WIP

src/AssemblyGeneratorX86.hpp

@@ -20,6 +20,7 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #pragma once
 
 #include "Instruction.hpp"
+#include "configuration.h"
 #include <sstream>
 
 namespace RandomX {

src/Instruction.hpp

@@ -78,6 +78,9 @@ namespace RandomX {
 		uint32_t getImm32() const {
 			return load32(&imm32);
 		}
+		void setImm32(uint32_t val) {
+			return store32(&imm32, val);
+		}
 		const char* getName() const {
 			return names[opcode];
 		}

src/JitCompilerX86.cpp

@@ -238,12 +238,7 @@ namespace RandomX {
 		emitByte(0xc0 + readReg1);
 		memcpy(code + codePos, codeLoopLoad, loopLoadSize);
 		codePos += loopLoadSize;
-		for (unsigned i = 0; i < RANDOMX_PROGRAM_SIZE; ++i) {
-			Instruction& instr = prog(i);
-			instr.src %= RegistersCount;
-			instr.dst %= RegistersCount;
-			generateCode(instr, i);
-		}
+		generateCode(prog);
 		emit(REX_MOV_RR);
 		emitByte(0xc0 + readReg2);
 		emit(REX_XOR_EAX);

src/JitCompilerX86.hpp

@@ -52,6 +52,16 @@ namespace RandomX {
 		uint8_t* code;
 		int32_t codePos;
 
+		template<class P>
+		void generateCode(P& prog) {
+			for (unsigned i = 0; i < prog.getSize(); ++i) {
+				Instruction& instr = prog(i);
+				instr.src %= RegistersCount;
+				instr.dst %= RegistersCount;
+				generateCode(instr, i);
+			}
+		}
+
 		void generateProgramPrologue(Program&);
 		void generateProgramEpilogue(Program&);
 		int getConditionRegister();

src/LightProgramGenerator.cpp

@@ -0,0 +1,342 @@
+/*
+Copyright (c) 2019 tevador
+
+This file is part of RandomX.
+
+RandomX is free software: you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation, either version 3 of the License, or
+(at your option) any later version.
+
+RandomX is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+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/>.
+*/
+
+#include "blake2/blake2.h"
+#include "configuration.h"
+#include "Program.hpp"
+#include "blake2/endian.h";
+#include <iostream>
+
+namespace RandomX {
+
+	namespace LightInstruction {
+		constexpr int IADD_R = 0;
+		constexpr int IADD_RC = 1;
+		constexpr int ISUB_R = 2;
+		constexpr int IMUL_9C = 3;
+		constexpr int IMUL_R = 4;
+		constexpr int IMULH_R = 5;
+		constexpr int ISMULH_R = 6;
+		constexpr int IMUL_RCP = 7;
+		constexpr int IXOR_R = 8;
+		constexpr int IROR_R = 9;
+		constexpr int COND_R = 10;
+		constexpr int COUNT = 11;
+	}
+
+	const int lightInstruction[] = {
+		LightInstruction::IADD_RC,
+		LightInstruction::IADD_RC,
+		LightInstruction::ISUB_R,
+		LightInstruction::ISUB_R,
+		LightInstruction::IMUL_9C,
+		LightInstruction::IMUL_R,
+		LightInstruction::IMUL_R,
+		LightInstruction::IMUL_R,
+		LightInstruction::IMULH_R,
+		LightInstruction::ISMULH_R,
+		LightInstruction::IMUL_RCP,
+		LightInstruction::IXOR_R,
+		LightInstruction::IXOR_R,
+		LightInstruction::IROR_R,
+		LightInstruction::IROR_R,
+		LightInstruction::COND_R
+	};
+
+	namespace LightInstructionOpcode {
+		constexpr int IADD_R = 0;
+		constexpr int IADD_RC = RANDOMX_FREQ_IADD_R + RANDOMX_FREQ_IADD_M;
+		constexpr int ISUB_R = IADD_RC + RANDOMX_FREQ_IADD_RC;
+		constexpr int IMUL_9C = ISUB_R + RANDOMX_FREQ_ISUB_R + RANDOMX_FREQ_ISUB_M;
+		constexpr int IMUL_R = IMUL_9C + RANDOMX_FREQ_IMUL_9C;
+		constexpr int IMULH_R = IMUL_R + RANDOMX_FREQ_IMUL_R + RANDOMX_FREQ_IMUL_M;
+		constexpr int ISMULH_R = IMULH_R + RANDOMX_FREQ_IMULH_R + RANDOMX_FREQ_IMULH_M;
+		constexpr int IMUL_RCP = ISMULH_R + RANDOMX_FREQ_ISMULH_R + RANDOMX_FREQ_ISMULH_M;;
+		constexpr int IXOR_R = IMUL_RCP + RANDOMX_FREQ_IMUL_RCP + RANDOMX_FREQ_INEG_R;
+		constexpr int IROR_R = IXOR_R + RANDOMX_FREQ_IXOR_R + RANDOMX_FREQ_IXOR_M;
+		constexpr int COND_R = IROR_R + RANDOMX_FREQ_IROR_R + RANDOMX_FREQ_IROL_R + RANDOMX_FREQ_ISWAP_R + RANDOMX_FREQ_FSWAP_R + RANDOMX_FREQ_FADD_R + RANDOMX_FREQ_FADD_M + RANDOMX_FREQ_FSUB_R + RANDOMX_FREQ_FSUB_M + RANDOMX_FREQ_FSCAL_R + RANDOMX_FREQ_FMUL_R + RANDOMX_FREQ_FDIV_M + RANDOMX_FREQ_FSQRT_R;
+	}
+
+	const int lightInstructionOpcode[] = {
+		LightInstructionOpcode::IADD_R,
+		LightInstructionOpcode::IADD_RC,
+		LightInstructionOpcode::ISUB_R,
+		LightInstructionOpcode::IMUL_9C,
+		LightInstructionOpcode::IMUL_R,
+		LightInstructionOpcode::IMULH_R,
+		LightInstructionOpcode::ISMULH_R,
+		LightInstructionOpcode::IMUL_RCP,
+		LightInstructionOpcode::IXOR_R,
+		LightInstructionOpcode::IROR_R,
+		LightInstructionOpcode::COND_R
+	};
+
+	constexpr int ALU_COUNT_MUL = 1;
+	constexpr int ALU_COUNT = 4;
+	constexpr int LIGHT_OPCODE_BITS = 4;
+	constexpr int V4_SRC_INDEX_BITS = 3;
+	constexpr int V4_DST_INDEX_BITS = 3;
+
+	static int blakeCounter = 0;
+
+	// If we don't have enough data available, generate more
+	static FORCE_INLINE void check_data(size_t& data_index, const size_t bytes_needed, uint8_t* data, const size_t data_size)
+	{
+		if (data_index + bytes_needed > data_size)
+		{
+			std::cout << "Calling Blake " << (++blakeCounter) << std::endl;
+			blake2b(data, data_size, data, data_size, nullptr, 0);
+			data_index = 0;
+		}
+	}
+
+	void generateLightProgram(LightProgram& prog, const void* seed, int indexRegister) {
+
+		// Source: https://www.agner.org/optimize/instruction_tables.pdf
+		const int op_latency[LightInstruction::COUNT] = { 1, 2, 1, 2, 3, 5, 5, 4, 1, 2, 5 };
+
+		// Instruction latencies for theoretical ASIC implementation
+		const int asic_op_latency[LightInstruction::COUNT] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 };
+
+		// Available ALUs for each instruction
+		const int op_ALUs[LightInstruction::COUNT] = { ALU_COUNT, ALU_COUNT, ALU_COUNT, ALU_COUNT, ALU_COUNT_MUL, ALU_COUNT_MUL, ALU_COUNT_MUL, ALU_COUNT_MUL, ALU_COUNT, ALU_COUNT, ALU_COUNT };
+
+		uint8_t data[64];
+		memset(data, 0, sizeof(data));
+		memcpy(data, seed, SeedSize);
+
+		// Set data_index past the last byte in data
+		// to trigger full data update with blake hash
+		// before we start using it
+		size_t data_index = sizeof(data);
+
+		int code_size;
+
+		do {
+			uint8_t opcode;
+			uint8_t dst_index;
+			uint8_t src_index;
+			uint32_t imm32 = 0;
+
+			int latency[8];
+			int asic_latency[9];
+
+			// Tracks previous instruction and value of the source operand for registers R0-R3 throughout code execution
+			// byte 0: current value of the destination register
+			// byte 1: instruction opcode
+			// byte 2: current value of the source register
+			//
+			// Registers R4-R8 are constant and are treated as having the same value because when we do
+			// the same operation twice with two constant source registers, it can be optimized into a single operation
+			uint64_t inst_data[8] = { 0, 1, 2, 3, 4, 5, 6, 7 };
+
+			bool alu_busy[RANDOMX_LPROG_LATENCY + 1][ALU_COUNT];
+			bool is_rotation[LightInstruction::COUNT];
+			bool rotated[8];
+			int rotate_count = 0;
+
+			memset(latency, 0, sizeof(latency));
+			memset(asic_latency, 0, sizeof(asic_latency));
+			memset(alu_busy, 0, sizeof(alu_busy));
+			memset(is_rotation, 0, sizeof(is_rotation));
+			memset(rotated, 0, sizeof(rotated));
+			is_rotation[LightInstruction::IROR_R] = true;
+
+			int num_retries = 0;
+			code_size = 0;
+
+			int total_iterations = 0;
+
+			// Generate random code to achieve minimal required latency for our abstract CPU
+			// Try to get this latency for all 4 registers
+			while (((latency[0] < RANDOMX_LPROG_LATENCY) || (latency[1] < RANDOMX_LPROG_LATENCY) || (latency[2] < RANDOMX_LPROG_LATENCY) || (latency[3] < RANDOMX_LPROG_LATENCY)
+				|| (latency[4] < RANDOMX_LPROG_LATENCY) || (latency[5] < RANDOMX_LPROG_LATENCY) || (latency[6] < RANDOMX_LPROG_LATENCY) || (latency[7] < RANDOMX_LPROG_LATENCY)) && (num_retries < 64))
+			{
+				// Fail-safe to guarantee loop termination
+				++total_iterations;
+				if (total_iterations > 1024) {
+					std::cout << "total_iterations = " << total_iterations << std::endl;
+					break;
+				}
+
+				check_data(data_index, 1, data, sizeof(data));
+				const uint8_t b1 = data[data_index++];
+				int instrType = lightInstruction[b1 & ((1 << LIGHT_OPCODE_BITS) - 1)];
+
+				check_data(data_index, 1, data, sizeof(data));
+				const uint8_t b2 = data[data_index++];
+				dst_index = b2 & ((1 << V4_DST_INDEX_BITS) - 1);
+				src_index = (b2 >> (V4_DST_INDEX_BITS)) & ((1 << V4_SRC_INDEX_BITS) - 1);
+
+				const int a = dst_index;
+				int b = src_index;
+
+				// Don't do rotation with the same destination twice because it's equal to a single rotation
+				if (is_rotation[instrType] && rotated[a])
+				{
+					continue;
+				}
+
+				// Don't do the same instruction (except MUL) with the same source value twice because all other cases can be optimized:
+				//	2x IADD_RC(a, b, C) = IADD_RC(a, b*2, C1+C2)
+				//	2x ISUB_R(a, b) = ISUB_R(a, 2*b)
+				//	2x IMUL_R(a, b) = IMUL_R(a, b*b)
+				//	2x IMUL_9C(a, C) = 9 * (9 * a + C1) + C2 = 81 * a + (9 * C1 + C2)
+				//	2x IMUL_RCP(a, C) = a * (C * C)
+				//	2x IXOR_R = NOP
+				//	2x IROR_R(a, b) = IROR_R(a, 2*b)
+				if (instrType != LightInstruction::IMULH_R && instrType != LightInstruction::ISMULH_R && ((inst_data[a] & 0xFFFF00) == (instrType << 8) + ((inst_data[b] & 255) << 16)))
+				{
+					continue;
+				}
+
+				if ((instrType == LightInstruction::IADD_RC) || (instrType == LightInstruction::IMUL_9C) || (instrType == LightInstruction::IMUL_RCP) || (instrType == LightInstruction::COND_R) || ((instrType != LightInstruction::IMULH_R) && (instrType != LightInstruction::ISMULH_R) && (a == b)))
+				{
+					check_data(data_index, 4, data, sizeof(data));
+					imm32 = load32(&data[data_index++]);
+				}
+
+				// Find which ALU is available (and when) for this instruction
+				int next_latency = (latency[a] > latency[b]) ? latency[a] : latency[b];
+				int alu_index = -1;
+				while (next_latency < RANDOMX_LPROG_LATENCY)
+				{
+					for (int i = op_ALUs[instrType] - 1; i >= 0; --i)
+					{
+						if (!alu_busy[next_latency][i])
+						{
+							// ADD is implemented as two 1-cycle instructions on a real CPU, so do an additional availability check
+							if ((instrType == LightInstruction::IADD_RC || instrType == LightInstruction::IMUL_9C || instrType == LightInstruction::IMULH_R || instrType == LightInstruction::ISMULH_R) && alu_busy[next_latency + 1][i])
+							{
+								continue;
+							}
+
+							// Rotation can only start when previous rotation is finished, so do an additional availability check
+							if (is_rotation[instrType] && (next_latency < rotate_count * op_latency[instrType]))
+							{
+								continue;
+							}
+
+							alu_index = i;
+							break;
+						}
+					}
+					if (alu_index >= 0)
+					{
+						break;
+					}
+					++next_latency;
+				}
+
+				// Don't generate instructions that leave some register unchanged for more than 15 cycles
+				if (next_latency > latency[a] + 15)
+				{
+					continue;
+				}
+
+				next_latency += op_latency[instrType];
+
+				if (next_latency <= RANDOMX_LPROG_LATENCY)
+				{
+					if (is_rotation[instrType])
+					{
+						++rotate_count;
+					}
+
+					// Mark ALU as busy only for the first cycle when it starts executing the instruction because ALUs are fully pipelined
+					alu_busy[next_latency - op_latency[instrType]][alu_index] = true;
+					latency[a] = next_latency;
+
+					// ASIC is supposed to have enough ALUs to run as many independent instructions per cycle as possible, so latency calculation for ASIC is simple
+					asic_latency[a] = ((asic_latency[a] > asic_latency[b]) ? asic_latency[a] : asic_latency[b]) + asic_op_latency[instrType];
+
+					rotated[a] = is_rotation[instrType];
+
+					inst_data[a] = code_size + (instrType << 8) + ((inst_data[b] & 255) << 16);
+
+					prog(code_size).opcode = lightInstructionOpcode[instrType];
+					prog(code_size).dst = dst_index;
+					prog(code_size).src = src_index;
+					prog(code_size).setImm32(imm32);
+
+					if (instrType == LightInstruction::IADD_RC || instrType == LightInstruction::IMUL_9C || instrType == LightInstruction::IMULH_R || instrType == LightInstruction::ISMULH_R)
+					{
+						// ADD instruction is implemented as two 1-cycle instructions on a real CPU, so mark ALU as busy for the next cycle too
+						alu_busy[next_latency - op_latency[instrType] + 1][alu_index] = true;
+					}
+
+					++code_size;
+					if (code_size >= RANDOMX_LPROG_MIN_SIZE)
+					{
+						break;
+					}
+				}
+				else
+				{
+					++num_retries;
+					std::cout << "Retry " << num_retries << " with code_size = " << code_size << ", next_latency = " << next_latency << std::endl;
+				}
+			}
+
+			// ASIC has more execution resources and can extract as much parallelism from the code as possible
+			// We need to add a few more MUL and ROR instructions to achieve minimal required latency for ASIC
+			// Get this latency for at least 1 of the 4 registers
+			const int prev_code_size = code_size;
+			if ((code_size < RANDOMX_LPROG_MAX_SIZE) && (asic_latency[indexRegister] < RANDOMX_LPROG_ASIC_LATENCY))
+			{
+				int min_idx = indexRegister;
+				int max_idx = 0;
+				for (int i = 1; i < 8; ++i)
+				{
+					//if (asic_latency[i] < asic_latency[min_idx]) min_idx = i;
+					if (asic_latency[i] > asic_latency[max_idx]) max_idx = i;
+				}
+
+				const int pattern[3] = { LightInstruction::IMUL_R, LightInstruction::IROR_R, LightInstruction::IMUL_R };
+				const int instrType = pattern[(code_size - prev_code_size) % 3];
+				latency[min_idx] = latency[max_idx] + op_latency[instrType];
+				asic_latency[min_idx] = asic_latency[max_idx] + asic_op_latency[instrType];
+
+				prog(code_size).opcode = lightInstructionOpcode[instrType];
+				prog(code_size).dst = min_idx;
+				prog(code_size).src = max_idx;
+
+				++code_size;
+			}
+
+			for (int i = 0; i < 8; ++i) {
+				std::cout << "Latency " << i << " = " << latency[i] << std::endl;
+			}
+
+			std::cout << "Code size = " << code_size << std::endl;
+			std::cout << "ALUs:" << std::endl;
+			for (int i = 0; i < RANDOMX_LPROG_LATENCY + 1; ++i) {
+				for (int j = 0; j < ALU_COUNT; ++j) {
+					std::cout << (alu_busy[i][j] ? '*' : '_');
+				}
+				std::cout << std::endl;
+			}
+
+			// There is ~98.15% chance that loop condition is false, so this loop will execute only 1 iteration most of the time
+			// It never does more than 4 iterations for all block heights < 10,000,000
+		} while ((code_size < RANDOMX_LPROG_MIN_SIZE) || (code_size > RANDOMX_LPROG_MAX_SIZE));
+
+		prog.setSize(code_size);
+	}
+}

src/LightProgramGenerator.hpp

@@ -0,0 +1,24 @@
+/*
+Copyright (c) 2019 tevador
+
+This file is part of RandomX.
+
+RandomX is free software: you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation, either version 3 of the License, or
+(at your option) any later version.
+
+RandomX is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+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/>.
+*/
+
+#include "Program.hpp"
+
+namespace RandomX {
+	void generateLightProgram(LightProgram& prog, const void* seed, int indexRegister);
+}

src/Program.cpp

@@ -21,7 +21,8 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #include "hashAes1Rx4.hpp"
 
 namespace RandomX {
-	void Program::print(std::ostream& os) const {
+	template<size_t PROGRAM_SIZE>
+	void ProgramBase::print(std::ostream& os) const {
 		for (int i = 0; i < RANDOMX_PROGRAM_SIZE; ++i) {
 			auto instr = programBuffer[i];
 			os << instr;

src/Program.hpp

@@ -39,11 +39,45 @@ namespace RandomX {
 		uint64_t getEntropy(int i) {
 			return load64(&entropyBuffer[i]);
 		}
+		uint32_t getSize() {
+			return RANDOMX_PROGRAM_SIZE;
+		}
 	private:
-		void print(std::ostream&) const;
+		void print(std::ostream& os) const {
+			for (int i = 0; i < RANDOMX_PROGRAM_SIZE; ++i) {
+				auto instr = programBuffer[i];
+				os << instr;
+			}
+		}
 		uint64_t entropyBuffer[16];
 		Instruction programBuffer[RANDOMX_PROGRAM_SIZE];
 	};
 
+	class LightProgram {
+	public:
+		Instruction& operator()(int pc) {
+			return programBuffer[pc];
+		}
+		friend std::ostream& operator<<(std::ostream& os, const LightProgram& p) {
+			p.print(os);
+			return os;
+		}
+		uint32_t getSize() {
+			return size;
+		}
+		void setSize(uint32_t val) {
+			size = val;
+		}
+	private:
+		void print(std::ostream& os) const {
+			for (unsigned i = 0; i < size; ++i) {
+				auto instr = programBuffer[i];
+				os << instr;
+			}
+		}
+		Instruction programBuffer[RANDOMX_LPROG_MAX_SIZE];
+		uint32_t size;
+	};
+
 	static_assert(sizeof(Program) % 64 == 0, "Invalid size of class Program");
 }

src/configuration.h

@@ -37,6 +37,11 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 //Number of random Cache accesses per Dataset block. Minimum is 2.
 #define RANDOMX_CACHE_ACCESSES     8
 
+#define RANDOMX_LPROG_LATENCY      168
+#define RANDOMX_LPROG_ASIC_LATENCY 84
+#define RANDOMX_LPROG_MIN_SIZE     225
+#define RANDOMX_LPROG_MAX_SIZE     512
+
 //Dataset size in bytes. Must be a power of 2.
 #define RANDOMX_DATASET_SIZE       (2ULL * 1024 * 1024 * 1024)
 

src/main.cpp

@@ -36,6 +36,7 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #include "dataset.hpp"
 #include "Cache.hpp"
 #include "hashAes1Rx4.hpp"
+#include "LightProgramGenerator.hpp"
 
 const uint8_t seed[32] = { 191, 182, 222, 175, 249, 89, 134, 104, 241, 68, 191, 62, 162, 166, 61, 64, 123, 191, 227, 193, 118, 60, 188, 53, 223, 133, 175, 24, 123, 230, 55, 74 };
 
@@ -203,7 +204,7 @@ void mine(RandomX::VirtualMachine* vm, std::atomic<uint32_t>& atomicNonce, Atomi
 }
 
 int main(int argc, char** argv) {
-	bool softAes, genAsm, miningMode, verificationMode, help, largePages, async, genNative, jit;
+	bool softAes, genAsm, miningMode, verificationMode, help, largePages, async, genNative, jit, genLight;
 	int programCount, threadCount, initThreadCount, epoch;
 
 	readOption("--softAes", argc, argv, softAes);
@@ -218,6 +219,14 @@ int main(int argc, char** argv) {
 	readOption("--jit", argc, argv, jit);
 	readOption("--genNative", argc, argv, genNative);
 	readOption("--help", argc, argv, help);
+	readOption("--genLight", argc, argv, genLight);
+
+	if (genLight) {
+		RandomX::LightProgram p;
+		RandomX::generateLightProgram(p, seed, 0);
+		std::cout << p << std::endl;
+		return 0;
+	}
 
 	if (genAsm) {
 		if (softAes)

src/variant4_random_math.h

@@ -0,0 +1,441 @@
+#ifndef VARIANT4_RANDOM_MATH_H
+#define VARIANT4_RANDOM_MATH_H
+
+// Register size can be configured to either 32 bit (uint32_t) or 64 bit (uint64_t)
+typedef uint32_t v4_reg;
+
+enum V4_Settings
+{
+	// Generate code with minimal theoretical latency = 45 cycles, which is equivalent to 15 multiplications
+	TOTAL_LATENCY = 15 * 3,
+	
+	// Always generate at least 60 instructions
+	NUM_INSTRUCTIONS_MIN = 60,
+
+	// Never generate more than 70 instructions (final RET instruction doesn't count here)
+	NUM_INSTRUCTIONS_MAX = 70,
+
+	// Available ALUs for MUL
+	// Modern CPUs typically have only 1 ALU which can do multiplications
+	ALU_COUNT_MUL = 1,
+
+	// Total available ALUs
+	// Modern CPUs have 4 ALUs, but we use only 3 because random math executes together with other main loop code
+	ALU_COUNT = 3,
+};
+
+enum V4_InstructionList
+{
+	MUL,	// a*b
+	ADD,	// a+b + C, C is an unsigned 32-bit constant
+	SUB,	// a-b
+	ROR,	// rotate right "a" by "b & 31" bits
+	ROL,	// rotate left "a" by "b & 31" bits
+	XOR,	// a^b
+	RET,	// finish execution
+	V4_INSTRUCTION_COUNT = RET,
+};
+
+// V4_InstructionDefinition is used to generate code from random data
+// Every random sequence of bytes is a valid code
+//
+// There are 9 registers in total:
+// - 4 variable registers
+// - 5 constant registers initialized from loop variables
+// This is why dst_index is 2 bits
+enum V4_InstructionDefinition
+{
+	V4_OPCODE_BITS = 3,
+	V4_DST_INDEX_BITS = 2,
+	V4_SRC_INDEX_BITS = 3,
+};
+
+struct V4_Instruction
+{
+	uint8_t opcode;
+	uint8_t dst_index;
+	uint8_t src_index;
+	uint32_t C;
+};
+
+#ifndef FORCEINLINE
+#if defined(__GNUC__)
+#define FORCEINLINE __attribute__((always_inline)) inline
+#elif defined(_MSC_VER)
+#define FORCEINLINE __forceinline
+#else
+#define FORCEINLINE inline
+#endif
+#endif
+
+#ifndef UNREACHABLE_CODE
+#if defined(__GNUC__)
+#define UNREACHABLE_CODE __builtin_unreachable()
+#elif defined(_MSC_VER)
+#define UNREACHABLE_CODE __assume(false)
+#else
+#define UNREACHABLE_CODE
+#endif
+#endif
+
+// Random math interpreter's loop is fully unrolled and inlined to achieve 100% branch prediction on CPU:
+// every switch-case will point to the same destination on every iteration of Cryptonight main loop
+//
+// This is about as fast as it can get without using low-level machine code generation
+static FORCEINLINE void v4_random_math(const struct V4_Instruction* code, v4_reg* r)
+{
+	enum
+	{
+		REG_BITS = sizeof(v4_reg) * 8,
+	};
+
+#define V4_EXEC(i) \
+	{ \
+		const struct V4_Instruction* op = code + i; \
+		const v4_reg src = r[op->src_index]; \
+		v4_reg* dst = r + op->dst_index; \
+		switch (op->opcode) \
+		{ \
+		case MUL: \
+			*dst *= src; \
+			break; \
+		case ADD: \
+			*dst += src + op->C; \
+			break; \
+		case SUB: \
+			*dst -= src; \
+			break; \
+		case ROR: \
+			{ \
+				const uint32_t shift = src % REG_BITS; \
+				*dst = (*dst >> shift) | (*dst << ((REG_BITS - shift) % REG_BITS)); \
+			} \
+			break; \
+		case ROL: \
+			{ \
+				const uint32_t shift = src % REG_BITS; \
+				*dst = (*dst << shift) | (*dst >> ((REG_BITS - shift) % REG_BITS)); \
+			} \
+			break; \
+		case XOR: \
+			*dst ^= src; \
+			break; \
+		case RET: \
+			return; \
+		default: \
+			UNREACHABLE_CODE; \
+			break; \
+		} \
+	}
+
+#define V4_EXEC_10(j) \
+	V4_EXEC(j + 0) \
+	V4_EXEC(j + 1) \
+	V4_EXEC(j + 2) \
+	V4_EXEC(j + 3) \
+	V4_EXEC(j + 4) \
+	V4_EXEC(j + 5) \
+	V4_EXEC(j + 6) \
+	V4_EXEC(j + 7) \
+	V4_EXEC(j + 8) \
+	V4_EXEC(j + 9)
+
+	// Generated program can have 60 + a few more (usually 2-3) instructions to achieve required latency
+	// I've checked all block heights < 10,000,000 and here is the distribution of program sizes:
+	//
+	// 60      27960
+	// 61      105054
+	// 62      2452759
+	// 63      5115997
+	// 64      1022269
+	// 65      1109635
+	// 66      153145
+	// 67      8550
+	// 68      4529
+	// 69      102
+
+	// Unroll 70 instructions here
+	V4_EXEC_10(0);		// instructions 0-9
+	V4_EXEC_10(10);		// instructions 10-19
+	V4_EXEC_10(20);		// instructions 20-29
+	V4_EXEC_10(30);		// instructions 30-39
+	V4_EXEC_10(40);		// instructions 40-49
+	V4_EXEC_10(50);		// instructions 50-59
+	V4_EXEC_10(60);		// instructions 60-69
+
+#undef V4_EXEC_10
+#undef V4_EXEC
+}
+
+// If we don't have enough data available, generate more
+static FORCEINLINE void check_data(size_t* data_index, const size_t bytes_needed, int8_t* data, const size_t data_size)
+{
+	if (*data_index + bytes_needed > data_size)
+	{
+		hash_extra_blake(data, data_size, (char*) data);
+		*data_index = 0;
+	}
+}
+
+// Generates as many random math operations as possible with given latency and ALU restrictions
+// "code" array must have space for NUM_INSTRUCTIONS_MAX+1 instructions
+static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_t height)
+{
+	// MUL is 3 cycles, 3-way addition and rotations are 2 cycles, SUB/XOR are 1 cycle
+	// These latencies match real-life instruction latencies for Intel CPUs starting from Sandy Bridge and up to Skylake/Coffee lake
+	//
+	// AMD Ryzen has the same latencies except 1-cycle ROR/ROL, so it'll be a bit faster than Intel Sandy Bridge and newer processors
+	// Surprisingly, Intel Nehalem also has 1-cycle ROR/ROL, so it'll also be faster than Intel Sandy Bridge and newer processors
+	// AMD Bulldozer has 4 cycles latency for MUL (slower than Intel) and 1 cycle for ROR/ROL (faster than Intel), so average performance will be the same
+	// Source: https://www.agner.org/optimize/instruction_tables.pdf
+	const int op_latency[V4_INSTRUCTION_COUNT] = { 3, 2, 1, 2, 2, 1 };
+
+	// Instruction latencies for theoretical ASIC implementation
+	const int asic_op_latency[V4_INSTRUCTION_COUNT] = { 3, 1, 1, 1, 1, 1 };
+
+	// Available ALUs for each instruction
+	const int op_ALUs[V4_INSTRUCTION_COUNT] = { ALU_COUNT_MUL, ALU_COUNT, ALU_COUNT, ALU_COUNT, ALU_COUNT, ALU_COUNT };
+
+	int8_t data[32];
+	memset(data, 0, sizeof(data));
+	uint64_t tmp = SWAP64LE(height);
+	memcpy(data, &tmp, sizeof(uint64_t));
+	data[20] = -38; // change seed
+
+	// Set data_index past the last byte in data
+	// to trigger full data update with blake hash
+	// before we start using it
+	size_t data_index = sizeof(data);
+
+	int code_size;
+
+	// There is a small chance (1.8%) that register R8 won't be used in the generated program
+	// So we keep track of it and try again if it's not used
+	bool r8_used;
+	do {
+		int latency[9];
+		int asic_latency[9];
+
+		// Tracks previous instruction and value of the source operand for registers R0-R3 throughout code execution
+		// byte 0: current value of the destination register
+		// byte 1: instruction opcode
+		// byte 2: current value of the source register
+		//
+		// Registers R4-R8 are constant and are treated as having the same value because when we do
+		// the same operation twice with two constant source registers, it can be optimized into a single operation
+		uint32_t inst_data[9] = { 0, 1, 2, 3, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF, 0xFFFFFF };
+
+		bool alu_busy[TOTAL_LATENCY + 1][ALU_COUNT];
+		bool is_rotation[V4_INSTRUCTION_COUNT];
+		bool rotated[4];
+		int rotate_count = 0;
+
+		memset(latency, 0, sizeof(latency));
+		memset(asic_latency, 0, sizeof(asic_latency));
+		memset(alu_busy, 0, sizeof(alu_busy));
+		memset(is_rotation, 0, sizeof(is_rotation));
+		memset(rotated, 0, sizeof(rotated));
+		is_rotation[ROR] = true;
+		is_rotation[ROL] = true;
+
+		int num_retries = 0;
+		code_size = 0;
+
+		int total_iterations = 0;
+		r8_used = false;
+
+		// Generate random code to achieve minimal required latency for our abstract CPU
+		// Try to get this latency for all 4 registers
+		while (((latency[0] < TOTAL_LATENCY) || (latency[1] < TOTAL_LATENCY) || (latency[2] < TOTAL_LATENCY) || (latency[3] < TOTAL_LATENCY)) && (num_retries < 64))
+		{
+			// Fail-safe to guarantee loop termination
+			++total_iterations;
+			if (total_iterations > 256)
+				break;
+
+			check_data(&data_index, 1, data, sizeof(data));
+
+			const uint8_t c = ((uint8_t*)data)[data_index++];
+
+			// MUL = opcodes 0-2
+			// ADD = opcode 3
+			// SUB = opcode 4
+			// ROR/ROL = opcode 5, shift direction is selected randomly
+			// XOR = opcodes 6-7
+			uint8_t opcode = c & ((1 << V4_OPCODE_BITS) - 1);
+			if (opcode == 5)
+			{
+				check_data(&data_index, 1, data, sizeof(data));
+				opcode = (data[data_index++] >= 0) ? ROR : ROL;
+			}
+			else if (opcode >= 6)
+			{
+				opcode = XOR;
+			}
+			else
+			{
+				opcode = (opcode <= 2) ? MUL : (opcode - 2);
+			}
+
+			uint8_t dst_index = (c >> V4_OPCODE_BITS) & ((1 << V4_DST_INDEX_BITS) - 1);
+			uint8_t src_index = (c >> (V4_OPCODE_BITS + V4_DST_INDEX_BITS)) & ((1 << V4_SRC_INDEX_BITS) - 1);
+
+			const int a = dst_index;
+			int b = src_index;
+
+			// Don't do ADD/SUB/XOR with the same register
+			if (((opcode == ADD) || (opcode == SUB) || (opcode == XOR)) && (a == b))
+			{
+				// Use register R8 as source instead
+				b = 8;
+				src_index = 8;
+			}
+
+			// Don't do rotation with the same destination twice because it's equal to a single rotation
+			if (is_rotation[opcode] && rotated[a])
+			{
+				continue;
+			}
+
+			// Don't do the same instruction (except MUL) with the same source value twice because all other cases can be optimized:
+			// 2xADD(a, b, C) = ADD(a, b*2, C1+C2), same for SUB and rotations
+			// 2xXOR(a, b) = NOP
+			if ((opcode != MUL) && ((inst_data[a] & 0xFFFF00) == (opcode << 8) + ((inst_data[b] & 255) << 16)))
+			{
+				continue;
+			}
+
+			// Find which ALU is available (and when) for this instruction
+			int next_latency = (latency[a] > latency[b]) ? latency[a] : latency[b];
+			int alu_index = -1;
+			while (next_latency < TOTAL_LATENCY)
+			{
+				for (int i = op_ALUs[opcode] - 1; i >= 0; --i)
+				{
+					if (!alu_busy[next_latency][i])
+					{
+						// ADD is implemented as two 1-cycle instructions on a real CPU, so do an additional availability check
+						if ((opcode == ADD) && alu_busy[next_latency + 1][i])
+						{
+							continue;
+						}
+
+						// Rotation can only start when previous rotation is finished, so do an additional availability check
+						if (is_rotation[opcode] && (next_latency < rotate_count * op_latency[opcode]))
+						{
+							continue;
+						}
+
+						alu_index = i;
+						break;
+					}
+				}
+				if (alu_index >= 0)
+				{
+					break;
+				}
+				++next_latency;
+			}
+
+			// Don't generate instructions that leave some register unchanged for more than 7 cycles
+			if (next_latency > latency[a] + 7)
+			{
+				continue;
+			}
+
+			next_latency += op_latency[opcode];
+
+			if (next_latency <= TOTAL_LATENCY)
+			{
+				if (is_rotation[opcode])
+				{
+					++rotate_count;
+				}
+
+				// Mark ALU as busy only for the first cycle when it starts executing the instruction because ALUs are fully pipelined
+				alu_busy[next_latency - op_latency[opcode]][alu_index] = true;
+				latency[a] = next_latency;
+
+				// ASIC is supposed to have enough ALUs to run as many independent instructions per cycle as possible, so latency calculation for ASIC is simple
+				asic_latency[a] = ((asic_latency[a] > asic_latency[b]) ? asic_latency[a] : asic_latency[b]) + asic_op_latency[opcode];
+
+				rotated[a] = is_rotation[opcode];
+
+				inst_data[a] = code_size + (opcode << 8) + ((inst_data[b] & 255) << 16);
+
+				code[code_size].opcode = opcode;
+				code[code_size].dst_index = dst_index;
+				code[code_size].src_index = src_index;
+				code[code_size].C = 0;
+
+				if (src_index == 8)
+				{
+					r8_used = true;
+				}
+
+				if (opcode == ADD)
+				{
+					// ADD instruction is implemented as two 1-cycle instructions on a real CPU, so mark ALU as busy for the next cycle too
+					alu_busy[next_latency - op_latency[opcode] + 1][alu_index] = true;
+
+					// ADD instruction requires 4 more random bytes for 32-bit constant "C" in "a = a + b + C"
+					check_data(&data_index, sizeof(uint32_t), data, sizeof(data));
+					uint32_t t;
+					memcpy(&t, data + data_index, sizeof(uint32_t));
+					code[code_size].C = SWAP32LE(t);
+					data_index += sizeof(uint32_t);
+				}
+
+				++code_size;
+				if (code_size >= NUM_INSTRUCTIONS_MIN)
+				{
+					break;
+				}
+			}
+			else
+			{
+				++num_retries;
+			}
+		}
+
+		// ASIC has more execution resources and can extract as much parallelism from the code as possible
+		// We need to add a few more MUL and ROR instructions to achieve minimal required latency for ASIC
+		// Get this latency for at least 1 of the 4 registers
+		const int prev_code_size = code_size;
+		while ((code_size < NUM_INSTRUCTIONS_MAX) && (asic_latency[0] < TOTAL_LATENCY) && (asic_latency[1] < TOTAL_LATENCY) && (asic_latency[2] < TOTAL_LATENCY) && (asic_latency[3] < TOTAL_LATENCY))
+		{
+			int min_idx = 0;
+			int max_idx = 0;
+			for (int i = 1; i < 4; ++i)
+			{
+				if (asic_latency[i] < asic_latency[min_idx]) min_idx = i;
+				if (asic_latency[i] > asic_latency[max_idx]) max_idx = i;
+			}
+
+			const uint8_t pattern[3] = { ROR, MUL, MUL };
+			const uint8_t opcode = pattern[(code_size - prev_code_size) % 3];
+			latency[min_idx] = latency[max_idx] + op_latency[opcode];
+			asic_latency[min_idx] = asic_latency[max_idx] + asic_op_latency[opcode];
+
+			code[code_size].opcode = opcode;
+			code[code_size].dst_index = min_idx;
+			code[code_size].src_index = max_idx;
+			code[code_size].C = 0;
+			++code_size;
+		}
+
+	// There is ~98.15% chance that loop condition is false, so this loop will execute only 1 iteration most of the time
+	// It never does more than 4 iterations for all block heights < 10,000,000
+	}  while (!r8_used || (code_size < NUM_INSTRUCTIONS_MIN) || (code_size > NUM_INSTRUCTIONS_MAX));
+
+	// It's guaranteed that NUM_INSTRUCTIONS_MIN <= code_size <= NUM_INSTRUCTIONS_MAX here
+	// Add final instruction to stop the interpreter
+	code[code_size].opcode = RET;
+	code[code_size].dst_index = 0;
+	code[code_size].src_index = 0;
+	code[code_size].C = 0;
+
+	return code_size;
+}
+
+#endif