Reworked addition instructions

Some bug fixes
This commit is contained in:
tevador 2019-04-03 14:06:59 +02:00
parent 2aaec84931
commit 690707ef49
2 changed files with 107 additions and 352 deletions

View file

@ -30,22 +30,20 @@ along with RandomX. If not, see<http://www.gnu.org/licenses/>.
namespace RandomX {
// Intel Ivy Bridge reference
namespace LightInstructionType { //uOPs (decode) execution ports latency code size
constexpr int IADD_R = 0; //1 p015 1 3
constexpr int IADD_C = 1; //1 p015 1 7
constexpr int IADD_RC = 2; //1 p1 3 8
constexpr int ISUB_R = 3; //1 p015 1 3
constexpr int IMUL_9C = 4; //1 p1 3 8
constexpr int IMUL_R = 5; //1 p1 3 4
constexpr int IMUL_C = 6; //1 p1 3 7
constexpr int IMULH_R = 7; //1+2+1 0+(p1,p5)+0 3 3+3+3
constexpr int ISMULH_R = 8; //1+2+1 0+(p1,p5)+0 3 3+3+3
constexpr int IMUL_RCP = 9; //1+1 p015+p1 4 10+4
constexpr int IXOR_R = 10; //1 p015 1 3
constexpr int IXOR_C = 11; //1 p015 1 7
constexpr int IROR_R = 12; //1+2 0+(p0,p5) 1 3+3
constexpr int IROR_C = 13; //1 p05 1 4
constexpr int COND_R = 14; //1+1+1+1+1+1 p015+p5+0+p015+p05+p015 3 7+13+3+7+3+3
constexpr int COUNT = 15;
constexpr int IADD_RS = 0; //1 p01 1 4
constexpr int ISUB_R = 1; //1 p015 1 3
constexpr int ISUB_C = 2; //1 p015 3 7
constexpr int IMUL_R = 3; //1 p1 3 4
constexpr int IMUL_C = 4; //1 p1 3 7
constexpr int IMULH_R = 5; //1+2+1 0+(p1,p5)+0 3 3+3+3
constexpr int ISMULH_R = 6; //1+2+1 0+(p1,p5)+0 3 3+3+3
constexpr int IMUL_RCP = 7; //1+1 p015+p1 4 10+4
constexpr int IXOR_R = 8; //1 p015 1 3
constexpr int IXOR_C = 9; //1 p015 1 7
constexpr int IROR_R = 10; //1+2 0+(p0,p5) 1 3+3
constexpr int IROR_C = 11; //1 p05 1 4
constexpr int COND_R = 12; //1+1+1+1+1+1 p015+p5+0+p015+p05+p015 3 7+13+3+7+3+3
constexpr int COUNT = 13;
}
namespace LightInstructionOpcode {
@ -62,8 +60,8 @@ namespace RandomX {
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;
}
static bool isMul(uint8_t opcode) {
return opcode == LightInstructionOpcode::IMUL_R || opcode == LightInstructionOpcode::IMULH_R || opcode == LightInstructionOpcode::ISMULH_R || opcode == LightInstructionOpcode::IMUL_RCP;
static bool isMul(int type) {
return type == LightInstructionType::IMUL_R || type == LightInstructionType::IMUL_C || type == LightInstructionType::IMULH_R || type == LightInstructionType::ISMULH_R || type == LightInstructionType::IMUL_RCP;
}
const int lightInstructionOpcode[] = {
@ -84,33 +82,15 @@ namespace RandomX {
LightInstructionOpcode::COND_R
};
const int lightInstruction[] = {
LightInstructionType::IADD_R,
LightInstructionType::IADD_C,
LightInstructionType::IADD_RC,
LightInstructionType::ISUB_R,
LightInstructionType::IMUL_9C,
LightInstructionType::IMUL_R,
LightInstructionType::IMUL_R,
LightInstructionType::IMUL_C,
LightInstructionType::IMULH_R,
LightInstructionType::ISMULH_R,
LightInstructionType::IMUL_RCP,
LightInstructionType::IXOR_R,
LightInstructionType::IXOR_C,
LightInstructionType::IROR_R,
LightInstructionType::IROR_C,
LightInstructionType::COND_R
};
namespace ExecutionPort {
using type = int;
constexpr type Null = 0;
constexpr type P0 = 1;
constexpr type P1 = 2;
constexpr type P5 = 3;
constexpr type P05 = 4;
constexpr type P015 = 5;
constexpr type P01 = 4;
constexpr type P05 = 5;
constexpr type P015 = 6;
}
class Blake2Generator {
@ -210,6 +190,7 @@ namespace RandomX {
static const MacroOp Add_ri;
static const MacroOp Lea_sib;
static const MacroOp Sub_rr;
static const MacroOp Sub_ri;
static const MacroOp Imul_rr;
static const MacroOp Imul_rri;
static const MacroOp Imul_r;
@ -238,8 +219,9 @@ namespace RandomX {
const MacroOp MacroOp::Add_rr = MacroOp("add r,r", 3, 1, ExecutionPort::P015);
const MacroOp MacroOp::Add_ri = MacroOp("add r,i", 7, 1, ExecutionPort::P015);
const MacroOp MacroOp::Lea_sib = MacroOp("lea r,m", 8, 3, ExecutionPort::P1);
const MacroOp MacroOp::Lea_sib = MacroOp("lea r,r+r*s", 4, 1, ExecutionPort::P01);
const MacroOp MacroOp::Sub_rr = MacroOp("sub r,r", 3, 1, ExecutionPort::P015);
const MacroOp MacroOp::Sub_ri = MacroOp("sub r,i", 7, 1, ExecutionPort::P015);
const MacroOp MacroOp::Imul_rr = MacroOp("imul r,r", 4, 3, ExecutionPort::P1);
const MacroOp MacroOp::Imul_rri = MacroOp("imul r,r,i", 7, 3, ExecutionPort::P1);
const MacroOp MacroOp::Imul_r = MacroOp("imul r", 3, 3, ExecutionPort::P1, ExecutionPort::P5);
@ -253,7 +235,7 @@ namespace RandomX {
const MacroOp MacroOp::Xor_self = MacroOp("xor rcx,rcx", 3);
const MacroOp MacroOp::Cmp_ri = MacroOp("cmp r,i", 7, 1, ExecutionPort::P015);
const MacroOp MacroOp::Setcc_r = MacroOp("setcc cl", 3, 1, ExecutionPort::P05);
const MacroOp MacroOp::TestJmp_fused = MacroOp("testjmp r,i", 13, 0, ExecutionPort::P5);
const MacroOp MacroOp::TestJmp_fused = MacroOp("testjz r,i", 13, 0, ExecutionPort::P5);
const MacroOp IMULH_R_ops_array[] = { MacroOp::Mov_rr, MacroOp::Mul_r, MacroOp::Mov_rr };
const MacroOp ISMULH_R_ops_array[] = { MacroOp::Mov_rr, MacroOp::Imul_r, MacroOp::Mov_rr };
@ -315,11 +297,9 @@ namespace RandomX {
int getSrcOp() const {
return srcOp_;
}
static const LightInstructionInfo IADD_R;
static const LightInstructionInfo IADD_C;
static const LightInstructionInfo IADD_RC;
static const LightInstructionInfo IADD_RS;
static const LightInstructionInfo ISUB_R;
static const LightInstructionInfo IMUL_9C;
static const LightInstructionInfo ISUB_C;
static const LightInstructionInfo IMUL_R;
static const LightInstructionInfo IMUL_C;
static const LightInstructionInfo IMULH_R;
@ -344,11 +324,9 @@ namespace RandomX {
: name_(name), type_(-1), latency_(0) {}
};
const LightInstructionInfo LightInstructionInfo::IADD_R = LightInstructionInfo("IADD_R", LightInstructionType::IADD_R, MacroOp::Add_rr, 0);
const LightInstructionInfo LightInstructionInfo::IADD_C = LightInstructionInfo("IADD_C", LightInstructionType::IADD_C, MacroOp::Add_ri, -1);
const LightInstructionInfo LightInstructionInfo::IADD_RC = LightInstructionInfo("IADD_RC", LightInstructionType::IADD_RC, MacroOp::Lea_sib, 0);
const LightInstructionInfo LightInstructionInfo::IADD_RS = LightInstructionInfo("IADD_RS", LightInstructionType::IADD_RS, MacroOp::Lea_sib, 0);
const LightInstructionInfo LightInstructionInfo::ISUB_R = LightInstructionInfo("ISUB_R", LightInstructionType::ISUB_R, MacroOp::Sub_rr, 0);
const LightInstructionInfo LightInstructionInfo::IMUL_9C = LightInstructionInfo("IMUL_9C", LightInstructionType::IMUL_9C, MacroOp::Lea_sib, -1);
const LightInstructionInfo LightInstructionInfo::ISUB_C = LightInstructionInfo("ISUB_C", LightInstructionType::ISUB_C, MacroOp::Sub_ri, -1);
const LightInstructionInfo LightInstructionInfo::IMUL_R = LightInstructionInfo("IMUL_R", LightInstructionType::IMUL_R, MacroOp::Imul_rr, 0);
const LightInstructionInfo LightInstructionInfo::IMUL_C = LightInstructionInfo("IMUL_C", LightInstructionType::IMUL_C, MacroOp::Imul_rri, -1);
const LightInstructionInfo LightInstructionInfo::IMULH_R = LightInstructionInfo("IMULH_R", LightInstructionType::IMULH_R, IMULH_R_ops_array, 1, 0, 1);
@ -364,7 +342,6 @@ namespace RandomX {
const int buffer0[] = { 3, 3, 10 };
const int buffer1[] = { 7, 3, 3, 3 };
const int buffer2[] = { 3, 3, 3, 7 };
const int buffer3[] = { 4, 8, 4 };
const int buffer4[] = { 4, 4, 4, 4 };
const int buffer5[] = { 3, 7, 3, 3 };
const int buffer6[] = { 3, 3, 7, 3 };
@ -390,18 +367,15 @@ namespace RandomX {
}
const DecoderBuffer& fetchNext(int prevType, Blake2Generator& gen) {
if (prevType == LightInstructionType::IMULH_R || prevType == LightInstructionType::ISMULH_R)
return decodeBuffers[0];
return decodeBuffer3310; //2-1-1 decode
if (index_ == 0) {
if ((gen.getByte() % 2) == 0)
return decodeBuffers[3];
else
return decodeBuffers[4];
return decodeBuffer4444; //IMUL_RCP end
}
if (index_ == 2) {
return decodeBuffers[7];
return decodeBuffer133; //COND_R middle
}
if (index_ == 7) {
return decodeBuffers[1];
return decodeBuffer7333; //COND_R end
}
return fetchNextDefault(gen);
}
@ -411,36 +385,49 @@ namespace RandomX {
const int* counts_;
int opsCount_;
DecoderBuffer() : index_(-1) {}
static const DecoderBuffer decodeBuffers[8];
static const DecoderBuffer decodeBuffer3310;
static const DecoderBuffer decodeBuffer7333;
static const DecoderBuffer decodeBuffer3337;
static const DecoderBuffer decodeBuffer4444;
static const DecoderBuffer decodeBuffer3733;
static const DecoderBuffer decodeBuffer3373;
static const DecoderBuffer decodeBuffer133;
static const DecoderBuffer* decodeBuffers[7];
const DecoderBuffer& fetchNextDefault(Blake2Generator& gen) {
int select;
do {
select = gen.getByte() & 7;
} while (select == 7);
return decodeBuffers[select];
return *decodeBuffers[select];
}
};
const DecoderBuffer DecoderBuffer::decodeBuffers[8] = {
DecoderBuffer("3,3,10", 0, buffer0),
DecoderBuffer("7,3,3,3", 1, buffer1),
DecoderBuffer("3,3,3,7", 2, buffer2),
DecoderBuffer("4,8,4", 3, buffer3),
DecoderBuffer("4,4,4,4", 4, buffer4),
DecoderBuffer("3,7,3,3", 5, buffer5),
DecoderBuffer("3,3,7,3", 6, buffer6),
DecoderBuffer("13,3", 7, buffer7),
const DecoderBuffer DecoderBuffer::decodeBuffer3310 = DecoderBuffer("3,3,10", 0, buffer0);
const DecoderBuffer DecoderBuffer::decodeBuffer7333 = DecoderBuffer("7,3,3,3", 1, buffer1);
const DecoderBuffer DecoderBuffer::decodeBuffer3337 = DecoderBuffer("3,3,3,7", 2, buffer2);
const DecoderBuffer DecoderBuffer::decodeBuffer4444 = DecoderBuffer("4,4,4,4", 4, buffer4);
const DecoderBuffer DecoderBuffer::decodeBuffer3733 = DecoderBuffer("3,7,3,3", 5, buffer5);
const DecoderBuffer DecoderBuffer::decodeBuffer3373 = DecoderBuffer("3,3,7,3", 6, buffer6);
const DecoderBuffer DecoderBuffer::decodeBuffer133 = DecoderBuffer("13,3", 7, buffer7);
const DecoderBuffer* DecoderBuffer::decodeBuffers[7] = {
&DecoderBuffer::decodeBuffer3310,
&DecoderBuffer::decodeBuffer7333,
&DecoderBuffer::decodeBuffer3337,
&DecoderBuffer::decodeBuffer4444,
&DecoderBuffer::decodeBuffer4444,
&DecoderBuffer::decodeBuffer3733,
&DecoderBuffer::decodeBuffer3373,
};
DecoderBuffer DecoderBuffer::Default = DecoderBuffer();
const LightInstructionInfo* slot_3[] = { &LightInstructionInfo::IADD_R, &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IADD_R };
const LightInstructionInfo* slot_3L[] = { &LightInstructionInfo::IADD_R, &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IMULH_R, &LightInstructionInfo::ISMULH_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IMULH_R, &LightInstructionInfo::ISMULH_R };
const LightInstructionInfo* slot_3F[] = { &LightInstructionInfo::IADD_R, &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IROR_R };
const LightInstructionInfo* slot_4[] = { &LightInstructionInfo::IMUL_R, &LightInstructionInfo::IROR_C };
const LightInstructionInfo* slot_7[] = { &LightInstructionInfo::IADD_C, &LightInstructionInfo::IMUL_C, &LightInstructionInfo::IXOR_C, &LightInstructionInfo::IADD_C };
const LightInstructionInfo* slot_3[] = { &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R };
const LightInstructionInfo* slot_3L[] = { &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IMULH_R, &LightInstructionInfo::ISMULH_R };
const LightInstructionInfo* slot_3C[] = { &LightInstructionInfo::ISUB_R, &LightInstructionInfo::IXOR_R, &LightInstructionInfo::IROR_R, &LightInstructionInfo::IXOR_R };
const LightInstructionInfo* slot_4[] = { &LightInstructionInfo::IMUL_R, &LightInstructionInfo::IROR_C, &LightInstructionInfo::IADD_RS, &LightInstructionInfo::IMUL_R };
const LightInstructionInfo* slot_7[] = { &LightInstructionInfo::ISUB_C, &LightInstructionInfo::IMUL_C, &LightInstructionInfo::IXOR_C, &LightInstructionInfo::ISUB_C };
const LightInstructionInfo* slot_7L = &LightInstructionInfo::COND_R;
const LightInstructionInfo* slot_8[] = { &LightInstructionInfo::IADD_RC, &LightInstructionInfo::IMUL_9C };
const LightInstructionInfo* slot_10 = &LightInstructionInfo::IMUL_RCP;
static bool selectRegister(std::vector<int>& availableRegisters, Blake2Generator& gen, int& reg) {
@ -469,21 +456,21 @@ namespace RandomX {
instr.setImm32(imm32_);
}
static LightInstruction createForSlot(Blake2Generator& gen, int slotSize, bool isLast = false, bool isFirst = false) {
static LightInstruction createForSlot(Blake2Generator& gen, int slotSize, bool isLast = false, bool complex = false) {
switch (slotSize)
{
case 3:
if (isLast) {
return create(slot_3L[gen.getByte() & 7], gen);
return create(slot_3L[gen.getByte() & 3], gen);
}
else if (isFirst) {
return create(slot_3F[gen.getByte() & 3], gen);
else if (complex) {
return create(slot_3C[gen.getByte() & 3], gen);
}
else {
return create(slot_3[gen.getByte() & 3], gen);
return create(slot_3[gen.getByte() & 1], gen);
}
case 4:
return create(slot_4[gen.getByte() & 1], gen);
return create(slot_4[gen.getByte() & 3], gen);
case 7:
if (isLast) {
return create(slot_7L, gen);
@ -491,12 +478,10 @@ namespace RandomX {
else {
return create(slot_7[gen.getByte() & 3], gen);
}
case 8:
return create(slot_8[gen.getByte() & 1], gen);
case 10:
return create(slot_10, gen);
default:
break;
throw std::runtime_error("Invalid slot");
}
}
@ -504,38 +489,24 @@ namespace RandomX {
LightInstruction li(info);
switch (info->getType())
{
case LightInstructionType::IADD_R: {
li.mod_ = 0;
case LightInstructionType::IADD_RS: {
li.mod_ = gen.getByte();
li.imm32_ = 0;
li.opGroup_ = LightInstructionType::IADD_R;
li.groupParIsSource_ = true;
} break;
case LightInstructionType::IADD_C: {
li.mod_ = 0;
li.imm32_ = gen.getInt32();
li.opGroup_ = LightInstructionType::IADD_R;
li.groupParIsSource_ = true;
} break;
case LightInstructionType::IADD_RC: {
li.mod_ = 0;
li.imm32_ = gen.getInt32();
li.opGroup_ = LightInstructionType::IADD_R;
li.opGroup_ = LightInstructionType::IADD_RS;
li.groupParIsSource_ = true;
} break;
case LightInstructionType::ISUB_R: {
li.mod_ = 0;
li.imm32_ = 0;
li.opGroup_ = LightInstructionType::IADD_R;
li.opGroup_ = LightInstructionType::IADD_RS;
li.groupParIsSource_ = true;
} break;
case LightInstructionType::IMUL_9C: {
case LightInstructionType::ISUB_C: {
li.mod_ = 0;
li.imm32_ = gen.getInt32();
li.opGroup_ = LightInstructionType::IMUL_C;
li.opGroup_ = LightInstructionType::ISUB_C;
li.opGroupPar_ = -1;
} break;
@ -721,7 +692,7 @@ namespace RandomX {
}
}
}
else if (mop.getUop1() == ExecutionPort::P05) {
else if (mop.getUop1() == ExecutionPort::P01) {
for (; cycle < CYCLE_MAP_SIZE; ++cycle) {
if (!portBusy[cycle][0]) {
if (commit) {
@ -730,6 +701,17 @@ namespace RandomX {
}
return cycle;
}
if (!portBusy[cycle][1]) {
if (commit) {
if (TRACE) std::cout << "; P1 at cycle " << cycle << std::endl;
portBusy[cycle][1] = mop.getUop1();
}
return cycle;
}
}
}
else if (mop.getUop1() == ExecutionPort::P05) {
for (; cycle < CYCLE_MAP_SIZE; ++cycle) {
if (!portBusy[cycle][2]) {
if (commit) {
if (TRACE) std::cout << "; P2 at cycle " << cycle << std::endl;
@ -737,17 +719,17 @@ namespace RandomX {
}
return cycle;
}
if (!portBusy[cycle][0]) {
if (commit) {
if (TRACE) std::cout << "; P0 at cycle " << cycle << std::endl;
portBusy[cycle][0] = mop.getUop1();
}
return cycle;
}
}
}
else {
for (; cycle < CYCLE_MAP_SIZE; ++cycle) {
if (!portBusy[cycle][0]) {
if (commit) {
if (TRACE) std::cout << "; P0 at cycle " << cycle << std::endl;
portBusy[cycle][0] = mop.getUop1();
}
return cycle;
}
if (!portBusy[cycle][2]) {
if (commit) {
if (TRACE) std::cout << "; P2 at cycle " << cycle << std::endl;
@ -755,6 +737,13 @@ namespace RandomX {
}
return cycle;
}
if (!portBusy[cycle][0]) {
if (commit) {
if (TRACE) std::cout << "; P0 at cycle " << cycle << std::endl;
portBusy[cycle][0] = mop.getUop1();
}
return cycle;
}
if (!portBusy[cycle][1]) {
if (commit) {
if (TRACE) std::cout << "; P1 at cycle " << cycle << std::endl;
@ -813,6 +802,7 @@ namespace RandomX {
bool portsSaturated = false;
int outIndex = 0;
int attempts = 0;
int mulCount = 0;
constexpr int MAX_ATTEMPTS = 4;
while(!portsSaturated) {
@ -872,6 +862,7 @@ namespace RandomX {
}
if (TRACE) std::cout << "; dst = r" << currentInstruction.getDestination() << std::endl;
}
scheduleCycle = scheduleUop<true>(mop, portBusy, scheduleCycle, scheduleCycle);
depCycle = scheduleCycle + mop.getLatency();
if (instrIndex == currentInstruction.getInfo().getResultOp()) {
int dst = currentInstruction.getDestination();
@ -882,7 +873,6 @@ namespace RandomX {
ri.lastOpPar = currentInstruction.getGroupPar();
if (TRACE) std::cout << "; RETIRED at cycle " << retireCycle << std::endl;
}
scheduleUop<true>(mop, portBusy, scheduleCycle, scheduleCycle);
codeSize += mop.getSize();
mopIndex++;
instrIndex++;
@ -893,13 +883,14 @@ namespace RandomX {
cycle = topCycle;
if (instrIndex >= currentInstruction.getInfo().getSize()) {
currentInstruction.toInstr(prog(outIndex++));
mulCount += isMul(currentInstruction.getType());
}
}
++cycle;
}
std::cout << "; ALU port utilization:" << std::endl;
std::cout << "; (*= in use, _ = idle)" << std::endl;
std::cout << "; (* = in use, _ = idle)" << std::endl;
int portCycles = 0;
for (int i = 0; i < CYCLE_MAP_SIZE; ++i) {
@ -920,14 +911,12 @@ namespace RandomX {
int asicLatency[8];
memset(asicLatency, 0, sizeof(asicLatency));
int mulCount = 0;
for (int i = 0; i < outIndex; ++i) {
Instruction& instr = prog(i);
int latDst = asicLatency[instr.dst] + 1;
int latSrc = instr.dst != instr.src ? asicLatency[instr.src] + 1 : 0;
asicLatency[instr.dst] = std::max(latDst, latSrc);
mulCount += isMul(instr.opcode);
}
std::cout << "; Multiplications: " << mulCount << std::endl;
@ -943,238 +932,4 @@ namespace RandomX {
prog.setSize(outIndex);
}
void generateLightProgram(LightProgram& prog, const void* seed, int indexRegister, int nonce) {
// Source: https://www.agner.org/optimize/instruction_tables.pdf
const int op_latency[LightInstructionType::COUNT] = { 1, 2, 1, 2, 3, 5, 5, 4, 1, 2, 5 };
// Instruction latencies for theoretical ASIC implementation
const int asic_op_latency[LightInstructionType::COUNT] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 };
// Available ALUs for each instruction
const int op_ALUs[LightInstructionType::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[LightInstructionType::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[LightInstructionType::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 != LightInstructionType::IMULH_R && instrType != LightInstructionType::ISMULH_R && ((inst_data[a] & 0xFFFF00) == (instrType << 8) + ((inst_data[b] & 255) << 16)))
{
continue;
}
if ((instrType == LightInstructionType::IADD_RC) || (instrType == LightInstructionType::IMUL_9C) || (instrType == LightInstructionType::IMUL_RCP) || (instrType == LightInstructionType::COND_R) || ((instrType != LightInstructionType::IMULH_R) && (instrType != LightInstructionType::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 == LightInstructionType::IADD_RC || instrType == LightInstructionType::IMUL_9C || instrType == LightInstructionType::IMULH_R || instrType == LightInstructionType::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 == LightInstructionType::IADD_RC || instrType == LightInstructionType::IMUL_9C || instrType == LightInstructionType::IMULH_R || instrType == LightInstructionType::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] = { LightInstructionType::IMUL_R, LightInstructionType::IROR_R, LightInstructionType::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);
}
}

View file

@ -224,10 +224,10 @@ int main(int argc, char** argv) {
if (genLight) {
RandomX::LightProgram p;
RandomX::generateLightProg2(p, seed, 0, programCount);
RandomX::AssemblyGeneratorX86 asmX86;
asmX86.generateProgram(p);
//RandomX::AssemblyGeneratorX86 asmX86;
//asmX86.generateProgram(p);
//std::ofstream file("lightProg2.asm");
asmX86.printCode(std::cout);
//asmX86.printCode(std::cout);
return 0;
}