adds faster decoding to tcc and cleans up others

This commit is contained in:
Eyck-Alexander Jentzsch 2023-07-29 11:42:46 +02:00
parent bd0d15f3a2
commit 6e52af168b
4 changed files with 180 additions and 187 deletions

View File

@ -218,7 +218,7 @@ private:
});
}
}
typename arch::traits<ARCH>::opcode_e decodeInstr(decoding_tree_node* node, code_word_t word){
typename arch::traits<ARCH>::opcode_e decode_instr(decoding_tree_node* node, code_word_t word){
if(!node->children.size()){
if(node->instrs.size() == 1) return node->instrs[0].op;
for(auto instr : node->instrs){
@ -228,7 +228,7 @@ private:
else{
for(auto child : node->children){
if (child->value == (node->submask&word)){
return decodeInstr(child, word);
return decode_instr(child, word);
}
}
}
@ -260,7 +260,6 @@ constexpr size_t bit_count(uint32_t u) {
template <typename ARCH>
vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id)
: vm_base<ARCH>(core, core_id, cluster_id) {
unsigned id=0;
root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for(auto instr:instr_descr){
root->instrs.push_back(instr);
@ -297,7 +296,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
} else {
if (is_jump_to_self_enabled(cond) &&
(instr == 0x0000006f || (instr&0xffff)==0xa001)) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
auto inst_id = decodeInstr(root, instr);
auto inst_id = decode_instr(root, instr);
// pre execution stuff
this->core.reg.last_branch = 0;
if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id));

View File

@ -121,57 +121,7 @@ protected:
}
}
// some compile time constants
// enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK16)) };
std::array<compile_func, LUT_SIZE> lut;
std::array<compile_func, LUT_SIZE_C> lut_00, lut_01, lut_10;
std::array<compile_func, LUT_SIZE> lut_11;
std::array<compile_func *, 4> qlut;
std::array<const uint32_t, 4> lutmasks = {{EXTR_MASK16, EXTR_MASK16, EXTR_MASK16, EXTR_MASK32}};
void expand_bit_mask(int pos, uint32_t mask, uint32_t value, uint32_t valid, uint32_t idx, compile_func lut[],
compile_func f) {
if (pos < 0) {
lut[idx] = f;
} else {
auto bitmask = 1UL << pos;
if ((mask & bitmask) == 0) {
expand_bit_mask(pos - 1, mask, value, valid, idx, lut, f);
} else {
if ((valid & bitmask) == 0) {
expand_bit_mask(pos - 1, mask, value, valid, (idx << 1), lut, f);
expand_bit_mask(pos - 1, mask, value, valid, (idx << 1) + 1, lut, f);
} else {
auto new_val = idx << 1;
if ((value & bitmask) != 0) new_val++;
expand_bit_mask(pos - 1, mask, value, valid, new_val, lut, f);
}
}
}
}
inline uint32_t extract_fields(uint32_t val) { return extract_fields(29, val >> 2, lutmasks[val & 0x3], 0); }
uint32_t extract_fields(int pos, uint32_t val, uint32_t mask, uint32_t lut_val) {
if (pos >= 0) {
auto bitmask = 1UL << pos;
if ((mask & bitmask) == 0) {
lut_val = extract_fields(pos - 1, val, mask, lut_val);
} else {
auto new_val = lut_val << 1;
if ((val & bitmask) != 0) new_val++;
lut_val = extract_fields(pos - 1, val, mask, new_val);
}
}
return lut_val;
}
template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
inline S sext(U from) {
auto mask = (1ULL<<W) - 1;
@ -183,14 +133,23 @@ private:
/****************************************************************************
* start opcode definitions
****************************************************************************/
struct InstructionDesriptor {
struct instruction_descriptor {
size_t length;
uint32_t value;
uint32_t mask;
compile_func op;
};
struct decoding_tree_node{
std::vector<instruction_descriptor> instrs;
std::vector<decoding_tree_node*> children;
uint32_t submask = std::numeric_limits<uint32_t>::max();
uint32_t value;
decoding_tree_node(uint32_t value) : value(value){}
};
const std::array<InstructionDesriptor, ${instructions.size}> instr_descr = {{
decoding_tree_node* root {nullptr};
const std::array<instruction_descriptor, ${instructions.size}> instr_descr = {{
/* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
/* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */
{${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
@ -228,11 +187,64 @@ private:
vm_impl::gen_trap_check(tu);
return BRANCH;
}
//decoding functionality
void populate_decoding_tree(decoding_tree_node* root){
//create submask
for(auto instr: root->instrs){
root->submask &= instr.mask;
}
//put each instr according to submask&encoding into children
for(auto instr: root->instrs){
bool foundMatch = false;
for(auto child: root->children){
//use value as identifying trait
if(child->value == (instr.value&root->submask)){
child->instrs.push_back(instr);
foundMatch = true;
}
}
if(!foundMatch){
decoding_tree_node* child = new decoding_tree_node(instr.value&root->submask);
child->instrs.push_back(instr);
root->children.push_back(child);
}
}
root->instrs.clear();
//call populate_decoding_tree for all children
if(root->children.size() >1)
for(auto child: root->children){
populate_decoding_tree(child);
}
else{
//sort instrs by value of the mask, this works bc we want to have the least restrictive one last
std::sort(root->children[0]->instrs.begin(), root->children[0]->instrs.end(), [](const instruction_descriptor& instr1, const instruction_descriptor& instr2) {
return instr1.mask > instr2.mask;
});
}
}
compile_func decode_instr(decoding_tree_node* node, code_word_t word){
if(!node->children.size()){
if(node->instrs.size() == 1) return node->instrs[0].op;
for(auto instr : node->instrs){
if((instr.mask&word) == instr.value) return instr.op;
}
}
else{
for(auto child : node->children){
if (child->value == (node->submask&word)){
return decode_instr(child, word);
}
}
}
return nullptr;
}
};
template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
volatile CODE_WORD x = insn;
insn = 2 * x;
template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
volatile CODE_WORD x = instr;
instr = 2 * x;
}
template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@ -240,14 +252,11 @@ template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
template <typename ARCH>
vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id)
: vm_base<ARCH>(core, core_id, cluster_id) {
qlut[0] = lut_00.data();
qlut[1] = lut_01.data();
qlut[2] = lut_10.data();
qlut[3] = lut_11.data();
for (auto instr : instr_descr) {
auto quantrant = instr.value & 0x3;
expand_bit_mask(29, lutmasks[quantrant], instr.value >> 2, instr.mask >> 2, 0, qlut[quantrant], instr.op);
root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for(auto instr:instr_descr){
root->instrs.push_back(instr);
}
populate_decoding_tree(root);
}
template <typename ARCH>
@ -255,30 +264,19 @@ std::tuple<continuation_e>
vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) {
// we fetch at max 4 byte, alignment is 2
enum {TRAP_ID=1<<16};
code_word_t insn = 0;
// const typename traits::addr_t upper_bits = ~traits::PGMASK;
code_word_t instr = 0;
phys_addr_t paddr(pc);
auto *const data = (uint8_t *)&insn;
paddr = this->core.v2p(pc);
// if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
// auto res = this->core.read(paddr, 2, data);
// if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
// if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
// res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
// }
// } else {
auto res = this->core.read(paddr, 4, data);
if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
// }
if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
auto res = this->core.read(paddr, 4, reinterpret_cast<uint8_t*>(&instr));
if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
if (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
// curr pc on stack
++inst_cnt;
auto lut_val = extract_fields(insn);
auto f = qlut[insn & 0x3][lut_val];
auto f = decode_instr(root, instr);
if (f == nullptr) {
f = &this_class::illegal_intruction;
}
return (this->*f)(pc, insn, tu);
return (this->*f)(pc, instr, tu);
}
template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {

View File

@ -298,7 +298,7 @@ private:
});
}
}
typename arch::traits<ARCH>::opcode_e decodeInstr(decoding_tree_node* node, code_word_t word){
typename arch::traits<ARCH>::opcode_e decode_instr(decoding_tree_node* node, code_word_t word){
if(!node->children.size()){
if(node->instrs.size() == 1) return node->instrs[0].op;
for(auto instr : node->instrs){
@ -308,7 +308,7 @@ private:
else{
for(auto child : node->children){
if (child->value == (node->submask&word)){
return decodeInstr(child, word);
return decode_instr(child, word);
}
}
}
@ -340,7 +340,6 @@ constexpr size_t bit_count(uint32_t u) {
template <typename ARCH>
vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id)
: vm_base<ARCH>(core, core_id, cluster_id) {
unsigned id=0;
root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for(auto instr:instr_descr){
root->instrs.push_back(instr);
@ -377,7 +376,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
} else {
if (is_jump_to_self_enabled(cond) &&
(instr == 0x0000006f || (instr&0xffff)==0xa001)) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
auto inst_id = decodeInstr(root, instr);
auto inst_id = decode_instr(root, instr);
// pre execution stuff
this->core.reg.last_branch = 0;
if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id));

View File

@ -121,57 +121,7 @@ protected:
}
}
// some compile time constants
// enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK16)) };
std::array<compile_func, LUT_SIZE> lut;
std::array<compile_func, LUT_SIZE_C> lut_00, lut_01, lut_10;
std::array<compile_func, LUT_SIZE> lut_11;
std::array<compile_func *, 4> qlut;
std::array<const uint32_t, 4> lutmasks = {{EXTR_MASK16, EXTR_MASK16, EXTR_MASK16, EXTR_MASK32}};
void expand_bit_mask(int pos, uint32_t mask, uint32_t value, uint32_t valid, uint32_t idx, compile_func lut[],
compile_func f) {
if (pos < 0) {
lut[idx] = f;
} else {
auto bitmask = 1UL << pos;
if ((mask & bitmask) == 0) {
expand_bit_mask(pos - 1, mask, value, valid, idx, lut, f);
} else {
if ((valid & bitmask) == 0) {
expand_bit_mask(pos - 1, mask, value, valid, (idx << 1), lut, f);
expand_bit_mask(pos - 1, mask, value, valid, (idx << 1) + 1, lut, f);
} else {
auto new_val = idx << 1;
if ((value & bitmask) != 0) new_val++;
expand_bit_mask(pos - 1, mask, value, valid, new_val, lut, f);
}
}
}
}
inline uint32_t extract_fields(uint32_t val) { return extract_fields(29, val >> 2, lutmasks[val & 0x3], 0); }
uint32_t extract_fields(int pos, uint32_t val, uint32_t mask, uint32_t lut_val) {
if (pos >= 0) {
auto bitmask = 1UL << pos;
if ((mask & bitmask) == 0) {
lut_val = extract_fields(pos - 1, val, mask, lut_val);
} else {
auto new_val = lut_val << 1;
if ((val & bitmask) != 0) new_val++;
lut_val = extract_fields(pos - 1, val, mask, new_val);
}
}
return lut_val;
}
template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
inline S sext(U from) {
auto mask = (1ULL<<W) - 1;
@ -183,14 +133,23 @@ private:
/****************************************************************************
* start opcode definitions
****************************************************************************/
struct InstructionDesriptor {
struct instruction_descriptor {
size_t length;
uint32_t value;
uint32_t mask;
compile_func op;
};
struct decoding_tree_node{
std::vector<instruction_descriptor> instrs;
std::vector<decoding_tree_node*> children;
uint32_t submask = std::numeric_limits<uint32_t>::max();
uint32_t value;
decoding_tree_node(uint32_t value) : value(value){}
};
const std::array<InstructionDesriptor, 87> instr_descr = {{
decoding_tree_node* root {nullptr};
const std::array<instruction_descriptor, 87> instr_descr = {{
/* entries are: size, valid value, valid mask, function ptr */
/* instruction LUI, encoding '0b00000000000000000000000000110111' */
{32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, &this_class::__lui},
@ -494,14 +453,14 @@ private:
this->gen_raise_trap(tu, 0, 2);
}
else{
auto new_pc = tu.assignment(tu.ext((tu.bitwise_and((tu.add(tu.load(rs1+ traits::X0, 0),tu.constant((int16_t)sext<12>(imm),16))),tu.constant(~ 0x1,8))),32,true),32);
auto new_pc = tu.assignment(tu.ext((tu.bitwise_and((tu.add(tu.load(rs1+ traits::X0, 0),tu.constant((int16_t)sext<12>(imm),16))),tu.constant(~0x1,8))),32,true),32);
tu.open_if(tu.srem(new_pc,tu.constant(static_cast<uint32_t>(traits:: INSTR_ALIGNMENT),32)));
this->gen_raise_trap(tu, 0, 0);
tu.open_else();
if(rd!= 0) {
tu.store(rd + traits::X0,tu.ext((tu.add(tu.ext(cur_pc_val,32,false),tu.constant( 4,8))),32,true));
}
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~ 0x1,8)),32);
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~0x1,8)),32);
tu.store(traits::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
tu.close_scope();
@ -1963,7 +1922,7 @@ private:
else{
auto xrd = tu.assignment(tu.read_mem(traits::CSR, csr, 32),32);
if(zimm!= 0) {
tu.write_mem(traits::CSR, csr, tu.bitwise_and(xrd,tu.constant(~ ((uint32_t)zimm),32)));
tu.write_mem(traits::CSR, csr, tu.bitwise_and(xrd,tu.constant(~((uint32_t)zimm),32)));
}
if(rd!= 0) {
tu.store(rd + traits::X0,xrd);
@ -2024,7 +1983,7 @@ private:
this->gen_raise_trap(tu, 0, 2);
}
else{
auto res = tu.assignment(tu.mul(tu.ext(tu.load(rs1+ traits::X0, 0),32,false),tu.ext(tu.load(rs2+ traits::X0, 0),32,false)),64);
auto res = tu.assignment(tu.ext((tu.mul(tu.ext(tu.ext(tu.load(rs1+ traits::X0, 0),32,true),64,false),tu.ext(tu.ext(tu.load(rs2+ traits::X0, 0),32,true),64,false))),64,false),64);
if(rd!=0) {
tu.store(rd + traits::X0,tu.ext(res,32,true));
}
@ -2058,7 +2017,7 @@ private:
this->gen_raise_trap(tu, 0, 2);
}
else{
auto res = tu.assignment(tu.mul(tu.ext(tu.load(rs1+ traits::X0, 0),32,false),tu.ext(tu.load(rs2+ traits::X0, 0),32,false)),64);
auto res = tu.assignment(tu.ext((tu.mul(tu.ext(tu.ext(tu.load(rs1+ traits::X0, 0),32,true),64,false),tu.ext(tu.ext(tu.load(rs2+ traits::X0, 0),32,true),64,false))),64,false),64);
if(rd!=0) {
tu.store(rd + traits::X0,tu.ext((tu.lshr(res,tu.constant(static_cast<uint32_t>(traits:: XLEN),32))),32,true));
}
@ -2092,7 +2051,7 @@ private:
this->gen_raise_trap(tu, 0, 2);
}
else{
auto res = tu.assignment(tu.mul(tu.ext(tu.load(rs1+ traits::X0, 0),32,false),tu.load(rs2+ traits::X0, 0)),64);
auto res = tu.assignment(tu.ext((tu.mul(tu.ext(tu.ext(tu.load(rs1+ traits::X0, 0),32,true),64,false),tu.ext(tu.load(rs2+ traits::X0, 0),64,true))),64,false),64);
if(rd!=0) {
tu.store(rd + traits::X0,tu.ext((tu.lshr(res,tu.constant(static_cast<uint32_t>(traits:: XLEN),32))),32,true));
}
@ -2126,7 +2085,7 @@ private:
this->gen_raise_trap(tu, 0, 2);
}
else{
auto res = tu.assignment(tu.mul(tu.load(rs1+ traits::X0, 0),tu.load(rs2+ traits::X0, 0)),64);
auto res = tu.assignment(tu.ext((tu.mul(tu.ext(tu.load(rs1+ traits::X0, 0),64,true),tu.ext(tu.load(rs2+ traits::X0, 0),64,true))),64,true),64);
if(rd!=0) {
tu.store(rd + traits::X0,tu.ext((tu.lshr(res,tu.constant(static_cast<uint32_t>(traits:: XLEN),32))),32,true));
}
@ -2164,13 +2123,13 @@ private:
auto divisor = tu.assignment(tu.ext(tu.load(rs2+ traits::X0, 0),32,false),32);
if(rd!= 0){ tu.open_if(tu.icmp(ICmpInst::ICMP_NE,divisor,tu.constant( 0,8)));
auto MMIN = tu.assignment(tu.constant(((uint32_t)1)<<(static_cast<uint32_t>(traits:: XLEN)-1),32),32);
tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,divisor,tu.constant(- 1,8))));
tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,divisor,tu.constant(-1,8))));
tu.store(rd + traits::X0,MMIN);
tu.open_else();
tu.store(rd + traits::X0,tu.ext((tu.sdiv(dividend,divisor)),32,true));
tu.close_scope();
tu.open_else();
tu.store(rd + traits::X0,tu.constant((uint32_t)- 1,32));
tu.store(rd + traits::X0,tu.constant((uint32_t)-1,32));
tu.close_scope();
}
}
@ -2209,7 +2168,7 @@ private:
}
tu.open_else();
if(rd!=0) {
tu.store(rd + traits::X0,tu.constant((uint32_t)- 1,32));
tu.store(rd + traits::X0,tu.constant((uint32_t)-1,32));
}
tu.close_scope();
}
@ -2244,7 +2203,7 @@ private:
else{
tu.open_if(tu.icmp(ICmpInst::ICMP_NE,tu.load(rs2+ traits::X0, 0),tu.constant( 0,8)));
auto MMIN = tu.assignment(tu.constant( 1<<(static_cast<uint32_t>(traits:: XLEN)-1),8),32);
tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,tu.ext(tu.load(rs2+ traits::X0, 0),32,false),tu.constant(- 1,8))));
tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,tu.ext(tu.load(rs2+ traits::X0, 0),32,false),tu.constant(-1,8))));
if(rd!=0) {
tu.store(rd + traits::X0,tu.constant( 0,8));
}
@ -2353,8 +2312,8 @@ private:
pc=pc+ 2;
gen_set_pc(tu, pc, traits::NEXT_PC);
tu.open_scope();
auto load_address = tu.assignment(tu.ext((tu.add(tu.load(rs1+ 8+ traits::X0, 0),tu.constant(uimm,8))),32,true),32);
tu.store(rd+ 8 + traits::X0,tu.ext(tu.ext(tu.read_mem(traits::MEM, load_address, 32),32,false),32,true));
auto offs = tu.assignment(tu.ext((tu.add(tu.load(rs1+ 8+ traits::X0, 0),tu.constant(uimm,8))),32,true),32);
tu.store(rd+ 8 + traits::X0,tu.ext(tu.ext(tu.read_mem(traits::MEM, offs, 32),32,false),32,true));
auto returnValue = std::make_tuple(CONT);
tu.close_scope();
vm_base<ARCH>::gen_sync(tu, POST_SYNC,58);
@ -2380,8 +2339,8 @@ private:
pc=pc+ 2;
gen_set_pc(tu, pc, traits::NEXT_PC);
tu.open_scope();
auto load_address = tu.assignment(tu.ext((tu.add(tu.load(rs1+ 8+ traits::X0, 0),tu.constant(uimm,8))),32,true),32);
tu.write_mem(traits::MEM, load_address, tu.ext(tu.load(rs2+ 8+ traits::X0, 0),32,true));
auto offs = tu.assignment(tu.ext((tu.add(tu.load(rs1+ 8+ traits::X0, 0),tu.constant(uimm,8))),32,true),32);
tu.write_mem(traits::MEM, offs, tu.ext(tu.load(rs2+ 8+ traits::X0, 0),32,true));
auto returnValue = std::make_tuple(CONT);
tu.close_scope();
vm_base<ARCH>::gen_sync(tu, POST_SYNC,59);
@ -2898,8 +2857,7 @@ private:
}
else{
auto offs = tu.assignment(tu.ext((tu.add(tu.load(2+ traits::X0, 0),tu.constant(uimm,8))),32,true),32);
auto res = tu.assignment(tu.ext(tu.read_mem(traits::MEM, offs, 32),32,false),32);
tu.store(rd + traits::X0,tu.ext(res,32,true));
tu.store(rd + traits::X0,tu.ext(tu.ext(tu.read_mem(traits::MEM, offs, 32),32,false),32,true));
}
auto returnValue = std::make_tuple(CONT);
tu.close_scope();
@ -2957,7 +2915,7 @@ private:
gen_set_pc(tu, pc, traits::NEXT_PC);
tu.open_scope();
if(rs1&&rs1<static_cast<uint32_t>(traits:: RFS)) {
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(tu.load(rs1%static_cast<uint32_t>(traits:: RFS)+ traits::X0, 0),tu.constant(~ 0x1,8)),32);
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(tu.load(rs1%static_cast<uint32_t>(traits:: RFS)+ traits::X0, 0),tu.constant(~0x1,8)),32);
tu.store(traits::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
}
@ -3045,7 +3003,7 @@ private:
else{
auto new_pc = tu.assignment(tu.load(rs1+ traits::X0, 0),32);
tu.store(1 + traits::X0,tu.ext((tu.add(tu.ext(cur_pc_val,32,false),tu.constant( 2,8))),32,true));
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~ 0x1,8)),32);
auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~0x1,8)),32);
tu.store(traits::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
}
@ -3138,11 +3096,64 @@ private:
vm_impl::gen_trap_check(tu);
return BRANCH;
}
//decoding functionality
void populate_decoding_tree(decoding_tree_node* root){
//create submask
for(auto instr: root->instrs){
root->submask &= instr.mask;
}
//put each instr according to submask&encoding into children
for(auto instr: root->instrs){
bool foundMatch = false;
for(auto child: root->children){
//use value as identifying trait
if(child->value == (instr.value&root->submask)){
child->instrs.push_back(instr);
foundMatch = true;
}
}
if(!foundMatch){
decoding_tree_node* child = new decoding_tree_node(instr.value&root->submask);
child->instrs.push_back(instr);
root->children.push_back(child);
}
}
root->instrs.clear();
//call populate_decoding_tree for all children
if(root->children.size() >1)
for(auto child: root->children){
populate_decoding_tree(child);
}
else{
//sort instrs by value of the mask, this works bc we want to have the least restrictive one last
std::sort(root->children[0]->instrs.begin(), root->children[0]->instrs.end(), [](const instruction_descriptor& instr1, const instruction_descriptor& instr2) {
return instr1.mask > instr2.mask;
});
}
}
compile_func decode_instr(decoding_tree_node* node, code_word_t word){
if(!node->children.size()){
if(node->instrs.size() == 1) return node->instrs[0].op;
for(auto instr : node->instrs){
if((instr.mask&word) == instr.value) return instr.op;
}
}
else{
for(auto child : node->children){
if (child->value == (node->submask&word)){
return decode_instr(child, word);
}
}
}
return nullptr;
}
};
template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
volatile CODE_WORD x = insn;
insn = 2 * x;
template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
volatile CODE_WORD x = instr;
instr = 2 * x;
}
template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@ -3150,14 +3161,11 @@ template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
template <typename ARCH>
vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id)
: vm_base<ARCH>(core, core_id, cluster_id) {
qlut[0] = lut_00.data();
qlut[1] = lut_01.data();
qlut[2] = lut_10.data();
qlut[3] = lut_11.data();
for (auto instr : instr_descr) {
auto quantrant = instr.value & 0x3;
expand_bit_mask(29, lutmasks[quantrant], instr.value >> 2, instr.mask >> 2, 0, qlut[quantrant], instr.op);
root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for(auto instr:instr_descr){
root->instrs.push_back(instr);
}
populate_decoding_tree(root);
}
template <typename ARCH>
@ -3165,30 +3173,19 @@ std::tuple<continuation_e>
vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) {
// we fetch at max 4 byte, alignment is 2
enum {TRAP_ID=1<<16};
code_word_t insn = 0;
// const typename traits::addr_t upper_bits = ~traits::PGMASK;
code_word_t instr = 0;
phys_addr_t paddr(pc);
auto *const data = (uint8_t *)&insn;
paddr = this->core.v2p(pc);
// if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
// auto res = this->core.read(paddr, 2, data);
// if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
// if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
// res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
// }
// } else {
auto res = this->core.read(paddr, 4, data);
if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
// }
if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
auto res = this->core.read(paddr, 4, reinterpret_cast<uint8_t*>(&instr));
if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
if (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
// curr pc on stack
++inst_cnt;
auto lut_val = extract_fields(insn);
auto f = qlut[insn & 0x3][lut_val];
auto f = decode_instr(root, instr);
if (f == nullptr) {
f = &this_class::illegal_intruction;
}
return (this->*f)(pc, insn, tu);
return (this->*f)(pc, instr, tu);
}
template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {