2 files changed, 360 insertions, 288 deletions

diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 01d7df405..a2d3d7823 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -14,50 +14,45 @@
 
 namespace Tegra::Engines {
 
+using namespace Texture;
+
 MaxwellDMA::MaxwellDMA(Core::System& system, MemoryManager& memory_manager)
     : system{system}, memory_manager{memory_manager} {}
 
 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");
+    ASSERT_MSG(method < NUM_REGS, "Invalid MaxwellDMA register");
 
     regs.reg_array[method] = method_argument;
 
-#define MAXWELLDMA_REG_INDEX(field_name)                                                           \
-    (offsetof(Tegra::Engines::MaxwellDMA::Regs, field_name) / sizeof(u32))
-
-    switch (method) {
-    case MAXWELLDMA_REG_INDEX(exec): {
-        HandleCopy();
-        break;
-    }
+    if (method == offsetof(Regs, launch_dma) / sizeof(u32)) {
+        Launch();
     }
-
-#undef MAXWELLDMA_REG_INDEX
 }
 
 void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                  u32 methods_pending) {
-    for (std::size_t i = 0; i < amount; i++) {
+    for (size_t i = 0; i < amount; ++i) {
         CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
     }
 }
 
-void MaxwellDMA::HandleCopy() {
-    LOG_TRACE(HW_GPU, "Requested a DMA copy");
-
-    const GPUVAddr source = regs.src_address.Address();
-    const GPUVAddr dest = regs.dst_address.Address();
+void MaxwellDMA::Launch() {
+    LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in),
+              static_cast<GPUVAddr>(regs.offset_out));
 
     // TODO(Subv): Perform more research and implement all features of this engine.
-    ASSERT(regs.exec.enable_swizzle == 0);
-    ASSERT(regs.exec.query_mode == Regs::QueryMode::None);
-    ASSERT(regs.exec.query_intr == Regs::QueryIntr::None);
-    ASSERT(regs.exec.copy_mode == Regs::CopyMode::Unk2);
-    ASSERT(regs.dst_params.pos_x == 0);
-    ASSERT(regs.dst_params.pos_y == 0);
-
-    if (!regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
+    const LaunchDMA& launch = regs.launch_dma;
+    ASSERT(launch.remap_enable == 0);
+    ASSERT(launch.semaphore_type == LaunchDMA::SemaphoreType::NONE);
+    ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE);
+    ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED);
+    ASSERT(regs.dst_params.origin.x == 0);
+    ASSERT(regs.dst_params.origin.y == 0);
+
+    const bool is_src_pitch = launch.src_memory_layout == LaunchDMA::MemoryLayout::PITCH;
+    const bool is_dst_pitch = launch.dst_memory_layout == LaunchDMA::MemoryLayout::PITCH;
+
+    if (!is_src_pitch && !is_dst_pitch) {
         // If both the source and the destination are in block layout, assert.
         UNREACHABLE_MSG("Tiled->Tiled DMA transfers are not yet implemented");
         return;
@@ -66,144 +61,161 @@ void MaxwellDMA::HandleCopy() {
     // All copies here update the main memory, so mark all rasterizer states as invalid.
     system.GPU().Maxwell3D().OnMemoryWrite();
 
-    if (regs.exec.is_dst_linear && regs.exec.is_src_linear) {
-        // When the enable_2d bit is disabled, the copy is performed as if we were copying a 1D
-        // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count,
-        // y_count).
-        if (!regs.exec.enable_2d) {
-            memory_manager.CopyBlock(dest, source, regs.x_count);
-            return;
-        }
+    if (is_src_pitch && is_dst_pitch) {
+        CopyPitchToPitch();
+    } else {
+        ASSERT(launch.multi_line_enable == 1);
 
-        // If both the source and the destination are in linear layout, perform a line-by-line
-        // copy. We're going to take a subrect of size (x_count, y_count) from the source
-        // rectangle. There is no need to manually flush/invalidate the regions because
-        // CopyBlock does that for us.
-        for (u32 line = 0; line < regs.y_count; ++line) {
-            const GPUVAddr source_line = source + line * regs.src_pitch;
-            const GPUVAddr dest_line = dest + line * regs.dst_pitch;
-            memory_manager.CopyBlock(dest_line, source_line, regs.x_count);
+        if (!is_src_pitch && is_dst_pitch) {
+            CopyBlockLinearToPitch();
+        } else {
+            CopyPitchToBlockLinear();
         }
-        return;
     }
+}
 
-    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(
-            true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y,
-            regs.src_params.size_z, regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
-
-        const std::size_t src_layer_size = Texture::CalculateSize(
-            true, bytes_per_pixel, regs.src_params.size_x, regs.src_params.size_y, 1,
-            regs.src_params.BlockHeight(), regs.src_params.BlockDepth());
-
-        const std::size_t dst_size = regs.dst_pitch * regs.y_count;
+void MaxwellDMA::CopyPitchToPitch() {
+    // When `multi_line_enable` bit is disabled the copy is performed as if we were copying a 1D
+    // buffer of length `line_length_in`.
+    // Otherwise we copy a 2D image of dimensions (line_length_in, line_count).
+    if (!regs.launch_dma.multi_line_enable) {
+        memory_manager.CopyBlock(regs.offset_out, regs.offset_in, regs.line_length_in);
+        return;
+    }
 
-        if (read_buffer.size() < src_size) {
-            read_buffer.resize(src_size);
-        }
+    // Perform a line-by-line copy.
+    // We're going to take a subrect of size (line_length_in, line_count) from the source rectangle.
+    // There is no need to manually flush/invalidate the regions because CopyBlock does that for us.
+    for (u32 line = 0; line < regs.line_count; ++line) {
+        const GPUVAddr source_line = regs.offset_in + static_cast<size_t>(line) * regs.pitch_in;
+        const GPUVAddr dest_line = regs.offset_out + static_cast<size_t>(line) * regs.pitch_out;
+        memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in);
+    }
+}
 
-        if (write_buffer.size() < dst_size) {
-            write_buffer.resize(dst_size);
-        }
+void MaxwellDMA::CopyBlockLinearToPitch() {
+    ASSERT(regs.src_params.block_size.depth == 0);
 
-        if (Settings::IsGPULevelExtreme()) {
-            memory_manager.ReadBlock(source, read_buffer.data(), src_size);
-            memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
-        } else {
-            memory_manager.ReadBlockUnsafe(source, read_buffer.data(), src_size);
-            memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
-        }
+    // Optimized path for micro copies.
+    const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
+    if (dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X) {
+        FastCopyBlockLinearToPitch();
+        return;
+    }
 
-        Texture::UnswizzleSubrect(
-            regs.x_count, regs.y_count, regs.dst_pitch, regs.src_params.size_x, bytes_per_pixel,
-            read_buffer.data() + src_layer_size * regs.src_params.pos_z, write_buffer.data(),
-            regs.src_params.BlockHeight(), regs.src_params.pos_x, regs.src_params.pos_y);
+    // Deswizzle the input and copy it over.
+    const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in;
+    const Parameters& src_params = regs.src_params;
+    const u32 width = src_params.width;
+    const u32 height = src_params.height;
+    const u32 depth = src_params.depth;
+    const u32 block_height = src_params.block_size.height;
+    const u32 block_depth = src_params.block_size.depth;
+    const size_t src_size =
+        CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);
+    const size_t src_layer_size =
+        CalculateSize(true, bytes_per_pixel, width, height, 1, block_height, block_depth);
+
+    if (read_buffer.size() < src_size) {
+        read_buffer.resize(src_size);
+    }
+    if (write_buffer.size() < dst_size) {
+        write_buffer.resize(dst_size);
+    }
 
-        memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
+    if (Settings::IsGPULevelExtreme()) {
+        memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
     } else {
-        ASSERT(regs.dst_params.BlockDepth() == 0);
+        memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
+    }
 
-        const u32 bytes_per_pixel = regs.src_pitch / regs.x_count;
+    UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, width, bytes_per_pixel,
+                     read_buffer.data() + src_layer_size * src_params.layer, write_buffer.data(),
+                     block_height, src_params.origin.x, src_params.origin.y);
 
-        const std::size_t dst_size = Texture::CalculateSize(
-            true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y,
-            regs.dst_params.size_z, regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
+    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
+}
 
-        const std::size_t dst_layer_size = Texture::CalculateSize(
-            true, bytes_per_pixel, regs.dst_params.size_x, regs.dst_params.size_y, 1,
-            regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth());
+void MaxwellDMA::CopyPitchToBlockLinear() {
+    const auto& dst_params = regs.dst_params;
+    const u32 bytes_per_pixel = regs.pitch_in / regs.line_length_in;
+    const u32 width = dst_params.width;
+    const u32 height = dst_params.height;
+    const u32 depth = dst_params.depth;
+    const u32 block_height = dst_params.block_size.height;
+    const u32 block_depth = dst_params.block_size.depth;
+    const size_t dst_size =
+        CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);
+    const size_t dst_layer_size =
+        CalculateSize(true, bytes_per_pixel, width, height, 1, block_height, block_depth);
+
+    const size_t src_size = static_cast<size_t>(regs.pitch_in) * regs.line_count;
+
+    if (read_buffer.size() < src_size) {
+        read_buffer.resize(src_size);
+    }
+    if (write_buffer.size() < dst_size) {
+        write_buffer.resize(dst_size);
+    }
 
-        const std::size_t src_size = regs.src_pitch * regs.y_count;
+    if (Settings::IsGPULevelExtreme()) {
+        memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+    } else {
+        memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
+    }
 
-        if (read_buffer.size() < src_size) {
-            read_buffer.resize(src_size);
-        }
+    // If the input is linear and the output is tiled, swizzle the input and copy it over.
+    if (regs.dst_params.block_size.depth > 0) {
+        ASSERT(dst_params.layer == 0);
+        SwizzleSliceToVoxel(regs.line_length_in, regs.line_count, regs.pitch_in, width, height,
+                            bytes_per_pixel, block_height, block_depth, dst_params.origin.x,
+                            dst_params.origin.y, write_buffer.data(), read_buffer.data());
+    } else {
+        SwizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_in, width, bytes_per_pixel,
+                       write_buffer.data() + dst_layer_size * dst_params.layer, read_buffer.data(),
+                       block_height, dst_params.origin.x, dst_params.origin.y);
+    }
 
-        if (write_buffer.size() < dst_size) {
-            write_buffer.resize(dst_size);
-        }
+    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
+}
 
-        if (Settings::IsGPULevelExtreme()) {
-            memory_manager.ReadBlock(source, read_buffer.data(), src_size);
-            memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
-        } else {
-            memory_manager.ReadBlockUnsafe(source, read_buffer.data(), src_size);
-            memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
-        }
+void MaxwellDMA::FastCopyBlockLinearToPitch() {
+    const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in;
+    const size_t src_size = GOB_SIZE;
+    const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
+    u32 pos_x = regs.src_params.origin.x;
+    u32 pos_y = regs.src_params.origin.y;
+    const u64 offset = GetGOBOffset(regs.src_params.width, regs.src_params.height, pos_x, pos_y,
+                                    regs.src_params.block_size.height, 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 the input is linear and the output is tiled, swizzle the input and copy it over.
-        Texture::SwizzleSubrect(
-            regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x, bytes_per_pixel,
-            write_buffer.data() + dst_layer_size * regs.dst_params.pos_z, read_buffer.data(),
-            regs.dst_params.BlockHeight(), regs.dst_params.pos_x, regs.dst_params.pos_y);
+    if (write_buffer.size() < dst_size) {
+        write_buffer.resize(dst_size);
+    }
 
-        memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
+    if (Settings::IsGPULevelExtreme()) {
+        memory_manager.ReadBlock(regs.offset_in + offset, read_buffer.data(), src_size);
+        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+    } else {
+        memory_manager.ReadBlockUnsafe(regs.offset_in + offset, read_buffer.data(), src_size);
+        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
     }
+
+    UnswizzleSubrect(regs.line_length_in, regs.line_count, regs.pitch_out, regs.src_params.width,
+                     bytes_per_pixel, read_buffer.data(), write_buffer.data(),
+                     regs.src_params.block_size.height, pos_x, pos_y);
+
+    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
 }
 
 } // namespace Tegra::Engines
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index 502dd8509..50f445efc 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -24,160 +24,190 @@ class MemoryManager;
 namespace Tegra::Engines {
 
 /**
- * This Engine is known as GK104_Copy. Documentation can be found in:
+ * This engine is known as gk104_copy. Documentation can be found in:
+ * https://github.com/NVIDIA/open-gpu-doc/blob/master/classes/dma-copy/clb0b5.h
  * https://github.com/envytools/envytools/blob/master/rnndb/fifo/gk104_copy.xml
  */
 
 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(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) override;
+    struct PackedGPUVAddr {
+        u32 upper;
+        u32 lower;
+
+        constexpr operator GPUVAddr() const noexcept {
+            return (static_cast<GPUVAddr>(upper & 0xff) << 32) | lower;
+        }
+    };
+
+    union BlockSize {
+        BitField<0, 4, u32> width;
+        BitField<4, 4, u32> height;
+        BitField<8, 4, u32> depth;
+        BitField<12, 4, u32> gob_height;
+    };
+    static_assert(sizeof(BlockSize) == 4);
+
+    union Origin {
+        BitField<0, 16, u32> x;
+        BitField<16, 16, u32> y;
+    };
+    static_assert(sizeof(Origin) == 4);
+
+    struct Parameters {
+        BlockSize block_size;
+        u32 width;
+        u32 height;
+        u32 depth;
+        u32 layer;
+        Origin origin;
+    };
+    static_assert(sizeof(Parameters) == 24);
+
+    struct Semaphore {
+        PackedGPUVAddr address;
+        u32 payload;
+    };
+    static_assert(sizeof(Semaphore) == 12);
+
+    struct RenderEnable {
+        enum class Mode : u32 {
+            FALSE = 0,
+            TRUE = 1,
+            CONDITIONAL = 2,
+            RENDER_IF_EQUAL = 3,
+            RENDER_IF_NOT_EQUAL = 4,
+        };
 
-    struct Regs {
-        static constexpr std::size_t NUM_REGS = 0x1D6;
+        PackedGPUVAddr address;
+        BitField<0, 3, Mode> mode;
+    };
+    static_assert(sizeof(RenderEnable) == 12);
+
+    enum class PhysModeTarget : u32 {
+        LOCAL_FB = 0,
+        COHERENT_SYSMEM = 1,
+        NONCOHERENT_SYSMEM = 2,
+    };
+    using PhysMode = BitField<0, 2, PhysModeTarget>;
+
+    union LaunchDMA {
+        enum class DataTransferType : u32 {
+            NONE = 0,
+            PIPELINED = 1,
+            NON_PIPELINED = 2,
+        };
 
-        struct Parameters {
-            union {
-                BitField<0, 4, u32> block_depth;
-                BitField<4, 4, u32> block_height;
-                BitField<8, 4, u32> block_width;
-            };
-            u32 size_x;
-            u32 size_y;
-            u32 size_z;
-            u32 pos_z;
-            union {
-                BitField<0, 16, u32> pos_x;
-                BitField<16, 16, u32> pos_y;
-            };
+        enum class SemaphoreType : u32 {
+            NONE = 0,
+            RELEASE_ONE_WORD_SEMAPHORE = 1,
+            RELEASE_FOUR_WORD_SEMAPHORE = 2,
+        };
 
-            u32 BlockHeight() const {
-                return block_height.Value();
-            }
+        enum class InterruptType : u32 {
+            NONE = 0,
+            BLOCKING = 1,
+            NON_BLOCKING = 2,
+        };
 
-            u32 BlockDepth() const {
-                return block_depth.Value();
-            }
+        enum class MemoryLayout : u32 {
+            BLOCKLINEAR = 0,
+            PITCH = 1,
         };
 
-        static_assert(sizeof(Parameters) == 24, "Parameters has wrong size");
+        enum class Type : u32 {
+            VIRTUAL = 0,
+            PHYSICAL = 1,
+        };
 
-        enum class ComponentMode : u32 {
-            Src0 = 0,
-            Src1 = 1,
-            Src2 = 2,
-            Src3 = 3,
-            Const0 = 4,
-            Const1 = 5,
-            Zero = 6,
+        enum class SemaphoreReduction : u32 {
+            IMIN = 0,
+            IMAX = 1,
+            IXOR = 2,
+            IAND = 3,
+            IOR = 4,
+            IADD = 5,
+            INC = 6,
+            DEC = 7,
+            FADD = 0xA,
         };
 
-        enum class CopyMode : u32 {
-            None = 0,
-            Unk1 = 1,
-            Unk2 = 2,
+        enum class SemaphoreReductionSign : u32 {
+            SIGNED = 0,
+            UNSIGNED = 1,
         };
 
-        enum class QueryMode : u32 {
-            None = 0,
-            Short = 1,
-            Long = 2,
+        enum class BypassL2 : u32 {
+            USE_PTE_SETTING = 0,
+            FORCE_VOLATILE = 1,
         };
 
-        enum class QueryIntr : u32 {
-            None = 0,
-            Block = 1,
-            NonBlock = 2,
+        BitField<0, 2, DataTransferType> data_transfer_type;
+        BitField<2, 1, u32> flush_enable;
+        BitField<3, 2, SemaphoreType> semaphore_type;
+        BitField<5, 2, InterruptType> interrupt_type;
+        BitField<7, 1, MemoryLayout> src_memory_layout;
+        BitField<8, 1, MemoryLayout> dst_memory_layout;
+        BitField<9, 1, u32> multi_line_enable;
+        BitField<10, 1, u32> remap_enable;
+        BitField<11, 1, u32> rmwdisable;
+        BitField<12, 1, Type> src_type;
+        BitField<13, 1, Type> dst_type;
+        BitField<14, 4, SemaphoreReduction> semaphore_reduction;
+        BitField<18, 1, SemaphoreReductionSign> semaphore_reduction_sign;
+        BitField<19, 1, u32> reduction_enable;
+        BitField<20, 1, BypassL2> bypass_l2;
+    };
+    static_assert(sizeof(LaunchDMA) == 4);
+
+    struct RemapConst {
+        enum Swizzle : u32 {
+            SRC_X = 0,
+            SRC_Y = 1,
+            SRC_Z = 2,
+            SRC_W = 3,
+            CONST_A = 4,
+            CONST_B = 5,
+            NO_WRITE = 6,
         };
 
-        union {
-            struct {
-                INSERT_UNION_PADDING_WORDS(0xC0);
-
-                struct {
-                    union {
-                        BitField<0, 2, CopyMode> copy_mode;
-                        BitField<2, 1, u32> flush;
-
-                        BitField<3, 2, QueryMode> query_mode;
-                        BitField<5, 2, QueryIntr> query_intr;
-
-                        BitField<7, 1, u32> is_src_linear;
-                        BitField<8, 1, u32> is_dst_linear;
-
-                        BitField<9, 1, u32> enable_2d;
-                        BitField<10, 1, u32> enable_swizzle;
-                    };
-                } exec;
-
-                INSERT_UNION_PADDING_WORDS(0x3F);
-
-                struct {
-                    u32 address_high;
-                    u32 address_low;
-
-                    GPUVAddr Address() const {
-                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
-                                                     address_low);
-                    }
-                } src_address;
-
-                struct {
-                    u32 address_high;
-                    u32 address_low;
-
-                    GPUVAddr Address() const {
-                        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
-                                                     address_low);
-                    }
-                } dst_address;
-
-                u32 src_pitch;
-                u32 dst_pitch;
-                u32 x_count;
-                u32 y_count;
-
-                INSERT_UNION_PADDING_WORDS(0xB8);
-
-                u32 const0;
-                u32 const1;
-                union {
-                    BitField<0, 4, ComponentMode> component0;
-                    BitField<4, 4, ComponentMode> component1;
-                    BitField<8, 4, ComponentMode> component2;
-                    BitField<12, 4, ComponentMode> component3;
-                    BitField<16, 2, u32> component_size;
-                    BitField<20, 3, u32> src_num_components;
-                    BitField<24, 3, u32> dst_num_components;
-
-                    u32 SrcBytePerPixel() const {
-                        return src_num_components.Value() * component_size.Value();
-                    }
-                    u32 DstBytePerPixel() const {
-                        return dst_num_components.Value() * component_size.Value();
-                    }
-                } swizzle_config;
+        PackedGPUVAddr address;
 
-                Parameters dst_params;
+        union {
+            BitField<0, 3, Swizzle> dst_x;
+            BitField<4, 3, Swizzle> dst_y;
+            BitField<8, 3, Swizzle> dst_z;
+            BitField<12, 3, Swizzle> dst_w;
+            BitField<16, 2, u32> component_size_minus_one;
+            BitField<20, 2, u32> num_src_components_minus_one;
+            BitField<24, 2, u32> num_dst_components_minus_one;
+        };
+    };
+    static_assert(sizeof(RemapConst) == 12);
 
-                INSERT_UNION_PADDING_WORDS(1);
+    explicit MaxwellDMA(Core::System& system, MemoryManager& memory_manager);
+    ~MaxwellDMA() = default;
 
-                Parameters src_params;
+    /// Write the value to the register identified by method.
+    void CallMethod(u32 method, u32 method_argument, bool is_last_call) override;
 
-                INSERT_UNION_PADDING_WORDS(0x13);
-            };
-            std::array<u32, NUM_REGS> reg_array;
-        };
-    } regs{};
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                         u32 methods_pending) override;
 
 private:
+    /// Performs the copy from the source buffer to the destination buffer as configured in the
+    /// registers.
+    void Launch();
+
+    void CopyPitchToPitch();
+
+    void CopyBlockLinearToPitch();
+
+    void CopyPitchToBlockLinear();
+
+    void FastCopyBlockLinearToPitch();
+
     Core::System& system;
 
     MemoryManager& memory_manager;
@@ -185,28 +215,58 @@ private:
     std::vector<u8> read_buffer;
     std::vector<u8> write_buffer;
 
-    /// Performs the copy from the source buffer to the destination buffer as configured in the
-    /// registers.
-    void HandleCopy();
-};
+    static constexpr std::size_t NUM_REGS = 0x800;
+    struct Regs {
+        union {
+            struct {
+                u32 reserved[0x40];
+                u32 nop;
+                u32 reserved01[0xf];
+                u32 pm_trigger;
+                u32 reserved02[0x3f];
+                Semaphore semaphore;
+                u32 reserved03[0x2];
+                RenderEnable render_enable;
+                PhysMode src_phys_mode;
+                PhysMode dst_phys_mode;
+                u32 reserved04[0x26];
+                LaunchDMA launch_dma;
+                u32 reserved05[0x3f];
+                PackedGPUVAddr offset_in;
+                PackedGPUVAddr offset_out;
+                u32 pitch_in;
+                u32 pitch_out;
+                u32 line_length_in;
+                u32 line_count;
+                u32 reserved06[0xb8];
+                RemapConst remap_const;
+                Parameters dst_params;
+                u32 reserved07[0x1];
+                Parameters src_params;
+                u32 reserved08[0x275];
+                u32 pm_trigger_end;
+                u32 reserved09[0x3ba];
+            };
+            std::array<u32, NUM_REGS> reg_array;
+        };
+    } regs{};
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
     static_assert(offsetof(MaxwellDMA::Regs, field_name) == position * 4,                          \
                   "Field " #field_name " has invalid position")
 
-ASSERT_REG_POSITION(exec, 0xC0);
-ASSERT_REG_POSITION(src_address, 0x100);
-ASSERT_REG_POSITION(dst_address, 0x102);
-ASSERT_REG_POSITION(src_pitch, 0x104);
-ASSERT_REG_POSITION(dst_pitch, 0x105);
-ASSERT_REG_POSITION(x_count, 0x106);
-ASSERT_REG_POSITION(y_count, 0x107);
-ASSERT_REG_POSITION(const0, 0x1C0);
-ASSERT_REG_POSITION(const1, 0x1C1);
-ASSERT_REG_POSITION(swizzle_config, 0x1C2);
-ASSERT_REG_POSITION(dst_params, 0x1C3);
-ASSERT_REG_POSITION(src_params, 0x1CA);
+    ASSERT_REG_POSITION(launch_dma, 0xC0);
+    ASSERT_REG_POSITION(offset_in, 0x100);
+    ASSERT_REG_POSITION(offset_out, 0x102);
+    ASSERT_REG_POSITION(pitch_in, 0x104);
+    ASSERT_REG_POSITION(pitch_out, 0x105);
+    ASSERT_REG_POSITION(line_length_in, 0x106);
+    ASSERT_REG_POSITION(line_count, 0x107);
+    ASSERT_REG_POSITION(remap_const, 0x1C0);
+    ASSERT_REG_POSITION(dst_params, 0x1C3);
+    ASSERT_REG_POSITION(src_params, 0x1CA);
 
 #undef ASSERT_REG_POSITION
+};
 
 } // namespace Tegra::Engines