merge commit

CMakeLists.txt

@@ -31,7 +31,11 @@ if(WITH_LLVM)
 		src/vm/llvm/fp_impl.cpp
 	)
 endif()
-
+if(WITH_ASMJIT)
+	list(APPEND LIB_SOURCES
+	   src/vm/asmjit/vm_tgc5c.cpp
+    )
+endif()
 # library files
 FILE(GLOB GEN_ISS_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src-gen/iss/arch/*.cpp)
 FILE(GLOB GEN_VM_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src-gen/vm/interp/vm_*.cpp)
@@ -53,7 +57,10 @@ if(WITH_TCC)
 	FILE(GLOB TCC_GEN_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src-gen/vm/tcc/vm_*.cpp)
 	list(APPEND LIB_SOURCES ${TCC_GEN_SOURCES})
 endif()
-
+if(WITH_ASMJIT)
+	FILE(GLOB TCC_GEN_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src-gen/vm/asmjit/vm_*.cpp)
+	list(APPEND LIB_SOURCES ${TCC_GEN_SOURCES})
+endif()
 if(TARGET yaml-cpp::yaml-cpp)
     list(APPEND LIB_SOURCES 
     	src/iss/plugin/cycle_estimate.cpp
@@ -182,6 +189,10 @@ if(BUILD_TESTING)
 	add_test(NAME tgc-sim-llvm
 	         COMMAND tgc-sim -f ${CMAKE_BINARY_DIR}/../../Firmwares/hello-world/hello --backend llvm)
 	endif()
+    if(WITH_ASMJIT)
+	add_test(NAME tgc-sim-asmjit
+	         COMMAND tgc-sim -f ${CMAKE_BINARY_DIR}/../../Firmwares/hello-world/hello --backend asmjit)
+	endif()
 endif()
 ###############################################################################
 #

gen_input/templates/asmjit/CORENAME.cpp.gtl

@@ -0,0 +1,280 @@
+/*******************************************************************************
+ * Copyright (C) 2017, 2023 MINRES Technologies GmbH
+ * 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 copyright holder 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 COPYRIGHT HOLDERS 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 COPYRIGHT HOLDER 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 <iss/arch/${coreDef.name.toLowerCase()}.h>
+#include <iss/debugger/gdb_session.h>
+#include <iss/debugger/server.h>
+#include <iss/iss.h>
+#include <iss/asmjit/vm_base.h>
+#include <asmjit/asmjit.h>
+#include <util/logging.h>
+
+#ifndef FMT_HEADER_ONLY
+#define FMT_HEADER_ONLY
+#endif
+#include <fmt/format.h>
+
+#include <array>
+#include <iss/debugger/riscv_target_adapter.h>
+
+namespace iss {
+namespace asmjit {
+
+
+namespace ${coreDef.name.toLowerCase()} {
+using namespace ::asmjit;
+using namespace iss::arch;
+using namespace iss::debugger;
+
+template <typename ARCH> class vm_impl : public iss::asmjit::vm_base<ARCH> {
+public:
+    using traits = arch::traits<ARCH>;
+    using super = typename iss::asmjit::vm_base<ARCH>;
+    using virt_addr_t = typename super::virt_addr_t;
+    using phys_addr_t = typename super::phys_addr_t;
+    using code_word_t = typename super::code_word_t;
+    using mem_type_e = typename super::mem_type_e;
+    using addr_t = typename super::addr_t;
+
+    vm_impl();
+
+    vm_impl(ARCH &core, unsigned core_id = 0, unsigned cluster_id = 0);
+
+    void enableDebug(bool enable) { super::sync_exec = super::ALL_SYNC; }
+
+    target_adapter_if *accquire_target_adapter(server_if *srv) override {
+        debugger_if::dbg_enabled = true;
+        if (vm_base<ARCH>::tgt_adapter == nullptr)
+            vm_base<ARCH>::tgt_adapter = new riscv_target_adapter<ARCH>(srv, this->get_arch());
+        return vm_base<ARCH>::tgt_adapter;
+    }
+
+protected:
+    using vm_base<ARCH>::get_reg_ptr;
+    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;
+    inline const char *name(size_t index){return traits::reg_aliases.at(index);}
+
+    template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
+    inline S sext(U from) {
+        auto mask = (1ULL<<W) - 1;
+        auto sign_mask = 1ULL<<(W-1);
+        return (from & mask) | ((from & sign_mask) ? ~mask : 0);
+    } 
+#include "helper_func.h"
+
+private:
+    /****************************************************************************
+     * start opcode definitions
+     ****************************************************************************/
+    struct instruction_descriptor {
+        size_t length;
+        uint32_t value;
+        uint32_t mask;
+        compile_func op;
+    };
+    struct decoding_tree_node{
+        std::vector<instruction_descriptor> instrs;
+        std::vector<decoding_tree_node*> children;
+        uint32_t submask = std::numeric_limits<uint32_t>::max();
+        uint32_t value;
+        decoding_tree_node(uint32_t value) : value(value){}
+    };
+
+    decoding_tree_node* root {nullptr};
+
+    const std::array<instruction_descriptor, ${instructions.size}> instr_descr = {{
+         /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
+        /* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */
+        {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
+    }};
+ 
+    /* 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){
+        uint64_t PC = pc.val;
+        <%instr.fields.eachLine{%>${it}
+        <%}%>if(this->disass_enabled){
+            /* generate disass */
+        }
+        x86::Compiler& cc = jh.cc;
+        //ideally only do this if necessary (someone / plugin needs it)
+        cc.mov(jh.pc,PC);
+        cc.comment(fmt::format("\\n${instr.name}_{:#x}:",pc.val).c_str());
+        this->gen_sync(jh, PRE_SYNC, ${idx});
+        pc=pc+ ${instr.length/8};
+       
+        gen_instr_prologue(jh, pc.val);
+        cc.comment("\\n//behavior:");
+        /*generate behavior*/
+        <%instr.behavior.eachLine{%>${it}
+        <%}%>
+        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 ) {
+
+        return BRANCH;
+    }    
+    //decoding functionality
+
+    void populate_decoding_tree(decoding_tree_node* root){
+        //create submask
+        for(auto instr: root->instrs){
+            root->submask &= instr.mask;
+        }
+        //put each instr according to submask&encoding into children
+        for(auto instr: root->instrs){
+            bool foundMatch = false;
+            for(auto child: root->children){
+                //use value as identifying trait
+                if(child->value == (instr.value&root->submask)){
+                    child->instrs.push_back(instr);
+                    foundMatch = true;
+                }
+            }
+            if(!foundMatch){
+                decoding_tree_node* child = new decoding_tree_node(instr.value&root->submask);
+                child->instrs.push_back(instr);
+                root->children.push_back(child);
+            }
+        }
+        root->instrs.clear();
+        //call populate_decoding_tree for all children
+        if(root->children.size() >1)
+            for(auto child: root->children){
+                populate_decoding_tree(child);      
+            }
+        else{
+            //sort instrs by value of the mask, this works bc we want to have the least restrictive one last
+            std::sort(root->children[0]->instrs.begin(), root->children[0]->instrs.end(), [](const instruction_descriptor& instr1, const instruction_descriptor& instr2) {
+            return instr1.mask > instr2.mask;
+            }); 
+        }
+    }
+    compile_func decode_instr(decoding_tree_node* node, code_word_t word){
+        if(!node->children.size()){
+            if(node->instrs.size() == 1) return node->instrs[0].op;
+            for(auto instr : node->instrs){
+                if((instr.mask&word) == instr.value) return instr.op;
+            }
+        }
+        else{
+            for(auto child : node->children){
+                if (child->value == (node->submask&word)){
+                    return decode_instr(child, word);
+                }  
+            }  
+        }
+        return nullptr;
+    }
+};
+
+template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
+    volatile CODE_WORD x = instr;
+    instr = 2 * x;
+}
+
+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);
+}
+
+template <typename ARCH>
+continuation_e
+vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, jit_holder& jh) {
+    enum {TRAP_ID=1<<16};
+    code_word_t instr = 0;
+    phys_addr_t paddr(pc);
+    auto *const data = (uint8_t *)&instr;
+    if(this->core.has_mmu())
+        paddr = this->core.virt2phys(pc);
+    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'
+    ++inst_cnt;
+    auto f = decode_instr(root, instr);
+    if (f == nullptr) 
+        f = &this_class::illegal_intruction;
+    return (this->*f)(pc, instr, jh);
+}
+
+
+
+} // namespace ${coreDef.name.toLowerCase()}
+
+template <>
+std::unique_ptr<vm_if> create<arch::${coreDef.name.toLowerCase()}>(arch::${coreDef.name.toLowerCase()} *core, unsigned short port, bool dump) {
+    auto ret = new ${coreDef.name.toLowerCase()}::vm_impl<arch::${coreDef.name.toLowerCase()}>(*core, dump);
+    if (port != 0) debugger::server<debugger::gdb_session>::run_server(ret, port);
+    return std::unique_ptr<vm_if>(ret);
+}
+} // namespace asmjit
+} // namespace iss
+
+#include <iss/factory.h>
+#include <iss/arch/riscv_hart_m_p.h>
+#include <iss/arch/riscv_hart_mu_p.h>
+namespace iss {
+namespace {
+volatile std::array<bool, 2> dummy = {
+        core_factory::instance().register_creator("${coreDef.name.toLowerCase()}|m_p|asmjit", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_m_p<iss::arch::${coreDef.name.toLowerCase()}>();
+            auto* vm = new asmjit::${coreDef.name.toLowerCase()}::vm_impl<arch::${coreDef.name.toLowerCase()}>(*cpu, false);
+            if (port != 0) debugger::server<debugger::gdb_session>::run_server(vm, port);
+            return {cpu_ptr{cpu}, vm_ptr{vm}};
+        }),
+        core_factory::instance().register_creator("${coreDef.name.toLowerCase()}|mu_p|asmjit", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_mu_p<iss::arch::${coreDef.name.toLowerCase()}>();
+            auto* vm = new asmjit::${coreDef.name.toLowerCase()}::vm_impl<arch::${coreDef.name.toLowerCase()}>(*cpu, false);
+            if (port != 0) debugger::server<debugger::gdb_session>::run_server(vm, port);
+            return {cpu_ptr{cpu}, vm_ptr{vm}};
+        })
+};
+}
+}

src/vm/asmjit/helper_func.h

@@ -0,0 +1,474 @@
+
+        
+x86::Mem get_reg_ptr(jit_holder& jh, unsigned idx){ 
+   
+        x86::Gp tmp_ptr = jh.cc.newUIntPtr("tmp_ptr");
+        jh.cc.mov(tmp_ptr, jh.regs_base_ptr);
+        jh.cc.add(tmp_ptr, traits::reg_byte_offsets[idx]);
+        switch(traits::reg_bit_widths[idx]){
+        case 8:
+            return x86::ptr_8(tmp_ptr);
+        case 16:
+            return x86::ptr_16(tmp_ptr);
+        case 32:
+            return x86::ptr_32(tmp_ptr);
+        case 64:
+            return x86::ptr_64(tmp_ptr);
+        default:
+            throw std::runtime_error("Invalid reg size in get_reg_ptr");
+        }
+}
+x86::Gp get_reg_for(jit_holder& jh, unsigned idx){
+    //can check for regs in jh and return them instead of creating new ones
+    switch(traits::reg_bit_widths[idx]){
+        case 8:
+            return jh.cc.newInt8();
+        case 16:
+            return jh.cc.newInt16();
+        case 32:
+            return jh.cc.newInt32();
+        case 64:
+            return jh.cc.newInt64();
+        default:
+            throw std::runtime_error("Invalid reg size in get_reg_ptr");
+        }
+}
+x86::Gp get_reg_for(jit_holder& jh, unsigned size, bool is_signed){
+    if(is_signed)
+        switch(size){
+            case 8:
+                return jh.cc.newInt8();
+            case 16:
+                return jh.cc.newInt16();
+            case 32:
+                return jh.cc.newInt32();
+            case 64:
+                return jh.cc.newInt64();
+            default:
+                throw std::runtime_error("Invalid reg size in get_reg_ptr");
+            }
+    else
+        switch(size){
+        case 8:
+            return jh.cc.newUInt8();
+        case 16:
+            return jh.cc.newUInt16();
+        case 32:
+            return jh.cc.newUInt32();
+        case 64:
+            return jh.cc.newUInt64();
+        default:
+            throw std::runtime_error("Invalid reg size in get_reg_ptr");
+        }
+}
+inline x86::Gp load_reg_from_mem(jit_holder& jh, unsigned idx){
+    auto ptr = get_reg_ptr(jh, idx);
+    auto reg = get_reg_for(jh, idx);
+    jh.cc.mov(reg, ptr);
+    return reg;
+}
+inline void write_reg_to_mem(jit_holder& jh, x86::Gp reg, unsigned idx){
+    auto ptr = get_reg_ptr(jh, idx);
+    jh.cc.mov(ptr, reg);
+}
+
+void gen_instr_prologue(jit_holder& jh, addr_t pc){
+    auto& cc = jh.cc;
+
+    cc.comment("\n//(*icount)++;");
+    cc.inc(get_reg_ptr(jh, traits::ICOUNT));
+
+    cc.comment("\n//*pc=*next_pc;");
+    cc.mov(get_reg_ptr(jh, traits::PC), jh.next_pc);
+
+    cc.comment("\n//*trap_state=*pending_trap;");
+    cc.mov(get_reg_ptr(jh, traits::PENDING_TRAP), jh.trap_state);
+    
+    cc.comment("\n//increment *next_pc");
+    cc.mov(jh.next_pc, pc);
+
+}
+void gen_instr_epilogue(jit_holder& jh){
+    auto& cc = jh.cc;
+   
+    cc.comment("\n//if(*trap_state!=0) goto trap_entry;");
+    cc.test(jh.trap_state, jh.trap_state);
+    cc.jnz(jh.trap_entry);
+
+    //Does this need to be done after every single instruction?
+    cc.comment("\n//write back regs to mem");
+    write_reg_to_mem(jh, jh.pc, traits::PC);
+    write_reg_to_mem(jh, jh.next_pc, traits::NEXT_PC);
+    write_reg_to_mem(jh, jh.trap_state, traits::TRAP_STATE);
+
+  
+}
+void gen_block_prologue(jit_holder& jh) override{
+
+    jh.pc = load_reg_from_mem(jh, traits::PC);
+    jh.next_pc = load_reg_from_mem(jh, traits::NEXT_PC);
+    jh.trap_state = load_reg_from_mem(jh, traits::TRAP_STATE);
+}
+void gen_block_epilogue(jit_holder& jh) override{
+    x86::Compiler& cc = jh.cc;
+    cc.comment("\n//return *next_pc;");
+    cc.ret(jh.next_pc);
+
+    cc.bind(jh.trap_entry);
+    cc.comment("\n//enter_trap(core_ptr, *trap_state, *pc, 0);");
+
+    x86::Gp current_trap_state =  get_reg_for(jh, traits::TRAP_STATE);
+    cc.mov(current_trap_state, get_reg_ptr(jh, traits::TRAP_STATE));
+
+    x86::Gp current_pc = get_reg_for(jh, traits::PC);
+    cc.mov(current_pc,  get_reg_ptr(jh, traits::PC));
+
+    x86::Gp instr = cc.newInt32("instr");
+    cc.mov(instr, 0);
+    InvokeNode* call_enter_trap;
+    cc.invoke(&call_enter_trap, &enter_trap, FuncSignatureT<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);
+    cc.comment("\n//*last_branch = std::numeric_limits<uint32_t>::max();");
+    cc.mov(get_reg_ptr(jh,traits::LAST_BRANCH), std::numeric_limits<uint32_t>::max());
+    cc.comment("\n//return *next_pc;");
+    cc.ret(jh.next_pc);
+
+}
+//TODO implement
+
+void gen_raise(jit_holder& jh, uint16_t trap_id, uint16_t cause) {
+    jh.cc.comment("//gen_raise");
+}
+void gen_wait(jit_holder& jh, unsigned type) {
+    jh.cc.comment("//gen_wait");
+}
+void gen_leave(jit_holder& jh, unsigned lvl){
+    jh.cc.comment("//gen_leave");
+}
+
+enum operation {add, sub, band, bor, bxor, shl, sar , shr};
+
+template <typename T, typename = std::enable_if_t<std::is_integral<T>::value || std::is_same<T, x86::Gp>::value>>
+x86::Gp gen_operation(jit_holder& jh, operation op, x86::Gp a, T b){
+    x86::Compiler& cc = jh.cc;
+    switch (op) {
+        case add: { cc.add(a, b); break; }
+        case sub: { cc.sub(a, b); break; }
+        case band: { cc.and_(a, b); break; }
+        case bor: { cc.or_(a, b); break; }
+        case bxor: { cc.xor_(a, b); break; }
+        case shl: { cc.shl(a, b); break; }
+        case sar: { cc.sar(a, b); break; }
+        case shr: { cc.shr(a, b); break; }
+        default: throw std::runtime_error(fmt::format("Current operation {} not supported in gen_operation (operation)", op));
+    }
+    return a;
+}
+
+enum three_operand_operation{imul, mul, idiv, div, srem, urem};
+
+x86::Gp gen_operation(jit_holder& jh, three_operand_operation op, x86::Gp a, x86::Gp b){
+    x86::Compiler& cc = jh.cc;
+    switch (op) {
+        case imul: {            
+            x86::Gp dummy = cc.newInt64();
+            cc.imul(dummy, a.r64(), b.r64());
+            return a;
+        }
+        case mul: {            
+            x86::Gp dummy = cc.newInt64();
+            cc.mul(dummy, a.r64(), b.r64());
+            return a;
+        }
+        case idiv: {
+            x86::Gp dummy = cc.newInt64();
+            cc.mov(dummy, 0);
+            cc.idiv(dummy, a.r64(), b.r64());
+            return a;
+        }
+        case div: {
+            x86::Gp dummy = cc.newInt64();
+            cc.mov(dummy, 0);
+            cc.div(dummy, a.r64(), b.r64());
+            return a;
+        }
+        case srem:{
+            x86::Gp rem = cc.newInt32();
+            cc.mov(rem, 0);
+            auto a_reg = cc.newInt32();
+            cc.mov(a_reg, a.r32());
+            cc.idiv(rem, a_reg, b.r32());
+            return rem;
+        }  
+        case urem:{
+            x86::Gp rem = cc.newInt32();
+            cc.mov(rem, 0);
+            auto a_reg = cc.newInt32();
+            cc.mov(a_reg, a.r32());
+            cc.div(rem, a_reg, b.r32());
+            return rem;
+        }         
+
+        default: throw std::runtime_error(fmt::format("Current operation {} not supported in gen_operation (three_operand)", op));
+    }
+    return a;
+}
+template <typename T, typename = std::enable_if_t<std::is_integral<T>::value>>
+x86::Gp gen_operation(jit_holder& jh, three_operand_operation op, x86::Gp a, T b){
+    x86::Gp b_reg = jh.cc.newInt32();
+/*     switch(a.size()){
+        case 1: b_reg = jh.cc.newInt8(); break;
+        case 2: b_reg = jh.cc.newInt16(); break;
+        case 4: b_reg = jh.cc.newInt32(); break;
+        case 8: b_reg = jh.cc.newInt64(); break;
+        default: throw std::runtime_error(fmt::format("Invalid size ({}) in gen operation", a.size()));
+    } */
+    jh.cc.mov(b_reg, b);
+    return gen_operation(jh, op, a, b_reg);
+}
+enum comparison_operation{land, lor, eq, ne, lt, ltu, gt, gtu, lte, lteu, gte, gteu};
+
+template <typename T, typename = std::enable_if_t<std::is_integral<T>::value || std::is_same<T, x86::Gp>::value>>
+x86::Gp gen_operation(jit_holder& jh, comparison_operation op, x86::Gp a, T b){
+    x86::Compiler& cc = jh.cc;
+    x86::Gp tmp = cc.newInt8();
+    cc.mov(tmp,1);
+    Label label_then = cc.newLabel();
+    cc.cmp(a,b);
+    switch (op) {
+    case eq: cc.je(label_then); break;
+    case ne: cc.jne(label_then); break;
+    case lt: cc.jl(label_then); break;
+    case ltu: cc.jb(label_then); break;
+    case gt: cc.jg(label_then); break;
+    case gtu: cc.ja(label_then); break;
+    case lte: cc.jle(label_then); break;
+    case lteu: cc.jbe(label_then); break;
+    case gte: cc.jge(label_then); break;
+    case gteu: cc.jae(label_then); break;
+    case land: {
+        Label label_false = cc.newLabel();
+        cc.cmp(a, 0);
+        cc.je(label_false);
+        auto b_reg = cc.newInt8();
+        cc.mov(b_reg, b);
+        cc.cmp(b_reg, 0);
+        cc.je(label_false);
+        cc.jmp(label_then);
+        cc.bind(label_false);
+        break;
+    }
+    case lor: {
+        cc.cmp(a, 0);
+        cc.jne(label_then);
+        auto b_reg = cc.newInt8();
+        cc.mov(b_reg, b);
+        cc.cmp(b_reg, 0);
+        cc.jne(label_then);   
+        break;
+    }
+    default: throw std::runtime_error(fmt::format("Current operation {} not supported in gen_operation (comparison)", op));
+    }
+    cc.mov(tmp,0);
+    cc.bind(label_then);
+    return tmp;
+}
+enum binary_operation{lnot, inc, dec, bnot, neg};
+
+x86::Gp gen_operation(jit_holder& jh, binary_operation op, x86::Gp a){
+    x86::Compiler& cc = jh.cc;
+    switch (op) {
+        case lnot: throw std::runtime_error("Current operation not supported in gen_operation(lnot)");
+        case inc: { cc.inc(a); break; }
+        case dec: { cc.dec(a); break; }
+        case bnot: { cc.not_(a); break; }
+        case neg: { cc.neg(a); break; }
+        default: throw std::runtime_error(fmt::format("Current operation {} not supported in gen_operation (unary)", op));
+    }
+    return a;
+}
+
+/* template <typename T>
+inline typename std::enable_if_t<std::is_unsigned<T>::value, x86::Gp> gen_ext(jit_holder& jh, T val, unsigned size, bool is_signed) const {
+    auto val_reg = get_reg_for(jh, sizeof(val)*8);
+    auto tmp = get_reg_for(jh, size);
+    jh.cc.mov(val_reg, val);
+    if(is_signed) jh.cc.movsx(tmp, val_reg);
+    else jh.cc.movzx(tmp,val_reg);
+    return tmp;
+}
+
+template <typename T>
+inline typename std::enable_if_t<std::is_signed<T>::value, x86::Gp> gen_ext(jit_holder& jh, T val, unsigned size, bool is_signed) const {
+    auto val_reg = get_reg_for(jh, sizeof(val)*8);
+    auto tmp = get_reg_for(jh, size);
+    jh.cc.mov(val_reg, val);
+    if(is_signed) jh.cc.movsx(tmp, val_reg);
+    else jh.cc.movzx(tmp,val_reg);
+    return tmp;
+} */
+template <typename T, typename = std::enable_if_t<std::is_integral<T>::value>>
+inline x86::Gp gen_ext(jit_holder& jh, T val, unsigned size, bool is_signed) {
+    auto val_reg = get_reg_for(jh, sizeof(val)*8);
+    jh.cc.mov(val_reg, val);
+    return gen_ext(jh, val_reg, size, is_signed);
+}
+//explicit Gp size cast
+inline x86::Gp gen_ext(jit_holder& jh, x86::Gp val, unsigned size, bool is_signed) {
+    auto& cc = jh.cc;
+    if(is_signed){
+        switch(val.size()){
+            case 1:    cc.cbw(val); break;
+            case 2:    cc.cwde(val); break;
+            case 4:    cc.cdqe(val); break;
+            case 8:    break;
+            default: throw std::runtime_error("Invalid register size in gen_ext");
+        }
+    }
+    switch(size){
+        case 8:  cc.and_(val,std::numeric_limits<uint8_t>::max()); return val.r8();
+        case 16: cc.and_(val,std::numeric_limits<uint16_t>::max()); return val.r16();
+        case 32: cc.and_(val,std::numeric_limits<uint32_t>::max()); return val.r32();
+        case 64: cc.and_(val,std::numeric_limits<uint64_t>::max()); return val.r64();
+        case 128: return val.r64();
+        default: throw std::runtime_error("Invalid size in gen_ext");
+    }
+}
+
+inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, uint32_t length){
+    x86::Compiler& cc = jh.cc;
+    auto ret_reg = cc.newInt32();
+
+    auto mem_type_reg = cc.newInt32();
+    cc.mov(mem_type_reg, type);
+
+    auto space_reg = cc.newInt32();
+    cc.mov(space_reg, static_cast<uint16_t>(iss::address_type::VIRTUAL));
+    
+    auto val_ptr = cc.newUIntPtr();
+    cc.mov(val_ptr, read_mem_buf);
+
+    InvokeNode* invokeNode;
+    uint64_t mask = 0;
+    x86::Gp val_reg = cc.newInt64();
+
+    switch(length){
+    case 1:{
+        cc.invoke(&invokeNode, &read_mem1, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uintptr_t>());
+        mask = std::numeric_limits<uint8_t>::max();
+        break;
+    }
+    case 2:{
+        cc.invoke(&invokeNode, &read_mem2, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uintptr_t>());
+        mask = std::numeric_limits<uint16_t>::max();
+        break;
+    }
+    case 4:{
+        cc.invoke(&invokeNode, &read_mem4, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uintptr_t>());
+        mask = std::numeric_limits<uint32_t>::max();
+        break;
+    }
+    case 8:{
+        cc.invoke(&invokeNode, &read_mem8, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uintptr_t>());
+        mask = std::numeric_limits<uint64_t>::max();
+        break;
+    }
+    default: throw std::runtime_error(fmt::format("Invalid length ({}) in gen_read_mem",length));
+    }
+
+    invokeNode->setRet(0, ret_reg);
+    invokeNode->setArg(0, jh.arch_if_ptr);
+    invokeNode->setArg(1, space_reg);
+    invokeNode->setArg(2, mem_type_reg);
+    invokeNode->setArg(3, addr);
+    invokeNode->setArg(4, val_ptr);
+
+    cc.mov(val_reg, x86::ptr_64(val_ptr));
+    cc.and_(val_reg, mask);
+    cc.cmp(ret_reg,0);
+    cc.jne(jh.trap_entry);
+    return val_reg;
+}
+inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp length){
+    uint32_t length_val = 0;
+    auto length_ptr = jh.cc.newIntPtr();
+    jh.cc.mov(length_ptr, &length_val);
+    jh.cc.mov(x86::ptr_32(length_ptr),length);
+
+    return gen_read_mem(jh, type, addr, length);
+}
+inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, uint64_t addr, x86::Gp length){
+    auto addr_reg = jh.cc.newInt64();
+    jh.cc.mov(addr_reg, addr);    
+
+    uint32_t length_val = 0;
+    auto length_ptr = jh.cc.newIntPtr();
+    jh.cc.mov(length_ptr, &length_val);
+    jh.cc.mov(x86::ptr_32(length_ptr),length);
+
+    return gen_read_mem(jh, type, addr_reg, length_val);
+}
+inline x86::Gp gen_read_mem(jit_holder& jh, mem_type_e type, uint64_t addr, uint32_t length){
+    auto addr_reg = jh.cc.newInt64();
+    jh.cc.mov(addr_reg, addr);
+
+    return gen_read_mem(jh, type, addr_reg, length);
+}
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, int64_t val){
+    auto val_reg = jh.cc.newInt64();
+    jh.cc.mov(val_reg, val);
+    gen_write_mem(jh, type, addr, val_reg);
+
+}
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, x86::Gp addr, x86::Gp val){
+    x86::Compiler& cc = jh.cc;
+
+    auto mem_type_reg = cc.newInt32();
+    jh.cc.mov(mem_type_reg, type);
+    auto space_reg = cc.newInt32();
+    jh.cc.mov(space_reg, static_cast<uint16_t>(iss::address_type::VIRTUAL));
+    auto ret_reg = cc.newInt32();
+    InvokeNode* invokeNode;
+
+    if(val.isGpb()){
+        cc.invoke(&invokeNode, &write_mem1, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint8_t>());
+    }
+    else if(val.isGpw()){
+        cc.invoke(&invokeNode, &write_mem2, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint16_t>());
+    }
+    else if(val.isGpd()){
+        cc.invoke(&invokeNode, &write_mem4, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint32_t>());
+    }
+    else if(val.isGpq()){
+        cc.invoke(&invokeNode, &write_mem8, FuncSignatureT<uint32_t, uint64_t, uint32_t, uint32_t, uint64_t, uint64_t>());
+    }
+    else throw std::runtime_error("Invalid register size in gen_write_mem");
+
+    invokeNode->setRet(0,ret_reg);
+    invokeNode->setArg(0, jh.arch_if_ptr);
+    invokeNode->setArg(1, space_reg);
+    invokeNode->setArg(2, mem_type_reg);
+    invokeNode->setArg(3, addr);
+    invokeNode->setArg(4, val);
+
+    cc.cmp(ret_reg,0);
+    cc.jne(jh.trap_entry);
+
+}
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, x86::Gp val){
+    auto addr_reg = jh.cc.newInt64();
+    jh.cc.mov(addr_reg, addr);
+    gen_write_mem(jh, type, addr_reg, val);
+}
+inline void gen_write_mem(jit_holder& jh, mem_type_e type, uint64_t addr, int64_t val){
+    auto val_reg = jh.cc.newInt64();
+    jh.cc.mov(val_reg, val);
+
+    auto addr_reg = jh.cc.newInt64();
+    jh.cc.mov(addr_reg, addr);
+    gen_write_mem(jh, type, addr_reg, val_reg);
+
+}