Merge branch 'ph4r05-pr/001-with-sse' into dev

CMakeLists.txt

@@ -24,6 +24,7 @@ option(WITH_ADL             "Enable ADL (AMD Display Library) or sysfs support (
 option(WITH_STRICT_CACHE    "Enable strict checks for OpenCL cache" ON)
 option(WITH_INTERLEAVE_DEBUG_LOG "Enable debug log for threads interleave" OFF)
 option(WITH_PROFILING       "Enable profiling for developers" OFF)
+option(WITH_SSE4_1          "Enable SSE 4.1 for Blake2" ON)
 
 option(BUILD_STATIC         "Build static binary" OFF)
 option(ARM_TARGET           "Force use specific ARM target 8 or 7" 0)

cmake/cpu.cmake

@@ -2,9 +2,10 @@ if (NOT CMAKE_SYSTEM_PROCESSOR)
     message(WARNING "CMAKE_SYSTEM_PROCESSOR not defined")
 endif()
 
-
 if (CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")
     add_definitions(/DRAPIDJSON_SSE2)
+else()
+    set(WITH_SSE4_1 OFF)
 endif()
 
 if (NOT ARM_TARGET)
@@ -41,3 +42,7 @@ if (ARM_TARGET AND ARM_TARGET GREATER 6)
         add_definitions(/DXMRIG_ARMv7)
     endif()
 endif()
+
+if (WITH_SSE4_1)
+    add_definitions(/DXMRIG_FEATURE_SSE4_1)
+endif()

cmake/randomx.cmake

@@ -19,7 +19,6 @@ if (WITH_RANDOMX)
         src/crypto/randomx/allocator.cpp
         src/crypto/randomx/blake2_generator.cpp
         src/crypto/randomx/blake2/blake2b.c
-        src/crypto/randomx/blake2/blake2b_sse41.c
         src/crypto/randomx/bytecode_machine.cpp
         src/crypto/randomx/dataset.cpp
         src/crypto/randomx/instructions_portable.cpp
@@ -65,9 +64,13 @@ if (WITH_RANDOMX)
         set_property(SOURCE src/crypto/randomx/jit_compiler_a64_static.S PROPERTY LANGUAGE C)
     endif()
 
+    if (WITH_SSE4_1)
+        list(APPEND SOURCES_CRYPTO src/crypto/randomx/blake2/blake2b_sse41.c)
+
         if (CMAKE_C_COMPILER_ID MATCHES GNU OR CMAKE_C_COMPILER_ID MATCHES Clang)
             set_source_files_properties(src/crypto/randomx/blake2/blake2b_sse41.c PROPERTIES COMPILE_FLAGS -msse4.1)
         endif()
+    endif()
 
     if (CMAKE_CXX_COMPILER_ID MATCHES Clang)
         set_source_files_properties(src/crypto/randomx/jit_compiler_x86.cpp PROPERTIES COMPILE_FLAGS -Wno-unused-const-variable)

doc/build/CMAKE_OPTIONS.md

@@ -22,6 +22,7 @@ This feature add external dependency to libhwloc (1.10.0+) (except MSVC builds).
 * **`-DWITH_EMBEDDED_CONFIG=ON`** Enable [embedded](https://github.com/xmrig/xmrig/issues/957) config support.
 * **`-DWITH_OPENCL=OFF`** Disable OpenCL backend.
 * **`-DWITH_CUDA=OFF`** Disable CUDA backend.
+* **`-DWITH_SSE4_1=OFF`** Disable SSE 4.1 for Blake2 (useful for arm builds).
 
 ## Debug options
 

src/crypto/randomx/blake2/blake2b.c

@@ -237,7 +237,7 @@ static void rx_blake2b_compress_integer(blake2b_state *S, const uint8_t *block)
 #undef ROUND
 }
 
-#if defined(_M_X64) || defined(__x86_64__)
+#if defined(XMRIG_FEATURE_SSE4_1)
 
 uint32_t rx_blake2b_use_sse41 = 0;
 void rx_blake2b_compress_sse41(blake2b_state* S, const uint8_t* block);

src/crypto/rx/RxVm.cpp

@@ -31,7 +31,7 @@
 #include "crypto/rx/RxVm.h"
 
 
-#if defined(_M_X64) || defined(__x86_64__)
+#if defined(XMRIG_FEATURE_SSE4_1)
 extern "C" uint32_t rx_blake2b_use_sse41;
 #endif
 
@@ -60,7 +60,7 @@ randomx_vm* xmrig::RxVm::create(RxDataset *dataset, uint8_t *scratchpad, bool so
         flags |= RANDOMX_FLAG_AMD;
     }
 
-#   if defined(_M_X64) || defined(__x86_64__)
+#   if defined(XMRIG_FEATURE_SSE4_1)
     rx_blake2b_use_sse41 = Cpu::info()->has(ICpuInfo::FLAG_SSE41) ? 1 : 0;
 #   endif