diff --git a/src/core/cpu_code_cache.cpp b/src/core/cpu_code_cache.cpp
index e42edb53a..bfb0f4b1e 100644
--- a/src/core/cpu_code_cache.cpp
+++ b/src/core/cpu_code_cache.cpp
@@ -4,6 +4,9 @@
 #include "cpu_disasm.h"
 #include "cpu_recompiler_code_generator.h"
 #include "cpu_recompiler_thunks.h"
+#include "system.h"
+#include <thread>
+#include <chrono>
 Log_SetChannel(CPU::CodeCache);
 
 namespace CPU {
@@ -15,8 +18,9 @@ CodeCache::CodeCache() = default;
 
 CodeCache::~CodeCache() = default;
 
-void CodeCache::Initialize(Core* core, Bus* bus)
+void CodeCache::Initialize(System* system, Core* core, Bus* bus)
 {
+  m_system = system;
   m_core = core;
   m_bus = bus;
 
@@ -122,9 +126,12 @@ bool CodeCache::CompileBlock(CodeBlock* block)
     }
 
     cbi.pc = pc;
-    cbi.is_branch = IsBranchInstruction(cbi.instruction);
     cbi.is_branch_delay_slot = is_branch_delay_slot;
     cbi.is_load_delay_slot = is_load_delay_slot;
+    cbi.is_branch_instruction = IsBranchInstruction(cbi.instruction);
+    cbi.is_load_instruction = IsMemoryLoadInstruction(cbi.instruction);
+    cbi.is_store_instruction = IsMemoryStoreInstruction(cbi.instruction);
+    cbi.has_load_delay = InstructionHasLoadDelay(cbi.instruction);
     cbi.can_trap = CanInstructionTrap(cbi.instruction, m_core->InUserMode());
 
     // instruction is decoded now
@@ -133,11 +140,14 @@ bool CodeCache::CompileBlock(CodeBlock* block)
 
     // if we're in a branch delay slot, the block is now done
     // except if this is a branch in a branch delay slot, then we grab the one after that, and so on...
-    if (is_branch_delay_slot && !cbi.is_branch)
+    if (is_branch_delay_slot && !cbi.is_branch_instruction)
       break;
 
     // if this is a branch, we grab the next instruction (delay slot), and then exit
-    is_branch_delay_slot = cbi.is_branch;
+    is_branch_delay_slot = cbi.is_branch_instruction;
+
+    // same for load delay
+    is_load_delay_slot = cbi.has_load_delay;
 
     // is this a non-branchy exit? (e.g. syscall)
     if (IsExitBlockInstruction(cbi.instruction))
@@ -254,10 +264,7 @@ void CodeCache::InterpretCachedBlock(const CodeBlock& block)
     m_core->ExecuteInstruction();
 
     // next load delay
-    m_core->m_load_delay_reg = m_core->m_next_load_delay_reg;
-    m_core->m_next_load_delay_reg = Reg::count;
-    m_core->m_load_delay_old_value = m_core->m_next_load_delay_old_value;
-    m_core->m_next_load_delay_old_value = 0;
+    m_core->UpdateLoadDelay();
 
     if (m_core->m_exception_raised)
       break;
@@ -294,10 +301,7 @@ void CodeCache::InterpretUncachedBlock()
     m_core->ExecuteInstruction();
 
     // next load delay
-    m_core->m_load_delay_reg = m_core->m_next_load_delay_reg;
-    m_core->m_next_load_delay_reg = Reg::count;
-    m_core->m_load_delay_old_value = m_core->m_next_load_delay_old_value;
-    m_core->m_next_load_delay_old_value = 0;
+    m_core->UpdateLoadDelay();
 
     const bool branch = IsBranchInstruction(m_core->m_current_instruction);
     if (m_core->m_exception_raised || (!branch && in_branch_delay_slot) ||
diff --git a/src/core/cpu_code_cache.h b/src/core/cpu_code_cache.h
index f68ea2bb1..5b9882032 100644
--- a/src/core/cpu_code_cache.h
+++ b/src/core/cpu_code_cache.h
@@ -9,6 +9,7 @@
 class JitCodeBuffer;
 
 class Bus;
+class System;
 
 namespace CPU {
 class Core;
@@ -23,7 +24,7 @@ public:
   CodeCache();
   ~CodeCache();
 
-  void Initialize(Core* core, Bus* bus);
+  void Initialize(System* system, Core* core, Bus* bus);
   void Reset();
   void Execute();
 
@@ -39,6 +40,7 @@ private:
   void InterpretCachedBlock(const CodeBlock& block);
   void InterpretUncachedBlock();
 
+  System* m_system;
   Core* m_core;
   Bus* m_bus;
 
diff --git a/src/core/cpu_core.cpp b/src/core/cpu_core.cpp
index 79625bbe1..5416edd48 100644
--- a/src/core/cpu_core.cpp
+++ b/src/core/cpu_core.cpp
@@ -97,8 +97,10 @@ bool Core::DoState(StateWrapper& sw)
   sw.Do(&m_next_instruction_is_branch_delay_slot);
   sw.Do(&m_branch_was_taken);
   sw.Do(&m_load_delay_reg);
+  sw.Do(&m_load_delay_value);
   sw.Do(&m_load_delay_old_value);
   sw.Do(&m_next_load_delay_reg);
+  sw.Do(&m_next_load_delay_value);
   sw.Do(&m_next_load_delay_old_value);
   sw.Do(&m_cache_control);
   sw.DoBytes(m_dcache.data(), m_dcache.size());
@@ -338,7 +340,7 @@ bool Core::HasPendingInterrupt()
   // const bool do_interrupt = m_cop0_regs.sr.IEc && ((m_cop0_regs.cause.Ip & m_cop0_regs.sr.Im) != 0);
   const bool do_interrupt =
     m_cop0_regs.sr.IEc && (((m_cop0_regs.cause.bits & m_cop0_regs.sr.bits) & (UINT32_C(0xFF) << 8)) != 0);
-  
+
   return do_interrupt;
 }
 
@@ -354,19 +356,16 @@ void Core::DispatchInterrupt()
                  m_next_instruction.cop.cop_n);
 }
 
-void Core::FlushLoadDelay()
-{
-  m_load_delay_reg = Reg::count;
-  m_load_delay_old_value = 0;
-  m_next_load_delay_reg = Reg::count;
-  m_next_load_delay_old_value = 0;
-}
-
 void Core::FlushPipeline()
 {
   // loads are flushed
-  FlushLoadDelay();
-
+  m_next_load_delay_reg = Reg::count;
+  if (m_load_delay_reg != Reg::count)
+  {
+    m_regs.r[static_cast<u8>(m_load_delay_reg)] = m_load_delay_value;
+    m_load_delay_reg = Reg::count;
+  }
+  
   // not in a branch delay slot
   m_branch_was_taken = false;
   m_next_instruction_is_branch_delay_slot = false;
@@ -383,13 +382,15 @@ void Core::FlushPipeline()
 
 u32 Core::ReadReg(Reg rs)
 {
-  return rs == m_load_delay_reg ? m_load_delay_old_value : m_regs.r[static_cast<u8>(rs)];
+  return m_regs.r[static_cast<u8>(rs)];
 }
 
 void Core::WriteReg(Reg rd, u32 value)
 {
-  if (rd != Reg::zero)
-    m_regs.r[static_cast<u8>(rd)] = value;
+  m_regs.r[static_cast<u8>(rd)] = value;
+
+  // prevent writes to $zero from going through - better than branching/cmov
+  m_regs.zero = 0;
 }
 
 void Core::WriteRegDelayed(Reg rd, u32 value)
@@ -398,10 +399,14 @@ void Core::WriteRegDelayed(Reg rd, u32 value)
   if (rd == Reg::zero)
     return;
 
-  // save the old value, this will be returned if the register is read in the next instruction
+  // double load delays ignore the first value
+  if (m_load_delay_reg == rd)
+    m_load_delay_reg = Reg::count;
+
+  // save the old value, if something else overwrites this reg we want to preserve it
   m_next_load_delay_reg = rd;
-  m_next_load_delay_old_value = ReadReg(rd);
-  m_regs.r[static_cast<u8>(rd)] = value;
+  m_next_load_delay_value = value;
+  m_next_load_delay_old_value = m_regs.r[static_cast<u8>(rd)];
 }
 
 std::optional<u32> Core::ReadCop0Reg(Cop0Reg reg)
@@ -608,10 +613,7 @@ void Core::Execute()
     ExecuteInstruction();
 
     // next load delay
-    m_load_delay_reg = m_next_load_delay_reg;
-    m_next_load_delay_reg = Reg::count;
-    m_load_delay_old_value = m_next_load_delay_old_value;
-    m_next_load_delay_old_value = 0;
+    UpdateLoadDelay();
   }
 }
 
@@ -643,12 +645,12 @@ void Core::ExecuteInstruction()
   }
 #endif
 
-//#ifdef _DEBUG
+  //#ifdef _DEBUG
   if (TRACE_EXECUTION)
     PrintInstruction(inst.bits, m_current_instruction_pc, this);
   if (LOG_EXECUTION)
     LogInstruction(inst.bits, m_current_instruction_pc, this);
-//#endif
+  //#endif
 
   switch (inst.op)
   {
@@ -1039,8 +1041,8 @@ void Core::ExecuteInstruction()
       if (!ReadMemoryWord(aligned_addr, &aligned_value))
         return;
 
-      // note: bypasses load delay on the read
-      const u32 existing_value = m_regs.r[static_cast<u8>(inst.i.rt.GetValue())];
+      // Bypasses load delay. No need to check the old value since this is the delay slot or it's not relevant.
+      const u32 existing_value = (inst.i.rt == m_load_delay_reg) ? m_load_delay_value : ReadReg(inst.i.rt);
       const u8 shift = (Truncate8(addr) & u8(3)) * u8(8);
       u32 new_value;
       if (inst.op == InstructionOp::lwl)
diff --git a/src/core/cpu_core.h b/src/core/cpu_core.h
index c0469025b..cd84b7ce6 100644
--- a/src/core/cpu_core.h
+++ b/src/core/cpu_core.h
@@ -14,11 +14,10 @@ namespace CPU {
 
 class CodeCache;
 
-namespace Recompiler
-{
+namespace Recompiler {
 class CodeGenerator;
 class Thunks;
-}
+} // namespace Recompiler
 
 class Core
 {
@@ -28,7 +27,7 @@ public:
   static constexpr PhysicalMemoryAddress DCACHE_LOCATION_MASK = UINT32_C(0xFFFFFC00);
   static constexpr PhysicalMemoryAddress DCACHE_OFFSET_MASK = UINT32_C(0x000003FF);
   static constexpr PhysicalMemoryAddress DCACHE_SIZE = UINT32_C(0x00000400);
-  
+
   friend CodeCache;
   friend Recompiler::CodeGenerator;
   friend Recompiler::Thunks;
@@ -103,6 +102,19 @@ private:
   void DisassembleAndLog(u32 addr);
   void DisassembleAndPrint(u32 addr, u32 instructions_before, u32 instructions_after);
 
+  // Updates load delays - call after each instruction
+  ALWAYS_INLINE void UpdateLoadDelay()
+  {
+    // the old value is needed in case the delay slot instruction overwrites the same register
+    if (m_load_delay_reg != Reg::count && m_regs.r[static_cast<u8>(m_load_delay_reg)] == m_load_delay_old_value)
+      m_regs.r[static_cast<u8>(m_load_delay_reg)] = m_load_delay_value;
+
+    m_load_delay_reg = m_next_load_delay_reg;
+    m_load_delay_value = m_next_load_delay_value;
+    m_load_delay_old_value = m_next_load_delay_old_value;
+    m_next_load_delay_reg = Reg::count;
+  }
+
   // Fetches the instruction at m_regs.npc
   bool FetchInstruction();
   void ExecuteInstruction();
@@ -117,9 +129,6 @@ private:
   bool HasPendingInterrupt();
   void DispatchInterrupt();
 
-  // flushes any load delays if present
-  void FlushLoadDelay();
-
   // clears pipeline of load/branch delays
   void FlushPipeline();
 
@@ -158,8 +167,10 @@ private:
 
   // load delays
   Reg m_load_delay_reg = Reg::count;
+  u32 m_load_delay_value = 0;
   u32 m_load_delay_old_value = 0;
   Reg m_next_load_delay_reg = Reg::count;
+  u32 m_next_load_delay_value = 0;
   u32 m_next_load_delay_old_value = 0;
 
   u32 m_cache_control = 0;
diff --git a/src/core/cpu_recompiler_code_generator.cpp b/src/core/cpu_recompiler_code_generator.cpp
index 13da8d3b5..e9f821169 100644
--- a/src/core/cpu_recompiler_code_generator.cpp
+++ b/src/core/cpu_recompiler_code_generator.cpp
@@ -28,11 +28,6 @@ bool CodeGenerator::CompileBlock(const CodeBlock* block, CodeBlock::HostCodePoin
   m_block_start = block->instructions.data();
   m_block_end = block->instructions.data() + block->instructions.size();
 
-  m_current_instruction_in_branch_delay_slot_dirty = true;
-  m_branch_was_taken_dirty = true;
-  m_current_instruction_was_branch_taken_dirty = false;
-  m_load_delay_dirty = true;
-
   EmitBeginBlock();
   BlockPrologue();
 
@@ -340,7 +335,11 @@ void CodeGenerator::BlockPrologue()
 {
   EmitStoreCPUStructField(offsetof(Core, m_exception_raised), Value::FromConstantU8(0));
 
-  // fetching of the first instruction...
+  // we don't know the state of the last block, so assume load delays might be in progress
+  m_current_instruction_in_branch_delay_slot_dirty = true;
+  m_branch_was_taken_dirty = true;
+  m_current_instruction_was_branch_taken_dirty = false;
+  m_load_delay_dirty = true;
 
   // sync m_current_instruction_pc so we can simply add to it
   SyncCurrentInstructionPC();
@@ -435,7 +434,7 @@ void CodeGenerator::InstructionPrologue(const CodeBlockInstruction& cbi, TickCou
     m_delayed_pc_add = 0;
   }
 
-  if (!cbi.is_branch)
+  if (!cbi.is_branch_instruction)
     m_delayed_pc_add = INSTRUCTION_SIZE;
 
   m_delayed_cycles_add += cycles;
@@ -445,38 +444,18 @@ void CodeGenerator::InstructionPrologue(const CodeBlockInstruction& cbi, TickCou
 void CodeGenerator::InstructionEpilogue(const CodeBlockInstruction& cbi)
 {
   // copy if the previous instruction was a load, reset the current value on the next instruction
-  if (m_load_delay_dirty)
+  if (m_next_load_delay_dirty)
   {
-    // cpu->m_load_delay_reg = cpu->m_next_load_delay_reg;
-    // cpu->m_next_load_delay_reg = Reg::count;
-    {
-      Value temp = m_register_cache.AllocateScratch(RegSize_8);
-      EmitLoadCPUStructField(temp.host_reg, RegSize_8, offsetof(Core, m_next_load_delay_reg));
-      EmitStoreCPUStructField(offsetof(Core, m_next_load_delay_reg),
-                              Value::FromConstantU8(static_cast<u8>(Reg::count)));
-      EmitStoreCPUStructField(offsetof(Core, m_load_delay_reg), temp);
-    }
-
-    // cpu->m_load_delay_old_value = cpu->m_next_load_delay_old_value;
-    // cpu->m_next_load_delay_old_value = 0;
-    {
-      Value temp = m_register_cache.AllocateScratch(RegSize_32);
-      EmitLoadCPUStructField(temp.host_reg, RegSize_32, offsetof(Core, m_next_load_delay_old_value));
-      EmitStoreCPUStructField(offsetof(Core, m_next_load_delay_old_value), Value::FromConstantU32(0));
-      EmitStoreCPUStructField(offsetof(Core, m_load_delay_old_value), temp);
-    }
-
-    m_load_delay_dirty = false;
-    m_next_load_delay_dirty = true;
-  }
-  else if (m_next_load_delay_dirty)
-  {
-    // cpu->m_load_delay_reg = Reg::count;
-    // cpu->m_load_delay_old_value = 0;
-    EmitStoreCPUStructField(offsetof(Core, m_load_delay_reg), Value::FromConstantU8(static_cast<u8>(Reg::count)));
-    EmitStoreCPUStructField(offsetof(Core, m_load_delay_old_value), Value::FromConstantU32(0));
-
+    Log_DebugPrint("Emitting delay slot flush (with move next)");
+    EmitDelaySlotUpdate(false, false, true);
     m_next_load_delay_dirty = false;
+    m_load_delay_dirty = true;
+  }
+  else if (m_load_delay_dirty)
+  {
+    Log_DebugPrint("Emitting delay slot flush");
+    EmitDelaySlotUpdate(true, false, false);
+    m_load_delay_dirty = false;
   }
 }
 
@@ -528,9 +507,9 @@ bool CodeGenerator::Compile_Fallback(const CodeBlockInstruction& cbi)
     EmitFunctionCall(nullptr, &Thunks::InterpretInstruction, m_register_cache.GetCPUPtr());
   }
 
-  m_current_instruction_in_branch_delay_slot_dirty = cbi.is_branch;
-  m_branch_was_taken_dirty = cbi.is_branch;
-  m_load_delay_dirty = true;
+  m_current_instruction_in_branch_delay_slot_dirty = cbi.is_branch_instruction;
+  m_branch_was_taken_dirty = cbi.is_branch_instruction;
+  m_next_load_delay_dirty = cbi.has_load_delay;
   InstructionEpilogue(cbi);
   return true;
 }
diff --git a/src/core/cpu_recompiler_code_generator.h b/src/core/cpu_recompiler_code_generator.h
index effff6b0a..2321db856 100644
--- a/src/core/cpu_recompiler_code_generator.h
+++ b/src/core/cpu_recompiler_code_generator.h
@@ -147,6 +147,7 @@ private:
   void SyncCurrentInstructionPC();
   void SyncPC();
   void AddPendingCycles();
+  void EmitDelaySlotUpdate(bool skip_check_for_delay, bool skip_check_old_value, bool move_next);
 
   //////////////////////////////////////////////////////////////////////////
   // Instruction Code Generators
@@ -172,14 +173,12 @@ private:
 
   std::array<Value, 3> m_operand_memory_addresses{};
 
-  Xbyak::Label m_block_exit_label;
-
   // whether various flags need to be reset.
   bool m_current_instruction_in_branch_delay_slot_dirty = false;
   bool m_branch_was_taken_dirty = false;
   bool m_current_instruction_was_branch_taken_dirty = false;
-  bool m_next_load_delay_dirty = false;
   bool m_load_delay_dirty = false;
+  bool m_next_load_delay_dirty = false;
 };
 
 } // namespace CPU_X86::Recompiler
diff --git a/src/core/cpu_recompiler_code_generator_x64.cpp b/src/core/cpu_recompiler_code_generator_x64.cpp
index 4f47d8f68..749c86c9d 100644
--- a/src/core/cpu_recompiler_code_generator_x64.cpp
+++ b/src/core/cpu_recompiler_code_generator_x64.cpp
@@ -151,9 +151,8 @@ void CodeGenerator::EmitBeginBlock()
 
 void CodeGenerator::EmitEndBlock()
 {
-  m_emit.L(m_block_exit_label);
   m_register_cache.FreeHostReg(RCPUPTR);
-  m_register_cache.PopCalleeSavedRegisters();
+  m_register_cache.PopCalleeSavedRegisters(true);
 
   m_emit.ret();
 }
@@ -166,9 +165,10 @@ void CodeGenerator::EmitBlockExitOnBool(const Value& value)
   m_emit.test(GetHostReg8(value), GetHostReg8(value));
   m_emit.jz(continue_label);
 
-  // flush current state
+  // flush current state and return
   m_register_cache.FlushAllGuestRegisters(false, false);
-  m_emit.jmp(m_block_exit_label, Xbyak::CodeGenerator::T_NEAR);
+  m_register_cache.PopCalleeSavedRegisters(false);
+  m_emit.ret();
 
   m_emit.L(continue_label);
 }
@@ -1303,6 +1303,62 @@ void CodeGenerator::EmitAddCPUStructField(u32 offset, const Value& value)
   }
 }
 
+void CodeGenerator::EmitDelaySlotUpdate(bool skip_check_for_delay, bool skip_check_old_value, bool move_next)
+{
+  Value reg = m_register_cache.AllocateScratch(RegSize_8);
+  Value value = m_register_cache.AllocateScratch(RegSize_32);
+
+  Xbyak::Label skip_flush;
+
+  auto load_delay_reg = m_emit.byte[GetCPUPtrReg() + offsetof(Core, m_load_delay_reg)];
+  auto load_delay_old_value = m_emit.dword[GetCPUPtrReg() + offsetof(Core, m_load_delay_old_value)];
+  auto load_delay_value = m_emit.dword[GetCPUPtrReg() + offsetof(Core, m_load_delay_value)];
+  auto reg_ptr = m_emit.dword[GetCPUPtrReg() + offsetof(Core, m_regs.r[0]) + GetHostReg64(reg.host_reg) * 4];
+
+  // reg = load_delay_reg
+  m_emit.movzx(GetHostReg32(reg.host_reg), load_delay_reg);
+  if (!skip_check_old_value)
+    m_emit.mov(GetHostReg32(value), load_delay_old_value);
+
+  if (!skip_check_for_delay)
+  {
+    // if load_delay_reg == Reg::count goto skip_flush
+    m_emit.cmp(GetHostReg32(reg.host_reg), static_cast<u8>(Reg::count));
+    m_emit.je(skip_flush);
+  }
+
+  if (!skip_check_old_value)
+  {
+    // if r[reg] != load_delay_old_value goto skip_flush
+    m_emit.cmp(GetHostReg32(value), reg_ptr);
+    m_emit.jne(skip_flush);
+  }
+
+  // r[reg] = load_delay_value
+  m_emit.mov(GetHostReg32(value), load_delay_value);
+  m_emit.mov(reg_ptr, GetHostReg32(value));
+
+  // if !move_next load_delay_reg = Reg::count
+  if (!move_next)
+    m_emit.mov(load_delay_reg, static_cast<u8>(Reg::count));
+
+  m_emit.L(skip_flush);
+
+  if (move_next)
+  {
+    auto next_load_delay_reg = m_emit.byte[GetCPUPtrReg() + offsetof(Core, m_next_load_delay_reg)];
+    auto next_load_delay_old_value = m_emit.dword[GetCPUPtrReg() + offsetof(Core, m_next_load_delay_old_value)];
+    auto next_load_delay_value = m_emit.dword[GetCPUPtrReg() + offsetof(Core, m_next_load_delay_value)];
+    m_emit.mov(GetHostReg32(value), next_load_delay_value);
+    m_emit.mov(GetHostReg8(reg), next_load_delay_reg);
+    m_emit.mov(load_delay_value, GetHostReg32(value));
+    m_emit.mov(GetHostReg32(value), next_load_delay_old_value);
+    m_emit.mov(load_delay_reg, GetHostReg8(reg));
+    m_emit.mov(load_delay_old_value, GetHostReg32(value));
+    m_emit.mov(next_load_delay_reg, static_cast<u8>(Reg::count));
+  }
+}
+
 #if 0
 class ThunkGenerator
 {
diff --git a/src/core/cpu_recompiler_register_cache.cpp b/src/core/cpu_recompiler_register_cache.cpp
index 3d8ae9f0f..e540c4b77 100644
--- a/src/core/cpu_recompiler_register_cache.cpp
+++ b/src/core/cpu_recompiler_register_cache.cpp
@@ -307,7 +307,7 @@ u32 RegisterCache::PopCallerSavedRegisters() const
   return count;
 }
 
-u32 RegisterCache::PopCalleeSavedRegisters()
+u32 RegisterCache::PopCalleeSavedRegisters(bool commit)
 {
   if (m_host_register_callee_saved_order_count == 0)
     return 0;
@@ -321,7 +321,8 @@ u32 RegisterCache::PopCalleeSavedRegisters()
                 (HostRegState::CalleeSaved | HostRegState::CalleeSavedAllocated));
 
     m_code_generator.EmitPopHostReg(reg);
-    m_host_register_state[reg] &= ~HostRegState::CalleeSavedAllocated;
+    if (commit)
+      m_host_register_state[reg] &= ~HostRegState::CalleeSavedAllocated;
     count++;
     i--;
   } while (i > 0);
diff --git a/src/core/cpu_recompiler_register_cache.h b/src/core/cpu_recompiler_register_cache.h
index b01637d25..a090c733f 100644
--- a/src/core/cpu_recompiler_register_cache.h
+++ b/src/core/cpu_recompiler_register_cache.h
@@ -176,7 +176,7 @@ public:
   u32 PopCallerSavedRegisters() const;
 
   /// Restore callee-saved registers. Call at the end of the function.
-  u32 PopCalleeSavedRegisters();
+  u32 PopCalleeSavedRegisters(bool commit);
 
   //////////////////////////////////////////////////////////////////////////
   // Scratch Register Allocation
diff --git a/src/core/cpu_recompiler_thunks.cpp b/src/core/cpu_recompiler_thunks.cpp
index cfdfa1e61..9d968da97 100644
--- a/src/core/cpu_recompiler_thunks.cpp
+++ b/src/core/cpu_recompiler_thunks.cpp
@@ -40,4 +40,9 @@ bool Thunks::InterpretInstruction(Core* cpu)
   return cpu->m_exception_raised;
 }
 
+void Thunks::UpdateLoadDelay(Core* cpu)
+{
+  cpu->UpdateLoadDelay();
+}
+
 } // namespace CPU::Recompiler
\ No newline at end of file
diff --git a/src/core/cpu_recompiler_thunks.h b/src/core/cpu_recompiler_thunks.h
index 4104d2cca..3cfbfdb14 100644
--- a/src/core/cpu_recompiler_thunks.h
+++ b/src/core/cpu_recompiler_thunks.h
@@ -20,6 +20,7 @@ public:
   static bool WriteMemoryHalfWord(Core* cpu, u32 address, u16 value);
   static bool WriteMemoryWord(Core* cpu, u32 address, u32 value);
   static bool InterpretInstruction(Core* cpu);
+  static void UpdateLoadDelay(Core* cpu);
 };
 
 class ASMFunctions
diff --git a/src/core/cpu_types.cpp b/src/core/cpu_types.cpp
index 927f83572..57d651c77 100644
--- a/src/core/cpu_types.cpp
+++ b/src/core/cpu_types.cpp
@@ -44,6 +44,76 @@ bool IsBranchInstruction(const Instruction& instruction)
   }
 }
 
+bool IsMemoryLoadInstruction(const Instruction& instruction)
+{
+  switch (instruction.op)
+  {
+    case InstructionOp::lb:
+    case InstructionOp::lh:
+    case InstructionOp::lw:
+    case InstructionOp::lbu:
+    case InstructionOp::lhu:
+    case InstructionOp::lwl:
+    case InstructionOp::lwr:
+      return true;
+
+    case InstructionOp::lwc2:
+      return true;
+
+    default:
+      return false;
+  }
+}
+
+bool IsMemoryStoreInstruction(const Instruction& instruction)
+{
+  switch (instruction.op)
+  {
+    case InstructionOp::sb:
+    case InstructionOp::sh:
+    case InstructionOp::sw:
+    case InstructionOp::swl:
+    case InstructionOp::swr:
+      return true;
+
+    case InstructionOp::swc2:
+      return true;
+
+    default:
+      return false;
+  }
+}
+
+bool InstructionHasLoadDelay(const Instruction& instruction)
+{
+  switch (instruction.op)
+  {
+    case InstructionOp::lb:
+    case InstructionOp::lh:
+    case InstructionOp::lw:
+    case InstructionOp::lbu:
+    case InstructionOp::lhu:
+    case InstructionOp::lwl:
+    case InstructionOp::lwr:
+      return true;
+
+    case InstructionOp::cop0:
+    case InstructionOp::cop2:
+    {
+      if (instruction.cop.IsCommonInstruction())
+      {
+        const CopCommonInstruction common_op = instruction.cop.CommonOp();
+        return (common_op == CopCommonInstruction::cfcn || common_op == CopCommonInstruction::mfcn);
+      }
+
+      return false;
+    }
+
+    default:
+      return false;
+  }
+}
+
 bool IsExitBlockInstruction(const Instruction& instruction)
 {
   switch (instruction.op)
@@ -167,26 +237,6 @@ bool CanInstructionTrap(const Instruction& instruction, bool in_user_mode)
   }
 }
 
-bool IsLoadDelayingInstruction(const Instruction& instruction)
-{
-  switch (instruction.op)
-  {
-    case InstructionOp::lb:
-    case InstructionOp::lh:
-    case InstructionOp::lw:
-    case InstructionOp::lbu:
-    case InstructionOp::lhu:
-      return true;
-
-    case InstructionOp::lwl:
-    case InstructionOp::lwr:
-      return false;
-
-    default:
-      return false;
-  }
-}
-
 bool IsInvalidInstruction(const Instruction& instruction)
 {
   // TODO
diff --git a/src/core/cpu_types.h b/src/core/cpu_types.h
index e7a74ef97..61395dd2b 100644
--- a/src/core/cpu_types.h
+++ b/src/core/cpu_types.h
@@ -208,9 +208,11 @@ union Instruction
 
 // Instruction helpers.
 bool IsBranchInstruction(const Instruction& instruction);
+bool IsMemoryLoadInstruction(const Instruction& instruction);
+bool IsMemoryStoreInstruction(const Instruction& instruction);
+bool InstructionHasLoadDelay(const Instruction& instruction);
 bool IsExitBlockInstruction(const Instruction& instruction);
 bool CanInstructionTrap(const Instruction& instruction, bool in_user_mode);
-bool IsLoadDelayingInstruction(const Instruction& instruction);
 bool IsInvalidInstruction(const Instruction& instruction);
 
 struct Registers
@@ -396,10 +398,13 @@ struct CodeBlockInstruction
   Instruction instruction;
   u32 pc;
 
-  bool is_branch : 1;
+  bool is_branch_instruction : 1;
   bool is_branch_delay_slot : 1;
+  bool is_load_instruction : 1;
+  bool is_store_instruction : 1;
   bool is_load_delay_slot : 1;
   bool is_last_instruction : 1;
+  bool has_load_delay : 1;
   bool can_trap : 1;
 };
 
diff --git a/src/core/system.cpp b/src/core/system.cpp
index c0877df77..64f89ae07 100644
--- a/src/core/system.cpp
+++ b/src/core/system.cpp
@@ -171,7 +171,7 @@ bool System::Boot(const char* filename)
 void System::InitializeComponents()
 {
   m_cpu->Initialize(m_bus.get());
-  m_cpu_code_cache->Initialize(m_cpu.get(), m_bus.get());
+  m_cpu_code_cache->Initialize(this, m_cpu.get(), m_bus.get());
   m_bus->Initialize(m_cpu.get(), m_cpu_code_cache.get(), m_dma.get(), m_interrupt_controller.get(), m_gpu.get(),
                     m_cdrom.get(), m_pad.get(), m_timers.get(), m_spu.get(), m_mdec.get());