Native load from the cache

src/InterpretedVirtualMachine.cpp

@@ -74,8 +74,9 @@ namespace RandomX {
 	}
 
 	static void print(__m128d f) {
-		uint64_t lo = *(((uint64_t*)&f) + 0);
-		uint64_t hi = *(((uint64_t*)&f) + 1);
+		uint64_t hi, lo;
+		memcpy(&lo, &f, 8);
+		memcpy(&hi, (uint8_t*)&f + 8, 8);
 		std::cout << std::hex << std::setw(16) << std::setfill('0') << hi << '-' << std::hex << std::setw(16) << std::setfill('0') << lo << std::endl;
 	}
 
@@ -417,6 +418,9 @@ namespace RandomX {
 		_mm_store_pd(&reg.e[1].lo, e[1]);
 		_mm_store_pd(&reg.e[2].lo, e[2]);
 		_mm_store_pd(&reg.e[3].lo, e[3]);
+
+		printState(r, f, e, a);
+		std::cout << "=================================================" << std::endl;
 	}
 
 	static int getConditionRegister(int(&registerUsage)[8]) {

src/blake2/endian.h

@@ -60,6 +60,12 @@ static FORCE_INLINE uint64_t load64(const void *src) {
 #endif
 }
 
+static FORCE_INLINE uint64_t load64np(const void *src) {
+	uint64_t w;
+	memcpy(&w, src, sizeof w);
+	return w;
+}
+
 static FORCE_INLINE void store32(void *dst, uint32_t w) {
 #if defined(NATIVE_LITTLE_ENDIAN)
 	memcpy(dst, &w, sizeof w);

src/dataset.cpp

@@ -57,14 +57,14 @@ namespace RandomX {
 	}
 
 	static FORCE_INLINE void mixCache(uint8_t* mixBlock, uint64_t& c0, uint64_t& c1, uint64_t& c2, uint64_t& c3, uint64_t& c4, uint64_t& c5, uint64_t& c6, uint64_t& c7) {
-		c0 ^= load64(mixBlock + 0);
-		c1 ^= load64(mixBlock + 8);
-		c2 ^= load64(mixBlock + 16);
-		c3 ^= load64(mixBlock + 24);
-		c4 ^= load64(mixBlock + 32);
-		c5 ^= load64(mixBlock + 40);
-		c6 ^= load64(mixBlock + 48);
-		c7 ^= load64(mixBlock + 56);
+		c0 ^= load64np(mixBlock + 0);
+		c1 ^= load64np(mixBlock + 8);
+		c2 ^= load64np(mixBlock + 16);
+		c3 ^= load64np(mixBlock + 24);
+		c4 ^= load64np(mixBlock + 32);
+		c5 ^= load64np(mixBlock + 40);
+		c6 ^= load64np(mixBlock + 48);
+		c7 ^= load64np(mixBlock + 56);
 	}
 
 	void initBlock(const Cache& cache, uint8_t* out, uint64_t blockNumber, unsigned iterations) {

src/main.cpp

@@ -16,7 +16,7 @@ GNU General Public License for more details.
 You should have received a copy of the GNU General Public License
 along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 */
-//#define TRACE
+#define TRACE
 #include "InterpretedVirtualMachine.hpp"
 #include "CompiledVirtualMachine.hpp"
 #include "CompiledLightVirtualMachine.hpp"
@@ -179,11 +179,21 @@ void mine(RandomX::VirtualMachine* vm, std::atomic<uint32_t>& atomicNonce, Atomi
 		for (int chain = 0; chain < RANDOMX_PROGRAM_COUNT - 1; ++chain) {
 			fillAes1Rx4<softAes>((void*)hash, sizeof(RandomX::Program), vm->getProgramBuffer());
 			vm->initialize();
+			if (RandomX::trace) {
+				std::cout << "Program " << chain << std::endl;
+				std::cout << *vm->getProgramBuffer();
+				std::cout << "=================================================" << std::endl;
+			}
 			vm->execute();
 			vm->getResult<false>(nullptr, 0, hash);
 		}
 		fillAes1Rx4<softAes>((void*)hash, sizeof(RandomX::Program), vm->getProgramBuffer());
 		vm->initialize();
+		if (RandomX::trace) {
+			std::cout << "Program 7" << std::endl;
+			std::cout << *vm->getProgramBuffer();
+			std::cout << "=================================================" << std::endl;
+		}
 		vm->execute();
 		/*if (RandomX::trace) {
 			for (int j = 0; j < RandomX::ProgramLength; ++j) {
@@ -191,6 +201,7 @@ void mine(RandomX::VirtualMachine* vm, std::atomic<uint32_t>& atomicNonce, Atomi
 				std::cout << std::hex << std::setw(16) << std::setfill('0') << res << std::endl;
 			}
 		}*/
+		dump((char*)scratchpad, RANDOMX_SCRATCHPAD_L3, "scratchpad");
 		vm->getResult<softAes>(scratchpad, RANDOMX_SCRATCHPAD_L3, hash);
 		result.xorWith(hash);
 		if (RandomX::trace) {
@@ -211,7 +222,7 @@ int main(int argc, char** argv) {
 	readOption("--mine", argc, argv, miningMode);
 	readOption("--verify", argc, argv, verificationMode);
 	readIntOption("--threads", argc, argv, threadCount, 1);
-	readIntOption("--nonces", argc, argv, programCount, 1000);
+	readIntOption("--nonces", argc, argv, programCount, 1);
 	readIntOption("--init", argc, argv, initThreadCount, 1);
 	readIntOption("--epoch", argc, argv, epoch, 0);
 	readOption("--largePages", argc, argv, largePages);