Begin to wire asm stubs into detection (extend asm_type); Add more CPU detection so asm_type can be matched up

Author: Spudz76

xmrstak/backend/cpu/cpuType.cpp

@@ -27,10 +27,10 @@ namespace cpu
 	#endif
 	}
 
-	int32_t get_masked(int32_t val, int32_t h, int32_t l)
+	uint32_t get_masked(int32_t val, int32_t h, int32_t l)
 	{
 		val &= (0x7FFFFFFF >> (31-(h-l))) << l;
-		return val >> l;
+		return static_cast<uint32_t>(val >> l);
 	}
 
 	bool has_feature(int32_t val, int32_t bit)
@@ -42,34 +42,86 @@ namespace cpu
 
 	Model getModel()
 	{
+		Model result;
+
+		int32_t cpu_HFP = 0;  // Highest Function Parameter
+		int32_t cpu_HEFP = 0; // Highest Extended Function Parameter
 		int32_t cpu_info[4];
 		char cpustr[13] = {0};
+		char brandstr[13] = {0};
 
 		cpuid(0, 0, cpu_info);
+		cpu_HFP = cpu_info[0];
 		std::memcpy(cpustr, &cpu_info[1], 4);
 		std::memcpy(cpustr+4, &cpu_info[3], 4);
 		std::memcpy(cpustr+8, &cpu_info[2], 4);
 
-		Model result;
-
 		cpuid(1, 0, cpu_info);
-
+		result.model = get_masked(cpu_info[0], 8, 4);
 		result.family = get_masked(cpu_info[0], 12, 8);
-		result.model = get_masked(cpu_info[0], 8, 4) | get_masked(cpu_info[0], 20, 16) << 4;
-		result.type_name = cpustr;
+		if(result.family == 0x6 || result.family == 0xF)
+		{
+			result.model += get_masked(cpu_info[0], 20, 16) << 4;
+		}
+		if(result.family != 0xF)
+		{
+			result.family += get_masked(cpu_info[0], 28, 20);
+		}
 
-		// feature bits https://en.wikipedia.org/wiki/CPUID
-		// sse2
+		// feature bits https://en.wikipedia.org/wiki/CPUID#EAX=1:_Processor_Info_and_Feature_Bits
+		// sse2/sse3/ssse3
 		result.sse2 = has_feature(cpu_info[3], 26);
+		result.sse3 = has_feature(cpu_info[2], 0);
+		result.ssse3 = has_feature(cpu_info[2], 9);
 		// aes-ni
 		result.aes = has_feature(cpu_info[2], 25);
 		// avx - 27 is the check if the OS overwrote cpu features
-		result.avx = has_feature(cpu_info[2], 28) && has_feature(cpu_info[2], 27) ;
+		result.avx = has_feature(cpu_info[2], 28) && has_feature(cpu_info[2], 27);
 
+		// extended feature bits https://en.wikipedia.org/wiki/CPUID#EAX=7,_ECX=0:_Extended_Features
+		if(cpu_HFP >= 7)
+		{
+			cpuid(7, 0, cpu_info);
+			result.avx2 = has_feature(cpu_info[1], 5);
+		}
+		// extended function support https://en.wikipedia.org/wiki/CPUID#EAX=80000000h:_Get_Highest_Extended_Function_Implemented
+		cpuid(0x80000000, 0, cpu_info);
+		cpu_HEFP = cpu_info[0];
+
+		// processor brand string https://en.wikipedia.org/wiki/CPUID#EAX=80000002h,80000003h,80000004h:_Processor_Brand_String
+		if(cpu_HEFP >= 0x80000004)
+		{
+			for(uint32_t efp=0x80000002; efp<0x80000004; efp++){
+				cpuid(0x80000002, 0, cpu_info);
+				std::memcpy(brandstr+(16*(efp-0x80000002)), &cpu_info, 16);
+			}
+			result.brand_name = brandstr;
+		}
+
+		if(strcmp(cpustr, "GenuineIntel") == 0)
+		{
+			if(result.family == 0x6){
+				result.isIntelXBridge = (
+					   result.model == 0x2A //Sandy Bridge
+					|| result.model == 0x3A //Ivy Bridge
+				);
+				result.isIntelXWell = (
+					   result.model == 0x3C || result.model == 0x45 || result.model == 0x46 //Haswell
+					|| result.model == 0x47 || result.model == 0x3D //Broadwell
+				);
+				result.isIntelXLake = (
+					   result.model == 0x4E || result.model == 0x5E //Skylake
+					|| result.model == 0x8E //Kaby/Coffee/Whiskey/Amber Lake
+					|| result.model == 0x9E //Kaby/Coffee Lake
+					|| result.model == 0x66 //Cannon Lake
+				);
+			}
+		}
 		if(strcmp(cpustr, "AuthenticAMD") == 0)
 		{
-			if(result.family == 0xF)
-				result.family += get_masked(cpu_info[0], 28, 20);
+			result.isAMDHammer    = (result.family != 0x15 && result.family >= 0xF && result.family <= 0x16);
+			result.isAMDBulldozer = (result.family == 0x15);
+			result.isAMDZen       = (result.family == 0x17);
 		}
 
 		return result;

xmrstak/backend/cpu/cpuType.hpp

@@ -12,10 +12,20 @@ namespace cpu
 	{
 		uint32_t family = 0u;
 		uint32_t model = 0u;
+		bool isIntelXBridge = false;
+		bool isIntelXWell = false;
+		bool isIntelXLake = false;
+		bool isAMDHammer = false;
+		bool isAMDBulldozer = false;
+		bool isAMDZen = false;
 		bool aes = false;
 		bool sse2 = false;
+		bool sse3 = false;
+		bool ssse3 = false;
 		bool avx = false;
+		bool avx2 = false;
 		std::string type_name = "unknown";
+		std::string brand_name = "unknown";
 	};
 
 	Model getModel();
@@ -25,7 +35,7 @@ namespace cpu
 	 * This enables us to put in values exactly like in the manual
 	 * For example EBX[30:22] is get_masked(cpu_info[1], 31, 22)
 	 */
-	int32_t get_masked(int32_t val, int32_t h, int32_t l);
+	uint32_t get_masked(int32_t val, int32_t h, int32_t l);
 
 
 } // namespace cpu

xmrstak/backend/cpu/crypto/cryptonight_aesni.h

@@ -1179,9 +1179,12 @@ struct Cryptonight_hash<5>
 	}
 };
 
-extern "C" void  cryptonight_v8_mainloop_ivybridge_asm(cryptonight_ctx* ctx0);
-extern "C" void  cryptonight_v8_mainloop_ryzen_asm(cryptonight_ctx* ctx0);
+extern "C" void cryptonight_v8_mainloop_ivybridge_asm(cryptonight_ctx* ctx0);
 extern "C" void cryptonight_v8_double_mainloop_sandybridge_asm(cryptonight_ctx* ctx0, cryptonight_ctx* ctx1);
+extern "C" void cryptonight_v8_mainloop_bulldozer_asm(cryptonight_ctx* ctx0);
+extern "C" void cryptonight_v8_mainloop_ryzen_asm(cryptonight_ctx* ctx0);
+extern "C" void cryptonight_v8_rwz_mainloop_asm(cryptonight_ctx* ctx0);
+extern "C" void cryptonight_v8_rwz_double_mainloop_asm(cryptonight_ctx* ctx0, cryptonight_ctx* ctx1);
 
 
 template< size_t N, size_t asm_version>
@@ -1298,7 +1301,7 @@ void* allocateExecutableMemory(size_t size)
 {
 
 #ifdef _WIN64
-return VirtualAlloc(0, size, MEM_COMMIT | MEM_RESERVE, PAGE_EXECUTE_READWRITE);
+return VirtualAlloc(nullptr, size, MEM_COMMIT | MEM_RESERVE, PAGE_EXECUTE_READWRITE);
 #else
 #   if defined(__APPLE__)
     return mmap(0, size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANON, -1, 0);
@@ -1356,7 +1359,7 @@ void patchAsmVariants(std::string selected_asm, cryptonight_ctx** ctx, const xmr
 
 	cn_mainloop_fun src_code = nullptr;
 
-	if(selected_asm == "intel_avx")
+	if(selected_asm == "intel_avx" || selected_asm == "ivybridge" || selected_asm == "sandybridge")
 	{
 		// Intel Ivy Bridge (Xeon v2, Core i7/i5/i3 3xxx, Pentium G2xxx, Celeron G1xxx)
 		if(N == 2)
@@ -1365,9 +1368,14 @@ void patchAsmVariants(std::string selected_asm, cryptonight_ctx** ctx, const xmr
 			src_code = cryptonight_v8_mainloop_ivybridge_asm;;
 	}
 	// supports only 1 thread per hash
-	if(selected_asm == "amd_avx")
+	if(selected_asm == "bulldozer")
 	{
-		// AMD Ryzen (1xxx and 2xxx series)
+		// AMD 15h "Bulldozer" - Orochi/Vishera etc; Bulldozer/Piledriver/Steamroller/Excavator 
+		src_code = cryptonight_v8_mainloop_bulldozer_asm;
+	}
+	if(selected_asm == "amd_avx" || selected_asm == "zen")
+	{
+		// AMD 17h "Zen" - Ryzen (1xxx and 2xxx series)
 		src_code = cryptonight_v8_mainloop_ryzen_asm;
 	}
 

xmrstak/backend/cpu/minethd.cpp

@@ -563,9 +563,9 @@ static std::string getAsmName(const uint32_t num_hashes)
 
 		if(cpu_model.avx && cpu_model.aes)
 		{
-			if(cpu_model.type_name.find("Intel") != std::string::npos)
+			if(cpu_model.isIntelXBridge || cpu_model.isIntelXWell || cpu_model.isIntelXLake)
 				asm_type = "intel_avx";
-			else if(cpu_model.type_name.find("AMD") != std::string::npos)
+			else if(cpu_model.isAMDBulldozer || cpu_model.isAMDZen)
 				asm_type = "amd_avx";
 		}
 	}