refactored interpreter backend structure

This commit is contained in:
Eyck Jentzsch 2022-03-05 20:59:17 +01:00
parent b37ef973de
commit 521f40a3d6
2 changed files with 2208 additions and 4020 deletions

View File

@ -94,7 +94,7 @@ protected:
inline const char *name(size_t index){return traits::reg_aliases.at(index);} inline const char *name(size_t index){return traits::reg_aliases.at(index);}
compile_func decode_inst(code_word_t instr) ; typename arch::traits<ARCH>::opcode_e decode_inst_id(code_word_t instr);
virt_addr_t execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit) override; virt_addr_t execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit) override;
// some compile time constants // some compile time constants
@ -114,7 +114,7 @@ protected:
struct instruction_pattern { struct instruction_pattern {
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
compile_func opc; typename arch::traits<ARCH>::opcode_e id;
}; };
std::array<std::vector<instruction_pattern>, 4> qlut; std::array<std::vector<instruction_pattern>, 4> qlut;
@ -201,74 +201,14 @@ private:
size_t length; size_t length;
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
compile_func op; typename arch::traits<ARCH>::opcode_e op;
}; };
const std::array<InstructionDesriptor, ${instructions.size}> instr_descr = {{ const std::array<InstructionDesriptor, ${instructions.size}> instr_descr = {{
/* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %> /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
/* instruction ${instr.instruction.name} */ {${instr.length}, ${instr.encoding}, ${instr.mask}, arch::traits<ARCH>::opcode_e::${instr.instruction.name}},<%}%>
{${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
}}; }};
/* instruction definitions */<%instructions.eachWithIndex{instr, idx -> %>
/* instruction ${idx}: ${instr.name} */
compile_ret_t __${generator.functionName(instr.name)}(virt_addr_t& pc, code_word_t instr){
// pre execution stuff
if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, ${idx});
<%instr.fields.eachLine{%>${it}
<%}%>if(this->disass_enabled){
/* generate console output when executing the command */
<%instr.disass.eachLine{%>${it}
<%}%>
}
auto* PC = reinterpret_cast<uint${addrDataWidth}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::PC]);
auto* NEXT_PC = reinterpret_cast<uint${addrDataWidth}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::NEXT_PC]);
auto* trap_state = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::TRAP_STATE]);
// used registers<%instr.usedVariables.each{ k,v->
if(v.isArray) {%>
auto* ${k} = reinterpret_cast<uint${nativeTypeSize(v.type.size)}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::${k}0]);<% }else{ %>
auto* ${k} = reinterpret_cast<uint${nativeTypeSize(v.type.size)}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::${k}]);
<%}}%>// calculate next pc value
*NEXT_PC = *PC + ${instr.length/8};
// execute instruction
try {
<%instr.behavior.eachLine{%>${it}
<%}%>} catch(...){}
// post execution stuff
process_spawn_blocks();
if(this->sync_exec && POST_SYNC) this->do_sync(POST_SYNC, ${idx});
// trap check
if(*trap_state!=0){
super::core.enter_trap(*trap_state, pc.val, instr);
} else {
(*reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::ICOUNT]))++;
(*reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::INSTRET]))++;
}
(*reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::CYCLE]))++;
pc.val=*NEXT_PC;
return pc;
}
<%}%>
/****************************************************************************
* end opcode definitions
****************************************************************************/
compile_ret_t illegal_intruction(virt_addr_t &pc, code_word_t instr) {
this->do_sync(PRE_SYNC, static_cast<unsigned>(arch::traits<ARCH>::opcode_e::MAX_OPCODE));
uint32_t* PC = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::PC]);
uint32_t* NEXT_PC = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::NEXT_PC]);
*NEXT_PC = *PC + ((instr & 3) == 3 ? 4 : 2);
raise(0, 2);
// post execution stuff
if(this->sync_exec && POST_SYNC) this->do_sync(POST_SYNC, static_cast<unsigned>(arch::traits<ARCH>::opcode_e::MAX_OPCODE));
auto* trap_state = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::TRAP_STATE]);
// trap check
if(*trap_state!=0){
super::core.enter_trap(*trap_state, pc.val, instr);
}
pc.val=*NEXT_PC;
return pc;
}
//static constexpr typename traits::addr_t upper_bits = ~traits::PGMASK; //static constexpr typename traits::addr_t upper_bits = ~traits::PGMASK;
iss::status fetch_ins(virt_addr_t pc, uint8_t * data){ iss::status fetch_ins(virt_addr_t pc, uint8_t * data){
auto phys_pc = this->core.v2p(pc); auto phys_pc = this->core.v2p(pc);
@ -307,6 +247,7 @@ constexpr size_t bit_count(uint32_t u) {
template <typename ARCH> template <typename ARCH>
vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id) vm_impl<ARCH>::vm_impl(ARCH &core, unsigned core_id, unsigned cluster_id)
: vm_base<ARCH>(core, core_id, cluster_id) { : vm_base<ARCH>(core, core_id, cluster_id) {
unsigned id=0;
for (auto instr : instr_descr) { for (auto instr : instr_descr) {
auto quadrant = instr.value & 0x3; auto quadrant = instr.value & 0x3;
qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op}); qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op});
@ -327,31 +268,74 @@ inline bool is_jump_to_self_enabled(finish_cond_e cond){
} }
template <typename ARCH> template <typename ARCH>
typename vm_impl<ARCH>::compile_func vm_impl<ARCH>::decode_inst(code_word_t instr){ typename arch::traits<ARCH>::opcode_e vm_impl<ARCH>::decode_inst_id(code_word_t instr){
for(auto& e: qlut[instr&0x3]){ for(auto& e: qlut[instr&0x3]){
if(!((instr&e.mask) ^ e.value )) return e.opc; if(!((instr&e.mask) ^ e.value )) return e.id;
} }
return &this_class::illegal_intruction; return arch::traits<ARCH>::opcode_e::MAX_OPCODE;
} }
template <typename ARCH> template <typename ARCH>
typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit){ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit){
// we fetch at max 4 byte, alignment is 2 // we fetch at max 4 byte, alignment is 2
code_word_t insn = 0; code_word_t instr = 0;
auto *const data = (uint8_t *)&insn; auto *const data = (uint8_t *)&instr;
auto pc=start; auto pc=start;
auto* PC = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::PC]);
auto* NEXT_PC = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::NEXT_PC]);
auto* trap_state = reinterpret_cast<uint32_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::TRAP_STATE]);
auto* icount = reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::ICOUNT]);
auto* instret = reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::INSTRET]);
while(!this->core.should_stop() && while(!this->core.should_stop() &&
!(is_count_limit_enabled(cond) && this->core.get_icount() >= icount_limit)){ !(is_count_limit_enabled(cond) && this->core.get_icount() >= icount_limit)){
auto res = fetch_ins(pc, data); if(fetch_ins(pc, data)!=iss::Ok){
if(res!=iss::Ok){
this->do_sync(POST_SYNC, std::numeric_limits<unsigned>::max()); this->do_sync(POST_SYNC, std::numeric_limits<unsigned>::max());
pc.val = super::core.enter_trap(std::numeric_limits<uint64_t>::max(), pc.val, 0); pc.val = super::core.enter_trap(std::numeric_limits<uint64_t>::max(), pc.val, 0);
} else { } else {
if (is_jump_to_self_enabled(cond) && if (is_jump_to_self_enabled(cond) &&
(insn == 0x0000006f || (insn&0xffff)==0xa001)) throw simulation_stopped(0); // 'J 0' or 'C.J 0' (instr == 0x0000006f || (instr&0xffff)==0xa001)) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
auto f = decode_inst(insn); auto inst_id = decode_inst_id(instr);
auto old_pc = pc.val; // pre execution stuff
pc = (this->*f)(pc, insn); if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id));
switch(inst_id){<%instructions.eachWithIndex{instr, idx -> %>
case arch::traits<ARCH>::opcode_e::${instr.name}: {
<%instr.fields.eachLine{%>${it}
<%}%>if(this->disass_enabled){
/* generate console output when executing the command */
<%instr.disass.eachLine{%>${it}
<%}%>
}
// used registers<%instr.usedVariables.each{ k,v->
if(v.isArray) {%>
auto* ${k} = reinterpret_cast<uint${nativeTypeSize(v.type.size)}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::${k}0]);<% }else{ %>
auto* ${k} = reinterpret_cast<uint${nativeTypeSize(v.type.size)}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::${k}]);
<%}}%>// calculate next pc value
*NEXT_PC = *PC + ${instr.length/8};
// execute instruction
try {
<%instr.behavior.eachLine{%>${it}
<%}%>} catch(...){}
}
break;<%}%>
default: {
*NEXT_PC = *PC + ((instr & 3) == 3 ? 4 : 2);
raise(0, 2);
}
}
// post execution stuff
process_spawn_blocks();
if(this->sync_exec && POST_SYNC) this->do_sync(POST_SYNC, 65);
// trap check
if(*trap_state!=0){
super::core.enter_trap(*trap_state, pc.val, instr);
} else {
(*icount)++;
(*instret)++;
}
(*reinterpret_cast<uint64_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::CYCLE]))++;
pc.val=*NEXT_PC;
this->core.reg.PC = this->core.reg.NEXT_PC; this->core.reg.PC = this->core.reg.NEXT_PC;
this->core.reg.trap_state = this->core.reg.pending_trap; this->core.reg.trap_state = this->core.reg.pending_trap;
} }
@ -359,7 +343,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
return pc; return pc;
} }
} // namespace mnrv32 }
template <> template <>
std::unique_ptr<vm_if> create<arch::${coreDef.name.toLowerCase()}>(arch::${coreDef.name.toLowerCase()} *core, unsigned short port, bool dump) { std::unique_ptr<vm_if> create<arch::${coreDef.name.toLowerCase()}>(arch::${coreDef.name.toLowerCase()} *core, unsigned short port, bool dump) {

File diff suppressed because it is too large Load Diff