regenerated sources and and add opcode enum to headers

Conflicts:
	gen_input/CoreDSL-Instruction-Set-Description

gen_input/CoreDSL-Instruction-Set-Description

@@ -1 +1 @@
-Subproject commit 3cfac1d86f6532140a04cdfc1501f1b8f3729632
+Subproject commit 89763fcb973838a634e48ee05dca28782807b344

gen_input/templates/interp/CORENAME.cpp.gtl

@@ -33,7 +33,7 @@
 def getRegisterSizes(){
 	def regs = registers.collect{it.size}
 	regs[-1]=64 // correct for NEXT_PC
-	regs+=[32, 32, 32, 32, 64] // append TRAP_STATE, PENDING_TRAP, MACHINE_STATE, LAST_BRANCH, ICOUNT
+	regs+=[32, 32, 64] // append TRAP_STATE, PENDING_TRAP, ICOUNT
     return regs
 }
 %>

gen_input/templates/interp/CORENAME.h.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2020 MINRES Technologies GmbH
+ * Copyright (C) 2017 - 2021 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -30,21 +30,22 @@
  *
  *******************************************************************************/
 <%
+def nativeTypeSize(int size){
+    if(size<=8) return 8; else if(size<=16) return 16; else if(size<=32) return 32; else return 64;
+}
 def getRegisterSizes(){
-    def regs = registers.collect{it.size}
+    def regs = registers.collect{nativeTypeSize(it.size)}
-    regs[-1]=pc.size // correct for NEXT_PC
+    regs+=[32,32, 64] // append TRAP_STATE, PENDING_TRAP, ICOUNT
-    regs+=[32, 32, 32, 32, 64] // append TRAP_STATE, PENDING_TRAP, MACHINE_STATE, LAST_BRANCH, ICOUNT
     return regs
 }
 def getRegisterOffsets(){
-    def regs = registers.collect{it.offset}
+    def offset = 0
-    def offs= regs[-1]
+    def offsets = []
-     // append TRAP_STATE, PENDING_TRAP, MACHINE_STATE, LAST_BRANCH, ICOUNT offsets starting with NEXT_PC size
+    getRegisterSizes().each { size ->
-    [pc.size/8, 4, 4, 4, 4].each{ sz ->
+        offsets<<offset
-        regs+=offs+sz
+        offset+=size/8
-        offs+=sz
     }
-    return regs
+    return offsets
 }
 def byteSize(int size){
     if(size<=8) return 8;
@@ -81,15 +82,10 @@ template <> struct traits<${coreDef.name.toLowerCase()}> {
 
     constexpr static unsigned FP_REGS_SIZE = ${constants.find {it.name=='FLEN'}?.value?:0};
 
-    enum reg_e {<%
+    enum reg_e {
-        registers.each { reg -> %>
+        ${registers.collect{it.name}.join(', ')}, NUM_REGS,
-        ${reg.name},<%  
+        TRAP_STATE=NUM_REGS,
-        }%>
-        NEXT_${pc.name}=NUM_REGS,
-        TRAP_STATE,
         PENDING_TRAP,
-        MACHINE_STATE,
-        LAST_BRANCH,
         ICOUNT
     };
 
@@ -168,7 +164,6 @@ protected:
         registers.each { reg -> if(reg.size>0) {%> 
         uint${byteSize(reg.size)}_t ${reg.name} = 0;<%
         }}%>
-        uint${byteSize(pc.size)}_t NEXT_${pc.name} = 0;
         uint32_t trap_state = 0, pending_trap = 0, machine_state = 0, last_branch = 0;
         uint64_t icount = 0;
     } reg;

gen_input/templates/interp/vm_CORENAME.cpp.gtl

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2020 MINRES Technologies GmbH
+ * Copyright (C) 2021 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -149,13 +149,14 @@ protected:
         this->core.leave_trap(lvl);
         auto pc_val = super::template read_mem<reg_t>(traits::CSR, (lvl << 8) + 0x41);
         this->template get_reg<reg_t>(traits::NEXT_PC) = pc_val;
-        this->template get_reg<uint32_t>(traits::LAST_BRANCH) = std::numeric_limits<uint32_t>::max();
     }
 
     void wait(unsigned type){
         this->core.wait_until(type);
     }
 
+    template<typename T>
+    T& pc_assign(T& val){super::ex_info.branch_taken=true; return val;}
     inline uint8_t readSpace1(typename super::mem_type_e space, uint64_t addr){return super::template read_mem<uint8_t>(space, addr);}
     inline uint16_t readSpace2(typename super::mem_type_e space, uint64_t addr){return super::template read_mem<uint16_t>(space, addr);}
     inline uint32_t readSpace4(typename super::mem_type_e space, uint64_t addr){return super::template read_mem<uint32_t>(space, addr);}
@@ -196,17 +197,15 @@ private:
         // prepare execution
         uint${addrDataWidth}_t* X = reinterpret_cast<uint${addrDataWidth}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::X0]);
         uint${addrDataWidth}_t* PC = reinterpret_cast<uint${addrDataWidth}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::PC]);
+        uint${addrDataWidth}_t* NEXT_PC = reinterpret_cast<uint${addrDataWidth}_t*>(this->regs_base_ptr+arch::traits<ARCH>::reg_byte_offsets[arch::traits<ARCH>::NEXT_PC]);
+        *NEXT_PC = *PC + ${instr.length/8};
         // execute instruction
         <%instr.behavior.eachLine{%>${it}
-        <%}%>// post execution stuff<% if(instr.modifiesPC) { %>
+        <%}%>// post execution stuff
-        super::template get_reg<reg_t>(arch::traits<ARCH>::NEXT_PC) = super::template get_reg<reg_t>(arch::traits<ARCH>::PC);<% } else { %>
-        super::template get_reg<reg_t>(arch::traits<ARCH>::NEXT_PC) = pc.val + ${instr.length/8};<% } %>
         if(this->sync_exec && POST_SYNC) this->do_sync(POST_SYNC, ${idx});
         auto& trap_state = super::template get_reg<uint32_t>(arch::traits<ARCH>::TRAP_STATE);
         // trap check
         if(trap_state!=0){
-            auto& last_br = super::template get_reg<uint32_t>(arch::traits<ARCH>::LAST_BRANCH);
-            last_br = std::numeric_limits<uint32_t>::max();
             super::core.enter_trap(trap_state, pc.val);
         }
         pc.val=super::template get_reg<reg_t>(arch::traits<ARCH>::NEXT_PC);

incl/iss/arch/riscv_hart_m_p.h

@@ -302,7 +302,7 @@ public:
 
         void write_mstatus(T val) {
             auto mask = get_mask();
-            auto new_val = (mstatus.st.value & ~mask) | (val & mask);
+            auto new_val = (mstatus.storage.val & ~mask) | (val & mask);
             mstatus = new_val;
         }
 

incl/iss/arch/tgf_b.h

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2020 MINRES Technologies GmbH
+ * Copyright (C) 2017 - 2021 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -48,56 +48,19 @@ template <> struct traits<tgf_b> {
     constexpr static char const* const core_type = "TGF_B";
     
     static constexpr std::array<const char*, 35> reg_names{
-        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "PRIV", "NUM_REGS"}};
+        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "NEXT_PC", "PRIV"}};
  
     static constexpr std::array<const char*, 35> reg_aliases{
-        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "PRIV", "NUM_REGS"}};
+        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "NEXT_PC", "PRIV"}};
 
     enum constants {XLEN=32, PCLEN=32, MISA_VAL=0b1000000000000000000000100000000, PGSIZE=0x1000, PGMASK=0xfff, CSR_SIZE=4096, fence=0, fencei=1, fencevmal=2, fencevmau=3};
 
     constexpr static unsigned FP_REGS_SIZE = 0;
 
     enum reg_e {
-        X0,
+        X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X12, X13, X14, X15, X16, X17, X18, X19, X20, X21, X22, X23, X24, X25, X26, X27, X28, X29, X30, X31, PC, NEXT_PC, PRIV, NUM_REGS,
-        X1,
+        TRAP_STATE=NUM_REGS,
-        X2,
-        X3,
-        X4,
-        X5,
-        X6,
-        X7,
-        X8,
-        X9,
-        X10,
-        X11,
-        X12,
-        X13,
-        X14,
-        X15,
-        X16,
-        X17,
-        X18,
-        X19,
-        X20,
-        X21,
-        X22,
-        X23,
-        X24,
-        X25,
-        X26,
-        X27,
-        X28,
-        X29,
-        X30,
-        X31,
-        PC,
-        PRIV,
-        NUM_REGS,
-        NEXT_PC=NUM_REGS,
-        TRAP_STATE,
         PENDING_TRAP,
-        MACHINE_STATE,
-        LAST_BRANCH,
         ICOUNT
     };
 
@@ -111,11 +74,11 @@ template <> struct traits<tgf_b> {
 
     using phys_addr_t = iss::typed_addr_t<iss::address_type::PHYSICAL>;
 
-    static constexpr std::array<const uint32_t, 40> reg_bit_widths{
+    static constexpr std::array<const uint32_t, 38> reg_bit_widths{
-        {32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,2,32,32,32,32,32,64}};
+        {32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,8,32,32,64}};
 
-    static constexpr std::array<const uint32_t, 40> reg_byte_offsets{
+    static constexpr std::array<const uint32_t, 38> reg_byte_offsets{
-        {0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,133,137,141,145,149,153}};
+        {0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,137,141,145}};
 
     static const uint64_t addr_mask = (reg_t(1) << (XLEN - 1)) | ((reg_t(1) << (XLEN - 1)) - 1);
 
@@ -257,8 +220,8 @@ protected:
         uint32_t X30 = 0; 
         uint32_t X31 = 0; 
         uint32_t PC = 0; 
-        uint8_t PRIV = 0;
         uint32_t NEXT_PC = 0; 
+        uint8_t PRIV = 0;
         uint32_t trap_state = 0, pending_trap = 0, machine_state = 0, last_branch = 0;
         uint64_t icount = 0;
     } reg;

incl/iss/arch/tgf_c.h

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (C) 2017 - 2020 MINRES Technologies GmbH
+ * Copyright (C) 2017 - 2021 MINRES Technologies GmbH
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -48,56 +48,19 @@ template <> struct traits<tgf_c> {
     constexpr static char const* const core_type = "TGF_C";
     
     static constexpr std::array<const char*, 35> reg_names{
-        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "PRIV", "NUM_REGS"}};
+        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "NEXT_PC", "PRIV"}};
  
     static constexpr std::array<const char*, 35> reg_aliases{
-        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "PRIV", "NUM_REGS"}};
+        {"X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", "X8", "X9", "X10", "X11", "X12", "X13", "X14", "X15", "X16", "X17", "X18", "X19", "X20", "X21", "X22", "X23", "X24", "X25", "X26", "X27", "X28", "X29", "X30", "X31", "PC", "NEXT_PC", "PRIV"}};
 
     enum constants {XLEN=32, PCLEN=32, MISA_VAL=0b1000000000000000001000100000100, PGSIZE=0x1000, PGMASK=0xfff, CSR_SIZE=4096, fence=0, fencei=1, fencevmal=2, fencevmau=3, MUL_LEN=64};
 
     constexpr static unsigned FP_REGS_SIZE = 0;
 
     enum reg_e {
-        X0,
+        X0, X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X12, X13, X14, X15, X16, X17, X18, X19, X20, X21, X22, X23, X24, X25, X26, X27, X28, X29, X30, X31, PC, NEXT_PC, PRIV, NUM_REGS,
-        X1,
+        TRAP_STATE=NUM_REGS,
-        X2,
-        X3,
-        X4,
-        X5,
-        X6,
-        X7,
-        X8,
-        X9,
-        X10,
-        X11,
-        X12,
-        X13,
-        X14,
-        X15,
-        X16,
-        X17,
-        X18,
-        X19,
-        X20,
-        X21,
-        X22,
-        X23,
-        X24,
-        X25,
-        X26,
-        X27,
-        X28,
-        X29,
-        X30,
-        X31,
-        PC,
-        PRIV,
-        NUM_REGS,
-        NEXT_PC=NUM_REGS,
-        TRAP_STATE,
         PENDING_TRAP,
-        MACHINE_STATE,
-        LAST_BRANCH,
         ICOUNT
     };
 
@@ -111,11 +74,11 @@ template <> struct traits<tgf_c> {
 
     using phys_addr_t = iss::typed_addr_t<iss::address_type::PHYSICAL>;
 
-    static constexpr std::array<const uint32_t, 40> reg_bit_widths{
+    static constexpr std::array<const uint32_t, 38> reg_bit_widths{
-        {32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,2,32,32,32,32,32,64}};
+        {32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,8,32,32,64}};
 
-    static constexpr std::array<const uint32_t, 40> reg_byte_offsets{
+    static constexpr std::array<const uint32_t, 38> reg_byte_offsets{
-        {0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,133,137,141,145,149,153}};
+        {0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,137,141,145}};
 
     static const uint64_t addr_mask = (reg_t(1) << (XLEN - 1)) | ((reg_t(1) << (XLEN - 1)) - 1);
 
@@ -293,8 +256,8 @@ protected:
         uint32_t X30 = 0; 
         uint32_t X31 = 0; 
         uint32_t PC = 0; 
-        uint8_t PRIV = 0;
         uint32_t NEXT_PC = 0; 
+        uint8_t PRIV = 0;
         uint32_t trap_state = 0, pending_trap = 0, machine_state = 0, last_branch = 0;
         uint64_t icount = 0;
     } reg;

incl/iss/plugin/cycle_estimate.h

@@ -76,7 +76,7 @@ public:
 
     sync_type get_sync() override { return POST_SYNC; };
 
-    void callback(instr_info_t instr_info) override;
+    void callback(instr_info_t instr_info, exec_info const&) override;
 
 private:
     iss::instrumentation_if *arch_instr;

incl/iss/plugin/instruction_count.h

@@ -69,7 +69,7 @@ public:
 
     sync_type get_sync() override { return POST_SYNC; };
 
-    void callback(instr_info_t instr_info) override;
+    void callback(instr_info_t, exec_info const&) override;
 
 private:
     Json::Value root;

src/iss/tgf_b.cpp

@@ -41,8 +41,8 @@ using namespace iss::arch;
 
 constexpr std::array<const char*, 35>    iss::arch::traits<iss::arch::tgf_b>::reg_names;
 constexpr std::array<const char*, 35>    iss::arch::traits<iss::arch::tgf_b>::reg_aliases;
-constexpr std::array<const uint32_t, 40> iss::arch::traits<iss::arch::tgf_b>::reg_bit_widths;
+constexpr std::array<const uint32_t, 38> iss::arch::traits<iss::arch::tgf_b>::reg_bit_widths;
-constexpr std::array<const uint32_t, 40> iss::arch::traits<iss::arch::tgf_b>::reg_byte_offsets;
+constexpr std::array<const uint32_t, 38> iss::arch::traits<iss::arch::tgf_b>::reg_byte_offsets;
 
 tgf_b::tgf_b() {
     reg.icount = 0;

src/iss/tgf_c.cpp

@@ -41,8 +41,8 @@ using namespace iss::arch;
 
 constexpr std::array<const char*, 35>    iss::arch::traits<iss::arch::tgf_c>::reg_names;
 constexpr std::array<const char*, 35>    iss::arch::traits<iss::arch::tgf_c>::reg_aliases;
-constexpr std::array<const uint32_t, 40> iss::arch::traits<iss::arch::tgf_c>::reg_bit_widths;
+constexpr std::array<const uint32_t, 38> iss::arch::traits<iss::arch::tgf_c>::reg_bit_widths;
-constexpr std::array<const uint32_t, 40> iss::arch::traits<iss::arch::tgf_c>::reg_byte_offsets;
+constexpr std::array<const uint32_t, 38> iss::arch::traits<iss::arch::tgf_c>::reg_byte_offsets;
 
 tgf_c::tgf_c() {
     reg.icount = 0;

src/plugin/cycle_estimate.cpp

@@ -83,7 +83,7 @@ bool iss::plugin::cycle_estimate::registration(const char* const version, vm_if&
 
 }
 
-void iss::plugin::cycle_estimate::callback(instr_info_t instr_info) {
+void iss::plugin::cycle_estimate::callback(instr_info_t instr_info, exec_info const&) {
     assert(arch_instr && "No instrumentation interface available but callback executed");
 	auto entry = delays[instr_info.instr_id];
 	bool taken = (arch_instr->get_next_pc()-arch_instr->get_pc()) != (entry.size/8);

src/plugin/instruction_count.cpp

@@ -90,6 +90,6 @@ bool iss::plugin::instruction_count::registration(const char* const version, vm_
 	return true;
 }
 
-void iss::plugin::instruction_count::callback(instr_info_t instr_info) {
+void iss::plugin::instruction_count::callback(instr_info_t instr_info, exec_info const&) {
 	rep_counts[instr_info.instr_id]++;
 }

src/vm/tcc/vm_tgf_b.cpp

@@ -377,7 +377,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 2);
         gen_trap_check(tu);
@@ -424,7 +423,6 @@ private:
             new_pc_val,
             tu.l_not(tu.constant(0x1, 32U))), 32);
         tu.store(PC_val_v, traits<ARCH>::NEXT_PC);
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32U), traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 3);
@@ -466,7 +464,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 4);
         gen_trap_check(tu);
@@ -507,7 +504,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 5);
         gen_trap_check(tu);
@@ -552,7 +548,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 6);
         gen_trap_check(tu);
@@ -597,7 +592,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 7);
         gen_trap_check(tu);
@@ -638,7 +632,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 8);
         gen_trap_check(tu);
@@ -679,7 +672,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 9);
         gen_trap_check(tu);
@@ -1621,7 +1613,6 @@ private:
             tu.constant(1, 64U),
             tu.trunc(tu.constant(imm, 32U), 32));
         tu.close_scope();
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
         gen_set_pc(tu, pc, traits<ARCH>::NEXT_PC);
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 38);
         gen_trap_check(tu);
@@ -2055,13 +2046,11 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {
     tu("  *trap_state = {:#x};", 0x80 << 24 | (cause << 16) | trap_id);
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_trap(tu_builder& tu, unsigned lvl) {
     tu("leave_trap(core_ptr, {});", lvl);
     tu.store(tu.read_mem(traits<ARCH>::CSR, (lvl << 8) + 0x41, traits<ARCH>::XLEN),traits<ARCH>::NEXT_PC);
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_wait(tu_builder& tu, unsigned type) {
@@ -2070,7 +2059,6 @@ template <typename ARCH> void vm_impl<ARCH>::gen_wait(tu_builder& tu, unsigned t
 template <typename ARCH> void vm_impl<ARCH>::gen_trap_behavior(tu_builder& tu) {
     tu("trap_entry:");
     tu("enter_trap(core_ptr, *trap_state, *pc);");
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(),32),traits<ARCH>::LAST_BRANCH);
     tu("return *next_pc;");
 }
 

src/vm/tcc/vm_tgf_c.cpp

@@ -449,7 +449,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 2);
         gen_trap_check(tu);
@@ -487,7 +486,6 @@ private:
             new_pc_val,
             tu.l_not(tu.constant(0x1, 32U))), 32);
         tu.store(PC_val_v, traits<ARCH>::NEXT_PC);
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32U), traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 3);
         gen_trap_check(tu);
@@ -528,7 +526,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 4);
         gen_trap_check(tu);
@@ -569,7 +566,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 5);
         gen_trap_check(tu);
@@ -614,7 +610,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 6);
         gen_trap_check(tu);
@@ -659,7 +654,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 7);
         gen_trap_check(tu);
@@ -700,7 +694,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 8);
         gen_trap_check(tu);
@@ -741,7 +734,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 9);
         gen_trap_check(tu);
@@ -1683,7 +1675,6 @@ private:
             tu.constant(1, 64U),
             tu.trunc(tu.constant(imm, 32U), 32));
         tu.close_scope();
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
         gen_set_pc(tu, pc, traits<ARCH>::NEXT_PC);
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 38);
         gen_trap_check(tu);
@@ -2562,7 +2553,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 65);
         gen_trap_check(tu);
@@ -2868,7 +2858,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 76);
         gen_trap_check(tu);
@@ -2908,7 +2897,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 77);
         gen_trap_check(tu);
@@ -2948,7 +2936,6 @@ private:
         auto is_cont_v = tu.choose(
             tu.icmp(ICmpInst::ICMP_NE, tu.ext(PC_val_v, 32U, true), tu.constant(pc.val, 32U)),
             tu.constant(0U, 32), tu.constant(1U, 32));
-        tu.store(is_cont_v, traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 78);
         gen_trap_check(tu);
@@ -3055,7 +3042,6 @@ private:
         tu.open_scope();
         auto PC_val_v = tu.assignment("PC_val", tu.load(rs1 + traits<ARCH>::X0, 0), 32);
         tu.store(PC_val_v, traits<ARCH>::NEXT_PC);
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32U), traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 82);
         gen_trap_check(tu);
@@ -3108,7 +3094,6 @@ private:
             tu.constant(2, 32U)), 1 + traits<ARCH>::X0);
         auto PC_val_v = tu.assignment("PC_val", tu.load(rs1 + traits<ARCH>::X0, 0), 32);
         tu.store(PC_val_v, traits<ARCH>::NEXT_PC);
-        tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32U), traits<ARCH>::LAST_BRANCH);
         tu.close_scope();
         vm_base<ARCH>::gen_sync(tu, POST_SYNC, 84);
         gen_trap_check(tu);
@@ -3247,13 +3232,11 @@ vm_impl<ARCH>::gen_single_inst_behavior(virt_addr_t &pc, unsigned int &inst_cnt,
 
 template <typename ARCH> void vm_impl<ARCH>::gen_raise_trap(tu_builder& tu, uint16_t trap_id, uint16_t cause) {
     tu("  *trap_state = {:#x};", 0x80 << 24 | (cause << 16) | trap_id);
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_leave_trap(tu_builder& tu, unsigned lvl) {
     tu("leave_trap(core_ptr, {});", lvl);
     tu.store(tu.read_mem(traits<ARCH>::CSR, (lvl << 8) + 0x41, traits<ARCH>::XLEN),traits<ARCH>::NEXT_PC);
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(), 32),traits<ARCH>::LAST_BRANCH);
 }
 
 template <typename ARCH> void vm_impl<ARCH>::gen_wait(tu_builder& tu, unsigned type) {
@@ -3262,7 +3245,6 @@ template <typename ARCH> void vm_impl<ARCH>::gen_wait(tu_builder& tu, unsigned t
 template <typename ARCH> void vm_impl<ARCH>::gen_trap_behavior(tu_builder& tu) {
     tu("trap_entry:");
     tu("enter_trap(core_ptr, *trap_state, *pc);");
-    tu.store(tu.constant(std::numeric_limits<uint32_t>::max(),32),traits<ARCH>::LAST_BRANCH);
     tu("return *next_pc;");
 }