Merge pull request #7780 from lioncash/macro

video_core/macro: Move impl classes into their cpp files

9 files changed, 204 insertions, 213 deletions

diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index b18b8a02a..c38ebd670 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -240,7 +240,7 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
         ((method - MacroRegistersStart) >> 1) % static_cast<u32>(macro_positions.size());
 
     // Execute the current macro.
-    macro_engine->Execute(*this, macro_positions[entry], parameters);
+    macro_engine->Execute(macro_positions[entry], parameters);
     if (mme_draw.current_mode != MMEDrawMode::Undefined) {
         FlushMMEInlineDraw();
     }
diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp
index d7fabe605..0aeda4ce8 100644
--- a/src/video_core/macro/macro.cpp
+++ b/src/video_core/macro/macro.cpp
@@ -2,12 +2,13 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <cstring>
 #include <optional>
+
 #include <boost/container_hash/hash.hpp>
+
 #include "common/assert.h"
-#include "common/logging/log.h"
 #include "common/settings.h"
-#include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
 #include "video_core/macro/macro_interpreter.h"
@@ -24,8 +25,7 @@ void MacroEngine::AddCode(u32 method, u32 data) {
     uploaded_macro_code[method].push_back(data);
 }
 
-void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method,
-                          const std::vector<u32>& parameters) {
+void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
     auto compiled_macro = macro_cache.find(method);
     if (compiled_macro != macro_cache.end()) {
         const auto& cache_info = compiled_macro->second;
@@ -66,10 +66,9 @@ void MacroEngine::Execute(Engines::Maxwell3D& maxwell3d, u32 method,
             cache_info.lle_program = Compile(code);
         }
 
-        auto hle_program = hle_macros->GetHLEProgram(cache_info.hash);
-        if (hle_program.has_value()) {
+        if (auto hle_program = hle_macros->GetHLEProgram(cache_info.hash)) {
             cache_info.has_hle_program = true;
-            cache_info.hle_program = std::move(hle_program.value());
+            cache_info.hle_program = std::move(hle_program);
             cache_info.hle_program->Execute(parameters, method);
         } else {
             cache_info.lle_program->Execute(parameters, method);
diff --git a/src/video_core/macro/macro.h b/src/video_core/macro/macro.h
index 31ee3440a..7aaa49286 100644
--- a/src/video_core/macro/macro.h
+++ b/src/video_core/macro/macro.h
@@ -119,7 +119,7 @@ public:
     void AddCode(u32 method, u32 data);
 
     // Compiles the macro if its not in the cache, and executes the compiled macro
-    void Execute(Engines::Maxwell3D& maxwell3d, u32 method, const std::vector<u32>& parameters);
+    void Execute(u32 method, const std::vector<u32>& parameters);
 
 protected:
     virtual std::unique_ptr<CachedMacro> Compile(const std::vector<u32>& code) = 0;
diff --git a/src/video_core/macro/macro_hle.cpp b/src/video_core/macro/macro_hle.cpp
index 70ac7c620..900ad23c9 100644
--- a/src/video_core/macro/macro_hle.cpp
+++ b/src/video_core/macro/macro_hle.cpp
@@ -5,12 +5,15 @@
 #include <array>
 #include <vector>
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
 #include "video_core/rasterizer_interface.h"
 
 namespace Tegra {
-
 namespace {
+
+using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters);
+
 // HLE'd functions
 void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
     const u32 instance_count = parameters[2] & maxwell3d.GetRegisterValue(0xD1B);
@@ -77,7 +80,6 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.CallMethodFromMME(0x8e5, 0x0);
     maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
-} // Anonymous namespace
 
 constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
     {0x771BB18C62444DA0, &HLE_771BB18C62444DA0},
@@ -85,25 +87,31 @@ constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
     {0x0217920100488FF7, &HLE_0217920100488FF7},
 }};
 
+class HLEMacroImpl final : public CachedMacro {
+public:
+    explicit HLEMacroImpl(Engines::Maxwell3D& maxwell3d_, HLEFunction func_)
+        : maxwell3d{maxwell3d_}, func{func_} {}
+
+    void Execute(const std::vector<u32>& parameters, u32 method) override {
+        func(maxwell3d, parameters);
+    }
+
+private:
+    Engines::Maxwell3D& maxwell3d;
+    HLEFunction func;
+};
+} // Anonymous namespace
+
 HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}
 HLEMacro::~HLEMacro() = default;
 
-std::optional<std::unique_ptr<CachedMacro>> HLEMacro::GetHLEProgram(u64 hash) const {
+std::unique_ptr<CachedMacro> HLEMacro::GetHLEProgram(u64 hash) const {
     const auto it = std::find_if(hle_funcs.cbegin(), hle_funcs.cend(),
                                  [hash](const auto& pair) { return pair.first == hash; });
     if (it == hle_funcs.end()) {
-        return std::nullopt;
+        return nullptr;
     }
     return std::make_unique<HLEMacroImpl>(maxwell3d, it->second);
 }
 
-HLEMacroImpl::~HLEMacroImpl() = default;
-
-HLEMacroImpl::HLEMacroImpl(Engines::Maxwell3D& maxwell3d_, HLEFunction func_)
-    : maxwell3d{maxwell3d_}, func{func_} {}
-
-void HLEMacroImpl::Execute(const std::vector<u32>& parameters, u32 method) {
-    func(maxwell3d, parameters);
-}
-
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_hle.h b/src/video_core/macro/macro_hle.h
index cb3bd1600..b86ba84a1 100644
--- a/src/video_core/macro/macro_hle.h
+++ b/src/video_core/macro/macro_hle.h
@@ -5,10 +5,7 @@
 #pragma once
 
 #include <memory>
-#include <optional>
-#include <vector>
 #include "common/common_types.h"
-#include "video_core/macro/macro.h"
 
 namespace Tegra {
 
@@ -16,29 +13,17 @@ namespace Engines {
 class Maxwell3D;
 }
 
-using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters);
-
 class HLEMacro {
 public:
     explicit HLEMacro(Engines::Maxwell3D& maxwell3d_);
     ~HLEMacro();
 
-    std::optional<std::unique_ptr<CachedMacro>> GetHLEProgram(u64 hash) const;
-
-private:
-    Engines::Maxwell3D& maxwell3d;
-};
-
-class HLEMacroImpl : public CachedMacro {
-public:
-    explicit HLEMacroImpl(Engines::Maxwell3D& maxwell3d, HLEFunction func);
-    ~HLEMacroImpl();
-
-    void Execute(const std::vector<u32>& parameters, u32 method) override;
+    // Allocates and returns a cached macro if the hash matches a known function.
+    // Returns nullptr otherwise.
+    [[nodiscard]] std::unique_ptr<CachedMacro> GetHLEProgram(u64 hash) const;
 
 private:
     Engines::Maxwell3D& maxwell3d;
-    HLEFunction func;
 };
 
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp
index 8da26fd59..fba755448 100644
--- a/src/video_core/macro/macro_interpreter.cpp
+++ b/src/video_core/macro/macro_interpreter.cpp
@@ -2,6 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <array>
+#include <optional>
+
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/microprofile.h"
@@ -11,16 +14,81 @@
 MICROPROFILE_DEFINE(MacroInterp, "GPU", "Execute macro interpreter", MP_RGB(128, 128, 192));
 
 namespace Tegra {
-MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d_)
-    : MacroEngine{maxwell3d_}, maxwell3d{maxwell3d_} {}
+namespace {
+class MacroInterpreterImpl final : public CachedMacro {
+public:
+    explicit MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d_, const std::vector<u32>& code_)
+        : maxwell3d{maxwell3d_}, code{code_} {}
 
-std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) {
-    return std::make_unique<MacroInterpreterImpl>(maxwell3d, code);
-}
+    void Execute(const std::vector<u32>& params, u32 method) override;
+
+private:
+    /// Resets the execution engine state, zeroing registers, etc.
+    void Reset();
+
+    /**
+     * Executes a single macro instruction located at the current program counter. Returns whether
+     * the interpreter should keep running.
+     *
+     * @param is_delay_slot Whether the current step is being executed due to a delay slot in a
+     *                      previous instruction.
+     */
+    bool Step(bool is_delay_slot);
+
+    /// Calculates the result of an ALU operation. src_a OP src_b;
+    u32 GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b);
+
+    /// Performs the result operation on the input result and stores it in the specified register
+    /// (if necessary).
+    void ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result);
+
+    /// Evaluates the branch condition and returns whether the branch should be taken or not.
+    bool EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const;
+
+    /// Reads an opcode at the current program counter location.
+    Macro::Opcode GetOpcode() const;
+
+    /// Returns the specified register's value. Register 0 is hardcoded to always return 0.
+    u32 GetRegister(u32 register_id) const;
+
+    /// Sets the register to the input value.
+    void SetRegister(u32 register_id, u32 value);
+
+    /// Sets the method address to use for the next Send instruction.
+    void SetMethodAddress(u32 address);
 
-MacroInterpreterImpl::MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d_,
-                                           const std::vector<u32>& code_)
-    : maxwell3d{maxwell3d_}, code{code_} {}
+    /// Calls a GPU Engine method with the input parameter.
+    void Send(u32 value);
+
+    /// Reads a GPU register located at the method address.
+    u32 Read(u32 method) const;
+
+    /// Returns the next parameter in the parameter queue.
+    u32 FetchParameter();
+
+    Engines::Maxwell3D& maxwell3d;
+
+    /// Current program counter
+    u32 pc{};
+    /// Program counter to execute at after the delay slot is executed.
+    std::optional<u32> delayed_pc;
+
+    /// General purpose macro registers.
+    std::array<u32, Macro::NUM_MACRO_REGISTERS> registers = {};
+
+    /// Method address to use for the next Send instruction.
+    Macro::MethodAddress method_address = {};
+
+    /// Input parameters of the current macro.
+    std::unique_ptr<u32[]> parameters;
+    std::size_t num_parameters = 0;
+    std::size_t parameters_capacity = 0;
+    /// Index of the next parameter that will be fetched by the 'parm' instruction.
+    u32 next_parameter_index = 0;
+
+    bool carry_flag = false;
+    const std::vector<u32>& code;
+};
 
 void MacroInterpreterImpl::Execute(const std::vector<u32>& params, u32 method) {
     MICROPROFILE_SCOPE(MacroInterp);
@@ -283,5 +351,13 @@ u32 MacroInterpreterImpl::FetchParameter() {
     ASSERT(next_parameter_index < num_parameters);
     return parameters[next_parameter_index++];
 }
+} // Anonymous namespace
+
+MacroInterpreter::MacroInterpreter(Engines::Maxwell3D& maxwell3d_)
+    : MacroEngine{maxwell3d_}, maxwell3d{maxwell3d_} {}
+
+std::unique_ptr<CachedMacro> MacroInterpreter::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroInterpreterImpl>(maxwell3d, code);
+}
 
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_interpreter.h b/src/video_core/macro/macro_interpreter.h
index d50c619ce..8a9648e46 100644
--- a/src/video_core/macro/macro_interpreter.h
+++ b/src/video_core/macro/macro_interpreter.h
@@ -3,10 +3,9 @@
 // Refer to the license.txt file included.
 
 #pragma once
-#include <array>
-#include <optional>
+
 #include <vector>
-#include "common/bit_field.h"
+
 #include "common/common_types.h"
 #include "video_core/macro/macro.h"
 
@@ -26,77 +25,4 @@ private:
     Engines::Maxwell3D& maxwell3d;
 };
 
-class MacroInterpreterImpl : public CachedMacro {
-public:
-    explicit MacroInterpreterImpl(Engines::Maxwell3D& maxwell3d_, const std::vector<u32>& code_);
-    void Execute(const std::vector<u32>& params, u32 method) override;
-
-private:
-    /// Resets the execution engine state, zeroing registers, etc.
-    void Reset();
-
-    /**
-     * Executes a single macro instruction located at the current program counter. Returns whether
-     * the interpreter should keep running.
-     *
-     * @param is_delay_slot Whether the current step is being executed due to a delay slot in a
-     *                      previous instruction.
-     */
-    bool Step(bool is_delay_slot);
-
-    /// Calculates the result of an ALU operation. src_a OP src_b;
-    u32 GetALUResult(Macro::ALUOperation operation, u32 src_a, u32 src_b);
-
-    /// Performs the result operation on the input result and stores it in the specified register
-    /// (if necessary).
-    void ProcessResult(Macro::ResultOperation operation, u32 reg, u32 result);
-
-    /// Evaluates the branch condition and returns whether the branch should be taken or not.
-    bool EvaluateBranchCondition(Macro::BranchCondition cond, u32 value) const;
-
-    /// Reads an opcode at the current program counter location.
-    Macro::Opcode GetOpcode() const;
-
-    /// Returns the specified register's value. Register 0 is hardcoded to always return 0.
-    u32 GetRegister(u32 register_id) const;
-
-    /// Sets the register to the input value.
-    void SetRegister(u32 register_id, u32 value);
-
-    /// Sets the method address to use for the next Send instruction.
-    void SetMethodAddress(u32 address);
-
-    /// Calls a GPU Engine method with the input parameter.
-    void Send(u32 value);
-
-    /// Reads a GPU register located at the method address.
-    u32 Read(u32 method) const;
-
-    /// Returns the next parameter in the parameter queue.
-    u32 FetchParameter();
-
-    Engines::Maxwell3D& maxwell3d;
-
-    /// Current program counter
-    u32 pc;
-    /// Program counter to execute at after the delay slot is executed.
-    std::optional<u32> delayed_pc;
-
-    /// General purpose macro registers.
-    std::array<u32, Macro::NUM_MACRO_REGISTERS> registers = {};
-
-    /// Method address to use for the next Send instruction.
-    Macro::MethodAddress method_address = {};
-
-    /// Input parameters of the current macro.
-    std::unique_ptr<u32[]> parameters;
-    std::size_t num_parameters = 0;
-    std::size_t parameters_capacity = 0;
-    /// Index of the next parameter that will be fetched by the 'parm' instruction.
-    u32 next_parameter_index = 0;
-
-    bool carry_flag = false;
-    const std::vector<u32>& code;
-};
-
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_jit_x64.cpp b/src/video_core/macro/macro_jit_x64.cpp
index c6b2b2109..924c9fe5c 100644
--- a/src/video_core/macro/macro_jit_x64.cpp
+++ b/src/video_core/macro/macro_jit_x64.cpp
@@ -2,9 +2,17 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <array>
+#include <bitset>
+#include <optional>
+
+#include <xbyak/xbyak.h>
+
 #include "common/assert.h"
+#include "common/bit_field.h"
 #include "common/logging/log.h"
 #include "common/microprofile.h"
+#include "common/x64/xbyak_abi.h"
 #include "common/x64/xbyak_util.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro_interpreter.h"
@@ -14,13 +22,14 @@ MICROPROFILE_DEFINE(MacroJitCompile, "GPU", "Compile macro JIT", MP_RGB(173, 255
 MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255, 0));
 
 namespace Tegra {
+namespace {
 constexpr Xbyak::Reg64 STATE = Xbyak::util::rbx;
 constexpr Xbyak::Reg32 RESULT = Xbyak::util::ebp;
 constexpr Xbyak::Reg64 PARAMETERS = Xbyak::util::r12;
 constexpr Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d;
 constexpr Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
 
-static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
+const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
     STATE,
     RESULT,
     PARAMETERS,
@@ -28,19 +37,75 @@ static const std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
     BRANCH_HOLDER,
 });
 
-MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d_)
-    : MacroEngine{maxwell3d_}, maxwell3d{maxwell3d_} {}
+// Arbitrarily chosen based on current booting games.
+constexpr size_t MAX_CODE_SIZE = 0x10000;
 
-std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
-    return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
+std::bitset<32> PersistentCallerSavedRegs() {
+    return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED;
 }
 
-MacroJITx64Impl::MacroJITx64Impl(Engines::Maxwell3D& maxwell3d_, const std::vector<u32>& code_)
-    : CodeGenerator{MAX_CODE_SIZE}, code{code_}, maxwell3d{maxwell3d_} {
-    Compile();
-}
+class MacroJITx64Impl final : public Xbyak::CodeGenerator, public CachedMacro {
+public:
+    explicit MacroJITx64Impl(Engines::Maxwell3D& maxwell3d_, const std::vector<u32>& code_)
+        : CodeGenerator{MAX_CODE_SIZE}, code{code_}, maxwell3d{maxwell3d_} {
+        Compile();
+    }
+
+    void Execute(const std::vector<u32>& parameters, u32 method) override;
+
+    void Compile_ALU(Macro::Opcode opcode);
+    void Compile_AddImmediate(Macro::Opcode opcode);
+    void Compile_ExtractInsert(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode);
+    void Compile_ExtractShiftLeftRegister(Macro::Opcode opcode);
+    void Compile_Read(Macro::Opcode opcode);
+    void Compile_Branch(Macro::Opcode opcode);
+
+private:
+    void Optimizer_ScanFlags();
+
+    void Compile();
+    bool Compile_NextInstruction();
+
+    Xbyak::Reg32 Compile_FetchParameter();
+    Xbyak::Reg32 Compile_GetRegister(u32 index, Xbyak::Reg32 dst);
+
+    void Compile_ProcessResult(Macro::ResultOperation operation, u32 reg);
+    void Compile_Send(Xbyak::Reg32 value);
 
-MacroJITx64Impl::~MacroJITx64Impl() = default;
+    Macro::Opcode GetOpCode() const;
+
+    struct JITState {
+        Engines::Maxwell3D* maxwell3d{};
+        std::array<u32, Macro::NUM_MACRO_REGISTERS> registers{};
+        u32 carry_flag{};
+    };
+    static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0");
+    using ProgramType = void (*)(JITState*, const u32*);
+
+    struct OptimizerState {
+        bool can_skip_carry{};
+        bool has_delayed_pc{};
+        bool zero_reg_skip{};
+        bool skip_dummy_addimmediate{};
+        bool optimize_for_method_move{};
+        bool enable_asserts{};
+    };
+    OptimizerState optimizer{};
+
+    std::optional<Macro::Opcode> next_opcode{};
+    ProgramType program{nullptr};
+
+    std::array<Xbyak::Label, MAX_CODE_SIZE> labels;
+    std::array<Xbyak::Label, MAX_CODE_SIZE> delay_skip;
+    Xbyak::Label end_of_code{};
+
+    bool is_delay_slot{};
+    u32 pc{};
+
+    const std::vector<u32>& code;
+    Engines::Maxwell3D& maxwell3d;
+};
 
 void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) {
     MICROPROFILE_SCOPE(MacroJitExecute);
@@ -307,11 +372,11 @@ void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
     Compile_ProcessResult(opcode.result_operation, opcode.dst);
 }
 
-static void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
+void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
     maxwell3d->CallMethodFromMME(method_address.address, value);
 }
 
-void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
+void MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
     Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
     mov(Common::X64::ABI_PARAM1, qword[STATE]);
     mov(Common::X64::ABI_PARAM2, METHOD_ADDRESS);
@@ -338,7 +403,7 @@ void Tegra::MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {
     L(dont_process);
 }
 
-void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
+void MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
     ASSERT_MSG(!is_delay_slot, "Executing a branch in a delay slot is not valid");
     const s32 jump_address =
         static_cast<s32>(pc) + static_cast<s32>(opcode.GetBranchTarget() / sizeof(s32));
@@ -392,7 +457,7 @@ void Tegra::MacroJITx64Impl::Compile_Branch(Macro::Opcode opcode) {
     L(end);
 }
 
-void Tegra::MacroJITx64Impl::Optimizer_ScanFlags() {
+void MacroJITx64Impl::Optimizer_ScanFlags() {
     optimizer.can_skip_carry = true;
     optimizer.has_delayed_pc = false;
     for (auto raw_op : code) {
@@ -534,7 +599,7 @@ bool MacroJITx64Impl::Compile_NextInstruction() {
     return true;
 }
 
-Xbyak::Reg32 Tegra::MacroJITx64Impl::Compile_FetchParameter() {
+Xbyak::Reg32 MacroJITx64Impl::Compile_FetchParameter() {
     mov(eax, dword[PARAMETERS]);
     add(PARAMETERS, sizeof(u32));
     return eax;
@@ -611,9 +676,12 @@ Macro::Opcode MacroJITx64Impl::GetOpCode() const {
     ASSERT(pc < code.size());
     return {code[pc]};
 }
+} // Anonymous namespace
 
-std::bitset<32> MacroJITx64Impl::PersistentCallerSavedRegs() const {
-    return PERSISTENT_REGISTERS & Common::X64::ABI_ALL_CALLER_SAVED;
-}
+MacroJITx64::MacroJITx64(Engines::Maxwell3D& maxwell3d_)
+    : MacroEngine{maxwell3d_}, maxwell3d{maxwell3d_} {}
 
+std::unique_ptr<CachedMacro> MacroJITx64::Compile(const std::vector<u32>& code) {
+    return std::make_unique<MacroJITx64Impl>(maxwell3d, code);
+}
 } // namespace Tegra
diff --git a/src/video_core/macro/macro_jit_x64.h b/src/video_core/macro/macro_jit_x64.h
index d03d480b4..773b037ae 100644
--- a/src/video_core/macro/macro_jit_x64.h
+++ b/src/video_core/macro/macro_jit_x64.h
@@ -4,12 +4,7 @@
 
 #pragma once
 
-#include <array>
-#include <bitset>
-#include <xbyak/xbyak.h>
-#include "common/bit_field.h"
 #include "common/common_types.h"
-#include "common/x64/xbyak_abi.h"
 #include "video_core/macro/macro.h"
 
 namespace Tegra {
@@ -18,9 +13,6 @@ namespace Engines {
 class Maxwell3D;
 }
 
-/// MAX_CODE_SIZE is arbitrarily chosen based on current booting games
-constexpr size_t MAX_CODE_SIZE = 0x10000;
-
 class MacroJITx64 final : public MacroEngine {
 public:
     explicit MacroJITx64(Engines::Maxwell3D& maxwell3d_);
@@ -32,67 +24,4 @@ private:
     Engines::Maxwell3D& maxwell3d;
 };
 
-class MacroJITx64Impl : public Xbyak::CodeGenerator, public CachedMacro {
-public:
-    explicit MacroJITx64Impl(Engines::Maxwell3D& maxwell3d_, const std::vector<u32>& code_);
-    ~MacroJITx64Impl();
-
-    void Execute(const std::vector<u32>& parameters, u32 method) override;
-
-    void Compile_ALU(Macro::Opcode opcode);
-    void Compile_AddImmediate(Macro::Opcode opcode);
-    void Compile_ExtractInsert(Macro::Opcode opcode);
-    void Compile_ExtractShiftLeftImmediate(Macro::Opcode opcode);
-    void Compile_ExtractShiftLeftRegister(Macro::Opcode opcode);
-    void Compile_Read(Macro::Opcode opcode);
-    void Compile_Branch(Macro::Opcode opcode);
-
-private:
-    void Optimizer_ScanFlags();
-
-    void Compile();
-    bool Compile_NextInstruction();
-
-    Xbyak::Reg32 Compile_FetchParameter();
-    Xbyak::Reg32 Compile_GetRegister(u32 index, Xbyak::Reg32 dst);
-
-    void Compile_ProcessResult(Macro::ResultOperation operation, u32 reg);
-    void Compile_Send(Xbyak::Reg32 value);
-
-    Macro::Opcode GetOpCode() const;
-    std::bitset<32> PersistentCallerSavedRegs() const;
-
-    struct JITState {
-        Engines::Maxwell3D* maxwell3d{};
-        std::array<u32, Macro::NUM_MACRO_REGISTERS> registers{};
-        u32 carry_flag{};
-    };
-    static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0");
-    using ProgramType = void (*)(JITState*, const u32*);
-
-    struct OptimizerState {
-        bool can_skip_carry{};
-        bool has_delayed_pc{};
-        bool zero_reg_skip{};
-        bool skip_dummy_addimmediate{};
-        bool optimize_for_method_move{};
-        bool enable_asserts{};
-    };
-    OptimizerState optimizer{};
-
-    std::optional<Macro::Opcode> next_opcode{};
-    ProgramType program{nullptr};
-
-    std::array<Xbyak::Label, MAX_CODE_SIZE> labels;
-    std::array<Xbyak::Label, MAX_CODE_SIZE> delay_skip;
-    Xbyak::Label end_of_code{};
-
-    bool is_delay_slot{};
-    u32 pc{};
-    std::optional<u32> delayed_pc;
-
-    const std::vector<u32>& code;
-    Engines::Maxwell3D& maxwell3d;
-};
-
 } // namespace Tegra