Merge branch 'develop' of

https://git.minres.com/DBT-RISE/DBT-RISE-TGC.git into develop

CMakeLists.txt

@@ -1,10 +1,12 @@
 cmake_minimum_required(VERSION 3.12)
+list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 ###############################################################################
 #
 ###############################################################################
 project(dbt-rise-tgc VERSION 1.0.0)
 
 include(GNUInstallDirs)
+include(flink)
 
 find_package(elfio QUIET)
 find_package(Boost COMPONENTS coroutine)
@@ -40,6 +42,9 @@ set(LIB_SOURCES
 if(WITH_TCC)
 	list(APPEND LIB_SOURCES src/vm/tcc/vm_tgc_c.cpp)
 endif()
+if(WITH_LLVM)
+	list(APPEND LIB_SOURCES src/vm/llvm/vm_tgc_c.cpp src/vm/llvm/fp_impl.cpp)
+endif()
 
 # library files
 FILE(GLOB GEN_ISS_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/src-gen/iss/arch/*.cpp)
@@ -67,10 +72,10 @@ if(TARGET RapidJSON OR TARGET RapidJSON::RapidJSON)
 endif()
 
 # Define the library
-add_library(${PROJECT_NAME} ${LIB_SOURCES})
+add_library(${PROJECT_NAME} SHARED ${LIB_SOURCES})
 
 if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
-     target_compile_options(${PROJECT_NAME} PRIVATE -Wno-shift-count-overflow)
+    target_compile_options(${PROJECT_NAME} PRIVATE -Wno-shift-count-overflow)
 elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
     target_compile_options(${PROJECT_NAME} PRIVATE /wd4293)
 endif()
@@ -82,11 +87,9 @@ if(TARGET jsoncpp::jsoncpp)
 else()
 	target_link_libraries(${PROJECT_NAME} PUBLIC jsoncpp)
 endif()
-if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU" AND BUILD_SHARED_LIBS)
+
-    target_link_libraries(${PROJECT_NAME} PUBLIC -Wl,--whole-archive dbt-rise-core -Wl,--no-whole-archive)
+target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-core)
-else()
+
-    target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-core)
-endif()
 if(TARGET elfio::elfio)
     target_link_libraries(${PROJECT_NAME} PUBLIC elfio::elfio)
 else()
@@ -101,6 +104,13 @@ if(TARGET RapidJSON::RapidJSON)
 elseif(TARGET RapidJSON)
     target_link_libraries(${PROJECT_NAME} PUBLIC RapidJSON)
 endif()
+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()
 
 
 set_target_properties(${PROJECT_NAME} PROPERTIES
@@ -154,7 +164,7 @@ if(WITH_TCC)
     target_compile_definitions(${PROJECT_NAME} PRIVATE WITH_TCC)
 endif()
 # Links the target exe against the libraries
-target_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-tgc)
+target_force_link_libraries(${PROJECT_NAME} PUBLIC dbt-rise-tgc)
 if(TARGET Boost::program_options)
     target_link_libraries(${PROJECT_NAME} PUBLIC Boost::program_options)
 else()

cmake/flink.cmake

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

gen_input/templates/interp/CORENAME.cpp.gtl

@@ -158,30 +158,81 @@ private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
-    struct InstructionDesriptor {
+    struct instruction_descriptor {
         size_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){}
+    };
 
-    const std::array<InstructionDesriptor, ${instructions.size}> instr_descr = {{
+    decoding_tree_node* root {nullptr};
+    const std::array<instruction_descriptor, ${instructions.size}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
         {${instr.length}, ${instr.encoding}, ${instr.mask}, arch::traits<ARCH>::opcode_e::${instr.instruction.name}},<%}%>
     }};
 
-    //static constexpr typename traits::addr_t upper_bits = ~traits::PGMASK;
     iss::status fetch_ins(virt_addr_t pc, uint8_t * data){
         auto phys_pc = this->core.v2p(pc);
-        //if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
-        //    if (this->core.read(phys_pc, 2, data) != iss::Ok) return iss::Err;
-        //    if ((data[0] & 0x3) == 0x3) // this is a 32bit instruction
-        //        if (this->core.read(this->core.v2p(pc + 2), 2, data + 2) != iss::Ok) return iss::Err;
-        //} else {
             if (this->core.read(phys_pc, 4, data) != iss::Ok)  return iss::Err;
-        //}
         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) {
@@ -208,16 +259,11 @@ 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) {
-    unsigned id=0;
+    root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
-    for (auto instr : instr_descr) {
+    for(auto instr:instr_descr){
-        auto quadrant = instr.value & 0x3;
+        root->instrs.push_back(instr);
-        qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op});
-    }
-    for(auto& lut: qlut){
-        std::sort(std::begin(lut), std::end(lut), [](instruction_pattern const& a, instruction_pattern const& b){
-            return bit_count(a.mask) > bit_count(b.mask);
-        });
     }
+    populate_decoding_tree(root);
 }
 
 inline bool is_count_limit_enabled(finish_cond_e cond){
@@ -228,14 +274,6 @@ inline bool is_jump_to_self_enabled(finish_cond_e cond){
     return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF;
 }
 
-template <typename ARCH>
-typename arch::traits<ARCH>::opcode_e vm_impl<ARCH>::decode_inst_id(code_word_t instr){
-    for(auto& e: qlut[instr&0x3]){
-        if(!((instr&e.mask) ^ e.value )) return e.id;
-    }
-    return arch::traits<ARCH>::opcode_e::MAX_OPCODE;
-}
-
 template <typename ARCH>
 typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit){
     auto pc=start;
@@ -257,7 +295,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
         } 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_inst_id(instr);
+            auto inst_id = decode_instr(root, instr);
             // 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));

gen_input/templates/llvm/CORENAME.cpp.gtl

@@ -30,10 +30,9 @@
  *
  *******************************************************************************/
 
+#include <iss/arch/${coreDef.name.toLowerCase()}.h>
 #include <iss/debugger/gdb_session.h>
 #include <iss/debugger/server.h>
-#include <iss/arch/${coreDef.name.toLowerCase()}.h>
-#include <iss/arch/riscv_hart_m_p.h>
 #include <iss/iss.h>
 #include <iss/llvm/vm_base.h>
 #include <util/logging.h>
@@ -111,7 +110,7 @@ protected:
     void gen_trap_check(BasicBlock *bb);
 
     inline Value *gen_reg_load(unsigned i, unsigned level = 0) {
-        return this->builder.CreateLoad(get_reg_ptr(i), false);
+        return this->builder.CreateLoad(this->get_typeptr(i), get_reg_ptr(i), false);
     }
 
     inline void gen_set_pc(virt_addr_t pc, unsigned reg_num) {
@@ -124,7 +123,7 @@ protected:
     // enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
     enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
     enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
-    enum { LUT_SIZE = 1 << util::bit_count(EXTR_MASK32), LUT_SIZE_C = 1 << util::bit_count(EXTR_MASK16) };
+    enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint64_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint64_t>(EXTR_MASK16)) };
 
     using this_class = vm_impl<ARCH>;
     using compile_func = std::tuple<continuation_e, BasicBlock *> (this_class::*)(virt_addr_t &pc,
@@ -204,10 +203,10 @@ private:
      ****************************************************************************/
     std::tuple<continuation_e, BasicBlock *> illegal_intruction(virt_addr_t &pc, code_word_t instr, BasicBlock *bb) {
 		this->gen_sync(iss::PRE_SYNC, instr_descr.size());
-        this->builder.CreateStore(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::NEXT_PC), true),
+        this->builder.CreateStore(this->builder.CreateLoad(this->get_typeptr(traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), true),
                                    get_reg_ptr(traits<ARCH>::PC), true);
         this->builder.CreateStore(
-            this->builder.CreateAdd(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::ICOUNT), true),
+            this->builder.CreateAdd(this->builder.CreateLoad(this->get_typeptr(traits<ARCH>::ICOUNT), get_reg_ptr(traits<ARCH>::ICOUNT), true),
                                      this->gen_const(64U, 1)),
             get_reg_ptr(traits<ARCH>::ICOUNT), true);
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
@@ -244,20 +243,21 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
     // we fetch at max 4 byte, alignment is 2
     enum {TRAP_ID=1<<16};
     code_word_t insn = 0;
-    const typename traits<ARCH>::addr_t upper_bits = ~traits<ARCH>::PGMASK;
+    // const typename traits<ARCH>::addr_t upper_bits = ~traits<ARCH>::PGMASK;
     phys_addr_t paddr(pc);
     auto *const data = (uint8_t *)&insn;
     paddr = this->core.v2p(pc);
-    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
+    //TODO: re-add page handling
-        auto res = this->core.read(paddr, 2, data);
+//    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
-        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
+//        auto res = this->core.read(paddr, 2, data);
-        if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
+//        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
+//        if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
-        }
+//            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
-    } else {
+//        }
+//    } else {
         auto res = this->core.read(paddr, 4, data);
         if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-    }
+//    }
     if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
     // curr pc on stack
     ++inst_cnt;
@@ -271,7 +271,7 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_behavior(BasicBlock *leave_blk) {
     this->builder.SetInsertPoint(leave_blk);
-    this->builder.CreateRet(this->builder.CreateLoad(get_reg_ptr(arch::traits<ARCH>::NEXT_PC), false));
+    this->builder.CreateRet(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(arch::traits<ARCH>::NEXT_PC), false));
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(uint16_t trap_id, uint16_t cause) {
@@ -295,18 +295,18 @@ template <typename ARCH> void vm_impl<ARCH>::gen_wait(unsigned type) {
 
 template <typename ARCH> void vm_impl<ARCH>::gen_trap_behavior(BasicBlock *trap_blk) {
     this->builder.SetInsertPoint(trap_blk);
-    auto *trap_state_val = this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
+    auto *trap_state_val = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
     this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()),
                               get_reg_ptr(traits<ARCH>::LAST_BRANCH), false);
     std::vector<Value *> args{this->core_ptr, this->adj_to64(trap_state_val),
-                              this->adj_to64(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::PC), false))};
+                              this->adj_to64(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::PC), get_reg_ptr(traits<ARCH>::PC), false))};
     this->builder.CreateCall(this->mod->getFunction("enter_trap"), args);
-    auto *trap_addr_val = this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::NEXT_PC), false);
+    auto *trap_addr_val = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), false);
     this->builder.CreateRet(trap_addr_val);
 }
 
 template <typename ARCH> inline void vm_impl<ARCH>::gen_trap_check(BasicBlock *bb) {
-    auto *v = this->builder.CreateLoad(get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), true);
+    auto *v = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), true);
     this->gen_cond_branch(this->builder.CreateICmp(
                               ICmpInst::ICMP_EQ, v,
                               ConstantInt::get(getContext(), APInt(v->getType()->getIntegerBitWidth(), 0))),
@@ -323,3 +323,25 @@ std::unique_ptr<vm_if> create<arch::${coreDef.name.toLowerCase()}>(arch::${coreD
 }
 } // namespace llvm
 } // 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("${coreDef.name.toLowerCase()}|m_p|llvm", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_m_p<iss::arch::${coreDef.name.toLowerCase()}>();
+            auto* vm = new llvm::${coreDef.name.toLowerCase()}::vm_impl<arch::${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("${coreDef.name.toLowerCase()}|mu_p|llvm", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_mu_p<iss::arch::${coreDef.name.toLowerCase()}>();
+            auto* vm = new llvm::${coreDef.name.toLowerCase()}::vm_impl<arch::${coreDef.name.toLowerCase()}>(*cpu, false);
+            if (port != 0) debugger::server<debugger::gdb_session>::run_server(vm, port);
+            return {cpu_ptr{cpu}, vm_ptr{vm}};
+        })
+};
+}
+}

gen_input/templates/llvm/CORENAME_cyles.txt.gtl

@@ -1,9 +0,0 @@
-{ 
-	"${coreDef.name}" : [<%instructions.eachWithIndex{instr,index -> %>${index==0?"":","}
-		{
-			"name"  : "${instr.name}",
-			"size"  : ${instr.length},
-			"delay" : ${generator.hasAttribute(instr.instruction, com.minres.coredsl.coreDsl.InstrAttribute.COND)?[1,1]:1}
-		}<%}%>
-	]
-}

gen_input/templates/llvm/incl-CORENAME.h.gtl

@@ -1,223 +0,0 @@
-/*******************************************************************************
- * Copyright (C) 2017, 2018 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.
- *
- *******************************************************************************/
-
-<% 
-import com.minres.coredsl.coreDsl.Register
-import com.minres.coredsl.coreDsl.RegisterFile
-import com.minres.coredsl.coreDsl.RegisterAlias
-def getTypeSize(size){
-	if(size > 32) 64 else if(size > 16) 32 else if(size > 8) 16 else 8
-}
-def getOriginalName(reg){
-    if( reg.original instanceof RegisterFile) {
-    	if( reg.index != null ) {
-        	return reg.original.name+generator.generateHostCode(reg.index)
-        } else {
-        	return reg.original.name
-        }
-    } else if(reg.original instanceof Register){
-        return reg.original.name
-    }
-}
-def getRegisterNames(){
-	def regNames = []
- 	allRegs.each { reg -> 
-		if( reg instanceof RegisterFile) {
-			(reg.range.right..reg.range.left).each{
-    			regNames+=reg.name.toLowerCase()+it
-            }
-        } else if(reg instanceof Register){
-    		regNames+=reg.name.toLowerCase()
-        }
-    }
-    return regNames
-}
-def getRegisterAliasNames(){
-	def regMap = allRegs.findAll{it instanceof RegisterAlias }.collectEntries {[getOriginalName(it), it.name]}
- 	return allRegs.findAll{it instanceof Register || it instanceof RegisterFile}.collect{reg ->
-		if( reg instanceof RegisterFile) {
-			return (reg.range.right..reg.range.left).collect{ (regMap[reg.name]?:regMap[reg.name+it]?:reg.name.toLowerCase()+it).toLowerCase() }
-        } else if(reg instanceof Register){
-    		regMap[reg.name]?:reg.name.toLowerCase()
-        }
- 	}.flatten()
-}
-%>
-#ifndef _${coreDef.name.toUpperCase()}_H_
-#define _${coreDef.name.toUpperCase()}_H_
-
-#include <array>
-#include <iss/arch/traits.h>
-#include <iss/arch_if.h>
-#include <iss/vm_if.h>
-
-namespace iss {
-namespace arch {
-
-struct ${coreDef.name.toLowerCase()};
-
-template <> struct traits<${coreDef.name.toLowerCase()}> {
-
-	constexpr static char const* const core_type = "${coreDef.name}";
-    
-  	static constexpr std::array<const char*, ${getRegisterNames().size}> reg_names{
- 		{"${getRegisterNames().join("\", \"")}"}};
- 
-  	static constexpr std::array<const char*, ${getRegisterAliasNames().size}> reg_aliases{
- 		{"${getRegisterAliasNames().join("\", \"")}"}};
-
-    enum constants {${coreDef.constants.collect{c -> c.name+"="+c.value}.join(', ')}};
-
-    constexpr static unsigned FP_REGS_SIZE = ${coreDef.constants.find {it.name=='FLEN'}?.value?:0};
-
-    enum reg_e {<%
-     	allRegs.each { reg -> 
-    		if( reg instanceof RegisterFile) {
-    			(reg.range.right..reg.range.left).each{%>
-        ${reg.name}${it},<%
-                }
-            } else if(reg instanceof Register){ %>
-        ${reg.name},<%  
-            }
-        }%>
-        NUM_REGS,
-        NEXT_${pc.name}=NUM_REGS,
-        TRAP_STATE,
-        PENDING_TRAP,
-        MACHINE_STATE,
-        LAST_BRANCH,
-        ICOUNT<% 
-     	allRegs.each { reg -> 
-    		if(reg instanceof RegisterAlias){ def aliasname=getOriginalName(reg)%>,
-        ${reg.name} = ${aliasname}<%
-            }
-        }%>
-    };
-
-    using reg_t = uint${regDataWidth}_t;
-
-    using addr_t = uint${addrDataWidth}_t;
-
-    using code_word_t = uint${addrDataWidth}_t; //TODO: check removal
-
-    using virt_addr_t = iss::typed_addr_t<iss::address_type::VIRTUAL>;
-
-    using phys_addr_t = iss::typed_addr_t<iss::address_type::PHYSICAL>;
-
- 	static constexpr std::array<const uint32_t, ${regSizes.size}> reg_bit_widths{
- 		{${regSizes.join(",")}}};
-
-    static constexpr std::array<const uint32_t, ${regOffsets.size}> reg_byte_offsets{
-    	{${regOffsets.join(",")}}};
-
-    static const uint64_t addr_mask = (reg_t(1) << (XLEN - 1)) | ((reg_t(1) << (XLEN - 1)) - 1);
-
-    enum sreg_flag_e { FLAGS };
-
-    enum mem_type_e { ${allSpaces.collect{s -> s.name}.join(', ')} };
-};
-
-struct ${coreDef.name.toLowerCase()}: public arch_if {
-
-    using virt_addr_t = typename traits<${coreDef.name.toLowerCase()}>::virt_addr_t;
-    using phys_addr_t = typename traits<${coreDef.name.toLowerCase()}>::phys_addr_t;
-    using reg_t =  typename traits<${coreDef.name.toLowerCase()}>::reg_t;
-    using addr_t = typename traits<${coreDef.name.toLowerCase()}>::addr_t;
-
-    ${coreDef.name.toLowerCase()}();
-    ~${coreDef.name.toLowerCase()}();
-
-    void reset(uint64_t address=0) override;
-
-    uint8_t* get_regs_base_ptr() override;
-    /// deprecated
-    void get_reg(short idx, std::vector<uint8_t>& value) override {}
-    void set_reg(short idx, const std::vector<uint8_t>& value) override {}
-    /// deprecated
-    bool get_flag(int flag) override {return false;}
-    void set_flag(int, bool value) override {};
-    /// deprecated
-    void update_flags(operations op, uint64_t opr1, uint64_t opr2) override {};
-
-    inline uint64_t get_icount() { return reg.icount; }
-
-    inline bool should_stop() { return interrupt_sim; }
-
-    inline uint64_t stop_code() { return interrupt_sim; }
-
-    inline phys_addr_t v2p(const iss::addr_t& addr){
-        if (addr.space != traits<${coreDef.name.toLowerCase()}>::MEM || addr.type == iss::address_type::PHYSICAL ||
-                addr_mode[static_cast<uint16_t>(addr.access)&0x3]==address_type::PHYSICAL) {
-            return phys_addr_t(addr.access, addr.space, addr.val&traits<${coreDef.name.toLowerCase()}>::addr_mask);
-        } else
-            return virt2phys(addr);
-    }
-
-    virtual phys_addr_t virt2phys(const iss::addr_t& addr);
-
-    virtual iss::sync_type needed_sync() const { return iss::NO_SYNC; }
-
-    inline uint32_t get_last_branch() { return reg.last_branch; }
-
-protected:
-    struct ${coreDef.name}_regs {<%
-     	allRegs.each { reg -> 
-    		if( reg instanceof RegisterFile) {
-    			(reg.range.right..reg.range.left).each{%>
-        uint${generator.getSize(reg)}_t ${reg.name}${it} = 0;<%
-                }
-            } else if(reg instanceof Register){ %>
-        uint${generator.getSize(reg)}_t ${reg.name} = 0;<%
-            }
-        }%>
-        uint${generator.getSize(pc)}_t NEXT_${pc.name} = 0;
-        uint32_t trap_state = 0, pending_trap = 0, machine_state = 0, last_branch = 0;
-        uint64_t icount = 0;
-    } reg;
-
-    std::array<address_type, 4> addr_mode;
-    
-    uint64_t interrupt_sim=0;
-<%
-def fcsr = allRegs.find {it.name=='FCSR'}
-if(fcsr != null) {%>
-	uint${generator.getSize(fcsr)}_t get_fcsr(){return reg.FCSR;}
-	void set_fcsr(uint${generator.getSize(fcsr)}_t val){reg.FCSR = val;}		
-<%} else { %>
-	uint32_t get_fcsr(){return 0;}
-	void set_fcsr(uint32_t val){}
-<%}%>
-};
-
-}
-}            
-#endif /* _${coreDef.name.toUpperCase()}_H_ */

gen_input/templates/llvm/src-CORENAME.cpp.gtl

@@ -1,107 +0,0 @@
-/*******************************************************************************
- * Copyright (C) 2017, 2018 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.
- *
- *******************************************************************************/
- <% 
-import com.minres.coredsl.coreDsl.Register
-import com.minres.coredsl.coreDsl.RegisterFile
-import com.minres.coredsl.coreDsl.RegisterAlias
-def getOriginalName(reg){
-    if( reg.original instanceof RegisterFile) {
-    	if( reg.index != null ) {
-        	return reg.original.name+generator.generateHostCode(reg.index)
-        } else {
-        	return reg.original.name
-        }
-    } else if(reg.original instanceof Register){
-        return reg.original.name
-    }
-}
-def getRegisterNames(){
-	def regNames = []
- 	allRegs.each { reg -> 
-		if( reg instanceof RegisterFile) {
-			(reg.range.right..reg.range.left).each{
-    			regNames+=reg.name.toLowerCase()+it
-            }
-        } else if(reg instanceof Register){
-    		regNames+=reg.name.toLowerCase()
-        }
-    }
-    return regNames
-}
-def getRegisterAliasNames(){
-	def regMap = allRegs.findAll{it instanceof RegisterAlias }.collectEntries {[getOriginalName(it), it.name]}
- 	return allRegs.findAll{it instanceof Register || it instanceof RegisterFile}.collect{reg ->
-		if( reg instanceof RegisterFile) {
-			return (reg.range.right..reg.range.left).collect{ (regMap[reg.name]?:regMap[reg.name+it]?:reg.name.toLowerCase()+it).toLowerCase() }
-        } else if(reg instanceof Register){
-    		regMap[reg.name]?:reg.name.toLowerCase()
-        }
- 	}.flatten()
-}
-%>
-#include "util/ities.h"
-#include <util/logging.h>
-#include <iss/arch/${coreDef.name.toLowerCase()}.h>
-#include <cstdio>
-#include <cstring>
-#include <fstream>
-
-using namespace iss::arch;
-
-constexpr std::array<const char*, ${getRegisterNames().size}>    iss::arch::traits<iss::arch::${coreDef.name.toLowerCase()}>::reg_names;
-constexpr std::array<const char*, ${getRegisterAliasNames().size}>    iss::arch::traits<iss::arch::${coreDef.name.toLowerCase()}>::reg_aliases;
-constexpr std::array<const uint32_t, ${regSizes.size}> iss::arch::traits<iss::arch::${coreDef.name.toLowerCase()}>::reg_bit_widths;
-constexpr std::array<const uint32_t, ${regOffsets.size}> iss::arch::traits<iss::arch::${coreDef.name.toLowerCase()}>::reg_byte_offsets;
-
-${coreDef.name.toLowerCase()}::${coreDef.name.toLowerCase()}() {
-    reg.icount = 0;
-}
-
-${coreDef.name.toLowerCase()}::~${coreDef.name.toLowerCase()}() = default;
-
-void ${coreDef.name.toLowerCase()}::reset(uint64_t address) {
-    for(size_t i=0; i<traits<${coreDef.name.toLowerCase()}>::NUM_REGS; ++i) set_reg(i, std::vector<uint8_t>(sizeof(traits<${coreDef.name.toLowerCase()}>::reg_t),0));
-    reg.PC=address;
-    reg.NEXT_PC=reg.PC;
-    reg.trap_state=0;
-    reg.machine_state=0x3;
-    reg.icount=0;
-}
-
-uint8_t *${coreDef.name.toLowerCase()}::get_regs_base_ptr() {
-	return reinterpret_cast<uint8_t*>(&reg);
-}
-
-${coreDef.name.toLowerCase()}::phys_addr_t ${coreDef.name.toLowerCase()}::virt2phys(const iss::addr_t &pc) {
-    return phys_addr_t(pc); // change logical address to physical address
-}
-

gen_input/templates/tcc/CORENAME.cpp.gtl

@@ -120,57 +120,7 @@ protected:
         }
     }
 
-    // some compile time constants
-    // enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
-    enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
-    enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
-    enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK16)) };
     
-    std::array<compile_func, LUT_SIZE> lut;
-
-    std::array<compile_func, LUT_SIZE_C> lut_00, lut_01, lut_10;
-    std::array<compile_func, LUT_SIZE> lut_11;
-
-    std::array<compile_func *, 4> qlut;
-
-    std::array<const uint32_t, 4> lutmasks = {{EXTR_MASK16, EXTR_MASK16, EXTR_MASK16, EXTR_MASK32}};
-
-    void expand_bit_mask(int pos, uint32_t mask, uint32_t value, uint32_t valid, uint32_t idx, compile_func lut[],
-                         compile_func f) {
-        if (pos < 0) {
-            lut[idx] = f;
-        } else {
-            auto bitmask = 1UL << pos;
-            if ((mask & bitmask) == 0) {
-                expand_bit_mask(pos - 1, mask, value, valid, idx, lut, f);
-            } else {
-                if ((valid & bitmask) == 0) {
-                    expand_bit_mask(pos - 1, mask, value, valid, (idx << 1), lut, f);
-                    expand_bit_mask(pos - 1, mask, value, valid, (idx << 1) + 1, lut, f);
-                } else {
-                    auto new_val = idx << 1;
-                    if ((value & bitmask) != 0) new_val++;
-                    expand_bit_mask(pos - 1, mask, value, valid, new_val, lut, f);
-                }
-            }
-        }
-    }
-
-    inline uint32_t extract_fields(uint32_t val) { return extract_fields(29, val >> 2, lutmasks[val & 0x3], 0); }
-
-    uint32_t extract_fields(int pos, uint32_t val, uint32_t mask, uint32_t lut_val) {
-        if (pos >= 0) {
-            auto bitmask = 1UL << pos;
-            if ((mask & bitmask) == 0) {
-                lut_val = extract_fields(pos - 1, val, mask, lut_val);
-            } else {
-                auto new_val = lut_val << 1;
-                if ((val & bitmask) != 0) new_val++;
-                lut_val = extract_fields(pos - 1, val, mask, new_val);
-            }
-        }
-        return lut_val;
-    }
     template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
     inline S sext(U from) {
         auto mask = (1ULL<<W) - 1;
@@ -182,14 +132,23 @@ private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
-    struct InstructionDesriptor {
+    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){}
+    };
 
-    const std::array<InstructionDesriptor, ${instructions.size}> instr_descr = {{
+    decoding_tree_node* root {nullptr};
+
+    const std::array<instruction_descriptor, ${instructions.size}> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */<%instructions.each{instr -> %>
         /* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */
         {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
@@ -227,11 +186,64 @@ private:
         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;
+    }
 };
 
-template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
+template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
-    volatile CODE_WORD x = insn;
+    volatile CODE_WORD x = instr;
-    insn = 2 * x;
+    instr = 2 * x;
 }
 
 template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@@ -239,14 +251,11 @@ 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) {
-    qlut[0] = lut_00.data();
+    root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
-    qlut[1] = lut_01.data();
+    for(auto instr:instr_descr){
-    qlut[2] = lut_10.data();
+        root->instrs.push_back(instr);
-    qlut[3] = lut_11.data();
-    for (auto instr : instr_descr) {
-        auto quantrant = instr.value & 0x3;
-        expand_bit_mask(29, lutmasks[quantrant], instr.value >> 2, instr.mask >> 2, 0, qlut[quantrant], instr.op);
     }
+    populate_decoding_tree(root);
 }
 
 template <typename ARCH>
@@ -254,30 +263,19 @@ std::tuple<continuation_e>
 vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) {
     // we fetch at max 4 byte, alignment is 2
     enum {TRAP_ID=1<<16};
-    code_word_t insn = 0;
+    code_word_t instr = 0;
-    // const typename traits::addr_t upper_bits = ~traits::PGMASK;
     phys_addr_t paddr(pc);
-    auto *const data = (uint8_t *)&insn;
     paddr = this->core.v2p(pc);
-//    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
+    auto res = this->core.read(paddr, 4, reinterpret_cast<uint8_t*>(&instr));
-//        auto res = this->core.read(paddr, 2, data);
+    if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-//        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
+    if (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
-//        if ((insn & 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 (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
     // curr pc on stack
     ++inst_cnt;
-    auto lut_val = extract_fields(insn);
+    auto f = decode_instr(root, instr);
-    auto f = qlut[insn & 0x3][lut_val];
     if (f == nullptr) {
         f = &this_class::illegal_intruction;
     }
-    return (this->*f)(pc, insn, tu);
+    return (this->*f)(pc, instr, tu);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {

src/vm/interp/vm_tgc_c.cpp

@@ -152,14 +152,22 @@ private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
-    struct InstructionDesriptor {
+    struct instruction_descriptor {
         size_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){}
+    };
 
-    const std::array<InstructionDesriptor, 87> instr_descr = {{
+    decoding_tree_node* root {nullptr};
+    const std::array<instruction_descriptor, 87> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */
         {32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::LUI},
         {32, 0b00000000000000000000000000010111, 0b00000000000000000000000001111111, arch::traits<ARCH>::opcode_e::AUIPC},
@@ -250,9 +258,9 @@ private:
         {16, 0b0000000000000000, 0b1111111111111111, arch::traits<ARCH>::opcode_e::DII},
     }};
 
-    //static constexpr typename traits::addr_t upper_bits = ~traits::PGMASK;
     iss::status fetch_ins(virt_addr_t pc, uint8_t * data){
         auto phys_pc = this->core.v2p(pc);
+        //TODO: re-add page handling
         //if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
         //    if (this->core.read(phys_pc, 2, data) != iss::Ok) return iss::Err;
         //    if ((data[0] & 0x3) == 0x3) // this is a 32bit instruction
@@ -262,6 +270,56 @@ 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) {
@@ -288,16 +346,11 @@ 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) {
-    unsigned id=0;
+    root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
-    for (auto instr : instr_descr) {
+    for(auto instr:instr_descr){
-        auto quadrant = instr.value & 0x3;
+        root->instrs.push_back(instr);
-        qlut[quadrant].push_back(instruction_pattern{instr.value, instr.mask, instr.op});
-    }
-    for(auto& lut: qlut){
-        std::sort(std::begin(lut), std::end(lut), [](instruction_pattern const& a, instruction_pattern const& b){
-            return bit_count(a.mask) > bit_count(b.mask);
-        });
     }
+    populate_decoding_tree(root);
 }
 
 inline bool is_count_limit_enabled(finish_cond_e cond){
@@ -308,14 +361,6 @@ inline bool is_jump_to_self_enabled(finish_cond_e cond){
     return (cond & finish_cond_e::JUMP_TO_SELF) == finish_cond_e::JUMP_TO_SELF;
 }
 
-template <typename ARCH>
-typename arch::traits<ARCH>::opcode_e vm_impl<ARCH>::decode_inst_id(code_word_t instr){
-    for(auto& e: qlut[instr&0x3]){
-        if(!((instr&e.mask) ^ e.value )) return e.id;
-    }
-    return arch::traits<ARCH>::opcode_e::MAX_OPCODE;
-}
-
 template <typename ARCH>
 typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e cond, virt_addr_t start, uint64_t icount_limit){
     auto pc=start;
@@ -337,7 +382,7 @@ typename vm_base<ARCH>::virt_addr_t vm_impl<ARCH>::execute_inst(finish_cond_e co
         } 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_inst_id(instr);
+            auto inst_id = decode_instr(root, instr);
             // 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));

src/vm/llvm/vm_tgc_c.cpp

@@ -111,7 +111,7 @@ protected:
     void gen_trap_check(BasicBlock *bb);
 
     inline Value *gen_reg_load(unsigned i, unsigned level = 0) {
-        return this->builder.CreateLoad(get_reg_ptr(i), false);
+        return this->builder.CreateLoad(this->get_typeptr(i), get_reg_ptr(i), false);
     }
 
     inline void gen_set_pc(virt_addr_t pc, unsigned reg_num) {
@@ -124,7 +124,7 @@ protected:
     // enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
     enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
     enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
-    enum { LUT_SIZE = 1 << util::bit_count(EXTR_MASK32), LUT_SIZE_C = 1 << util::bit_count(EXTR_MASK16) };
+    enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint64_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint64_t>(EXTR_MASK16)) };
 
     using this_class = vm_impl<ARCH>;
     using compile_func = std::tuple<continuation_e, BasicBlock *> (this_class::*)(virt_addr_t &pc,
@@ -4042,10 +4042,10 @@ private:
      ****************************************************************************/
     std::tuple<continuation_e, BasicBlock *> illegal_intruction(virt_addr_t &pc, code_word_t instr, BasicBlock *bb) {
 		this->gen_sync(iss::PRE_SYNC, instr_descr.size());
-        this->builder.CreateStore(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::NEXT_PC), true),
+        this->builder.CreateStore(this->builder.CreateLoad(this->get_typeptr(traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), true),
                                    get_reg_ptr(traits<ARCH>::PC), true);
         this->builder.CreateStore(
-            this->builder.CreateAdd(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::ICOUNT), true),
+            this->builder.CreateAdd(this->builder.CreateLoad(this->get_typeptr(traits<ARCH>::ICOUNT), get_reg_ptr(traits<ARCH>::ICOUNT), true),
                                      this->gen_const(64U, 1)),
             get_reg_ptr(traits<ARCH>::ICOUNT), true);
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
@@ -4082,20 +4082,21 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
     // we fetch at max 4 byte, alignment is 2
     enum {TRAP_ID=1<<16};
     code_word_t insn = 0;
-    const typename traits<ARCH>::addr_t upper_bits = ~traits<ARCH>::PGMASK;
+    // const typename traits<ARCH>::addr_t upper_bits = ~traits<ARCH>::PGMASK;
     phys_addr_t paddr(pc);
     auto *const data = (uint8_t *)&insn;
     paddr = this->core.v2p(pc);
-    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
+    //TODO: re-add page handling
-        auto res = this->core.read(paddr, 2, data);
+//    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
-        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
+//        auto res = this->core.read(paddr, 2, data);
-        if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
+//        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
+//        if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
-        }
+//            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
-    } else {
+//        }
+//    } else {
         auto res = this->core.read(paddr, 4, data);
         if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-    }
+//    }
     if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
     // curr pc on stack
     ++inst_cnt;
@@ -4109,7 +4110,7 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_behavior(BasicBlock *leave_blk) {
     this->builder.SetInsertPoint(leave_blk);
-    this->builder.CreateRet(this->builder.CreateLoad(get_reg_ptr(arch::traits<ARCH>::NEXT_PC), false));
+    this->builder.CreateRet(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(arch::traits<ARCH>::NEXT_PC), false));
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(uint16_t trap_id, uint16_t cause) {
@@ -4133,18 +4134,18 @@ template <typename ARCH> void vm_impl<ARCH>::gen_wait(unsigned type) {
 
 template <typename ARCH> void vm_impl<ARCH>::gen_trap_behavior(BasicBlock *trap_blk) {
     this->builder.SetInsertPoint(trap_blk);
-    auto *trap_state_val = this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
+    auto *trap_state_val = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
     this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()),
                               get_reg_ptr(traits<ARCH>::LAST_BRANCH), false);
     std::vector<Value *> args{this->core_ptr, this->adj_to64(trap_state_val),
-                              this->adj_to64(this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::PC), false))};
+                              this->adj_to64(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::PC), get_reg_ptr(traits<ARCH>::PC), false))};
     this->builder.CreateCall(this->mod->getFunction("enter_trap"), args);
-    auto *trap_addr_val = this->builder.CreateLoad(get_reg_ptr(traits<ARCH>::NEXT_PC), false);
+    auto *trap_addr_val = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), false);
     this->builder.CreateRet(trap_addr_val);
 }
 
 template <typename ARCH> inline void vm_impl<ARCH>::gen_trap_check(BasicBlock *bb) {
-    auto *v = this->builder.CreateLoad(get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), true);
+    auto *v = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), true);
     this->gen_cond_branch(this->builder.CreateICmp(
                               ICmpInst::ICMP_EQ, v,
                               ConstantInt::get(getContext(), APInt(v->getType()->getIntegerBitWidth(), 0))),
@@ -4161,3 +4162,25 @@ std::unique_ptr<vm_if> create<arch::tgc_c>(arch::tgc_c *core, unsigned short por
 }
 } // namespace llvm
 } // 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("tgc_c|m_p|llvm", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_m_p<iss::arch::tgc_c>();
+            auto* vm = new llvm::tgc_c::vm_impl<arch::tgc_c>(*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("tgc_c|mu_p|llvm", [](unsigned port, void*) -> std::tuple<cpu_ptr, vm_ptr>{
+            auto* cpu = new iss::arch::riscv_hart_mu_p<iss::arch::tgc_c>();
+            auto* vm = new llvm::tgc_c::vm_impl<arch::tgc_c>(*cpu, false);
+            if (port != 0) debugger::server<debugger::gdb_session>::run_server(vm, port);
+            return {cpu_ptr{cpu}, vm_ptr{vm}};
+        })
+};
+}
+}

src/vm/tcc/vm_tgc_c.cpp

@@ -120,57 +120,7 @@ protected:
         }
     }
 
-    // some compile time constants
-    // enum { MASK16 = 0b1111110001100011, MASK32 = 0b11111111111100000111000001111111 };
-    enum { MASK16 = 0b1111111111111111, MASK32 = 0b11111111111100000111000001111111 };
-    enum { EXTR_MASK16 = MASK16 >> 2, EXTR_MASK32 = MASK32 >> 2 };
-    enum { LUT_SIZE = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK32)), LUT_SIZE_C = 1 << util::bit_count(static_cast<uint32_t>(EXTR_MASK16)) };
     
-    std::array<compile_func, LUT_SIZE> lut;
-
-    std::array<compile_func, LUT_SIZE_C> lut_00, lut_01, lut_10;
-    std::array<compile_func, LUT_SIZE> lut_11;
-
-    std::array<compile_func *, 4> qlut;
-
-    std::array<const uint32_t, 4> lutmasks = {{EXTR_MASK16, EXTR_MASK16, EXTR_MASK16, EXTR_MASK32}};
-
-    void expand_bit_mask(int pos, uint32_t mask, uint32_t value, uint32_t valid, uint32_t idx, compile_func lut[],
-                         compile_func f) {
-        if (pos < 0) {
-            lut[idx] = f;
-        } else {
-            auto bitmask = 1UL << pos;
-            if ((mask & bitmask) == 0) {
-                expand_bit_mask(pos - 1, mask, value, valid, idx, lut, f);
-            } else {
-                if ((valid & bitmask) == 0) {
-                    expand_bit_mask(pos - 1, mask, value, valid, (idx << 1), lut, f);
-                    expand_bit_mask(pos - 1, mask, value, valid, (idx << 1) + 1, lut, f);
-                } else {
-                    auto new_val = idx << 1;
-                    if ((value & bitmask) != 0) new_val++;
-                    expand_bit_mask(pos - 1, mask, value, valid, new_val, lut, f);
-                }
-            }
-        }
-    }
-
-    inline uint32_t extract_fields(uint32_t val) { return extract_fields(29, val >> 2, lutmasks[val & 0x3], 0); }
-
-    uint32_t extract_fields(int pos, uint32_t val, uint32_t mask, uint32_t lut_val) {
-        if (pos >= 0) {
-            auto bitmask = 1UL << pos;
-            if ((mask & bitmask) == 0) {
-                lut_val = extract_fields(pos - 1, val, mask, lut_val);
-            } else {
-                auto new_val = lut_val << 1;
-                if ((val & bitmask) != 0) new_val++;
-                lut_val = extract_fields(pos - 1, val, mask, new_val);
-            }
-        }
-        return lut_val;
-    }
     template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
     inline S sext(U from) {
         auto mask = (1ULL<<W) - 1;
@@ -182,14 +132,23 @@ private:
     /****************************************************************************
      * start opcode definitions
      ****************************************************************************/
-    struct InstructionDesriptor {
+    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){}
+    };
 
-    const std::array<InstructionDesriptor, 87> instr_descr = {{
+    decoding_tree_node* root {nullptr};
+
+    const std::array<instruction_descriptor, 87> instr_descr = {{
          /* entries are: size, valid value, valid mask, function ptr */
         /* instruction LUI, encoding '0b00000000000000000000000000110111' */
         {32, 0b00000000000000000000000000110111, 0b00000000000000000000000001111111, &this_class::__lui},
@@ -493,14 +452,14 @@ private:
             this->gen_raise_trap(tu, 0,  2);
         }
         else{
-        	auto new_pc = tu.assignment(tu.ext((tu.bitwise_and((tu.add(tu.load(rs1+ traits::X0, 0),tu.constant((int16_t)sext<12>(imm),16))),tu.constant(~ 0x1,8))),32,true),32);
+        	auto new_pc = tu.assignment(tu.ext((tu.bitwise_and((tu.add(tu.load(rs1+ traits::X0, 0),tu.constant((int16_t)sext<12>(imm),16))),tu.constant(~0x1,8))),32,true),32);
         	tu.open_if(tu.srem(new_pc,tu.constant(static_cast<uint32_t>(traits:: INSTR_ALIGNMENT),32)));
         	this->gen_raise_trap(tu, 0,  0);
         	tu.open_else();
         	if(rd!= 0) {
         	    tu.store(rd + traits::X0,tu.ext((tu.add(tu.ext(cur_pc_val,32,false),tu.constant( 4,8))),32,true));
         	}
-        	auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~ 0x1,8)),32);
+        	auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~0x1,8)),32);
         	tu.store(traits::NEXT_PC, PC_val_v);
         	tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
         	tu.close_scope();
@@ -1962,7 +1921,7 @@ private:
         else{
         	auto xrd = tu.assignment(tu.read_mem(traits::CSR, csr, 32),32);
         	if(zimm!= 0) {
-        	    tu.write_mem(traits::CSR, csr, tu.bitwise_and(xrd,tu.constant(~ ((uint32_t)zimm),32)));
+        	    tu.write_mem(traits::CSR, csr, tu.bitwise_and(xrd,tu.constant(~((uint32_t)zimm),32)));
         	}
         	if(rd!= 0) {
         	    tu.store(rd + traits::X0,xrd);
@@ -2163,13 +2122,13 @@ private:
         	auto divisor = tu.assignment(tu.ext(tu.load(rs2+ traits::X0, 0),32,false),32);
         	if(rd!= 0){ tu.open_if(tu.icmp(ICmpInst::ICMP_NE,divisor,tu.constant( 0,8)));
         	auto MMIN = tu.assignment(tu.constant(((uint32_t)1)<<(static_cast<uint32_t>(traits:: XLEN)-1),32),32);
-        	tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,divisor,tu.constant(- 1,8))));
+        	tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,divisor,tu.constant(-1,8))));
         	tu.store(rd + traits::X0,MMIN);
         	tu.open_else();
         	tu.store(rd + traits::X0,tu.ext((tu.sdiv(dividend,divisor)),32,true));
         	tu.close_scope();
         	tu.open_else();
-        	tu.store(rd + traits::X0,tu.constant((uint32_t)- 1,32));
+        	tu.store(rd + traits::X0,tu.constant((uint32_t)-1,32));
         	tu.close_scope();
         	}
         }
@@ -2208,7 +2167,7 @@ private:
         	}
         	tu.open_else();
         	if(rd!=0) {
-        	    tu.store(rd + traits::X0,tu.constant((uint32_t)- 1,32));
+        	    tu.store(rd + traits::X0,tu.constant((uint32_t)-1,32));
         	}
         	tu.close_scope();
         }
@@ -2243,7 +2202,7 @@ private:
         else{
         	tu.open_if(tu.icmp(ICmpInst::ICMP_NE,tu.load(rs2+ traits::X0, 0),tu.constant( 0,8)));
         	auto MMIN = tu.assignment(tu.constant( 1<<(static_cast<uint32_t>(traits:: XLEN)-1),8),32);
-        	tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,tu.ext(tu.load(rs2+ traits::X0, 0),32,false),tu.constant(- 1,8))));
+        	tu.open_if(tu.logical_and(tu.icmp(ICmpInst::ICMP_EQ,tu.load(rs1+ traits::X0, 0),MMIN),tu.icmp(ICmpInst::ICMP_EQ,tu.ext(tu.load(rs2+ traits::X0, 0),32,false),tu.constant(-1,8))));
         	if(rd!=0) {
         	    tu.store(rd + traits::X0,tu.constant( 0,8));
         	}
@@ -2955,7 +2914,7 @@ private:
         gen_set_pc(tu, pc, traits::NEXT_PC);
         tu.open_scope();
         if(rs1&&rs1<static_cast<uint32_t>(traits:: RFS)) {
-            auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(tu.load(rs1%static_cast<uint32_t>(traits:: RFS)+ traits::X0, 0),tu.constant(~ 0x1,8)),32);
+            auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(tu.load(rs1%static_cast<uint32_t>(traits:: RFS)+ traits::X0, 0),tu.constant(~0x1,8)),32);
             tu.store(traits::NEXT_PC, PC_val_v);
             tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
         }
@@ -3043,7 +3002,7 @@ private:
         else{
         	auto new_pc = tu.assignment(tu.load(rs1+ traits::X0, 0),32);
         	tu.store(1 + traits::X0,tu.ext((tu.add(tu.ext(cur_pc_val,32,false),tu.constant( 2,8))),32,true));
-        	auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~ 0x1,8)),32);
+        	auto PC_val_v = tu.assignment("PC_val", tu.bitwise_and(new_pc,tu.constant(~0x1,8)),32);
         	tu.store(traits::NEXT_PC, PC_val_v);
         	tu.store(traits::LAST_BRANCH, tu.constant(2U, 2));
         }
@@ -3136,11 +3095,64 @@ private:
         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;
+    }
 };
 
-template <typename CODE_WORD> void debug_fn(CODE_WORD insn) {
+template <typename CODE_WORD> void debug_fn(CODE_WORD instr) {
-    volatile CODE_WORD x = insn;
+    volatile CODE_WORD x = instr;
-    insn = 2 * x;
+    instr = 2 * x;
 }
 
 template <typename ARCH> vm_impl<ARCH>::vm_impl() { this(new ARCH()); }
@@ -3148,14 +3160,11 @@ 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) {
-    qlut[0] = lut_00.data();
+    root = new decoding_tree_node(std::numeric_limits<uint32_t>::max());
-    qlut[1] = lut_01.data();
+    for(auto instr:instr_descr){
-    qlut[2] = lut_10.data();
+        root->instrs.push_back(instr);
-    qlut[3] = lut_11.data();
-    for (auto instr : instr_descr) {
-        auto quantrant = instr.value & 0x3;
-        expand_bit_mask(29, lutmasks[quantrant], instr.value >> 2, instr.mask >> 2, 0, qlut[quantrant], instr.op);
     }
+    populate_decoding_tree(root);
 }
 
 template <typename ARCH>
@@ -3163,11 +3172,10 @@ std::tuple<continuation_e>
 vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, tu_builder& tu) {
     // we fetch at max 4 byte, alignment is 2
     enum {TRAP_ID=1<<16};
-    code_word_t insn = 0;
+    code_word_t instr = 0;
-    // const typename traits::addr_t upper_bits = ~traits::PGMASK;
     phys_addr_t paddr(pc);
-    auto *const data = (uint8_t *)&insn;
     paddr = this->core.v2p(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);
@@ -3175,18 +3183,17 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
 //            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
 //        }
 //    } else {
-        auto res = this->core.read(paddr, 4, data);
+        auto res = this->core.read(paddr, 4, reinterpret_cast<uint8_t*>(&instr));
         if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
 //    }
-    if (insn == 0x0000006f || (insn&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
+    if (instr == 0x0000006f || (instr&0xffff)==0xa001) throw simulation_stopped(0); // 'J 0' or 'C.J 0'
     // curr pc on stack
     ++inst_cnt;
-    auto lut_val = extract_fields(insn);
+    auto f = decode_instr(root, instr);
-    auto f = qlut[insn & 0x3][lut_val];
     if (f == nullptr) {
         f = &this_class::illegal_intruction;
     }
-    return (this->*f)(pc, insn, tu);
+    return (this->*f)(pc, instr, tu);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {