ARM: Fixed several dyncom bugs.

- Fixed NZCVT flags to properly save state when function returns.
- Fixed counter to keep track of the actual number of instructions executed.
- Fixed single-step mode to only execute one instruction at a time.
- DefaultIni: Removed comment that no longer applied to dyncom.

4 files changed, 26 insertions, 18 deletions

diff --git a/src/citra/default_ini.h b/src/citra/default_ini.h
index 7352c70c2..557da881b 100644
--- a/src/citra/default_ini.h
+++ b/src/citra/default_ini.h
@@ -28,7 +28,7 @@ pad_sright =
 
 [Core]
 cpu_core = ## 0: Interpreter (default), 1: FastInterpreter (experimental)
-gpu_refresh_rate = ## 60 (default), 1024 or 2048 may work better on the FastInterpreter
+gpu_refresh_rate = ## 60 (default)
 
 [Data Storage]
 use_virtual_sd =
diff --git a/src/core/arm/dyncom/arm_dyncom.cpp b/src/core/arm/dyncom/arm_dyncom.cpp
index 669b612fc..0b5dcccb9 100644
--- a/src/core/arm/dyncom/arm_dyncom.cpp
+++ b/src/core/arm/dyncom/arm_dyncom.cpp
@@ -110,9 +110,12 @@ u64 ARM_DynCom::GetTicks() const {
  * @param num_instructions Number of instructions to executes
  */
 void ARM_DynCom::ExecuteInstructions(int num_instructions) {
-    ticks += num_instructions;
     state->NumInstrsToExecute = num_instructions;
-    InterpreterMainLoop(state.get());
+
+    // Dyncom only breaks on instruction dispatch. This only happens on every instruction when
+    // executing one instruction at a time. Otherwise, if a block is being executed, more 
+    // instructions may actually be executed than specified.
+    ticks += InterpreterMainLoop(state.get());
 }
 
 /**
diff --git a/src/core/arm/dyncom/arm_dyncom_interpreter.cpp b/src/core/arm/dyncom/arm_dyncom_interpreter.cpp
index fe1501b59..f56a84518 100644
--- a/src/core/arm/dyncom/arm_dyncom_interpreter.cpp
+++ b/src/core/arm/dyncom/arm_dyncom_interpreter.cpp
@@ -3718,7 +3718,7 @@ static bool InAPrivilegedMode(arm_core_t *core)
 }
 
 /* r15 = r15 + 8 */
-void InterpreterMainLoop(ARMul_State* state)
+unsigned InterpreterMainLoop(ARMul_State* state)
 {
         #define CRn                             inst_cream->crn
         #define OPCODE_2                        inst_cream->opcode_2
@@ -3754,9 +3754,15 @@ void InterpreterMainLoop(ARMul_State* state)
 // GCC and Clang have a C++ extension to support a lookup table of labels. Otherwise, fallback to a
 // clunky switch statement.
 #if defined __GNUC__ || defined __clang__
-#define GOTO_NEXT_INST                  goto *InstLabel[inst_base->idx]
+#define GOTO_NEXT_INST \
+    if (num_instrs >= cpu->NumInstrsToExecute) goto END; \
+    num_instrs++; \
+    goto *InstLabel[inst_base->idx]
 #else
-#define GOTO_NEXT_INST switch(inst_base->idx) { \
+#define GOTO_NEXT_INST \
+    if (num_instrs >= cpu->NumInstrsToExecute) goto END; \
+    num_instrs++; \
+    switch(inst_base->idx) { \
     case 0: goto VMLA_INST; \
     case 1: goto VMLS_INST; \
     case 2: goto VNMLA_INST; \
@@ -4028,20 +4034,15 @@ void InterpreterMainLoop(ARMul_State* state)
         unsigned int addr;
         unsigned int phys_addr;
         unsigned int last_pc = 0;
+        unsigned int num_instrs = 0;
         fault_t fault;
         static unsigned int last_physical_base = 0, last_logical_base = 0;
         int ptr;
+        bool single_step = (cpu->NumInstrsToExecute == 1);
 
         LOAD_NZCVT;
         DISPATCH:
         {
-        if (cpu->NumInstrsToExecute == 0)
-            return;
-
-        cpu->NumInstrsToExecute--;
-
-        //NOTICE_LOG(ARM11, "instr!");
-
                 if (!cpu->NirqSig) {
                         if (!(cpu->Cpsr & 0x80)) {
                                 goto END;
@@ -4393,7 +4394,8 @@ void InterpreterMainLoop(ARMul_State* state)
                         #define CP_ACCESS_ALLOW 0
                         if(CP_ACCESS_ALLOW){
                                 /* undefined instruction here */
-                                return;
+                                cpu->NumInstrsToExecute = 0;
+                                return num_instrs;
                         }
                         ERROR_LOG(ARM11, "CDP insn inst=0x%x, pc=0x%x\n", inst_cream->inst, cpu->Reg[15]);
                         unsigned cpab = (cpu->CDP[inst_cream->cp_num]) (cpu, ARMul_FIRST, inst_cream->inst);
@@ -6532,12 +6534,14 @@ void InterpreterMainLoop(ARMul_State* state)
                 cpu->AbortAddr = addr;
                 cpu->CP15[CP15(CP15_FAULT_STATUS)] = fault & 0xff;
                 cpu->CP15[CP15(CP15_FAULT_ADDRESS)] = addr;
-                return;
+                cpu->NumInstrsToExecute = 0;
+                return num_instrs;
         }
         END:
         {
                 SAVE_NZCVT;
-                return;
+                cpu->NumInstrsToExecute = 0;
+                return num_instrs;
         }
         INIT_INST_LENGTH:
         {
@@ -6557,7 +6561,8 @@ void InterpreterMainLoop(ARMul_State* state)
         DEBUG_LOG(ARM11, "%llx\n", InstLabel[1]);
         DEBUG_LOG(ARM11, "%lld\n", (char *)InstEndLabel[1] - (char *)InstLabel[1]);
 #endif
-        return;
+        cpu->NumInstrsToExecute = 0;
+        return num_instrs;
         }
 }
 
diff --git a/src/core/arm/dyncom/arm_dyncom_interpreter.h b/src/core/arm/dyncom/arm_dyncom_interpreter.h
index d73f8f65f..c65eb23f7 100644
--- a/src/core/arm/dyncom/arm_dyncom_interpreter.h
+++ b/src/core/arm/dyncom/arm_dyncom_interpreter.h
@@ -4,4 +4,4 @@
 
 #pragma once
 
-void InterpreterMainLoop(ARMul_State* state);
+unsigned InterpreterMainLoop(ARMul_State* state);