checkin from eve

this relates to https://github.com/Minres/DBT-RISE-RISCV/issues/8 :
Debugger loses control when trap vector fetch fails

and https://github.com/Minres/DBT-RISE-RISCV/issues/7 : Two debugger
single-steps are required at reset vector

CMakeLists.txt

@@ -20,6 +20,7 @@ set(LIB_SOURCES
     src/iss/arch/tgc5c.cpp
     src/vm/interp/vm_tgc5c.cpp
     src/vm/fp_functions.cpp
+    src/iss/debugger/csr_names.cpp
     src/iss/semihosting/semihosting.cpp
 )
 
@@ -108,15 +109,14 @@ if(TARGET yaml-cpp::yaml-cpp)
     target_link_libraries(${PROJECT_NAME} PUBLIC yaml-cpp::yaml-cpp)
 endif()
 
-if(WITH_LLVM)
-    find_package(LLVM)
-    target_compile_definitions(${PROJECT_NAME} PUBLIC ${LLVM_DEFINITIONS})
-    target_include_directories(${PROJECT_NAME} PUBLIC ${LLVM_INCLUDE_DIRS})
+# if(WITH_LLVM)
+#     target_compile_definitions(${PROJECT_NAME} PUBLIC ${LLVM_DEFINITIONS})
+#     target_include_directories(${PROJECT_NAME} PUBLIC ${LLVM_INCLUDE_DIRS})
 
-    if(BUILD_SHARED_LIBS)
-        target_link_libraries(${PROJECT_NAME} PUBLIC ${LLVM_LIBRARIES})
-    endif()
-endif()
+#     if(BUILD_SHARED_LIBS)
+#         target_link_libraries(${PROJECT_NAME} PUBLIC ${LLVM_LIBRARIES})
+#     endif()
+# endif()
 
 set_target_properties(${PROJECT_NAME} PROPERTIES
     VERSION ${PROJECT_VERSION}
@@ -261,3 +261,9 @@ if(TARGET scc-sysc)
         INCLUDES DESTINATION ${CMAKE_INSTALL_INCLUDEDIR} # headers
     )
 endif()
+
+project(elfio-test)
+find_package(Boost COMPONENTS program_options thread REQUIRED)
+
+add_executable(${PROJECT_NAME} src/elfio.cpp)
+target_link_libraries(${PROJECT_NAME} PUBLIC elfio::elfio)

gen_input/templates/CORENAME.cpp.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2020 MINRES Technologies GmbH
+ * Copyright (C) 2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without

gen_input/templates/CORENAME.h.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2021 MINRES Technologies GmbH
+ * Copyright (C) 2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -131,8 +131,6 @@ struct ${coreDef.name.toLowerCase()}: public arch_if {
 
     uint8_t* get_regs_base_ptr() override;
 
-    inline uint64_t get_icount() { return reg.icount; }
-
     inline bool should_stop() { return interrupt_sim; }
 
     inline uint64_t stop_code() { return interrupt_sim; }
@@ -141,8 +139,6 @@ struct ${coreDef.name.toLowerCase()}: public arch_if {
 
     virtual iss::sync_type needed_sync() const { return iss::NO_SYNC; }
 
-    inline uint32_t get_last_branch() { return reg.last_branch; }
-
 
 #pragma pack(push, 1)
     struct ${coreDef.name}_regs {<%

gen_input/templates/CORENAME_sysc.cpp.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2023 MINRES Technologies GmbH
+ * Copyright (C) 2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -45,17 +45,17 @@ namespace interp {
 using namespace sysc;
 volatile std::array<bool, ${array_count}> ${coreDef.name.toLowerCase()}_init = {
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|m_p|interp", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         }),
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p|interp", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%if(coreDef.name.toLowerCase()=="tgc5d" || coreDef.name.toLowerCase()=="tgc5e") {%>,
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p_clic_pmp|interp", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}, (iss::arch::features_e)(iss::arch::FEAT_PMP | iss::arch::FEAT_EXT_N | iss::arch::FEAT_CLIC)>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%}%>
@@ -66,17 +66,17 @@ namespace llvm {
 using namespace sysc;
 volatile std::array<bool, ${array_count}> ${coreDef.name.toLowerCase()}_init = {
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|m_p|llvm", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         }),
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p|llvm", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%if(coreDef.name.toLowerCase()=="tgc5d" || coreDef.name.toLowerCase()=="tgc5e") {%>,
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p_clic_pmp|llvm", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}, (iss::arch::features_e)(iss::arch::FEAT_PMP | iss::arch::FEAT_EXT_N | iss::arch::FEAT_CLIC)>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%}%>
@@ -88,17 +88,17 @@ namespace tcc {
 using namespace sysc;
 volatile std::array<bool, ${array_count}> ${coreDef.name.toLowerCase()}_init = {
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|m_p|tcc", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         }),
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p|tcc", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%if(coreDef.name.toLowerCase()=="tgc5d" || coreDef.name.toLowerCase()=="tgc5e") {%>,
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p_clic_pmp|tcc", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}, (iss::arch::features_e)(iss::arch::FEAT_PMP | iss::arch::FEAT_EXT_N | iss::arch::FEAT_CLIC)>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%}%>
@@ -110,17 +110,17 @@ namespace asmjit {
 using namespace sysc;
 volatile std::array<bool, ${array_count}> ${coreDef.name.toLowerCase()}_init = {
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|m_p|asmjit", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         }),
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p|asmjit", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%if(coreDef.name.toLowerCase()=="tgc5d" || coreDef.name.toLowerCase()=="tgc5e") {%>,
         iss_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p_clic_pmp|asmjit", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-            auto* cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+            auto* cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
             auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::${coreDef.name.toLowerCase()}, (iss::arch::features_e)(iss::arch::FEAT_PMP | iss::arch::FEAT_EXT_N | iss::arch::FEAT_CLIC)>>(cc);
             return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::${coreDef.name.toLowerCase()}*>(cpu), gdb_port)}};
         })<%}%>

gen_input/templates/asmjit/CORENAME.cpp.gtl

@@ -37,7 +37,10 @@
 #include <iss/asmjit/vm_base.h>
 #include <asmjit/asmjit.h>
 #include <util/logging.h>
-
+#include <iss/instruction_decoder.h>
+<%def fcsr = registers.find {it.name=='FCSR'}
+if(fcsr != null) {%>
+#include <vm/fp_functions.h><%}%>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
 #endif
@@ -80,28 +83,32 @@ public:
 
 protected:
     using super::get_ptr_for;
-using super::get_reg;
     using super::get_reg_for;
+    using super::get_reg_for_Gp;
     using super::load_reg_from_mem;
+    using super::load_reg_from_mem_Gp;
     using super::write_reg_to_mem;
-    using super::gen_ext;
     using super::gen_read_mem;
     using super::gen_write_mem;
-    using super::gen_wait;
     using super::gen_leave;
-    using super::gen_operation;
+    using super::gen_sync;
    
     using this_class = vm_impl<ARCH>;
     using compile_func = continuation_e (this_class::*)(virt_addr_t&, code_word_t, jit_holder&);
 
     continuation_e gen_single_inst_behavior(virt_addr_t&, unsigned int &, jit_holder&) override;
+    enum globals_e {TVAL = 0, GLOBALS_SIZE};
     void gen_block_prologue(jit_holder& jh) override;
     void gen_block_epilogue(jit_holder& jh) override;
     inline const char *name(size_t index){return traits::reg_aliases.at(index);}
-
+<%if(fcsr != null) {%>
+    inline const char *fname(size_t index){return index < 32?name(index+traits::F0):"illegal";}   
+<%}%>
     void gen_instr_prologue(jit_holder& jh);
     void gen_instr_epilogue(jit_holder& jh);
     inline void gen_raise(jit_holder& jh, uint16_t trap_id, uint16_t cause);
+    template <typename T, typename = typename std::enable_if<std::is_integral<T>::value>::type> void gen_set_tval(jit_holder& jh, T new_tval) ;
+    void gen_set_tval(jit_holder& jh, x86_reg_t _new_tval) ;
 
     template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
     inline S sext(U from) {
@@ -109,25 +116,19 @@ using super::get_reg;
         auto sign_mask = 1ULL<<(W-1);
         return (from & mask) | ((from & sign_mask) ? ~mask : 0);
     }
+<%functions.each{ it.eachLine { %>
+    ${it}<%}%>
+<%}%>
 private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
     struct instruction_descriptor {
-        size_t length;
+        uint32_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){}
-    };
-
-    decoding_tree_node* root {nullptr};
 
     const std::array<instruction_descriptor, ${instructions.size()}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
@@ -135,6 +136,9 @@ private:
         {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
     }};
 
+    //needs to be declared after instr_descr
+    decoder instr_decoder;
+
     /* instruction definitions */<%instructions.eachWithIndex{instr, idx -> %>
     /* instruction ${idx}: ${instr.name} */
     continuation_e __${generator.functionName(instr.name)}(virt_addr_t& pc, code_word_t instr, jit_holder& jh){
@@ -155,87 +159,49 @@ private:
         }
         x86::Compiler& cc = jh.cc;
         cc.comment(fmt::format("${instr.name}_{:#x}:",pc.val).c_str());
-        this->gen_sync(jh, PRE_SYNC, ${idx});
-        cc.mov(jh.pc, pc.val);
+        gen_sync(jh, PRE_SYNC, ${idx});
+        mov(cc, jh.pc, pc.val);
+        gen_set_tval(jh, instr);
         pc = pc+${instr.length/8};
-        cc.mov(jh.next_pc, pc.val);
+        mov(cc, jh.next_pc, pc.val);
 
         gen_instr_prologue(jh);
         cc.comment("//behavior:");
         /*generate behavior*/
         <%instr.behavior.eachLine{%>${it}
         <%}%>
+        gen_sync(jh, POST_SYNC, ${idx});
         gen_instr_epilogue(jh);
-        this->gen_sync(jh, POST_SYNC, ${idx});
     	return returnValue;        
     }
     <%}%>
     /****************************************************************************
      * end opcode definitions
      ****************************************************************************/
-    continuation_e illegal_intruction(virt_addr_t &pc, code_word_t instr, jit_holder& jh ) {
+    continuation_e illegal_instruction(virt_addr_t &pc, code_word_t instr, jit_holder& jh ) {
         x86::Compiler& cc = jh.cc;
-        cc.comment(fmt::format("illegal_intruction{:#x}:",pc.val).c_str());
-        this->gen_sync(jh, PRE_SYNC, instr_descr.size());
+        if(this->disass_enabled){          
+            auto mnemonic = std::string("illegal_instruction");
+            InvokeNode* call_print_disass;
+            char* mnemonic_ptr = strdup(mnemonic.c_str());
+            jh.disass_collection.push_back(mnemonic_ptr);
+            jh.cc.invoke(&call_print_disass, &print_disass, FuncSignature::build<void, void *, uint64_t, char *>());
+            call_print_disass->setArg(0, jh.arch_if_ptr);
+            call_print_disass->setArg(1, pc.val);
+            call_print_disass->setArg(2, mnemonic_ptr);
+        }
+        cc.comment(fmt::format("illegal_instruction{:#x}:",pc.val).c_str());
+        gen_sync(jh, PRE_SYNC, instr_descr.size());
+        mov(cc, jh.pc, pc.val);
+        gen_set_tval(jh, instr);
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
+        mov(cc, jh.next_pc, pc.val);
         gen_instr_prologue(jh);
         cc.comment("//behavior:");
+        gen_raise(jh, 0, 2);
+        gen_sync(jh, POST_SYNC, instr_descr.size());
         gen_instr_epilogue(jh);
-        this->gen_sync(jh, POST_SYNC, instr_descr.size());
-        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;
+        return ILLEGAL_INSTR;
     }
 };
 
@@ -243,13 +209,16 @@ 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) {
-    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);
+: vm_base<ARCH>(core, core_id, cluster_id)
+, instr_decoder([this]() {
+        std::vector<generic_instruction_descriptor> g_instr_descr;
+        g_instr_descr.reserve(instr_descr.size());
+        for (uint32_t i = 0; i < instr_descr.size(); ++i) {
+            generic_instruction_descriptor new_instr_descr {instr_descr[i].value, instr_descr[i].mask, i};
+            g_instr_descr.push_back(new_instr_descr);
         }
+        return std::move(g_instr_descr);
+    }()) {}
 
 template <typename ARCH>
 continuation_e vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, jit_holder& jh) {
@@ -261,13 +230,16 @@ continuation_e vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned
         paddr = this->core.virt2phys(pc);
     auto res = this->core.read(paddr, 4, data);
     if (res != iss::Ok)
-        throw trap_access(TRAP_ID, pc.val);
+        return ILLEGAL_FETCH;
     if (instr == 0x0000006f || (instr&0xffff)==0xa001)
-        throw simulation_stopped(0); // 'J 0' or 'C.J 0'
+        return JUMP_TO_SELF;
     ++inst_cnt;
-    auto f = decode_instr(root, instr);
+    uint32_t inst_index = instr_decoder.decode_instr(instr);
+    compile_func f = nullptr;
+    if(inst_index < instr_descr.size())
+        f = instr_descr[inst_index].op;
     if (f == nullptr) 
-        f = &this_class::illegal_intruction;
+        f = &this_class::illegal_instruction;
     return (this->*f)(pc, instr, jh);
 }
 template <typename ARCH>
@@ -275,11 +247,10 @@ void vm_impl<ARCH>::gen_instr_prologue(jit_holder& jh) {
     auto& cc = jh.cc;
 
     cc.comment("//gen_instr_prologue");
-    cc.inc(get_ptr_for(jh, traits::ICOUNT));
 
-    x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
-    cc.mov(current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
-    cc.mov(get_ptr_for(jh, traits::PENDING_TRAP), current_trap_state);
+    x86_reg_t current_trap_state = get_reg_for(cc, traits::TRAP_STATE);
+    mov(cc, current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
+    mov(cc, get_ptr_for(jh, traits::PENDING_TRAP), current_trap_state);
 
 }
 template <typename ARCH>
@@ -287,16 +258,19 @@ void vm_impl<ARCH>::gen_instr_epilogue(jit_holder& jh) {
     auto& cc = jh.cc;
 
     cc.comment("//gen_instr_epilogue");
-    x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
-    cc.mov(current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
-    cc.cmp(current_trap_state, 0);
+    x86_reg_t current_trap_state = get_reg_for(cc, traits::TRAP_STATE);
+    mov(cc, current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
+    cmp(cc, current_trap_state, 0);
     cc.jne(jh.trap_entry);
+    cc.inc(get_ptr_for(jh, traits::ICOUNT));
+    cc.inc(get_ptr_for(jh, traits::CYCLE));
 }
 template <typename ARCH>
 void vm_impl<ARCH>::gen_block_prologue(jit_holder& jh){
-
-    jh.pc = load_reg_from_mem(jh, traits::PC);
-    jh.next_pc = load_reg_from_mem(jh, traits::NEXT_PC);
+    jh.pc = load_reg_from_mem_Gp(jh, traits::PC);
+    jh.next_pc = load_reg_from_mem_Gp(jh, traits::NEXT_PC);
+    jh.globals.resize(GLOBALS_SIZE);
+    jh.globals[TVAL] = get_reg_Gp(jh.cc, 64, false);
 }
 template <typename ARCH>
 void vm_impl<ARCH>::gen_block_epilogue(jit_holder& jh){
@@ -306,39 +280,52 @@ void vm_impl<ARCH>::gen_block_epilogue(jit_holder& jh){
 
     cc.bind(jh.trap_entry);
     this->write_back(jh);
-    this->gen_sync(jh, POST_SYNC, -1);
 
-    x86::Gp current_trap_state = get_reg_for(jh, traits::TRAP_STATE);
-    cc.mov(current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
+    x86::Gp current_trap_state = get_reg_for_Gp(cc, traits::TRAP_STATE);
+    mov(cc, current_trap_state, get_ptr_for(jh, traits::TRAP_STATE));
 
-    x86::Gp current_pc = get_reg_for(jh, traits::PC);
-    cc.mov(current_pc, get_ptr_for(jh, traits::PC));
+    x86::Gp current_pc = get_reg_for_Gp(cc, traits::PC);
+    mov(cc, current_pc, get_ptr_for(jh, traits::PC));
 
-    x86::Gp instr = cc.newInt32("instr");
-    cc.mov(instr, 0); // FIXME:this is not correct
     cc.comment("//enter trap call;");
     InvokeNode* call_enter_trap;
     cc.invoke(&call_enter_trap, &enter_trap, FuncSignature::build<uint64_t, void*, uint64_t, uint64_t, uint64_t>());
     call_enter_trap->setArg(0, jh.arch_if_ptr);
     call_enter_trap->setArg(1, current_trap_state);
     call_enter_trap->setArg(2, current_pc);
-    call_enter_trap->setArg(3, instr);
+    call_enter_trap->setArg(3, jh.globals[TVAL]);
 
-    x86::Gp current_next_pc = get_reg_for(jh, traits::NEXT_PC);
-    cc.mov(current_next_pc, get_ptr_for(jh, traits::NEXT_PC));
-    cc.mov(jh.next_pc, current_next_pc);
+    x86_reg_t current_next_pc = get_reg_for(cc, traits::NEXT_PC);
+    mov(cc, current_next_pc, get_ptr_for(jh, traits::NEXT_PC));
+    mov(cc, jh.next_pc, current_next_pc);
 
-    cc.mov(get_ptr_for(jh, traits::LAST_BRANCH), std::numeric_limits<uint32_t>::max());
+    mov(cc, get_ptr_for(jh, traits::LAST_BRANCH), static_cast<int>(UNKNOWN_JUMP));
     cc.ret(jh.next_pc);
 }
 template <typename ARCH>
 inline void vm_impl<ARCH>::gen_raise(jit_holder& jh, uint16_t trap_id, uint16_t cause) {
     auto& cc = jh.cc;
     cc.comment("//gen_raise");
-    auto tmp1 = get_reg_for(jh, traits::TRAP_STATE);
-    cc.mov(tmp1, 0x80ULL << 24 | (cause << 16) | trap_id);
-    cc.mov(get_ptr_for(jh, traits::TRAP_STATE), tmp1);
-    cc.mov(jh.next_pc, std::numeric_limits<uint32_t>::max());
+    auto tmp1 = get_reg_for(cc, traits::TRAP_STATE);
+    mov(cc, tmp1, 0x80ULL << 24 | (cause << 16) | trap_id);
+    mov(cc, get_ptr_for(jh, traits::TRAP_STATE), tmp1);
+    cc.jmp(jh.trap_entry);
+}
+template <typename ARCH>
+template <typename T, typename>
+void vm_impl<ARCH>::gen_set_tval(jit_holder& jh, T new_tval) {
+        mov(jh.cc, jh.globals[TVAL], new_tval);
+    }
+template <typename ARCH>
+void vm_impl<ARCH>::gen_set_tval(jit_holder& jh, x86_reg_t _new_tval) {
+    if(nonstd::holds_alternative<x86::Gp>(_new_tval)) {
+        x86::Gp new_tval = nonstd::get<x86::Gp>(_new_tval);
+        if(new_tval.size() < 8)
+            new_tval = gen_ext_Gp(jh.cc, new_tval, 64, false);
+        mov(jh.cc, jh.globals[TVAL], new_tval);
+    } else {
+        throw std::runtime_error("Variant not supported in gen_set_tval");
+    }
 }
 
 } // namespace tgc5c

gen_input/templates/interp/CORENAME.cpp.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 20217-2024 MINRES Technologies GmbH
+ * Copyright (C) 2017-2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -35,6 +35,7 @@ def nativeTypeSize(int size){
 }
 %>
 // clang-format off
+#include <cstdint>
 #include <iss/arch/${coreDef.name.toLowerCase()}.h>
 #include <iss/debugger/gdb_session.h>
 #include <iss/debugger/server.h>
@@ -47,6 +48,8 @@ def nativeTypeSize(int size){
 #include <exception>
 #include <vector>
 #include <sstream>
+#include <iss/instruction_decoder.h>
+
 
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
@@ -97,7 +100,12 @@ protected:
     using compile_ret_t = virt_addr_t;
     using compile_func = compile_ret_t (this_class::*)(virt_addr_t &pc, code_word_t instr);
 
-    inline const char *name(size_t index){return index<traits::reg_aliases.size()?traits::reg_aliases[index]:"illegal";}
+    inline const char *name(size_t index){return traits::reg_aliases.at(index);}
+<%
+def fcsr = registers.find {it.name=='FCSR'}
+if(fcsr != null) {%>
+    inline const char *fname(size_t index){return index < 32?name(index+traits::F0):"illegal";}     
+<%}%>
 
     virt_addr_t execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit) override;
 
@@ -106,7 +114,6 @@ protected:
     inline void raise(uint16_t trap_id, uint16_t cause){
         auto trap_val =  0x80ULL << 24 | (cause << 16) | trap_id;
         this->core.reg.trap_state = trap_val;
-        this->template get_reg<uint${addrDataWidth}_t>(traits::NEXT_PC) = std::numeric_limits<uint${addrDataWidth}_t>::max();
     }
 
     inline void leave(unsigned lvl){
@@ -117,7 +124,12 @@ protected:
         this->core.wait_until(type);
     }
 
+    inline void set_tval(uint64_t new_tval){
+        tval = new_tval;
+    }
+
     uint64_t fetch_count{0};
+    uint64_t tval{0};
 
     using yield_t = boost::coroutines2::coroutine<void>::push_type;
     using coro_t = boost::coroutines2::coroutine<void>::pull_type;
@@ -148,25 +160,20 @@ private:
      * start opcode definitions
      ****************************************************************************/
     struct instruction_descriptor {
-        size_t length;
+        uint32_t length;
         uint32_t value;
         uint32_t mask;
         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){}
-    };
 
-    decoding_tree_node* root {nullptr};
     const std::array<instruction_descriptor, ${instructions.size()}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
         {${instr.length}, ${instr.encoding}, ${instr.mask}, arch::traits<ARCH>::opcode_e::${instr.instruction.name}},<%}%>
     }};
 
+    //needs to be declared after instr_descr
+    decoder instr_decoder;
+
     iss::status fetch_ins(virt_addr_t pc, uint8_t * data){
         if(this->core.has_mmu()) {
             auto phys_pc = this->core.virt2phys(pc);
@@ -186,58 +193,6 @@ private:
         }
         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) {
@@ -263,13 +218,16 @@ 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) {
-    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);
+: vm_base<ARCH>(core, core_id, cluster_id)
+, instr_decoder([this]() {
+        std::vector<generic_instruction_descriptor> g_instr_descr;
+        g_instr_descr.reserve(instr_descr.size());
+        for (uint32_t i = 0; i < instr_descr.size(); ++i) {
+            generic_instruction_descriptor new_instr_descr {instr_descr[i].value, instr_descr[i].mask, i};
+            g_instr_descr.push_back(new_instr_descr);
         }
+        return std::move(g_instr_descr);
+    }()) {}
 
 inline bool is_icount_limit_enabled(finish_cond_e cond){
     return (cond & finish_cond_e::ICOUNT_LIMIT) == finish_cond_e::ICOUNT_LIMIT;
@@ -299,14 +257,22 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
     while(!this->core.should_stop() &&
             !(is_icount_limit_enabled(cond) && icount >= count_limit) &&
             !(is_fcount_limit_enabled(cond) && fetch_count >= count_limit)){
-        fetch_count++;
+        if(this->debugging_enabled())
+            this->tgt_adapter->check_continue(*PC);
+        pc.val=*PC;
         if(fetch_ins(pc, data)!=iss::Ok){
-            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);
+            if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, std::numeric_limits<unsigned>::max());
+            process_spawn_blocks();
+            if(this->sync_exec && POST_SYNC) this->do_sync(PRE_SYNC, std::numeric_limits<unsigned>::max());
+            pc.val = super::core.enter_trap(arch::traits<ARCH>::RV_CAUSE_FETCH_ACCESS<<16, pc.val, 0);
         } 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 = decode_instr(root, instr);
+            uint32_t inst_index = instr_decoder.decode_instr(instr);
+            opcode_e inst_id = arch::traits<ARCH>::opcode_e::MAX_OPCODE;;
+            if(inst_index <instr_descr.size())
+                inst_id = instr_descr[inst_index].op;
+
             // 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));
@@ -317,6 +283,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                     <%}%>if(this->disass_enabled){
                         /* generate console output when executing the command */<%instr.disass.eachLine{%>
                         ${it}<%}%>
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers<%instr.usedVariables.each{ k,v->
                     if(v.isArray) {%>
@@ -341,16 +308,18 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
             //    this->core.reg.trap_state =  this->core.reg.pending_trap;
             // trap check
             if(trap_state!=0){
-                super::core.enter_trap(trap_state, pc.val, instr);
+                //In case of Instruction address misaligned (cause = 0 and trapid = 0) need the targeted addr (in tval)
+                auto mcause = (trap_state>>16) & 0xff; 
+                super::core.enter_trap(trap_state, pc.val, mcause ? instr:tval);
             } else {
                 icount++;
                 instret++;
             }
-            cycle++;
-            pc.val=*NEXT_PC;
-            this->core.reg.PC = this->core.reg.NEXT_PC;
+            *PC = *NEXT_PC;
             this->core.reg.trap_state =  this->core.reg.pending_trap;
         }
+        fetch_count++;
+        cycle++;
     }
     return pc;
 }

gen_input/templates/llvm/CORENAME.cpp.gtl

@@ -36,7 +36,10 @@
 #include <iss/iss.h>
 #include <iss/llvm/vm_base.h>
 #include <util/logging.h>
-
+#include <iss/instruction_decoder.h>
+<%def fcsr = registers.find {it.name=='FCSR'}
+if(fcsr != null) {%>
+#include <vm/fp_functions.h><%}%>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
 #endif
@@ -82,7 +85,9 @@ protected:
     using vm_base<ARCH>::get_reg_ptr;
 
     inline const char *name(size_t index){return traits::reg_aliases.at(index);}
-
+<%if(fcsr != null) {%>
+    inline const char *fname(size_t index){return index < 32?name(index+traits::F0):"illegal";}   
+<%}%>
     template <typename T> inline ConstantInt *size(T type) {
         return ConstantInt::get(getContext(), APInt(32, type->getType()->getScalarSizeInBits()));
     }
@@ -102,7 +107,10 @@ protected:
     void gen_raise_trap(uint16_t trap_id, uint16_t cause);
     void gen_leave_trap(unsigned lvl);
     void gen_wait(unsigned type);
+    void set_tval(uint64_t new_tval);
+    void set_tval(Value* new_tval);
     void gen_trap_behavior(BasicBlock *) override;
+    void gen_instr_prologue();
     void gen_instr_epilogue(BasicBlock *bb);
 
     inline Value *gen_reg_load(unsigned i, unsigned level = 0) {
@@ -127,26 +135,19 @@ protected:
         auto sign_mask = 1ULL<<(W-1);
         return (from & mask) | ((from & sign_mask) ? ~mask : 0);
     }
-
+<%functions.each{ it.eachLine { %>
+    ${it}<%}%>
+<%}%>
 private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
     struct instruction_descriptor {
-        size_t length;
+        uint32_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){}
-    };
-
-    decoding_tree_node* root {nullptr};
 
     const std::array<instruction_descriptor, ${instructions.size()}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
@@ -154,6 +155,9 @@ private:
         {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
     }};
 
+    //needs to be declared after instr_descr
+    decoder instr_decoder;
+
     /* instruction definitions */<%instructions.eachWithIndex{instr, idx -> %>
     /* instruction ${idx}: ${instr.name} */
     std::tuple<continuation_e, BasicBlock*> __${generator.functionName(instr.name)}(virt_addr_t& pc, code_word_t instr, BasicBlock* bb){
@@ -162,18 +166,27 @@ private:
         <%}%>if(this->disass_enabled){
             /* generate console output when executing the command */<%instr.disass.eachLine{%>
             ${it}<%}%>
+            std::vector<Value*> args {
+                this->core_ptr,
+                this->gen_const(64, pc.val),
+                this->builder.CreateGlobalStringPtr(mnemonic),
+            };
+            this->builder.CreateCall(this->mod->getFunction("print_disass"), args);
         }
         bb->setName(fmt::format("${instr.name}_0x{:X}",pc.val));
         this->gen_sync(PRE_SYNC,${idx});
-        auto cur_pc_val = this->gen_const(32,pc.val);
+        
+        this->gen_set_pc(pc, traits::PC);
+        this->set_tval(instr);
         pc=pc+ ${instr.length/8};
         this->gen_set_pc(pc, traits::NEXT_PC);
         
+        this->gen_instr_prologue();
         /*generate behavior*/
         <%instr.behavior.eachLine{%>${it}
         <%}%>
-        this->gen_instr_epilogue(bb);
         this->gen_sync(POST_SYNC, ${idx});
+        this->gen_instr_epilogue(bb);
         this->builder.CreateBr(bb);
     	return returnValue;        
     }
@@ -181,7 +194,16 @@ private:
     /****************************************************************************
      * end opcode definitions
      ****************************************************************************/
-    std::tuple<continuation_e, BasicBlock *> illegal_intruction(virt_addr_t &pc, code_word_t instr, BasicBlock *bb) {
+    std::tuple<continuation_e, BasicBlock *> illegal_instruction(virt_addr_t &pc, code_word_t instr, BasicBlock *bb) {
+        if(this->disass_enabled){
+            auto mnemonic = std::string("illegal_instruction");
+            std::vector<Value*> args {
+                this->core_ptr,
+                this->gen_const(64, pc.val),
+                this->builder.CreateGlobalStringPtr(mnemonic),
+            };
+            this->builder.CreateCall(this->mod->getFunction("print_disass"), args);
+        }
         this->gen_sync(iss::PRE_SYNC, instr_descr.size());
         this->builder.CreateStore(this->builder.CreateLoad(this->get_typeptr(traits::NEXT_PC), get_reg_ptr(traits::NEXT_PC), true),
                                    get_reg_ptr(traits::PC), true);
@@ -190,62 +212,13 @@ private:
                                      this->gen_const(64U, 1)),
             get_reg_ptr(traits::ICOUNT), true);
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
+        this->set_tval(instr);
         this->gen_raise_trap(0, 2);     // illegal instruction trap
 		this->gen_sync(iss::POST_SYNC, instr_descr.size());
-        this->gen_instr_epilogue(this->leave_blk);
-        return std::make_tuple(BRANCH, nullptr);
-    }    
-    //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;
+        bb = this->leave_blk;
+        this->gen_instr_epilogue(bb);
+        this->builder.CreateBr(bb);
+        return std::make_tuple(ILLEGAL_INSTR, nullptr);
     }    
 };
 
@@ -258,13 +231,16 @@ 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) {
-    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);
+: vm_base<ARCH>(core, core_id, cluster_id)
+, instr_decoder([this]() {
+        std::vector<generic_instruction_descriptor> g_instr_descr;
+        g_instr_descr.reserve(instr_descr.size());
+        for (uint32_t i = 0; i < instr_descr.size(); ++i) {
+            generic_instruction_descriptor new_instr_descr {instr_descr[i].value, instr_descr[i].mask, i};
+            g_instr_descr.push_back(new_instr_descr);
         }
+        return std::move(g_instr_descr);
+    }()) {}
 
 template <typename ARCH>
 std::tuple<continuation_e, BasicBlock *>
@@ -277,23 +253,18 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
     auto *const data = (uint8_t *)&instr;
     if(this->core.has_mmu())
         paddr = this->core.virt2phys(pc);
-    //TODO: re-add page handling
-//    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 ((instr & 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 (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
-    // curr pc on stack
+    if (res != iss::Ok) 
+        return std::make_tuple(ILLEGAL_FETCH, nullptr);
+    if (instr == 0x0000006f || (instr&0xffff)==0xa001)
+        return std::make_tuple(JUMP_TO_SELF, nullptr);
     ++inst_cnt;
-    auto f = decode_instr(root, instr);
+    uint32_t inst_index = instr_decoder.decode_instr(instr);
+    compile_func f = nullptr;
+    if(inst_index < instr_descr.size())
+        f = instr_descr[inst_index].op;
     if (f == nullptr) {
-        f = &this_class::illegal_intruction;
+        f = &this_class::illegal_instruction;
     }
     return (this->*f)(pc, instr, this_block);
 }
@@ -308,16 +279,14 @@ template <typename ARCH>
 void vm_impl<ARCH>::gen_raise_trap(uint16_t trap_id, uint16_t cause) {
     auto *TRAP_val = this->gen_const(32, 0x80 << 24 | (cause << 16) | trap_id);
     this->builder.CreateStore(TRAP_val, get_reg_ptr(traits::TRAP_STATE), true);
-    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits::LAST_BRANCH), false);
+    this->builder.CreateBr(this->trap_blk);
 }
 
 template <typename ARCH>
 void vm_impl<ARCH>::gen_leave_trap(unsigned lvl) {
     std::vector<Value *> args{ this->core_ptr, ConstantInt::get(getContext(), APInt(64, lvl)) };
     this->builder.CreateCall(this->mod->getFunction("leave_trap"), args);
-    auto *PC_val = this->gen_read_mem(traits::CSR, (lvl << 8) + 0x41, traits::XLEN / 8);
-    this->builder.CreateStore(PC_val, get_reg_ptr(traits::NEXT_PC), false);
-    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits::LAST_BRANCH), false);
+    this->builder.CreateStore(this->gen_const(32U, static_cast<int>(UNKNOWN_JUMP)), get_reg_ptr(traits::LAST_BRANCH), false);
 }
 
 template <typename ARCH>
@@ -326,19 +295,37 @@ void vm_impl<ARCH>::gen_wait(unsigned type) {
     this->builder.CreateCall(this->mod->getFunction("wait"), args);
 }
 
+template <typename ARCH>
+inline void vm_impl<ARCH>::set_tval(uint64_t tval) {
+    auto tmp_tval = this->gen_const(64, tval);
+    this->set_tval(tmp_tval);
+}
+template <typename ARCH>
+inline void vm_impl<ARCH>::set_tval(Value* new_tval) {
+    this->builder.CreateStore(this->gen_ext(new_tval, 64, false), this->tval);
+}
 template <typename ARCH> 
 void vm_impl<ARCH>::gen_trap_behavior(BasicBlock *trap_blk) {
     this->builder.SetInsertPoint(trap_blk);
-    this->gen_sync(POST_SYNC, -1); //TODO get right InstrId
     auto *trap_state_val = this->builder.CreateLoad(this->get_typeptr(traits::TRAP_STATE), get_reg_ptr(traits::TRAP_STATE), true);
-    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()),
-                              get_reg_ptr(traits::LAST_BRANCH), false);
-    std::vector<Value *> args{this->core_ptr, this->adj_to64(trap_state_val),
-                              this->adj_to64(this->builder.CreateLoad(this->get_typeptr(traits::PC), get_reg_ptr(traits::PC), false))};
+    auto *cur_pc_val = this->builder.CreateLoad(this->get_typeptr(traits::PC), get_reg_ptr(traits::PC), true);
+    std::vector<Value *> args{this->core_ptr,
+                                this->adj_to64(trap_state_val),
+                                this->adj_to64(cur_pc_val),
+                              this->adj_to64(this->builder.CreateLoad(this->get_type(64),this->tval))};
     this->builder.CreateCall(this->mod->getFunction("enter_trap"), args);
+    this->builder.CreateStore(this->gen_const(32U, static_cast<int>(UNKNOWN_JUMP)), get_reg_ptr(traits::LAST_BRANCH), false);
+
     auto *trap_addr_val = this->builder.CreateLoad(this->get_typeptr(traits::NEXT_PC), get_reg_ptr(traits::NEXT_PC), false);
     this->builder.CreateRet(trap_addr_val);
 }
+template <typename ARCH>
+void vm_impl<ARCH>::gen_instr_prologue() {
+    auto* trap_val =
+        this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::PENDING_TRAP), get_reg_ptr(arch::traits<ARCH>::PENDING_TRAP));
+    this->builder.CreateStore(trap_val, get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), false);
+}
+            
 
 template <typename ARCH>
 void vm_impl<ARCH>::gen_instr_epilogue(BasicBlock *bb) {
@@ -349,6 +336,10 @@ void vm_impl<ARCH>::gen_instr_epilogue(BasicBlock *bb) {
                               ConstantInt::get(getContext(), APInt(v->getType()->getIntegerBitWidth(), 0))),
                           target_bb, this->trap_blk, 1);
     this->builder.SetInsertPoint(target_bb);
+    // update icount
+    auto* icount_val = this->builder.CreateAdd(
+        this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::ICOUNT), get_reg_ptr(arch::traits<ARCH>::ICOUNT)), this->gen_const(64U, 1));
+    this->builder.CreateStore(icount_val, get_reg_ptr(arch::traits<ARCH>::ICOUNT), false);
 }
 
 } // namespace ${coreDef.name.toLowerCase()}

gen_input/templates/tcc/CORENAME.cpp.gtl

@@ -37,7 +37,10 @@
 #include <iss/tcc/vm_base.h>
 #include <util/logging.h>
 #include <sstream>
-
+#include <iss/instruction_decoder.h>
+<%def fcsr = registers.find {it.name=='FCSR'}
+if(fcsr != null) {%>
+#include <vm/fp_functions.h><%}%>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
 #endif
@@ -80,16 +83,21 @@ protected:
     using vm_base<ARCH>::get_reg_ptr;
 
     using this_class = vm_impl<ARCH>;
-    using compile_ret_t = std::tuple<continuation_e>;
+    using compile_ret_t = continuation_e;
     using compile_func = compile_ret_t (this_class::*)(virt_addr_t &pc, code_word_t instr, tu_builder&);
 
     inline const char *name(size_t index){return traits::reg_aliases.at(index);}
+<%
+if(fcsr != null) {%>
+    inline const char *fname(size_t index){return index < 32?name(index+traits::F0):"illegal";}   
+<%}%>
+    void add_prologue(tu_builder& tu) override;
 
     void setup_module(std::string m) override {
         super::setup_module(m);
     }
 
-    compile_ret_t gen_single_inst_behavior(virt_addr_t &, unsigned int &, tu_builder&) override;
+    compile_ret_t gen_single_inst_behavior(virt_addr_t &, tu_builder&) override;
 
     void gen_trap_behavior(tu_builder& tu) override;
 
@@ -97,7 +105,9 @@ protected:
 
     void gen_leave_trap(tu_builder& tu, unsigned lvl);
 
-    void gen_wait(tu_builder& tu, unsigned type);
+    inline void gen_set_tval(tu_builder& tu, uint64_t new_tval);
+
+    inline void gen_set_tval(tu_builder& tu, value new_tval);
 
     inline void gen_trap_check(tu_builder& tu) {
         tu("if(*trap_state!=0) goto trap_entry;");
@@ -128,25 +138,19 @@ protected:
         return (from & mask) | ((from & sign_mask) ? ~mask : 0);
     }
 
+<%functions.each{ it.eachLine { %>
+    ${it}<%}%>
+<%}%>
 private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
     struct instruction_descriptor {
-        size_t length;
+        uint32_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){}
-    };
-
-    decoding_tree_node* root {nullptr};
 
     const std::array<instruction_descriptor, ${instructions.size()}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
@@ -154,6 +158,9 @@ private:
         {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
     }};
 
+    //needs to be declared after instr_descr
+    decoder instr_decoder;
+
     /* 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, tu_builder& tu){
@@ -164,82 +171,36 @@ private:
         <%}%>if(this->disass_enabled){
             /* generate console output when executing the command */<%instr.disass.eachLine{%>
             ${it}<%}%>
+            tu("print_disass(core_ptr, {:#x}, \"{}\");", pc.val, mnemonic);
         }
         auto cur_pc_val = tu.constant(pc.val, traits::reg_bit_widths[traits::PC]);
         pc=pc+ ${instr.length/8};
         gen_set_pc(tu, pc, traits::NEXT_PC);
         tu.open_scope();
+        this->gen_set_tval(tu, instr);
         <%instr.behavior.eachLine{%>${it}
         <%}%>
         tu.close_scope();
-        gen_trap_check(tu);        
         vm_base<ARCH>::gen_sync(tu, POST_SYNC,${idx});
+        gen_trap_check(tu);        
         return returnValue;
     }
     <%}%>
     /****************************************************************************
      * end opcode definitions
      ****************************************************************************/
-    compile_ret_t illegal_intruction(virt_addr_t &pc, code_word_t instr, tu_builder& tu) {
+    compile_ret_t illegal_instruction(virt_addr_t &pc, code_word_t instr, tu_builder& tu) {
         vm_impl::gen_sync(tu, iss::PRE_SYNC, instr_descr.size());
+        if(this->disass_enabled){
+            /* generate console output when executing the command */
+            tu("print_disass(core_ptr, {:#x}, \"{}\");", pc.val, std::string("illegal_instruction"));
+        }
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
-        gen_raise_trap(tu, 0, 2);     // illegal instruction trap
+        gen_raise_trap(tu, 0, static_cast<int32_t>(traits:: RV_CAUSE_ILLEGAL_INSTRUCTION));
+        this->gen_set_tval(tu, instr);
         vm_impl::gen_sync(tu, iss::POST_SYNC, instr_descr.size());
         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;
+        return ILLEGAL_INSTR;
     }
 };
 
@@ -252,65 +213,102 @@ 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) {
-    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);
+: vm_base<ARCH>(core, core_id, cluster_id)
+, instr_decoder([this]() {
+        std::vector<generic_instruction_descriptor> g_instr_descr;
+        g_instr_descr.reserve(instr_descr.size());
+        for (uint32_t i = 0; i < instr_descr.size(); ++i) {
+            generic_instruction_descriptor new_instr_descr {instr_descr[i].value, instr_descr[i].mask, i};
+            g_instr_descr.push_back(new_instr_descr);
         }
+        return std::move(g_instr_descr);
+    }()) {}
 
 template <typename ARCH>
-std::tuple<continuation_e>
-vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) {
+continuation_e
+vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, tu_builder& tu) {
     // we fetch at max 4 byte, alignment is 2
     enum {TRAP_ID=1<<16};
     code_word_t instr = 0;
     phys_addr_t paddr(pc);
     if(this->core.has_mmu())
         paddr = this->core.virt2phys(pc);
-    //TODO: re-add page handling
-//    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, 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 f = decode_instr(root, instr);
+    if (res != iss::Ok)
+        return ILLEGAL_FETCH;
+    if (instr == 0x0000006f || (instr&0xffff)==0xa001) 
+        return JUMP_TO_SELF;
+    uint32_t inst_index = instr_decoder.decode_instr(instr);
+    compile_func f = nullptr;
+    if(inst_index < instr_descr.size())
+        f = instr_descr[inst_index].op;
     if (f == nullptr) {
-        f = &this_class::illegal_intruction;
+        f = &this_class::illegal_instruction;
     }
     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) {
     tu("  *trap_state = {:#x};", 0x80 << 24 | (cause << 16) | trap_id);
-    tu.store(traits::NEXT_PC, tu.constant(std::numeric_limits<uint32_t>::max(), 32));
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_trap(tu_builder& tu, unsigned lvl) {
     tu("leave_trap(core_ptr, {});", lvl);
     tu.store(traits::NEXT_PC, tu.read_mem(traits::CSR, (lvl << 8) + 0x41, traits::XLEN));
-    tu.store(traits::LAST_BRANCH, tu.constant(std::numeric_limits<uint32_t>::max(), 32));
+    tu.store(traits::LAST_BRANCH, tu.constant(static_cast<int>(UNKNOWN_JUMP), 32));
 }
 
-template <typename ARCH> void vm_impl<ARCH>::gen_wait(tu_builder& tu, unsigned type) {
+template <typename ARCH> void vm_impl<ARCH>::gen_set_tval(tu_builder& tu, uint64_t new_tval) {
+    tu(fmt::format("tval = {};", new_tval));
+}
+template <typename ARCH> void vm_impl<ARCH>::gen_set_tval(tu_builder& tu, value new_tval) {
+    tu(fmt::format("tval = {};", new_tval.str));
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_trap_behavior(tu_builder& tu) {
     tu("trap_entry:");
     this->gen_sync(tu, POST_SYNC, -1);    
-    tu("enter_trap(core_ptr, *trap_state, *pc, 0);");
-    tu.store(traits::LAST_BRANCH, tu.constant(std::numeric_limits<uint32_t>::max(),32));
+    tu("enter_trap(core_ptr, *trap_state, *pc, tval);");
+    tu.store(traits::LAST_BRANCH, tu.constant(static_cast<int>(UNKNOWN_JUMP),32));
     tu("return *next_pc;");
 }
+<%
+
+template <typename ARCH> void vm_impl<ARCH>::add_prologue(tu_builder& tu){
+    std::ostringstream os;
+    os << add_reg_ptr("trap_state", arch::traits<ARCH>::TRAP_STATE);
+    os << add_reg_ptr("pending_trap", arch::traits<ARCH>::PENDING_TRAP);
+if(fcsr != null) {%>
+    os << "uint32_t (*fget_flags)()=" << (uintptr_t)&fget_flags << ";\\n";
+    os << "uint32_t (*fadd_s)(uint32_t v1, uint32_t v2, uint8_t mode)=" << (uintptr_t)&fadd_s << ";\\n";
+    os << "uint32_t (*fsub_s)(uint32_t v1, uint32_t v2, uint8_t mode)=" << (uintptr_t)&fsub_s << ";\\n";
+    os << "uint32_t (*fmul_s)(uint32_t v1, uint32_t v2, uint8_t mode)=" << (uintptr_t)&fmul_s << ";\\n";
+    os << "uint32_t (*fdiv_s)(uint32_t v1, uint32_t v2, uint8_t mode)=" << (uintptr_t)&fdiv_s << ";\\n";
+    os << "uint32_t (*fsqrt_s)(uint32_t v1, uint8_t mode)=" << (uintptr_t)&fsqrt_s << ";\\n";
+    os << "uint32_t (*fcmp_s)(uint32_t v1, uint32_t v2, uint32_t op)=" << (uintptr_t)&fcmp_s << ";\\n";
+    os << "uint32_t (*fcvt_s)(uint32_t v1, uint32_t op, uint8_t mode)=" << (uintptr_t)&fcvt_s << ";\\n";
+    os << "uint32_t (*fmadd_s)(uint32_t v1, uint32_t v2, uint32_t v3, uint32_t op, uint8_t mode)=" << (uintptr_t)&fmadd_s << ";\\n";
+    os << "uint32_t (*fsel_s)(uint32_t v1, uint32_t v2, uint32_t op)=" << (uintptr_t)&fsel_s << ";\\n";
+    os << "uint32_t (*fclass_s)( uint32_t v1 )=" << (uintptr_t)&fclass_s << ";\\n";
+    os << "uint32_t (*fconv_d2f)(uint64_t v1, uint8_t mode)=" << (uintptr_t)&fconv_d2f << ";\\n";
+    os << "uint64_t (*fconv_f2d)(uint32_t v1, uint8_t mode)=" << (uintptr_t)&fconv_f2d << ";\\n";
+    os << "uint64_t (*fadd_d)(uint64_t v1, uint64_t v2, uint8_t mode)=" << (uintptr_t)&fadd_d << ";\\n";
+    os << "uint64_t (*fsub_d)(uint64_t v1, uint64_t v2, uint8_t mode)=" << (uintptr_t)&fsub_d << ";\\n";
+    os << "uint64_t (*fmul_d)(uint64_t v1, uint64_t v2, uint8_t mode)=" << (uintptr_t)&fmul_d << ";\\n";
+    os << "uint64_t (*fdiv_d)(uint64_t v1, uint64_t v2, uint8_t mode)=" << (uintptr_t)&fdiv_d << ";\\n";
+    os << "uint64_t (*fsqrt_d)(uint64_t v1, uint8_t mode)=" << (uintptr_t)&fsqrt_d << ";\\n";
+    os << "uint64_t (*fcmp_d)(uint64_t v1, uint64_t v2, uint32_t op)=" << (uintptr_t)&fcmp_d << ";\\n";
+    os << "uint64_t (*fcvt_d)(uint64_t v1, uint32_t op, uint8_t mode)=" << (uintptr_t)&fcvt_d << ";\\n";
+    os << "uint64_t (*fmadd_d)(uint64_t v1, uint64_t v2, uint64_t v3, uint32_t op, uint8_t mode)=" << (uintptr_t)&fmadd_d << ";\\n";
+    os << "uint64_t (*fsel_d)(uint64_t v1, uint64_t v2, uint32_t op)=" << (uintptr_t)&fsel_d << ";\\n";
+    os << "uint64_t (*fclass_d)(uint64_t v1  )=" << (uintptr_t)&fclass_d << ";\\n";
+    os << "uint64_t (*fcvt_32_64)(uint32_t v1, uint32_t op, uint8_t mode)=" << (uintptr_t)&fcvt_32_64 << ";\\n";
+    os << "uint32_t (*fcvt_64_32)(uint64_t v1, uint32_t op, uint8_t mode)=" << (uintptr_t)&fcvt_64_32 << ";\\n";
+    os << "uint32_t (*unbox_s)(uint64_t v)=" << (uintptr_t)&unbox_s << ";\\n";
+    <%}%>
+    tu.add_prologue(os.str());
+}
+
 
 } // namespace ${coreDef.name.toLowerCase()}
 

softfloat/.gitignore

@@ -0,0 +1,2 @@
+build/*/*.o
+build/*/*.a

softfloat/README.md

@@ -0,0 +1,24 @@
+
+Package Overview for Berkeley SoftFloat Release 3e
+==================================================
+
+John R. Hauser<br>
+2018 January 20
+
+
+Berkeley SoftFloat is a software implementation of binary floating-point
+that conforms to the IEEE Standard for Floating-Point Arithmetic.  SoftFloat
+is distributed in the form of C source code.  Building the SoftFloat sources
+generates a library file (typically `softfloat.a` or `libsoftfloat.a`)
+containing the floating-point subroutines.
+
+
+The SoftFloat package is documented in the following files in the `doc`
+subdirectory:
+
+* [SoftFloat.html](http://www.jhauser.us/arithmetic/SoftFloat-3/doc/SoftFloat.html) Documentation for using the SoftFloat functions.
+* [SoftFloat-source.html](http://www.jhauser.us/arithmetic/SoftFloat-3/doc/SoftFloat-source.html) Documentation for building SoftFloat.
+* [SoftFloat-history.html](http://www.jhauser.us/arithmetic/SoftFloat-3/doc/SoftFloat-history.html) History of the major changes to SoftFloat.
+
+Other files in the package comprise the source code for SoftFloat.
+

softfloat/build/Linux-386-GCC/platform.h

@@ -50,3 +50,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *----------------------------------------------------------------------------*/
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #include "opts-GCC.h"
+

softfloat/build/Linux-386-SSE2-GCC/platform.h

@@ -50,3 +50,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *----------------------------------------------------------------------------*/
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #include "opts-GCC.h"
+

softfloat/build/Linux-ARM-VFPv2-GCC/platform.h

@@ -50,3 +50,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *----------------------------------------------------------------------------*/
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #include "opts-GCC.h"
+

softfloat/build/Linux-RISCV64-GCC/Makefile

@@ -0,0 +1,399 @@
+
+#=============================================================================
+#
+# This Makefile is part of the SoftFloat IEEE Floating-Point Arithmetic
+# Package, Release 3e, by John R. Hauser.
+#
+# Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+# University of California.  All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+#  1. Redistributions of source code must retain the above copyright notice,
+#     this list of conditions, and the following disclaimer.
+#
+#  2. Redistributions in binary form must reproduce the above copyright
+#     notice, this list of conditions, and the following disclaimer in the
+#     documentation and/or other materials provided with the distribution.
+#
+#  3. Neither the name of the University nor the names of its contributors
+#     may be used to endorse or promote products derived from this software
+#     without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+# DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+# DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+# (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+# ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+# THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+#=============================================================================
+
+SOURCE_DIR ?= ../../source
+SPECIALIZE_TYPE ?= RISCV
+MARCH ?= rv64gcv_zfh_zfhmin
+MABI ?= lp64d
+
+SOFTFLOAT_OPTS ?= \
+  -DSOFTFLOAT_ROUND_ODD -DINLINE_LEVEL=5 -DSOFTFLOAT_FAST_DIV32TO16 \
+  -DSOFTFLOAT_FAST_DIV64TO32
+
+DELETE = rm -f
+C_INCLUDES = -I. -I$(SOURCE_DIR)/$(SPECIALIZE_TYPE) -I$(SOURCE_DIR)/include
+COMPILE_C = \
+  riscv64-unknown-linux-gnu-gcc -c -march=$(MARCH) -mabi=$(MABI) -Werror-implicit-function-declaration -DSOFTFLOAT_FAST_INT64 \
+    $(SOFTFLOAT_OPTS) $(C_INCLUDES) -O2 -o $@
+MAKELIB = ar crs $@
+
+OBJ = .o
+LIB = .a
+
+OTHER_HEADERS = $(SOURCE_DIR)/include/opts-GCC.h
+
+.PHONY: all
+all: softfloat$(LIB)
+
+OBJS_PRIMITIVES = \
+  s_eq128$(OBJ) \
+  s_le128$(OBJ) \
+  s_lt128$(OBJ) \
+  s_shortShiftLeft128$(OBJ) \
+  s_shortShiftRight128$(OBJ) \
+  s_shortShiftRightJam64$(OBJ) \
+  s_shortShiftRightJam64Extra$(OBJ) \
+  s_shortShiftRightJam128$(OBJ) \
+  s_shortShiftRightJam128Extra$(OBJ) \
+  s_shiftRightJam32$(OBJ) \
+  s_shiftRightJam64$(OBJ) \
+  s_shiftRightJam64Extra$(OBJ) \
+  s_shiftRightJam128$(OBJ) \
+  s_shiftRightJam128Extra$(OBJ) \
+  s_shiftRightJam256M$(OBJ) \
+  s_countLeadingZeros8$(OBJ) \
+  s_countLeadingZeros16$(OBJ) \
+  s_countLeadingZeros32$(OBJ) \
+  s_countLeadingZeros64$(OBJ) \
+  s_add128$(OBJ) \
+  s_add256M$(OBJ) \
+  s_sub128$(OBJ) \
+  s_sub256M$(OBJ) \
+  s_mul64ByShifted32To128$(OBJ) \
+  s_mul64To128$(OBJ) \
+  s_mul128By32$(OBJ) \
+  s_mul128To256M$(OBJ) \
+  s_approxRecip_1Ks$(OBJ) \
+  s_approxRecip32_1$(OBJ) \
+  s_approxRecipSqrt_1Ks$(OBJ) \
+  s_approxRecipSqrt32_1$(OBJ) \
+
+OBJS_SPECIALIZE = \
+  softfloat_raiseFlags$(OBJ) \
+  s_f16UIToCommonNaN$(OBJ) \
+  s_commonNaNToF16UI$(OBJ) \
+  s_propagateNaNF16UI$(OBJ) \
+  s_bf16UIToCommonNaN$(OBJ) \
+  s_commonNaNToBF16UI$(OBJ) \
+  s_f32UIToCommonNaN$(OBJ) \
+  s_commonNaNToF32UI$(OBJ) \
+  s_propagateNaNF32UI$(OBJ) \
+  s_f64UIToCommonNaN$(OBJ) \
+  s_commonNaNToF64UI$(OBJ) \
+  s_propagateNaNF64UI$(OBJ) \
+  extF80M_isSignalingNaN$(OBJ) \
+  s_extF80UIToCommonNaN$(OBJ) \
+  s_commonNaNToExtF80UI$(OBJ) \
+  s_propagateNaNExtF80UI$(OBJ) \
+  f128M_isSignalingNaN$(OBJ) \
+  s_f128UIToCommonNaN$(OBJ) \
+  s_commonNaNToF128UI$(OBJ) \
+  s_propagateNaNF128UI$(OBJ) \
+
+OBJS_OTHERS = \
+  s_roundToUI32$(OBJ) \
+  s_roundToUI64$(OBJ) \
+  s_roundToI32$(OBJ) \
+  s_roundToI64$(OBJ) \
+  s_normSubnormalBF16Sig$(OBJ) \
+  s_roundPackToBF16$(OBJ) \
+  s_normSubnormalF16Sig$(OBJ) \
+  s_roundPackToF16$(OBJ) \
+  s_normRoundPackToF16$(OBJ) \
+  s_addMagsF16$(OBJ) \
+  s_subMagsF16$(OBJ) \
+  s_mulAddF16$(OBJ) \
+  s_normSubnormalF32Sig$(OBJ) \
+  s_roundPackToF32$(OBJ) \
+  s_normRoundPackToF32$(OBJ) \
+  s_addMagsF32$(OBJ) \
+  s_subMagsF32$(OBJ) \
+  s_mulAddF32$(OBJ) \
+  s_normSubnormalF64Sig$(OBJ) \
+  s_roundPackToF64$(OBJ) \
+  s_normRoundPackToF64$(OBJ) \
+  s_addMagsF64$(OBJ) \
+  s_subMagsF64$(OBJ) \
+  s_mulAddF64$(OBJ) \
+  s_normSubnormalExtF80Sig$(OBJ) \
+  s_roundPackToExtF80$(OBJ) \
+  s_normRoundPackToExtF80$(OBJ) \
+  s_addMagsExtF80$(OBJ) \
+  s_subMagsExtF80$(OBJ) \
+  s_normSubnormalF128Sig$(OBJ) \
+  s_roundPackToF128$(OBJ) \
+  s_normRoundPackToF128$(OBJ) \
+  s_addMagsF128$(OBJ) \
+  s_subMagsF128$(OBJ) \
+  s_mulAddF128$(OBJ) \
+  softfloat_state$(OBJ) \
+  ui32_to_f16$(OBJ) \
+  ui32_to_f32$(OBJ) \
+  ui32_to_f64$(OBJ) \
+  ui32_to_extF80$(OBJ) \
+  ui32_to_extF80M$(OBJ) \
+  ui32_to_f128$(OBJ) \
+  ui32_to_f128M$(OBJ) \
+  ui64_to_f16$(OBJ) \
+  ui64_to_f32$(OBJ) \
+  ui64_to_f64$(OBJ) \
+  ui64_to_extF80$(OBJ) \
+  ui64_to_extF80M$(OBJ) \
+  ui64_to_f128$(OBJ) \
+  ui64_to_f128M$(OBJ) \
+  i32_to_f16$(OBJ) \
+  i32_to_f32$(OBJ) \
+  i32_to_f64$(OBJ) \
+  i32_to_extF80$(OBJ) \
+  i32_to_extF80M$(OBJ) \
+  i32_to_f128$(OBJ) \
+  i32_to_f128M$(OBJ) \
+  i64_to_f16$(OBJ) \
+  i64_to_f32$(OBJ) \
+  i64_to_f64$(OBJ) \
+  i64_to_extF80$(OBJ) \
+  i64_to_extF80M$(OBJ) \
+  i64_to_f128$(OBJ) \
+  i64_to_f128M$(OBJ) \
+  bf16_isSignalingNaN$(OBJ) \
+  bf16_to_f32$(OBJ) \
+  f16_to_ui32$(OBJ) \
+  f16_to_ui64$(OBJ) \
+  f16_to_i32$(OBJ) \
+  f16_to_i64$(OBJ) \
+  f16_to_ui32_r_minMag$(OBJ) \
+  f16_to_ui64_r_minMag$(OBJ) \
+  f16_to_i32_r_minMag$(OBJ) \
+  f16_to_i64_r_minMag$(OBJ) \
+  f16_to_f32$(OBJ) \
+  f16_to_f64$(OBJ) \
+  f16_to_extF80$(OBJ) \
+  f16_to_extF80M$(OBJ) \
+  f16_to_f128$(OBJ) \
+  f16_to_f128M$(OBJ) \
+  f16_roundToInt$(OBJ) \
+  f16_add$(OBJ) \
+  f16_sub$(OBJ) \
+  f16_mul$(OBJ) \
+  f16_mulAdd$(OBJ) \
+  f16_div$(OBJ) \
+  f16_rem$(OBJ) \
+  f16_sqrt$(OBJ) \
+  f16_eq$(OBJ) \
+  f16_le$(OBJ) \
+  f16_lt$(OBJ) \
+  f16_eq_signaling$(OBJ) \
+  f16_le_quiet$(OBJ) \
+  f16_lt_quiet$(OBJ) \
+  f16_isSignalingNaN$(OBJ) \
+  f32_to_ui32$(OBJ) \
+  f32_to_ui64$(OBJ) \
+  f32_to_i32$(OBJ) \
+  f32_to_i64$(OBJ) \
+  f32_to_ui32_r_minMag$(OBJ) \
+  f32_to_ui64_r_minMag$(OBJ) \
+  f32_to_i32_r_minMag$(OBJ) \
+  f32_to_i64_r_minMag$(OBJ) \
+  f32_to_bf16$(OBJ) \
+  f32_to_f16$(OBJ) \
+  f32_to_f64$(OBJ) \
+  f32_to_extF80$(OBJ) \
+  f32_to_extF80M$(OBJ) \
+  f32_to_f128$(OBJ) \
+  f32_to_f128M$(OBJ) \
+  f32_roundToInt$(OBJ) \
+  f32_add$(OBJ) \
+  f32_sub$(OBJ) \
+  f32_mul$(OBJ) \
+  f32_mulAdd$(OBJ) \
+  f32_div$(OBJ) \
+  f32_rem$(OBJ) \
+  f32_sqrt$(OBJ) \
+  f32_eq$(OBJ) \
+  f32_le$(OBJ) \
+  f32_lt$(OBJ) \
+  f32_eq_signaling$(OBJ) \
+  f32_le_quiet$(OBJ) \
+  f32_lt_quiet$(OBJ) \
+  f32_isSignalingNaN$(OBJ) \
+  f64_to_ui32$(OBJ) \
+  f64_to_ui64$(OBJ) \
+  f64_to_i32$(OBJ) \
+  f64_to_i64$(OBJ) \
+  f64_to_ui32_r_minMag$(OBJ) \
+  f64_to_ui64_r_minMag$(OBJ) \
+  f64_to_i32_r_minMag$(OBJ) \
+  f64_to_i64_r_minMag$(OBJ) \
+  f64_to_f16$(OBJ) \
+  f64_to_f32$(OBJ) \
+  f64_to_extF80$(OBJ) \
+  f64_to_extF80M$(OBJ) \
+  f64_to_f128$(OBJ) \
+  f64_to_f128M$(OBJ) \
+  f64_roundToInt$(OBJ) \
+  f64_add$(OBJ) \
+  f64_sub$(OBJ) \
+  f64_mul$(OBJ) \
+  f64_mulAdd$(OBJ) \
+  f64_div$(OBJ) \
+  f64_rem$(OBJ) \
+  f64_sqrt$(OBJ) \
+  f64_eq$(OBJ) \
+  f64_le$(OBJ) \
+  f64_lt$(OBJ) \
+  f64_eq_signaling$(OBJ) \
+  f64_le_quiet$(OBJ) \
+  f64_lt_quiet$(OBJ) \
+  f64_isSignalingNaN$(OBJ) \
+  extF80_to_ui32$(OBJ) \
+  extF80_to_ui64$(OBJ) \
+  extF80_to_i32$(OBJ) \
+  extF80_to_i64$(OBJ) \
+  extF80_to_ui32_r_minMag$(OBJ) \
+  extF80_to_ui64_r_minMag$(OBJ) \
+  extF80_to_i32_r_minMag$(OBJ) \
+  extF80_to_i64_r_minMag$(OBJ) \
+  extF80_to_f16$(OBJ) \
+  extF80_to_f32$(OBJ) \
+  extF80_to_f64$(OBJ) \
+  extF80_to_f128$(OBJ) \
+  extF80_roundToInt$(OBJ) \
+  extF80_add$(OBJ) \
+  extF80_sub$(OBJ) \
+  extF80_mul$(OBJ) \
+  extF80_div$(OBJ) \
+  extF80_rem$(OBJ) \
+  extF80_sqrt$(OBJ) \
+  extF80_eq$(OBJ) \
+  extF80_le$(OBJ) \
+  extF80_lt$(OBJ) \
+  extF80_eq_signaling$(OBJ) \
+  extF80_le_quiet$(OBJ) \
+  extF80_lt_quiet$(OBJ) \
+  extF80_isSignalingNaN$(OBJ) \
+  extF80M_to_ui32$(OBJ) \
+  extF80M_to_ui64$(OBJ) \
+  extF80M_to_i32$(OBJ) \
+  extF80M_to_i64$(OBJ) \
+  extF80M_to_ui32_r_minMag$(OBJ) \
+  extF80M_to_ui64_r_minMag$(OBJ) \
+  extF80M_to_i32_r_minMag$(OBJ) \
+  extF80M_to_i64_r_minMag$(OBJ) \
+  extF80M_to_f16$(OBJ) \
+  extF80M_to_f32$(OBJ) \
+  extF80M_to_f64$(OBJ) \
+  extF80M_to_f128M$(OBJ) \
+  extF80M_roundToInt$(OBJ) \
+  extF80M_add$(OBJ) \
+  extF80M_sub$(OBJ) \
+  extF80M_mul$(OBJ) \
+  extF80M_div$(OBJ) \
+  extF80M_rem$(OBJ) \
+  extF80M_sqrt$(OBJ) \
+  extF80M_eq$(OBJ) \
+  extF80M_le$(OBJ) \
+  extF80M_lt$(OBJ) \
+  extF80M_eq_signaling$(OBJ) \
+  extF80M_le_quiet$(OBJ) \
+  extF80M_lt_quiet$(OBJ) \
+  f128_to_ui32$(OBJ) \
+  f128_to_ui64$(OBJ) \
+  f128_to_i32$(OBJ) \
+  f128_to_i64$(OBJ) \
+  f128_to_ui32_r_minMag$(OBJ) \
+  f128_to_ui64_r_minMag$(OBJ) \
+  f128_to_i32_r_minMag$(OBJ) \
+  f128_to_i64_r_minMag$(OBJ) \
+  f128_to_f16$(OBJ) \
+  f128_to_f32$(OBJ) \
+  f128_to_extF80$(OBJ) \
+  f128_to_f64$(OBJ) \
+  f128_roundToInt$(OBJ) \
+  f128_add$(OBJ) \
+  f128_sub$(OBJ) \
+  f128_mul$(OBJ) \
+  f128_mulAdd$(OBJ) \
+  f128_div$(OBJ) \
+  f128_rem$(OBJ) \
+  f128_sqrt$(OBJ) \
+  f128_eq$(OBJ) \
+  f128_le$(OBJ) \
+  f128_lt$(OBJ) \
+  f128_eq_signaling$(OBJ) \
+  f128_le_quiet$(OBJ) \
+  f128_lt_quiet$(OBJ) \
+  f128_isSignalingNaN$(OBJ) \
+  f128M_to_ui32$(OBJ) \
+  f128M_to_ui64$(OBJ) \
+  f128M_to_i32$(OBJ) \
+  f128M_to_i64$(OBJ) \
+  f128M_to_ui32_r_minMag$(OBJ) \
+  f128M_to_ui64_r_minMag$(OBJ) \
+  f128M_to_i32_r_minMag$(OBJ) \
+  f128M_to_i64_r_minMag$(OBJ) \
+  f128M_to_f16$(OBJ) \
+  f128M_to_f32$(OBJ) \
+  f128M_to_extF80M$(OBJ) \
+  f128M_to_f64$(OBJ) \
+  f128M_roundToInt$(OBJ) \
+  f128M_add$(OBJ) \
+  f128M_sub$(OBJ) \
+  f128M_mul$(OBJ) \
+  f128M_mulAdd$(OBJ) \
+  f128M_div$(OBJ) \
+  f128M_rem$(OBJ) \
+  f128M_sqrt$(OBJ) \
+  f128M_eq$(OBJ) \
+  f128M_le$(OBJ) \
+  f128M_lt$(OBJ) \
+  f128M_eq_signaling$(OBJ) \
+  f128M_le_quiet$(OBJ) \
+  f128M_lt_quiet$(OBJ) \
+
+OBJS_ALL = $(OBJS_PRIMITIVES) $(OBJS_SPECIALIZE) $(OBJS_OTHERS)
+
+$(OBJS_ALL): \
+  $(OTHER_HEADERS) platform.h $(SOURCE_DIR)/include/primitiveTypes.h \
+  $(SOURCE_DIR)/include/primitives.h
+$(OBJS_SPECIALIZE) $(OBJS_OTHERS): \
+  $(SOURCE_DIR)/include/softfloat_types.h $(SOURCE_DIR)/include/internals.h \
+  $(SOURCE_DIR)/$(SPECIALIZE_TYPE)/specialize.h \
+  $(SOURCE_DIR)/include/softfloat.h
+
+$(OBJS_PRIMITIVES) $(OBJS_OTHERS): %$(OBJ): $(SOURCE_DIR)/%.c
+	$(COMPILE_C) $(SOURCE_DIR)/$*.c
+
+$(OBJS_SPECIALIZE): %$(OBJ): $(SOURCE_DIR)/$(SPECIALIZE_TYPE)/%.c
+	$(COMPILE_C) $(SOURCE_DIR)/$(SPECIALIZE_TYPE)/$*.c
+
+softfloat$(LIB): $(OBJS_ALL)
+	$(DELETE) $@
+	$(MAKELIB) $^
+
+.PHONY: clean
+clean:
+	$(DELETE) $(OBJS_ALL) softfloat$(LIB)
+

softfloat/build/Linux-RISCV64-GCC/platform.h

@@ -0,0 +1,54 @@
+
+/*============================================================================
+
+This C header file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+/*----------------------------------------------------------------------------
+*----------------------------------------------------------------------------*/
+#define LITTLEENDIAN 1
+
+/*----------------------------------------------------------------------------
+*----------------------------------------------------------------------------*/
+#ifdef __GNUC_STDC_INLINE__
+#define INLINE inline
+#else
+#define INLINE extern inline
+#endif
+
+/*----------------------------------------------------------------------------
+*----------------------------------------------------------------------------*/
+#define SOFTFLOAT_BUILTIN_CLZ 1
+#define SOFTFLOAT_INTRINSIC_INT128 1
+#include "opts-GCC.h"
+

softfloat/build/Linux-x86_64-GCC/Makefile

@@ -94,6 +94,8 @@ OBJS_SPECIALIZE = \
   s_f16UIToCommonNaN$(OBJ) \
   s_commonNaNToF16UI$(OBJ) \
   s_propagateNaNF16UI$(OBJ) \
+  s_bf16UIToCommonNaN$(OBJ) \
+  s_commonNaNToBF16UI$(OBJ) \
   s_f32UIToCommonNaN$(OBJ) \
   s_commonNaNToF32UI$(OBJ) \
   s_propagateNaNF32UI$(OBJ) \
@@ -114,6 +116,8 @@ OBJS_OTHERS = \
   s_roundToUI64$(OBJ) \
   s_roundToI32$(OBJ) \
   s_roundToI64$(OBJ) \
+  s_normSubnormalBF16Sig$(OBJ) \
+  s_roundPackToBF16$(OBJ) \
   s_normSubnormalF16Sig$(OBJ) \
   s_roundPackToF16$(OBJ) \
   s_normRoundPackToF16$(OBJ) \
@@ -172,6 +176,8 @@ OBJS_OTHERS = \
   i64_to_extF80M$(OBJ) \
   i64_to_f128$(OBJ) \
   i64_to_f128M$(OBJ) \
+  bf16_isSignalingNaN$(OBJ) \
+  bf16_to_f32$(OBJ) \
   f16_to_ui32$(OBJ) \
   f16_to_ui64$(OBJ) \
   f16_to_i32$(OBJ) \
@@ -209,6 +215,7 @@ OBJS_OTHERS = \
   f32_to_ui64_r_minMag$(OBJ) \
   f32_to_i32_r_minMag$(OBJ) \
   f32_to_i64_r_minMag$(OBJ) \
+  f32_to_bf16$(OBJ) \
   f32_to_f16$(OBJ) \
   f32_to_f64$(OBJ) \
   f32_to_extF80$(OBJ) \

softfloat/build/Win32-MinGW/platform.h

@@ -50,3 +50,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *----------------------------------------------------------------------------*/
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #include "opts-GCC.h"
+

softfloat/build/Win32-SSE2-MinGW/platform.h

@@ -50,3 +50,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *----------------------------------------------------------------------------*/
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #include "opts-GCC.h"
+

softfloat/build/Win64-MinGW-w64/platform.h

@@ -51,3 +51,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SOFTFLOAT_BUILTIN_CLZ 1
 #define SOFTFLOAT_INTRINSIC_INT128 1
 #include "opts-GCC.h"
+

softfloat/build/template-FAST_INT64/Makefile

@@ -115,6 +115,8 @@ OBJS_OTHERS = \
   s_roundToUI64$(OBJ) \
   s_roundToI32$(OBJ) \
   s_roundToI64$(OBJ) \
+  s_normSubnormalBF16Sig$(OBJ) \
+  s_roundPackToBF16$(OBJ) \
   s_normSubnormalF16Sig$(OBJ) \
   s_roundPackToF16$(OBJ) \
   s_normRoundPackToF16$(OBJ) \
@@ -173,6 +175,8 @@ OBJS_OTHERS = \
   i64_to_extF80M$(OBJ) \
   i64_to_f128$(OBJ) \
   i64_to_f128M$(OBJ) \
+  bf16_isSignalingNaN$(OBJ) \
+  bf16_to_f32$(OBJ) \
   f16_to_ui32$(OBJ) \
   f16_to_ui64$(OBJ) \
   f16_to_i32$(OBJ) \
@@ -210,6 +214,7 @@ OBJS_OTHERS = \
   f32_to_ui64_r_minMag$(OBJ) \
   f32_to_i32_r_minMag$(OBJ) \
   f32_to_i64_r_minMag$(OBJ) \
+  f32_to_bf16$(OBJ) \
   f32_to_f16$(OBJ) \
   f32_to_f64$(OBJ) \
   f32_to_extF80$(OBJ) \

softfloat/build/template-FAST_INT64/platform.h

@@ -47,3 +47,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /*----------------------------------------------------------------------------
 *----------------------------------------------------------------------------*/
 ==> #define THREAD_LOCAL _Thread_local
+

softfloat/build/template-not-FAST_INT64/platform.h

@@ -47,3 +47,4 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /*----------------------------------------------------------------------------
 *----------------------------------------------------------------------------*/
 ==> #define THREAD_LOCAL _Thread_local
+

softfloat/doc/SoftFloat.html

@@ -508,7 +508,7 @@ significant extra cost.
 On computers where the word size is <NOBR>64 bits</NOBR> or larger, both
 function versions (<CODE>f128M_add</CODE> and <CODE>f128_add</CODE>) are
 provided, because the cost of passing by value is then more reasonable.
-Applications that must be portable accross both classes of computers must use
+Applications that must be portable across both classes of computers must use
 the pointer-based functions, as these are always implemented.
 However, if it is known that SoftFloat includes the by-value functions for all
 platforms of interest, programmers can use whichever version they prefer.

softfloat/source/8086-SSE/s_bf16UIToCommonNaN.c

@@ -0,0 +1,59 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdint.h>
+#include "platform.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+/*----------------------------------------------------------------------------
+| Assuming `uiA' has the bit pattern of a BF16 NaN, converts
+| this NaN to the common NaN form, and stores the resulting common NaN at the
+| location pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid
+| exception is raised.
+*----------------------------------------------------------------------------*/
+void softfloat_bf16UIToCommonNaN( uint_fast16_t uiA, struct commonNaN *zPtr )
+{
+
+    if ( softfloat_isSigNaNBF16UI( uiA ) ) {
+        softfloat_raiseFlags( softfloat_flag_invalid );
+    }
+    zPtr->sign = uiA>>15;
+    zPtr->v64  = (uint_fast64_t) uiA<<56;
+    zPtr->v0   = 0;
+
+}
+

softfloat/source/8086-SSE/s_commonNaNToBF16UI.c

@@ -0,0 +1,51 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdint.h>
+#include "platform.h"
+#include "specialize.h"
+
+/*----------------------------------------------------------------------------
+| Converts the common NaN pointed to by `aPtr' into a BF16 NaN, and 
+| returns the bit pattern of this value as an unsigned integer.
+*----------------------------------------------------------------------------*/
+uint_fast16_t softfloat_commonNaNToBF16UI( const struct commonNaN *aPtr )
+{
+
+    return (uint_fast16_t) aPtr->sign<<15 | 0x7FC0 | aPtr->v64>>56;
+
+}
+

softfloat/source/8086-SSE/specialize.h

@@ -116,6 +116,27 @@ uint_fast16_t softfloat_commonNaNToF16UI(const struct commonNaN* aPtr);
 *----------------------------------------------------------------------------*/
 uint_fast16_t softfloat_propagateNaNF16UI(uint_fast16_t uiA, uint_fast16_t uiB);
 
+/*----------------------------------------------------------------------------
+| Returns true when 16-bit unsigned integer 'uiA' has the bit pattern of a
+| 16-bit brain floating-point (BF16) signaling NaN.
+| Note:  This macro evaluates its argument more than once.
+*----------------------------------------------------------------------------*/
+#define softfloat_isSigNaNBF16UI(uiA) ((((uiA)&0x7FC0) == 0x7F80) && ((uiA)&0x003F))
+
+/*----------------------------------------------------------------------------
+| Assuming 'uiA' has the bit pattern of a 16-bit BF16 floating-point NaN, converts
+| this NaN to the common NaN form, and stores the resulting common NaN at the
+| location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
+| exception is raised.
+*----------------------------------------------------------------------------*/
+void softfloat_bf16UIToCommonNaN(uint_fast16_t uiA, struct commonNaN* zPtr);
+
+/*----------------------------------------------------------------------------
+| Converts the common NaN pointed to by 'aPtr' into a 16-bit floating-point
+| NaN, and returns the bit pattern of this value as an unsigned integer.
+*----------------------------------------------------------------------------*/
+uint_fast16_t softfloat_commonNaNToBF16UI(const struct commonNaN* aPtr);
+
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 32-bit floating-point NaN.
 *----------------------------------------------------------------------------*/

softfloat/source/RISCV/s_bf16UIToCommonNaN.c

@@ -0,0 +1,5 @@
+
+/*----------------------------------------------------------------------------
+| This file intentionally contains no code.
+*----------------------------------------------------------------------------*/
+

softfloat/source/RISCV/s_commonNaNToBF16UI.c

@@ -0,0 +1,5 @@
+
+/*----------------------------------------------------------------------------
+| This file intentionally contains no code.
+*----------------------------------------------------------------------------*/
+

softfloat/source/RISCV/s_commonNaNToExtF80M.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,9 +34,10 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdint.h>
 #include "platform.h"
-#include "internals.h"
+#include "softfloat_types.h"
+
+#define softfloat_commonNaNToExtF80M softfloat_commonNaNToExtF80M
 #include "specialize.h"
 
 /*----------------------------------------------------------------------------
@@ -49,8 +50,8 @@ void
      const struct commonNaN *aPtr, struct extFloat80M *zSPtr )
 {
 
-    zSPtr->signExp = packToExtF80UI64( aPtr->sign, 0x7FFF );
-    zSPtr->signif = UINT64_C( 0xC000000000000000 ) | aPtr->v64>>1;
+    zSPtr->signExp = defaultNaNExtF80UI64;
+    zSPtr->signif  = defaultNaNExtF80UI0;
 
 }
 

softfloat/source/RISCV/s_commonNaNToExtF80UI.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,9 +34,10 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdint.h>
 #include "platform.h"
-#include "primitives.h"
+#include "primitiveTypes.h"
+
+#define softfloat_commonNaNToExtF80UI softfloat_commonNaNToExtF80UI
 #include "specialize.h"
 
 /*----------------------------------------------------------------------------
@@ -48,8 +49,8 @@ struct uint128 softfloat_commonNaNToExtF80UI( const struct commonNaN *aPtr )
 {
     struct uint128 uiZ;
 
-    uiZ.v64 = (uint_fast16_t) aPtr->sign<<15 | 0x7FFF;
-    uiZ.v0 = UINT64_C( 0xC000000000000000 ) | aPtr->v64>>1;
+    uiZ.v64 = defaultNaNExtF80UI64;
+    uiZ.v0  = defaultNaNExtF80UI0;
     return uiZ;
 
 }

softfloat/source/RISCV/s_commonNaNToF128M.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -36,7 +36,9 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include <stdint.h>
 #include "platform.h"
-#include "primitives.h"
+#include "primitiveTypes.h"
+
+#define softfloat_commonNaNToF128M softfloat_commonNaNToF128M
 #include "specialize.h"
 
 /*----------------------------------------------------------------------------
@@ -49,8 +51,10 @@ void
  softfloat_commonNaNToF128M( const struct commonNaN *aPtr, uint32_t *zWPtr )
 {
 
-    softfloat_shortShiftRight128M( (const uint32_t *) &aPtr->v0, 16, zWPtr );
-    zWPtr[indexWordHi( 4 )] |= (uint32_t) aPtr->sign<<31 | 0x7FFF8000;
+    zWPtr[indexWord( 4, 3 )] = defaultNaNF128UI96;
+    zWPtr[indexWord( 4, 2 )] = defaultNaNF128UI64;
+    zWPtr[indexWord( 4, 1 )] = defaultNaNF128UI32;
+    zWPtr[indexWord( 4, 0 )] = defaultNaNF128UI0;
 
 }
 

softfloat/source/RISCV/s_commonNaNToF128UI.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,9 +34,10 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdint.h>
 #include "platform.h"
-#include "primitives.h"
+#include "primitiveTypes.h"
+
+#define softfloat_commonNaNToF128UI softfloat_commonNaNToF128UI
 #include "specialize.h"
 
 /*----------------------------------------------------------------------------
@@ -47,8 +48,8 @@ struct uint128 softfloat_commonNaNToF128UI( const struct commonNaN *aPtr )
 {
     struct uint128 uiZ;
 
-    uiZ = softfloat_shortShiftRight128( aPtr->v64, aPtr->v0, 16 );
-    uiZ.v64 |= (uint_fast64_t) aPtr->sign<<63 | UINT64_C( 0x7FFF800000000000 );
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0  = defaultNaNF128UI0;
     return uiZ;
 
 }

softfloat/source/RISCV/s_commonNaNToF16UI.c

@@ -1,51 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
-California.  All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-
 /*----------------------------------------------------------------------------
-| Converts the common NaN pointed to by `aPtr' into a 16-bit floating-point
-| NaN, and returns the bit pattern of this value as an unsigned integer.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-uint_fast16_t softfloat_commonNaNToF16UI( const struct commonNaN *aPtr )
-{
-
-    return (uint_fast16_t) aPtr->sign<<15 | 0x7E00 | aPtr->v64>>54;
-
-}
 

softfloat/source/RISCV/s_commonNaNToF32UI.c

@@ -1,51 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-
 /*----------------------------------------------------------------------------
-| Converts the common NaN pointed to by `aPtr' into a 32-bit floating-point
-| NaN, and returns the bit pattern of this value as an unsigned integer.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-uint_fast32_t softfloat_commonNaNToF32UI( const struct commonNaN *aPtr )
-{
-
-    return (uint_fast32_t) aPtr->sign<<31 | 0x7FC00000 | aPtr->v64>>41;
-
-}
 

softfloat/source/RISCV/s_commonNaNToF64UI.c

@@ -1,53 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-
 /*----------------------------------------------------------------------------
-| Converts the common NaN pointed to by `aPtr' into a 64-bit floating-point
-| NaN, and returns the bit pattern of this value as an unsigned integer.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-uint_fast64_t softfloat_commonNaNToF64UI( const struct commonNaN *aPtr )
-{
-
-    return
-        (uint_fast64_t) aPtr->sign<<63 | UINT64_C( 0x7FF8000000000000 )
-            | aPtr->v64>>12;
-
-}
 

softfloat/source/RISCV/s_extF80MToCommonNaN.c

@@ -1,62 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "internals.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming the 80-bit extended floating-point value pointed to by `aSPtr' is
-| a NaN, converts this NaN to the common NaN form, and stores the resulting
-| common NaN at the location pointed to by `zPtr'.  If the NaN is a signaling
-| NaN, the invalid exception is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void
- softfloat_extF80MToCommonNaN(
-     const struct extFloat80M *aSPtr, struct commonNaN *zPtr )
-{
-
-    if ( extF80M_isSignalingNaN( (const extFloat80_t *) aSPtr ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = signExtF80UI64( aSPtr->signExp );
-    zPtr->v64 = aSPtr->signif<<1;
-    zPtr->v0  = 0;
-
-}
 

softfloat/source/RISCV/s_extF80UIToCommonNaN.c

@@ -1,62 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming the unsigned integer formed from concatenating `uiA64' and `uiA0'
-| has the bit pattern of an 80-bit extended floating-point NaN, converts
-| this NaN to the common NaN form, and stores the resulting common NaN at the
-| location pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid
-| exception is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void
- softfloat_extF80UIToCommonNaN(
-     uint_fast16_t uiA64, uint_fast64_t uiA0, struct commonNaN *zPtr )
-{
-
-    if ( softfloat_isSigNaNExtF80UI( uiA64, uiA0 ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = uiA64>>15;
-    zPtr->v64  = uiA0<<1;
-    zPtr->v0   = 0;
-
-}
 

softfloat/source/RISCV/s_f128MToCommonNaN.c

@@ -1,62 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "primitives.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming the 128-bit floating-point value pointed to by `aWPtr' is a NaN,
-| converts this NaN to the common NaN form, and stores the resulting common
-| NaN at the location pointed to by `zPtr'.  If the NaN is a signaling NaN,
-| the invalid exception is raised.  Argument `aWPtr' points to an array of
-| four 32-bit elements that concatenate in the platform's normal endian order
-| to form a 128-bit floating-point value.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void
- softfloat_f128MToCommonNaN( const uint32_t *aWPtr, struct commonNaN *zPtr )
-{
-
-    if ( f128M_isSignalingNaN( (const float128_t *) aWPtr ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = aWPtr[indexWordHi( 4 )]>>31;
-    softfloat_shortShiftLeft128M( aWPtr, 16, (uint32_t *) &zPtr->v0 );
-
-}
 

softfloat/source/RISCV/s_f128UIToCommonNaN.c

@@ -1,65 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "primitives.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming the unsigned integer formed from concatenating `uiA64' and `uiA0'
-| has the bit pattern of a 128-bit floating-point NaN, converts this NaN to
-| the common NaN form, and stores the resulting common NaN at the location
-| pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid exception
-| is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void
- softfloat_f128UIToCommonNaN(
-     uint_fast64_t uiA64, uint_fast64_t uiA0, struct commonNaN *zPtr )
-{
-    struct uint128 NaNSig;
-
-    if ( softfloat_isSigNaNF128UI( uiA64, uiA0 ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    NaNSig = softfloat_shortShiftLeft128( uiA64, uiA0, 16 );
-    zPtr->sign = uiA64>>63;
-    zPtr->v64  = NaNSig.v64;
-    zPtr->v0   = NaNSig.v0;
-
-}
 

softfloat/source/RISCV/s_f16UIToCommonNaN.c

@@ -1,59 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
-California.  All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming `uiA' has the bit pattern of a 16-bit floating-point NaN, converts
-| this NaN to the common NaN form, and stores the resulting common NaN at the
-| location pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid
-| exception is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void softfloat_f16UIToCommonNaN( uint_fast16_t uiA, struct commonNaN *zPtr )
-{
-
-    if ( softfloat_isSigNaNF16UI( uiA ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = uiA>>15;
-    zPtr->v64  = (uint_fast64_t) uiA<<54;
-    zPtr->v0   = 0;
-
-}
 

softfloat/source/RISCV/s_f32UIToCommonNaN.c

@@ -1,59 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming `uiA' has the bit pattern of a 32-bit floating-point NaN, converts
-| this NaN to the common NaN form, and stores the resulting common NaN at the
-| location pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid
-| exception is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void softfloat_f32UIToCommonNaN( uint_fast32_t uiA, struct commonNaN *zPtr )
-{
-
-    if ( softfloat_isSigNaNF32UI( uiA ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = uiA>>31;
-    zPtr->v64  = (uint_fast64_t) uiA<<41;
-    zPtr->v0   = 0;
-
-}
 

softfloat/source/RISCV/s_f64UIToCommonNaN.c

@@ -1,59 +1,5 @@
 
-/*============================================================================
-
-This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
-Package, Release 3e, by John R. Hauser.
-
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
- 1. Redistributions of source code must retain the above copyright notice,
-    this list of conditions, and the following disclaimer.
-
- 2. Redistributions in binary form must reproduce the above copyright notice,
-    this list of conditions, and the following disclaimer in the documentation
-    and/or other materials provided with the distribution.
-
- 3. Neither the name of the University nor the names of its contributors may
-    be used to endorse or promote products derived from this software without
-    specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
-EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
-DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
-DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
-ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-=============================================================================*/
-
-#include <stdint.h>
-#include "platform.h"
-#include "specialize.h"
-#include "softfloat.h"
-
 /*----------------------------------------------------------------------------
-| Assuming `uiA' has the bit pattern of a 64-bit floating-point NaN, converts
-| this NaN to the common NaN form, and stores the resulting common NaN at the
-| location pointed to by `zPtr'.  If the NaN is a signaling NaN, the invalid
-| exception is raised.
+| This file intentionally contains no code.
 *----------------------------------------------------------------------------*/
-void softfloat_f64UIToCommonNaN( uint_fast64_t uiA, struct commonNaN *zPtr )
-{
-
-    if ( softfloat_isSigNaNF64UI( uiA ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
-    }
-    zPtr->sign = uiA>>63;
-    zPtr->v64  = uiA<<12;
-    zPtr->v0   = 0;
-
-}
 

softfloat/source/RISCV/s_propagateNaNExtF80M.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,10 +34,9 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
+#include "primitiveTypes.h"
 #include "specialize.h"
 #include "softfloat.h"
 
@@ -54,54 +53,22 @@ void
      struct extFloat80M *zSPtr
  )
 {
-    bool isSigNaNA;
-    const struct extFloat80M *sPtr;
-    bool isSigNaNB;
-    uint_fast16_t uiB64;
-    uint64_t uiB0;
-    uint_fast16_t uiA64;
-    uint64_t uiA0;
-    uint_fast16_t uiMagA64, uiMagB64;
+    uint_fast16_t ui64;
+    uint_fast64_t ui0;
 
-    isSigNaNA = extF80M_isSignalingNaN( (const extFloat80_t *) aSPtr );
-    sPtr = aSPtr;
-    if ( ! bSPtr ) {
-        if ( isSigNaNA ) softfloat_raiseFlags( softfloat_flag_invalid );
-        goto copy;
-    }
-    isSigNaNB = extF80M_isSignalingNaN( (const extFloat80_t *) bSPtr );
-    if ( isSigNaNA | isSigNaNB ) {
+    ui64 = aSPtr->signExp;
+    ui0  = aSPtr->signif;
+    if (
+        softfloat_isSigNaNExtF80UI( ui64, ui0 )
+            || (bSPtr
+                    && (ui64 = bSPtr->signExp,
+                        ui0  = bSPtr->signif,
+                        softfloat_isSigNaNExtF80UI( ui64, ui0 )))
+    ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) {
-            uiB64 = bSPtr->signExp;
-            if ( isSigNaNB ) goto returnLargerUIMag;
-            uiB0 = bSPtr->signif;
-            if ( isNaNExtF80UI( uiB64, uiB0 ) ) goto copyB;
-            goto copy;
-        } else {
-            uiA64 = aSPtr->signExp;
-            uiA0 = aSPtr->signif;
-            if ( isNaNExtF80UI( uiA64, uiA0 ) ) goto copy;
-            goto copyB;
     }
-    }
-    uiB64 = bSPtr->signExp;
- returnLargerUIMag:
-    uiA64 = aSPtr->signExp;
-    uiMagA64 = uiA64 & 0x7FFF;
-    uiMagB64 = uiB64 & 0x7FFF;
-    if ( uiMagA64 < uiMagB64 ) goto copyB;
-    if ( uiMagB64 < uiMagA64 ) goto copy;
-    uiA0 = aSPtr->signif;
-    uiB0 = bSPtr->signif;
-    if ( uiA0 < uiB0 ) goto copyB;
-    if ( uiB0 < uiA0 ) goto copy;
-    if ( uiA64 < uiB64 ) goto copy;
- copyB:
-    sPtr = bSPtr;
- copy:
-    zSPtr->signExp = sPtr->signExp;
-    zSPtr->signif = sPtr->signif | UINT64_C( 0xC000000000000000 );
+    zSPtr->signExp = defaultNaNExtF80UI64;
+    zSPtr->signif  = defaultNaNExtF80UI0;
 
 }
 

softfloat/source/RISCV/s_propagateNaNExtF80UI.c

@@ -4,7 +4,7 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014, 2018 The Regents of the University of
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
 California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -34,17 +34,16 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
+#include "primitiveTypes.h"
 #include "specialize.h"
 #include "softfloat.h"
 
 /*----------------------------------------------------------------------------
-| Interpreting the unsigned integer formed from concatenating 'uiA64' and
-| 'uiA0' as an 80-bit extended floating-point value, and likewise interpreting
-| the unsigned integer formed from concatenating 'uiB64' and 'uiB0' as another
+| Interpreting the unsigned integer formed from concatenating `uiA64' and
+| `uiA0' as an 80-bit extended floating-point value, and likewise interpreting
+| the unsigned integer formed from concatenating `uiB64' and `uiB0' as another
 | 80-bit extended floating-point value, and assuming at least on of these
 | floating-point values is a NaN, returns the bit pattern of the combined NaN
 | result.  If either original floating-point value is a signaling NaN, the
@@ -58,48 +57,16 @@ struct uint128
      uint_fast64_t uiB0
  )
 {
-    bool isSigNaNA, isSigNaNB;
-    uint_fast64_t uiNonsigA0, uiNonsigB0;
-    uint_fast16_t uiMagA64, uiMagB64;
     struct uint128 uiZ;
 
-    /*------------------------------------------------------------------------
-    *------------------------------------------------------------------------*/
-    isSigNaNA = softfloat_isSigNaNExtF80UI( uiA64, uiA0 );
-    isSigNaNB = softfloat_isSigNaNExtF80UI( uiB64, uiB0 );
-    /*------------------------------------------------------------------------
-    | Make NaNs non-signaling.
-    *------------------------------------------------------------------------*/
-    uiNonsigA0 = uiA0 | UINT64_C( 0xC000000000000000 );
-    uiNonsigB0 = uiB0 | UINT64_C( 0xC000000000000000 );
-    /*------------------------------------------------------------------------
-    *------------------------------------------------------------------------*/
-    if ( isSigNaNA | isSigNaNB ) {
+    if (
+           softfloat_isSigNaNExtF80UI( uiA64, uiA0 )
+        || softfloat_isSigNaNExtF80UI( uiB64, uiB0 )
+    ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) {
-            if ( isSigNaNB ) goto returnLargerMag;
-            if ( isNaNExtF80UI( uiB64, uiB0 ) ) goto returnB;
-            goto returnA;
-        } else {
-            if ( isNaNExtF80UI( uiA64, uiA0 ) ) goto returnA;
-            goto returnB;
     }
-    }
- returnLargerMag:
-    uiMagA64 = uiA64 & 0x7FFF;
-    uiMagB64 = uiB64 & 0x7FFF;
-    if ( uiMagA64 < uiMagB64 ) goto returnB;
-    if ( uiMagB64 < uiMagA64 ) goto returnA;
-    if ( uiA0 < uiB0 ) goto returnB;
-    if ( uiB0 < uiA0 ) goto returnA;
-    if ( uiA64 < uiB64 ) goto returnA;
- returnB:
-    uiZ.v64 = uiB64;
-    uiZ.v0  = uiNonsigB0;
-    return uiZ;
- returnA:
-    uiZ.v64 = uiA64;
-    uiZ.v0  = uiNonsigA0;
+    uiZ.v64 = defaultNaNExtF80UI64;
+    uiZ.v0  = defaultNaNExtF80UI0;
     return uiZ;
 
 }

softfloat/source/RISCV/s_propagateNaNF128M.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015, 2018 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,43 +34,35 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
+#include "primitiveTypes.h"
 #include "specialize.h"
 #include "softfloat.h"
 
 /*----------------------------------------------------------------------------
 | Assuming at least one of the two 128-bit floating-point values pointed to by
-| `aWPtr' and `bWPtr' is a NaN, stores the combined NaN result at the location
-| pointed to by `zWPtr'.  If either original floating-point value is a
-| signaling NaN, the invalid exception is raised.  Each of `aWPtr', `bWPtr',
-| and `zWPtr' points to an array of four 32-bit elements that concatenate in
+| 'aWPtr' and 'bWPtr' is a NaN, stores the combined NaN result at the location
+| pointed to by 'zWPtr'.  If either original floating-point value is a
+| signaling NaN, the invalid exception is raised.  Each of 'aWPtr', 'bWPtr',
+| and 'zWPtr' points to an array of four 32-bit elements that concatenate in
 | the platform's normal endian order to form a 128-bit floating-point value.
 *----------------------------------------------------------------------------*/
 void
  softfloat_propagateNaNF128M(
      const uint32_t *aWPtr, const uint32_t *bWPtr, uint32_t *zWPtr )
 {
-    bool isSigNaNA;
-    const uint32_t *ptr;
 
-    ptr = aWPtr;
-    isSigNaNA = f128M_isSignalingNaN( (const float128_t *) aWPtr );
     if (
-        isSigNaNA
+        f128M_isSignalingNaN( (const float128_t *) aWPtr )
             || (bWPtr && f128M_isSignalingNaN( (const float128_t *) bWPtr ))
     ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) goto copy;
     }
-    if ( ! softfloat_isNaNF128M( aWPtr ) ) ptr = bWPtr;
- copy:
-    zWPtr[indexWordHi( 4 )] = ptr[indexWordHi( 4 )] | 0x00008000;
-    zWPtr[indexWord( 4, 2 )] = ptr[indexWord( 4, 2 )];
-    zWPtr[indexWord( 4, 1 )] = ptr[indexWord( 4, 1 )];
-    zWPtr[indexWord( 4, 0 )] = ptr[indexWord( 4, 0 )];
+    zWPtr[indexWord( 4, 3 )] = defaultNaNF128UI96;
+    zWPtr[indexWord( 4, 2 )] = defaultNaNF128UI64;
+    zWPtr[indexWord( 4, 1 )] = defaultNaNF128UI32;
+    zWPtr[indexWord( 4, 0 )] = defaultNaNF128UI0;
 
 }
 

softfloat/source/RISCV/s_propagateNaNF128UI.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,10 +34,9 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
+#include "primitiveTypes.h"
 #include "specialize.h"
 #include "softfloat.h"
 
@@ -58,23 +57,16 @@ struct uint128
      uint_fast64_t uiB0
  )
 {
-    bool isSigNaNA;
     struct uint128 uiZ;
 
-    isSigNaNA = softfloat_isSigNaNF128UI( uiA64, uiA0 );
-    if ( isSigNaNA || softfloat_isSigNaNF128UI( uiB64, uiB0 ) ) {
+    if (
+           softfloat_isSigNaNF128UI( uiA64, uiA0 )
+        || softfloat_isSigNaNF128UI( uiB64, uiB0 )
+    ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) goto returnNonsigA;
     }
-    if ( isNaNF128UI( uiA64, uiA0 ) ) {
- returnNonsigA:
-        uiZ.v64 = uiA64;
-        uiZ.v0  = uiA0;
-    } else {
-        uiZ.v64 = uiB64;
-        uiZ.v0  = uiB0;
-    }
-    uiZ.v64 |= UINT64_C( 0x0000800000000000 );
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0  = defaultNaNF128UI0;
     return uiZ;
 
 }

softfloat/source/RISCV/s_propagateNaNF16UI.c

@@ -4,7 +4,7 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+Copyright 2011, 2012, 2013, 2014, 2015, 2016 The Regents of the University of
 California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -34,10 +34,8 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
 #include "specialize.h"
 #include "softfloat.h"
 
@@ -50,14 +48,11 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 uint_fast16_t
  softfloat_propagateNaNF16UI( uint_fast16_t uiA, uint_fast16_t uiB )
 {
-    bool isSigNaNA;
 
-    isSigNaNA = softfloat_isSigNaNF16UI( uiA );
-    if ( isSigNaNA || softfloat_isSigNaNF16UI( uiB ) ) {
+    if ( softfloat_isSigNaNF16UI( uiA ) || softfloat_isSigNaNF16UI( uiB ) ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) return uiA | 0x0200;
     }
-    return (isNaNF16UI( uiA ) ? uiA : uiB) | 0x0200;
+    return defaultNaNF16UI;
 
 }
 

softfloat/source/RISCV/s_propagateNaNF32UI.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,10 +34,8 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
 #include "specialize.h"
 #include "softfloat.h"
 
@@ -50,14 +48,11 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 uint_fast32_t
  softfloat_propagateNaNF32UI( uint_fast32_t uiA, uint_fast32_t uiB )
 {
-    bool isSigNaNA;
 
-    isSigNaNA = softfloat_isSigNaNF32UI( uiA );
-    if ( isSigNaNA || softfloat_isSigNaNF32UI( uiB ) ) {
+    if ( softfloat_isSigNaNF32UI( uiA ) || softfloat_isSigNaNF32UI( uiB ) ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) return uiA | 0x00400000;
     }
-    return (isNaNF32UI( uiA ) ? uiA : uiB) | 0x00400000;
+    return defaultNaNF32UI;
 
 }
 

softfloat/source/RISCV/s_propagateNaNF64UI.c

@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.
 
-Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
-All rights reserved.
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,10 +34,8 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 =============================================================================*/
 
-#include <stdbool.h>
 #include <stdint.h>
 #include "platform.h"
-#include "internals.h"
 #include "specialize.h"
 #include "softfloat.h"
 
@@ -50,14 +48,11 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 uint_fast64_t
  softfloat_propagateNaNF64UI( uint_fast64_t uiA, uint_fast64_t uiB )
 {
-    bool isSigNaNA;
 
-    isSigNaNA = softfloat_isSigNaNF64UI( uiA );
-    if ( isSigNaNA || softfloat_isSigNaNF64UI( uiB ) ) {
+    if ( softfloat_isSigNaNF64UI( uiA ) || softfloat_isSigNaNF64UI( uiB ) ) {
         softfloat_raiseFlags( softfloat_flag_invalid );
-        if ( isSigNaNA ) return uiA | UINT64_C( 0x0008000000000000 );
     }
-    return (isNaNF64UI( uiA ) ? uiA : uiB) | UINT64_C( 0x0008000000000000 );
+    return defaultNaNF64UI;
 
 }
 

softfloat/source/RISCV/specialize.h

@@ -51,19 +51,19 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 | The values to return on conversions to 32-bit integer formats that raise an
 | invalid exception.
 *----------------------------------------------------------------------------*/
-#define ui32_fromPosOverflow UINT32_C(0xFFFFFFFF)
-#define ui32_fromNegOverflow UINT32_C(0x0)
-#define ui32_fromNaN UINT32_C(0xFFFFFFFF)
-#define i32_fromPosOverflow INT64_C(0x7FFFFFFF)
-#define i32_fromNegOverflow (-INT64_C(0x7FFFFFFF) - 1)
-#define i32_fromNaN INT64_C(0x7FFFFFFF)
+#define ui32_fromPosOverflow 0xFFFFFFFF
+#define ui32_fromNegOverflow 0
+#define ui32_fromNaN 0xFFFFFFFF
+#define i32_fromPosOverflow 0x7FFFFFFF
+#define i32_fromNegOverflow (-0x7FFFFFFF - 1)
+#define i32_fromNaN 0x7FFFFFFF
 
 /*----------------------------------------------------------------------------
 | The values to return on conversions to 64-bit integer formats that raise an
 | invalid exception.
 *----------------------------------------------------------------------------*/
 #define ui64_fromPosOverflow UINT64_C(0xFFFFFFFFFFFFFFFF)
-#define ui64_fromNegOverflow UINT64_C(0x0)
+#define ui64_fromNegOverflow 0
 #define ui64_fromNaN UINT64_C(0xFFFFFFFFFFFFFFFF)
 #define i64_fromPosOverflow INT64_C(0x7FFFFFFFFFFFFFFF)
 #define i64_fromNegOverflow (-INT64_C(0x7FFFFFFFFFFFFFFF) - 1)
@@ -74,18 +74,13 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 | to another.
 *----------------------------------------------------------------------------*/
 struct commonNaN {
-    bool sign;
-#ifdef LITTLEENDIAN
-    uint64_t v0, v64;
-#else
-    uint64_t v64, v0;
-#endif
+    char _unused;
 };
 
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 16-bit floating-point NaN.
 *----------------------------------------------------------------------------*/
-#define defaultNaNF16UI 0xFE00
+#define defaultNaNF16UI 0x7E00
 
 /*----------------------------------------------------------------------------
 | Returns true when 16-bit unsigned integer 'uiA' has the bit pattern of a
@@ -94,19 +89,38 @@ struct commonNaN {
 *----------------------------------------------------------------------------*/
 #define softfloat_isSigNaNF16UI(uiA) ((((uiA)&0x7E00) == 0x7C00) && ((uiA)&0x01FF))
 
+/*----------------------------------------------------------------------------
+| Returns true when 16-bit unsigned integer 'uiA' has the bit pattern of a
+| 16-bit brain floating-point (BF16) signaling NaN.
+| Note:  This macro evaluates its argument more than once.
+*----------------------------------------------------------------------------*/
+#define softfloat_isSigNaNBF16UI(uiA) ((((uiA)&0x7FC0) == 0x7F80) && ((uiA)&0x003F))
+
 /*----------------------------------------------------------------------------
 | Assuming 'uiA' has the bit pattern of a 16-bit floating-point NaN, converts
 | this NaN to the common NaN form, and stores the resulting common NaN at the
 | location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
 | exception is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_f16UIToCommonNaN(uint_fast16_t uiA, struct commonNaN* zPtr);
+#define softfloat_f16UIToCommonNaN(uiA, zPtr)                                                                                              \
+    if(!((uiA)&0x0200))                                                                                                                    \
+    softfloat_raiseFlags(softfloat_flag_invalid)
+
+/*----------------------------------------------------------------------------
+| Assuming 'uiA' has the bit pattern of a 16-bit BF16 floating-point NaN, converts
+| this NaN to the common NaN form, and stores the resulting common NaN at the
+| location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
+| exception is raised.
+*----------------------------------------------------------------------------*/
+#define softfloat_bf16UIToCommonNaN(uiA, zPtr)                                                                                             \
+    if(!((uiA)&0x0040))                                                                                                                    \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into a 16-bit floating-point
 | NaN, and returns the bit pattern of this value as an unsigned integer.
 *----------------------------------------------------------------------------*/
-uint_fast16_t softfloat_commonNaNToF16UI(const struct commonNaN* aPtr);
+#define softfloat_commonNaNToF16UI(aPtr) ((uint_fast16_t)defaultNaNF16UI)
 
 /*----------------------------------------------------------------------------
 | Interpreting 'uiA' and 'uiB' as the bit patterns of two 16-bit floating-
@@ -116,6 +130,17 @@ uint_fast16_t softfloat_commonNaNToF16UI(const struct commonNaN* aPtr);
 *----------------------------------------------------------------------------*/
 uint_fast16_t softfloat_propagateNaNF16UI(uint_fast16_t uiA, uint_fast16_t uiB);
 
+/*----------------------------------------------------------------------------
+| The bit pattern for a default generated 16-bit BF16 floating-point NaN.
+*----------------------------------------------------------------------------*/
+#define defaultNaNBF16UI 0x7FC0
+
+/*----------------------------------------------------------------------------
+| Converts the common NaN pointed to by 'aPtr' into a 16-bit floating-point
+| NaN, and returns the bit pattern of this value as an unsigned integer.
+*----------------------------------------------------------------------------*/
+#define softfloat_commonNaNToBF16UI(aPtr) ((uint_fast16_t)defaultNaNBF16UI)
+
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 32-bit floating-point NaN.
 *----------------------------------------------------------------------------*/
@@ -134,13 +159,15 @@ uint_fast16_t softfloat_propagateNaNF16UI(uint_fast16_t uiA, uint_fast16_t uiB);
 | location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
 | exception is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_f32UIToCommonNaN(uint_fast32_t uiA, struct commonNaN* zPtr);
+#define softfloat_f32UIToCommonNaN(uiA, zPtr)                                                                                              \
+    if(!((uiA)&0x00400000))                                                                                                                \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into a 32-bit floating-point
 | NaN, and returns the bit pattern of this value as an unsigned integer.
 *----------------------------------------------------------------------------*/
-uint_fast32_t softfloat_commonNaNToF32UI(const struct commonNaN* aPtr);
+#define softfloat_commonNaNToF32UI(aPtr) ((uint_fast32_t)defaultNaNF32UI)
 
 /*----------------------------------------------------------------------------
 | Interpreting 'uiA' and 'uiB' as the bit patterns of two 32-bit floating-
@@ -169,13 +196,15 @@ uint_fast32_t softfloat_propagateNaNF32UI(uint_fast32_t uiA, uint_fast32_t uiB);
 | location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
 | exception is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_f64UIToCommonNaN(uint_fast64_t uiA, struct commonNaN* zPtr);
+#define softfloat_f64UIToCommonNaN(uiA, zPtr)                                                                                              \
+    if(!((uiA)&UINT64_C(0x0008000000000000)))                                                                                              \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into a 64-bit floating-point
 | NaN, and returns the bit pattern of this value as an unsigned integer.
 *----------------------------------------------------------------------------*/
-uint_fast64_t softfloat_commonNaNToF64UI(const struct commonNaN* aPtr);
+#define softfloat_commonNaNToF64UI(aPtr) ((uint_fast64_t)defaultNaNF64UI)
 
 /*----------------------------------------------------------------------------
 | Interpreting 'uiA' and 'uiB' as the bit patterns of two 64-bit floating-
@@ -188,7 +217,7 @@ uint_fast64_t softfloat_propagateNaNF64UI(uint_fast64_t uiA, uint_fast64_t uiB);
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 80-bit extended floating-point NaN.
 *----------------------------------------------------------------------------*/
-#define defaultNaNExtF80UI64 0xFFFF
+#define defaultNaNExtF80UI64 0x7FFF
 #define defaultNaNExtF80UI0 UINT64_C(0xC000000000000000)
 
 /*----------------------------------------------------------------------------
@@ -214,14 +243,26 @@ uint_fast64_t softfloat_propagateNaNF64UI(uint_fast64_t uiA, uint_fast64_t uiB);
 | location pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid
 | exception is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_extF80UIToCommonNaN(uint_fast16_t uiA64, uint_fast64_t uiA0, struct commonNaN* zPtr);
+#define softfloat_extF80UIToCommonNaN(uiA64, uiA0, zPtr)                                                                                   \
+    if(!((uiA0)&UINT64_C(0x4000000000000000)))                                                                                             \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into an 80-bit extended
 | floating-point NaN, and returns the bit pattern of this value as an unsigned
 | integer.
 *----------------------------------------------------------------------------*/
+#if defined INLINE && !defined softfloat_commonNaNToExtF80UI
+INLINE
+struct uint128 softfloat_commonNaNToExtF80UI(const struct commonNaN* aPtr) {
+    struct uint128 uiZ;
+    uiZ.v64 = defaultNaNExtF80UI64;
+    uiZ.v0 = defaultNaNExtF80UI0;
+    return uiZ;
+}
+#else
 struct uint128 softfloat_commonNaNToExtF80UI(const struct commonNaN* aPtr);
+#endif
 
 /*----------------------------------------------------------------------------
 | Interpreting the unsigned integer formed from concatenating 'uiA64' and
@@ -237,7 +278,7 @@ struct uint128 softfloat_propagateNaNExtF80UI(uint_fast16_t uiA64, uint_fast64_t
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 128-bit floating-point NaN.
 *----------------------------------------------------------------------------*/
-#define defaultNaNF128UI64 UINT64_C(0xFFFF800000000000)
+#define defaultNaNF128UI64 UINT64_C(0x7FFF800000000000)
 #define defaultNaNF128UI0 UINT64_C(0)
 
 /*----------------------------------------------------------------------------
@@ -256,13 +297,25 @@ struct uint128 softfloat_propagateNaNExtF80UI(uint_fast16_t uiA64, uint_fast64_t
 | pointed to by 'zPtr'.  If the NaN is a signaling NaN, the invalid exception
 | is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_f128UIToCommonNaN(uint_fast64_t uiA64, uint_fast64_t uiA0, struct commonNaN* zPtr);
+#define softfloat_f128UIToCommonNaN(uiA64, uiA0, zPtr)                                                                                     \
+    if(!((uiA64)&UINT64_C(0x0000800000000000)))                                                                                            \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into a 128-bit floating-point
 | NaN, and returns the bit pattern of this value as an unsigned integer.
 *----------------------------------------------------------------------------*/
+#if defined INLINE && !defined softfloat_commonNaNToF128UI
+INLINE
+struct uint128 softfloat_commonNaNToF128UI(const struct commonNaN* aPtr) {
+    struct uint128 uiZ;
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0 = defaultNaNF128UI0;
+    return uiZ;
+}
+#else
 struct uint128 softfloat_commonNaNToF128UI(const struct commonNaN*);
+#endif
 
 /*----------------------------------------------------------------------------
 | Interpreting the unsigned integer formed from concatenating 'uiA64' and
@@ -288,14 +341,24 @@ struct uint128 softfloat_propagateNaNF128UI(uint_fast64_t uiA64, uint_fast64_t u
 | common NaN at the location pointed to by 'zPtr'.  If the NaN is a signaling
 | NaN, the invalid exception is raised.
 *----------------------------------------------------------------------------*/
-void softfloat_extF80MToCommonNaN(const struct extFloat80M* aSPtr, struct commonNaN* zPtr);
+#define softfloat_extF80MToCommonNaN(aSPtr, zPtr)                                                                                          \
+    if(!((aSPtr)->signif & UINT64_C(0x4000000000000000)))                                                                                  \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into an 80-bit extended
 | floating-point NaN, and stores this NaN at the location pointed to by
 | 'zSPtr'.
 *----------------------------------------------------------------------------*/
+#if defined INLINE && !defined softfloat_commonNaNToExtF80M
+INLINE
+void softfloat_commonNaNToExtF80M(const struct commonNaN* aPtr, struct extFloat80M* zSPtr) {
+    zSPtr->signExp = defaultNaNExtF80UI64;
+    zSPtr->signif = defaultNaNExtF80UI0;
+}
+#else
 void softfloat_commonNaNToExtF80M(const struct commonNaN* aPtr, struct extFloat80M* zSPtr);
+#endif
 
 /*----------------------------------------------------------------------------
 | Assuming at least one of the two 80-bit extended floating-point values
@@ -308,7 +371,7 @@ void softfloat_propagateNaNExtF80M(const struct extFloat80M* aSPtr, const struct
 /*----------------------------------------------------------------------------
 | The bit pattern for a default generated 128-bit floating-point NaN.
 *----------------------------------------------------------------------------*/
-#define defaultNaNF128UI96 0xFFFF8000
+#define defaultNaNF128UI96 0x7FFF8000
 #define defaultNaNF128UI64 0
 #define defaultNaNF128UI32 0
 #define defaultNaNF128UI0 0
@@ -321,7 +384,9 @@ void softfloat_propagateNaNExtF80M(const struct extFloat80M* aSPtr, const struct
 | four 32-bit elements that concatenate in the platform's normal endian order
 | to form a 128-bit floating-point value.
 *----------------------------------------------------------------------------*/
-void softfloat_f128MToCommonNaN(const uint32_t* aWPtr, struct commonNaN* zPtr);
+#define softfloat_f128MToCommonNaN(aWPtr, zPtr)                                                                                            \
+    if(!((aWPtr)[indexWordHi(4)] & UINT64_C(0x0000800000000000)))                                                                          \
+    softfloat_raiseFlags(softfloat_flag_invalid)
 
 /*----------------------------------------------------------------------------
 | Converts the common NaN pointed to by 'aPtr' into a 128-bit floating-point
@@ -329,7 +394,17 @@ void softfloat_f128MToCommonNaN(const uint32_t* aWPtr, struct commonNaN* zPtr);
 | 'zWPtr' points to an array of four 32-bit elements that concatenate in the
 | platform's normal endian order to form a 128-bit floating-point value.
 *----------------------------------------------------------------------------*/
+#if defined INLINE && !defined softfloat_commonNaNToF128M
+INLINE
+void softfloat_commonNaNToF128M(const struct commonNaN* aPtr, uint32_t* zWPtr) {
+    zWPtr[indexWord(4, 3)] = defaultNaNF128UI96;
+    zWPtr[indexWord(4, 2)] = defaultNaNF128UI64;
+    zWPtr[indexWord(4, 1)] = defaultNaNF128UI32;
+    zWPtr[indexWord(4, 0)] = defaultNaNF128UI0;
+}
+#else
 void softfloat_commonNaNToF128M(const struct commonNaN* aPtr, uint32_t* zWPtr);
+#endif
 
 /*----------------------------------------------------------------------------
 | Assuming at least one of the two 128-bit floating-point values pointed to by

softfloat/source/bf16_isSignalingNaN.c

@@ -0,0 +1,51 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+bool bf16_isSignalingNaN( bfloat16_t a )
+{
+    union ui16_bf16 uA;
+
+    uA.f = a;
+    return softfloat_isSigNaNBF16UI( uA.ui );
+
+}
+

softfloat/source/bf16_to_f32.c

@@ -0,0 +1,90 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float32_t bf16_to_f32( bfloat16_t a )
+{
+    union ui16_bf16 uA;
+    uint_fast16_t uiA;
+    bool sign;
+    int_fast16_t exp;
+    uint_fast16_t frac;
+    struct commonNaN commonNaN;
+    uint_fast32_t uiZ;
+    struct exp8_sig16 normExpSig;
+    union ui32_f32 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA = uA.ui;
+    sign = signBF16UI( uiA );
+    exp  = expBF16UI( uiA );
+    frac = fracBF16UI( uiA );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    // NaN or Inf
+    if ( exp == 0xFF ) {
+        if ( frac ) {
+            softfloat_bf16UIToCommonNaN( uiA, &commonNaN );
+            uiZ = softfloat_commonNaNToF32UI( &commonNaN );
+        } else {
+            uiZ = packToF32UI( sign, 0xFF, 0 );
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    // packToF32UI simply packs bitfields without any numerical change
+    // which means it can be used directly for any BF16 to f32 conversions which
+    // does not require bits manipulation
+    // (that is everything where the 16-bit are just padded right with 16 zeros, including
+    //  subnormal numbers)
+    uiZ = packToF32UI( sign, exp, ((uint_fast32_t) frac) <<16 );
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
+
+

softfloat/source/f32_to_bf16.c

@@ -0,0 +1,105 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+#include <inttypes.h>
+#include <stdio.h>
+
+bfloat16_t f32_to_bf16( float32_t a )
+{
+    union ui32_f32 uA;
+    uint_fast32_t uiA;
+    bool sign;
+    int_fast16_t exp;
+    uint_fast32_t frac;
+    struct commonNaN commonNaN;
+    uint_fast16_t uiZ, frac16;
+    union ui16_bf16 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA = uA.ui;
+    sign = signF32UI( uiA );
+    exp  = expF32UI( uiA );
+    frac = fracF32UI( uiA );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    // infinity or NaN cases
+    if ( exp == 0xFF ) {
+        if ( frac ) {
+            // NaN case
+            softfloat_f32UIToCommonNaN( uiA, &commonNaN );
+            uiZ = softfloat_commonNaNToBF16UI( &commonNaN );
+        } else {
+            // infinity case
+            uiZ = packToBF16UI( sign, 0xFF, 0 );
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    // frac is a 24-bit mantissa, right shifted by 9
+    // In the normal case, (24-9) = 15 are set 
+    frac16 = frac>>9 | ((frac & 0x1FF) != 0);
+    if ( ! (exp | frac16) ) {
+        uiZ = packToBF16UI( sign, 0, 0 );
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    // softfloat_roundPackToBF16 exponent argument (2nd argument)
+    // must correspond to the exponent of fracIn[13] bits
+    // (fracIn is the 3rd and last argument) 
+    uint_fast32_t mask = exp ? 0x4000 : 0x0; // implicit one mask added if input is a normal number
+    // exponent for the lowest normal and largest subnormal should be equal
+    // but is not in IEEE encoding so mantissa must be partially normalized
+    // (by one bit) for subnormal numbers. Such that (exp - 1) corresponds
+    // to the exponent of frac16[13]
+    frac16 = frac16 << (exp ? 0 : 1);
+    return softfloat_roundPackToBF16( sign, exp - 1, frac16 | mask );
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+

softfloat/source/f32_to_f16.c

@@ -72,6 +72,9 @@ float16_t f32_to_f16( float32_t a )
     }
     /*------------------------------------------------------------------------
     *------------------------------------------------------------------------*/
+    // frac is a 24-bit significand, the bottom 9 bits LSB are extracted and OR-red
+    // into a sticky flag, the top 15 MSBs are extracted, the LSB of this top slice
+    // is OR-red with the sticky 
     frac16 = frac>>9 | ((frac & 0x1FF) != 0);
     if ( ! (exp | frac16) ) {
         uiZ = packToF16UI( sign, 0, 0 );

softfloat/source/include/internals.h

@@ -46,6 +46,10 @@ union ui16_f16 {
     uint16_t ui;
     float16_t f;
 };
+union ui16_bf16 {
+    uint16_t ui;
+    bfloat16_t f;
+};
 union ui32_f32 {
     uint32_t ui;
     float32_t f;
@@ -108,6 +112,18 @@ float16_t softfloat_addMagsF16(uint_fast16_t, uint_fast16_t);
 float16_t softfloat_subMagsF16(uint_fast16_t, uint_fast16_t);
 float16_t softfloat_mulAddF16(uint_fast16_t, uint_fast16_t, uint_fast16_t, uint_fast8_t);
 
+/*----------------------------------------------------------------------------
+ *----------------------------------------------------------------------------*/
+#define signBF16UI(a) ((bool)((uint16_t)(a) >> 15))
+#define expBF16UI(a) ((int_fast16_t)((a) >> 7) & 0xFF)
+#define fracBF16UI(a) ((a)&0x07F)
+#define packToBF16UI(sign, exp, sig) (((uint16_t)(sign) << 15) + ((uint16_t)(exp) << 7) + (sig))
+
+#define isNaNBF16UI(a) (((~(a)&0x7FC0) == 0) && ((a)&0x07F))
+
+bfloat16_t softfloat_roundPackToBF16(bool, int_fast16_t, uint_fast16_t);
+struct exp8_sig16 softfloat_normSubnormalBF16Sig(uint_fast16_t);
+
 /*----------------------------------------------------------------------------
  *----------------------------------------------------------------------------*/
 #define signF32UI(a) ((bool)((uint32_t)(a) >> 31))

softfloat/source/include/softfloat.h

@@ -76,13 +76,13 @@ enum {
 | Software floating-point exception flags.
 *----------------------------------------------------------------------------*/
 extern THREAD_LOCAL uint_fast8_t softfloat_exceptionFlags;
-enum {
+typedef enum {
     softfloat_flag_inexact = 1,
     softfloat_flag_underflow = 2,
     softfloat_flag_overflow = 4,
     softfloat_flag_infinite = 8,
     softfloat_flag_invalid = 16
-};
+} exceptionFlag_t;
 
 /*----------------------------------------------------------------------------
 | Routine to raise any or all of the software floating-point exception flags.
@@ -164,6 +164,13 @@ bool f16_le_quiet(float16_t, float16_t);
 bool f16_lt_quiet(float16_t, float16_t);
 bool f16_isSignalingNaN(float16_t);
 
+/*----------------------------------------------------------------------------
+| 16-bit (brain float 16) floating-point operations.
+*----------------------------------------------------------------------------*/
+float32_t bf16_to_f32(bfloat16_t);
+bfloat16_t f32_to_bf16(float32_t);
+bool bf16_isSignalingNaN(bfloat16_t);
+
 /*----------------------------------------------------------------------------
 | 32-bit (single-precision) floating-point operations.
 *----------------------------------------------------------------------------*/

softfloat/source/include/softfloat_types.h

@@ -50,6 +50,9 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 typedef struct {
     uint16_t v;
 } float16_t;
+typedef struct {
+    uint16_t v;
+} bfloat16_t;
 typedef struct {
     uint32_t v;
 } float32_t;

softfloat/source/s_mulAddF32.c

@@ -221,4 +221,3 @@ float32_t
     return uZ.f;
 
 }
-

softfloat/source/s_normSubnormalBF16Sig.c

@@ -0,0 +1,52 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+
+struct exp8_sig16 softfloat_normSubnormalBF16Sig( uint_fast16_t sig )
+{
+    int_fast8_t shiftDist;
+    struct exp8_sig16 z;
+
+    shiftDist = softfloat_countLeadingZeros16( sig ) - 8;
+    z.exp = 1 - shiftDist;
+    z.sig = sig<<shiftDist;
+    return z;
+
+}
+

softfloat/source/s_roundPackToBF16.c

@@ -0,0 +1,114 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2017 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+/** sig last significant bit is sig[7], the 7 LSBs will be used for rounding */
+bfloat16_t
+ softfloat_roundPackToBF16( bool sign, int_fast16_t exp, uint_fast16_t sig )
+{
+    uint_fast8_t roundingMode;
+    bool roundNearEven;
+    uint_fast8_t roundIncrement, roundBits;
+    bool isTiny;
+    uint_fast16_t uiZ;
+    union ui16_bf16 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    roundingMode = softfloat_roundingMode;
+    roundNearEven = (roundingMode == softfloat_round_near_even);
+    roundIncrement = 0x40;
+    if ( ! roundNearEven && (roundingMode != softfloat_round_near_maxMag) ) {
+        roundIncrement =
+            (roundingMode
+                 == (sign ? softfloat_round_min : softfloat_round_max))
+                ? 0x7F
+                : 0;
+    }
+    roundBits = sig & 0x7F;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( 0xFD <= (unsigned int) exp ) {
+        if ( exp < 0 ) {
+            /*----------------------------------------------------------------
+            *----------------------------------------------------------------*/
+            isTiny =
+                (softfloat_detectTininess == softfloat_tininess_beforeRounding)
+                    || (exp < -1) || (sig + roundIncrement < 0x8000);
+            sig = softfloat_shiftRightJam32( sig, -exp );
+            exp = 0;
+            roundBits = sig & 0x7F;
+            if ( isTiny && roundBits ) {
+                softfloat_raiseFlags( softfloat_flag_underflow );
+            }
+        } else if ( (0xFD < exp) || (0x8000 <= sig + roundIncrement) ) {
+            /*----------------------------------------------------------------
+            *----------------------------------------------------------------*/
+            softfloat_raiseFlags(
+                softfloat_flag_overflow | softfloat_flag_inexact );
+            uiZ = packToBF16UI( sign, 0xFF, 0 ) - ! roundIncrement;
+            goto uiZ;
+        }
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    sig = (sig + roundIncrement)>>7;
+    if ( roundBits ) {
+        softfloat_exceptionFlags |= softfloat_flag_inexact;
+#ifdef SOFTFLOAT_ROUND_ODD
+        if ( roundingMode == softfloat_round_odd ) {
+            sig |= 1;
+            goto packReturn;
+        }
+#endif
+    }
+    sig &= ~(uint_fast16_t) (! (roundBits ^ 0x40) & roundNearEven);
+    if ( ! sig ) exp = 0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ packReturn:
+    uiZ = packToBF16UI( sign, exp, sig );
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+

src/elfio.cpp

@@ -0,0 +1,35 @@
+#ifdef _MSC_VER
+#define _SCL_SECURE_NO_WARNINGS
+#define ELFIO_NO_INTTYPES
+#endif
+
+#include <elfio/elfio_dump.hpp>
+#include <iostream>
+
+using namespace ELFIO;
+
+int main(int argc, char** argv) {
+    if(argc != 2) {
+        printf("Usage: elfdump <file_name>\n");
+        return 1;
+    }
+
+    elfio reader;
+
+    if(!reader.load(argv[1])) {
+        printf("File %s is not found or it is not an ELF file\n", argv[1]);
+        return 1;
+    }
+
+    dump::header(std::cout, reader);
+    dump::section_headers(std::cout, reader);
+    dump::segment_headers(std::cout, reader);
+    dump::symbol_tables(std::cout, reader);
+    dump::notes(std::cout, reader);
+    dump::modinfo(std::cout, reader);
+    dump::dynamic_tags(std::cout, reader);
+    dump::section_datas(std::cout, reader);
+    dump::segment_datas(std::cout, reader);
+
+    return 0;
+}

src/iss/arch/hwl.h

@@ -51,8 +51,8 @@ public:
     virtual ~hwl() = default;
 
 protected:
-    iss::status read_custom_csr_reg(unsigned addr, reg_t& val) override;
-    iss::status write_custom_csr_reg(unsigned addr, reg_t val) override;
+    iss::status read_custom_csr(unsigned addr, reg_t& val) override;
+    iss::status write_custom_csr(unsigned addr, reg_t val) override;
 };
 
 template <typename BASE>
@@ -68,7 +68,7 @@ inline hwl<BASE>::hwl(feature_config cfg)
     }
 }
 
-template <typename BASE> inline iss::status iss::arch::hwl<BASE>::read_custom_csr_reg(unsigned addr, reg_t& val) {
+template <typename BASE> inline iss::status iss::arch::hwl<BASE>::read_custom_csr(unsigned addr, reg_t& val) {
     switch(addr) {
     case 0x800:
         val = this->reg.lpstart0;
@@ -92,7 +92,7 @@ template <typename BASE> inline iss::status iss::arch::hwl<BASE>::read_custom_cs
     return iss::Ok;
 }
 
-template <typename BASE> inline iss::status iss::arch::hwl<BASE>::write_custom_csr_reg(unsigned addr, reg_t val) {
+template <typename BASE> inline iss::status iss::arch::hwl<BASE>::write_custom_csr(unsigned addr, reg_t val) {
     switch(addr) {
     case 0x800:
         this->reg.lpstart0 = val;

src/iss/arch/riscv_hart_common.h

@@ -35,6 +35,7 @@
 #ifndef _RISCV_HART_COMMON
 #define _RISCV_HART_COMMON
 
+#include "iss/vm_types.h"
 #include <cstdint>
 #include <elfio/elfio.hpp>
 #include <fmt/format.h>
@@ -44,6 +45,14 @@
 #include <unordered_map>
 #include <util/logging.h>
 
+#if defined(__GNUC__)
+#define likely(x) ::__builtin_expect(!!(x), 1)
+#define unlikely(x) ::__builtin_expect(!!(x), 0)
+#else
+#define likely(x) x
+#define unlikely(x) x
+#endif
+
 namespace iss {
 namespace arch {
 
@@ -306,31 +315,36 @@ struct riscv_hart_common {
     riscv_hart_common(){};
     ~riscv_hart_common(){};
     std::unordered_map<std::string, uint64_t> symbol_table;
+    uint64_t entry_address{0};
+    uint64_t tohost = tohost_dflt;
+    uint64_t fromhost = fromhost_dflt;
 
-    std::unordered_map<std::string, uint64_t> get_sym_table(std::string name) {
-        if(!symbol_table.empty())
-            return symbol_table;
-        FILE* fp = fopen(name.c_str(), "r");
-        if(fp) {
-            std::array<char, 5> buf;
-            auto n = fread(buf.data(), 1, 4, fp);
-            fclose(fp);
-            if(n != 4)
-                throw std::runtime_error("input file has insufficient size");
-            buf[4] = 0;
-            if(strcmp(buf.data() + 1, "ELF") == 0) {
+    bool read_elf_file(std::string name, uint8_t expected_elf_class,
+                       std::function<iss::status(uint64_t, uint64_t, const uint8_t* const)> cb) {
         // Create elfio reader
         ELFIO::elfio reader;
         // Load ELF data
-                if(!reader.load(name))
-                    throw std::runtime_error("could not process elf file");
+        if(reader.load(name)) {
             // check elf properties
-                if(reader.get_type() != ET_EXEC)
-                    throw std::runtime_error("wrong elf type in file");
-                if(reader.get_machine() != EM_RISCV)
-                    throw std::runtime_error("wrong elf machine in file");
+            if(reader.get_class() != expected_elf_class)
+                return false;
+            if(reader.get_type() != ELFIO::ET_EXEC)
+                return false;
+            if(reader.get_machine() != ELFIO::EM_RISCV)
+                return false;
+            entry_address = reader.get_entry();
+            for(const auto& pseg : reader.segments) {
+                const auto fsize = pseg->get_file_size(); // 0x42c/0x0
+                const auto seg_data = pseg->get_data();
+                const auto type = pseg->get_type();
+                if(type == 1 && fsize > 0) {
+                    auto res = cb(pseg->get_physical_address(), fsize, reinterpret_cast<const uint8_t* const>(seg_data));
+                    if(res != iss::Ok)
+                        CPPLOG(ERR) << "problem writing " << fsize << "bytes to 0x" << std::hex << pseg->get_physical_address();
+                }
+            }
             const auto sym_sec = reader.sections[".symtab"];
-                if(SHT_SYMTAB == sym_sec->get_type() || SHT_DYNSYM == sym_sec->get_type()) {
+            if(ELFIO::SHT_SYMTAB == sym_sec->get_type() || ELFIO::SHT_DYNSYM == sym_sec->get_type()) {
                 ELFIO::symbol_section_accessor symbols(reader, sym_sec);
                 auto sym_no = symbols.get_symbols_num();
                 std::string name;
@@ -349,12 +363,19 @@ struct riscv_hart_common {
 #endif
                     }
                 }
+                try {
+                    tohost = symbol_table.at("tohost");
+                    try {
+                        fromhost = symbol_table.at("fromhost");
+                    } catch(std::out_of_range& e) {
+                        fromhost = tohost + 0x40;
                     }
-                return symbol_table;
+                } catch(std::out_of_range& e) {
                 }
-            throw std::runtime_error(fmt::format("memory load file {} is not a valid elf file", name));
-        } else
-            throw std::runtime_error(fmt::format("memory load file not found, check if {} is a valid file", name));
+            }
+            return true;
+        }
+        return false;
     };
 };
 

src/iss/arch/riscv_hart_m_p.h

@@ -39,7 +39,9 @@
 #include "iss/instrumentation_if.h"
 #include "iss/log_categories.h"
 #include "iss/vm_if.h"
+#include "iss/vm_types.h"
 #include "riscv_hart_common.h"
+#include <elfio/elf_types.hpp>
 #include <stdexcept>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
@@ -58,18 +60,11 @@
 
 #include <iss/semihosting/semihosting.h>
 
-#if defined(__GNUC__)
-#define likely(x) __builtin_expect(!!(x), 1)
-#define unlikely(x) __builtin_expect(!!(x), 0)
-#else
-#define likely(x) x
-#define unlikely(x) x
-#endif
-
 namespace iss {
 namespace arch {
 
-template <typename BASE, features_e FEAT = FEAT_NONE> class riscv_hart_m_p : public BASE, public riscv_hart_common {
+template <typename BASE, features_e FEAT = FEAT_NONE, typename LOGCAT = logging::disass>
+class riscv_hart_m_p : public BASE, public riscv_hart_common {
 protected:
     const std::array<const char, 4> lvl = {{'U', 'S', 'H', 'M'}};
     const std::array<const char*, 16> trap_str = {{""
@@ -96,7 +91,7 @@ protected:
 
 public:
     using core = BASE;
-    using this_class = riscv_hart_m_p<BASE, FEAT>;
+    using this_class = riscv_hart_m_p<BASE, FEAT, LOGCAT>;
     using phys_addr_t = typename core::phys_addr_t;
     using reg_t = typename core::reg_t;
     using addr_t = typename core::addr_t;
@@ -283,8 +278,8 @@ public:
     void set_mhartid(reg_t mhartid) { mhartid_reg = mhartid; };
 
     void disass_output(uint64_t pc, const std::string instr) override {
-        CLOG(INFO, disass) << fmt::format("0x{:016x}    {:40} [s:0x{:x};c:{}]", pc, instr, (reg_t)state.mstatus,
-                                          this->reg.icount + cycle_offset);
+        NSCLOG(INFO, LOGCAT) << fmt::format("0x{:016x}    {:40} [s:0x{:x};c:{}]", pc, instr, (reg_t)state.mstatus,
+                                            this->reg.cycle + cycle_offset);
     };
 
     iss::instrumentation_if* get_instrumentation_if() override { return &instr_if; }
@@ -298,7 +293,7 @@ public:
 protected:
     struct riscv_instrumentation_if : public iss::instrumentation_if {
 
-        riscv_instrumentation_if(riscv_hart_m_p<BASE, FEAT>& arch)
+        riscv_instrumentation_if(riscv_hart_m_p<BASE, FEAT, LOGCAT>& arch)
         : arch(arch) {}
         /**
          * get the name of this architecture
@@ -317,7 +312,7 @@ protected:
 
         uint64_t get_pendig_traps() override { return arch.reg.trap_state; }
 
-        uint64_t get_total_cycles() override { return arch.reg.icount + arch.cycle_offset; }
+        uint64_t get_total_cycles() override { return arch.reg.cycle + arch.cycle_offset; }
 
         void update_last_instr_cycles(unsigned cycles) override { arch.cycle_offset += cycles - 1; }
 
@@ -327,9 +322,9 @@ protected:
 
         unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
 
-        std::unordered_map<std::string, uint64_t> get_symbol_table(std::string name) override { return arch.get_sym_table(name); }
+        std::unordered_map<std::string, uint64_t> const& get_symbol_table(std::string name) override { return arch.symbol_table; }
 
-        riscv_hart_m_p<BASE, FEAT>& arch;
+        riscv_hart_m_p<BASE, FEAT, LOGCAT>& arch;
     };
 
     friend struct riscv_instrumentation_if;
@@ -349,8 +344,6 @@ protected:
     int64_t instret_offset{0};
     uint64_t minstret_csr{0};
     reg_t fault_data;
-    uint64_t tohost = tohost_dflt;
-    uint64_t fromhost = fromhost_dflt;
     bool tohost_lower_written = false;
     riscv_instrumentation_if instr_if;
 
@@ -383,8 +376,8 @@ protected:
 
     std::vector<uint8_t> tcm;
 
-    iss::status read_csr_reg(unsigned addr, reg_t& val);
-    iss::status write_csr_reg(unsigned addr, reg_t val);
+    iss::status read_plain(unsigned addr, reg_t& val);
+    iss::status write_plain(unsigned addr, reg_t val);
     iss::status read_null(unsigned addr, reg_t& val);
     iss::status write_null(unsigned addr, reg_t val) { return iss::status::Ok; }
     iss::status read_cycle(unsigned addr, reg_t& val);
@@ -405,17 +398,19 @@ protected:
     iss::status read_intstatus(unsigned addr, reg_t& val);
     iss::status write_intthresh(unsigned addr, reg_t val);
     iss::status write_xtvt(unsigned addr, reg_t val);
-    iss::status write_dcsr_dcsr(unsigned addr, reg_t val);
-    iss::status read_dcsr_reg(unsigned addr, reg_t& val);
-    iss::status write_dcsr_reg(unsigned addr, reg_t val);
-    iss::status read_dpc_reg(unsigned addr, reg_t& val);
-    iss::status write_dpc_reg(unsigned addr, reg_t val);
+    iss::status write_dcsr(unsigned addr, reg_t val);
+    iss::status read_debug(unsigned addr, reg_t& val);
+    iss::status write_dscratch(unsigned addr, reg_t val);
+    iss::status read_dpc(unsigned addr, reg_t& val);
+    iss::status write_dpc(unsigned addr, reg_t val);
+    iss::status read_fcsr(unsigned addr, reg_t& val);
+    iss::status write_fcsr(unsigned addr, reg_t val);
 
-    virtual iss::status read_custom_csr_reg(unsigned addr, reg_t& val) { return iss::status::Err; };
-    virtual iss::status write_custom_csr_reg(unsigned addr, reg_t val) { return iss::status::Err; };
+    virtual iss::status read_custom_csr(unsigned addr, reg_t& val) { return iss::status::Err; };
+    virtual iss::status write_custom_csr(unsigned addr, reg_t val) { return iss::status::Err; };
 
-    void register_custom_csr_rd(unsigned addr) { csr_rd_cb[addr] = &this_class::read_custom_csr_reg; }
-    void register_custom_csr_wr(unsigned addr) { csr_wr_cb[addr] = &this_class::write_custom_csr_reg; }
+    void register_custom_csr_rd(unsigned addr) { csr_rd_cb[addr] = &this_class::read_custom_csr; }
+    void register_custom_csr_wr(unsigned addr) { csr_wr_cb[addr] = &this_class::write_custom_csr; }
 
     reg_t mhartid_reg{0x0};
 
@@ -440,8 +435,8 @@ protected:
     std::function<mem_write_f> hart_mem_wr_delegate;
 };
 
-template <typename BASE, features_e FEAT>
-riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
+template <typename BASE, features_e FEAT, typename LOGCAT>
+riscv_hart_m_p<BASE, FEAT, LOGCAT>::riscv_hart_m_p(feature_config cfg)
 : state()
 , instr_if(*this)
 , cfg(cfg) {
@@ -452,18 +447,22 @@ riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
     csr[mimpid] = 1;
 
     uart_buf.str("");
+    if(traits<BASE>::FLEN > 0) {
+        csr_rd_cb[fcsr] = &this_class::read_fcsr;
+        csr_wr_cb[fcsr] = &this_class::write_fcsr;
+    }
     for(unsigned addr = mhpmcounter3; addr <= mhpmcounter31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     if(traits<BASE>::XLEN == 32)
         for(unsigned addr = mhpmcounter3h; addr <= mhpmcounter31h; ++addr) {
             csr_rd_cb[addr] = &this_class::read_null;
-            csr_wr_cb[addr] = &this_class::write_csr_reg;
+            csr_wr_cb[addr] = &this_class::write_plain;
         }
     for(unsigned addr = mhpmevent3; addr <= mhpmevent31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     for(unsigned addr = hpmcounter3; addr <= hpmcounter31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
@@ -471,18 +470,17 @@ riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
     if(traits<BASE>::XLEN == 32)
         for(unsigned addr = hpmcounter3h; addr <= hpmcounter31h; ++addr) {
             csr_rd_cb[addr] = &this_class::read_null;
-            // csr_wr_cb[addr] = &this_class::write_csr_reg;
         }
     // common regs
     const std::array<unsigned, 4> roaddrs{{misa, mvendorid, marchid, mimpid}};
     for(auto addr : roaddrs) {
-        csr_rd_cb[addr] = &this_class::read_csr_reg;
+        csr_rd_cb[addr] = &this_class::read_plain;
         csr_wr_cb[addr] = &this_class::write_null;
     }
     const std::array<unsigned, 4> rwaddrs{{mepc, mtvec, mscratch, mtval}};
     for(auto addr : rwaddrs) {
-        csr_rd_cb[addr] = &this_class::read_csr_reg;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_rd_cb[addr] = &this_class::read_plain;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     // special handling & overrides
     csr_rd_cb[time] = &this_class::read_time;
@@ -523,7 +521,7 @@ riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
     csr_wr_cb[marchid] = &this_class::write_null;
     csr_wr_cb[mimpid] = &this_class::write_null;
     if(FEAT & FEAT_CLIC) {
-        csr_rd_cb[mtvt] = &this_class::read_csr_reg;
+        csr_rd_cb[mtvt] = &this_class::read_plain;
         csr_wr_cb[mtvt] = &this_class::write_xtvt;
         //        csr_rd_cb[mxnti] = &this_class::read_csr_reg;
         //        csr_wr_cb[mxnti] = &this_class::write_csr_reg;
@@ -533,7 +531,7 @@ riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
         //        csr_wr_cb[mscratchcsw] = &this_class::write_csr_reg;
         //        csr_rd_cb[mscratchcswl] = &this_class::read_csr_reg;
         //        csr_wr_cb[mscratchcswl] = &this_class::write_csr_reg;
-        csr_rd_cb[mintthresh] = &this_class::read_csr_reg;
+        csr_rd_cb[mintthresh] = &this_class::read_plain;
         csr_wr_cb[mintthresh] = &this_class::write_intthresh;
         clic_int_reg.resize(cfg.clic_num_irq, clic_int_reg_t{.raw = 0});
         clic_cfg_reg = 0x20;
@@ -559,75 +557,33 @@ riscv_hart_m_p<BASE, FEAT>::riscv_hart_m_p(feature_config cfg)
         insert_mem_range(cfg.tcm_base, cfg.tcm_size, read_clic_cb, write_clic_cb);
     }
     if(FEAT & FEAT_DEBUG) {
-        csr_wr_cb[dscratch0] = &this_class::write_dcsr_reg;
-        csr_rd_cb[dscratch0] = &this_class::read_dcsr_reg;
-        csr_wr_cb[dscratch1] = &this_class::write_dcsr_reg;
-        csr_rd_cb[dscratch1] = &this_class::read_dcsr_reg;
-        csr_wr_cb[dpc] = &this_class::write_dpc_reg;
-        csr_rd_cb[dpc] = &this_class::read_dpc_reg;
-        csr_wr_cb[dcsr] = &this_class::write_dcsr_dcsr;
-        csr_rd_cb[dcsr] = &this_class::read_dcsr_reg;
+        csr_wr_cb[dscratch0] = &this_class::write_dscratch;
+        csr_rd_cb[dscratch0] = &this_class::read_debug;
+        csr_wr_cb[dscratch1] = &this_class::write_dscratch;
+        csr_rd_cb[dscratch1] = &this_class::read_debug;
+        csr_wr_cb[dpc] = &this_class::write_dpc;
+        csr_rd_cb[dpc] = &this_class::read_dpc;
+        csr_wr_cb[dcsr] = &this_class::write_dcsr;
+        csr_rd_cb[dcsr] = &this_class::read_debug;
     }
     hart_mem_rd_delegate = [this](phys_addr_t a, unsigned l, uint8_t* const d) -> iss::status { return this->read_mem(a, l, d); };
     hart_mem_wr_delegate = [this](phys_addr_t a, unsigned l, uint8_t const* const d) -> iss::status { return this->write_mem(a, l, d); };
 }
 
-template <typename BASE, features_e FEAT> std::pair<uint64_t, bool> riscv_hart_m_p<BASE, FEAT>::load_file(std::string name, int type) {
-    get_sym_table(name);
-    try {
-        tohost = symbol_table.at("tohost");
-        fromhost = symbol_table.at("fromhost");
-    } catch(std::out_of_range& e) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+std::pair<uint64_t, bool> riscv_hart_m_p<BASE, FEAT, LOGCAT>::load_file(std::string name, int type) {
+    if(read_elf_file(name, sizeof(reg_t) == 4 ? ELFIO::ELFCLASS32 : ELFIO::ELFCLASS64,
+                     [this](uint64_t addr, uint64_t size, const uint8_t* const data) -> iss::status {
+                         return this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM, addr, size,
+                                            data);
+                     })) {
+        return std::make_pair(entry_address, true);
     }
-    FILE* fp = fopen(name.c_str(), "r");
-    if(fp) {
-        std::array<char, 5> buf;
-        auto n = fread(buf.data(), 1, 4, fp);
-        fclose(fp);
-        if(n != 4)
-            throw std::runtime_error("input file has insufficient size");
-        buf[4] = 0;
-        if(strcmp(buf.data() + 1, "ELF") == 0) {
-            // Create elfio reader
-            ELFIO::elfio reader;
-            // Load ELF data
-            if(!reader.load(name))
-                throw std::runtime_error("could not process elf file");
-            // check elf properties
-            if(reader.get_class() != ELFCLASS32)
-                if(sizeof(reg_t) == 4)
-                    throw std::runtime_error("wrong elf class in file");
-            if(reader.get_type() != ET_EXEC)
-                throw std::runtime_error("wrong elf type in file");
-            if(reader.get_machine() != EM_RISCV)
-                throw std::runtime_error("wrong elf machine in file");
-            auto entry = reader.get_entry();
-            for(const auto pseg : reader.segments) {
-                const auto fsize = pseg->get_file_size(); // 0x42c/0x0
-                const auto seg_data = pseg->get_data();
-                const auto type = pseg->get_type();
-                if(type==1 && fsize > 0) {
-                    auto res = this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM,
-                                           pseg->get_physical_address(), fsize, reinterpret_cast<const uint8_t* const>(seg_data));
-                    if(res != iss::Ok)
-                        CPPLOG(ERR) << "problem writing " << fsize << "bytes to 0x" << std::hex << pseg->get_physical_address();
-                }
-            }
-            for(const auto sec : reader.sections) {
-                if(sec->get_name() == ".tohost") {
-                    tohost = sec->get_address();
-                    fromhost = tohost + 0x40;
-                }
-            }
-            return std::make_pair(entry, true);
-        }
-        throw std::runtime_error(fmt::format("memory load file {} is not a valid elf file", name));
-    }
-    throw std::runtime_error(fmt::format("memory load file not found, check if {} is a valid file", name));
+    return std::make_pair(entry_address, false);
 }
 
-template <typename BASE, features_e FEAT>
-inline void riscv_hart_m_p<BASE, FEAT>::insert_mem_range(uint64_t base, uint64_t size, std::function<mem_read_f> rd_f,
+template <typename BASE, features_e FEAT, typename LOGCAT>
+inline void riscv_hart_m_p<BASE, FEAT, LOGCAT>::insert_mem_range(uint64_t base, uint64_t size, std::function<mem_read_f> rd_f,
                                                                  std::function<mem_write_f> wr_fn) {
     std::tuple<uint64_t, uint64_t> entry{base, size};
     auto it = std::upper_bound(
@@ -639,9 +595,9 @@ inline void riscv_hart_m_p<BASE, FEAT>::insert_mem_range(uint64_t base, uint64_t
     memfn_write.insert(std::begin(memfn_write) + idx, wr_fn);
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::read(const address_type type, const access_type access, const uint32_t space, const uint64_t addr,
-                                             const unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read(const address_type type, const access_type access, const uint32_t space,
+                                                     const uint64_t addr, const unsigned length, uint8_t* const data) {
 #ifndef NDEBUG
     if(access && iss::access_type::DEBUG) {
         CPPLOG(TRACEALL) << "debug read of " << length << " bytes @addr 0x" << std::hex << addr;
@@ -670,7 +626,7 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read(const address_type type, const acce
                 }
                 phys_addr_t phys_addr{access, 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;
@@ -689,8 +645,10 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read(const address_type type, const acce
                 }
                 return res;
             } catch(trap_access& ta) {
+                if((access & access_type::DEBUG) == 0) {
                     this->reg.trap_state = (1UL << 31) | ta.id;
                     fault_data = ta.addr;
+                }
                 return iss::Err;
             }
         } break;
@@ -717,15 +675,17 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read(const address_type type, const acce
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const access_type access, const uint32_t space, const uint64_t addr,
-                                              const unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write(const address_type type, const access_type access, const uint32_t space,
+                                                      const uint64_t addr, const unsigned length, const uint8_t* const data) {
 #ifndef NDEBUG
     const char* prefix = (access && iss::access_type::DEBUG) ? "debug " : "";
     switch(length) {
@@ -746,7 +706,7 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const acc
                       << std::hex << addr;
         break;
     default:
-        CPPLOG(TRACE) << prefix << "write of " << length << " bytes @addr " << addr;
+        CPPLOG(TRACE) << prefix << "write of " << length << " bytes @addr 0x" << std::hex << addr;
     }
 #endif
     try {
@@ -767,7 +727,7 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const acc
                 }
                 phys_addr_t phys_addr{access, 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;
@@ -798,8 +758,6 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const acc
             case 0x10023000: // UART1 base, TXFIFO reg
                 uart_buf << (char)data[0];
                 if(((char)data[0]) == '\n' || data[0] == 0) {
-                    // CPPLOG(INFO)<<"UART"<<((paddr.val>>16)&0x3)<<" send
-                    // '"<<uart_buf.str()<<"'";
                     std::cout << uart_buf.str();
                     uart_buf.str("");
                 }
@@ -850,13 +808,16 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write(const address_type type, const acc
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_csr(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_csr(unsigned addr, reg_t& val) {
     if(addr >= csr.size())
         return iss::Err;
     auto req_priv_lvl = (addr >> 8) & 0x3;
@@ -868,7 +829,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return (this->*(it->second))(addr, val);
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_csr(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_csr(unsigned addr, reg_t val) {
     if(addr >= csr.size())
         return iss::Err;
     auto req_priv_lvl = (addr >> 8) & 0x3;
@@ -882,23 +844,27 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return (this->*(it->second))(addr, val);
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_csr_reg(unsigned addr, reg_t& val) {
-    val = csr[addr];
-    return iss::Ok;
-}
-
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_null(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_null(unsigned addr, reg_t& val) {
     val = 0;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_csr_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_plain(unsigned addr, reg_t& val) {
+    val = csr[addr];
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_plain(unsigned addr, reg_t val) {
     csr[addr] = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_cycle(unsigned addr, reg_t& val) {
-    auto cycle_val = this->reg.icount + cycle_offset;
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_cycle(unsigned addr, reg_t& val) {
+    auto cycle_val = this->reg.cycle + cycle_offset;
     if(addr == mcycle) {
         val = static_cast<reg_t>(cycle_val);
     } else if(addr == mcycleh) {
@@ -907,7 +873,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_cycle(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_cycle(unsigned addr, reg_t val) {
     if(sizeof(typename traits<BASE>::reg_t) != 4) {
         mcycle_csr = static_cast<uint64_t>(val);
     } else {
@@ -917,11 +884,12 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
             mcycle_csr = (static_cast<uint64_t>(val) << 32) + (mcycle_csr & 0xffffffff);
         }
     }
-    cycle_offset = mcycle_csr - this->reg.icount; // TODO: relying on wrap-around
+    cycle_offset = mcycle_csr - this->reg.cycle; // TODO: relying on wrap-around
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_instret(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_instret(unsigned addr, reg_t& val) {
     if((addr & 0xff) == (minstret & 0xff)) {
         val = static_cast<reg_t>(this->reg.instret);
     } else if((addr & 0xff) == (minstreth & 0xff)) {
@@ -930,7 +898,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_instret(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_instret(unsigned addr, reg_t val) {
     if(sizeof(typename traits<BASE>::reg_t) != 4) {
         this->reg.instret = static_cast<uint64_t>(val);
     } else {
@@ -944,8 +913,9 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_time(unsigned addr, reg_t& val) {
-    uint64_t time_val = this->reg.icount / (100000000 / 32768 - 1); //-> ~3052;
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_time(unsigned addr, reg_t& val) {
+    uint64_t time_val = this->reg.cycle / (100000000 / 32768 - 1); //-> ~3052;
     if(addr == time) {
         val = static_cast<reg_t>(time_val);
     } else if(addr == timeh) {
@@ -956,23 +926,27 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_tvec(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_tvec(unsigned addr, reg_t& val) {
     val = FEAT & features_e::FEAT_CLIC ? csr[addr] : csr[addr] & ~2;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_status(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_status(unsigned addr, reg_t& val) {
     val = state.mstatus & hart_state_type::get_mask();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_status(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_status(unsigned addr, reg_t val) {
     state.write_mstatus(val);
     check_interrupt();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_cause(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_cause(unsigned addr, reg_t& val) {
     if((FEAT & features_e::FEAT_CLIC) && (csr[mtvec] & 0x3) == 3) {
         val = csr[addr] & ((1UL << (traits<BASE>::XLEN - 1)) | (mcause_max_irq - 1) | (0xfUL << 16));
         val |= clic_mprev_lvl << 16;
@@ -983,7 +957,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_cause(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_cause(unsigned addr, reg_t val) {
     if((FEAT & features_e::FEAT_CLIC) && (csr[mtvec] & 0x3) == 3) {
         auto mask = ((1UL << (traits<BASE>::XLEN - 1)) | (mcause_max_irq - 1) | (0xfUL << 16));
         csr[addr] = (val & mask) | (csr[addr] & ~mask);
@@ -997,36 +972,42 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_hartid(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_hartid(unsigned addr, reg_t& val) {
     val = mhartid_reg;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_ie(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_ie(unsigned addr, reg_t& val) {
     auto mask = get_irq_mask();
     val = csr[mie] & mask;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_ie(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_ie(unsigned addr, reg_t val) {
     auto mask = get_irq_mask();
     csr[mie] = (csr[mie] & ~mask) | (val & mask);
     check_interrupt();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_ip(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_ip(unsigned addr, reg_t& val) {
     auto mask = get_irq_mask();
     val = csr[mip] & mask;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_epc(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_epc(unsigned addr, reg_t val) {
     csr[addr] = val & get_pc_mask();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_dcsr_dcsr(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_dcsr(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     //                  +-------------- ebreakm
@@ -1037,51 +1018,70 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_dcsr_reg(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_debug(unsigned addr, reg_t& val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     val = csr[addr];
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_dcsr_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_dscratch(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     csr[addr] = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_dpc_reg(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_dpc(unsigned addr, reg_t& val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     val = this->reg.DPC;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_dpc_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_dpc(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     this->reg.DPC = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::read_intstatus(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_intstatus(unsigned addr, reg_t& val) {
     val = (clic_mact_lvl & 0xff) << 24;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_intthresh(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_fcsr(unsigned addr, reg_t& val) {
+    val = this->get_fcsr();
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_fcsr(unsigned addr, reg_t val) {
+    this->set_fcsr(val);
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_intthresh(unsigned addr, reg_t val) {
     csr[addr] = (val & 0xff) | (1 << (cfg.clic_int_ctl_bits)) - 1;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_m_p<BASE, FEAT>::write_xtvt(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_xtvt(unsigned addr, reg_t val) {
     csr[addr] = val & ~0x3fULL;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::read_mem(phys_addr_t paddr, unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_mem(phys_addr_t paddr, unsigned length, uint8_t* const data) {
     switch(paddr.val) {
     default: {
         for(auto offs = 0U; offs < length; ++offs) {
@@ -1092,8 +1092,8 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read_mem(phys_addr_t paddr, unsigned len
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::write_mem(phys_addr_t paddr, unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_mem(phys_addr_t paddr, unsigned length, const uint8_t* const data) {
     switch(paddr.val) {
     // TODO remove UART, Peripherals should not be part of the ISS
     case 0xFFFF0000: // UART0 base, TXFIFO reg
@@ -1151,8 +1151,8 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write_mem(phys_addr_t paddr, unsigned le
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::read_clic(uint64_t addr, unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::read_clic(uint64_t addr, unsigned length, uint8_t* const data) {
     if(addr == cfg.clic_base) { // cliccfg
         *data = clic_cfg_reg;
         for(auto i = 1; i < length; ++i)
@@ -1171,8 +1171,8 @@ iss::status riscv_hart_m_p<BASE, FEAT>::read_clic(uint64_t addr, unsigned length
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_m_p<BASE, FEAT>::write_clic(uint64_t addr, unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_m_p<BASE, FEAT, LOGCAT>::write_clic(uint64_t addr, unsigned length, const uint8_t* const data) {
     if(addr == cfg.clic_base) { // cliccfg
         clic_cfg_reg = (clic_cfg_reg & ~0x1e) | (*data & 0x1e);
     } else if(addr >= (cfg.clic_base + 0x40) && (addr + length) <= (cfg.clic_base + 0x40 + cfg.clic_num_trigger * 4)) { // clicinttrig
@@ -1187,12 +1187,12 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write_clic(uint64_t addr, unsigned lengt
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> inline void riscv_hart_m_p<BASE, FEAT>::reset(uint64_t address) {
+template <typename BASE, features_e FEAT, typename LOGCAT> inline void riscv_hart_m_p<BASE, FEAT, LOGCAT>::reset(uint64_t address) {
     BASE::reset(address);
     state.mstatus = hart_state_type::mstatus_reset_val;
 }
 
-template <typename BASE, features_e FEAT> void riscv_hart_m_p<BASE, FEAT>::check_interrupt() {
+template <typename BASE, features_e FEAT, typename LOGCAT> void riscv_hart_m_p<BASE, FEAT, LOGCAT>::check_interrupt() {
     // TODO: Implement CLIC functionality
     // auto ideleg = csr[mideleg];
     // Multiple simultaneous interrupts and traps at the same privilege level are
@@ -1215,7 +1215,8 @@ template <typename BASE, features_e FEAT> void riscv_hart_m_p<BASE, FEAT>::check
     }
 }
 
-template <typename BASE, features_e FEAT> uint64_t riscv_hart_m_p<BASE, FEAT>::enter_trap(uint64_t flags, uint64_t addr, uint64_t instr) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+uint64_t riscv_hart_m_p<BASE, FEAT, LOGCAT>::enter_trap(uint64_t flags, uint64_t addr, uint64_t tval) {
     // flags are ACTIVE[31:31], CAUSE[30:16], TRAPID[15:0]
     // calculate and write mcause val
     auto const trap_id = bit_sub<0, 16>(flags);
@@ -1236,10 +1237,10 @@ template <typename BASE, features_e FEAT> uint64_t riscv_hart_m_p<BASE, FEAT>::e
          */
         switch(cause) {
         case 0:
-            csr[mtval] = static_cast<reg_t>(addr);
+            csr[mtval] = static_cast<reg_t>(tval);
             break;
         case 2:
-            csr[mtval] = (!has_compressed() || (instr & 0x3) == 3) ? instr : instr & 0xffff;
+            csr[mtval] = (!has_compressed() || (tval & 0x3) == 3) ? tval : tval & 0xffff;
             break;
         case 3:
             if((FEAT & FEAT_DEBUG) && (csr[dcsr] & 0x8000)) {
@@ -1268,7 +1269,7 @@ template <typename BASE, features_e FEAT> uint64_t riscv_hart_m_p<BASE, FEAT>::e
 #else
                     sprintf(buffer.data(), "0x%016lx", addr);
 #endif
-                    CLOG(INFO, disass) << "Semihosting call at address " << buffer.data() << " occurred ";
+                    NSCLOG(INFO, LOGCAT) << "Semihosting call at address " << buffer.data() << " occurred ";
 
                     semihosting_cb(this, &(this->reg.X10) /*a0*/, &(this->reg.X11) /*a1*/);
                     return this->reg.NEXT_PC;
@@ -1324,18 +1325,18 @@ template <typename BASE, features_e FEAT> uint64_t riscv_hart_m_p<BASE, FEAT>::e
     sprintf(buffer.data(), "0x%016lx", addr);
 #endif
     if((flags & 0xffffffff) != 0xffffffff)
-        CLOG(INFO, disass) << (trap_id ? "Interrupt" : "Trap") << " with cause '" << (trap_id ? irq_str[cause] : trap_str[cause]) << "' ("
+        NSCLOG(INFO, LOGCAT) << (trap_id ? "Interrupt" : "Trap") << " with cause '" << (trap_id ? irq_str[cause] : trap_str[cause]) << "' ("
                              << cause << ")"
                              << " at address " << buffer.data() << " occurred";
     return this->reg.NEXT_PC;
 }
 
-template <typename BASE, features_e FEAT> uint64_t riscv_hart_m_p<BASE, FEAT>::leave_trap(uint64_t flags) {
+template <typename BASE, features_e FEAT, typename LOGCAT> uint64_t riscv_hart_m_p<BASE, FEAT, LOGCAT>::leave_trap(uint64_t flags) {
     state.mstatus.MIE = state.mstatus.MPIE;
     state.mstatus.MPIE = 1;
     // sets the pc to the value stored in the x epc register.
     this->reg.NEXT_PC = csr[mepc] & get_pc_mask();
-    CLOG(INFO, disass) << "Executing xRET";
+    NSCLOG(INFO, LOGCAT) << "Executing xRET";
     check_interrupt();
     this->reg.trap_state = this->reg.pending_trap;
     return this->reg.NEXT_PC;

src/iss/arch/riscv_hart_msu_vp.h

@@ -39,7 +39,9 @@
 #include "iss/instrumentation_if.h"
 #include "iss/log_categories.h"
 #include "iss/vm_if.h"
+#include "iss/vm_types.h"
 #include "riscv_hart_common.h"
+#include <stdexcept>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
 #endif
@@ -57,14 +59,6 @@
 
 #include <iss/semihosting/semihosting.h>
 
-#if defined(__GNUC__)
-#define likely(x) __builtin_expect(!!(x), 1)
-#define unlikely(x) __builtin_expect(!!(x), 0)
-#else
-#define likely(x) x
-#define unlikely(x) x
-#endif
-
 namespace iss {
 namespace arch {
 
@@ -334,7 +328,7 @@ public:
 
     void disass_output(uint64_t pc, const std::string instr) override {
         CLOG(INFO, disass) << fmt::format("0x{:016x}    {:40} [p:{};s:0x{:x};c:{}]", pc, instr, lvl[this->reg.PRIV], (reg_t)state.mstatus,
-                                          this->reg.icount + cycle_offset);
+                                          this->reg.cycle + cycle_offset);
     };
 
     iss::instrumentation_if* get_instrumentation_if() override { return &instr_if; }
@@ -367,7 +361,7 @@ protected:
 
         uint64_t get_pendig_traps() override { return arch.reg.trap_state; }
 
-        uint64_t get_total_cycles() override { return arch.reg.icount + arch.cycle_offset; }
+        uint64_t get_total_cycles() override { return arch.reg.cycle + arch.cycle_offset; }
 
         void update_last_instr_cycles(unsigned cycles) override { arch.cycle_offset += cycles - 1; }
 
@@ -377,7 +371,7 @@ protected:
 
         unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
 
-        std::unordered_map<std::string, uint64_t> get_symbol_table(std::string name) override { return arch.get_sym_table(name); }
+        std::unordered_map<std::string, uint64_t> const& get_symbol_table(std::string name) override { return arch.symbol_table; }
 
         riscv_hart_msu_vp<BASE>& arch;
     };
@@ -399,8 +393,6 @@ protected:
     uint64_t minstret_csr{0};
     reg_t fault_data;
     std::array<vm_info, 2> vm;
-    uint64_t tohost = tohost_dflt;
-    uint64_t fromhost = fromhost_dflt;
     bool tohost_lower_written = false;
     riscv_instrumentation_if instr_if;
 
@@ -563,71 +555,14 @@ riscv_hart_msu_vp<BASE>::riscv_hart_msu_vp()
 }
 
 template <typename BASE> std::pair<uint64_t, bool> riscv_hart_msu_vp<BASE>::load_file(std::string name, int type) {
-    FILE* fp = fopen(name.c_str(), "r");
-    if(fp) {
-        std::array<char, 5> buf;
-        auto n = fread(buf.data(), 1, 4, fp);
-        fclose(fp);
-        if(n != 4)
-            throw std::runtime_error("input file has insufficient size");
-        buf[4] = 0;
-        if(strcmp(buf.data() + 1, "ELF") == 0) {
-            // Create elfio reader
-            ELFIO::elfio reader;
-            // Load ELF data
-            if(!reader.load(name))
-                throw std::runtime_error("could not process elf file");
-            // check elf properties
-            if(reader.get_class() != ELFCLASS32)
-                if(sizeof(reg_t) == 4)
-                    throw std::runtime_error("wrong elf class in file");
-            if(reader.get_type() != ET_EXEC)
-                throw std::runtime_error("wrong elf type in file");
-            if(reader.get_machine() != EM_RISCV)
-                throw std::runtime_error("wrong elf machine in file");
-            auto entry = reader.get_entry();
-            for(const auto pseg : reader.segments) {
-                const auto fsize = pseg->get_file_size(); // 0x42c/0x0
-                const auto seg_data = pseg->get_data();
-                const auto type = pseg->get_type();
-                if(type==1 && fsize > 0) {
-                    auto res = this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM,
-                                           pseg->get_physical_address(), fsize, reinterpret_cast<const uint8_t* const>(seg_data));
-                    if(res != iss::Ok)
-                        CPPLOG(ERR) << "problem writing " << fsize << "bytes to 0x" << std::hex << pseg->get_physical_address();
+    if(read_elf_file(name, sizeof(reg_t) == 4 ? ELFIO::ELFCLASS32 : ELFIO::ELFCLASS64,
+                     [this](uint64_t addr, uint64_t size, const uint8_t* const data) -> iss::status {
+                         return this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM, addr, size,
+                                            data);
+                     })) {
+        return std::make_pair(entry_address, true);
     }
-            }
-            for(const auto sec : reader.sections) {
-                if(sec->get_name() == ".symtab") {
-                    if(SHT_SYMTAB == sec->get_type() || SHT_DYNSYM == sec->get_type()) {
-                        ELFIO::symbol_section_accessor symbols(reader, sec);
-                        auto sym_no = symbols.get_symbols_num();
-                        std::string name;
-                        ELFIO::Elf64_Addr value = 0;
-                        ELFIO::Elf_Xword size = 0;
-                        unsigned char bind = 0;
-                        unsigned char type = 0;
-                        ELFIO::Elf_Half section = 0;
-                        unsigned char other = 0;
-                        for(auto i = 0U; i < sym_no; ++i) {
-                            symbols.get_symbol(i, name, value, size, bind, type, section, other);
-                            if(name == "tohost") {
-                                tohost = value;
-                            } else if(name == "fromhost") {
-                                fromhost = value;
-                            }
-                        }
-                    }
-                } else if(sec->get_name() == ".tohost") {
-                    tohost = sec->get_address();
-                    fromhost = tohost + 0x40;
-                }
-            }
-            return std::make_pair(entry, true);
-        }
-        throw std::runtime_error(fmt::format("memory load file {} is not a valid elf file", name));
-    }
-    throw std::runtime_error(fmt::format("memory load file not found, check if {} is a valid file", name));
+    return std::make_pair(entry_address, false);
 }
 
 template <typename BASE>
@@ -677,8 +612,10 @@ iss::status riscv_hart_msu_vp<BASE>::read(const address_type type, const access_
                 }
                 return res;
             } catch(trap_access& ta) {
-                this->reg.trap_state = (1 << 31) | ta.id;
+                if((access & access_type::DEBUG) == 0) {
+                    this->reg.trap_state = (1UL << 31) | ta.id;
                     fault_data = ta.addr;
+                }
                 return iss::Err;
             }
         } break;
@@ -716,8 +653,10 @@ iss::status riscv_hart_msu_vp<BASE>::read(const address_type type, const access_
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
@@ -847,8 +786,10 @@ iss::status riscv_hart_msu_vp<BASE>::write(const address_type type, const access
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
@@ -895,7 +836,7 @@ template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_reg(unsigned
 }
 
 template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_cycle(unsigned addr, reg_t& val) {
-    auto cycle_val = this->reg.icount + cycle_offset;
+    auto cycle_val = this->reg.cycle + cycle_offset;
     if(addr == mcycle) {
         val = static_cast<reg_t>(cycle_val);
     } else if(addr == mcycleh) {
@@ -916,7 +857,7 @@ template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_cycle(unsign
             mcycle_csr = (static_cast<uint64_t>(val) << 32) + (mcycle_csr & 0xffffffff);
         }
     }
-    cycle_offset = mcycle_csr - this->reg.icount; // TODO: relying on wrap-around
+    cycle_offset = mcycle_csr - this->reg.cycle; // TODO: relying on wrap-around
     return iss::Ok;
 }
 
@@ -944,7 +885,7 @@ template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::write_instret(unsi
 }
 
 template <typename BASE> iss::status riscv_hart_msu_vp<BASE>::read_time(unsigned addr, reg_t& val) {
-    uint64_t time_val = this->reg.icount / (100000000 / 32768 - 1); //-> ~3052;
+    uint64_t time_val = this->reg.cycle / (100000000 / 32768 - 1); //-> ~3052;
     if(addr == time) {
         val = static_cast<reg_t>(time_val);
     } else if(addr == timeh) {

src/iss/arch/riscv_hart_mu_p.h

@@ -39,7 +39,9 @@
 #include "iss/instrumentation_if.h"
 #include "iss/log_categories.h"
 #include "iss/vm_if.h"
+#include "iss/vm_types.h"
 #include "riscv_hart_common.h"
+#include <stdexcept>
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
 #endif
@@ -57,18 +59,11 @@
 
 #include <iss/semihosting/semihosting.h>
 
-#if defined(__GNUC__)
-#define likely(x) __builtin_expect(!!(x), 1)
-#define unlikely(x) __builtin_expect(!!(x), 0)
-#else
-#define likely(x) x
-#define unlikely(x) x
-#endif
-
 namespace iss {
 namespace arch {
 
-template <typename BASE, features_e FEAT = FEAT_NONE> class riscv_hart_mu_p : public BASE, public riscv_hart_common {
+template <typename BASE, features_e FEAT = FEAT_NONE, typename LOGCAT = logging::disass>
+class riscv_hart_mu_p : public BASE, public riscv_hart_common {
 protected:
     const std::array<const char, 4> lvl = {{'U', 'S', 'H', 'M'}};
     const std::array<const char*, 16> trap_str = {{""
@@ -95,7 +90,7 @@ protected:
 
 public:
     using core = BASE;
-    using this_class = riscv_hart_mu_p<BASE, FEAT>;
+    using this_class = riscv_hart_mu_p<BASE, FEAT, LOGCAT>;
     using phys_addr_t = typename core::phys_addr_t;
     using reg_t = typename core::reg_t;
     using addr_t = typename core::addr_t;
@@ -309,8 +304,8 @@ public:
     void set_mhartid(reg_t mhartid) { mhartid_reg = mhartid; };
 
     void disass_output(uint64_t pc, const std::string instr) override {
-        CLOG(INFO, disass) << fmt::format("0x{:016x}    {:40} [p:{};s:0x{:x};c:{}]", pc, instr, lvl[this->reg.PRIV], (reg_t)state.mstatus,
-                                          this->reg.icount + cycle_offset);
+        NSCLOG(INFO, LOGCAT) << fmt::format("0x{:016x}    {:40} [p:{};s:0x{:x};c:{}]", pc, instr, lvl[this->reg.PRIV], (reg_t)state.mstatus,
+                                            this->reg.cycle + cycle_offset);
     };
 
     iss::instrumentation_if* get_instrumentation_if() override { return &instr_if; }
@@ -324,7 +319,7 @@ public:
 protected:
     struct riscv_instrumentation_if : public iss::instrumentation_if {
 
-        riscv_instrumentation_if(riscv_hart_mu_p<BASE, FEAT>& arch)
+        riscv_instrumentation_if(riscv_hart_mu_p<BASE, FEAT, LOGCAT>& arch)
         : arch(arch) {}
         /**
          * get the name of this architecture
@@ -343,7 +338,7 @@ protected:
 
         uint64_t get_pendig_traps() override { return arch.reg.trap_state; }
 
-        uint64_t get_total_cycles() override { return arch.reg.icount + arch.cycle_offset; }
+        uint64_t get_total_cycles() override { return arch.reg.cycle + arch.cycle_offset; }
 
         void update_last_instr_cycles(unsigned cycles) override { arch.cycle_offset += cycles - 1; }
 
@@ -353,9 +348,9 @@ protected:
 
         unsigned get_reg_size(unsigned num) override { return traits<BASE>::reg_bit_widths[num]; }
 
-        std::unordered_map<std::string, uint64_t> get_symbol_table(std::string name) override { return arch.get_sym_table(name); }
+        std::unordered_map<std::string, uint64_t> const& get_symbol_table(std::string name) override { return arch.symbol_table; }
 
-        riscv_hart_mu_p<BASE, FEAT>& arch;
+        riscv_hart_mu_p<BASE, FEAT, LOGCAT>& arch;
     };
 
     friend struct riscv_instrumentation_if;
@@ -375,8 +370,6 @@ protected:
     int64_t instret_offset{0};
     uint64_t minstret_csr{0};
     reg_t fault_data;
-    uint64_t tohost = tohost_dflt;
-    uint64_t fromhost = fromhost_dflt;
     bool tohost_lower_written = false;
     riscv_instrumentation_if instr_if;
 
@@ -409,8 +402,8 @@ protected:
 
     std::vector<uint8_t> tcm;
 
-    iss::status read_csr_reg(unsigned addr, reg_t& val);
-    iss::status write_csr_reg(unsigned addr, reg_t val);
+    iss::status read_plain(unsigned addr, reg_t& val);
+    iss::status write_plain(unsigned addr, reg_t val);
     iss::status read_null(unsigned addr, reg_t& val);
     iss::status write_null(unsigned addr, reg_t val) { return iss::status::Ok; }
     iss::status read_cycle(unsigned addr, reg_t& val);
@@ -433,15 +426,17 @@ protected:
     iss::status read_intstatus(unsigned addr, reg_t& val);
     iss::status write_intthresh(unsigned addr, reg_t val);
     iss::status write_xtvt(unsigned addr, reg_t val);
-    iss::status write_dcsr_dcsr(unsigned addr, reg_t val);
-    iss::status read_dcsr_reg(unsigned addr, reg_t& val);
-    iss::status write_dcsr_reg(unsigned addr, reg_t val);
-    iss::status read_dpc_reg(unsigned addr, reg_t& val);
-    iss::status write_dpc_reg(unsigned addr, reg_t val);
-    iss::status write_pmpcfg_reg(unsigned addr, reg_t val);
+    iss::status write_dcsr(unsigned addr, reg_t val);
+    iss::status read_debug(unsigned addr, reg_t& val);
+    iss::status write_dscratch(unsigned addr, reg_t val);
+    iss::status read_dpc(unsigned addr, reg_t& val);
+    iss::status write_dpc(unsigned addr, reg_t val);
+    iss::status read_fcsr(unsigned addr, reg_t& val);
+    iss::status write_fcsr(unsigned addr, reg_t val);
+    iss::status write_pmpcfg(unsigned addr, reg_t val);
 
-    virtual iss::status read_custom_csr_reg(unsigned addr, reg_t& val) { return iss::status::Err; };
-    virtual iss::status write_custom_csr_reg(unsigned addr, reg_t val) { return iss::status::Err; };
+    virtual iss::status read_custom_csr(unsigned addr, reg_t& val) { return iss::status::Err; };
+    virtual iss::status write_custom_csr(unsigned addr, reg_t val) { return iss::status::Err; };
 
     void register_custom_csr_rd(unsigned addr) { csr_rd_cb[addr] = &this_class::read_custom_csr_reg; }
     void register_custom_csr_wr(unsigned addr) { csr_wr_cb[addr] = &this_class::write_custom_csr_reg; }
@@ -469,8 +464,8 @@ protected:
     std::function<mem_write_f> hart_mem_wr_delegate;
 };
 
-template <typename BASE, features_e FEAT>
-riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
+template <typename BASE, features_e FEAT, typename LOGCAT>
+riscv_hart_mu_p<BASE, FEAT, LOGCAT>::riscv_hart_mu_p(feature_config cfg)
 : state()
 , instr_if(*this)
 , cfg(cfg) {
@@ -481,18 +476,22 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
     csr[mimpid] = 1;
 
     uart_buf.str("");
+    if(traits<BASE>::FLEN > 0) {
+        csr_rd_cb[fcsr] = &this_class::read_fcsr;
+        csr_wr_cb[fcsr] = &this_class::write_fcsr;
+    }
     for(unsigned addr = mhpmcounter3; addr <= mhpmcounter31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     if(traits<BASE>::XLEN == 32)
         for(unsigned addr = mhpmcounter3h; addr <= mhpmcounter31h; ++addr) {
             csr_rd_cb[addr] = &this_class::read_null;
-            csr_wr_cb[addr] = &this_class::write_csr_reg;
+            csr_wr_cb[addr] = &this_class::write_plain;
         }
     for(unsigned addr = mhpmevent3; addr <= mhpmevent31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     for(unsigned addr = hpmcounter3; addr <= hpmcounter31; ++addr) {
         csr_rd_cb[addr] = &this_class::read_null;
@@ -500,12 +499,11 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
     if(traits<BASE>::XLEN == 32)
         for(unsigned addr = hpmcounter3h; addr <= hpmcounter31h; ++addr) {
             csr_rd_cb[addr] = &this_class::read_null;
-            // csr_wr_cb[addr] = &this_class::write_csr_reg;
         }
     // common regs
     const std::array<unsigned, 4> roaddrs{{misa, mvendorid, marchid, mimpid}};
     for(auto addr : roaddrs) {
-        csr_rd_cb[addr] = &this_class::read_csr_reg;
+        csr_rd_cb[addr] = &this_class::read_plain;
         csr_wr_cb[addr] = &this_class::write_null;
     }
     const std::array<unsigned, 8> rwaddrs{{
@@ -519,8 +517,8 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
         utval,
     }};
     for(auto addr : rwaddrs) {
-        csr_rd_cb[addr] = &this_class::read_csr_reg;
-        csr_wr_cb[addr] = &this_class::write_csr_reg;
+        csr_rd_cb[addr] = &this_class::read_plain;
+        csr_wr_cb[addr] = &this_class::write_plain;
     }
     // special handling & overrides
     csr_rd_cb[time] = &this_class::read_time;
@@ -565,18 +563,18 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
 
     if(FEAT & FEAT_PMP) {
         for(size_t i = pmpaddr0; i <= pmpaddr15; ++i) {
-            csr_rd_cb[i] = &this_class::read_csr_reg;
-            csr_wr_cb[i] = &this_class::write_csr_reg;
+            csr_rd_cb[i] = &this_class::read_plain;
+            csr_wr_cb[i] = &this_class::write_plain;
         }
         for(size_t i = pmpcfg0; i < pmpcfg0 + 16 / sizeof(reg_t); ++i) {
-            csr_rd_cb[i] = &this_class::read_csr_reg;
-            csr_wr_cb[i] = &this_class::write_pmpcfg_reg;
+            csr_rd_cb[i] = &this_class::read_plain;
+            csr_wr_cb[i] = &this_class::write_pmpcfg;
         }
     }
     if(FEAT & FEAT_EXT_N) {
-        csr_rd_cb[mideleg] = &this_class::read_csr_reg;
+        csr_rd_cb[mideleg] = &this_class::read_plain;
         csr_wr_cb[mideleg] = &this_class::write_ideleg;
-        csr_rd_cb[medeleg] = &this_class::read_csr_reg;
+        csr_rd_cb[medeleg] = &this_class::read_plain;
         csr_wr_cb[medeleg] = &this_class::write_edeleg;
         csr_rd_cb[uie] = &this_class::read_ie;
         csr_wr_cb[uie] = &this_class::write_ie;
@@ -590,7 +588,7 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
         csr_rd_cb[utvec] = &this_class::read_tvec;
     }
     if(FEAT & FEAT_CLIC) {
-        csr_rd_cb[mtvt] = &this_class::read_csr_reg;
+        csr_rd_cb[mtvt] = &this_class::read_plain;
         csr_wr_cb[mtvt] = &this_class::write_xtvt;
         //        csr_rd_cb[mxnti] = &this_class::read_csr_reg;
         //        csr_wr_cb[mxnti] = &this_class::write_csr_reg;
@@ -600,14 +598,14 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
         //        csr_wr_cb[mscratchcsw] = &this_class::write_csr_reg;
         //        csr_rd_cb[mscratchcswl] = &this_class::read_csr_reg;
         //        csr_wr_cb[mscratchcswl] = &this_class::write_csr_reg;
-        csr_rd_cb[mintthresh] = &this_class::read_csr_reg;
+        csr_rd_cb[mintthresh] = &this_class::read_plain;
         csr_wr_cb[mintthresh] = &this_class::write_intthresh;
         if(FEAT & FEAT_EXT_N) {
-            csr_rd_cb[utvt] = &this_class::read_csr_reg;
+            csr_rd_cb[utvt] = &this_class::read_plain;
             csr_wr_cb[utvt] = &this_class::write_xtvt;
             csr_rd_cb[uintstatus] = &this_class::read_intstatus;
             csr_wr_cb[uintstatus] = &this_class::write_null;
-            csr_rd_cb[uintthresh] = &this_class::read_csr_reg;
+            csr_rd_cb[uintthresh] = &this_class::read_plain;
             csr_wr_cb[uintthresh] = &this_class::write_intthresh;
         }
         clic_int_reg.resize(cfg.clic_num_irq, clic_int_reg_t{.raw = 0});
@@ -636,89 +634,33 @@ riscv_hart_mu_p<BASE, FEAT>::riscv_hart_mu_p(feature_config cfg)
         insert_mem_range(cfg.tcm_base, cfg.tcm_size, read_clic_cb, write_clic_cb);
     }
     if(FEAT & FEAT_DEBUG) {
-        csr_wr_cb[dscratch0] = &this_class::write_dcsr_reg;
-        csr_rd_cb[dscratch0] = &this_class::read_dcsr_reg;
-        csr_wr_cb[dscratch1] = &this_class::write_dcsr_reg;
-        csr_rd_cb[dscratch1] = &this_class::read_dcsr_reg;
-        csr_wr_cb[dpc] = &this_class::write_dpc_reg;
-        csr_rd_cb[dpc] = &this_class::read_dpc_reg;
-        csr_wr_cb[dcsr] = &this_class::write_dcsr_dcsr;
-        csr_rd_cb[dcsr] = &this_class::read_dcsr_reg;
+        csr_wr_cb[dscratch0] = &this_class::write_dscratch;
+        csr_rd_cb[dscratch0] = &this_class::read_debug;
+        csr_wr_cb[dscratch1] = &this_class::write_dscratch;
+        csr_rd_cb[dscratch1] = &this_class::read_debug;
+        csr_wr_cb[dpc] = &this_class::write_dpc;
+        csr_rd_cb[dpc] = &this_class::read_dpc;
+        csr_wr_cb[dcsr] = &this_class::write_dcsr;
+        csr_rd_cb[dcsr] = &this_class::read_debug;
     }
     hart_mem_rd_delegate = [this](phys_addr_t a, unsigned l, uint8_t* const d) -> iss::status { return this->read_mem(a, l, d); };
     hart_mem_wr_delegate = [this](phys_addr_t a, unsigned l, uint8_t const* const d) -> iss::status { return this->write_mem(a, l, d); };
 }
 
-template <typename BASE, features_e FEAT> std::pair<uint64_t, bool> riscv_hart_mu_p<BASE, FEAT>::load_file(std::string name, int type) {
-    FILE* fp = fopen(name.c_str(), "r");
-    if(fp) {
-        std::array<char, 5> buf;
-        auto n = fread(buf.data(), 1, 4, fp);
-        fclose(fp);
-        if(n != 4)
-            throw std::runtime_error("input file has insufficient size");
-        buf[4] = 0;
-        if(strcmp(buf.data() + 1, "ELF") == 0) {
-            // Create elfio reader
-            ELFIO::elfio reader;
-            // Load ELF data
-            if(!reader.load(name))
-                throw std::runtime_error("could not process elf file");
-            // check elf properties
-            if(reader.get_class() != ELFCLASS32)
-                if(sizeof(reg_t) == 4)
-                    throw std::runtime_error("wrong elf class in file");
-            if(reader.get_type() != ET_EXEC)
-                throw std::runtime_error("wrong elf type in file");
-            if(reader.get_machine() != EM_RISCV)
-                throw std::runtime_error("wrong elf machine in file");
-            auto entry = reader.get_entry();
-            for(const auto pseg : reader.segments) {
-                const auto fsize = pseg->get_file_size(); // 0x42c/0x0
-                const auto seg_data = pseg->get_data();
-                const auto type = pseg->get_type();
-                if(type==1 && fsize > 0) {
-                    auto res = this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM,
-                                           pseg->get_physical_address(), fsize, reinterpret_cast<const uint8_t* const>(seg_data));
-                    if(res != iss::Ok)
-                        CPPLOG(ERR) << "problem writing " << fsize << "bytes to 0x" << std::hex << pseg->get_physical_address();
+template <typename BASE, features_e FEAT, typename LOGCAT>
+std::pair<uint64_t, bool> riscv_hart_mu_p<BASE, FEAT, LOGCAT>::load_file(std::string name, int type) {
+    if(read_elf_file(name, sizeof(reg_t) == 4 ? ELFIO::ELFCLASS32 : ELFIO::ELFCLASS64,
+                     [this](uint64_t addr, uint64_t size, const uint8_t* const data) -> iss::status {
+                         return this->write(iss::address_type::PHYSICAL, iss::access_type::DEBUG_WRITE, traits<BASE>::MEM, addr, size,
+                                            data);
+                     })) {
+        return std::make_pair(entry_address, true);
     }
-            }
-            for(const auto sec : reader.sections) {
-                if(sec->get_name() == ".symtab") {
-                    if(SHT_SYMTAB == sec->get_type() || SHT_DYNSYM == sec->get_type()) {
-                        ELFIO::symbol_section_accessor symbols(reader, sec);
-                        auto sym_no = symbols.get_symbols_num();
-                        std::string name;
-                        ELFIO::Elf64_Addr value = 0;
-                        ELFIO::Elf_Xword size = 0;
-                        unsigned char bind = 0;
-                        unsigned char type = 0;
-                        ELFIO::Elf_Half section = 0;
-                        unsigned char other = 0;
-                        for(auto i = 0U; i < sym_no; ++i) {
-                            symbols.get_symbol(i, name, value, size, bind, type, section, other);
-                            if(name == "tohost") {
-                                tohost = value;
-                            } else if(name == "fromhost") {
-                                fromhost = value;
-                            }
-                        }
-                    }
-                } else if(sec->get_name() == ".tohost") {
-                    tohost = sec->get_address();
-                    fromhost = tohost + 0x40;
-                }
-            }
-            return std::make_pair(entry, true);
-        }
-        throw std::runtime_error(fmt::format("memory load file {} is not a valid elf file", name));
-    }
-    throw std::runtime_error(fmt::format("memory load file not found, check if {} is a valid file", name));
+    return std::make_pair(entry_address, false);
 }
 
-template <typename BASE, features_e FEAT>
-inline void riscv_hart_mu_p<BASE, FEAT>::insert_mem_range(uint64_t base, uint64_t size, std::function<mem_read_f> rd_f,
+template <typename BASE, features_e FEAT, typename LOGCAT>
+inline void riscv_hart_mu_p<BASE, FEAT, LOGCAT>::insert_mem_range(uint64_t base, uint64_t size, std::function<mem_read_f> rd_f,
                                                                   std::function<mem_write_f> wr_fn) {
     std::tuple<uint64_t, uint64_t> entry{base, size};
     auto it = std::upper_bound(
@@ -730,13 +672,14 @@ inline void riscv_hart_mu_p<BASE, FEAT>::insert_mem_range(uint64_t base, uint64_
     memfn_write.insert(std::begin(memfn_write) + idx, wr_fn);
 }
 
-template <typename BASE, features_e FEAT> inline iss::status riscv_hart_mu_p<BASE, FEAT>::write_pmpcfg_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+inline iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_pmpcfg(unsigned addr, reg_t val) {
     csr[addr] = val & 0x9f9f9f9f;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-bool riscv_hart_mu_p<BASE, FEAT>::pmp_check(const access_type type, const uint64_t addr, const unsigned len) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+bool riscv_hart_mu_p<BASE, FEAT, LOGCAT>::pmp_check(const access_type type, const uint64_t addr, const unsigned len) {
     constexpr auto PMP_SHIFT = 2U;
     constexpr auto PMP_R = 0x1U;
     constexpr auto PMP_W = 0x2U;
@@ -816,9 +759,9 @@ bool riscv_hart_mu_p<BASE, FEAT>::pmp_check(const access_type type, const uint64
     return !any_active || this->reg.PRIV == PRIV_M;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::read(const address_type type, const access_type access, const uint32_t space, const uint64_t addr,
-                                              const unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read(const address_type type, const access_type access, const uint32_t space,
+                                                      const uint64_t addr, const unsigned length, uint8_t* const data) {
 #ifndef NDEBUG
     if(access && iss::access_type::DEBUG) {
         CPPLOG(TRACEALL) << "debug read of " << length << " bytes @addr 0x" << std::hex << addr;
@@ -875,8 +818,10 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::read(const address_type type, const acc
                 }
                 return res;
             } catch(trap_access& ta) {
+                if((access & access_type::DEBUG) == 0) {
                     this->reg.trap_state = (1UL << 31) | ta.id;
                     fault_data = ta.addr;
+                }
                 return iss::Err;
             }
         } break;
@@ -903,15 +848,17 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::read(const address_type type, const acc
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::write(const address_type type, const access_type access, const uint32_t space, const uint64_t addr,
-                                               const unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write(const address_type type, const access_type access, const uint32_t space,
+                                                       const uint64_t addr, const unsigned length, const uint8_t* const data) {
 #ifndef NDEBUG
     const char* prefix = (access && iss::access_type::DEBUG) ? "debug " : "";
     switch(length) {
@@ -993,8 +940,6 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::write(const address_type type, const ac
             case 0x10023000: // UART1 base, TXFIFO reg
                 uart_buf << (char)data[0];
                 if(((char)data[0]) == '\n' || data[0] == 0) {
-                    // CPPLOG(INFO)<<"UART"<<((addr>>16)&0x3)<<" send
-                    // '"<<uart_buf.str()<<"'";
                     std::cout << uart_buf.str();
                     uart_buf.str("");
                 }
@@ -1045,13 +990,16 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::write(const address_type type, const ac
         }
         return iss::Ok;
     } catch(trap_access& ta) {
+        if((access & access_type::DEBUG) == 0) {
             this->reg.trap_state = (1UL << 31) | ta.id;
             fault_data = ta.addr;
+        }
         return iss::Err;
     }
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_csr(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_csr(unsigned addr, reg_t& val) {
     if(addr >= csr.size())
         return iss::Err;
     auto req_priv_lvl = (addr >> 8) & 0x3;
@@ -1063,7 +1011,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return (this->*(it->second))(addr, val);
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_csr(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_csr(unsigned addr, reg_t val) {
     if(addr >= csr.size())
         return iss::Err;
     auto req_priv_lvl = (addr >> 8) & 0x3;
@@ -1077,23 +1026,27 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return (this->*(it->second))(addr, val);
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_csr_reg(unsigned addr, reg_t& val) {
-    val = csr[addr];
-    return iss::Ok;
-}
-
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_null(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_null(unsigned addr, reg_t& val) {
     val = 0;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_csr_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_plain(unsigned addr, reg_t& val) {
+    val = csr[addr];
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_plain(unsigned addr, reg_t val) {
     csr[addr] = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_cycle(unsigned addr, reg_t& val) {
-    auto cycle_val = this->reg.icount + cycle_offset;
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_cycle(unsigned addr, reg_t& val) {
+    auto cycle_val = this->reg.cycle + cycle_offset;
     if(addr == mcycle) {
         val = static_cast<reg_t>(cycle_val);
     } else if(addr == mcycleh) {
@@ -1102,7 +1055,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_cycle(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_cycle(unsigned addr, reg_t val) {
     if(sizeof(typename traits<BASE>::reg_t) != 4) {
         mcycle_csr = static_cast<uint64_t>(val);
     } else {
@@ -1112,11 +1066,12 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
             mcycle_csr = (static_cast<uint64_t>(val) << 32) + (mcycle_csr & 0xffffffff);
         }
     }
-    cycle_offset = mcycle_csr - this->reg.icount; // TODO: relying on wrap-around
+    cycle_offset = mcycle_csr - this->reg.cycle; // TODO: relying on wrap-around
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_instret(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_instret(unsigned addr, reg_t& val) {
     if((addr & 0xff) == (minstret & 0xff)) {
         val = static_cast<reg_t>(this->reg.instret);
     } else if((addr & 0xff) == (minstreth & 0xff)) {
@@ -1125,7 +1080,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_instret(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_instret(unsigned addr, reg_t val) {
     if(sizeof(typename traits<BASE>::reg_t) != 4) {
         this->reg.instret = static_cast<uint64_t>(val);
     } else {
@@ -1139,8 +1095,9 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_time(unsigned addr, reg_t& val) {
-    uint64_t time_val = this->reg.icount / (100000000 / 32768 - 1); //-> ~3052;
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_time(unsigned addr, reg_t& val) {
+    uint64_t time_val = this->reg.cycle / (100000000 / 32768 - 1); //-> ~3052;
     if(addr == time) {
         val = static_cast<reg_t>(time_val);
     } else if(addr == timeh) {
@@ -1151,22 +1108,27 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_tvec(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_tvec(unsigned addr, reg_t& val) {
     val = FEAT & features_e::FEAT_CLIC ? csr[addr] : csr[addr] & ~2;
     return iss::Ok;
 }
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_status(unsigned addr, reg_t& val) {
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_status(unsigned addr, reg_t& val) {
     val = state.mstatus & hart_state_type::get_mask((addr >> 8) & 0x3);
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_status(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_status(unsigned addr, reg_t val) {
     state.write_mstatus(val, (addr >> 8) & 0x3);
     check_interrupt();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_cause(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_cause(unsigned addr, reg_t& val) {
     if((FEAT & features_e::FEAT_CLIC) && (csr[mtvec] & 0x3) == 3) {
         val = csr[addr] & ((1UL << (traits<BASE>::XLEN - 1)) | (mcause_max_irq - 1) | (0xfUL << 16));
         auto mode = (addr >> 8) & 0x3;
@@ -1186,7 +1148,8 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_cause(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_cause(unsigned addr, reg_t val) {
     if((FEAT & features_e::FEAT_CLIC) && (csr[mtvec] & 0x3) == 3) {
         auto mask = ((1UL << (traits<BASE>::XLEN - 1)) | (mcause_max_irq - 1) | (0xfUL << 16));
         csr[addr] = (val & mask) | (csr[addr] & ~mask);
@@ -1209,12 +1172,14 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_hartid(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_hartid(unsigned addr, reg_t& val) {
     val = mhartid_reg;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_ie(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_ie(unsigned addr, reg_t& val) {
     auto mask = get_irq_mask((addr >> 8) & 0x3);
     val = csr[mie] & mask;
     if(this->reg.PRIV != 3)
@@ -1222,14 +1187,16 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_ie(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_ie(unsigned addr, reg_t val) {
     auto mask = get_irq_mask((addr >> 8) & 0x3);
     csr[mie] = (csr[mie] & ~mask) | (val & mask);
     check_interrupt();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_ip(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_ip(unsigned addr, reg_t& val) {
     auto mask = get_irq_mask((addr >> 8) & 0x3);
     val = csr[mip] & mask;
     if(this->reg.PRIV != 3)
@@ -1237,24 +1204,28 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_ideleg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_ideleg(unsigned addr, reg_t val) {
     auto mask = 0b000100010001; // only U mode supported
     csr[mideleg] = (csr[mideleg] & ~mask) | (val & mask);
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_edeleg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_edeleg(unsigned addr, reg_t val) {
     auto mask = 0b1011001111110111; // bit 14/10 (reserved), bit 11 (Env call), and 3 (break) are hardwired to 0
     csr[medeleg] = (csr[medeleg] & ~mask) | (val & mask);
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_epc(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_epc(unsigned addr, reg_t val) {
     csr[addr] = val & get_pc_mask();
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_dcsr_dcsr(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_dcsr(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     //                  +-------------- ebreakm
@@ -1265,35 +1236,40 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_dcsr_reg(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_debug(unsigned addr, reg_t& val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     val = csr[addr];
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_dcsr_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_dscratch(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     csr[addr] = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_dpc_reg(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_dpc(unsigned addr, reg_t& val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     val = this->reg.DPC;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_dpc_reg(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_dpc(unsigned addr, reg_t val) {
     if(!debug_mode_active())
         throw illegal_instruction_fault(this->fault_data);
     this->reg.DPC = val;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::read_intstatus(unsigned addr, reg_t& val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_intstatus(unsigned addr, reg_t& val) {
     auto mode = (addr >> 8) & 0x3;
     val = clic_uact_lvl & 0xff;
     if(mode == 0x3)
@@ -1301,18 +1277,32 @@ template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_intthresh(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_fcsr(unsigned addr, reg_t& val) {
+    val = this->get_fcsr();
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_fcsr(unsigned addr, reg_t val) {
+    this->set_fcsr(val);
+    return iss::Ok;
+}
+
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_intthresh(unsigned addr, reg_t val) {
     csr[addr] = (val & 0xff) | (1 << (cfg.clic_int_ctl_bits)) - 1;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> iss::status riscv_hart_mu_p<BASE, FEAT>::write_xtvt(unsigned addr, reg_t val) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_xtvt(unsigned addr, reg_t val) {
     csr[addr] = val & ~0x3fULL;
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::read_mem(phys_addr_t paddr, unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_mem(phys_addr_t paddr, unsigned length, uint8_t* const data) {
     switch(paddr.val) {
     default: {
         for(auto offs = 0U; offs < length; ++offs) {
@@ -1323,8 +1313,8 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::read_mem(phys_addr_t paddr, unsigned le
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::write_mem(phys_addr_t paddr, unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_mem(phys_addr_t paddr, unsigned length, const uint8_t* const data) {
     switch(paddr.val) {
     // TODO remove UART, Peripherals should not be part of the ISS
     case 0xFFFF0000: // UART0 base, TXFIFO reg
@@ -1385,8 +1375,8 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::write_mem(phys_addr_t paddr, unsigned l
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::read_clic(uint64_t addr, unsigned length, uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::read_clic(uint64_t addr, unsigned length, uint8_t* const data) {
     if(addr == cfg.clic_base) { // cliccfg
         *data = clic_cfg_reg;
         for(auto i = 1; i < length; ++i)
@@ -1405,8 +1395,8 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::read_clic(uint64_t addr, unsigned lengt
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT>
-iss::status riscv_hart_mu_p<BASE, FEAT>::write_clic(uint64_t addr, unsigned length, const uint8_t* const data) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+iss::status riscv_hart_mu_p<BASE, FEAT, LOGCAT>::write_clic(uint64_t addr, unsigned length, const uint8_t* const data) {
     if(addr == cfg.clic_base) { // cliccfg
         clic_cfg_reg = (clic_cfg_reg & ~0x1e) | (*data & 0x1e);
     } else if(addr >= (cfg.clic_base + 0x40) && (addr + length) <= (cfg.clic_base + 0x40 + cfg.clic_num_trigger * 4)) { // clicinttrig
@@ -1421,12 +1411,12 @@ iss::status riscv_hart_mu_p<BASE, FEAT>::write_clic(uint64_t addr, unsigned leng
     return iss::Ok;
 }
 
-template <typename BASE, features_e FEAT> inline void riscv_hart_mu_p<BASE, FEAT>::reset(uint64_t address) {
+template <typename BASE, features_e FEAT, typename LOGCAT> inline void riscv_hart_mu_p<BASE, FEAT, LOGCAT>::reset(uint64_t address) {
     BASE::reset(address);
     state.mstatus = hart_state_type::mstatus_reset_val;
 }
 
-template <typename BASE, features_e FEAT> void riscv_hart_mu_p<BASE, FEAT>::check_interrupt() {
+template <typename BASE, features_e FEAT, typename LOGCAT> void riscv_hart_mu_p<BASE, FEAT, LOGCAT>::check_interrupt() {
     // TODO: Implement CLIC functionality
     auto ideleg = csr[mideleg];
     // Multiple simultaneous interrupts and traps at the same privilege level are
@@ -1449,7 +1439,8 @@ template <typename BASE, features_e FEAT> void riscv_hart_mu_p<BASE, FEAT>::chec
     }
 }
 
-template <typename BASE, features_e FEAT> uint64_t riscv_hart_mu_p<BASE, FEAT>::enter_trap(uint64_t flags, uint64_t addr, uint64_t instr) {
+template <typename BASE, features_e FEAT, typename LOGCAT>
+uint64_t riscv_hart_mu_p<BASE, FEAT, LOGCAT>::enter_trap(uint64_t flags, uint64_t addr, uint64_t instr) {
     // flags are ACTIVE[31:31], CAUSE[30:16], TRAPID[15:0]
     // calculate and write mcause val
     if(flags == std::numeric_limits<uint64_t>::max())
@@ -1583,7 +1574,7 @@ template <typename BASE, features_e FEAT> uint64_t riscv_hart_mu_p<BASE, FEAT>::
     return this->reg.NEXT_PC;
 }
 
-template <typename BASE, features_e FEAT> uint64_t riscv_hart_mu_p<BASE, FEAT>::leave_trap(uint64_t flags) {
+template <typename BASE, features_e FEAT, typename LOGCAT> uint64_t riscv_hart_mu_p<BASE, FEAT, LOGCAT>::leave_trap(uint64_t flags) {
     auto cur_priv = this->reg.PRIV;
     auto inst_priv = (flags & 0x3) ? 3 : 0;
     if(inst_priv > cur_priv) {

src/iss/arch/tgc5c.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2020 MINRES Technologies GmbH
+ * Copyright (C) 2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without

src/iss/arch/wt_cache.h

@@ -87,7 +87,7 @@ public:
     virtual ~wt_cache() = default;
 
     unsigned size{4096};
-    unsigned line_sz{32};
+    unsigned line_sz{64};
     unsigned ways{1};
     uint64_t io_address{0xf0000000};
     uint64_t io_addr_mask{0xf0000000};
@@ -119,7 +119,7 @@ template <typename BASE> iss::status iss::arch::wt_cache<BASE>::read_cache(phys_
         icache_ptr.reset(new cache::cache(size, line_sz, ways));
         dcache_ptr.reset(new cache::cache(size, line_sz, ways));
     }
-    if((a.val & io_addr_mask) != io_address) {
+    if((a.access & iss::access_type::FETCH) == iss::access_type::FETCH || (a.val & io_addr_mask) != io_address) {
         auto set_addr = (a.val & (size - 1)) >> util::ilog2(line_sz * ways);
         auto tag_addr = a.val >> util::ilog2(line_sz);
         auto& set = (is_fetch(a.access) ? icache_ptr : dcache_ptr)->sets[set_addr];

src/iss/debugger/riscv_target_adapter.h

@@ -30,8 +30,8 @@
  *
  *******************************************************************************/
 
-#ifndef _ISS_DEBUGGER_RISCV_TARGET_ADAPTER_H_
-#define _ISS_DEBUGGER_RISCV_TARGET_ADAPTER_H_
+#ifndef _ISS_ARCH_DEBUGGER_RISCV_TARGET_ADAPTER_H_
+#define _ISS_ARCH_DEBUGGER_RISCV_TARGET_ADAPTER_H_
 
 #include "iss/arch_if.h"
 #include <iss/arch/traits.h>
@@ -48,6 +48,10 @@
 
 namespace iss {
 namespace debugger {
+
+char const* const get_csr_name(unsigned);
+constexpr auto csr_offset = 100U;
+
 using namespace iss::arch;
 using namespace iss::debugger;
 
@@ -129,11 +133,17 @@ public:
 
 protected:
     static inline constexpr addr_t map_addr(const addr_t& i) { return i; }
-
+    std::string csr_xml;
     iss::arch_if* core;
     rp_thread_ref thread_idx;
 };
 
+template <typename ARCH> typename std::enable_if<iss::arch::traits<ARCH>::FLEN != 0, unsigned>::type get_f0_offset() {
+    return iss::arch::traits<ARCH>::F0;
+}
+
+template <typename ARCH> typename std::enable_if<iss::arch::traits<ARCH>::FLEN == 0, unsigned>::type get_f0_offset() { return 0; }
+
 template <typename ARCH> status riscv_target_adapter<ARCH>::set_gen_thread(rp_thread_ref& thread) {
     thread_idx = thread;
     return Ok;
@@ -175,12 +185,29 @@ template <typename ARCH> status riscv_target_adapter<ARCH>::current_thread_query
 
 template <typename ARCH> status riscv_target_adapter<ARCH>::read_registers(std::vector<uint8_t>& data, std::vector<uint8_t>& avail) {
     CPPLOG(TRACE) << "reading target registers";
-    // return idx<0?:;
     data.clear();
     avail.clear();
     const uint8_t* reg_base = core->get_regs_base_ptr();
     auto start_reg = arch::traits<ARCH>::X0;
-    for(size_t reg_no = start_reg; reg_no < start_reg + 33 /*arch::traits<ARCH>::NUM_REGS*/; ++reg_no) {
+    for(size_t i = 0; i < 33; ++i) {
+        if(i < arch::traits<ARCH>::RFS || i == arch::traits<ARCH>::PC) {
+            auto reg_no = i < 32 ? start_reg + i : arch::traits<ARCH>::PC;
+            unsigned offset = traits<ARCH>::reg_byte_offsets[reg_no];
+            for(size_t j = 0; j < arch::traits<ARCH>::XLEN / 8; ++j) {
+                data.push_back(*(reg_base + offset + j));
+                avail.push_back(0xff);
+            }
+        } else {
+            for(size_t j = 0; j < arch::traits<ARCH>::XLEN / 8; ++j) {
+                data.push_back(0);
+                avail.push_back(0);
+            }
+        }
+    }
+    if(iss::arch::traits<ARCH>::FLEN > 0) {
+        auto fstart_reg = get_f0_offset<ARCH>();
+        for(size_t i = 0; i < 32; ++i) {
+            auto reg_no = fstart_reg + i;
             auto reg_width = arch::traits<ARCH>::reg_bit_widths[reg_no] / 8;
             unsigned offset = traits<ARCH>::reg_byte_offsets[reg_no];
             for(size_t j = 0; j < reg_width; ++j) {
@@ -188,21 +215,7 @@ template <typename ARCH> status riscv_target_adapter<ARCH>::read_registers(std::
                 avail.push_back(0xff);
             }
         }
-    // work around fill with F type registers
-    //    if (arch::traits<ARCH>::NUM_REGS < 65) {
-    //        auto reg_width = sizeof(typename arch::traits<ARCH>::reg_t);
-    //        for (size_t reg_no = 0; reg_no < 33; ++reg_no) {
-    //            for (size_t j = 0; j < reg_width; ++j) {
-    //                data.push_back(0x0);
-    //                avail.push_back(0x00);
-    //            }
-    //            // if(arch::traits<ARCH>::XLEN < 64)
-    //            //     for(unsigned j=0; j<4; ++j){
-    //            //         data.push_back(0x0);
-    //            //         avail.push_back(0x00);
-    //            //     }
-    //        }
-    //    }
+    }
     return Ok;
 }
 
@@ -210,25 +223,25 @@ template <typename ARCH> status riscv_target_adapter<ARCH>::write_registers(cons
     auto start_reg = arch::traits<ARCH>::X0;
     auto* reg_base = core->get_regs_base_ptr();
     auto iter = data.data();
-    bool e_ext = arch::traits<ARCH>::PC < 32;
-    for(size_t reg_no = 0; reg_no < start_reg + 33 /*arch::traits<ARCH>::NUM_REGS*/; ++reg_no) {
-        if(e_ext && reg_no > 15) {
-            if(reg_no == 32) {
-                auto reg_width = arch::traits<ARCH>::reg_bit_widths[arch::traits<ARCH>::PC] / 8;
+    auto iter_end = data.data() + data.size();
+    for(size_t i = 0; i < 33 && iter < iter_end; ++i) {
+        auto reg_width = arch::traits<ARCH>::XLEN / 8;
+        if(i < arch::traits<ARCH>::RFS) {
+            auto offset = traits<ARCH>::reg_byte_offsets[start_reg + i];
+            std::copy(iter, iter + reg_width, reg_base + offset);
+        } else if(i == 32) {
             auto offset = traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::PC];
-                std::copy(iter, iter + reg_width, reg_base);
-            } else {
-                const uint64_t zero_val = 0;
-                auto reg_width = arch::traits<ARCH>::reg_bit_widths[15] / 8;
-                auto iter = (uint8_t*)&zero_val;
-                std::copy(iter, iter + reg_width, reg_base);
+            std::copy(iter, iter + reg_width, reg_base + offset);
         }
-        } else {
-            auto reg_width = arch::traits<ARCH>::reg_bit_widths[reg_no] / 8;
-            auto offset = traits<ARCH>::reg_byte_offsets[reg_no];
-            std::copy(iter, iter + reg_width, reg_base);
-            iter += 4;
-            reg_base += offset;
+        iter += reg_width;
+    }
+    if(iss::arch::traits<ARCH>::FLEN > 0) {
+        auto fstart_reg = get_f0_offset<ARCH>();
+        auto reg_width = arch::traits<ARCH>::FLEN / 8;
+        for(size_t i = 0; i < 32 && iter < iter_end; ++i) {
+            unsigned offset = traits<ARCH>::reg_byte_offsets[fstart_reg + i];
+            std::copy(iter, iter + reg_width, reg_base + offset);
+            iter += reg_width;
         }
     }
     return Ok;
@@ -236,7 +249,7 @@ template <typename ARCH> status riscv_target_adapter<ARCH>::write_registers(cons
 
 template <typename ARCH>
 status riscv_target_adapter<ARCH>::read_single_register(unsigned int reg_no, std::vector<uint8_t>& data, std::vector<uint8_t>& avail) {
-    if(reg_no < 65) {
+    if(reg_no < csr_offset) {
         // auto reg_size = arch::traits<ARCH>::reg_bit_width(static_cast<typename
         // arch::traits<ARCH>::reg_e>(reg_no))/8;
         auto* reg_base = core->get_regs_base_ptr();
@@ -247,23 +260,24 @@ status riscv_target_adapter<ARCH>::read_single_register(unsigned int reg_no, std
         std::copy(reg_base + offset, reg_base + offset + reg_width, data.begin());
         std::fill(avail.begin(), avail.end(), 0xff);
     } else {
-        typed_addr_t<iss::address_type::PHYSICAL> a(iss::access_type::DEBUG_READ, traits<ARCH>::CSR, reg_no - 65);
+        typed_addr_t<iss::address_type::PHYSICAL> a(iss::access_type::DEBUG_READ, traits<ARCH>::CSR, reg_no - csr_offset);
         data.resize(sizeof(typename traits<ARCH>::reg_t));
         avail.resize(sizeof(typename traits<ARCH>::reg_t));
         std::fill(avail.begin(), avail.end(), 0xff);
         core->read(a, data.size(), data.data());
+        std::fill(avail.begin(), avail.end(), 0xff);
     }
     return data.size() > 0 ? Ok : Err;
 }
 
 template <typename ARCH> status riscv_target_adapter<ARCH>::write_single_register(unsigned int reg_no, const std::vector<uint8_t>& data) {
-    if(reg_no < 65) {
+    if(reg_no < csr_offset) {
         auto* reg_base = core->get_regs_base_ptr();
         auto reg_width = arch::traits<ARCH>::reg_bit_widths[static_cast<typename arch::traits<ARCH>::reg_e>(reg_no)] / 8;
         auto offset = traits<ARCH>::reg_byte_offsets[reg_no];
         std::copy(data.begin(), data.begin() + reg_width, reg_base + offset);
     } else {
-        typed_addr_t<iss::address_type::PHYSICAL> a(iss::access_type::DEBUG_WRITE, traits<ARCH>::CSR, reg_no - 65);
+        typed_addr_t<iss::address_type::PHYSICAL> a(iss::access_type::DEBUG_WRITE, traits<ARCH>::CSR, reg_no - csr_offset);
         core->write(a, data.size(), data.data());
     }
     return Ok;
@@ -276,7 +290,7 @@ template <typename ARCH> status riscv_target_adapter<ARCH>::read_mem(uint64_t ad
 }
 
 template <typename ARCH> status riscv_target_adapter<ARCH>::write_mem(uint64_t addr, const std::vector<uint8_t>& data) {
-    auto a = map_addr({iss::access_type::DEBUG_READ, iss::address_type::VIRTUAL, 0, addr});
+    auto a = map_addr({iss::access_type::DEBUG_WRITE, iss::address_type::VIRTUAL, 0, addr});
     auto f = [&]() -> status { return core->write(a, data.size(), data.data()); };
     return srv->execute_syncronized(f);
 }
@@ -369,93 +383,57 @@ status riscv_target_adapter<ARCH>::resume_from_addr(bool step, int sig, uint64_t
 }
 
 template <typename ARCH> status riscv_target_adapter<ARCH>::target_xml_query(std::string& out_buf) {
-    const std::string res{"<?xml version=\"1.0\"?><!DOCTYPE target SYSTEM \"gdb-target.dtd\">"
-                          "<target><architecture>riscv:rv32</architecture>"
-                          //"  <feature name=\"org.gnu.gdb.riscv.rv32i\">\n"
-                          //"    <reg name=\"x0\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x1\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x2\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x3\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x4\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x5\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x6\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x7\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x8\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x9\"  bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x10\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x11\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x12\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x13\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x14\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x15\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x16\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x17\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x18\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x19\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x20\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x21\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x22\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x23\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x24\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x25\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x26\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x27\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x28\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x29\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x30\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"    <reg name=\"x31\" bitsize=\"32\" group=\"general\"/>\n"
-                          //"  </feature>\n"
-                          "</target>"};
-    out_buf = res;
+    if(!csr_xml.size()) {
+        std::ostringstream oss;
+        oss << "<?xml version=\"1.0\"?><!DOCTYPE feature SYSTEM \"gdb-target.dtd\"><target version=\"1.0\">\n";
+        if(iss::arch::traits<ARCH>::XLEN == 32)
+            oss << "<architecture>riscv:rv32</architecture>\n";
+        else if(iss::arch::traits<ARCH>::XLEN == 64)
+            oss << "  <architectureriscv:rv64</architecture>\n";
+        oss << "  <feature name=\"org.gnu.gdb.riscv.cpu\">\n";
+        auto reg_base_num = iss::arch::traits<ARCH>::X0;
+        for(auto i = 0U; i < iss::arch::traits<ARCH>::RFS; ++i) {
+            oss << "    <reg name=\"x" << i << "\" bitsize=\"" << iss::arch::traits<ARCH>::reg_bit_widths[reg_base_num + i]
+                << "\" type=\"int\" regnum=\"" << i << "\"/>\n";
+        }
+        oss << "    <reg name=\"pc\" bitsize=\"" << iss::arch::traits<ARCH>::reg_bit_widths[iss::arch::traits<ARCH>::PC]
+            << "\" type=\"code_ptr\" regnum=\"" << 32U << "\"/>\n";
+        oss << "  </feature>\n";
+        if(iss::arch::traits<ARCH>::FLEN > 0) {
+            oss << "  <feature name=\"org.gnu.gdb.riscv.fpu\">\n";
+            auto reg_base_num = get_f0_offset<ARCH>();
+            auto type = iss::arch::traits<ARCH>::FLEN == 32 ? "ieee_single" : "riscv_double";
+            for(auto i = 0U; i < 32; ++i) {
+                oss << "    <reg name=\"f" << i << "\" bitsize=\"" << iss::arch::traits<ARCH>::reg_bit_widths[reg_base_num + i]
+                    << "\" type=\"" << type << "\" regnum=\"" << i + 33 << "\"/>\n";
+            }
+            oss << "    <reg name=\"fcsr\" bitsize=\"" << iss::arch::traits<ARCH>::XLEN << "\" regnum=\"103\" type int/>\n";
+            oss << "    <reg name=\"fflags\" bitsize=\"" << iss::arch::traits<ARCH>::XLEN << "\" regnum=\"101\" type int/>\n";
+            oss << "    <reg name=\"frm\" bitsize=\"" << iss::arch::traits<ARCH>::XLEN << "\" regnum=\"102\" type int/>\n";
+            oss << "  </feature>\n";
+        }
+        oss << "  <feature name=\"org.gnu.gdb.riscv.csr\">\n";
+        std::vector<uint8_t> data;
+        std::vector<uint8_t> avail;
+        data.resize(sizeof(typename traits<ARCH>::reg_t));
+        avail.resize(sizeof(typename traits<ARCH>::reg_t));
+        for(auto i = 0U; i < 4096; ++i) {
+            typed_addr_t<iss::address_type::PHYSICAL> a(iss::access_type::DEBUG_READ, traits<ARCH>::CSR, i);
+            std::fill(avail.begin(), avail.end(), 0xff);
+            auto res = core->read(a, data.size(), data.data());
+            if(res == iss::Ok) {
+                oss << "    <reg name=\"" << get_csr_name(i) << "\" bitsize=\"" << iss::arch::traits<ARCH>::XLEN
+                    << "\"  type=\"int\" regnum=\"" << (i + csr_offset) << "\"/>\n";
+            }
+        }
+        oss << "  </feature>\n";
+        oss << "</target>\n";
+        csr_xml = oss.str();
+    }
+    out_buf = csr_xml;
     return Ok;
 }
-
-/*
- *
-<?xml version="1.0"?>
-<!DOCTYPE target SYSTEM "gdb-target.dtd">
-<target>
-  <architecture>riscv:rv32</architecture>
-
-  <feature name="org.gnu.gdb.riscv.rv32i">
-    <reg name="x0"  bitsize="32" group="general"/>
-    <reg name="x1"  bitsize="32" group="general"/>
-    <reg name="x2"  bitsize="32" group="general"/>
-    <reg name="x3"  bitsize="32" group="general"/>
-    <reg name="x4"  bitsize="32" group="general"/>
-    <reg name="x5"  bitsize="32" group="general"/>
-    <reg name="x6"  bitsize="32" group="general"/>
-    <reg name="x7"  bitsize="32" group="general"/>
-    <reg name="x8"  bitsize="32" group="general"/>
-    <reg name="x9"  bitsize="32" group="general"/>
-    <reg name="x10" bitsize="32" group="general"/>
-    <reg name="x11" bitsize="32" group="general"/>
-    <reg name="x12" bitsize="32" group="general"/>
-    <reg name="x13" bitsize="32" group="general"/>
-    <reg name="x14" bitsize="32" group="general"/>
-    <reg name="x15" bitsize="32" group="general"/>
-    <reg name="x16" bitsize="32" group="general"/>
-    <reg name="x17" bitsize="32" group="general"/>
-    <reg name="x18" bitsize="32" group="general"/>
-    <reg name="x19" bitsize="32" group="general"/>
-    <reg name="x20" bitsize="32" group="general"/>
-    <reg name="x21" bitsize="32" group="general"/>
-    <reg name="x22" bitsize="32" group="general"/>
-    <reg name="x23" bitsize="32" group="general"/>
-    <reg name="x24" bitsize="32" group="general"/>
-    <reg name="x25" bitsize="32" group="general"/>
-    <reg name="x26" bitsize="32" group="general"/>
-    <reg name="x27" bitsize="32" group="general"/>
-    <reg name="x28" bitsize="32" group="general"/>
-    <reg name="x29" bitsize="32" group="general"/>
-    <reg name="x30" bitsize="32" group="general"/>
-    <reg name="x31" bitsize="32" group="general"/>
-  </feature>
-
-</target>
-
- */
 } // namespace debugger
 } // namespace iss
 
-#endif /* _ISS_DEBUGGER_RISCV_TARGET_ADAPTER_H_ */
+#endif /* _ISS_ARCH_DEBUGGER_RISCV_TARGET_ADAPTER_H_ */

src/main.cpp

@@ -69,7 +69,8 @@ int main(int argc, char* argv[]) {
         ("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")
+        ("ilimit,i", po::value<uint64_t>()->default_value(std::numeric_limits<uint64_t>::max()), "max. number of instructions to simulate")
+        ("flimit", po::value<uint64_t>()->default_value(std::numeric_limits<uint64_t>::max()), "max. number of fetches to simulate")
         ("reset,r", po::value<std::string>(), "reset address")
         ("dump-ir", "dump the intermediate representation")
         ("elf,f", po::value<std::vector<std::string>>(), "ELF file(s) to load")
@@ -140,7 +141,10 @@ int main(int argc, char* argv[]) {
             std::tie(cpu, vm) = f.create(isa_opt, clim["gdb-port"].as<unsigned>(), &semihosting_cb);
         }
         if(!cpu) {
-            CPPLOG(ERR) << "Could not create cpu for isa " << isa_opt << " and backend " << clim["backend"].as<std::string>() << std::endl;
+            auto list = f.get_names();
+            std::sort(std::begin(list), std::end(list));
+            CPPLOG(ERR) << "Could not create cpu for isa " << isa_opt << " and backend " << clim["backend"].as<std::string>() << "\n"
+                        << "Available implementations (core|platform|backend):\n  - " << util::join(list, "\n  - ") << std::endl;
             return 127;
         }
         if(!vm) {
@@ -202,21 +206,36 @@ int main(int argc, char* argv[]) {
         if(clim.count("elf"))
             for(std::string input : clim["elf"].as<std::vector<std::string>>()) {
                 auto start_addr = vm->get_arch()->load_file(input);
-                if(start_addr.second) // FIXME: this always evaluates to true as load file always returns <sth, true>
+                if(start_addr.second)
                     start_address = start_addr.first;
+                else {
+                    LOG(ERR) << "Error occured while loading file " << input << std::endl;
+                    return 1;
+                }
             }
         for(std::string input : args) {
             auto start_addr = vm->get_arch()->load_file(input); // treat remaining arguments as elf files
-            if(start_addr.second) // FIXME: this always evaluates to true as load file always returns <sth, true>
+            if(start_addr.second)
                 start_address = start_addr.first;
+            else {
+                LOG(ERR) << "Error occured while loading file " << input << std::endl;
+                return 1;
+            }
         }
         if(clim.count("reset")) {
             auto str = clim["reset"].as<std::string>();
             start_address = str.find("0x") == 0 ? std::stoull(str.substr(2), nullptr, 16) : std::stoull(str, nullptr, 10);
         }
         vm->reset(start_address);
-        auto cycles = clim["instructions"].as<uint64_t>();
-        res = vm->start(cycles, dump);
+        auto limit = clim["ilimit"].as<uint64_t>();
+        auto cond = iss::finish_cond_e::JUMP_TO_SELF;
+        if(clim.count("flimit")) {
+            cond = cond | iss::finish_cond_e::FCOUNT_LIMIT;
+            limit = clim["flimit"].as<uint64_t>();
+        } else {
+            cond = cond | iss::finish_cond_e::ICOUNT_LIMIT;
+        }
+        res = vm->start(limit, dump, cond);
 
         auto instr_if = vm->get_arch()->get_instrumentation_if();
         // this assumes a single input file

src/sysc/core_complex.cpp

@@ -42,7 +42,6 @@
 #include <iss/plugin/loader.h>
 #endif
 #include "sc_core_adapter_if.h"
-#include <iss/arch/tgc_mapper.h>
 #include <scc/report.h>
 #include <util/ities.h>
 #include <iostream>
@@ -125,7 +124,7 @@ using vm_ptr = std::unique_ptr<iss::vm_if>;
 
 class core_wrapper {
 public:
-    core_wrapper(core_complex* owner)
+    core_wrapper(core_complex_if* owner)
     : owner(owner) {}
 
     void reset(uint64_t addr) { vm->reset(addr); }
@@ -181,7 +180,7 @@ public:
                                              "SystemC sub-commands: break <time>, print_time"});
     }
 
-    core_complex* const owner;
+    core_complex_if* const owner;
     vm_ptr vm{nullptr};
     sc_cpu_ptr cpu{nullptr};
     iss::debugger::target_adapter_if* tgt_adapter{nullptr};
@@ -197,9 +196,9 @@ struct core_trace {
     scv_tr_handle tr_handle;
 };
 
-SC_HAS_PROCESS(core_complex); // NOLINT
 #ifndef CWR_SYSTEMC
-core_complex::core_complex(sc_module_name const& name)
+template <unsigned int BUSWIDTH>
+core_complex<BUSWIDTH>::core_complex(sc_module_name const& name)
 : sc_module(name)
 , fetch_lut(tlm_dmi_ext())
 , read_lut(tlm_dmi_ext())
@@ -208,7 +207,7 @@ core_complex::core_complex(sc_module_name const& name)
 }
 #endif
 
-void core_complex::init() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::init() {
     trc = new core_trace();
     ibus.register_invalidate_direct_mem_ptr([=](uint64_t start, uint64_t end) -> void {
         auto lut_entry = fetch_lut.getEntry(start);
@@ -227,6 +226,7 @@ void core_complex::init() {
         }
     });
 
+    SC_HAS_PROCESS(core_complex<BUSWIDTH>); // NOLINT
     SC_THREAD(run);
     SC_METHOD(rst_cb);
     sensitive << rst_i;
@@ -252,16 +252,16 @@ void core_complex::init() {
 #endif
 }
 
-core_complex::~core_complex() {
+template <unsigned int BUSWIDTH> core_complex<BUSWIDTH>::~core_complex() {
     delete cpu;
     delete trc;
     for(auto* p : plugin_list)
         delete p;
 }
 
-void core_complex::trace(sc_trace_file* trf) const {}
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::trace(sc_trace_file* trf) const {}
 
-void core_complex::before_end_of_elaboration() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::before_end_of_elaboration() {
     SCCDEBUG(SCMOD) << "instantiating iss::arch::tgf with " << GET_PROP_VALUE(backend) << " backend";
     // cpu = scc::make_unique<core_wrapper>(this);
     cpu = new core_wrapper(this);
@@ -302,7 +302,7 @@ void core_complex::before_end_of_elaboration() {
     }
 }
 
-void core_complex::start_of_simulation() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::start_of_simulation() {
     // quantum_keeper.reset();
     if(GET_PROP_VALUE(elf_file).size() > 0) {
         istringstream is(GET_PROP_VALUE(elf_file));
@@ -325,7 +325,7 @@ void core_complex::start_of_simulation() {
     }
 }
 
-bool core_complex::disass_output(uint64_t pc, const std::string instr_str) {
+template <unsigned int BUSWIDTH> bool core_complex<BUSWIDTH>::disass_output(uint64_t pc, const std::string instr_str) {
     if(trc->m_db == nullptr)
         return false;
     if(trc->tr_handle.is_active())
@@ -339,7 +339,7 @@ bool core_complex::disass_output(uint64_t pc, const std::string instr_str) {
     return true;
 }
 
-void core_complex::forward() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::forward() {
 #ifndef CWR_SYSTEMC
     set_clock_period(clk_i.read());
 #else
@@ -348,24 +348,24 @@ void core_complex::forward() {
 #endif
 }
 
-void core_complex::set_clock_period(sc_core::sc_time period) {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::set_clock_period(sc_core::sc_time period) {
     curr_clk = period;
     if(period == SC_ZERO_TIME)
         cpu->set_interrupt_execution(true);
 }
 
-void core_complex::rst_cb() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::rst_cb() {
     if(rst_i.read())
         cpu->set_interrupt_execution(true);
 }
 
-void core_complex::sw_irq_cb() { cpu->local_irq(3, sw_irq_i.read()); }
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::sw_irq_cb() { cpu->local_irq(3, sw_irq_i.read()); }
 
-void core_complex::timer_irq_cb() { cpu->local_irq(7, timer_irq_i.read()); }
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::timer_irq_cb() { cpu->local_irq(7, timer_irq_i.read()); }
 
-void core_complex::ext_irq_cb() { cpu->local_irq(11, ext_irq_i.read()); }
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::ext_irq_cb() { cpu->local_irq(11, ext_irq_i.read()); }
 
-void core_complex::local_irq_cb() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::local_irq_cb() {
     for(auto i = 0U; i < local_irq_i.size(); ++i) {
         if(local_irq_i[i].event()) {
             cpu->local_irq(16 + i, local_irq_i[i].read());
@@ -373,7 +373,7 @@ void core_complex::local_irq_cb() {
     }
 }
 
-void core_complex::run() {
+template <unsigned int BUSWIDTH> void core_complex<BUSWIDTH>::run() {
     wait(SC_ZERO_TIME); // separate from elaboration phase
     do {
         wait(SC_ZERO_TIME);
@@ -391,7 +391,7 @@ void core_complex::run() {
     sc_stop();
 }
 
-bool core_complex::read_mem(uint64_t addr, unsigned length, uint8_t* const data, bool is_fetch) {
+template <unsigned int BUSWIDTH> bool core_complex<BUSWIDTH>::read_mem(uint64_t addr, unsigned length, uint8_t* const data, bool is_fetch) {
     auto& dmi_lut = is_fetch ? fetch_lut : read_lut;
     auto lut_entry = dmi_lut.getEntry(addr);
     if(lut_entry.get_granted_access() != tlm::tlm_dmi::DMI_ACCESS_NONE && addr + length <= lut_entry.get_end_address() + 1) {
@@ -449,7 +449,7 @@ bool core_complex::read_mem(uint64_t addr, unsigned length, uint8_t* const data,
     }
 }
 
-bool core_complex::write_mem(uint64_t addr, unsigned length, const uint8_t* const data) {
+template <unsigned int BUSWIDTH> bool core_complex<BUSWIDTH>::write_mem(uint64_t addr, unsigned length, const uint8_t* const data) {
     auto lut_entry = write_lut.getEntry(addr);
     if(lut_entry.get_granted_access() != tlm::tlm_dmi::DMI_ACCESS_NONE && addr + length <= lut_entry.get_end_address() + 1) {
         auto offset = addr - lut_entry.get_start_address();
@@ -497,7 +497,7 @@ bool core_complex::write_mem(uint64_t addr, unsigned length, const uint8_t* cons
     }
 }
 
-bool core_complex::read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const data) {
+template <unsigned int BUSWIDTH> bool core_complex<BUSWIDTH>::read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const data) {
     tlm::tlm_generic_payload gp;
     gp.set_command(tlm::TLM_READ_COMMAND);
     gp.set_address(addr);
@@ -507,7 +507,7 @@ bool core_complex::read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const d
     return dbus->transport_dbg(gp) == length;
 }
 
-bool core_complex::write_mem_dbg(uint64_t addr, unsigned length, const uint8_t* const data) {
+template <unsigned int BUSWIDTH> bool core_complex<BUSWIDTH>::write_mem_dbg(uint64_t addr, unsigned length, const uint8_t* const data) {
     write_buf.resize(length);
     std::copy(data, data + length, write_buf.begin()); // need to copy as TLM does not guarantee data integrity
     tlm::tlm_generic_payload gp;
@@ -518,5 +518,10 @@ bool core_complex::write_mem_dbg(uint64_t addr, unsigned length, const uint8_t*
     gp.set_streaming_width(length);
     return dbus->transport_dbg(gp) == length;
 }
+
+template class core_complex<scc::LT>;
+template class core_complex<32>;
+template class core_complex<64>;
+
 } /* namespace tgfs */
 } /* namespace sysc */

src/sysc/core_complex.h

@@ -33,6 +33,7 @@
 #ifndef _SYSC_CORE_COMPLEX_H_
 #define _SYSC_CORE_COMPLEX_H_
 
+#include <scc/signal_opt_ports.h>
 #include <scc/tick2time.h>
 #include <scc/traceable.h>
 #include <scc/utilities.h>
@@ -40,10 +41,8 @@
 #include <tlm/scc/scv/tlm_rec_initiator_socket.h>
 #ifdef CWR_SYSTEMC
 #include <scmlinc/scml_property.h>
-#define SOCKET_WIDTH 32
 #else
 #include <cci_configuration>
-#define SOCKET_WIDTH scc::LT
 #endif
 #include <memory>
 #include <tlm>
@@ -68,12 +67,35 @@ public:
 namespace tgfs {
 class core_wrapper;
 struct core_trace;
+struct core_complex_if {
 
-class core_complex : public sc_core::sc_module, public scc::traceable {
+    virtual ~core_complex_if() = default;
+
+    virtual bool read_mem(uint64_t addr, unsigned length, uint8_t* const data, bool is_fetch) = 0;
+
+    virtual bool write_mem(uint64_t addr, unsigned length, const uint8_t* const data) = 0;
+
+    virtual bool read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const data) = 0;
+
+    virtual bool write_mem_dbg(uint64_t addr, unsigned length, const uint8_t* const data) = 0;
+
+    virtual bool disass_output(uint64_t pc, const std::string instr) = 0;
+
+    virtual unsigned get_last_bus_cycles() = 0;
+
+    //! Allow quantum keeper handling
+    virtual void sync(uint64_t) = 0;
+
+    virtual char const* hier_name() = 0;
+
+    scc::sc_in_opt<uint64_t> mtime_i{"mtime_i"};
+};
+
+template <unsigned int BUSWIDTH = scc::LT> class core_complex : public sc_core::sc_module, public scc::traceable, public core_complex_if {
 public:
-    tlm::scc::initiator_mixin<tlm::tlm_initiator_socket<SOCKET_WIDTH>> ibus{"ibus"};
+    tlm::scc::initiator_mixin<tlm::tlm_initiator_socket<BUSWIDTH>> ibus{"ibus"};
 
-    tlm::scc::initiator_mixin<tlm::tlm_initiator_socket<SOCKET_WIDTH>> dbus{"dbus"};
+    tlm::scc::initiator_mixin<tlm::tlm_initiator_socket<BUSWIDTH>> dbus{"dbus"};
 
     sc_core::sc_in<bool> rst_i{"rst_i"};
 
@@ -88,8 +110,6 @@ public:
 #ifndef CWR_SYSTEMC
     sc_core::sc_in<sc_core::sc_time> clk_i{"clk_i"};
 
-    sc_core::sc_port<tlm::tlm_peek_if<uint64_t>, 1, sc_core::SC_ZERO_OR_MORE_BOUND> mtime_o{"mtime_o"};
-
     cci::cci_param<std::string> elf_file{"elf_file", ""};
 
     cci::cci_param<bool> enable_disass{"enable_disass", false};
@@ -115,8 +135,6 @@ public:
 #else
     sc_core::sc_in<bool> clk_i{"clk_i"};
 
-    sc_core::sc_in<uint64_t> mtime_i{"mtime_i"};
-
     scml_property<std::string> elf_file{"elf_file", ""};
 
     scml_property<bool> enable_disass{"enable_disass", false};
@@ -159,13 +177,13 @@ public:
 
     ~core_complex();
 
-    inline unsigned get_last_bus_cycles() {
+    unsigned get_last_bus_cycles() override {
         auto mem_incr = std::max(ibus_inc, dbus_inc);
         ibus_inc = dbus_inc = 0;
         return mem_incr > 1 ? mem_incr : 1;
     }
 
-    inline void sync(uint64_t cycle) {
+    void sync(uint64_t cycle) override {
         auto core_inc = curr_clk * (cycle - last_sync_cycle);
         quantum_keeper.inc(core_inc);
         if(quantum_keeper.need_sync()) {
@@ -175,20 +193,22 @@ public:
         last_sync_cycle = cycle;
     }
 
-    bool read_mem(uint64_t addr, unsigned length, uint8_t* const data, bool is_fetch);
+    bool read_mem(uint64_t addr, unsigned length, uint8_t* const data, bool is_fetch) override;
 
-    bool write_mem(uint64_t addr, unsigned length, const uint8_t* const data);
+    bool write_mem(uint64_t addr, unsigned length, const uint8_t* const data) override;
 
-    bool read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const data);
+    bool read_mem_dbg(uint64_t addr, unsigned length, uint8_t* const data) override;
 
-    bool write_mem_dbg(uint64_t addr, unsigned length, const uint8_t* const data);
+    bool write_mem_dbg(uint64_t addr, unsigned length, const uint8_t* const data) override;
 
     void trace(sc_core::sc_trace_file* trf) const override;
 
-    bool disass_output(uint64_t pc, const std::string instr);
+    bool disass_output(uint64_t pc, const std::string instr) override;
 
     void set_clock_period(sc_core::sc_time period);
 
+    char const* hier_name() override { return name(); }
+
 protected:
     void before_end_of_elaboration() override;
     void start_of_simulation() override;

src/sysc/register_tgc_c.cpp

@@ -46,12 +46,12 @@ using namespace sysc;
 volatile std::array<bool, 2> tgc_init = {
     iss_factory::instance().register_creator("tgc5c|m_p|interp",
                                              [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
                                                  auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::tgc5c>>(cc);
                                                  return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
                                              }),
     iss_factory::instance().register_creator("tgc5c|mu_p|interp", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-        auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+        auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
         auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::tgc5c>>(cc);
         return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
     })};
@@ -62,12 +62,12 @@ using namespace sysc;
 volatile std::array<bool, 2> tgc_init = {
     iss_factory::instance().register_creator("tgc5c|m_p|llvm",
                                              [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
                                                  auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::tgc5c>>(cc);
                                                  return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
                                              }),
     iss_factory::instance().register_creator("tgc5c|mu_p|llvm", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-        auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+        auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
         auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::tgc5c>>(cc);
         return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
     })};
@@ -79,12 +79,12 @@ using namespace sysc;
 volatile std::array<bool, 2> tgc_init = {
     iss_factory::instance().register_creator("tgc5c|m_p|tcc",
                                              [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
                                                  auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::tgc5c>>(cc);
                                                  return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
                                              }),
     iss_factory::instance().register_creator("tgc5c|mu_p|tcc", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-        auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+        auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
         auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::tgc5c>>(cc);
         return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
     })};
@@ -96,12 +96,12 @@ using namespace sysc;
 volatile std::array<bool, 2> tgc_init = {
     iss_factory::instance().register_creator("tgc5c|m_p|asmjit",
                                              [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+                                                 auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
                                                  auto* cpu = new sc_core_adapter<arch::riscv_hart_m_p<arch::tgc5c>>(cc);
                                                  return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
                                              }),
     iss_factory::instance().register_creator("tgc5c|mu_p|asmjit", [](unsigned gdb_port, void* data) -> iss_factory::base_t {
-        auto cc = reinterpret_cast<sysc::tgfs::core_complex*>(data);
+        auto cc = reinterpret_cast<sysc::tgfs::core_complex_if*>(data);
         auto* cpu = new sc_core_adapter<arch::riscv_hart_mu_p<arch::tgc5c>>(cc);
         return {sysc::sc_cpu_ptr{cpu}, vm_ptr{create(static_cast<arch::tgc5c*>(cpu), gdb_port)}};
     })};

src/sysc/sc_core_adapter.h

@@ -21,7 +21,7 @@ public:
     using reg_t = typename iss::arch::traits<typename PLAT::core>::reg_t;
     using phys_addr_t = typename iss::arch::traits<typename PLAT::core>::phys_addr_t;
     using heart_state_t = typename PLAT::hart_state_type;
-    sc_core_adapter(sysc::tgfs::core_complex* owner)
+    sc_core_adapter(sysc::tgfs::core_complex_if* owner)
     : owner(owner) {}
 
     iss::arch_if* get_arch_if() override { return this; }
@@ -54,9 +54,9 @@ public:
             std::stringstream s;
             s << "[p:" << lvl[this->reg.PRIV] << ";s:0x" << std::hex << std::setfill('0') << std::setw(sizeof(reg_t) * 2)
               << (reg_t)this->state.mstatus << std::dec << ";c:" << this->reg.icount + this->cycle_offset << "]";
-            SCCDEBUG(owner->name()) << "disass: "
-                                    << "0x" << std::setw(16) << std::right << std::setfill('0') << std::hex << pc << "\t\t" << std::setw(40)
-                                    << std::setfill(' ') << std::left << instr << s.str();
+            SCCDEBUG(owner->hier_name()) << "disass: "
+                                         << "0x" << std::setw(16) << std::right << std::setfill('0') << std::hex << pc << "\t\t"
+                                         << std::setw(40) << std::setfill(' ') << std::left << instr << s.str();
         }
     };
 
@@ -79,10 +79,10 @@ public:
                     switch(hostvar >> 48) {
                     case 0:
                         if(hostvar != 0x1) {
-                            SCCINFO(owner->name())
+                            SCCINFO(owner->hier_name())
                                 << "tohost value is 0x" << std::hex << hostvar << std::dec << " (" << hostvar << "), stopping simulation";
                         } else {
-                            SCCINFO(owner->name())
+                            SCCINFO(owner->hier_name())
                                 << "tohost value is 0x" << std::hex << hostvar << std::dec << " (" << hostvar << "), stopping simulation";
                         }
                         this->reg.trap_state = std::numeric_limits<uint32_t>::max();
@@ -112,21 +112,8 @@ public:
     }
 
     iss::status read_csr(unsigned addr, reg_t& val) override {
-#ifndef CWR_SYSTEMC
-        if((addr == iss::arch::time || addr == iss::arch::timeh) && owner->mtime_o.get_interface(0)) {
-            uint64_t time_val;
-            bool ret = owner->mtime_o->nb_peek(time_val);
-            if(addr == iss::arch::time) {
-                val = static_cast<reg_t>(time_val);
-            } else if(addr == iss::arch::timeh) {
-                if(sizeof(reg_t) != 4)
-                    return iss::Err;
-                val = static_cast<reg_t>(time_val >> 32);
-            }
-            return ret ? iss::Ok : iss::Err;
-#else
         if((addr == iss::arch::time || addr == iss::arch::timeh)) {
-            uint64_t time_val = owner->mtime_i.read();
+            uint64_t time_val = owner->mtime_i.get_interface() ? owner->mtime_i.read() : 0;
             if(addr == iss::arch::time) {
                 val = static_cast<reg_t>(time_val);
             } else if(addr == iss::arch::timeh) {
@@ -135,14 +122,13 @@ public:
                 val = static_cast<reg_t>(time_val >> 32);
             }
             return iss::Ok;
-#endif
         } else {
             return PLAT::read_csr(addr, val);
         }
     }
 
     void wait_until(uint64_t flags) override {
-        SCCDEBUG(owner->name()) << "Sleeping until interrupt";
+        SCCDEBUG(owner->hier_name()) << "Sleeping until interrupt";
         while(this->reg.pending_trap == 0 && (this->csr[iss::arch::mip] & this->csr[iss::arch::mie]) == 0) {
             sc_core::wait(wfi_evt);
         }
@@ -173,11 +159,11 @@ public:
             this->csr[iss::arch::mip] &= ~mask;
         this->check_interrupt();
         if(value)
-            SCCTRACE(owner->name()) << "Triggering interrupt " << id << " Pending trap: " << this->reg.pending_trap;
+            SCCTRACE(owner->hier_name()) << "Triggering interrupt " << id << " Pending trap: " << this->reg.pending_trap;
     }
 
 private:
-    sysc::tgfs::core_complex* const owner;
+    sysc::tgfs::core_complex_if* const owner{nullptr};
     sc_core::sc_event wfi_evt;
     uint64_t hostvar{std::numeric_limits<uint64_t>::max()};
     unsigned to_host_wr_cnt = 0;

src/vm/fp_functions.cpp

@@ -33,6 +33,7 @@
 ////////////////////////////////////////////////////////////////////////////////
 
 #include "fp_functions.h"
+#include <array>
 
 extern "C" {
 #include "internals.h"
@@ -43,9 +44,10 @@ extern "C" {
 #include <limits>
 
 using this_t = uint8_t*;
-const uint8_t rmm_map[] = {
+// this does not inlcude any reserved rm or the DYN rm, as DYN rm should be taken care of in the vm_impl
+const std::array<uint8_t, 5> rmm_map = {
     softfloat_round_near_even /*RNE*/, softfloat_round_minMag /*RTZ*/, softfloat_round_min /*RDN*/, softfloat_round_max /*RUP?*/,
-    softfloat_round_near_maxMag /*RMM*/, softfloat_round_max /*RTZ*/,    softfloat_round_max /*RTZ*/, softfloat_round_max /*RTZ*/,
+    softfloat_round_near_maxMag /*RMM*/
 };
 
 const uint32_t quiet_nan32 = 0x7fC00000;
@@ -56,7 +58,7 @@ uint32_t fget_flags() { return softfloat_exceptionFlags & 0x1f; }
 
 uint32_t fadd_s(uint32_t v1, uint32_t v2, uint8_t mode) {
     float32_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float32_t r = f32_add(v1f, v2f);
     return r.v;
@@ -64,7 +66,7 @@ uint32_t fadd_s(uint32_t v1, uint32_t v2, uint8_t mode) {
 
 uint32_t fsub_s(uint32_t v1, uint32_t v2, uint8_t mode) {
     float32_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float32_t r = f32_sub(v1f, v2f);
     return r.v;
@@ -72,7 +74,7 @@ uint32_t fsub_s(uint32_t v1, uint32_t v2, uint8_t mode) {
 
 uint32_t fmul_s(uint32_t v1, uint32_t v2, uint8_t mode) {
     float32_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float32_t r = f32_mul(v1f, v2f);
     return r.v;
@@ -80,7 +82,7 @@ uint32_t fmul_s(uint32_t v1, uint32_t v2, uint8_t mode) {
 
 uint32_t fdiv_s(uint32_t v1, uint32_t v2, uint8_t mode) {
     float32_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float32_t r = f32_div(v1f, v2f);
     return r.v;
@@ -88,7 +90,7 @@ uint32_t fdiv_s(uint32_t v1, uint32_t v2, uint8_t mode) {
 
 uint32_t fsqrt_s(uint32_t v1, uint8_t mode) {
     float32_t v1f{v1};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float32_t r = f32_sqrt(v1f);
     return r.v;
@@ -130,18 +132,18 @@ uint32_t fcvt_s(uint32_t v1, uint32_t op, uint8_t mode) {
     softfloat_exceptionFlags = 0;
     float32_t r;
     switch(op) {
-    case 0: { // w->s, fp to int32
-        uint_fast32_t res = f32_to_i32(v1f, rmm_map[mode & 0x7], true);
+    case 0: { // FCVT__W__S
+        uint_fast32_t res = f32_to_i32(v1f, rmm_map.at(mode), true);
         return (uint32_t)res;
     }
-    case 1: { // wu->s
-        uint_fast32_t res = f32_to_ui32(v1f, rmm_map[mode & 0x7], true);
+    case 1: { // FCVT__WU__S
+        uint_fast32_t res = f32_to_ui32(v1f, rmm_map.at(mode), true);
         return (uint32_t)res;
     }
-    case 2: // s->w
-        r = i32_to_f32(v1);
+    case 2: // FCVT__S__W
+        r = i32_to_f32((int32_t)v1);
         return r.v;
-    case 3: // s->wu
+    case 3: // FCVT__S__WU
         r = ui32_to_f32(v1);
         return r.v;
     }
@@ -149,12 +151,24 @@ uint32_t fcvt_s(uint32_t v1, uint32_t op, uint8_t mode) {
 }
 
 uint32_t fmadd_s(uint32_t v1, uint32_t v2, uint32_t v3, uint32_t op, uint8_t mode) {
-    // op should be {softfloat_mulAdd_subProd(2), softfloat_mulAdd_subC(1)}
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    uint32_t F32_SIGN = 1UL << 31;
+    switch(op) {
+    case 0: // FMADD_S
+        break;
+    case 1: // FMSUB_S
+        v3 ^= F32_SIGN;
+        break;
+    case 2: // FNMADD_S
+        v1 ^= F32_SIGN;
+        v3 ^= F32_SIGN;
+        break;
+    case 3: // FNMSUB_S
+        v1 ^= F32_SIGN;
+        break;
+    }
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
-    float32_t res = softfloat_mulAddF32(v1, v2, v3, op & 0x1);
-    if(op > 1)
-        res.v ^= 1ULL << 31;
+    float32_t res = softfloat_mulAddF32(v1, v2, v3, 0);
     return res.v;
 }
 
@@ -189,8 +203,8 @@ uint32_t fclass_s(uint32_t v1) {
     uA.f = a;
     uiA = uA.ui;
 
-    uint_fast16_t infOrNaN = expF32UI(uiA) == 0xFF;
-    uint_fast16_t subnormalOrZero = expF32UI(uiA) == 0;
+    bool infOrNaN = expF32UI(uiA) == 0xFF;
+    bool subnormalOrZero = expF32UI(uiA) == 0;
     bool sign = signF32UI(uiA);
     bool fracZero = fracF32UI(uiA) == 0;
     bool isNaN = isNaNF32UI(uiA);
@@ -203,9 +217,13 @@ uint32_t fclass_s(uint32_t v1) {
 }
 
 uint32_t fconv_d2f(uint64_t v1, uint8_t mode) {
-    softfloat_roundingMode = rmm_map[mode & 0x7];
-    bool nan = (v1 & defaultNaNF64UI) == defaultNaNF64UI;
-    if(nan) {
+    bool isNan = isNaNF64UI(v1);
+    bool isSNaN = softfloat_isSigNaNF64UI(v1);
+    softfloat_roundingMode = rmm_map.at(mode);
+    softfloat_exceptionFlags = 0;
+    if(isNan) {
+        if(isSNaN)
+            softfloat_raiseFlags(softfloat_flag_invalid);
         return defaultNaNF32UI;
     } else {
         float32_t res = f64_to_f32(float64_t{v1});
@@ -214,11 +232,11 @@ uint32_t fconv_d2f(uint64_t v1, uint8_t mode) {
 }
 
 uint64_t fconv_f2d(uint32_t v1, uint8_t mode) {
-    bool nan = (v1 & defaultNaNF32UI) == defaultNaNF32UI;
-    if(nan) {
+    bool infOrNaN = expF32UI(v1) == 0xFF;
+    bool subnormalOrZero = expF32UI(v1) == 0;
+    if(infOrNaN || subnormalOrZero) {
         return defaultNaNF64UI;
     } else {
-        softfloat_roundingMode = rmm_map[mode & 0x7];
         float64_t res = f32_to_f64(float32_t{v1});
         return res.v;
     }
@@ -228,7 +246,7 @@ uint64_t fadd_d(uint64_t v1, uint64_t v2, uint8_t mode) {
     bool nan = (v1 & defaultNaNF32UI) == quiet_nan32;
     bool snan = softfloat_isSigNaNF32UI(v1);
     float64_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float64_t r = f64_add(v1f, v2f);
     return r.v;
@@ -236,7 +254,7 @@ uint64_t fadd_d(uint64_t v1, uint64_t v2, uint8_t mode) {
 
 uint64_t fsub_d(uint64_t v1, uint64_t v2, uint8_t mode) {
     float64_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float64_t r = f64_sub(v1f, v2f);
     return r.v;
@@ -244,7 +262,7 @@ uint64_t fsub_d(uint64_t v1, uint64_t v2, uint8_t mode) {
 
 uint64_t fmul_d(uint64_t v1, uint64_t v2, uint8_t mode) {
     float64_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float64_t r = f64_mul(v1f, v2f);
     return r.v;
@@ -252,7 +270,7 @@ uint64_t fmul_d(uint64_t v1, uint64_t v2, uint8_t mode) {
 
 uint64_t fdiv_d(uint64_t v1, uint64_t v2, uint8_t mode) {
     float64_t v1f{v1}, v2f{v2};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float64_t r = f64_div(v1f, v2f);
     return r.v;
@@ -260,7 +278,7 @@ uint64_t fdiv_d(uint64_t v1, uint64_t v2, uint8_t mode) {
 
 uint64_t fsqrt_d(uint64_t v1, uint8_t mode) {
     float64_t v1f{v1};
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
     float64_t r = f64_sqrt(v1f);
     return r.v;
@@ -298,22 +316,23 @@ uint64_t fcmp_d(uint64_t v1, uint64_t v2, uint32_t op) {
 }
 
 uint64_t fcvt_d(uint64_t v1, uint32_t op, uint8_t mode) {
+
     float64_t v1f{v1};
     softfloat_exceptionFlags = 0;
     float64_t r;
     switch(op) {
-    case 0: { // l->d, fp to int32
-        int64_t res = f64_to_i64(v1f, rmm_map[mode & 0x7], true);
+    case 0: { // l from d
+        int64_t res = f64_to_i64(v1f, rmm_map.at(mode), true);
         return (uint64_t)res;
     }
-    case 1: { // lu->s
-        uint64_t res = f64_to_ui64(v1f, rmm_map[mode & 0x7], true);
+    case 1: { // lu from d
+        uint64_t res = f64_to_ui64(v1f, rmm_map.at(mode), true);
         return res;
     }
-    case 2: // s->l
+    case 2: // d from l
         r = i64_to_f64(v1);
         return r.v;
-    case 3: // s->lu
+    case 3: // d from lu
         r = ui64_to_f64(v1);
         return r.v;
     }
@@ -321,12 +340,24 @@ uint64_t fcvt_d(uint64_t v1, uint32_t op, uint8_t mode) {
 }
 
 uint64_t fmadd_d(uint64_t v1, uint64_t v2, uint64_t v3, uint32_t op, uint8_t mode) {
-    // op should be {softfloat_mulAdd_subProd(2), softfloat_mulAdd_subC(1)}
-    softfloat_roundingMode = rmm_map[mode & 0x7];
+    uint64_t F64_SIGN = 1ULL << 63;
+    switch(op) {
+    case 0: // FMADD_D
+        break;
+    case 1: // FMSUB_D
+        v3 ^= F64_SIGN;
+        break;
+    case 2: // FNMADD_D
+        v1 ^= F64_SIGN;
+        v3 ^= F64_SIGN;
+        break;
+    case 3: // FNMSUB_D
+        v1 ^= F64_SIGN;
+        break;
+    }
+    softfloat_roundingMode = rmm_map.at(mode);
     softfloat_exceptionFlags = 0;
-    float64_t res = softfloat_mulAddF64(v1, v2, v3, op & 0x1);
-    if(op > 1)
-        res.v ^= 1ULL << 63;
+    float64_t res = softfloat_mulAddF64(v1, v2, v3, 0);
     return res.v;
 }
 
@@ -362,8 +393,8 @@ uint64_t fclass_d(uint64_t v1) {
     uA.f = a;
     uiA = uA.ui;
 
-    uint_fast16_t infOrNaN = expF64UI(uiA) == 0x7FF;
-    uint_fast16_t subnormalOrZero = expF64UI(uiA) == 0;
+    bool infOrNaN = expF64UI(uiA) == 0x7FF;
+    bool subnormalOrZero = expF64UI(uiA) == 0;
     bool sign = signF64UI(uiA);
     bool fracZero = fracF64UI(uiA) == 0;
     bool isNaN = isNaNF64UI(uiA);
@@ -381,9 +412,9 @@ uint64_t fcvt_32_64(uint32_t v1, uint32_t op, uint8_t mode) {
     float64_t r;
     switch(op) {
     case 0: // l->s, fp to int32
-        return f32_to_i64(v1f, rmm_map[mode & 0x7], true);
+        return f32_to_i64(v1f, rmm_map.at(mode), true);
     case 1: // wu->s
-        return f32_to_ui64(v1f, rmm_map[mode & 0x7], true);
+        return f32_to_ui64(v1f, rmm_map.at(mode), true);
     case 2: // s->w
         r = i32_to_f64(v1);
         return r.v;
@@ -399,11 +430,11 @@ uint32_t fcvt_64_32(uint64_t v1, uint32_t op, uint8_t mode) {
     float32_t r;
     switch(op) {
     case 0: { // wu->s
-        int32_t r = f64_to_i32(float64_t{v1}, rmm_map[mode & 0x7], true);
+        int32_t r = f64_to_i32(float64_t{v1}, rmm_map.at(mode), true);
         return r;
     }
     case 1: { // wu->s
-        uint32_t r = f64_to_ui32(float64_t{v1}, rmm_map[mode & 0x7], true);
+        uint32_t r = f64_to_ui32(float64_t{v1}, rmm_map.at(mode), true);
         return r;
     }
     case 2: // l->s, fp to int32

src/vm/interp/vm_tgc5c.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 20217-2024 MINRES Technologies GmbH
+ * Copyright (C) 2017-2024 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -31,6 +31,7 @@
  *******************************************************************************/
 
 // clang-format off
+#include <cstdint>
 #include <iss/arch/tgc5c.h>
 #include <iss/debugger/gdb_session.h>
 #include <iss/debugger/server.h>
@@ -43,6 +44,8 @@
 #include <exception>
 #include <vector>
 #include <sstream>
+#include <iss/instruction_decoder.h>
+
 
 #ifndef FMT_HEADER_ONLY
 #define FMT_HEADER_ONLY
@@ -93,7 +96,8 @@ protected:
     using compile_ret_t = virt_addr_t;
     using compile_func = compile_ret_t (this_class::*)(virt_addr_t &pc, code_word_t instr);
 
-    inline const char *name(size_t index){return index<traits::reg_aliases.size()?traits::reg_aliases[index]:"illegal";}
+    inline const char *name(size_t index){return traits::reg_aliases.at(index);}
+
 
     virt_addr_t execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit) override;
 
@@ -102,7 +106,6 @@ protected:
     inline void raise(uint16_t trap_id, uint16_t cause){
         auto trap_val =  0x80ULL << 24 | (cause << 16) | trap_id;
         this->core.reg.trap_state = trap_val;
-        this->template get_reg<uint32_t>(traits::NEXT_PC) = std::numeric_limits<uint32_t>::max();
     }
 
     inline void leave(unsigned lvl){
@@ -113,7 +116,12 @@ protected:
         this->core.wait_until(type);
     }
 
+    inline void set_tval(uint64_t new_tval){
+        tval = new_tval;
+    }
+
     uint64_t fetch_count{0};
+    uint64_t tval{0};
 
     using yield_t = boost::coroutines2::coroutine<void>::push_type;
     using coro_t = boost::coroutines2::coroutine<void>::pull_type;
@@ -142,20 +150,12 @@ private:
      * start opcode definitions
      ****************************************************************************/
     struct instruction_descriptor {
-        size_t length;
+        uint32_t length;
         uint32_t value;
         uint32_t mask;
         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){}
-    };
 
-    decoding_tree_node* root {nullptr};
     const std::array<instruction_descriptor, 87> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */
         {32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::LUI},
@@ -247,6 +247,9 @@ private:
         {16, 0b0000000000000000, 0b1111111111111111, arch::traits<ARCH>::opcode_e::DII},
     }};
 
+    //needs to be declared after instr_descr
+    decoder instr_decoder;
+
     iss::status fetch_ins(virt_addr_t pc, uint8_t * data){
         if(this->core.has_mmu()) {
             auto phys_pc = this->core.virt2phys(pc);
@@ -266,67 +269,12 @@ private:
         }
         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) {
     volatile CODE_WORD x = insn;
     insn = 2 * x;
 }
-
-template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
-
 // according to
 // https://stackoverflow.com/questions/8871204/count-number-of-1s-in-binary-representation
 #ifdef __GCC__
@@ -343,13 +291,16 @@ 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) {
-    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);
+: vm_base<ARCH>(core, core_id, cluster_id)
+, instr_decoder([this]() {
+        std::vector<generic_instruction_descriptor> g_instr_descr;
+        g_instr_descr.reserve(instr_descr.size());
+        for (uint32_t i = 0; i < instr_descr.size(); ++i) {
+            generic_instruction_descriptor new_instr_descr {instr_descr[i].value, instr_descr[i].mask, i};
+            g_instr_descr.push_back(new_instr_descr);
         }
+        return std::move(g_instr_descr);
+    }()) {}
 
 inline bool is_icount_limit_enabled(finish_cond_e cond){
     return (cond & finish_cond_e::ICOUNT_LIMIT) == finish_cond_e::ICOUNT_LIMIT;
@@ -379,14 +330,22 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
     while(!this->core.should_stop() &&
             !(is_icount_limit_enabled(cond) && icount >= count_limit) &&
             !(is_fcount_limit_enabled(cond) && fetch_count >= count_limit)){
-        fetch_count++;
+        if(this->debugging_enabled())
+            this->tgt_adapter->check_continue(*PC);
+        pc.val=*PC;
         if(fetch_ins(pc, data)!=iss::Ok){
-            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);
+            if(this->sync_exec && PRE_SYNC) this->do_sync(PRE_SYNC, std::numeric_limits<unsigned>::max());
+            process_spawn_blocks();
+            if(this->sync_exec && POST_SYNC) this->do_sync(PRE_SYNC, std::numeric_limits<unsigned>::max());
+            pc.val = super::core.enter_trap(arch::traits<ARCH>::RV_CAUSE_FETCH_ACCESS<<16, pc.val, 0);
         } 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 = decode_instr(root, instr);
+            uint32_t inst_index = instr_decoder.decode_instr(instr);
+            opcode_e inst_id = arch::traits<ARCH>::opcode_e::MAX_OPCODE;;
+            if(inst_index <instr_descr.size())
+                inst_id = instr_descr[inst_index].op;
+
             // 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));
@@ -463,14 +422,16 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                         raise(0, traits::RV_CAUSE_ILLEGAL_INSTRUCTION);
                                     }
                                     else {
-                                        if(imm % traits::INSTR_ALIGNMENT) {
+                                        uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int32_t)sext<21>(imm) ));
+                                        if(new_pc % traits::INSTR_ALIGNMENT) {
+                                            set_tval(new_pc);
                                             raise(0, 0);
                                         }
                                         else {
                                             if(rd != 0) {
                                                 *(X+rd) = (uint32_t)((uint64_t)(*PC ) + (uint64_t)(4 ));
                                             }
-                                            *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int32_t)sext<21>(imm) ));
+                                            *NEXT_PC = new_pc;
                                             this->core.reg.last_branch = 1;
                                         }
                                     }
@@ -500,6 +461,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                         uint32_t addr_mask = (uint32_t)- 2;
                                         uint32_t new_pc = (uint32_t)(((uint64_t)(*(X+rs1) ) + (uint64_t)((int16_t)sext<12>(imm) )) & (int64_t)(addr_mask ));
                                         if(new_pc % traits::INSTR_ALIGNMENT) {
+                                            set_tval(new_pc);
                                             raise(0, 0);
                                         }
                                         else {
@@ -534,11 +496,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if(*(X+rs1) == *(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -567,11 +531,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if(*(X+rs1) != *(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -600,11 +566,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if((int32_t)*(X+rs1) < (int32_t)*(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -633,11 +601,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if((int32_t)*(X+rs1) >= (int32_t)*(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -666,11 +636,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if(*(X+rs1) < *(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -699,11 +671,13 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                     }
                                     else {
                                         if(*(X+rs1) >= *(X+rs2)) {
-                                            if((uint32_t)(imm ) % traits::INSTR_ALIGNMENT) {
+                                            uint32_t new_pc = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                            if(new_pc % traits::INSTR_ALIGNMENT) {
+                                                set_tval(new_pc);
                                                 raise(0, 0);
                                             }
                                             else {
-                                                *NEXT_PC = (uint32_t)((uint64_t)(*PC ) + (uint64_t)((int16_t)sext<13>(imm) ));
+                                                *NEXT_PC = new_pc;
                                                 this->core.reg.last_branch = 1;
                                             }
                                         }
@@ -1485,7 +1459,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::ECALL: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "ecall");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "ecall";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 4;
@@ -1498,7 +1474,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::EBREAK: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "ebreak");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "ebreak";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 4;
@@ -1511,7 +1489,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::MRET: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "mret");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "mret";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 4;
@@ -1524,7 +1504,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::WFI: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "wfi");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "wfi";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 4;
@@ -1748,7 +1730,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
                         auto mnemonic = fmt::format(
-                            "{mnemonic:10} {rs1}, {rd}, {imm}", fmt::arg("mnemonic", "fence.i"),
+                            "{mnemonic:10} {rs1}, {rd}, {imm}", fmt::arg("mnemonic", "fence_i"),
                             fmt::arg("rs1", name(rs1)), fmt::arg("rd", name(rd)), fmt::arg("imm", imm));
                         this->core.disass_output(pc.val, mnemonic);
                     }
@@ -2122,7 +2104,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                     uint8_t nzimm = ((bit_sub<2,5>(instr)) | (bit_sub<12,1>(instr) << 5));
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "c.nop");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "c.nop";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 2;
@@ -2228,7 +2212,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                     uint8_t rd = ((bit_sub<7,5>(instr)));
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, ".reserved_clui");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = ".reserved_clui";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 2;
@@ -2534,7 +2520,8 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                     // execute instruction
                     {
                                     if(rs1 && rs1 < traits::RFS) {
-                                        *NEXT_PC = *(X+(uint32_t)(rs1 ) % traits::RFS) & (uint32_t)(~ 1 );
+                                        uint32_t addr_mask = (uint32_t)- 2;
+                                        *NEXT_PC = *(X+(uint32_t)(rs1 ) % traits::RFS) & addr_mask;
                                         this->core.reg.last_branch = 1;
                                     }
                                     else {
@@ -2546,7 +2533,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::__reserved_cmv: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, ".reserved_cmv");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = ".reserved_cmv";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 2;
@@ -2600,9 +2589,10 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                                         raise(0, traits::RV_CAUSE_ILLEGAL_INSTRUCTION);
                                     }
                                     else {
+                                        uint32_t addr_mask = (uint32_t)- 2;
                                         uint32_t new_pc = *(X+rs1);
                                         *(X+1) = (uint32_t)((uint64_t)(*PC ) + (uint64_t)(2 ));
-                                        *NEXT_PC = new_pc & (uint32_t)(~ 1 );
+                                        *NEXT_PC = new_pc & addr_mask;
                                         this->core.reg.last_branch = 1;
                                     }
                                 }
@@ -2611,7 +2601,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::C__EBREAK: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "c.ebreak");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "c.ebreak";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 2;
@@ -2650,7 +2642,9 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
                 case arch::traits<ARCH>::opcode_e::DII: {
                     if(this->disass_enabled){
                         /* generate console output when executing the command */
-                        this->core.disass_output(pc.val, "dii");
+                        //No disass specified, using instruction name
+                        std::string mnemonic = "dii";
+                        this->core.disass_output(pc.val, mnemonic);
                     }
                     // used registers// calculate next pc value
                     *NEXT_PC = *PC + 2;
@@ -2673,16 +2667,18 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
             //    this->core.reg.trap_state =  this->core.reg.pending_trap;
             // trap check
             if(trap_state!=0){
-                super::core.enter_trap(trap_state, pc.val, instr);
+                //In case of Instruction address misaligned (cause = 0 and trapid = 0) need the targeted addr (in tval)
+                auto mcause = (trap_state>>16) & 0xff; 
+                super::core.enter_trap(trap_state, pc.val, mcause ? instr:tval);
             } else {
                 icount++;
                 instret++;
             }
-            cycle++;
-            pc.val=*NEXT_PC;
-            this->core.reg.PC = this->core.reg.NEXT_PC;
+            *PC = *NEXT_PC;
             this->core.reg.trap_state =  this->core.reg.pending_trap;
         }
+        fetch_count++;
+        cycle++;
     }
     return pc;
 }