Merge branch 'develop' of

https://git.minres.com/DBT-RISE/DBT-RISE-TGC.git into develop
This commit is contained in:
Eyck Jentzsch 2023-07-30 09:14:58 +02:00
commit e68f9c573f
6 changed files with 336 additions and 215 deletions

View File

@ -1,10 +1,12 @@
cmake_minimum_required(VERSION 3.12) cmake_minimum_required(VERSION 3.12)
list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
############################################################################### ###############################################################################
# #
############################################################################### ###############################################################################
project(dbt-rise-tgc VERSION 1.0.0) project(dbt-rise-tgc VERSION 1.0.0)
include(GNUInstallDirs) include(GNUInstallDirs)
include(flink)
find_package(elfio QUIET) find_package(elfio QUIET)
find_package(Boost COMPONENTS coroutine) find_package(Boost COMPONENTS coroutine)
@ -70,7 +72,7 @@ if(TARGET RapidJSON OR TARGET RapidJSON::RapidJSON)
endif() endif()
# Define the library # Define the library
add_library(${PROJECT_NAME} ${LIB_SOURCES}) add_library(${PROJECT_NAME} SHARED ${LIB_SOURCES})
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU") if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
target_compile_options(${PROJECT_NAME} PRIVATE -Wno-shift-count-overflow) target_compile_options(${PROJECT_NAME} PRIVATE -Wno-shift-count-overflow)
@ -85,11 +87,9 @@ if(TARGET jsoncpp::jsoncpp)
else() else()
target_link_libraries(${PROJECT_NAME} PUBLIC jsoncpp) target_link_libraries(${PROJECT_NAME} PUBLIC jsoncpp)
endif() endif()
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND BUILD_SHARED_LIBS)
target_link_libraries(${PROJECT_NAME} PUBLIC -Wl,--whole-archive dbt-rise-core -Wl,--no-whole-archive) target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-core)
else()
target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-core)
endif()
if(TARGET elfio::elfio) if(TARGET elfio::elfio)
target_link_libraries(${PROJECT_NAME} PUBLIC elfio::elfio) target_link_libraries(${PROJECT_NAME} PUBLIC elfio::elfio)
else() else()
@ -164,7 +164,7 @@ if(WITH_TCC)
target_compile_definitions(${PROJECT_NAME} PRIVATE WITH_TCC) target_compile_definitions(${PROJECT_NAME} PRIVATE WITH_TCC)
endif() endif()
# Links the target exe against the libraries # Links the target exe against the libraries
target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-tgc) target_force_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-tgc)
if(TARGET Boost::program_options) if(TARGET Boost::program_options)
target_link_libraries(${PROJECT_NAME} PUBLIC Boost::program_options) target_link_libraries(${PROJECT_NAME} PUBLIC Boost::program_options)
else() else()

35
cmake/flink.cmake Normal file
View File

@ -0,0 +1,35 @@
# according to https://github.com/horance-liu/flink.cmake/tree/master
# SPDX-License-Identifier: Apache-2.0
include(CMakeParseArguments)
function(target_do_force_link_libraries target visibility lib)
if(MSVC)
target_link_libraries(${target} ${visibility} "/WHOLEARCHIVE:${lib}")
elseif(APPLE)
target_link_libraries(${target} ${visibility} -Wl,-force_load ${lib})
else()
target_link_libraries(${target} ${visibility} -Wl,--whole-archive ${lib} -Wl,--no-whole-archive)
endif()
endfunction()
function(target_force_link_libraries target)
cmake_parse_arguments(FLINK
""
""
"PUBLIC;INTERFACE;PRIVATE"
${ARGN}
)
foreach(lib IN LISTS FLINK_PUBLIC)
target_do_force_link_libraries(${target} PUBLIC ${lib})
endforeach()
foreach(lib IN LISTS FLINK_INTERFACE)
target_do_force_link_libraries(${target} INTERFACE ${lib})
endforeach()
foreach(lib IN LISTS FLINK_PRIVATE)
target_do_force_link_libraries(${target} PRIVATE ${lib})
endforeach()
endfunction()

View File

@ -158,30 +158,81 @@ private:
/**************************************************************************** /****************************************************************************
* start opcode definitions * start opcode definitions
****************************************************************************/ ****************************************************************************/
struct InstructionDesriptor { struct instruction_descriptor {
size_t length; size_t length;
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
typename arch::traits<ARCH>::opcode_e op; typename arch::traits<ARCH>::opcode_e 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, ${instructions.size}> instr_descr = {{
/* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %> /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
{${instr.length}, ${instr.encoding}, ${instr.mask}, arch::traits<ARCH>::opcode_e::${instr.instruction.name}},<%}%> {${instr.length}, ${instr.encoding}, ${instr.mask}, arch::traits<ARCH>::opcode_e::${instr.instruction.name}},<%}%>
}}; }};
//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);
//if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
// if (this->core.read(phys_pc, 2, data) != iss::Ok) return iss::Err;
// if ((data[0] & 0x3) == 0x3) // this is a 32bit instruction
// if (this->core.read(this->core.v2p(pc + 2), 2, data + 2) != iss::Ok) return iss::Err;
//} else {
if (this->core.read(phys_pc, 4, data) != iss::Ok) return iss::Err; if (this->core.read(phys_pc, 4, data) != iss::Ok) return iss::Err;
//}
return iss::Ok; return iss::Ok;
} }
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;
});
}
}
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){
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 arch::traits<ARCH>::opcode_e::MAX_OPCODE;
}
}; };
template <typename CODE_WORD> void debug_fn(CODE_WORD insn) { template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
@ -208,16 +259,11 @@ 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; root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for (auto instr : instr_descr) { for(auto instr:instr_descr){
auto quadrant = instr.value & 0x3; root->instrs.push_back(instr);
qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op});
}
for(auto& lut: qlut){
std::sort(std::begin(lut), std::end(lut), [](instruction_pattern const& a, instruction_pattern const& b){
return bit_count(a.mask) > bit_count(b.mask);
});
} }
populate_decoding_tree(root);
} }
inline bool is_count_limit_enabled(finish_cond_e cond){ inline bool is_count_limit_enabled(finish_cond_e cond){
@ -228,14 +274,6 @@ inline bool is_jump_to_self_enabled(finish_cond_e cond){
return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF; return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF;
} }
template <typename ARCH>
typename arch::traits<ARCH>::opcode_e vm_impl<ARCH>::decode_inst_id(code_word_t instr){
for(auto& e: qlut[instr&0x3]){
if(!((instr&e.mask) ^ e.value )) return e.id;
}
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){
auto pc=start; auto pc=start;
@ -257,7 +295,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
} else { } else {
if (is_jump_to_self_enabled(cond) && if (is_jump_to_self_enabled(cond) &&
(instr == 0x0000006f || (instr&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 inst_id = decode_inst_id(instr); auto inst_id = decode_instr(root, instr);
// pre execution stuff // pre execution stuff
this->core.reg.last_branch = 0; this->core.reg.last_branch = 0;
if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id)); if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id));

View File

@ -120,57 +120,7 @@ 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> template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
inline S sext(U from) { inline S sext(U from) {
auto mask = (1ULL<<W) - 1; auto mask = (1ULL<<W) - 1;
@ -182,14 +132,23 @@ private:
/**************************************************************************** /****************************************************************************
* start opcode definitions * start opcode definitions
****************************************************************************/ ****************************************************************************/
struct InstructionDesriptor { struct instruction_descriptor {
size_t length; size_t length;
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
compile_func op; 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 -> %> /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
/* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */ /* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */
{${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%> {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
@ -227,11 +186,64 @@ private:
vm_impl::gen_trap_check(tu); vm_impl::gen_trap_check(tu);
return BRANCH; 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) { template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
volatile CODE_WORD x = insn; volatile CODE_WORD x = instr;
insn = 2 * x; instr = 2 * x;
} }
template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); } template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@ -239,14 +251,11 @@ template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
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) {
qlut[0] = lut_00.data(); root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
qlut[1] = lut_01.data(); for(auto instr:instr_descr){
qlut[2] = lut_10.data(); root->instrs.push_back(instr);
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);
} }
populate_decoding_tree(root);
} }
template <typename ARCH> template <typename ARCH>
@ -254,30 +263,19 @@ std::tuple<continuation_e>
vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) { 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 // we fetch at max 4 byte, alignment is 2
enum {TRAP_ID=1<<16}; enum {TRAP_ID=1<<16};
code_word_t insn = 0; code_word_t instr = 0;
// const typename traits::addr_t upper_bits = ~traits::PGMASK;
phys_addr_t paddr(pc); phys_addr_t paddr(pc);
auto *const data = (uint8_t *)&insn;
paddr = this->core.v2p(pc); 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, 4, reinterpret_cast<uint8_t*>(&instr));
// 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 (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'
if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
// curr pc on stack // curr pc on stack
++inst_cnt; ++inst_cnt;
auto lut_val = extract_fields(insn); auto f = decode_instr(root, instr);
auto f = qlut[insn & 0x3][lut_val];
if (f == nullptr) { if (f == nullptr) {
f = &this_class::illegal_intruction; 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) { template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {

View File

@ -152,14 +152,22 @@ private:
/**************************************************************************** /****************************************************************************
* start opcode definitions * start opcode definitions
****************************************************************************/ ****************************************************************************/
struct InstructionDesriptor { struct instruction_descriptor {
size_t length; size_t length;
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
typename arch::traits<ARCH>::opcode_e op; typename arch::traits<ARCH>::opcode_e 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 */ /* entries are: size, valid value, valid mask, function ptr */
{32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::LUI}, {32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::LUI},
{32, 0b00000000000000000000000000010111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::AUIPC}, {32, 0b00000000000000000000000000010111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::AUIPC},
@ -250,7 +258,6 @@ private:
{16, 0b0000000000000000, 0b1111111111111111, arch::traits<ARCH>::opcode_e::DII}, {16, 0b0000000000000000, 0b1111111111111111, arch::traits<ARCH>::opcode_e::DII},
}}; }};
//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);
//TODO: re-add page handling //TODO: re-add page handling
@ -263,6 +270,56 @@ private:
//} //}
return iss::Ok; return iss::Ok;
} }
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;
});
}
}
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){
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 arch::traits<ARCH>::opcode_e::MAX_OPCODE;
}
}; };
template <typename CODE_WORD> void debug_fn(CODE_WORD insn) { template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
@ -289,16 +346,11 @@ 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; root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
for (auto instr : instr_descr) { for(auto instr:instr_descr){
auto quadrant = instr.value & 0x3; root->instrs.push_back(instr);
qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op});
}
for(auto& lut: qlut){
std::sort(std::begin(lut), std::end(lut), [](instruction_pattern const& a, instruction_pattern const& b){
return bit_count(a.mask) > bit_count(b.mask);
});
} }
populate_decoding_tree(root);
} }
inline bool is_count_limit_enabled(finish_cond_e cond){ inline bool is_count_limit_enabled(finish_cond_e cond){
@ -309,14 +361,6 @@ inline bool is_jump_to_self_enabled(finish_cond_e cond){
return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF; return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF;
} }
template <typename ARCH>
typename arch::traits<ARCH>::opcode_e vm_impl<ARCH>::decode_inst_id(code_word_t instr){
for(auto& e: qlut[instr&0x3]){
if(!((instr&e.mask) ^ e.value )) return e.id;
}
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){
auto pc=start; auto pc=start;
@ -338,7 +382,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
} else { } else {
if (is_jump_to_self_enabled(cond) && if (is_jump_to_self_enabled(cond) &&
(instr == 0x0000006f || (instr&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 inst_id = decode_inst_id(instr); auto inst_id = decode_instr(root, instr);
// pre execution stuff // pre execution stuff
this->core.reg.last_branch = 0; this->core.reg.last_branch = 0;
if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id)); if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, static_cast<unsigned>(inst_id));

View File

@ -120,57 +120,7 @@ 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> template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
inline S sext(U from) { inline S sext(U from) {
auto mask = (1ULL<<W) - 1; auto mask = (1ULL<<W) - 1;
@ -182,14 +132,23 @@ private:
/**************************************************************************** /****************************************************************************
* start opcode definitions * start opcode definitions
****************************************************************************/ ****************************************************************************/
struct InstructionDesriptor { struct instruction_descriptor {
size_t length; size_t length;
uint32_t value; uint32_t value;
uint32_t mask; uint32_t mask;
compile_func op; 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 */ /* entries are: size, valid value, valid mask, function ptr */
/* instruction LUI, encoding '0b00000000000000000000000000110111' */ /* instruction LUI, encoding '0b00000000000000000000000000110111' */
{32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, &this_class::__lui}, {32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, &this_class::__lui},
@ -493,14 +452,14 @@ private:
this->gen_raise_trap(tu, 0, 2); this->gen_raise_trap(tu, 0, 2);
} }
else{ 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))); tu.open_if(tu.srem(new_pc,tu.constant(static_cast<uint32_t>(traits:: INSTR_ALIGNMENT),32)));
this->gen_raise_trap(tu, 0, 0); this->gen_raise_trap(tu, 0, 0);
tu.open_else(); tu.open_else();
if(rd!= 0) { 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)); 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::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2)); tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
tu.close_scope(); tu.close_scope();
@ -1962,7 +1921,7 @@ private:
else{ else{
auto xrd = tu.assignment(tu.read_mem(traits::CSR, csr, 32),32); auto xrd = tu.assignment(tu.read_mem(traits::CSR, csr, 32),32);
if(zimm!= 0) { 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) { if(rd!= 0) {
tu.store(rd + traits::X0,xrd); tu.store(rd + traits::X0,xrd);
@ -2163,13 +2122,13 @@ private:
auto divisor = tu.assignment(tu.ext(tu.load(rs2+ traits::X0, 0),32,false),32); 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))); 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); 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.store(rd + traits::X0,MMIN);
tu.open_else(); tu.open_else();
tu.store(rd + traits::X0,tu.ext((tu.sdiv(dividend,divisor)),32,true)); tu.store(rd + traits::X0,tu.ext((tu.sdiv(dividend,divisor)),32,true));
tu.close_scope(); tu.close_scope();
tu.open_else(); 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(); tu.close_scope();
} }
} }
@ -2208,7 +2167,7 @@ private:
} }
tu.open_else(); tu.open_else();
if(rd!=0) { 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(); tu.close_scope();
} }
@ -2243,7 +2202,7 @@ private:
else{ else{
tu.open_if(tu.icmp(ICmpInst::ICMP_NE,tu.load(rs2+ traits::X0, 0),tu.constant( 0,8))); 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); 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) { if(rd!=0) {
tu.store(rd + traits::X0,tu.constant( 0,8)); tu.store(rd + traits::X0,tu.constant( 0,8));
} }
@ -2955,7 +2914,7 @@ private:
gen_set_pc(tu, pc, traits::NEXT_PC); gen_set_pc(tu, pc, traits::NEXT_PC);
tu.open_scope(); tu.open_scope();
if(rs1&&rs1<static_cast<uint32_t>(traits:: RFS)) { 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::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2)); tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
} }
@ -3043,7 +3002,7 @@ private:
else{ else{
auto new_pc = tu.assignment(tu.load(rs1+ traits::X0, 0),32); 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)); 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::NEXT_PC, PC_val_v);
tu.store(traits::LAST_BRANCH, tu.constant(2U, 2)); tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
} }
@ -3136,11 +3095,64 @@ private:
vm_impl::gen_trap_check(tu); vm_impl::gen_trap_check(tu);
return BRANCH; 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) { template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
volatile CODE_WORD x = insn; volatile CODE_WORD x = instr;
insn = 2 * x; instr = 2 * x;
} }
template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); } template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@ -3148,14 +3160,11 @@ template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
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) {
qlut[0] = lut_00.data(); root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
qlut[1] = lut_01.data(); for(auto instr:instr_descr){
qlut[2] = lut_10.data(); root->instrs.push_back(instr);
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);
} }
populate_decoding_tree(root);
} }
template <typename ARCH> template <typename ARCH>
@ -3163,10 +3172,8 @@ std::tuple<continuation_e>
vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) { 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 // we fetch at max 4 byte, alignment is 2
enum {TRAP_ID=1<<16}; enum {TRAP_ID=1<<16};
code_word_t insn = 0; code_word_t instr = 0;
// const typename traits::addr_t upper_bits = ~traits::PGMASK;
phys_addr_t paddr(pc); phys_addr_t paddr(pc);
auto *const data = (uint8_t *)&insn;
paddr = this->core.v2p(pc); paddr = this->core.v2p(pc);
//TODO: re-add page handling //TODO: re-add page handling
// if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary // if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
@ -3176,18 +3183,17 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
// res = this->core.read(this->core.v2p(pc + 2), 2, data + 2); // res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
// } // }
// } else { // } else {
auto res = this->core.read(paddr, 4, data); auto res = this->core.read(paddr, 4, reinterpret_cast<uint8_t*>(&instr));
if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val); 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' if (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
// curr pc on stack // curr pc on stack
++inst_cnt; ++inst_cnt;
auto lut_val = extract_fields(insn); auto f = decode_instr(root, instr);
auto f = qlut[insn & 0x3][lut_val];
if (f == nullptr) { if (f == nullptr) {
f = &this_class::illegal_intruction; 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) { template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {