Merge pull request #3815 from FernandoS27/command-list-2

GPU: More optimizations to GPU Command List Processing and DMA Copy Optimizations

18 files changed, 198 insertions, 66 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index ff53282c9..d23c53843 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -8,6 +8,7 @@ add_library(video_core STATIC
     dma_pusher.h
     engines/const_buffer_engine_interface.h
     engines/const_buffer_info.h
+    engines/engine_interface.h
     engines/engine_upload.cpp
     engines/engine_upload.h
     engines/fermi_2d.cpp
diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index 16311f05e..bdc023d54 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -27,6 +27,8 @@ void DmaPusher::DispatchCalls() {
 
     dma_pushbuffer_subindex = 0;
 
+    dma_state.is_last_call = true;
+
     while (system.IsPoweredOn()) {
         if (!Step()) {
             break;
@@ -82,9 +84,11 @@ bool DmaPusher::Step() {
                     index);
                 CallMultiMethod(&command_header.argument, max_write);
                 dma_state.method_count -= max_write;
+                dma_state.is_last_call = true;
                 index += max_write;
                 continue;
             } else {
+                dma_state.is_last_call = dma_state.method_count <= 1;
                 CallMethod(command_header.argument);
             }
 
@@ -144,12 +148,22 @@ void DmaPusher::SetState(const CommandHeader& command_header) {
 }
 
 void DmaPusher::CallMethod(u32 argument) const {
-    gpu.CallMethod({dma_state.method, argument, dma_state.subchannel, dma_state.method_count});
+    if (dma_state.method < non_puller_methods) {
+        gpu.CallMethod({dma_state.method, argument, dma_state.subchannel, dma_state.method_count});
+    } else {
+        subchannels[dma_state.subchannel]->CallMethod(dma_state.method, argument,
+                                                      dma_state.is_last_call);
+    }
 }
 
 void DmaPusher::CallMultiMethod(const u32* base_start, u32 num_methods) const {
-    gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods,
-                        dma_state.method_count);
+    if (dma_state.method < non_puller_methods) {
+        gpu.CallMultiMethod(dma_state.method, dma_state.subchannel, base_start, num_methods,
+                            dma_state.method_count);
+    } else {
+        subchannels[dma_state.subchannel]->CallMultiMethod(dma_state.method, base_start,
+                                                           num_methods, dma_state.method_count);
+    }
 }
 
 } // namespace Tegra
diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h
index 6cef71306..e8b714e94 100644
--- a/src/video_core/dma_pusher.h
+++ b/src/video_core/dma_pusher.h
@@ -4,11 +4,13 @@
 
 #pragma once
 
+#include <array>
 #include <vector>
 #include <queue>
 
 #include "common/bit_field.h"
 #include "common/common_types.h"
+#include "video_core/engines/engine_interface.h"
 
 namespace Core {
 class System;
@@ -69,7 +71,13 @@ public:
 
     void DispatchCalls();
 
+    void BindSubchannel(Tegra::Engines::EngineInterface* engine, u32 subchannel_id) {
+        subchannels[subchannel_id] = engine;
+    }
+
 private:
+    static constexpr u32 non_puller_methods = 0x40;
+    static constexpr u32 max_subchannels = 8;
     bool Step();
 
     void SetState(const CommandHeader& command_header);
@@ -88,6 +96,7 @@ private:
         u32 method_count;      ///< Current method count
         u32 length_pending;    ///< Large NI command length pending
         bool non_incrementing; ///< Current command's NI flag
+        bool is_last_call;
     };
 
     DmaState dma_state{};
@@ -96,6 +105,8 @@ private:
     GPUVAddr dma_mget{};  ///< main pushbuffer last read address
     bool ib_enable{true}; ///< IB mode enabled
 
+    std::array<Tegra::Engines::EngineInterface*, max_subchannels> subchannels{};
+
     GPU& gpu;
     Core::System& system;
 };
diff --git a/src/video_core/engines/engine_interface.h b/src/video_core/engines/engine_interface.h
new file mode 100644
index 000000000..18a9db7e6
--- /dev/null
+++ b/src/video_core/engines/engine_interface.h
@@ -0,0 +1,22 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <type_traits>
+#include "common/common_types.h"
+
+namespace Tegra::Engines {
+
+class EngineInterface {
+public:
+    /// Write the value to the register identified by method.
+    virtual void CallMethod(u32 method, u32 method_argument, bool is_last_call) = 0;
+
+    /// Write multiple values to the register identified by method.
+    virtual void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                 u32 methods_pending) = 0;
+};
+
+} // namespace Tegra::Engines
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 8a47614d2..ff10ff40d 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -12,13 +12,13 @@ namespace Tegra::Engines {
 
 Fermi2D::Fermi2D(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {}
 
-void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
-    ASSERT_MSG(method_call.method < Regs::NUM_REGS,
+void Fermi2D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Fermi2D register, increase the size of the Regs structure");
 
-    regs.reg_array[method_call.method] = method_call.argument;
+    regs.reg_array[method] = method_argument;
 
-    switch (method_call.method) {
+    switch (method) {
     // Trigger the surface copy on the last register write. This is blit_src_y, but this is 64-bit,
     // so trigger on the second 32-bit write.
     case FERMI2D_REG_INDEX(blit_src_y) + 1: {
@@ -30,7 +30,7 @@ void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
 
 void Fermi2D::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) {
     for (std::size_t i = 0; i < amount; i++) {
-        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
     }
 }
 
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 939a5966d..8f37d053f 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -10,6 +10,7 @@
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/math_util.h"
+#include "video_core/engines/engine_interface.h"
 #include "video_core/gpu.h"
 
 namespace Tegra {
@@ -31,16 +32,17 @@ namespace Tegra::Engines {
 #define FERMI2D_REG_INDEX(field_name)                                                              \
     (offsetof(Tegra::Engines::Fermi2D::Regs, field_name) / sizeof(u32))
 
-class Fermi2D final {
+class Fermi2D final : public EngineInterface {
 public:
     explicit Fermi2D(VideoCore::RasterizerInterface& rasterizer);
     ~Fermi2D() = default;
 
     /// Write the value to the register identified by method.
-    void CallMethod(const GPU::MethodCall& method_call);
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
     /// Write multiple values to the register identified by method.
-    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
     enum class Origin : u32 {
         Center = 0,
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index 00a12175f..f6237fc6a 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -24,20 +24,19 @@ KeplerCompute::KeplerCompute(Core::System& system, VideoCore::RasterizerInterfac
 
 KeplerCompute::~KeplerCompute() = default;
 
-void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
-    ASSERT_MSG(method_call.method < Regs::NUM_REGS,
+void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid KeplerCompute register, increase the size of the Regs structure");
 
-    regs.reg_array[method_call.method] = method_call.argument;
+    regs.reg_array[method] = method_argument;
 
-    switch (method_call.method) {
+    switch (method) {
     case KEPLER_COMPUTE_REG_INDEX(exec_upload): {
         upload_state.ProcessExec(regs.exec_upload.linear != 0);
         break;
     }
     case KEPLER_COMPUTE_REG_INDEX(data_upload): {
-        const bool is_last_call = method_call.IsLastCall();
-        upload_state.ProcessData(method_call.argument, is_last_call);
+        upload_state.ProcessData(method_argument, is_last_call);
         if (is_last_call) {
             system.GPU().Maxwell3D().OnMemoryWrite();
         }
@@ -54,7 +53,7 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
 void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                     u32 methods_pending) {
     for (std::size_t i = 0; i < amount; i++) {
-        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
     }
 }
 
diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h
index fe55fdfd0..18ceedfaf 100644
--- a/src/video_core/engines/kepler_compute.h
+++ b/src/video_core/engines/kepler_compute.h
@@ -11,6 +11,7 @@
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "video_core/engines/const_buffer_engine_interface.h"
+#include "video_core/engines/engine_interface.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/gpu.h"
@@ -39,7 +40,7 @@ namespace Tegra::Engines {
 #define KEPLER_COMPUTE_REG_INDEX(field_name)                                                       \
     (offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32))
 
-class KeplerCompute final : public ConstBufferEngineInterface {
+class KeplerCompute final : public ConstBufferEngineInterface, public EngineInterface {
 public:
     explicit KeplerCompute(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                            MemoryManager& memory_manager);
@@ -200,10 +201,11 @@ public:
                   "KeplerCompute LaunchParams has wrong size");
 
     /// Write the value to the register identified by method.
-    void CallMethod(const GPU::MethodCall& method_call);
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
     /// Write multiple values to the register identified by method.
-    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
     Texture::FullTextureInfo GetTexture(std::size_t offset) const;
 
diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp
index 586ff15dc..dc71b2eec 100644
--- a/src/video_core/engines/kepler_memory.cpp
+++ b/src/video_core/engines/kepler_memory.cpp
@@ -19,20 +19,19 @@ KeplerMemory::KeplerMemory(Core::System& system, MemoryManager& memory_manager)
 
 KeplerMemory::~KeplerMemory() = default;
 
-void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
-    ASSERT_MSG(method_call.method < Regs::NUM_REGS,
+void KeplerMemory::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid KeplerMemory register, increase the size of the Regs structure");
 
-    regs.reg_array[method_call.method] = method_call.argument;
+    regs.reg_array[method] = method_argument;
 
-    switch (method_call.method) {
+    switch (method) {
     case KEPLERMEMORY_REG_INDEX(exec): {
         upload_state.ProcessExec(regs.exec.linear != 0);
         break;
     }
     case KEPLERMEMORY_REG_INDEX(data): {
-        const bool is_last_call = method_call.IsLastCall();
-        upload_state.ProcessData(method_call.argument, is_last_call);
+        upload_state.ProcessData(method_argument, is_last_call);
         if (is_last_call) {
             system.GPU().Maxwell3D().OnMemoryWrite();
         }
@@ -44,7 +43,7 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
 void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                    u32 methods_pending) {
     for (std::size_t i = 0; i < amount; i++) {
-        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
     }
 }
 
diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h
index bb26fb030..5b7f71a00 100644
--- a/src/video_core/engines/kepler_memory.h
+++ b/src/video_core/engines/kepler_memory.h
@@ -10,6 +10,7 @@
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "video_core/engines/engine_interface.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/gpu.h"
 
@@ -32,16 +33,17 @@ namespace Tegra::Engines {
 #define KEPLERMEMORY_REG_INDEX(field_name)                                                         \
     (offsetof(Tegra::Engines::KeplerMemory::Regs, field_name) / sizeof(u32))
 
-class KeplerMemory final {
+class KeplerMemory final : public EngineInterface {
 public:
     KeplerMemory(Core::System& system, MemoryManager& memory_manager);
     ~KeplerMemory();
 
     /// Write the value to the register identified by method.
-    void CallMethod(const GPU::MethodCall& method_call);
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
     /// Write multiple values to the register identified by method.
-    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
     struct Regs {
         static constexpr size_t NUM_REGS = 0x7F;
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 7db055ea0..33936e209 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -125,12 +125,10 @@ void Maxwell3D::CallMacroMethod(u32 method, std::size_t num_parameters, const u3
     }
 }
 
-void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
-    const u32 method = method_call.method;
-
+void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
     if (method == cb_data_state.current) {
-        regs.reg_array[method] = method_call.argument;
-        ProcessCBData(method_call.argument);
+        regs.reg_array[method] = method_argument;
+        ProcessCBData(method_argument);
         return;
     } else if (cb_data_state.current != null_cb_data) {
         FinishCBData();
@@ -153,10 +151,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
             executing_macro = method;
         }
 
-        macro_params.push_back(method_call.argument);
+        macro_params.push_back(method_argument);
 
         // Call the macro when there are no more parameters in the command buffer
-        if (method_call.IsLastCall()) {
+        if (is_last_call) {
             CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
             macro_params.clear();
         }
@@ -166,7 +164,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Maxwell3D register, increase the size of the Regs structure");
 
-    u32 arg = method_call.argument;
+    u32 arg = method_argument;
     // Keep track of the register value in shadow_state when requested.
     if (shadow_state.shadow_ram_control == Regs::ShadowRamControl::Track ||
         shadow_state.shadow_ram_control == Regs::ShadowRamControl::TrackWithFilter) {
@@ -189,7 +187,7 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         break;
     }
     case MAXWELL3D_REG_INDEX(shadow_ram_control): {
-        shadow_state.shadow_ram_control = static_cast<Regs::ShadowRamControl>(method_call.argument);
+        shadow_state.shadow_ram_control = static_cast<Regs::ShadowRamControl>(method_argument);
         break;
     }
     case MAXWELL3D_REG_INDEX(macros.data): {
@@ -272,7 +270,6 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         break;
     }
     case MAXWELL3D_REG_INDEX(data_upload): {
-        const bool is_last_call = method_call.IsLastCall();
         upload_state.ProcessData(arg, is_last_call);
         if (is_last_call) {
             OnMemoryWrite();
@@ -330,7 +327,7 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
     }
     default: {
         for (std::size_t i = 0; i < amount; i++) {
-            CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+            CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
         }
     }
     }
@@ -360,16 +357,15 @@ void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) {
     StepInstance(expected_mode, count);
 }
 
-void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) {
-    const u32 method = method_call.method;
+void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) {
     if (mme_inline[method]) {
-        regs.reg_array[method] = method_call.argument;
+        regs.reg_array[method] = method_argument;
         if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) ||
             method == MAXWELL3D_REG_INDEX(index_array.count)) {
             const MMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count)
                                                   ? MMEDrawMode::Array
                                                   : MMEDrawMode::Indexed;
-            StepInstance(expected_mode, method_call.argument);
+            StepInstance(expected_mode, method_argument);
         } else if (method == MAXWELL3D_REG_INDEX(draw.vertex_begin_gl)) {
             mme_draw.instance_mode =
                 (regs.draw.instance_next != 0) || (regs.draw.instance_cont != 0);
@@ -381,7 +377,7 @@ void Maxwell3D::CallMethodFromMME(const GPU::MethodCall& method_call) {
         if (mme_draw.current_mode != MMEDrawMode::Undefined) {
             FlushMMEInlineDraw();
         }
-        CallMethod(method_call);
+        CallMethod(method, method_argument, true);
     }
 }
 
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 864924ff3..1a5df05ce 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -19,6 +19,7 @@
 #include "common/math_util.h"
 #include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/const_buffer_info.h"
+#include "video_core/engines/engine_interface.h"
 #include "video_core/engines/engine_upload.h"
 #include "video_core/engines/shader_type.h"
 #include "video_core/gpu.h"
@@ -48,7 +49,7 @@ namespace Tegra::Engines {
 #define MAXWELL3D_REG_INDEX(field_name)                                                            \
     (offsetof(Tegra::Engines::Maxwell3D::Regs, field_name) / sizeof(u32))
 
-class Maxwell3D final : public ConstBufferEngineInterface {
+class Maxwell3D final : public ConstBufferEngineInterface, public EngineInterface {
 public:
     explicit Maxwell3D(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
                        MemoryManager& memory_manager);
@@ -1360,13 +1361,14 @@ public:
     u32 GetRegisterValue(u32 method) const;
 
     /// Write the value to the register identified by method.
-    void CallMethod(const GPU::MethodCall& method_call);
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
     /// Write multiple values to the register identified by method.
-    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
     /// Write the value to the register identified by method.
-    void CallMethodFromMME(const GPU::MethodCall& method_call);
+    void CallMethodFromMME(u32 method, u32 method_argument);
 
     void FlushMMEInlineDraw();
 
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 6630005b0..01d7df405 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -17,16 +17,16 @@ namespace Tegra::Engines {
 MaxwellDMA::MaxwellDMA(Core::System& system, MemoryManager& memory_manager)
     : system{system}, memory_manager{memory_manager} {}
 
-void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
-    ASSERT_MSG(method_call.method < Regs::NUM_REGS,
+void MaxwellDMA::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid MaxwellDMA register, increase the size of the Regs structure");
 
-    regs.reg_array[method_call.method] = method_call.argument;
+    regs.reg_array[method] = method_argument;
 
 #define MAXWELLDMA_REG_INDEX(field_name)                                                           \
     (offsetof(Tegra::Engines::MaxwellDMA::Regs, field_name) / sizeof(u32))
 
-    switch (method_call.method) {
+    switch (method) {
     case MAXWELLDMA_REG_INDEX(exec): {
         HandleCopy();
         break;
@@ -39,7 +39,7 @@ void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) {
 void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                  u32 methods_pending) {
     for (std::size_t i = 0; i < amount; i++) {
-        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
     }
 }
 
@@ -90,7 +90,47 @@ void MaxwellDMA::HandleCopy() {
     ASSERT(regs.exec.enable_2d == 1);
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
+
         ASSERT(regs.src_params.BlockDepth() == 0);
+        // Optimized path for micro copies.
+        if (regs.dst_pitch * regs.y_count < Texture::GetGOBSize() && regs.dst_pitch <= 64) {
+            const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
+            const std::size_t src_size = Texture::GetGOBSize();
+            const std::size_t dst_size = regs.dst_pitch * regs.y_count;
+            u32 pos_x = regs.src_params.pos_x;
+            u32 pos_y = regs.src_params.pos_y;
+            const u64 offset =
+                Texture::GetGOBOffset(regs.src_params.size_x, regs.src_params.size_y, pos_x, pos_y,
+                                      regs.src_params.BlockDepth(), bytes_per_pixel);
+            const u32 x_in_gob = 64 / bytes_per_pixel;
+            pos_x = pos_x % x_in_gob;
+            pos_y = pos_y % 8;
+
+            if (read_buffer.size() < src_size) {
+                read_buffer.resize(src_size);
+            }
+
+            if (write_buffer.size() < dst_size) {
+                write_buffer.resize(dst_size);
+            }
+
+            if (Settings::IsGPULevelExtreme()) {
+                memory_manager.ReadBlock(source + offset, read_buffer.data(), src_size);
+                memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
+            } else {
+                memory_manager.ReadBlockUnsafe(source + offset, read_buffer.data(), src_size);
+                memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
+            }
+
+            Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
+                                      regs.src_params.size_x, bytes_per_pixel, read_buffer.data(),
+                                      write_buffer.data(), regs.src_params.BlockHeight(), pos_x,
+                                      pos_y);
+
+            memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
+
+            return;
+        }
         // If the input is tiled and the output is linear, deswizzle the input and copy it over.
         const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
         const std::size_t src_size = Texture::CalculateSize(
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index c43ed8194..502dd8509 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -10,6 +10,7 @@
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "video_core/engines/engine_interface.h"
 #include "video_core/gpu.h"
 
 namespace Core {
@@ -27,16 +28,17 @@ namespace Tegra::Engines {
  * https://github.com/envytools/envytools/blob/master/rnndb/fifo/gk104_copy.xml
  */
 
-class MaxwellDMA final {
+class MaxwellDMA final : public EngineInterface {
 public:
     explicit MaxwellDMA(Core::System& system, MemoryManager& memory_manager);
     ~MaxwellDMA() = default;
 
     /// Write the value to the register identified by method.
-    void CallMethod(const GPU::MethodCall& method_call);
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
     /// Write multiple values to the register identified by method.
-    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
     struct Regs {
         static constexpr std::size_t NUM_REGS = 0x1D6;
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index b87fd873d..8eb017f65 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -299,19 +299,21 @@ void GPU::CallEngineMethod(const MethodCall& method_call) {
 
     switch (engine) {
     case EngineID::FERMI_TWOD_A:
-        fermi_2d->CallMethod(method_call);
+        fermi_2d->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
         break;
     case EngineID::MAXWELL_B:
-        maxwell_3d->CallMethod(method_call);
+        maxwell_3d->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
         break;
     case EngineID::KEPLER_COMPUTE_B:
-        kepler_compute->CallMethod(method_call);
+        kepler_compute->CallMethod(method_call.method, method_call.argument,
+                                   method_call.IsLastCall());
         break;
     case EngineID::MAXWELL_DMA_COPY_A:
-        maxwell_dma->CallMethod(method_call);
+        maxwell_dma->CallMethod(method_call.method, method_call.argument, method_call.IsLastCall());
         break;
     case EngineID::KEPLER_INLINE_TO_MEMORY_B:
-        kepler_memory->CallMethod(method_call);
+        kepler_memory->CallMethod(method_call.method, method_call.argument,
+                                  method_call.IsLastCall());
         break;
     default:
         UNIMPLEMENTED_MSG("Unimplemented engine");
@@ -347,7 +349,27 @@ void GPU::ProcessBindMethod(const MethodCall& method_call) {
     // Bind the current subchannel to the desired engine id.
     LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel,
               method_call.argument);
-    bound_engines[method_call.subchannel] = static_cast<EngineID>(method_call.argument);
+    const auto engine_id = static_cast<EngineID>(method_call.argument);
+    bound_engines[method_call.subchannel] = static_cast<EngineID>(engine_id);
+    switch (engine_id) {
+    case EngineID::FERMI_TWOD_A:
+        dma_pusher->BindSubchannel(fermi_2d.get(), method_call.subchannel);
+        break;
+    case EngineID::MAXWELL_B:
+        dma_pusher->BindSubchannel(maxwell_3d.get(), method_call.subchannel);
+        break;
+    case EngineID::KEPLER_COMPUTE_B:
+        dma_pusher->BindSubchannel(kepler_compute.get(), method_call.subchannel);
+        break;
+    case EngineID::MAXWELL_DMA_COPY_A:
+        dma_pusher->BindSubchannel(maxwell_dma.get(), method_call.subchannel);
+        break;
+    case EngineID::KEPLER_INLINE_TO_MEMORY_B:
+        dma_pusher->BindSubchannel(kepler_memory.get(), method_call.subchannel);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", static_cast<u32>(engine_id));
+    }
 }
 
 void GPU::ProcessSemaphoreTriggerMethod() {
diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp
index 42031d80a..947364928 100644
--- a/src/video_core/macro_interpreter.cpp
+++ b/src/video_core/macro_interpreter.cpp
@@ -328,7 +328,7 @@ void MacroInterpreter::SetMethodAddress(u32 address) {
 }
 
 void MacroInterpreter::Send(u32 value) {
-    maxwell3d.CallMethodFromMME({method_address.address, value});
+    maxwell3d.CallMethodFromMME(method_address.address, value);
     // Increment the method address by the method increment.
     method_address.address.Assign(method_address.address.Value() +
                                   method_address.increment.Value());
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index fae8638ec..548e4c3fe 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -382,4 +382,18 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height
     }
 }
 
+u64 GetGOBOffset(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height,
+                 u32 bytes_per_pixel) {
+    auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
+    const u32 gobs_in_block = 1 << block_height;
+    const u32 y_blocks = gob_size_y << block_height;
+    const u32 x_per_gob = gob_size_x / bytes_per_pixel;
+    const u32 x_blocks = div_ceil(width, x_per_gob);
+    const u32 block_size = gob_size * gobs_in_block;
+    const u32 stride = block_size * x_blocks;
+    const u32 base = (dst_y / y_blocks) * stride + (dst_x / x_per_gob) * block_size;
+    const u32 relative_y = dst_y % y_blocks;
+    return base + (relative_y / gob_size_y) * gob_size;
+}
+
 } // namespace Tegra::Texture
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index 9f2d6d308..06f3ebf87 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -59,4 +59,8 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32
 void SwizzleKepler(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height,
                    std::size_t copy_size, const u8* source_data, u8* swizzle_data);
 
+/// Obtains the offset of the gob for positions 'dst_x' & 'dst_y'
+u64 GetGOBOffset(u32 width, u32 height, u32 dst_x, u32 dst_y, u32 block_height,
+                 u32 bytes_per_pixel);
+
 } // namespace Tegra::Texture