diff --git a/src/engine/rv64/isa/mips32.c b/src/engine/rv64/isa/mips32.c
index 3abecdd8853904ec04335aa99583df7801ff3c5c..e5024a4f48b43cc374411bd4f5dc113a7ced07e8 100644
--- a/src/engine/rv64/isa/mips32.c
+++ b/src/engine/rv64/isa/mips32.c
@@ -68,10 +68,6 @@ uint32_t reg_ptr2idx(DecodeExecState *s, const rtlreg_t* dest) {
 
   // if not mapped, spill out one tmp_reg and remap
   idx = spill_out_and_remap(s, tmp_idx);
-
-  // for (int i = 0; i < TMP_REG_NUM; i++) {
-  //   printf("[%d] %d %d\n", i, tmp_regs[i].map_ptr, tmp_regs[i].used);
-  // }
   return idx;
 }
 
@@ -81,7 +77,7 @@ void guest_getregs(CPU_state *mips32) {
   int i;
   for (i = 0; i < 32; i ++) {
     switch (i) {
-      case 28: case 1: case 25: case 26: case 27: continue;
+      case 28: case 1: case TMP_REG_ADDR: case TMP_REG_1: case TMP_REG_2: continue;
     }
     mips32->gpr[i]._32 = r.gpr[i]._64;
   }
@@ -95,7 +91,7 @@ void guest_setregs(const CPU_state *mips32) {
   int i;
   for (i = 0; i < 32; i ++) {
     switch (i) {
-      case 28: case 1: case 25: case 26: case 27: continue;
+      case 28: case 1: case TMP_REG_ADDR: case TMP_REG_1: case TMP_REG_2: continue;
     }
     r.gpr[i]._64 = mips32->gpr[i]._32;
   }
diff --git a/src/engine/rv64/rv64-backend/exec.c b/src/engine/rv64/rv64-backend/exec.c
index 5d6f7dfb43fd3dc6eba7d1150d1debe7e2b6a20f..0271222c89cfaef907c982ae853e1afa4d7ebb81 100644
--- a/src/engine/rv64/rv64-backend/exec.c
+++ b/src/engine/rv64/rv64-backend/exec.c
@@ -1,5 +1,6 @@
 #include <isa/riscv64.h>
 #include "../tran.h"
+#include "../spill.h"
 
 #define RV64_EXEC_PC (riscv64_PMEM_BASE + BBL_MAX_SIZE) // skip bbl
 
@@ -19,7 +20,7 @@ vaddr_t rv64_exec_trans_buffer(void *buf, int nr_instr) {
   // if the basic block is end with a branch instruction,
   // execute until the branch instruction
   // see rtl_jrelop() at rtl-basic.c
-  int nr_exec = (tran_next_pc == NEXT_PC_BRANCH ? nr_instr - 5 : nr_instr);
+  int nr_exec = (tran_next_pc == NEXT_PC_BRANCH ? nr_instr - (5+2*suffix_inst) : nr_instr);
   backend_exec_code(RV64_EXEC_PC, nr_exec);
 
   riscv64_CPU_state r;
@@ -33,9 +34,13 @@ vaddr_t rv64_exec_trans_buffer(void *buf, int nr_instr) {
 
   if (tran_next_pc == NEXT_PC_BRANCH) {
     // execute the branch instruction and load x86.pc to x30
-    backend_exec(3);
+    backend_exec(suffix_inst+3);
     backend_getregs(&r);
   }
 
+#ifdef REG_SPILLING
+  tmp_regs_reset();
+#endif
+
   return r.gpr[tmp0]._64;
 }
diff --git a/src/engine/rv64/rv64-backend/rtl-basic.c b/src/engine/rv64/rv64-backend/rtl-basic.c
index 7b2ba0beefbfc6934ea3301a70b0a96b5b99f058..bd5e12961a469018cc53296d5ec51b139c9b4b30 100644
--- a/src/engine/rv64/rv64-backend/rtl-basic.c
+++ b/src/engine/rv64/rv64-backend/rtl-basic.c
@@ -358,22 +358,28 @@ make_rtl(host_sm, void *addr, const rtlreg_t *src1, int len) {
 
 // we use tmp0 to store x86.pc of the next basic block
 make_rtl(j, vaddr_t target) {
+#ifdef REG_SPILLING
   spill_out_all();
+#endif
   if (!load_imm_big(tmp0, target)) rv64_addiw(tmp0, tmp0, target & 0xfff);
   tran_next_pc = NEXT_PC_JMP;
 }
 
 make_rtl(jr, rtlreg_t *target) {
-  spill_out_all();
   rv64_addi(tmp0, reg_ptr2idx(s, target), 0);
+#ifdef REG_SPILLING
+  spill_out_all();
+#endif
   tran_next_pc = NEXT_PC_JMP;
 }
 
 make_rtl(jrelop, uint32_t relop, const rtlreg_t *src1, const rtlreg_t *src2, vaddr_t target) {
-  spill_out_all();
   uint32_t rs1 = reg_ptr2idx(s, src1);
   uint32_t rs2 = reg_ptr2idx(s, src2);
-  uint32_t offset = 12; // branch two instructions
+#ifdef REG_SPILLING
+  cal_suffix_inst();
+#endif
+  uint32_t offset = 12 + 4*suffix_inst; // branch two instructions
   extern int trans_buffer_index;
   int old_idx = trans_buffer_index;
 
@@ -397,13 +403,19 @@ make_rtl(jrelop, uint32_t relop, const rtlreg_t *src1, const rtlreg_t *src2, vad
 
   // generate instrutions to load the not-taken target
   load_imm_no_opt(tmp0, s->seq_pc);  // only two instructions
+#ifdef REG_SPILLING
+  spill_out_all();
+#endif
   // generate instrutions to load the taken target
   load_imm_no_opt(tmp0, target);     // only two instructions
+#ifdef REG_SPILLING
+  spill_out_all();
+#endif
 
   tran_next_pc = NEXT_PC_BRANCH;
 
   int new_idx = trans_buffer_index;
-  Assert(new_idx - old_idx == 5, "if this condition is broken, "
+  Assert(new_idx - old_idx == suffix_inst*2+5, "if this condition is broken, "
       "you should also modify rv64_exec_trans_buffer() in exec.c");
 }
 
diff --git a/src/engine/rv64/spill.c b/src/engine/rv64/spill.c
index 9f21575b5b35d3b6f66c0a49ce5e21b8108844c8..c14d8991dbec34f68b564c35fc0bddd25a295e33 100644
--- a/src/engine/rv64/spill.c
+++ b/src/engine/rv64/spill.c
@@ -7,25 +7,23 @@
 
 Tmp_reg tmp_regs[TMP_REG_NUM];
 Tmp_reg spill_tmp_reg;
+uint32_t suffix_inst = 0;
 
-static inline bool load_imm_big(uint32_t r, const uint32_t imm) {
+static inline void load_imm_no_opt(uint32_t r, const sword_t imm) {
   RV_IMM rv_imm = { .val = imm };
   uint32_t lui_imm = rv_imm.imm_31_12 + (rv_imm.imm_11_0 >> 11);
-  if (lui_imm == 0) return false;
-  else {
-    rv64_lui(r, lui_imm);
-    if (rv_imm.imm_11_0 != 0) rv64_addiw(r, r, rv_imm.imm_11_0);
-    return true;
-  }
+  rv64_lui(r, lui_imm);
+  rv64_addiw(r, r, rv_imm.imm_11_0);
 }
 
 void tmp_regs_init() {
+    suffix_inst = 0;
     if (TMP_REG_NUM == 2) {
         tmp_regs[0].idx = TMP_REG_1;
-        tmp_regs[0].map_ptr = 3;
+        tmp_regs[0].map_ptr = 0;
         tmp_regs[0].used = 0;
         tmp_regs[1].idx = TMP_REG_2;
-        tmp_regs[1].map_ptr = 4;
+        tmp_regs[1].map_ptr = 0;
         tmp_regs[1].used = 0;
     } else {
         panic("Other TMP_REG_NUM!\n");
@@ -34,6 +32,17 @@ void tmp_regs_init() {
     spill_tmp_reg.map_ptr = 2;
 }
 
+void tmp_regs_reset() {
+    if (TMP_REG_NUM == 2) {
+        tmp_regs[0].map_ptr = 0;
+        tmp_regs[0].used = 0;
+        tmp_regs[1].map_ptr = 0;
+        tmp_regs[1].used = 0;
+    } else {
+        panic("Other TMP_REG_NUM!\n");
+    }
+}
+
 uint32_t check_tmp_reg(uint32_t tmp_idx) {
     for (int i = 0; i < TMP_REG_NUM; i++) {
         if (tmp_regs[i].map_ptr == tmp_idx) {
@@ -59,11 +68,11 @@ uint32_t spill_out_and_remap(DecodeExecState *s, uint32_t tmp_idx) {
     }
 
     addr = SCRATCHPAD_BASE_ADDR + 4 * (tmp_regs[ptr].map_ptr);
-    load_imm_big(spill_tmp_reg.idx, addr);
+    load_imm_no_opt(spill_tmp_reg.idx, addr);
     rv64_sw(tmp_regs[ptr].idx, spill_tmp_reg.idx, 0);
-    
+
     addr = SCRATCHPAD_BASE_ADDR + 4 * (tmp_idx);
-    load_imm_big(spill_tmp_reg.idx, addr);
+    load_imm_no_opt(spill_tmp_reg.idx, addr);
     rv64_lw(tmp_regs[ptr].idx, spill_tmp_reg.idx, 0);
 
     tmp_regs[ptr].map_ptr = tmp_idx;
@@ -72,21 +81,23 @@ uint32_t spill_out_and_remap(DecodeExecState *s, uint32_t tmp_idx) {
     return tmp_regs[ptr].idx;
 }
 
-void spill_out_all() {
+void cal_suffix_inst() {
+    suffix_inst = 0;
+    for (int i = 0; i < TMP_REG_NUM; i++) {
+        if (tmp_regs[i].map_ptr != 0) {
+            suffix_inst += 3;
+        }
+    }
+}
+
+void spill_out_all() {  // can be 0/3/6 inst
     uint32_t addr;
     for (int i = 0; i < TMP_REG_NUM; i++) {
-        addr = SCRATCHPAD_BASE_ADDR + 4 * (tmp_regs[i].map_ptr);
-        //printf("used: %d, %x\n", tmp_regs[i].used, addr);
-        load_imm_big(spill_tmp_reg.idx, addr);
-        rv64_sw(tmp_regs[i].idx, spill_tmp_reg.idx, 0);
-        
-        tmp_regs[i].map_ptr = 3+i;
-
-        addr = SCRATCHPAD_BASE_ADDR + 4 * (tmp_regs[i].map_ptr);
-        load_imm_big(spill_tmp_reg.idx, addr);
-        rv64_lw(tmp_regs[i].idx, spill_tmp_reg.idx, 0);
-
-        tmp_regs[i].used = 0;
+        if (tmp_regs[i].map_ptr != 0) {
+            addr = SCRATCHPAD_BASE_ADDR + 4 * (tmp_regs[i].map_ptr);
+            load_imm_no_opt(spill_tmp_reg.idx, addr);
+            rv64_sw(tmp_regs[i].idx, spill_tmp_reg.idx, 0);
+        }
     }
 }
 
@@ -97,4 +108,10 @@ void spill_clean(uint32_t tmp_idx) {
             return;
         }
     }
+}
+
+void spill_cleanall() {
+    for (int i = 0; i < TMP_REG_NUM; i++) {
+      tmp_regs[i].used = 0;
+    }
 }
\ No newline at end of file
diff --git a/src/engine/rv64/spill.h b/src/engine/rv64/spill.h
index 72c0a5e5ca84f731b6c3b57a2f82ab2cab8be89f..962ebac95ec15141c0cce161ef41097048e48a28 100644
--- a/src/engine/rv64/spill.h
+++ b/src/engine/rv64/spill.h
@@ -23,11 +23,15 @@ typedef struct {
 
 extern Tmp_reg tmp_regs[TMP_REG_NUM];
 extern Tmp_reg spill_tmp_reg;
+extern uint32_t suffix_inst;
 
 void tmp_regs_init();
+void tmp_regs_reset();
 uint32_t check_tmp_reg(uint32_t);
 uint32_t spill_out_and_remap(DecodeExecState*, uint32_t);
 void spill_clean(uint32_t);
+void spill_cleanall();
+void cal_suffix_inst();
 void spill_out_all();
 
 #endif
\ No newline at end of file
diff --git a/src/engine/rv64/tran.c b/src/engine/rv64/tran.c
index 2bcae5a024a39241701b976cdee604ef27856c93..8f998c4e56d09944e1a92b90810ed01717a3d54d 100644
--- a/src/engine/rv64/tran.c
+++ b/src/engine/rv64/tran.c
@@ -19,7 +19,6 @@ void guest_getregs(CPU_state *cpu);
 
 void write_ins(uint32_t ins) {
   assert(trans_buffer_index < BUF_SIZE);
-  //printf("instr: %x\n", ins);
   trans_buffer[trans_buffer_index++]=ins;
 }
 
@@ -33,11 +32,8 @@ void mainloop() {
     __attribute__((unused)) vaddr_t ori_pc = cpu.pc;
     __attribute__((unused)) vaddr_t seq_pc = isa_exec_once();
 #ifdef REG_SPILLING
-    for (int i = 0; i < TMP_REG_NUM; i++) {
-      tmp_regs[i].used = 0;
-    }
+    spill_cleanall();
 #endif
-    //printf("PC: %x\n", seq_pc);
 
     if (nemu_state.state != NEMU_RUNNING) tran_next_pc = NEXT_PC_END;
 
@@ -47,9 +43,6 @@ void mainloop() {
 
 #ifndef DIFF_TEST
     if (tran_next_pc != NEXT_PC_SEQ) {
-#endif
-#ifdef REG_SPILLING
-      tmp_regs_init();
 #endif
       vaddr_t next_pc = rv64_exec_trans_buffer(trans_buffer, trans_buffer_index);
       total_instr += trans_buffer_index;
diff --git a/src/isa/riscv64/exec/exec.c b/src/isa/riscv64/exec/exec.c
index d5b5828292ece93237456615fa694f5e20abee10..98c0adda7097ad572a21a00469fbf6749573e358 100644
--- a/src/isa/riscv64/exec/exec.c
+++ b/src/isa/riscv64/exec/exec.c
@@ -176,7 +176,6 @@ static inline void exec(DecodeExecState *s) {
     // in-page instructions, fetch 4 byte and
     // see whether it is an RVC instruction later
     s->isa.instr.val = instr_fetch(&s->seq_pc, 4);
-    //printf("[instr]%x\n", s->isa.instr.val);
   }
 
   check_mem_ex();