8 files changed, 367 insertions, 71 deletions

diff --git a/src/core/arm/arm_interface.h b/src/core/arm/arm_interface.h
index 47b964eb7..57eae839e 100644
--- a/src/core/arm/arm_interface.h
+++ b/src/core/arm/arm_interface.h
@@ -25,7 +25,20 @@ public:
     explicit ARM_Interface(System& system_) : system{system_} {}
     virtual ~ARM_Interface() = default;
 
-    struct ThreadContext {
+    struct ThreadContext32 {
+        std::array<u32, 16> cpu_registers;
+        u32 cpsr;
+        std::array<u8, 4> padding;
+        std::array<u64, 32> fprs;
+        u32 fpscr;
+        u32 fpexc;
+        u32 tpidr;
+    };
+    // Internally within the kernel, it expects the AArch32 version of the
+    // thread context to be 344 bytes in size.
+    static_assert(sizeof(ThreadContext32) == 0x158);
+
+    struct ThreadContext64 {
         std::array<u64, 31> cpu_registers;
         u64 sp;
         u64 pc;
@@ -38,7 +51,7 @@ public:
     };
     // Internally within the kernel, it expects the AArch64 version of the
     // thread context to be 800 bytes in size.
-    static_assert(sizeof(ThreadContext) == 0x320);
+    static_assert(sizeof(ThreadContext64) == 0x320);
 
     /// Runs the CPU until an event happens
     virtual void Run() = 0;
@@ -130,17 +143,10 @@ public:
      */
     virtual void SetTPIDR_EL0(u64 value) = 0;
 
-    /**
-     * Saves the current CPU context
-     * @param ctx Thread context to save
-     */
-    virtual void SaveContext(ThreadContext& ctx) = 0;
-
-    /**
-     * Loads a CPU context
-     * @param ctx Thread context to load
-     */
-    virtual void LoadContext(const ThreadContext& ctx) = 0;
+    virtual void SaveContext(ThreadContext32& ctx) = 0;
+    virtual void SaveContext(ThreadContext64& ctx) = 0;
+    virtual void LoadContext(const ThreadContext32& ctx) = 0;
+    virtual void LoadContext(const ThreadContext64& ctx) = 0;
 
     /// Clears the exclusive monitor's state.
     virtual void ClearExclusiveState() = 0;
diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
new file mode 100644
index 000000000..187a972ac
--- /dev/null
+++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
@@ -0,0 +1,208 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cinttypes>
+#include <memory>
+#include <dynarmic/A32/a32.h>
+#include <dynarmic/A32/config.h>
+#include <dynarmic/A32/context.h>
+#include "common/microprofile.h"
+#include "core/arm/dynarmic/arm_dynarmic_32.h"
+#include "core/arm/dynarmic/arm_dynarmic_64.h"
+#include "core/arm/dynarmic/arm_dynarmic_cp15.h"
+#include "core/core.h"
+#include "core/core_manager.h"
+#include "core/core_timing.h"
+#include "core/hle/kernel/svc.h"
+#include "core/memory.h"
+
+namespace Core {
+
+class DynarmicCallbacks32 : public Dynarmic::A32::UserCallbacks {
+public:
+    explicit DynarmicCallbacks32(ARM_Dynarmic_32& parent) : parent(parent) {}
+
+    u8 MemoryRead8(u32 vaddr) override {
+        return parent.system.Memory().Read8(vaddr);
+    }
+    u16 MemoryRead16(u32 vaddr) override {
+        return parent.system.Memory().Read16(vaddr);
+    }
+    u32 MemoryRead32(u32 vaddr) override {
+        return parent.system.Memory().Read32(vaddr);
+    }
+    u64 MemoryRead64(u32 vaddr) override {
+        return parent.system.Memory().Read64(vaddr);
+    }
+
+    void MemoryWrite8(u32 vaddr, u8 value) override {
+        parent.system.Memory().Write8(vaddr, value);
+    }
+    void MemoryWrite16(u32 vaddr, u16 value) override {
+        parent.system.Memory().Write16(vaddr, value);
+    }
+    void MemoryWrite32(u32 vaddr, u32 value) override {
+        parent.system.Memory().Write32(vaddr, value);
+    }
+    void MemoryWrite64(u32 vaddr, u64 value) override {
+        parent.system.Memory().Write64(vaddr, value);
+    }
+
+    void InterpreterFallback(u32 pc, std::size_t num_instructions) override {
+        UNIMPLEMENTED();
+    }
+
+    void ExceptionRaised(u32 pc, Dynarmic::A32::Exception exception) override {
+        switch (exception) {
+        case Dynarmic::A32::Exception::UndefinedInstruction:
+        case Dynarmic::A32::Exception::UnpredictableInstruction:
+            break;
+        case Dynarmic::A32::Exception::Breakpoint:
+            break;
+        }
+        LOG_CRITICAL(HW_GPU, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})",
+                     static_cast<std::size_t>(exception), pc, MemoryReadCode(pc));
+        UNIMPLEMENTED();
+    }
+
+    void CallSVC(u32 swi) override {
+        Kernel::CallSVC(parent.system, swi);
+    }
+
+    void AddTicks(u64 ticks) override {
+        // Divide the number of ticks by the amount of CPU cores. TODO(Subv): This yields only a
+        // rough approximation of the amount of executed ticks in the system, it may be thrown off
+        // if not all cores are doing a similar amount of work. Instead of doing this, we should
+        // device a way so that timing is consistent across all cores without increasing the ticks 4
+        // times.
+        u64 amortized_ticks = (ticks - num_interpreted_instructions) / Core::NUM_CPU_CORES;
+        // Always execute at least one tick.
+        amortized_ticks = std::max<u64>(amortized_ticks, 1);
+
+        parent.system.CoreTiming().AddTicks(amortized_ticks);
+        num_interpreted_instructions = 0;
+    }
+    u64 GetTicksRemaining() override {
+        return std::max(parent.system.CoreTiming().GetDowncount(), {});
+    }
+
+    ARM_Dynarmic_32& parent;
+    std::size_t num_interpreted_instructions{};
+    u64 tpidrro_el0{};
+    u64 tpidr_el0{};
+};
+
+std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable& page_table,
+                                                             std::size_t address_space_bits) const {
+    Dynarmic::A32::UserConfig config;
+    config.callbacks = cb.get();
+    // TODO(bunnei): Implement page table for 32-bit
+    // config.page_table = &page_table.pointers;
+    config.coprocessors[15] = std::make_shared<DynarmicCP15>((u32*)&CP15_regs[0]);
+    config.define_unpredictable_behaviour = true;
+    return std::make_unique<Dynarmic::A32::Jit>(config);
+}
+
+MICROPROFILE_DEFINE(ARM_Jit_Dynarmic_32, "ARM JIT", "Dynarmic", MP_RGB(255, 64, 64));
+
+void ARM_Dynarmic_32::Run() {
+    MICROPROFILE_SCOPE(ARM_Jit_Dynarmic_32);
+    jit->Run();
+}
+
+void ARM_Dynarmic_32::Step() {
+    cb->InterpreterFallback(jit->Regs()[15], 1);
+}
+
+ARM_Dynarmic_32::ARM_Dynarmic_32(System& system, ExclusiveMonitor& exclusive_monitor,
+                                 std::size_t core_index)
+    : ARM_Interface{system},
+      cb(std::make_unique<DynarmicCallbacks32>(*this)), core_index{core_index},
+      exclusive_monitor{dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {}
+
+ARM_Dynarmic_32::~ARM_Dynarmic_32() = default;
+
+void ARM_Dynarmic_32::SetPC(u64 pc) {
+    jit->Regs()[15] = static_cast<u32>(pc);
+}
+
+u64 ARM_Dynarmic_32::GetPC() const {
+    return jit->Regs()[15];
+}
+
+u64 ARM_Dynarmic_32::GetReg(int index) const {
+    return jit->Regs()[index];
+}
+
+void ARM_Dynarmic_32::SetReg(int index, u64 value) {
+    jit->Regs()[index] = static_cast<u32>(value);
+}
+
+u128 ARM_Dynarmic_32::GetVectorReg(int index) const {
+    return {};
+}
+
+void ARM_Dynarmic_32::SetVectorReg(int index, u128 value) {}
+
+u32 ARM_Dynarmic_32::GetPSTATE() const {
+    return jit->Cpsr();
+}
+
+void ARM_Dynarmic_32::SetPSTATE(u32 cpsr) {
+    jit->SetCpsr(cpsr);
+}
+
+u64 ARM_Dynarmic_32::GetTlsAddress() const {
+    return CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)];
+}
+
+void ARM_Dynarmic_32::SetTlsAddress(VAddr address) {
+    CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)] = static_cast<u32>(address);
+}
+
+u64 ARM_Dynarmic_32::GetTPIDR_EL0() const {
+    return cb->tpidr_el0;
+}
+
+void ARM_Dynarmic_32::SetTPIDR_EL0(u64 value) {
+    cb->tpidr_el0 = value;
+}
+
+void ARM_Dynarmic_32::SaveContext(ThreadContext32& ctx) {
+    Dynarmic::A32::Context context;
+    jit->SaveContext(context);
+    ctx.cpu_registers = context.Regs();
+    ctx.cpsr = context.Cpsr();
+}
+
+void ARM_Dynarmic_32::LoadContext(const ThreadContext32& ctx) {
+    Dynarmic::A32::Context context;
+    context.Regs() = ctx.cpu_registers;
+    context.SetCpsr(ctx.cpsr);
+    jit->LoadContext(context);
+}
+
+void ARM_Dynarmic_32::PrepareReschedule() {
+    jit->HaltExecution();
+}
+
+void ARM_Dynarmic_32::ClearInstructionCache() {
+    jit->ClearCache();
+}
+
+void ARM_Dynarmic_32::ClearExclusiveState() {}
+
+void ARM_Dynarmic_32::PageTableChanged(Common::PageTable& page_table,
+                                       std::size_t new_address_space_size_in_bits) {
+    auto key = std::make_pair(&page_table, new_address_space_size_in_bits);
+    auto iter = jit_cache.find(key);
+    if (iter != jit_cache.end()) {
+        jit = iter->second;
+        return;
+    }
+    jit = MakeJit(page_table, new_address_space_size_in_bits);
+    jit_cache.emplace(key, jit);
+}
+
+} // namespace Core
diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.h b/src/core/arm/dynarmic/arm_dynarmic_32.h
new file mode 100644
index 000000000..143e46e4d
--- /dev/null
+++ b/src/core/arm/dynarmic/arm_dynarmic_32.h
@@ -0,0 +1,77 @@
+// Copyright 2020 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <unordered_map>
+
+#include <dynarmic/A32/a32.h>
+#include <dynarmic/A64/a64.h>
+#include <dynarmic/A64/exclusive_monitor.h>
+#include "common/common_types.h"
+#include "common/hash.h"
+#include "core/arm/arm_interface.h"
+#include "core/arm/exclusive_monitor.h"
+
+namespace Memory {
+class Memory;
+}
+
+namespace Core {
+
+class DynarmicCallbacks32;
+class DynarmicExclusiveMonitor;
+class System;
+
+class ARM_Dynarmic_32 final : public ARM_Interface {
+public:
+    ARM_Dynarmic_32(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index);
+    ~ARM_Dynarmic_32() override;
+
+    void SetPC(u64 pc) override;
+    u64 GetPC() const override;
+    u64 GetReg(int index) const override;
+    void SetReg(int index, u64 value) override;
+    u128 GetVectorReg(int index) const override;
+    void SetVectorReg(int index, u128 value) override;
+    u32 GetPSTATE() const override;
+    void SetPSTATE(u32 pstate) override;
+    void Run() override;
+    void Step() override;
+    VAddr GetTlsAddress() const override;
+    void SetTlsAddress(VAddr address) override;
+    void SetTPIDR_EL0(u64 value) override;
+    u64 GetTPIDR_EL0() const override;
+
+    void SaveContext(ThreadContext32& ctx) override;
+    void SaveContext(ThreadContext64& ctx) override {}
+    void LoadContext(const ThreadContext32& ctx) override;
+    void LoadContext(const ThreadContext64& ctx) override {}
+
+    void PrepareReschedule() override;
+    void ClearExclusiveState() override;
+
+    void ClearInstructionCache() override;
+    void PageTableChanged(Common::PageTable& new_page_table,
+                          std::size_t new_address_space_size_in_bits) override;
+
+private:
+    std::shared_ptr<Dynarmic::A32::Jit> MakeJit(Common::PageTable& page_table,
+                                                std::size_t address_space_bits) const;
+
+    using JitCacheKey = std::pair<Common::PageTable*, std::size_t>;
+    using JitCacheType =
+        std::unordered_map<JitCacheKey, std::shared_ptr<Dynarmic::A32::Jit>, Common::PairHash>;
+
+    friend class DynarmicCallbacks32;
+    std::unique_ptr<DynarmicCallbacks32> cb;
+    JitCacheType jit_cache;
+    std::shared_ptr<Dynarmic::A32::Jit> jit;
+    std::size_t core_index;
+    DynarmicExclusiveMonitor& exclusive_monitor;
+    std::array<u32, 84> CP15_regs{};
+};
+
+} // namespace Core
diff --git a/src/core/arm/dynarmic/arm_dynarmic.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp
index 7c9d59ab8..a53a58ba0 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp
@@ -8,7 +8,7 @@
 #include <dynarmic/A64/config.h>
 #include "common/logging/log.h"
 #include "common/microprofile.h"
-#include "core/arm/dynarmic/arm_dynarmic.h"
+#include "core/arm/dynarmic/arm_dynarmic_64.h"
 #include "core/core.h"
 #include "core/core_manager.h"
 #include "core/core_timing.h"
@@ -25,9 +25,9 @@ namespace Core {
 
 using Vector = Dynarmic::A64::Vector;
 
-class ARM_Dynarmic_Callbacks : public Dynarmic::A64::UserCallbacks {
+class DynarmicCallbacks64 : public Dynarmic::A64::UserCallbacks {
 public:
-    explicit ARM_Dynarmic_Callbacks(ARM_Dynarmic& parent) : parent(parent) {}
+    explicit DynarmicCallbacks64(ARM_Dynarmic_64& parent) : parent(parent) {}
 
     u8 MemoryRead8(u64 vaddr) override {
         return parent.system.Memory().Read8(vaddr);
@@ -68,7 +68,7 @@ public:
         LOG_INFO(Core_ARM, "Unicorn fallback @ 0x{:X} for {} instructions (instr = {:08X})", pc,
                  num_instructions, MemoryReadCode(pc));
 
-        ARM_Interface::ThreadContext ctx;
+        ARM_Interface::ThreadContext64 ctx;
         parent.SaveContext(ctx);
         parent.inner_unicorn.LoadContext(ctx);
         parent.inner_unicorn.ExecuteInstructions(num_instructions);
@@ -90,7 +90,7 @@ public:
                 parent.jit->HaltExecution();
                 parent.SetPC(pc);
                 Kernel::Thread* const thread = parent.system.CurrentScheduler().GetCurrentThread();
-                parent.SaveContext(thread->GetContext());
+                parent.SaveContext(thread->GetContext64());
                 GDBStub::Break();
                 GDBStub::SendTrap(thread, 5);
                 return;
@@ -126,14 +126,14 @@ public:
         return Timing::CpuCyclesToClockCycles(parent.system.CoreTiming().GetTicks());
     }
 
-    ARM_Dynarmic& parent;
+    ARM_Dynarmic_64& parent;
     std::size_t num_interpreted_instructions = 0;
     u64 tpidrro_el0 = 0;
     u64 tpidr_el0 = 0;
 };
 
-std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit(Common::PageTable& page_table,
-                                                          std::size_t address_space_bits) const {
+std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable& page_table,
+                                                             std::size_t address_space_bits) const {
     Dynarmic::A64::UserConfig config;
 
     // Callbacks
@@ -162,76 +162,76 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit(Common::PageTable& pag
     return std::make_shared<Dynarmic::A64::Jit>(config);
 }
 
-MICROPROFILE_DEFINE(ARM_Jit_Dynarmic, "ARM JIT", "Dynarmic", MP_RGB(255, 64, 64));
+MICROPROFILE_DEFINE(ARM_Jit_Dynarmic_64, "ARM JIT", "Dynarmic", MP_RGB(255, 64, 64));
 
-void ARM_Dynarmic::Run() {
-    MICROPROFILE_SCOPE(ARM_Jit_Dynarmic);
+void ARM_Dynarmic_64::Run() {
+    MICROPROFILE_SCOPE(ARM_Jit_Dynarmic_64);
 
     jit->Run();
 }
 
-void ARM_Dynarmic::Step() {
+void ARM_Dynarmic_64::Step() {
     cb->InterpreterFallback(jit->GetPC(), 1);
 }
 
-ARM_Dynarmic::ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor,
-                           std::size_t core_index)
+ARM_Dynarmic_64::ARM_Dynarmic_64(System& system, ExclusiveMonitor& exclusive_monitor,
+                                 std::size_t core_index)
     : ARM_Interface{system},
-      cb(std::make_unique<ARM_Dynarmic_Callbacks>(*this)), inner_unicorn{system},
+      cb(std::make_unique<DynarmicCallbacks64>(*this)), inner_unicorn{system},
       core_index{core_index}, exclusive_monitor{
                                   dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {}
 
-ARM_Dynarmic::~ARM_Dynarmic() = default;
+ARM_Dynarmic_64::~ARM_Dynarmic_64() = default;
 
-void ARM_Dynarmic::SetPC(u64 pc) {
+void ARM_Dynarmic_64::SetPC(u64 pc) {
     jit->SetPC(pc);
 }
 
-u64 ARM_Dynarmic::GetPC() const {
+u64 ARM_Dynarmic_64::GetPC() const {
     return jit->GetPC();
 }
 
-u64 ARM_Dynarmic::GetReg(int index) const {
+u64 ARM_Dynarmic_64::GetReg(int index) const {
     return jit->GetRegister(index);
 }
 
-void ARM_Dynarmic::SetReg(int index, u64 value) {
+void ARM_Dynarmic_64::SetReg(int index, u64 value) {
     jit->SetRegister(index, value);
 }
 
-u128 ARM_Dynarmic::GetVectorReg(int index) const {
+u128 ARM_Dynarmic_64::GetVectorReg(int index) const {
     return jit->GetVector(index);
 }
 
-void ARM_Dynarmic::SetVectorReg(int index, u128 value) {
+void ARM_Dynarmic_64::SetVectorReg(int index, u128 value) {
     jit->SetVector(index, value);
 }
 
-u32 ARM_Dynarmic::GetPSTATE() const {
+u32 ARM_Dynarmic_64::GetPSTATE() const {
     return jit->GetPstate();
 }
 
-void ARM_Dynarmic::SetPSTATE(u32 pstate) {
+void ARM_Dynarmic_64::SetPSTATE(u32 pstate) {
     jit->SetPstate(pstate);
 }
 
-u64 ARM_Dynarmic::GetTlsAddress() const {
+u64 ARM_Dynarmic_64::GetTlsAddress() const {
     return cb->tpidrro_el0;
 }
 
-void ARM_Dynarmic::SetTlsAddress(VAddr address) {
+void ARM_Dynarmic_64::SetTlsAddress(VAddr address) {
     cb->tpidrro_el0 = address;
 }
 
-u64 ARM_Dynarmic::GetTPIDR_EL0() const {
+u64 ARM_Dynarmic_64::GetTPIDR_EL0() const {
     return cb->tpidr_el0;
 }
 
-void ARM_Dynarmic::SetTPIDR_EL0(u64 value) {
+void ARM_Dynarmic_64::SetTPIDR_EL0(u64 value) {
     cb->tpidr_el0 = value;
 }
 
-void ARM_Dynarmic::SaveContext(ThreadContext& ctx) {
+void ARM_Dynarmic_64::SaveContext(ThreadContext64& ctx) {
     ctx.cpu_registers = jit->GetRegisters();
     ctx.sp = jit->GetSP();
     ctx.pc = jit->GetPC();
@@ -242,7 +242,7 @@ void ARM_Dynarmic::SaveContext(ThreadContext& ctx) {
     ctx.tpidr = cb->tpidr_el0;
 }
 
-void ARM_Dynarmic::LoadContext(const ThreadContext& ctx) {
+void ARM_Dynarmic_64::LoadContext(const ThreadContext64& ctx) {
     jit->SetRegisters(ctx.cpu_registers);
     jit->SetSP(ctx.sp);
     jit->SetPC(ctx.pc);
@@ -253,20 +253,20 @@ void ARM_Dynarmic::LoadContext(const ThreadContext& ctx) {
     SetTPIDR_EL0(ctx.tpidr);
 }
 
-void ARM_Dynarmic::PrepareReschedule() {
+void ARM_Dynarmic_64::PrepareReschedule() {
     jit->HaltExecution();
 }
 
-void ARM_Dynarmic::ClearInstructionCache() {
+void ARM_Dynarmic_64::ClearInstructionCache() {
     jit->ClearCache();
 }
 
-void ARM_Dynarmic::ClearExclusiveState() {
+void ARM_Dynarmic_64::ClearExclusiveState() {
     jit->ClearExclusiveState();
 }
 
-void ARM_Dynarmic::PageTableChanged(Common::PageTable& page_table,
-                                    std::size_t new_address_space_size_in_bits) {
+void ARM_Dynarmic_64::PageTableChanged(Common::PageTable& page_table,
+                                       std::size_t new_address_space_size_in_bits) {
     auto key = std::make_pair(&page_table, new_address_space_size_in_bits);
     auto iter = jit_cache.find(key);
     if (iter != jit_cache.end()) {
@@ -277,8 +277,8 @@ void ARM_Dynarmic::PageTableChanged(Common::PageTable& page_table,
     jit_cache.emplace(key, jit);
 }
 
-DynarmicExclusiveMonitor::DynarmicExclusiveMonitor(Memory::Memory& memory_, std::size_t core_count)
-    : monitor(core_count), memory{memory_} {}
+DynarmicExclusiveMonitor::DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count)
+    : monitor(core_count), memory{memory} {}
 
 DynarmicExclusiveMonitor::~DynarmicExclusiveMonitor() = default;
 
diff --git a/src/core/arm/dynarmic/arm_dynarmic.h b/src/core/arm/dynarmic/arm_dynarmic_64.h
index ffbb69d76..e71240a96 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.h
+++ b/src/core/arm/dynarmic/arm_dynarmic_64.h
@@ -21,18 +21,14 @@ class Memory;
 
 namespace Core {
 
-class ARM_Dynarmic_Callbacks;
+class DynarmicCallbacks64;
 class DynarmicExclusiveMonitor;
 class System;
 
-using JitCacheKey = std::pair<Common::PageTable*, std::size_t>;
-using JitCacheType =
-    std::unordered_map<JitCacheKey, std::shared_ptr<Dynarmic::A64::Jit>, Common::PairHash>;
-
-class ARM_Dynarmic final : public ARM_Interface {
+class ARM_Dynarmic_64 final : public ARM_Interface {
 public:
-    ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index);
-    ~ARM_Dynarmic() override;
+    ARM_Dynarmic_64(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index);
+    ~ARM_Dynarmic_64() override;
 
     void SetPC(u64 pc) override;
     u64 GetPC() const override;
@@ -49,8 +45,10 @@ public:
     void SetTPIDR_EL0(u64 value) override;
     u64 GetTPIDR_EL0() const override;
 
-    void SaveContext(ThreadContext& ctx) override;
-    void LoadContext(const ThreadContext& ctx) override;
+    void SaveContext(ThreadContext32& ctx) override {}
+    void SaveContext(ThreadContext64& ctx) override;
+    void LoadContext(const ThreadContext32& ctx) override {}
+    void LoadContext(const ThreadContext64& ctx) override;
 
     void PrepareReschedule() override;
     void ClearExclusiveState() override;
@@ -63,8 +61,12 @@ private:
     std::shared_ptr<Dynarmic::A64::Jit> MakeJit(Common::PageTable& page_table,
                                                 std::size_t address_space_bits) const;
 
-    friend class ARM_Dynarmic_Callbacks;
-    std::unique_ptr<ARM_Dynarmic_Callbacks> cb;
+    using JitCacheKey = std::pair<Common::PageTable*, std::size_t>;
+    using JitCacheType =
+        std::unordered_map<JitCacheKey, std::shared_ptr<Dynarmic::A64::Jit>, Common::PairHash>;
+
+    friend class DynarmicCallbacks64;
+    std::unique_ptr<DynarmicCallbacks64> cb;
     JitCacheType jit_cache;
     std::shared_ptr<Dynarmic::A64::Jit> jit;
     ARM_Unicorn inner_unicorn;
@@ -75,7 +77,7 @@ private:
 
 class DynarmicExclusiveMonitor final : public ExclusiveMonitor {
 public:
-    explicit DynarmicExclusiveMonitor(Memory::Memory& memory_, std::size_t core_count);
+    explicit DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count);
     ~DynarmicExclusiveMonitor() override;
 
     void SetExclusive(std::size_t core_index, VAddr addr) override;
@@ -88,7 +90,7 @@ public:
     bool ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) override;
 
 private:
-    friend class ARM_Dynarmic;
+    friend class ARM_Dynarmic_64;
     Dynarmic::A64::ExclusiveMonitor monitor;
     Memory::Memory& memory;
 };
diff --git a/src/core/arm/exclusive_monitor.cpp b/src/core/arm/exclusive_monitor.cpp
index 94570e520..b32401e0b 100644
--- a/src/core/arm/exclusive_monitor.cpp
+++ b/src/core/arm/exclusive_monitor.cpp
@@ -3,7 +3,7 @@
 // Refer to the license.txt file included.
 
 #ifdef ARCHITECTURE_x86_64
-#include "core/arm/dynarmic/arm_dynarmic.h"
+#include "core/arm/dynarmic/arm_dynarmic_64.h"
 #endif
 #include "core/arm/exclusive_monitor.h"
 #include "core/memory.h"
diff --git a/src/core/arm/unicorn/arm_unicorn.cpp b/src/core/arm/unicorn/arm_unicorn.cpp
index f99ad5802..8a9800a96 100644
--- a/src/core/arm/unicorn/arm_unicorn.cpp
+++ b/src/core/arm/unicorn/arm_unicorn.cpp
@@ -53,7 +53,7 @@ static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int si
                                void* user_data) {
     auto* const system = static_cast<System*>(user_data);
 
-    ARM_Interface::ThreadContext ctx{};
+    ARM_Interface::ThreadContext64 ctx{};
     system->CurrentArmInterface().SaveContext(ctx);
     ASSERT_MSG(false, "Attempted to read from unmapped memory: 0x{:X}, pc=0x{:X}, lr=0x{:X}", addr,
                ctx.pc, ctx.cpu_registers[30]);
@@ -179,7 +179,7 @@ void ARM_Unicorn::ExecuteInstructions(std::size_t num_instructions) {
         }
 
         Kernel::Thread* const thread = system.CurrentScheduler().GetCurrentThread();
-        SaveContext(thread->GetContext());
+        SaveContext(thread->GetContext64());
         if (last_bkpt_hit || GDBStub::IsMemoryBreak() || GDBStub::GetCpuStepFlag()) {
             last_bkpt_hit = false;
             GDBStub::Break();
@@ -188,7 +188,7 @@ void ARM_Unicorn::ExecuteInstructions(std::size_t num_instructions) {
     }
 }
 
-void ARM_Unicorn::SaveContext(ThreadContext& ctx) {
+void ARM_Unicorn::SaveContext(ThreadContext64& ctx) {
     int uregs[32];
     void* tregs[32];
 
@@ -215,7 +215,7 @@ void ARM_Unicorn::SaveContext(ThreadContext& ctx) {
     CHECKED(uc_reg_read_batch(uc, uregs, tregs, 32));
 }
 
-void ARM_Unicorn::LoadContext(const ThreadContext& ctx) {
+void ARM_Unicorn::LoadContext(const ThreadContext64& ctx) {
     int uregs[32];
     void* tregs[32];
 
diff --git a/src/core/arm/unicorn/arm_unicorn.h b/src/core/arm/unicorn/arm_unicorn.h
index 3c5b155f9..f30d13cb6 100644
--- a/src/core/arm/unicorn/arm_unicorn.h
+++ b/src/core/arm/unicorn/arm_unicorn.h
@@ -30,8 +30,6 @@ public:
     void SetTlsAddress(VAddr address) override;
     void SetTPIDR_EL0(u64 value) override;
     u64 GetTPIDR_EL0() const override;
-    void SaveContext(ThreadContext& ctx) override;
-    void LoadContext(const ThreadContext& ctx) override;
     void PrepareReschedule() override;
     void ClearExclusiveState() override;
     void ExecuteInstructions(std::size_t num_instructions);
@@ -41,6 +39,11 @@ public:
     void PageTableChanged(Common::PageTable&, std::size_t) override {}
     void RecordBreak(GDBStub::BreakpointAddress bkpt);
 
+    void SaveContext(ThreadContext32& ctx) override {}
+    void SaveContext(ThreadContext64& ctx) override;
+    void LoadContext(const ThreadContext32& ctx) override {}
+    void LoadContext(const ThreadContext64& ctx) override;
+
 private:
     static void InterruptHook(uc_engine* uc, u32 int_no, void* user_data);