adds initial working version of llvm backend

CMakeLists.txt

@@ -95,6 +95,7 @@ if(TARGET RapidJSON)
 endif()
 
 if(WITH_LLVM)
+    find_package(LLVM)
 	target_compile_definitions(${PROJECT_NAME} PUBLIC ${LLVM_DEFINITIONS})
 	target_include_directories(${PROJECT_NAME} PUBLIC ${LLVM_INCLUDE_DIRS})
 	if(BUILD_SHARED_LIBS)

gen_input/templates/llvm/CORENAME.cpp.gtl

@@ -58,6 +58,7 @@ using namespace iss::debugger;
 
 template <typename ARCH> class vm_impl : public iss::llvm::vm_base<ARCH> {
 public:
+    using traits = arch::traits<ARCH>;
     using super = typename iss::llvm::vm_base<ARCH>;
     using virt_addr_t = typename super::virt_addr_t;
     using phys_addr_t = typename super::phys_addr_t;
@@ -80,7 +81,7 @@ public:
 protected:
     using vm_base<ARCH>::get_reg_ptr;
 
-    inline const char *name(size_t index){return traits<ARCH>::reg_aliases.at(index);}
+    inline const char *name(size_t index){return traits::reg_aliases.at(index);}
 
     template <typename T> inline ConstantInt *size(T type) {
         return ConstantInt::get(getContext(), APInt(32, type->getType()->getScalarSizeInBits()));
@@ -88,7 +89,7 @@ protected:
 
     void setup_module(Module* m) override {
         super::setup_module(m);
-        iss::llvm::fp_impl::add_fp_functions_2_module(m, traits<ARCH>::FP_REGS_SIZE, traits<ARCH>::XLEN);
+        iss::llvm::fp_impl::add_fp_functions_2_module(m, traits::FP_REGS_SIZE, traits::XLEN);
     }
 
     inline Value *gen_choose(Value *cond, Value *trueVal, Value *falseVal, unsigned size) {
@@ -114,88 +115,70 @@ protected:
     }
 
     inline void gen_set_pc(virt_addr_t pc, unsigned reg_num) {
-        Value *next_pc_v = this->builder.CreateSExtOrTrunc(this->gen_const(traits<ARCH>::XLEN, pc.val),
+        Value *next_pc_v = this->builder.CreateSExtOrTrunc(this->gen_const(traits::XLEN, pc.val),
-                                                           this->get_type(traits<ARCH>::XLEN));
+                                                           this->get_type(traits::XLEN));
         this->builder.CreateStore(next_pc_v, get_reg_ptr(reg_num), true);
     }
 
     // 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<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,
                                                                                   code_word_t instr,
                                                                                   BasicBlock *bb);
-    std::array<compile_func, LUT_SIZE> lut;
+    template<unsigned W, typename U, typename S = typename std::make_signed<U>::type>
-
+    inline S sext(U from) {
-    std::array<compile_func, LUT_SIZE_C> lut_00, lut_01, lut_10;
+        auto mask = (1ULL<<W) - 1;
-    std::array<compile_func, LUT_SIZE> lut_11;
+        auto sign_mask = 1ULL<<(W-1);
-
+        return (from & mask) | ((from & sign_mask) ? ~mask : 0);
-	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;
-    }
 
 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} */
+        /* instruction ${instr.instruction.name}, encoding '${instr.encoding}' */
-        {${instr.length}, ${instr.value}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
+        {${instr.length}, ${instr.encoding}, ${instr.mask}, &this_class::__${generator.functionName(instr.name)}},<%}%>
     }};
  
     /* instruction definitions */<%instructions.eachWithIndex{instr, idx -> %>
     /* instruction ${idx}: ${instr.name} */
-    std::tuple<continuation_e, BasicBlock*> __${generator.functionName(instr.name)}(virt_addr_t& pc, code_word_t instr, BasicBlock* bb){<%instr.code.eachLine{%>
+    std::tuple<continuation_e, BasicBlock*> __${generator.functionName(instr.name)}(virt_addr_t& pc, code_word_t instr, BasicBlock* bb){
-    	${it}<%}%>
+        bb->setName(fmt::format("${instr.name}_0x{:X}",pc.val));
+        this->gen_sync(PRE_SYNC,${idx});
+        uint64_t PC = pc.val;
+        <%instr.fields.eachLine{%>${it}
+        <%}%>if(this->disass_enabled){
+            /* generate console output when executing the command */<%instr.disass.eachLine{%>
+            ${it}<%}%>
+        }
+        auto cur_pc_val = this->gen_const(32,pc.val);
+        pc=pc+ ${instr.length/8};
+        this->gen_set_pc(pc, traits::NEXT_PC);
+        <%instr.behavior.eachLine{%>${it}
+        <%}%>
+        this->gen_trap_check(bb);
+    	this->gen_sync(POST_SYNC, ${idx});
+        this->builder.CreateBr(bb);
+    	return returnValue;        
     }
     <%}%>
     /****************************************************************************
@@ -203,23 +186,75 @@ 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(this->get_typeptr(traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), true),
+        this->builder.CreateStore(this->builder.CreateLoad(this->get_typeptr(traits::NEXT_PC), get_reg_ptr(traits::NEXT_PC), true),
-                                   get_reg_ptr(traits<ARCH>::PC), true);
+                                   get_reg_ptr(traits::PC), true);
         this->builder.CreateStore(
-            this->builder.CreateAdd(this->builder.CreateLoad(this->get_typeptr(traits<ARCH>::ICOUNT), get_reg_ptr(traits<ARCH>::ICOUNT), true),
+            this->builder.CreateAdd(this->builder.CreateLoad(this->get_typeptr(traits::ICOUNT), get_reg_ptr(traits::ICOUNT), true),
                                      this->gen_const(64U, 1)),
-            get_reg_ptr(traits<ARCH>::ICOUNT), true);
+            get_reg_ptr(traits::ICOUNT), true);
         pc = pc + ((instr & 3) == 3 ? 4 : 2);
         this->gen_raise_trap(0, 2);     // illegal instruction trap
 		this->gen_sync(iss::POST_SYNC, instr_descr.size());
         this->gen_trap_check(this->leave_blk);
         return std::make_tuple(BRANCH, nullptr);
+    }    
+    //decoding functionality
+
+    void populate_decoding_tree(decoding_tree_node* root){
+        //create submask
+        for(auto instr: root->instrs){
+            root->submask &= instr.mask;
+        }
+        //put each instr according to submask&encoding into children
+        for(auto instr: root->instrs){
+            bool foundMatch = false;
+            for(auto child: root->children){
+                //use value as identifying trait
+                if(child->value == (instr.value&root->submask)){
+                    child->instrs.push_back(instr);
+                    foundMatch = true;
+                }
+            }
+            if(!foundMatch){
+                decoding_tree_node* child = new decoding_tree_node(instr.value&root->submask);
+                child->instrs.push_back(instr);
+                root->children.push_back(child);
+            }
+        }
+        root->instrs.clear();
+        //call populate_decoding_tree for all children
+        if(root->children.size() >1)
+            for(auto child: root->children){
+                populate_decoding_tree(child);      
+            }
+        else{
+            //sort instrs by value of the mask, this works bc we want to have the least restrictive one last
+            std::sort(root->children[0]->instrs.begin(), root->children[0]->instrs.end(), [](const instruction_descriptor& instr1, const instruction_descriptor& instr2) {
+            return instr1.mask > instr2.mask;
+            }); 
+        }
+    }
+    compile_func decode_instr(decoding_tree_node* node, code_word_t word){
+        if(!node->children.size()){
+            if(node->instrs.size() == 1) return node->instrs[0].op;
+            for(auto instr : node->instrs){
+                if((instr.mask&word) == instr.value) return instr.op;
+            }
+        }
+        else{
+            for(auto child : node->children){
+                if (child->value == (node->submask&word)){
+                    return decode_instr(child, word);
+                }  
+            }  
+        }
+        return nullptr;
     }
 };
 
-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()); }
@@ -227,14 +262,11 @@ 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>
@@ -242,50 +274,50 @@ std::tuple<continuation_e, BasicBlock *>
 vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt, BasicBlock *this_block) {
     // 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<ARCH>::addr_t upper_bits = ~traits<ARCH>::PGMASK;
+    // const typename traits::addr_t upper_bits = ~traits::PGMASK;
     phys_addr_t paddr(pc);
-    auto *const data = (uint8_t *)&insn;
+    auto *const data = (uint8_t *)&instr;
-    paddr = this->core.v2p(pc);
+    if(this->core.has_mmu())
+        paddr = this->core.virt2phys(pc);
     //TODO: re-add page handling
 //    if ((pc.val & upper_bits) != ((pc.val + 2) & upper_bits)) { // we may cross a page boundary
 //        auto res = this->core.read(paddr, 2, data);
 //        if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
-//        if ((insn & 0x3) == 0x3) { // this is a 32bit instruction
+//        if ((instr & 0x3) == 0x3) { // this is a 32bit instruction
 //            res = this->core.read(this->core.v2p(pc + 2), 2, data + 2);
 //        }
 //    } else {
         auto res = this->core.read(paddr, 4, data);
         if (res != iss::Ok) throw trap_access(TRAP_ID, pc.val);
 //    }
-    if (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, this_block);
+    return (this->*f)(pc, instr, this_block);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_behavior(BasicBlock *leave_blk) {
     this->builder.SetInsertPoint(leave_blk);
-    this->builder.CreateRet(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(arch::traits<ARCH>::NEXT_PC), false));
+    this->builder.CreateRet(this->builder.CreateLoad(this->get_typeptr(traits::NEXT_PC),get_reg_ptr(traits::NEXT_PC), false));
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(uint16_t trap_id, uint16_t cause) {
     auto *TRAP_val = this->gen_const(32, 0x80 << 24 | (cause << 16) | trap_id);
-    this->builder.CreateStore(TRAP_val, get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
+    this->builder.CreateStore(TRAP_val, get_reg_ptr(traits::TRAP_STATE), true);
-    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits<ARCH>::LAST_BRANCH), false);
+    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits::LAST_BRANCH), false);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_trap(unsigned lvl) {
     std::vector<Value *> args{ this->core_ptr, ConstantInt::get(getContext(), APInt(64, lvl)) };
     this->builder.CreateCall(this->mod->getFunction("leave_trap"), args);
-    auto *PC_val = this->gen_read_mem(traits<ARCH>::CSR, (lvl << 8) + 0x41, traits<ARCH>::XLEN / 8);
+    auto *PC_val = this->gen_read_mem(traits::CSR, (lvl << 8) + 0x41, traits::XLEN / 8);
-    this->builder.CreateStore(PC_val, get_reg_ptr(traits<ARCH>::NEXT_PC), false);
+    this->builder.CreateStore(PC_val, get_reg_ptr(traits::NEXT_PC), false);
-    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits<ARCH>::LAST_BRANCH), false);
+    this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()), get_reg_ptr(traits::LAST_BRANCH), false);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_wait(unsigned type) {
@@ -295,22 +327,25 @@ 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(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(traits<ARCH>::TRAP_STATE), true);
+    this->gen_sync(POST_SYNC, -1); //TODO get right InstrId
+    auto *trap_state_val = this->builder.CreateLoad(this->get_typeptr(traits::TRAP_STATE), get_reg_ptr(traits::TRAP_STATE), true);
     this->builder.CreateStore(this->gen_const(32U, std::numeric_limits<uint32_t>::max()),
-                              get_reg_ptr(traits<ARCH>::LAST_BRANCH), false);
+                              get_reg_ptr(traits::LAST_BRANCH), false);
     std::vector<Value *> args{this->core_ptr, this->adj_to64(trap_state_val),
-                              this->adj_to64(this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::PC), get_reg_ptr(traits<ARCH>::PC), false))};
+                              this->adj_to64(this->builder.CreateLoad(this->get_typeptr(traits::PC), get_reg_ptr(traits::PC), false))};
     this->builder.CreateCall(this->mod->getFunction("enter_trap"), args);
-    auto *trap_addr_val = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::NEXT_PC), get_reg_ptr(traits<ARCH>::NEXT_PC), false);
+    auto *trap_addr_val = this->builder.CreateLoad(this->get_typeptr(traits::NEXT_PC), get_reg_ptr(traits::NEXT_PC), false);
     this->builder.CreateRet(trap_addr_val);
 }
 
 template <typename ARCH> inline void vm_impl<ARCH>::gen_trap_check(BasicBlock *bb) {
-    auto *v = this->builder.CreateLoad(this->get_typeptr(arch::traits<ARCH>::TRAP_STATE), get_reg_ptr(arch::traits<ARCH>::TRAP_STATE), true);
+    auto* target_bb = BasicBlock::Create(this->mod->getContext(), "", this->func, bb);
+    auto *v = this->builder.CreateLoad(this->get_typeptr(traits::TRAP_STATE), get_reg_ptr(traits::TRAP_STATE), true);
     this->gen_cond_branch(this->builder.CreateICmp(
                               ICmpInst::ICMP_EQ, v,
                               ConstantInt::get(getContext(), APInt(v->getType()->getIntegerBitWidth(), 0))),
-                          bb, this->trap_blk, 1);
+                          target_bb, this->trap_blk, 1);
+    this->builder.SetInsertPoint(target_bb);
 }
 
 } // namespace ${coreDef.name.toLowerCase()}

src/iss/arch/riscv_hart_m_p.h

@@ -1113,8 +1113,10 @@ iss::status riscv_hart_m_p<BASE, FEAT>::write_mem(phys_addr_t paddr, unsigned le
                         }
                         this->reg.trap_state=std::numeric_limits<uint32_t>::max();
                         this->interrupt_sim=hostvar;
+#ifndef WITH_TCC
+                        throw(iss::simulation_stopped(hostvar));
+#endif
                         break;
-                        //throw(iss::simulation_stopped(hostvar));
                     case 0x0101: {
                         char c = static_cast<char>(hostvar & 0xff);
                         if (c == '\n' || c == 0) {