engine,rv64,spill: no need to writeback s0/s1

* they are only valid during the translation of a single instruction

src/engine/rv64/spill.c

@@ -16,11 +16,14 @@ void spill_init() {
   tmp_regs[1].rvidx = TMP_REG_2;
 }
 
+void spill_flush(int tmpidx) {
+  tmp_regs[tmpidx].spmidx = 0;
+  tmp_regs[tmpidx].dirty = 0;
+}
+
 void spill_reset() {
-  tmp_regs[0].spmidx = 0;
-  tmp_regs[0].dirty = 0;
-  tmp_regs[1].spmidx = 0;
-  tmp_regs[1].dirty = 0;
+  spill_flush(0);
+  spill_flush(1);
 }
 
 int spmidx2tmpidx(uint32_t spmidx) {
@@ -146,3 +149,19 @@ void spill_set_dirty_rvidx(uint32_t rvidx) {
   int tmpidx = rvidx2tmpidx(rvidx);
   if (tmpidx != -1) spill_set_dirty(tmpidx);
 }
+
+// this will be called after every translation of an instruction
+// to flush RTL tmp registers, since their life-cycle is only
+// valid during the translation of a single instruction
+static void spill_flush_local_internal(DecodeExecState *s, const rtlreg_t *dest) {
+  uint32_t varidx = rtlreg2varidx(s, dest);
+  for (int i = 0; i < TMP_REG_NUM; i++) {
+    if (tmp_regs[i].spmidx == varidx && tmp_regs[i].dirty) spill_flush(i);
+  }
+}
+void spill_flush_local() {
+  DecodeExecState state; // only used in rtlreg2varidx()
+  DecodeExecState *s = &state;
+  spill_flush_local_internal(s, s0);
+  spill_flush_local_internal(s, s1);
+}

src/engine/rv64/spill.h

@@ -12,7 +12,6 @@
 
 uint32_t spmidx2rvidx(uint32_t);
 uint32_t spill_out_and_remap(DecodeExecState*, uint32_t);
-void spill_flush(uint32_t);
 void spill_flush_all();
 void cal_suffix_inst();
 void spill_writeback_all();

src/engine/rv64/tran.c

@@ -18,6 +18,7 @@ void asm_print(vaddr_t ori_pc, int instr_len, bool print_flag);
 vaddr_t rv64_exec_trans_buffer(void *buf, int nr_instr, int npc_type);
 void guest_getregs(CPU_state *cpu);
 void spill_reset();
+void spill_flush_local();
 
 typedef struct TB {
   vaddr_t pc;
@@ -103,6 +104,8 @@ void mainloop() {
         __attribute__((unused)) vaddr_t seq_pc = isa_exec_once();
         guest_nr_instr ++;
 
+        spill_flush_local();
+
         if (nemu_state.state != NEMU_RUNNING) tran_next_pc = NEXT_PC_END;
 
 #ifdef DEBUG