Cypher_Stack/RandomWOW
2
0
Fork 0
mirror of https://git.wownero.com/wownero/RandomWOW.git synced 2024-12-22 15:58:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

Refactoring

Browse Source
This commit is contained in:
tevador 2019-04-12 14:56:20 +02:00
parent d49302561f
commit 9404516dd8
6 changed files with 154 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
src/LightProgramGenerator.cpp
View File

@ -507,8 +507,16 @@ namespace RandomX {
bool selectDestination(int cycle, RegisterInfo (&registers)[8], Blake2Generator& gen) { bool selectDestination(int cycle, RegisterInfo (&registers)[8], Blake2Generator& gen) {
std::vector<int> availableRegisters; std::vector<int> availableRegisters;
//Conditions for the destination register:
// * value must be ready at the required cycle
// * cannot be the same as the source register unless the instruction allows it
// - this avoids optimizable instructions such as "xor r, r" or "sub r, r"
// * either the last instruction applied to the register or its source must be different than this instruction
// - this avoids optimizable instruction sequences such as "xor r1, r2; xor r1, r2" or "ror r, C1; ror r, C2" or "add r, C1; add r, C2"
// - it also avoids accumulation of trailing zeroes in registers due to excessive multiplication
// * register r5 cannot be the destination of the IADD_RS instruction (limitation of the x86 lea instruction)
for (unsigned i = 0; i < 8; ++i) { for (unsigned i = 0; i < 8; ++i) {
if (registers[i].latency <= cycle && (canReuse_ || i != src_) && (registers[i].lastOpGroup != opGroup_ || registers[i].lastOpPar != opGroupPar_) && (info_->getType() != SuperscalarInstructionType::IADD_RS || i != 5)) if (registers[i].latency <= cycle && (canReuse_ || i != src_) && (registers[i].lastOpGroup != opGroup_ || registers[i].lastOpPar != opGroupPar_) && (info_->getType() != SuperscalarInstructionType::IADD_RS || i != LimitedAddressRegister))
availableRegisters.push_back(i); availableRegisters.push_back(i);
} }
return selectRegister(availableRegisters, gen, dst_); return selectRegister(availableRegisters, gen, dst_);
@ -516,13 +524,15 @@ namespace RandomX {
bool selectSource(int cycle, RegisterInfo(&registers)[8], Blake2Generator& gen) { bool selectSource(int cycle, RegisterInfo(&registers)[8], Blake2Generator& gen) {
std::vector<int> availableRegisters; std::vector<int> availableRegisters;
//all registers that are ready at the cycle
for (unsigned i = 0; i < 8; ++i) { for (unsigned i = 0; i < 8; ++i) {
if (registers[i].latency <= cycle) if (registers[i].latency <= cycle)
availableRegisters.push_back(i); availableRegisters.push_back(i);
} }
//if there are only 2 available registers for IADD_RS and one of them is r5, select it as the source because it cannot be the destination
if (availableRegisters.size() == 2 && info_->getType() == SuperscalarInstructionType::IADD_RS) { if (availableRegisters.size() == 2 && info_->getType() == SuperscalarInstructionType::IADD_RS) {
if (availableRegisters[0] == 5 || availableRegisters[1] == 5) { if (availableRegisters[0] == LimitedAddressRegister || availableRegisters[1] == LimitedAddressRegister) {
opGroupPar_ = src_ = 5; opGroupPar_ = src_ = LimitedAddressRegister;
return true; return true;
} }
} }
@ -656,7 +666,7 @@ namespace RandomX {
return -1; return -1;
} }
double generateLightProg2(LightProgram& prog, Blake2Generator& gen) { double generateSuperscalar(LightProgram& prog, Blake2Generator& gen) {
ExecutionPort::type portBusy[CYCLE_MAP_SIZE][3]; ExecutionPort::type portBusy[CYCLE_MAP_SIZE][3];
memset(portBusy, 0, sizeof(portBusy)); memset(portBusy, 0, sizeof(portBusy));
@ -674,6 +684,7 @@ namespace RandomX {
int programSize = 0; int programSize = 0;
int mulCount = 0; int mulCount = 0;
int decodeCycle; int decodeCycle;
int throwAwayCount = 0;
//decode instructions for RANDOMX_SUPERSCALAR_LATENCY cycles or until an execution port is saturated. //decode instructions for RANDOMX_SUPERSCALAR_LATENCY cycles or until an execution port is saturated.
//Each decode cycle decodes 16 bytes of x86 code. //Each decode cycle decodes 16 bytes of x86 code.
@ -722,12 +733,20 @@ namespace RandomX {
} }
//if no register was found, throw the instruction away and try another one //if no register was found, throw the instruction away and try another one
if (forward == LOOK_FORWARD_CYCLES) { if (forward == LOOK_FORWARD_CYCLES) {
if (throwAwayCount < MAX_THROWAWAY_COUNT) {
throwAwayCount++;
macroOpIndex = currentInstruction.getInfo().getSize(); macroOpIndex = currentInstruction.getInfo().getSize();
if (TRACE) std::cout << "; THROW away " << currentInstruction.getInfo().getName() << std::endl; if (TRACE) std::cout << "; THROW away " << currentInstruction.getInfo().getName() << std::endl;
continue; continue;
} }
//abort this decode buffer
/*if (TRACE)*/ std::cout << "Aborting at cycle " << cycle << " with decode buffer " << decodeBuffer->getName() << " - source registers not available" << std::endl;
currentInstruction = LightInstruction::Null;
break;
}
if (TRACE) std::cout << "; src = r" << currentInstruction.getSource() << std::endl; if (TRACE) std::cout << "; src = r" << currentInstruction.getSource() << std::endl;
} }
throwAwayCount = 0;
//find a destination register that will be ready when this instruction executes //find a destination register that will be ready when this instruction executes
if (macroOpIndex == currentInstruction.getInfo().getDstOp()) { if (macroOpIndex == currentInstruction.getInfo().getDstOp()) {
int forward; int forward;
@ -737,12 +756,20 @@ namespace RandomX {
++cycle; ++cycle;
} }
if (forward == LOOK_FORWARD_CYCLES) { //throw instruction away if (forward == LOOK_FORWARD_CYCLES) { //throw instruction away
if (throwAwayCount < MAX_THROWAWAY_COUNT) {
throwAwayCount++;
macroOpIndex = currentInstruction.getInfo().getSize(); macroOpIndex = currentInstruction.getInfo().getSize();
if (TRACE) std::cout << "; THROW away " << currentInstruction.getInfo().getName() << std::endl; if (TRACE) std::cout << "; THROW away " << currentInstruction.getInfo().getName() << std::endl;
continue; continue;
} }
//abort this decode buffer
/*if (TRACE)*/ std::cout << "Aborting at cycle " << cycle << " with decode buffer " << decodeBuffer->getName() << " - destination registers not available" << std::endl;
currentInstruction = LightInstruction::Null;
break;
}
if (TRACE) std::cout << "; dst = r" << currentInstruction.getDestination() << std::endl; if (TRACE) std::cout << "; dst = r" << currentInstruction.getDestination() << std::endl;
} }
throwAwayCount = 0;
//recalculate when the instruction can be scheduled for execution based on operand availability //recalculate when the instruction can be scheduled for execution based on operand availability
scheduleCycle = scheduleMop<true>(mop, portBusy, scheduleCycle, scheduleCycle); scheduleCycle = scheduleMop<true>(mop, portBusy, scheduleCycle, scheduleCycle);

2
src/LightProgramGenerator.hpp
View File

@ -54,5 +54,5 @@ namespace RandomX {
void checkData(const size_t); void checkData(const size_t);
}; };
double generateLightProg2(LightProgram& prog, Blake2Generator& gen); double generateSuperscalar(LightProgram& prog, Blake2Generator& gen);
} }

1
src/common.hpp
View File

@ -95,6 +95,7 @@ namespace RandomX {
constexpr int ScratchpadL3Mask = (ScratchpadL3 - 1) * 8; constexpr int ScratchpadL3Mask = (ScratchpadL3 - 1) * 8;
constexpr int ScratchpadL3Mask64 = (ScratchpadL3 / 8 - 1) * 64; constexpr int ScratchpadL3Mask64 = (ScratchpadL3 / 8 - 1) * 64;
constexpr int RegistersCount = 8; constexpr int RegistersCount = 8;
constexpr int LimitedAddressRegister = 5; //x86 r13 register
struct Cache { struct Cache {
uint8_t* memory; uint8_t* memory;

6
src/main.cpp
View File

@ -228,11 +228,11 @@ int main(int argc, char** argv) {
if (genSuperscalar) { if (genSuperscalar) {
RandomX::LightProgram p; RandomX::LightProgram p;
RandomX::Blake2Generator gen(seed, programCount); RandomX::Blake2Generator gen(seed, programCount);
RandomX::generateLightProg2(p, gen); RandomX::generateSuperscalar(p, gen);
RandomX::AssemblyGeneratorX86 asmX86; RandomX::AssemblyGeneratorX86 asmX86;
asmX86.generateAsm(p); asmX86.generateAsm(p);
//std::ofstream file("lightProg2.asm"); //std::ofstream file("lightProg2.asm");
asmX86.printCode(std::cout); //asmX86.printCode(std::cout);
return 0; return 0;
} }
@ -288,7 +288,7 @@ int main(int argc, char** argv) {
if (!legacy) { if (!legacy) {
RandomX::Blake2Generator gen(seed, programCount); RandomX::Blake2Generator gen(seed, programCount);
for (int i = 0; i < RANDOMX_CACHE_ACCESSES; ++i) { for (int i = 0; i < RANDOMX_CACHE_ACCESSES; ++i) {
RandomX::generateLightProg2(programs[i], gen); RandomX::generateSuperscalar(programs[i], gen);
} }
} }
if (!miningMode) { if (!miningMode) {

29
vcxproj/randomx.vcxproj
View File

@ -149,6 +149,35 @@
<MASM Include="..\src\JitCompilerX86-static.asm" /> <MASM Include="..\src\JitCompilerX86-static.asm" />
<MASM Include="..\src\squareHash.asm" /> <MASM Include="..\src\squareHash.asm" />
</ItemGroup> </ItemGroup>
<ItemGroup>
<ClInclude Include="..\src\argon2.h" />
<ClInclude Include="..\src\argon2_core.h" />
<ClInclude Include="..\src\AssemblyGeneratorX86.hpp" />
<ClInclude Include="..\src\Cache.hpp" />
<ClInclude Include="..\src\catch.hpp" />
<ClInclude Include="..\src\common.hpp" />
<ClInclude Include="..\src\CompiledLightVirtualMachine.hpp" />
<ClInclude Include="..\src\CompiledVirtualMachine.hpp" />
<ClInclude Include="..\src\configuration.h" />
<ClInclude Include="..\src\dataset.hpp" />
<ClInclude Include="..\src\hashAes1Rx4.hpp" />
<ClInclude Include="..\src\Instruction.hpp" />
<ClInclude Include="..\src\instructionWeights.hpp" />
<ClInclude Include="..\src\InterpretedVirtualMachine.hpp" />
<ClInclude Include="..\src\intrinPortable.h" />
<ClInclude Include="..\src\JitCompilerX86-static.hpp" />
<ClInclude Include="..\src\JitCompilerX86.hpp" />
<ClInclude Include="..\src\LightClientAsyncWorker.hpp" />
<ClInclude Include="..\src\LightProgramGenerator.hpp" />
<ClInclude Include="..\src\Program.hpp" />
<ClInclude Include="..\src\reciprocal.h" />
<ClInclude Include="..\src\softAes.h" />
<ClInclude Include="..\src\squareHash.h" />
<ClInclude Include="..\src\Stopwatch.hpp" />
<ClInclude Include="..\src\variant4_random_math.h" />
<ClInclude Include="..\src\VirtualMachine.hpp" />
<ClInclude Include="..\src\virtualMemory.hpp" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" /> <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets"> <ImportGroup Label="ExtensionTargets">
<Import Project="$(VCTargetsPath)\BuildCustomizations\masm.targets" /> <Import Project="$(VCTargetsPath)\BuildCustomizations\masm.targets" />

83
vcxproj/randomx.vcxproj.filters
View File

@ -84,4 +84,87 @@
<Filter>Source Files</Filter> <Filter>Source Files</Filter>
</MASM> </MASM>
</ItemGroup> </ItemGroup>
<ItemGroup>
<ClInclude Include="..\src\argon2.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\argon2_core.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\AssemblyGeneratorX86.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\Cache.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\catch.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\common.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\CompiledLightVirtualMachine.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\CompiledVirtualMachine.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\configuration.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\dataset.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\hashAes1Rx4.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\Instruction.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\instructionWeights.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\InterpretedVirtualMachine.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\intrinPortable.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\JitCompilerX86.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\JitCompilerX86-static.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\LightClientAsyncWorker.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\LightProgramGenerator.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\Program.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\reciprocal.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\softAes.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\squareHash.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\Stopwatch.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\variant4_random_math.h">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\VirtualMachine.hpp">
<Filter>Header Files</Filter>
</ClInclude>
<ClInclude Include="..\src\virtualMemory.hpp">
<Filter>Header Files</Filter>
</ClInclude>
</ItemGroup>
</Project> </Project>