->compMode,
-        xmrig::cn_select_memory(algo),
-        static_cast<int>(algo),
-        ctx->unrollFactor,
-        ctx->amdDriverMajorVersion,
-        worksize(ctx, xmrig::VARIANT_GPU)
-    );
+#   ifdef XMRIG_ALGO_RANDOMX
+    if (algo == xmrig::RANDOM_X) {
+        uint32_t workSize;
+        switch (ctx->workSize)
+        {
+        case 2:
+        case 4:
+        case 8:
+        case 16:
+            workSize = ctx->workSize;
+            break;
+
+        default:
+            workSize = 8;
+        }
+
+        snprintf(options, options_size, "-DWORKERS_PER_HASH=%u", workSize);
+    }
+    else
+#   endif
+    {
+        snprintf(options, options_size, "-DITERATIONS=%u -DMASK=%u -DWORKSIZE=%zu -DSTRIDED_INDEX=%d -DMEM_CHUNK_EXPONENT=%d -DCOMP_MODE=%d -DMEMORY=%zu "
+            "-DALGO=%d -DUNROLL_FACTOR=%d -DOPENCL_DRIVER_MAJOR=%d -DWORKSIZE_GPU=%zu -cl-fp32-correctly-rounded-divide-sqrt",
+            xmrig::cn_select_iter(algo, xmrig::VARIANT_AUTO),
+            xmrig::cn_select_mask(algo),
+            ctx->workSize,
+            ctx->stridedIndex,
+            static_cast<int>(1u << ctx->memChunk),
+            ctx->compMode,
+            xmrig::cn_select_memory(algo),
+            static_cast<int>(algo),
+            ctx->unrollFactor,
+            ctx->amdDriverMajorVersion,
+            worksize(ctx, xmrig::VARIANT_GPU)
+        );
+    }
 }
 
 bool OclCache::load()