From 987a17066514dbab8e02acae20d4ea32c4f502eb Mon Sep 17 00:00:00 2001
From: Subv <subv2112@gmail.com>
Date: Sun, 10 Jun 2018 17:02:33 -0500
Subject: [PATCH] GPU: Partially implemented the Maxwell DMA engine.

Only tiled->linear and linear->tiled copies that aren't offsetted are supported for now. Queries are not supported. Swizzled copies are not supported.
---
 src/video_core/CMakeLists.txt          |   2 +
 src/video_core/command_processor.cpp   |   6 +-
 src/video_core/engines/fermi_2d.cpp    |   1 +
 src/video_core/engines/maxwell_dma.cpp |  69 +++++++++++
 src/video_core/engines/maxwell_dma.h   | 155 +++++++++++++++++++++++++
 src/video_core/gpu.cpp                 |   2 +
 src/video_core/gpu.h                   |   3 +
 7 files changed, 237 insertions(+), 1 deletion(-)
 create mode 100644 src/video_core/engines/maxwell_dma.cpp
 create mode 100644 src/video_core/engines/maxwell_dma.h

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 281810357..6e193e7e1 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -9,6 +9,8 @@ add_library(video_core STATIC
     engines/maxwell_3d.h
     engines/maxwell_compute.cpp
     engines/maxwell_compute.h
+    engines/maxwell_dma.cpp
+    engines/maxwell_dma.h
     engines/shader_bytecode.h
     gpu.cpp
     gpu.h
diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp
index d72d6f760..cec9cb9f3 100644
--- a/src/video_core/command_processor.cpp
+++ b/src/video_core/command_processor.cpp
@@ -16,6 +16,7 @@
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_compute.h"
+#include "video_core/engines/maxwell_dma.h"
 #include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
 #include "video_core/video_core.h"
@@ -60,8 +61,11 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params)
     case EngineID::MAXWELL_COMPUTE_B:
         maxwell_compute->WriteReg(method, value);
         break;
+    case EngineID::MAXWELL_DMA_COPY_A:
+        maxwell_dma->WriteReg(method, value);
+        break;
     default:
-        UNIMPLEMENTED();
+        UNIMPLEMENTED_MSG("Unimplemented engine");
     }
 }
 
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 6b9382f06..998b7c843 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -47,6 +47,7 @@ void Fermi2D::HandleSurfaceCopy() {
 
     if (regs.src.linear == regs.dst.linear) {
         // If the input layout and the output layout are the same, just perform a raw copy.
+        ASSERT(regs.src.BlockHeight() == regs.dst.BlockHeight());
         Memory::CopyBlock(dest_cpu, source_cpu,
                           src_bytes_per_pixel * regs.dst.width * regs.dst.height);
         return;
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
new file mode 100644
index 000000000..442138988
--- /dev/null
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -0,0 +1,69 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/memory.h"
+#include "video_core/engines/maxwell_dma.h"
+#include "video_core/textures/decoders.h"
+
+namespace Tegra {
+namespace Engines {
+
+MaxwellDMA::MaxwellDMA(MemoryManager& memory_manager) : memory_manager(memory_manager) {}
+
+void MaxwellDMA::WriteReg(u32 method, u32 value) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
+               "Invalid MaxwellDMA register, increase the size of the Regs structure");
+
+    regs.reg_array[method] = value;
+
+#define MAXWELLDMA_REG_INDEX(field_name)                                                           \
+    (offsetof(Tegra::Engines::MaxwellDMA::Regs, field_name) / sizeof(u32))
+
+    switch (method) {
+    case MAXWELLDMA_REG_INDEX(exec): {
+        HandleCopy();
+        break;
+    }
+    }
+
+#undef MAXWELLDMA_REG_INDEX
+}
+
+void MaxwellDMA::HandleCopy() {
+    NGLOG_WARNING(HW_GPU, "Requested a DMA copy");
+
+    const GPUVAddr source = regs.src_address.Address();
+    const GPUVAddr dest = regs.dst_address.Address();
+
+    const VAddr source_cpu = *memory_manager.GpuToCpuAddress(source);
+    const VAddr dest_cpu = *memory_manager.GpuToCpuAddress(dest);
+
+    // TODO(Subv): Perform more research and implement all features of this engine.
+    ASSERT(regs.exec.enable_swizzle == 0);
+    ASSERT(regs.exec.enable_2d == 1);
+    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.src_params.pos_x == 0);
+    ASSERT(regs.src_params.pos_y == 0);
+    ASSERT(regs.dst_params.pos_x == 0);
+    ASSERT(regs.dst_params.pos_y == 0);
+    ASSERT(regs.exec.is_dst_linear != regs.exec.is_src_linear);
+
+    u8* src_buffer = Memory::GetPointer(source_cpu);
+    u8* dst_buffer = Memory::GetPointer(dest_cpu);
+
+    if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
+        // If the input is tiled and the output is linear, deswizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, 1, 1, src_buffer,
+                                  dst_buffer, true, regs.src_params.BlockHeight());
+    } else {
+        // If the input is linear and the output is tiled, swizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, 1, 1, dst_buffer,
+                                  src_buffer, false, regs.dst_params.BlockHeight());
+    }
+}
+
+} // namespace Engines
+} // namespace Tegra
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
new file mode 100644
index 000000000..905749bde
--- /dev/null
+++ b/src/video_core/engines/maxwell_dma.h
@@ -0,0 +1,155 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include "common/assert.h"
+#include "common/bit_field.h"
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
+
+namespace Tegra {
+namespace Engines {
+
+class MaxwellDMA final {
+public:
+    explicit MaxwellDMA(MemoryManager& memory_manager);
+    ~MaxwellDMA() = default;
+
+    /// Write the value to the register identified by method.
+    void WriteReg(u32 method, u32 value);
+
+    struct Regs {
+        static constexpr size_t NUM_REGS = 0x1D6;
+
+        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;
+            };
+
+            u32 BlockHeight() const {
+                return 1 << block_height;
+            }
+        };
+
+        static_assert(sizeof(Parameters) == 24, "Parameters has wrong size");
+
+        enum class CopyMode : u32 {
+            None = 0,
+            Unk1 = 1,
+            Unk2 = 2,
+        };
+
+        enum class QueryMode : u32 {
+            None = 0,
+            Short = 1,
+            Long = 2,
+        };
+
+        enum class QueryIntr : u32 {
+            None = 0,
+            Block = 1,
+            NonBlock = 2,
+        };
+
+        union {
+            struct {
+                INSERT_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_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_PADDING_WORDS(0xBB);
+
+                Parameters dst_params;
+
+                INSERT_PADDING_WORDS(1);
+
+                Parameters src_params;
+
+                INSERT_PADDING_WORDS(0x13);
+            };
+            std::array<u32, NUM_REGS> reg_array;
+        };
+    } regs{};
+
+    MemoryManager& memory_manager;
+
+private:
+    /// Performs the copy from the source buffer to the destination buffer as configured in the
+    /// registers.
+    void HandleCopy();
+};
+
+#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(dst_params, 0x1C3);
+ASSERT_REG_POSITION(src_params, 0x1CA);
+
+#undef ASSERT_REG_POSITION
+
+} // namespace Engines
+} // namespace Tegra
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 66351fe6e..e36483145 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -5,6 +5,7 @@
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_compute.h"
+#include "video_core/engines/maxwell_dma.h"
 #include "video_core/gpu.h"
 
 namespace Tegra {
@@ -14,6 +15,7 @@ GPU::GPU() {
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(*memory_manager);
     fermi_2d = std::make_unique<Engines::Fermi2D>(*memory_manager);
     maxwell_compute = std::make_unique<Engines::MaxwellCompute>();
+    maxwell_dma = std::make_unique<Engines::MaxwellDMA>(*memory_manager);
 }
 
 GPU::~GPU() = default;
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 5852b9619..7b4e9b842 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -63,6 +63,7 @@ namespace Engines {
 class Fermi2D;
 class Maxwell3D;
 class MaxwellCompute;
+class MaxwellDMA;
 } // namespace Engines
 
 enum class EngineID {
@@ -103,6 +104,8 @@ private:
     std::unique_ptr<Engines::Fermi2D> fermi_2d;
     /// Compute engine
     std::unique_ptr<Engines::MaxwellCompute> maxwell_compute;
+    /// DMA engine
+    std::unique_ptr<Engines::MaxwellDMA> maxwell_dma;
 };
 
 } // namespace Tegra