rx-wow-fix-3: Revert "Increase the frequency of CBRANCH (#118)"

This reverts commit 22689eda49.

README.md

@@ -48,6 +48,8 @@ cmake -DARCH=native ..
 make
 ```
 
+To build portable binaries, omit the `ARCH` option when executing cmake.
+
 ### Windows
 
 On Windows, it is possible to build using MinGW (same procedure as on Linux) or using Visual Studio (solution file is provided).

doc/design.md

@@ -255,7 +255,7 @@ The Scratchpad is split into 3 levels to mimic the typical CPU cache hierarchy [
 |----------------|----------|----------|----------|------|
 ARM Cortex A55|2|6|-|[[24](https://www.anandtech.com/show/11441/dynamiq-and-arms-new-cpus-cortex-a75-a55/4)]
 |AMD Zen+|4|12|40|[[25](https://en.wikichip.org/wiki/amd/microarchitectures/zen%2B#Memory_Hierarchy)]|
-|Intel Skylake|4|12|42|[[26](https://en.wikichip.org/wiki/amd/microarchitectures/zen%2B#Memory_Hierarchy)]
+|Intel Skylake|4|12|42|[[26](https://en.wikichip.org/wiki/intel/microarchitectures/skylake_(client)#Memory_Hierarchy)]
 
 The L3 cache is much larger and located further from the CPU core. As a result, its access latencies are much higher and can cause stalls in program execution.
 
@@ -638,7 +638,7 @@ state3 = 00000000000000000000000000000000
 
 [25] AMD Zen+ Microarchitecture - https://en.wikichip.org/wiki/amd/microarchitectures/zen%2B#Memory_Hierarchy
 
-[26] Intel Skylake Microarchitecture - https://en.wikichip.org/wiki/amd/microarchitectures/zen%2B#Memory_Hierarchy
+[26] Intel Skylake Microarchitecture - https://en.wikichip.org/wiki/intel/microarchitectures/skylake_(client)#Memory_Hierarchy
 
 [27] Biryukov et al.: Fast and Tradeoff-Resilient Memory-Hard Functions for
 Cryptocurrencies and Password Hashing - https://eprint.iacr.org/2015/430.pdf Table 2, page 8
@@ -647,4 +647,4 @@ Cryptocurrencies and Password Hashing - https://eprint.iacr.org/2015/430.pdf Tab
 
 [29] 7-Zip File archiver - https://www.7-zip.org/
 
-[30] TestU01 library - http://simul.iro.umontreal.ca/testu01/tu01.html
+[30] TestU01 library - http://simul.iro.umontreal.ca/testu01/tu01.html

src/instructions_portable.cpp

@@ -157,6 +157,21 @@ void rx_set_rounding_mode(uint32_t mode) {
 	}
 }
 
+uint32_t rx_get_rounding_mode() {
+	switch (fegetround()) {
+	case FE_DOWNWARD:
+		return RoundDown;
+	case FE_UPWARD:
+		return RoundUp;
+	case FE_TOWARDZERO:
+		return RoundToZero;
+	case FE_TONEAREST:
+		return RoundToNearest;
+	default:
+		UNREACHABLE;
+	}
+}
+
 #endif
 
 #ifdef RANDOMX_USE_X87

src/intrin_portable.h

@@ -173,6 +173,10 @@ FORCE_INLINE void rx_set_rounding_mode(uint32_t mode) {
 	_mm_setcsr(rx_mxcsr_default | (mode << 13));
 }
 
+FORCE_INLINE uint32_t rx_get_rounding_mode() {
+	return (_mm_getcsr() >> 13) & 3;
+}
+
 #elif defined(__PPC64__) && defined(__ALTIVEC__) && defined(__VSX__) //sadly only POWER7 and newer will be able to use SIMD acceleration. Earlier processors cant use doubles or 64 bit integers with SIMD
 #include <cstdint>
 #include <stdexcept>
@@ -736,6 +740,8 @@ void rx_reset_float_state();
 
 void rx_set_rounding_mode(uint32_t mode);
 
+uint32_t rx_get_rounding_mode();
+
 #endif
 
 double loadDoublePortable(const void* addr);

src/jit_compiler.hpp

@@ -35,3 +35,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #else
 #include "jit_compiler_fallback.hpp"
 #endif
+
+#if defined(__OpenBSD__) || defined(__NetBSD__)
+#define RANDOMX_FORCE_SECURE
+#endif

src/jit_compiler_x86.cpp

@@ -295,7 +295,7 @@ namespace randomx {
 
 	void JitCompilerX86::generateProgramPrologue(Program& prog, ProgramConfiguration& pcfg) {
 		instructionOffsets.clear();
-		for (unsigned i = 0; i < 8; ++i) {
+		for (unsigned i = 0; i < RegistersCount; ++i) {
 			registerUsage[i] = -1;
 		}
 

src/randomx.cpp

@@ -36,13 +36,14 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "cpu.hpp"
 #include <cassert>
 #include <limits>
+#include <cfenv>
 
 extern "C" {
 
 	randomx_flags randomx_get_flags() {
 		randomx_flags flags = RANDOMX_HAVE_COMPILER ? RANDOMX_FLAG_JIT : RANDOMX_FLAG_DEFAULT;
 		randomx::Cpu cpu;
-#ifdef __OpenBSD__
+#ifdef RANDOMX_FORCE_SECURE
 		if (flags == RANDOMX_FLAG_JIT) {
 			flags |= RANDOMX_FLAG_SECURE;
 		}
@@ -328,7 +329,7 @@ extern "C" {
 	void randomx_vm_set_cache(randomx_vm *machine, randomx_cache* cache) {
 		assert(machine != nullptr);
 		assert(cache != nullptr && cache->isInitialized());
-		if (machine->cacheKey != cache->cacheKey) {
+		if (machine->cacheKey != cache->cacheKey || machine->getMemory() != cache->memory) {
 			machine->setCache(cache);
 			machine->cacheKey = cache->cacheKey;
 		}
@@ -349,6 +350,8 @@ extern "C" {
 		assert(machine != nullptr);
 		assert(inputSize == 0 || input != nullptr);
 		assert(output != nullptr);
+		fenv_t fpstate;
+		fegetenv(&fpstate);
 		alignas(16) uint64_t tempHash[8];
 		int blakeResult = blake2b(tempHash, sizeof(tempHash), input, inputSize, nullptr, 0);
 		assert(blakeResult == 0);
@@ -361,6 +364,7 @@ extern "C" {
 		}
 		machine->run(&tempHash);
 		machine->getFinalResult(output, RANDOMX_HASH_SIZE);
+		fesetenv(&fpstate);
 	}
 
 	void randomx_calculate_hash_first(randomx_vm* machine, const void* input, size_t inputSize) {
@@ -380,4 +384,14 @@ extern "C" {
 		blake2b(machine->tempHash, sizeof(machine->tempHash), nextInput, nextInputSize, nullptr, 0);
 		machine->hashAndFill(output, RANDOMX_HASH_SIZE, machine->tempHash);
 	}
+
+	void randomx_calculate_hash_last(randomx_vm* machine, void* output) {
+		machine->resetRoundingMode();
+		for (int chain = 0; chain < RANDOMX_PROGRAM_COUNT - 1; ++chain) {
+			machine->run(machine->tempHash);
+			blake2b(machine->tempHash, sizeof(machine->tempHash), machine->getRegisterFile(), sizeof(randomx::RegisterFile), nullptr, 0);
+		}
+		machine->run(machine->tempHash);
+		machine->getFinalResult(output, RANDOMX_HASH_SIZE);
+	}
 }

src/randomx.h

@@ -240,9 +240,13 @@ RANDOMX_EXPORT void randomx_destroy_vm(randomx_vm *machine);
 RANDOMX_EXPORT void randomx_calculate_hash(randomx_vm *machine, const void *input, size_t inputSize, void *output);
 
 /**
- * Paired functions used to calculate multiple RandomX hashes more efficiently.
- * randomx_calculate_hash_first is called for the first input value.
- * randomx_calculate_hash_next will output the hash value of the previous input.
+ * Set of functions used to calculate multiple RandomX hashes more efficiently.
+ * randomx_calculate_hash_first will begin a hash calculation.
+ * randomx_calculate_hash_next  will output the hash value of the previous input
+ *                              and begin the calculation of the next hash.
+ * randomx_calculate_hash_last  will output the hash value of the previous input.
+ *
+ * WARNING: These functions may alter the floating point rounding mode of the calling thread.
  *
  * @param machine is a pointer to a randomx_vm structure. Must not be NULL.
  * @param input is a pointer to memory to be hashed. Must not be NULL.
@@ -254,6 +258,7 @@ RANDOMX_EXPORT void randomx_calculate_hash(randomx_vm *machine, const void *inpu
 */
 RANDOMX_EXPORT void randomx_calculate_hash_first(randomx_vm* machine, const void* input, size_t inputSize);
 RANDOMX_EXPORT void randomx_calculate_hash_next(randomx_vm* machine, const void* nextInput, size_t nextInputSize, void* output);
+RANDOMX_EXPORT void randomx_calculate_hash_last(randomx_vm* machine, void* output);
 
 #if defined(__cplusplus)
 }

src/tests/affinity.cpp

@@ -65,7 +65,7 @@ set_thread_affinity(std::thread::native_handle_type thread,
             (thread_policy_t)&policy, 1);
 #elif defined(_WIN32) || defined(__CYGWIN__)
     rc = SetThreadAffinityMask(reinterpret_cast<HANDLE>(thread), 1ULL << cpuid) == 0 ? -2 : 0;
-#elif !defined(__OpenBSD__) && !defined(__FreeBSD__) && !defined(__ANDROID__)
+#elif !defined(__OpenBSD__) && !defined(__FreeBSD__) && !defined(__ANDROID__) && !defined(__NetBSD__)
     cpu_set_t cs;
     CPU_ZERO(&cs);
     CPU_SET(cpuid, &cs);

src/tests/benchmark.cpp

@@ -40,9 +40,10 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "../dataset.hpp"
 #include "../blake2/endian.h"
 #include "../common.hpp"
+#include "../jit_compiler.hpp"
 #ifdef _WIN32
 #include <windows.h>
-#include <VersionHelpers.h>
+#include <versionhelpers.h>
 #endif
 #include "affinity.hpp"
 
@@ -94,6 +95,7 @@ void printUsage(const char* executable) {
 	std::cout << "  --ssse3       use optimized Argon2 for SSSE3 CPUs" << std::endl;
 	std::cout << "  --avx2        use optimized Argon2 for AVX2 CPUs" << std::endl;
 	std::cout << "  --auto        select the best options for the current CPU" << std::endl;
+	std::cout << "  --noBatch     calculate hashes one by one (default: batch)" << std::endl;
 }
 
 struct MemoryException : public std::exception {
@@ -109,11 +111,14 @@ struct DatasetAllocException : public MemoryException {
 	}
 };
 
-void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result, uint32_t noncesCount, int thread, int cpuid=-1) {
+using MineFunc = void(randomx_vm * vm, std::atomic<uint32_t> & atomicNonce, AtomicHash & result, uint32_t noncesCount, int thread, int cpuid);
+
+template<bool batch>
+void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result, uint32_t noncesCount, int thread, int cpuid = -1) {
 	if (cpuid >= 0) {
 		int rc = set_thread_affinity(cpuid);
 		if (rc) {
-			std::cerr << "Failed to set thread affinity for thread " << thread << " (error=" << rc << ")" <<  std::endl;
+			std::cerr << "Failed to set thread affinity for thread " << thread << " (error=" << rc << ")" << std::endl;
 		}
 	}
 	uint64_t hash[RANDOMX_HASH_SIZE / sizeof(uint64_t)];
@@ -122,19 +127,27 @@ void mine(randomx_vm* vm, std::atomic<uint32_t>& atomicNonce, AtomicHash& result
 	void* noncePtr = blockTemplate + 39;
 	auto nonce = atomicNonce.fetch_add(1);
 
-	store32(noncePtr, nonce);
-	randomx_calculate_hash_first(vm, blockTemplate, sizeof(blockTemplate));
+	if (batch) {
+		store32(noncePtr, nonce);
+		randomx_calculate_hash_first(vm, blockTemplate, sizeof(blockTemplate));
+	}
 
 	while (nonce < noncesCount) {
-		nonce = atomicNonce.fetch_add(1);
+		if (batch) {
+			nonce = atomicNonce.fetch_add(1);
+		}
 		store32(noncePtr, nonce);
-		randomx_calculate_hash_next(vm, blockTemplate, sizeof(blockTemplate), &hash);
+		(batch ? randomx_calculate_hash_next : randomx_calculate_hash)(vm, blockTemplate, sizeof(blockTemplate), &hash);
 		result.xorWith(hash);
+		if (!batch) {
+			nonce = atomicNonce.fetch_add(1);
+		}
 	}
 }
 
 int main(int argc, char** argv) {
-	bool softAes, miningMode, verificationMode, help, largePages, jit, secure, ssse3, avx2, autoFlags;
+	bool softAes, miningMode, verificationMode, help, largePages, jit, secure;
+	bool ssse3, avx2, autoFlags, noBatch;
 	int noncesCount, threadCount, initThreadCount;
 	uint64_t threadAffinity;
 	int32_t seedValue;
@@ -158,10 +171,11 @@ int main(int argc, char** argv) {
 	readOption("--ssse3", argc, argv, ssse3);
 	readOption("--avx2", argc, argv, avx2);
 	readOption("--auto", argc, argv, autoFlags);
+	readOption("--noBatch", argc, argv, noBatch);
 
 	store32(&seed, seedValue);
 
-	std::cout << "RandomX benchmark v1.1.7" << std::endl;
+	std::cout << "RandomX benchmark v1.1.8" << std::endl;
 
 	if (help) {
 		printUsage(argv[0]);
@@ -199,7 +213,7 @@ int main(int argc, char** argv) {
 		}
 		if (jit) {
 			flags |= RANDOMX_FLAG_JIT;
-#ifdef __OpenBSD__
+#ifdef RANDOMX_FORCE_SECURE
 			flags |= RANDOMX_FLAG_SECURE;
 #endif
 		}
@@ -211,7 +225,7 @@ int main(int argc, char** argv) {
 	if (miningMode) {
 		flags |= RANDOMX_FLAG_FULL_MEM;
 	}
-#ifndef __OpenBSD__
+#ifndef RANDOMX_FORCE_SECURE
 	if (secure) {
 		flags |= RANDOMX_FLAG_SECURE;
 	}
@@ -263,6 +277,16 @@ int main(int argc, char** argv) {
 		std::cout << " - thread affinity (" << mask_to_string(threadAffinity) << ")" << std::endl;
 	}
 
+	MineFunc* func;
+
+	if (noBatch) {
+		func = &mine<false>;
+	}
+	else {
+		func = &mine<true>;
+		std::cout << " - batch mode" << std::endl;
+	}
+
 	std::cout << "Initializing";
 	if (miningMode)
 		std::cout << " (" << initThreadCount << " thread" << (initThreadCount > 1 ? "s)" : ")");
@@ -333,14 +357,14 @@ int main(int argc, char** argv) {
 				int cpuid = -1;
 				if (threadAffinity)
 					cpuid = cpuid_from_mask(threadAffinity, i);
-				threads.push_back(std::thread(&mine, vms[i], std::ref(atomicNonce), std::ref(result), noncesCount, i, cpuid));
+				threads.push_back(std::thread(func, vms[i], std::ref(atomicNonce), std::ref(result), noncesCount, i, cpuid));
 			}
 			for (unsigned i = 0; i < threads.size(); ++i) {
 				threads[i].join();
 			}
 		}
 		else {
-			mine(vms[0], std::ref(atomicNonce), std::ref(result), noncesCount, 0);
+			func(vms[0], std::ref(atomicNonce), std::ref(result), noncesCount, 0, -1);
 		}
 
 		double elapsed = sw.getElapsed();

src/tests/tests.cpp

@@ -143,7 +143,7 @@ int main() {
 		randomx::JitCompiler jit;
 		jit.generateSuperscalarHash(cache->programs, cache->reciprocalCache);
 		jit.generateDatasetInitCode();
-#ifdef __OpenBSD__
+#ifdef RANDOMX_FORCE_SECURE
 		jit.enableExecution();
 #else
 		jit.enableAll();
@@ -954,7 +954,7 @@ int main() {
 		assert(ibc.memMask == randomx::ScratchpadL3Mask);
 	});
 
-#ifdef __OpenBSD__
+#ifdef RANDOMX_FORCE_SECURE
 	vm = randomx_create_vm(RANDOMX_FLAG_DEFAULT | RANDOMX_FLAG_SECURE, cache, nullptr);
 #else
 	vm = randomx_create_vm(RANDOMX_FLAG_DEFAULT, cache, nullptr);
@@ -1009,10 +1009,10 @@ int main() {
 		vm = nullptr;
 		cache = randomx_alloc_cache(RANDOMX_FLAG_JIT);
 		initCache("test key 000");
-#ifdef __OpenBSD__
-		vm = randomx_create_vm(RANDOMX_FLAG_DEFAULT | RANDOMX_FLAG_SECURE, cache, nullptr);
+#ifdef RANDOMX_FORCE_SECURE
+		vm = randomx_create_vm(RANDOMX_FLAG_JIT | RANDOMX_FLAG_SECURE, cache, nullptr);
 #else
-		vm = randomx_create_vm(RANDOMX_FLAG_DEFAULT, cache, nullptr);
+		vm = randomx_create_vm(RANDOMX_FLAG_JIT, cache, nullptr);
 #endif
 	}
 
@@ -1026,9 +1026,6 @@ int main() {
 
 	runTest("Hash test 2e (compiler)", RANDOMX_HAVE_COMPILER && stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), test_e);
 
-	randomx_destroy_vm(vm);
-	vm = nullptr;
-
 	auto flags = randomx_get_flags();
 
 	randomx_release_cache(cache);
@@ -1054,6 +1051,40 @@ int main() {
 		assert(cacheMemory[33554431] == 0x1f47f056d05cd99b);
 	});
 
+	if (cache != nullptr)
+		randomx_release_cache(cache);
+	cache = randomx_alloc_cache(RANDOMX_FLAG_DEFAULT);
+
+	runTest("Hash batch test", RANDOMX_HAVE_COMPILER && stringsEqual(RANDOMX_ARGON_SALT, "RandomX\x03"), []() {
+		char hash1[RANDOMX_HASH_SIZE];
+		char hash2[RANDOMX_HASH_SIZE];
+		char hash3[RANDOMX_HASH_SIZE];
+		initCache("test key 000");
+		char input1[] = "This is a test";
+		char input2[] = "Lorem ipsum dolor sit amet";
+		char input3[] = "sed do eiusmod tempor incididunt ut labore et dolore magna aliqua";
+
+		randomx_calculate_hash_first(vm, input1, sizeof(input1) - 1);
+		randomx_calculate_hash_next(vm, input2, sizeof(input2) - 1, &hash1);
+		randomx_calculate_hash_next(vm, input3, sizeof(input3) - 1, &hash2);
+		randomx_calculate_hash_last(vm, &hash3);
+
+		assert(equalsHex(hash1, "639183aae1bf4c9a35884cb46b09cad9175f04efd7684e7262a0ac1c2f0b4e3f"));
+		assert(equalsHex(hash2, "300a0adb47603dedb42228ccb2b211104f4da45af709cd7547cd049e9489c969"));
+		assert(equalsHex(hash3, "c36d4ed4191e617309867ed66a443be4075014e2b061bcdaf9ce7b721d2b77a8"));
+	});
+
+	runTest("Preserve rounding mode", RANDOMX_FREQ_CFROUND > 0, []() {
+		rx_set_rounding_mode(RoundToNearest);
+		char hash[RANDOMX_HASH_SIZE];
+		calcStringHash("test key 000", "Lorem ipsum dolor sit amet", &hash);
+		assert(equalsHex(hash, "300a0adb47603dedb42228ccb2b211104f4da45af709cd7547cd049e9489c969"));
+		assert(rx_get_rounding_mode() == RoundToNearest);
+	});
+
+	randomx_destroy_vm(vm);
+	vm = nullptr;
+
 	if (cache != nullptr)
 		randomx_release_cache(cache);
 

src/virtual_machine.hpp

@@ -54,6 +54,9 @@ public:
 	{
 		return program;
 	}
+	const uint8_t* getMemory() const {
+		return mem.memory;
+	}
 protected:
 	void initialize();
 	alignas(64) randomx::Program program;

src/virtual_memory.cpp

@@ -94,7 +94,12 @@ void* allocMemoryPages(std::size_t bytes) {
 	if (mem == nullptr)
 		throw std::runtime_error(getErrorMessage("allocMemoryPages - VirtualAlloc"));
 #else
-	mem = mmap(nullptr, bytes, PAGE_READWRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
+	#if defined(__NetBSD__)
+		#define RESERVED_FLAGS PROT_MPROTECT(PROT_EXEC)
+	#else
+		#define RESERVED_FLAGS 0
+	#endif
+	mem = mmap(nullptr, bytes, PAGE_READWRITE | RESERVED_FLAGS, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
 	if (mem == MAP_FAILED)
 		throw std::runtime_error("allocMemoryPages - mmap failed");
 #endif
@@ -141,7 +146,7 @@ void* allocLargePagesMemory(std::size_t bytes) {
 	mem = mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, VM_FLAGS_SUPERPAGE_SIZE_2MB, 0);
 #elif defined(__FreeBSD__)
 	mem = mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_ALIGNED_SUPER, -1, 0);
-#elif defined(__OpenBSD__)
+#elif defined(__OpenBSD__) || defined(__NetBSD__)
 	mem = MAP_FAILED; // OpenBSD does not support huge pages
 #else
 	mem = mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, -1, 0);

vcxproj/randomx-dll.vcxproj

@@ -54,12 +54,17 @@
   <ItemGroup>
     <ClCompile Include="..\src\aes_hash.cpp" />
     <ClCompile Include="..\src\allocator.cpp" />
+    <ClCompile Include="..\src\argon2_avx2.c">
+      <EnableEnhancedInstructionSet Condition="'$(Configuration)|$(Platform)'=='Release|x64'">AdvancedVectorExtensions2</EnableEnhancedInstructionSet>
+    </ClCompile>
     <ClCompile Include="..\src\argon2_core.c" />
     <ClCompile Include="..\src\argon2_ref.c" />
+    <ClCompile Include="..\src\argon2_ssse3.c" />
     <ClCompile Include="..\src\assembly_generator_x86.cpp" />
     <ClCompile Include="..\src\blake2\blake2b.c" />
     <ClCompile Include="..\src\blake2_generator.cpp" />
     <ClCompile Include="..\src\bytecode_machine.cpp" />
+    <ClCompile Include="..\src\cpu.cpp" />
     <ClCompile Include="..\src\dataset.cpp" />
     <ClCompile Include="..\src\instruction.cpp" />
     <ClCompile Include="..\src\instructions_portable.cpp" />

vcxproj/randomx-dll.vcxproj.filters

@@ -172,5 +172,14 @@
     <ClCompile Include="..\src\bytecode_machine.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\src\argon2_avx2.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="..\src\argon2_ssse3.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="..\src\cpu.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
 </Project>