adds support for compressed instructions

CMakeLists.txt

@@ -83,6 +83,10 @@ elseif(TARGET elfio::elfio)
 else()
     message(FATAL_ERROR "No elfio library found, maybe a find_package() call is missing")
 endif()
+if(TARGET lz4::lz4)
+    target_compile_definitions(${PROJECT_NAME} PUBLIC WITH_LZ4)
+    target_link_libraries(${PROJECT_NAME} PUBLIC lz4::lz4)
+endif()
 if(TARGET RapidJSON)
     target_link_libraries(${PROJECT_NAME} PUBLIC RapidJSON)
 endif()

gen_input/templates/CORENAME.h.gtl

@@ -163,13 +163,13 @@ struct ${coreDef.name.toLowerCase()}: public arch_if {
         uint${byteSize(reg.size)}_t ${reg.name} = 0;<%
         }}%>
     } reg;
+#pragma pack(pop)
     uint32_t trap_state = 0, pending_trap = 0;
     uint64_t icount = 0;
     uint64_t cycle = 0;
     uint64_t instret = 0;
     uint32_t instruction = 0;
     uint32_t last_branch = 0;
-#pragma pack(pop)
     std::array<address_type, 4> addr_mode;
     
     uint64_t interrupt_sim=0;

gen_input/templates/interp/CORENAME.cpp.gtl

@@ -314,11 +314,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
 		        <%}}%>// calculate next pc value
 		        *NEXT_PC = *PC + ${instr.length/8};
 		        // execute instruction
-		        try {
 		        <%instr.behavior.eachLine{%>${it}
-		        <%}%>} catch(...){}
+		        <%}%>TRAP_${instr.name}:break;
-	    	}
+	    	}// @suppress("No break at end of case")<%}%>
-	    	break;<%}%>
             default: {
                 *NEXT_PC = *PC + ((instr & 3) == 3 ? 4 : 2);
                 raise(0,  2);

incl/iss/arch/riscv_hart_m_p.h

@@ -567,22 +567,22 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read(const address_type type, const acce
     try {
         switch (space) {
         case traits<BASE>::MEM: {
-            if (unlikely((access == iss::access_type::FETCH || access == iss::access_type::DEBUG_FETCH) && (addr & 0x1) == 1)) {
+            auto alignment = is_fetch(access)? (traits<BASE>::MISA_VAL&0x100? 2 : 4) : length;
+            if (unlikely(is_fetch(access) && (addr&(alignment-1)))) {
                 fault_data = addr;
-                if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
+                if (is_debug(access)) throw trap_access(0, addr);
                 this->trap_state = (1 << 31); // issue trap 0
                 return iss::Err;
             }
             try {
-                auto alignment = access == iss::access_type::FETCH? (traits<BASE>::MISA_VAL&0x100? 2 : 4) : length;
+                if(!is_debug(access) && (addr&(alignment-1))){
-                if(alignment>1 && (addr&(alignment-1))){
                     this->trap_state = 1<<31 | 4<<16;
                     fault_data=addr;
                     return iss::Err;
                 }
                 auto phys_addr = type==iss::address_type::PHYSICAL?phys_addr_t{access, space, addr}:BASE::v2p(iss::addr_t{access, type, space, addr});
                 auto res = iss::Err;
-                if(access != access_type::FETCH && memfn_range.size()){
+                if(!is_fetch(access) && memfn_range.size()){
                     auto it = std::find_if(std::begin(memfn_range), std::end(memfn_range), [phys_addr](std::tuple<uint64_t, uint64_t> const& a){
                         return std::get<0>(a)<=phys_addr.val && (std::get<0>(a)+std::get<1>(a))>phys_addr.val;
                     });

incl/iss/arch/riscv_hart_msu_vp.h

@@ -611,13 +611,19 @@ iss::status riscv_hart_msu_vp<BASE>::read(const address_type type, const access_
     try {
         switch (space) {
         case traits<BASE>::MEM: {
-            if (unlikely((access == iss::access_type::FETCH || access == iss::access_type::DEBUG_FETCH) && (addr & 0x1) == 1)) {
+            auto alignment = is_fetch(access)? (traits<BASE>::MISA_VAL&0x100? 2 : 4) : length;
+            if (unlikely(is_fetch(access) && (addr&(alignment-1)))) {
                 fault_data = addr;
                 if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
                 this->trap_state = (1 << 31); // issue trap 0
                 return iss::Err;
             }
             try {
+                if(!is_debug(access)  && (addr&(alignment-1))){
+                    this->trap_state = 1<<31 | 4<<16;
+                    fault_data=addr;
+                    return iss::Err;
+                }
                 if (unlikely((addr & ~PGMASK) != ((addr + length - 1) & ~PGMASK))) { // we may cross a page boundary
                     vm_info vm = hart_state_type::decode_vm_info(this->reg.PRIV, state.satp);
                     if (vm.levels != 0) { // VM is active

incl/iss/arch/riscv_hart_mu_p.h

@@ -705,22 +705,22 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::read(const address_type type, const acc
         switch (space) {
         case traits<BASE>::MEM: {
             if(FEAT & FEAT_PMP){
-                if(!pmp_check(access, addr, length) && (access&access_type::DEBUG) != access_type::DEBUG) {
+                if(!pmp_check(access, addr, length) && !is_debug(access)) {
                     fault_data = addr;
-                    if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
+                    if (is_debug(access)) throw trap_access(0, addr);
                     this->trap_state = (1 << 31) | ((access==access_type::FETCH?1:5) << 16); // issue trap 1
                     return iss::Err;
                 }
             }
-            if (unlikely((access == iss::access_type::FETCH || access == iss::access_type::DEBUG_FETCH) && (addr & 0x1) == 1)) {
+            auto alignment = is_fetch(access)? (traits<BASE>::MISA_VAL&0x100? 2 : 4) : length;
+            if (unlikely(is_fetch(access) && (addr&(alignment-1)))) {
                 fault_data = addr;
-                if (access && iss::access_type::DEBUG) throw trap_access(0, addr);
+                if (is_debug(access)) throw trap_access(0, addr);
                 this->trap_state = (1 << 31); // issue trap 0
                 return iss::Err;
             }
             try {
-                auto alignment = access == iss::access_type::FETCH? (traits<BASE>::MISA_VAL&0x100? 2 : 4) : length;
+                if(!is_debug(access)  && (addr&(alignment-1))){
-                if(alignment>1 && (addr&(alignment-1))){
                     this->trap_state = 1<<31 | 4<<16;
                     fault_data=addr;
                     return iss::Err;

incl/iss/arch/tgc_c.h

@@ -252,13 +252,13 @@ struct tgc_c: public arch_if {
         uint8_t PRIV = 0; 
         uint32_t DPC = 0;
     } reg;
+#pragma pack(pop)
     uint32_t trap_state = 0, pending_trap = 0;
     uint64_t icount = 0;
     uint64_t cycle = 0;
     uint64_t instret = 0;
     uint32_t instruction = 0;
     uint32_t last_branch = 0;
-#pragma pack(pop)
     std::array<address_type, 4> addr_mode;
     
     uint64_t interrupt_sim=0;

incl/iss/plugin/cycle_estimate.h

@@ -81,7 +81,7 @@ public:
     void callback(instr_info_t instr_info, exec_info const&) override;
 
 private:
-    iss::instrumentation_if *arch_instr;
+    iss::instrumentation_if *instr_if;
     std::vector<instr_desc> delays;
     struct pair_hash {
         size_t operator()(const std::pair<uint64_t, uint64_t> &p) const {

incl/iss/plugin/pctrace.h

@@ -44,7 +44,7 @@
 
 namespace iss {
 namespace plugin {
-
+class lz4compress_steambuf;
 class cov : public iss::vm_plugin {
     struct instr_delay {
         std::string instr_name;
@@ -88,10 +88,13 @@ public:
 private:
     iss::instrumentation_if *instr_if  {nullptr};
     std::ofstream output;
+#ifdef WITH_LZ4
+    std::unique_ptr<lz4compress_steambuf> strbuf;
+    std::ostream ostr;
+#endif
     std::string filename;
     std::vector<instr_desc> delays;
-    bool jumped, first;
+    bool jumped{false}, first{true};
-
 };
 }
 }

src/main.cpp

@@ -60,13 +60,13 @@ int main(int argc, char *argv[]) {
     desc.add_options()
         ("help,h", "Print help message")
         ("verbose,v", po::value<int>()->implicit_value(0), "Sets logging verbosity")
-        ("logfile,f", po::value<std::string>(), "Sets default log file.")
+        ("logfile,l", po::value<std::string>(), "Sets default log file.")
         ("disass,d", po::value<std::string>()->implicit_value(""), "Enables disassembly")
         ("gdb-port,g", po::value<unsigned>()->default_value(0), "enable gdb server and specify port to use")
         ("instructions,i", po::value<uint64_t>()->default_value(std::numeric_limits<uint64_t>::max()), "max. number of instructions to simulate")
         ("reset,r", po::value<std::string>(), "reset address")
         ("dump-ir", "dump the intermediate representation")
-        ("elf", po::value<std::vector<std::string>>(), "ELF file(s) to load")
+        ("elf,f", po::value<std::vector<std::string>>(), "ELF file(s) to load")
         ("mem,m", po::value<std::string>(), "the memory input file")
         ("plugin,p", po::value<std::vector<std::string>>(), "plugin to activate")
         ("backend", po::value<std::string>()->default_value("interp"), "the memory input file")

src/plugin/cycle_estimate.cpp

@@ -48,7 +48,7 @@ using namespace rapidjson;
 using namespace std;
 
 iss::plugin::cycle_estimate::cycle_estimate(string const& config_file_name)
-: arch_instr(nullptr)
+: instr_if(nullptr)
 , config_file_name(config_file_name)
 {
 }
@@ -57,9 +57,9 @@ iss::plugin::cycle_estimate::~cycle_estimate() {
 }
 
 bool iss::plugin::cycle_estimate::registration(const char* const version, vm_if& vm) {
-    arch_instr = vm.get_arch()->get_instrumentation_if();
+    instr_if = vm.get_arch()->get_instrumentation_if();
-    if(!arch_instr) return false;
+    if(!instr_if) return false;
-    const string  core_name = arch_instr->core_type_name();
+    const string  core_name = instr_if->core_type_name();
     if (config_file_name.length() > 0) {
         ifstream is(config_file_name);
         if (is.is_open()) {
@@ -108,11 +108,11 @@ bool iss::plugin::cycle_estimate::registration(const char* const version, vm_if&
 }
 
 void iss::plugin::cycle_estimate::callback(instr_info_t instr_info, exec_info const& exc_info) {
-    assert(arch_instr && "No instrumentation interface available but callback executed");
+    assert(instr_if && "No instrumentation interface available but callback executed");
     auto entry = delays[instr_info.instr_id];
     bool taken = exc_info.branch_taken;
     if (exc_info.branch_taken && (entry.taken > 1))
-        arch_instr->set_curr_instr_cycles(entry.taken);
+        instr_if->set_curr_instr_cycles(entry.taken);
     else if (entry.not_taken > 1)
-        arch_instr->set_curr_instr_cycles(entry.not_taken);
+        instr_if->set_curr_instr_cycles(entry.not_taken);
 }

src/plugin/pctrace.cpp

@@ -9,27 +9,99 @@
 #include <rapidjson/ostreamwrapper.h>
 #include <rapidjson/error/en.h>
 #include <fstream>
-
 #include <iostream>
+#ifdef WITH_LZ4
+#include <lz4frame.h>
+#endif
 
+namespace iss {
+namespace plugin {
 
 using namespace rapidjson;
 using namespace std;
 
-iss::plugin::cov::cov(std::string const &filename)
+#ifdef WITH_LZ4
-    : instr_if(nullptr)
+class lz4compress_steambuf: public std::streambuf {
-    , filename(filename)
+public:
-{
+    lz4compress_steambuf(const lz4compress_steambuf&) = delete;
-    output.open("output.trc");
+    lz4compress_steambuf& operator=(const lz4compress_steambuf&) = delete;
-    jumped = false;
+    lz4compress_steambuf(std::ostream &sink, size_t buf_size)
-    first = true;
+    : sink(sink)
-}
+    , src_buf(buf_size)
+    , dest_buf(LZ4F_compressBound(buf_size, nullptr))
+    {
+        auto errCode = LZ4F_createCompressionContext(&ctx, LZ4F_VERSION);
+        if (LZ4F_isError(errCode) != 0)
+            throw std::runtime_error(std::string("Failed to create LZ4 context: ") + LZ4F_getErrorName(errCode));
+        size_t ret = LZ4F_compressBegin(ctx, &dest_buf.front(), dest_buf.capacity(), nullptr);
+        if (LZ4F_isError(ret) != 0)
+            throw std::runtime_error(std::string("Failed to start LZ4 compression: ") + LZ4F_getErrorName(ret));
+        setp(src_buf.data(), src_buf.data() + src_buf.size() - 1);
+        sink.write(dest_buf.data(), ret);
+    }
 
-iss::plugin::cov::~cov() {
+    ~lz4compress_steambuf() {
-    output.close();
+        close();
-}
+    }
 
-bool iss::plugin::cov::registration(const char *const version, vm_if& vm) {
+    void close() {
+        if (closed)
+            return;
+        sync();
+        auto ret = LZ4F_compressEnd(ctx, dest_buf.data(), dest_buf.capacity(), nullptr);
+        if (LZ4F_isError(ret) != 0)
+            throw std::runtime_error(std::string("Failed to finish LZ4 compression: ") + LZ4F_getErrorName(ret));
+        sink.write(dest_buf.data(), ret);
+        LZ4F_freeCompressionContext(ctx);
+        closed = true;
+    }
+
+private:
+    int_type overflow(int_type ch) override {
+        compress_and_write();
+        *pptr() = static_cast<char_type>(ch);
+        pbump(1);
+        return ch;
+    }
+
+    int_type sync() override {
+        compress_and_write();
+        return 0;
+    }
+
+    void compress_and_write() {
+        if (closed)
+            throw std::runtime_error("Cannot write to closed stream");
+        if(auto orig_size = pptr() - pbase()){
+            auto ret = LZ4F_compressUpdate(ctx, dest_buf.data(), dest_buf.capacity(), pbase(), orig_size, nullptr);
+            if (LZ4F_isError(ret) != 0)
+                throw std::runtime_error(std::string("LZ4 compression failed: ") + LZ4F_getErrorName(ret));
+            if(ret) sink.write(dest_buf.data(), ret);
+            pbump(-orig_size);
+        }
+    }
+
+    std::ostream &sink;
+    std::vector<char> src_buf;
+    std::vector<char> dest_buf;
+    LZ4F_compressionContext_t ctx{ nullptr };
+    bool closed{ false };
+};
+#endif
+
+cov::cov(std::string const &filename)
+: instr_if(nullptr)
+, filename(filename)
+, output("output.trc")
+#ifdef WITH_LZ4
+, strbuf(new lz4compress_steambuf(output, 4096))
+, ostr(strbuf.get())
+#endif
+{ }
+
+cov::~cov() { }
+
+bool cov::registration(const char *const version, vm_if& vm) {
     instr_if = vm.get_arch()->get_instrumentation_if();
     if(!instr_if) return false;
     const string  core_name = instr_if->core_type_name();
@@ -78,60 +150,31 @@ bool iss::plugin::cov::registration(const char *const version, vm_if& vm) {
         }
     }
     return true;
-
-
-
 }
 
-inline string formatPC(uint64_t pc) {
+void cov::callback(instr_info_t iinfo, const exec_info& einfo) {
-    stringstream stream;
-    stream << "0x" << std::hex << pc;
-    return stream.str();
-}
-
-void iss::plugin::cov::callback(instr_info_t iinfo, const exec_info& einfo) {
-//    auto delay = 0;
-//    auto entry = delays[iinfo.instr_id];
-//    bool taken = einfo.branch_taken;
-//    if (einfo.branch_taken)
-//        delay = entry.taken;
-//    else
-//        delay = entry.not_taken;
-//
-//    if (first){
-//        output << formatPC(instr_if->get_pc()) << "," << delay;
-//        first = false;
-//    }
-//    if(instr_if->get_next_pc()-instr_if->get_pc() != delays[iinfo.instr_id].size/8){
-//        //The goal is to keep the output in start-target pairs, so after a jump the target address needs to get written
-//        //to the output. If the target happens to also be a start, we keep the pairing by adding a 0-delay entry.
-//        if (jumped)
-//            output <<"\n" <<formatPC(instr_if->get_pc()) << "," << 0;
-//        output <<"\n" << formatPC(instr_if->get_pc()) << "," << delay;
-//        jumped = true;
-//    }
-//    else{
-//        if (jumped){
-//            output <<"\n" << formatPC(instr_if->get_pc()) << "," << delay;
-//            jumped = false;
-//        }
-//        else if(delay!=1){
-//            output <<"\n" << formatPC(instr_if->get_pc()) << "," << delay;
-//            output <<"\n" << formatPC(instr_if->get_pc()) << "," << 0;
-//        }
-//
-//    }
-
-//source code for the full output
     auto delay = 0;
     size_t id = iinfo.instr_id;
     auto entry = delays[id];
     auto instr = instr_if->get_instr_word();
-    auto call = (id==2 || id==3) && bit_sub<7,5>(instr)!=0;
+    auto call = id==65 || id ==86 || ((id==2 || id==3) && bit_sub<7,5>(instr)!=0) ;//not taking care of tail calls (jalr with loading x6)
     bool taken = einfo.branch_taken;
-    if (einfo.branch_taken)
+    bool compressed = (instr&0x3)!=0x3;
+    if (einfo.branch_taken) {
         delay = entry.taken;
-    else
+        if(entry.taken > 1)
+            instr_if->set_curr_instr_cycles(entry.taken);
+    } else {
         delay = entry.not_taken;
-    output<<std::hex <<"0x" << instr_if->get_pc() <<"," << delay << "\n";
+        if (entry.not_taken > 1)
+            instr_if->set_curr_instr_cycles(entry.not_taken);
+    }
+#ifndef WITH_LZ4
+    output<<std::hex <<"0x" << instr_if->get_pc() <<"," << delay <<"," << call<<","<<(compressed?2:4) <<"\n";
+#else
+    auto rdbuf=ostr.rdbuf();
+    ostr<<std::hex <<"0x" << instr_if->get_pc() <<"," << delay <<"," << call<<","<<(compressed?2:4) <<"\n";
+#endif
+}
+}
 }