wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fixed simX multicore support, added shared memory

Browse Source
This commit is contained in:
Blaise Tine
2021-03-04 20:45:27 -08:00
parent 5e3a949d2d
commit 8a86bddd3e
18 changed files with 824 additions and 717 deletions
Download Patch File Download Diff File Expand all files Collapse all files

410
simX/execute.cpp
View File

@@ -159,6 +159,7 @@ uint8_t fpBinIsInf(uint32_t din) {
void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
assert(tmask_.any());
Word nextPC = PC_;
bool updatePC = false;
bool runOnce = false;
@@ -167,20 +168,21 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
Word func6 = instr.getFunc6();
Word func7 = instr.getFunc7();
Opcode opcode = instr.getOpcode();
int rdest = instr.getRDest();
int rsrc0 = instr.getRSrc(0);
int rsrc1 = instr.getRSrc(1);
int rsrc2 = instr.getRSrc(2);
Word immsrc = instr.getImm();
bool vmask = instr.getVmask();
auto opcode = instr.getOpcode();
int rdest = instr.getRDest();
int rsrc0 = instr.getRSrc(0);
int rsrc1 = instr.getRSrc(1);
int rsrc2 = instr.getRSrc(2);
Word immsrc= instr.getImm();
bool vmask = instr.getVmask();
for (std::size_t t = 0; t < tmask_.count(); t++) {
if (runOnce)
int num_threads = core_->arch().num_threads();
for (int t = 0; t < num_threads; t++) {
if (!tmask_.test(t) || runOnce)
continue;
auto &iregs = iRegFile_[t];
auto &fregs = fRegFile_[t];
auto &iregs = iRegFile_.at(t);
auto &fregs = fRegFile_.at(t);
++insts_;
@@ -197,7 +199,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 0:
// MUL
D(3, "MUL: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = ((int)iregs[rsrc0]) * ((int)iregs[rsrc1]);
if (rdest) iregs[rdest] = ((int)iregs[rsrc0]) * ((int)iregs[rsrc1]);
break;
case 1:
// MULH
@@ -213,7 +215,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
}
// cout << "mulh: " << std::dec << first << " * " << second;
uint64_t result = first * second;
iregs[rdest] = (result >> 32) & 0xFFFFFFFF;
if (rdest) iregs[rdest] = (result >> 32) & 0xFFFFFFFF;
// cout << " = " << result << " or " << iregs[rdest] << "\n";
}
break;
@@ -226,7 +228,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
first = first | 0xFFFFFFFF00000000;
}
int64_t second = (int64_t)iregs[rsrc1];
iregs[rdest] = ((first * second) >> 32) & 0xFFFFFFFF;
if (rdest) iregs[rdest] = ((first * second) >> 32) & 0xFFFFFFFF;
}
break;
case 3:
@@ -236,46 +238,46 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
uint64_t first = (uint64_t)iregs[rsrc0];
uint64_t second = (uint64_t)iregs[rsrc1];
// cout << "MULHU\n";
iregs[rdest] = ((first * second) >> 32) & 0xFFFFFFFF;
if (rdest) iregs[rdest] = ((first * second) >> 32) & 0xFFFFFFFF;
}
break;
case 4:
// DIV
D(3, "DIV: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (iregs[rsrc1] == 0) {
iregs[rdest] = -1;
if (rdest) iregs[rdest] = -1;
break;
}
// cout << "dividing: " << std::dec << ((int) iregs[rsrc0]) << " / " << ((int) iregs[rsrc1]);
iregs[rdest] = ((int)iregs[rsrc0]) / ((int)iregs[rsrc1]);
if (rdest) iregs[rdest] = ((int)iregs[rsrc0]) / ((int)iregs[rsrc1]);
// cout << " = " << ((int) iregs[rdest]) << "\n";
break;
case 5:
// DIVU
D(3, "DIVU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (iregs[rsrc1] == 0) {
iregs[rdest] = -1;
if (rdest) iregs[rdest] = -1;
break;
}
iregs[rdest] = ((uint32_t)iregs[rsrc0]) / ((uint32_t)iregs[rsrc1]);
if (rdest) iregs[rdest] = ((uint32_t)iregs[rsrc0]) / ((uint32_t)iregs[rsrc1]);
break;
case 6:
// REM
D(3, "REM: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (iregs[rsrc1] == 0) {
iregs[rdest] = iregs[rsrc0];
if (rdest) iregs[rdest] = iregs[rsrc0];
break;
}
iregs[rdest] = ((int)iregs[rsrc0]) % ((int)iregs[rsrc1]);
if (rdest) iregs[rdest] = ((int)iregs[rsrc0]) % ((int)iregs[rsrc1]);
break;
case 7:
// REMU
D(3, "REMU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (iregs[rsrc1] == 0) {
iregs[rdest] = iregs[rsrc0];
if (rdest) iregs[rdest] = iregs[rsrc0];
break;
}
iregs[rdest] = ((uint32_t)iregs[rsrc0]) % ((uint32_t)iregs[rsrc1]);
if (rdest) iregs[rdest] = ((uint32_t)iregs[rsrc0]) % ((uint32_t)iregs[rsrc1]);
break;
default:
std::cout << "unsupported MUL/DIV instr\n";
@@ -287,52 +289,52 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 0:
if (func7) {
D(3, "SUBI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] - iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] - iregs[rsrc1];
} else {
D(3, "ADDI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] + iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] + iregs[rsrc1];
}
break;
case 1:
D(3, "SLLI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] << iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] << iregs[rsrc1];
break;
case 2:
D(3, "SLTI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (int(iregs[rsrc0]) < int(iregs[rsrc1])) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
break;
case 3:
D(3, "SLTU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
if (Word(iregs[rsrc0]) < Word(iregs[rsrc1])) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
break;
case 4:
D(3, "XORI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] ^ iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] ^ iregs[rsrc1];
break;
case 5:
if (func7) {
D(3, "SRLI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = int(iregs[rsrc0]) >> int(iregs[rsrc1]);
if (rdest) iregs[rdest] = int(iregs[rsrc0]) >> int(iregs[rsrc1]);
} else {
D(3, "SRLU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = Word(iregs[rsrc0]) >> Word(iregs[rsrc1]);
if (rdest) iregs[rdest] = Word(iregs[rsrc0]) >> Word(iregs[rsrc1]);
}
break;
case 6:
D(3, "ORI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] | iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] | iregs[rsrc1];
break;
case 7:
D(3, "AND: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
iregs[rdest] = iregs[rsrc0] & iregs[rsrc1];
if (rdest) iregs[rdest] = iregs[rsrc0] & iregs[rsrc1];
break;
default:
std::cout << "ERROR: UNSUPPORTED R INST\n";
@@ -346,58 +348,58 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 0:
// ADDI
D(3, "ADDI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=" << immsrc);
iregs[rdest] = iregs[rsrc0] + immsrc;
if (rdest) iregs[rdest] = iregs[rsrc0] + immsrc;
break;
case 2:
// SLTI
D(3, "SLTI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=" << immsrc);
if (int(iregs[rsrc0]) < int(immsrc)) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
break;
case 3: {
// SLTIU
D(3, "SLTIU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=" << immsrc);
if (unsigned(iregs[rsrc0]) < unsigned(immsrc)) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
} break;
case 4:
// XORI
D(3, "XORI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = iregs[rsrc0] ^ immsrc;
if (rdest) iregs[rdest] = iregs[rsrc0] ^ immsrc;
break;
case 6:
// ORI
D(3, "ORI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = iregs[rsrc0] | immsrc;
if (rdest) iregs[rdest] = iregs[rsrc0] | immsrc;
break;
case 7:
// ANDI
D(3, "ANDI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = iregs[rsrc0] & immsrc;
if (rdest) iregs[rdest] = iregs[rsrc0] & immsrc;
break;
case 1:
// SLLI
D(3, "SLLI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = iregs[rsrc0] << immsrc;
if (rdest) iregs[rdest] = iregs[rsrc0] << immsrc;
break;
case 5:
if ((func7 == 0)) {
// SRLI
D(3, "SRLI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=" << immsrc);
Word result = Word(iregs[rsrc0]) >> Word(immsrc);
iregs[rdest] = result;
if (rdest) iregs[rdest] = result;
} else {
// SRAI
D(3, "SRAI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=" << immsrc);
Word op1 = iregs[rsrc0];
Word op2 = immsrc;
iregs[rdest] = op1 >> op2;
if (rdest) iregs[rdest] = op1 >> op2;
}
break;
default:
@@ -409,34 +411,34 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
++loads_;
Word memAddr = ((iregs[rsrc0] + immsrc) & 0xFFFFFFFC);
Word shift_by = ((iregs[rsrc0] + immsrc) & 0x00000003) * 8;
Word data_read = core_->mem().read(memAddr, 0);
Word data_read = core_->dcache_read(memAddr, 0);
trace_inst->is_lw = true;
trace_inst->mem_addresses[t] = memAddr;
switch (func3) {
case 0:
// LBI
D(3, "LBI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = signExt((data_read >> shift_by) & 0xFF, 8, 0xFF);
if (rdest) iregs[rdest] = signExt((data_read >> shift_by) & 0xFF, 8, 0xFF);
break;
case 1:
// LWI
D(3, "LHI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = signExt((data_read >> shift_by) & 0xFFFF, 16, 0xFFFF);
if (rdest) iregs[rdest] = signExt((data_read >> shift_by) & 0xFFFF, 16, 0xFFFF);
break;
case 2:
// LDI
D(3, "LWI: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = int(data_read & 0xFFFFFFFF);
if (rdest) iregs[rdest] = int(data_read & 0xFFFFFFFF);
break;
case 4:
// LBU
D(3, "LBU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = unsigned((data_read >> shift_by) & 0xFF);
if (rdest) iregs[rdest] = unsigned((data_read >> shift_by) & 0xFF);
break;
case 5:
// LWU
D(3, "LHU: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
iregs[rdest] = unsigned((data_read >> shift_by) & 0xFFFF);
if (rdest) iregs[rdest] = unsigned((data_read >> shift_by) & 0xFFFF);
break;
default:
std::cout << "ERROR: UNSUPPORTED L INST\n";
@@ -453,17 +455,17 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 0:
// SB
D(3, "SB: r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1] << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
core_->mem().write(memAddr, iregs[rsrc1] & 0x000000FF, 0, 1);
core_->dcache_write(memAddr, iregs[rsrc1] & 0x000000FF, 0, 1);
break;
case 1:
// SH
D(3, "SH: r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1] << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
core_->mem().write(memAddr, iregs[rsrc1], 0, 2);
core_->dcache_write(memAddr, iregs[rsrc1], 0, 2);
break;
case 2:
// SW
D(3, "SW: r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1] << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
core_->mem().write(memAddr, iregs[rsrc1], 0, 4);
core_->dcache_write(memAddr, iregs[rsrc1], 0, 4);
break;
default:
std::cout << "ERROR: UNSUPPORTED S INST\n";
@@ -532,11 +534,11 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
break;
case LUI_INST:
D(3, "LUI: r" << std::dec << rdest << " <- imm=0x" << std::hex << immsrc);
iregs[rdest] = (immsrc << 12) & 0xfffff000;
if (rdest) iregs[rdest] = (immsrc << 12) & 0xfffff000;
break;
case AUIPC_INST:
D(3, "AUIPC: r" << std::dec << rdest << " <- imm=0x" << std::hex << immsrc);
iregs[rdest] = ((immsrc << 12) & 0xfffff000) + (PC_ - 4);
if (rdest) iregs[rdest] = ((immsrc << 12) & 0xfffff000) + (PC_ - 4);
break;
case JAL_INST:
D(3, "JAL: r" << std::dec << rdest << " <- imm=0x" << std::hex << immsrc);
@@ -545,9 +547,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
nextPC = (PC_ - 4) + immsrc;
//std::cout << "JAL... SETTING PC: " << nextPC << "\n";
}
if (rdest != 0) {
iregs[rdest] = PC_;
}
if (rdest) iregs[rdest] = PC_;
updatePC = true;
break;
case JALR_INST:
@@ -557,109 +557,53 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
nextPC = iregs[rsrc0] + immsrc;
//std::cout << "JALR... SETTING PC: " << nextPC << "\n";
}
if (rdest != 0) {
iregs[rdest] = PC_;
}
if (rdest) iregs[rdest] = PC_;
updatePC = true;
break;
case SYS_INST: {
D(3, "SYS_INST: r" << std::dec << rdest << " <- r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", imm=0x" << std::hex << immsrc);
Word rs1 = iregs[rsrc0];
Word csr_addr = immsrc & 0x00000FFF;
// GPGPU CSR extension
if (csr_addr == CSR_WTID) {
// Warp threadID
iregs[rdest] = t;
} else if (csr_addr == CSR_LTID) {
// Core threadID
iregs[rdest] = t + (id_ * core_->arch().num_threads());
} else if (csr_addr == CSR_GTID) {
// Processor threadID
iregs[rdest] = t + (id_ * core_->arch().num_threads()) +
(core_->arch().num_threads() * core_->arch().num_warps() * core_->id());
} else if (csr_addr == CSR_LWID) {
// Core warpID
iregs[rdest] = id_;
} else if (csr_addr == CSR_GWID) {
// Processor warpID
iregs[rdest] = id_ + (core_->arch().num_warps() * core_->id());
} else if (csr_addr == CSR_GCID) {
// Processor coreID
iregs[rdest] = core_->id();
} else if (csr_addr == CSR_NT) {
// Number of threads per warp
iregs[rdest] = core_->arch().num_threads();
} else if (csr_addr == CSR_NW) {
// Number of warps per core
iregs[rdest] = core_->arch().num_warps();
} else if (csr_addr == CSR_NC) {
// Number of cores
iregs[rdest] = core_->arch().num_cores();
} else if (csr_addr == CSR_INSTRET) {
// NumInsts
iregs[rdest] = (Word)core_->num_insts();
} else if (csr_addr == CSR_INSTRET_H) {
// NumInsts
iregs[rdest] = (Word)(core_->num_insts() >> 32);
} else if (csr_addr == CSR_CYCLE) {
// NumCycles
iregs[rdest] = (Word)core_->num_steps();
} else if (csr_addr == CSR_CYCLE_H) {
// NumCycles
iregs[rdest] = (Word)(core_->num_steps() >> 32);
} else {
switch (func3) {
case 0:
if (csr_addr < 2) {
// ECALL/EBREAK
tmask_.reset();
}
break;
case 1:
// CSRRW
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rs1;
break;
case 2:
// CSRRS
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rs1 | csrs_[csr_addr];
break;
case 3:
// CSRRC
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rs1 & (~csrs_[csr_addr]);
break;
case 5:
// CSRRWI
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rsrc0;
break;
case 6:
// CSRRSI
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rsrc0 | csrs_[csr_addr];
break;
case 7:
// CSRRCI
if (rdest != 0) {
iregs[rdest] = csrs_[csr_addr];
}
csrs_[csr_addr] = rsrc0 & (~csrs_[csr_addr]);
break;
default:
break;
switch (func3) {
case 0:
if (csr_addr < 2) {
// ECALL/EBREAK
tmask_.reset();
active_ = tmask_.any();
}
break;
case 1:
// CSRRW
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rs1);
break;
case 2:
// CSRRS
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rs1 | core_->get_csr(csr_addr, t, id_));
break;
case 3:
// CSRRC
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rs1 & ~core_->get_csr(csr_addr, t, id_));
break;
case 5:
// CSRRWI
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rsrc0);
break;
case 6:
// CSRRSI
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rsrc0 | core_->get_csr(csr_addr, t, id_));
break;
case 7:
// CSRRCI
if (rdest) iregs[rdest] = core_->get_csr(csr_addr, t, id_);
core_->set_csr(csr_addr, rsrc0 & ~core_->get_csr(csr_addr, t, id_));
break;
default:
break;
}
} break;
case FENCE:
@@ -684,6 +628,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
for (int i = 0; i < active_threads; ++i) {
tmask_[i] = true;
}
active_ = tmask_.any();
runOnce = true;
} break;
case 1: {
@@ -726,6 +671,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
for (unsigned i = 0; i < e.tmask.size(); ++i) {
tmask_[i] = !e.tmask[i] && tmask_[i];
}
active_ = tmask_.any();
D(3, "Split: New TM");
DX( for (int i = 0; i < core_->arch().num_threads(); ++i) D(3, tmask_[i] << " "); )
@@ -741,6 +687,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
D(2, "Uni branch at join");
printf("NEW DOMESTACK: \n");
tmask_ = domStack_.top().tmask;
active_ = tmask_.any();
domStack_.pop();
break;
}
@@ -757,6 +704,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
DX( for (int i = 0; i < core_->arch().num_threads(); ++i) D(3, tmask_[i] << " "); )
std::cout << "\n";
tmask_ = domStack_.top().tmask;
active_ = tmask_.any();
D(3, "Join: New TM: ");
DX( for (int i = 0; i < core_->arch().num_threads(); ++i) D(3, tmask_[i] << " "); )
@@ -765,7 +713,10 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
runOnce = true;
} break;
case 4: {
// is_barrier
// BAR
D(3, "BAR: r" << std::dec << rsrc0 << "=0x" << std::hex << iregs[rsrc0] << ", r" << std::dec << rsrc1 << "=0x" << std::hex << iregs[rsrc1]);
active_ = false;
core_->barrier(iregs[rsrc0], iregs[rsrc1], id_);
trace_inst->stall_warp = true;
runOnce = true;
} break;
@@ -1667,7 +1618,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
} else if (s0 >= (2 * VLMAX)) {
vl_ = VLMAX;
}
iregs[rdest] = vl_;
if (rdest) iregs[rdest] = vl_;
} break;
default: {
std::abort();
@@ -1681,7 +1632,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
// rs1 is integer is register!
Word memAddr = ((iregs[rsrc0] + immsrc) & 0xFFFFFFFC); // alignment
D(9,"something weird happen!");
Word data_read = core_->mem().read(memAddr, 0);
Word data_read = core_->dcache_read(memAddr, 0);
D(3, "Memaddr");
DPN(3, ' ' << std::setw(8) << std::hex << memAddr << std::endl);
trace_inst->is_lw = true;
@@ -1698,9 +1649,8 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
}
D(3, "LOAD MEM ADDRESS: " << std::hex << memAddr);
} else {
int VLEN = core_->arch().vsize() * 8;
D(3, "Executing vector load");
D(4, "lmul: " << vtype_.vlmul << " VLEN:" << VLEN << "sew: " << vtype_.vsew);
D(4, "lmul: " << vtype_.vlmul << " VLEN:" << (core_->arch().vsize() * 8) << "sew: " << vtype_.vsew);
D(4, "src: " << rsrc0 << " " << iregs[rsrc0]);
D(4, "dest" << rdest);
D(4, "width" << instr.getVlsWidth());
@@ -1711,7 +1661,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 6: { //load word and unit strided (not checking for unit stride)
for (int i = 0; i < vl_; i++) {
Word memAddr = ((iregs[rsrc0]) & 0xFFFFFFFC) + (i * vtype_.vsew / 8);
Word data_read = core_->mem().read(memAddr, 0);
Word data_read = core_->dcache_read(memAddr, 0);
D(4, "Mem addr: " << std::hex << memAddr << " Data read " << data_read);
int *result_ptr = (int *)(vd.data() + i);
*result_ptr = data_read;
@@ -1748,7 +1698,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
break;
case 2:
// //std::cout << std::hex << "FSW: about to write: " << fregs[rsrc1] << " to " << memAddr << "\n";
core_->mem().write(memAddr, fregs[rsrc1], 0, 4);
core_->dcache_write(memAddr, fregs[rsrc1], 0, 4);
break;
case 3:
std::cout << "ERROR: UNSUPPORTED FS INST\n";
@@ -1780,7 +1730,7 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 6: //store word and unit strided (not checking for unit stride)
{
uint32_t value = *(uint32_t *)(vRegFile_[instr.getVs3()].data() + i);
core_->mem().write(memAddr, value, 0, 4);
core_->dcache_write(memAddr, value, 0, 4);
D(4, "store: " << memAddr << " value:" << value);
} break;
default:
@@ -1802,8 +1752,8 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
D(3, "one or two rsrc is NaN!");
// one of them is not quiet NaN, them set FCSR
if ((fpBinIsNan(fregs[rsrc0])==2) | (fpBinIsNan(fregs[rsrc1])==2)) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
if (fpBinIsNan(fregs[rsrc0]) && fpBinIsNan(fregs[rsrc1]))
fregs[rdest] = 0x7fc00000; // canonical(quiet) NaN
@@ -1836,28 +1786,28 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
// fcsr defined in riscv
if (fetestexcept(FE_INEXACT)) {
csrs_[0x003] = csrs_[0x003] | 0x1; // set NX bit
csrs_[0x001] = csrs_[0x001] | 0x1; // set NX bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x1); // set NX bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x1); // set NX bit
}
if (fetestexcept(FE_UNDERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x2; // set UF bit
csrs_[0x001] = csrs_[0x001] | 0x2; // set UF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x2); // set UF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x2); // set UF bit
}
if (fetestexcept(FE_OVERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x4; // set OF bit
csrs_[0x001] = csrs_[0x001] | 0x4; // set OF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x4); // set OF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x4); // set OF bit
}
if (fetestexcept(FE_DIVBYZERO)) {
csrs_[0x003] = csrs_[0x003] | 0x8; // set DZ bit
csrs_[0x001] = csrs_[0x001] | 0x8; // set DZ bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x8); // set DZ bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x8); // set DZ bit
}
if (fetestexcept(FE_INVALID)) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NX bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NX bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NX bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NX bit
}
D(4, "fpOut: " << fpOut);
@@ -1897,8 +1847,8 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
if (fpBinIsNan(fregs[rsrc0]) || fpBinIsNan(fregs[rsrc1])) { // if one of src is NaN
// one of them is not quiet NaN, them set FCSR
if ((fpBinIsNan(fregs[rsrc0])==2) | (fpBinIsNan(fregs[rsrc1])==2)) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
if (fpBinIsNan(fregs[rsrc0]) && fpBinIsNan(fregs[rsrc1]))
fregs[rdest] = 0x7fc00000; // canonical(quiet) NaN
@@ -1982,31 +1932,31 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
// fcsr defined in riscv
if (fetestexcept(FE_INEXACT)) {
csrs_[0x003] = csrs_[0x003] | 0x1; // set NX bit
csrs_[0x001] = csrs_[0x001] | 0x1; // set NX bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x1); // set NX bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x1); // set NX bit
}
if (fetestexcept(FE_UNDERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x2; // set UF bit
csrs_[0x001] = csrs_[0x001] | 0x2; // set UF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x2); // set UF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x2); // set UF bit
}
if (fetestexcept(FE_OVERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x4; // set OF bit
csrs_[0x001] = csrs_[0x001] | 0x4; // set OF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x4); // set OF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x4); // set OF bit
}
if (fetestexcept(FE_DIVBYZERO)) {
csrs_[0x003] = csrs_[0x003] | 0x8; // set DZ bit
csrs_[0x001] = csrs_[0x001] | 0x8; // set DZ bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x8); // set DZ bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x8); // set DZ bit
}
if (fetestexcept(FE_INVALID) || outOfRange) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
iregs[rdest] = result;
if (rdest) iregs[rdest] = result;
} break;
// FMV.X.W FCLASS.S
@@ -2015,31 +1965,33 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
if (func3) {
// Examine the value in fpReg rs1 and write to integer rd
// a 10-bit mask to indicate the class of the fp number
iregs[rdest] = 0; // clear all bits
if (rdest) iregs[rdest] = 0; // clear all bits
bool fsign = fregs[rsrc0] & 0x80000000;
uint32_t expo = (fregs[rsrc0]>>23) & 0x000000FF;
uint32_t fraction = fregs[rsrc0] & 0x007FFFFF;
if ((expo==0) && (fraction==0))
iregs[rdest] = fsign? (1<<3) : (1<<4); // +/- 0
else if ((expo==0) && (fraction!=0))
iregs[rdest] = fsign? (1<<2) : (1<<5); // +/- subnormal
else if ((expo==0xFF) && (fraction==0))
iregs[rdest] = fsign? (1<<0) : (1<<7); // +/- infinity
else if ((expo==0xFF) && (fraction!=0))
if (!fsign && (fraction == 0x00400000))
iregs[rdest] = (1<<9); // quiet NaN
else
iregs[rdest] = (1<<8); // signaling NaN
else
iregs[rdest] = fsign? (1<<1) : (1<<6); // +/- normal
if ((expo==0) && (fraction==0)) {
if (rdest) iregs[rdest] = fsign? (1<<3) : (1<<4); // +/- 0
} else if ((expo==0) && (fraction!=0)) {
if (rdest) iregs[rdest] = fsign? (1<<2) : (1<<5); // +/- subnormal
} else if ((expo==0xFF) && (fraction==0)) {
if (rdest) iregs[rdest] = fsign? (1<<0) : (1<<7); // +/- infinity
} else if ((expo==0xFF) && (fraction!=0)) {
if (!fsign && (fraction == 0x00400000)) {
if (rdest) iregs[rdest] = (1<<9); // quiet NaN
} else {
if (rdest) iregs[rdest] = (1<<8); // signaling NaN
}
} else {
if (rdest) iregs[rdest] = fsign? (1<<1) : (1<<6); // +/- normal
}
} else {
// FMV.X.W
// Move bit values from floating-point register rs1 to integer register rd
// Since we are using integer register to represent floating point register,
// just simply assign here.
iregs[rdest] = fregs[rsrc0];
if (rdest) iregs[rdest] = fregs[rsrc0];
}
} break;
@@ -2052,35 +2004,35 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
// FLE.S or FLT.S
if (func3 == 0 || func3 == 1) {
// If either input is NaN, set NV bit
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
} else { // FEQ.S
// Only set NV bit if it is signaling NaN
if (fpBinIsNan(fregs[rsrc0]) == 2 || fpBinIsNan(fregs[rsrc1]) == 2) {
// If either input is NaN, set NV bit
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
}
// The result is 0 if either operand is NaN
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
} else {
switch(func3) {
case 0: {
// FLE.S
if (intregToFloat(fregs[rsrc0]) <= intregToFloat(fregs[rsrc1])) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
} break;
case 1: {
// FLT.S
if (intregToFloat(fregs[rsrc0]) < intregToFloat(fregs[rsrc1])) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
}
break;
@@ -2088,9 +2040,9 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case 2: {
// FEQ.S
if (intregToFloat(fregs[rsrc0]) == intregToFloat(fregs[rsrc1])) {
iregs[rdest] = 1;
if (rdest) iregs[rdest] = 1;
} else {
iregs[rdest] = 0;
if (rdest) iregs[rdest] = 0;
}
}
break;
@@ -2131,16 +2083,16 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
case FMNMSUB: {
// multiplicands are infinity and zero, them set FCSR
if (fpBinIsZero(fregs[rsrc0])|| fpBinIsZero(fregs[rsrc1])|| fpBinIsInf(fregs[rsrc0]) || fpBinIsInf(fregs[rsrc1])) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
if (fpBinIsNan(fregs[rsrc0]) || fpBinIsNan(fregs[rsrc1]) || fpBinIsNan(fregs[rsrc2])) {
// if one of op is NaN
// if addend is not quiet NaN, them set FCSR
if ((fpBinIsNan(fregs[rsrc0])==2) | (fpBinIsNan(fregs[rsrc1])==2) | (fpBinIsNan(fregs[rsrc1])==2)) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
fregs[rdest] = 0x7fc00000; // canonical(quiet) NaN
} else {
@@ -2170,28 +2122,28 @@ void Warp::execute(Instr &instr, trace_inst_t *trace_inst) {
// fcsr defined in riscv
if (fetestexcept(FE_INEXACT)) {
csrs_[0x003] = csrs_[0x003] | 0x1; // set NX bit
csrs_[0x001] = csrs_[0x001] | 0x1; // set NX bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x1); // set NX bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x1); // set NX bit
}
if (fetestexcept(FE_UNDERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x2; // set UF bit
csrs_[0x001] = csrs_[0x001] | 0x2; // set UF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x2); // set UF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x2); // set UF bit
}
if (fetestexcept(FE_OVERFLOW)) {
csrs_[0x003] = csrs_[0x003] | 0x4; // set OF bit
csrs_[0x001] = csrs_[0x001] | 0x4; // set OF bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x4); // set OF bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x4); // set OF bit
}
if (fetestexcept(FE_DIVBYZERO)) {
csrs_[0x003] = csrs_[0x003] | 0x8; // set DZ bit
csrs_[0x001] = csrs_[0x001] | 0x8; // set DZ bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x8); // set DZ bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x8); // set DZ bit
}
if (fetestexcept(FE_INVALID)) {
csrs_[0x003] = csrs_[0x003] | 0x10; // set NV bit
csrs_[0x001] = csrs_[0x001] | 0x10; // set NV bit
core_->set_csr(0x003, core_->get_csr(0x003, t, id_) | 0x10); // set NV bit
core_->set_csr(0x001, core_->get_csr(0x001, t, id_) | 0x10); // set NV bit
}
fregs[rdest] = floatToBin(fpOut);