10 files changed, 516 insertions, 6 deletions

diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index e56253c4c..3dab81769 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -1,6 +1,8 @@
 add_library(video_core STATIC
     command_processor.cpp
     command_processor.h
+    debug_utils/debug_utils.cpp
+    debug_utils/debug_utils.h
     engines/fermi_2d.cpp
     engines/fermi_2d.h
     engines/maxwell_3d.cpp
@@ -31,6 +33,9 @@ add_library(video_core STATIC
     renderer_opengl/gl_stream_buffer.h
     renderer_opengl/renderer_opengl.cpp
     renderer_opengl/renderer_opengl.h
+    textures/decoders.cpp
+    textures/decoders.h
+    textures/texture.h
     utils.h
     video_core.cpp
     video_core.h
diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
new file mode 100644
index 000000000..22d44aab2
--- /dev/null
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -0,0 +1,64 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <condition_variable>
+#include <cstdint>
+#include <cstring>
+#include <fstream>
+#include <map>
+#include <mutex>
+#include <string>
+
+#include "common/assert.h"
+#include "common/bit_field.h"
+#include "common/color.h"
+#include "common/common_types.h"
+#include "common/file_util.h"
+#include "common/logging/log.h"
+#include "common/math_util.h"
+#include "common/vector_math.h"
+#include "video_core/debug_utils/debug_utils.h"
+
+namespace Tegra {
+
+void DebugContext::DoOnEvent(Event event, void* data) {
+    {
+        std::unique_lock<std::mutex> lock(breakpoint_mutex);
+
+        // TODO(Subv): Commit the rasterizer's caches so framebuffers, render targets, etc. will
+        // show on debug widgets
+
+        // TODO: Should stop the CPU thread here once we multithread emulation.
+
+        active_breakpoint = event;
+        at_breakpoint = true;
+
+        // Tell all observers that we hit a breakpoint
+        for (auto& breakpoint_observer : breakpoint_observers) {
+            breakpoint_observer->OnMaxwellBreakPointHit(event, data);
+        }
+
+        // Wait until another thread tells us to Resume()
+        resume_from_breakpoint.wait(lock, [&] { return !at_breakpoint; });
+    }
+}
+
+void DebugContext::Resume() {
+    {
+        std::lock_guard<std::mutex> lock(breakpoint_mutex);
+
+        // Tell all observers that we are about to resume
+        for (auto& breakpoint_observer : breakpoint_observers) {
+            breakpoint_observer->OnMaxwellResume();
+        }
+
+        // Resume the waiting thread (i.e. OnEvent())
+        at_breakpoint = false;
+    }
+
+    resume_from_breakpoint.notify_one();
+}
+
+} // namespace Tegra
diff --git a/src/video_core/debug_utils/debug_utils.h b/src/video_core/debug_utils/debug_utils.h
new file mode 100644
index 000000000..bbba8e380
--- /dev/null
+++ b/src/video_core/debug_utils/debug_utils.h
@@ -0,0 +1,163 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <algorithm>
+#include <array>
+#include <condition_variable>
+#include <iterator>
+#include <list>
+#include <map>
+#include <memory>
+#include <mutex>
+#include <string>
+#include <utility>
+#include <vector>
+#include "common/common_types.h"
+#include "common/vector_math.h"
+
+namespace Tegra {
+
+class DebugContext {
+public:
+    enum class Event {
+        FirstEvent = 0,
+
+        MaxwellCommandLoaded = FirstEvent,
+        MaxwellCommandProcessed,
+        IncomingPrimitiveBatch,
+        FinishedPrimitiveBatch,
+
+        NumEvents
+    };
+
+    /**
+     * Inherit from this class to be notified of events registered to some debug context.
+     * Most importantly this is used for our debugger GUI.
+     *
+     * To implement event handling, override the OnMaxwellBreakPointHit and OnMaxwellResume methods.
+     * @warning All BreakPointObservers need to be on the same thread to guarantee thread-safe state
+     * access
+     * @todo Evaluate an alternative interface, in which there is only one managing observer and
+     * multiple child observers running (by design) on the same thread.
+     */
+    class BreakPointObserver {
+    public:
+        /// Constructs the object such that it observes events of the given DebugContext.
+        BreakPointObserver(std::shared_ptr<DebugContext> debug_context)
+            : context_weak(debug_context) {
+            std::unique_lock<std::mutex> lock(debug_context->breakpoint_mutex);
+            debug_context->breakpoint_observers.push_back(this);
+        }
+
+        virtual ~BreakPointObserver() {
+            auto context = context_weak.lock();
+            if (context) {
+                std::unique_lock<std::mutex> lock(context->breakpoint_mutex);
+                context->breakpoint_observers.remove(this);
+
+                // If we are the last observer to be destroyed, tell the debugger context that
+                // it is free to continue. In particular, this is required for a proper yuzu
+                // shutdown, when the emulation thread is waiting at a breakpoint.
+                if (context->breakpoint_observers.empty())
+                    context->Resume();
+            }
+        }
+
+        /**
+         * Action to perform when a breakpoint was reached.
+         * @param event Type of event which triggered the breakpoint
+         * @param data Optional data pointer (if unused, this is a nullptr)
+         * @note This function will perform nothing unless it is overridden in the child class.
+         */
+        virtual void OnMaxwellBreakPointHit(Event event, void* data) {}
+
+        /**
+         * Action to perform when emulation is resumed from a breakpoint.
+         * @note This function will perform nothing unless it is overridden in the child class.
+         */
+        virtual void OnMaxwellResume() {}
+
+    protected:
+        /**
+         * Weak context pointer. This need not be valid, so when requesting a shared_ptr via
+         * context_weak.lock(), always compare the result against nullptr.
+         */
+        std::weak_ptr<DebugContext> context_weak;
+    };
+
+    /**
+     * Simple structure defining a breakpoint state
+     */
+    struct BreakPoint {
+        bool enabled = false;
+    };
+
+    /**
+     * Static constructor used to create a shared_ptr of a DebugContext.
+     */
+    static std::shared_ptr<DebugContext> Construct() {
+        return std::shared_ptr<DebugContext>(new DebugContext);
+    }
+
+    /**
+     * Used by the emulation core when a given event has happened. If a breakpoint has been set
+     * for this event, OnEvent calls the event handlers of the registered breakpoint observers.
+     * The current thread then is halted until Resume() is called from another thread (or until
+     * emulation is stopped).
+     * @param event Event which has happened
+     * @param data Optional data pointer (pass nullptr if unused). Needs to remain valid until
+     * Resume() is called.
+     */
+    void OnEvent(Event event, void* data) {
+        // This check is left in the header to allow the compiler to inline it.
+        if (!breakpoints[(int)event].enabled)
+            return;
+        // For the rest of event handling, call a separate function.
+        DoOnEvent(event, data);
+    }
+
+    void DoOnEvent(Event event, void* data);
+
+    /**
+     * Resume from the current breakpoint.
+     * @warning Calling this from the same thread that OnEvent was called in will cause a deadlock.
+     * Calling from any other thread is safe.
+     */
+    void Resume();
+
+    /**
+     * Delete all set breakpoints and resume emulation.
+     */
+    void ClearBreakpoints() {
+        for (auto& bp : breakpoints) {
+            bp.enabled = false;
+        }
+        Resume();
+    }
+
+    // TODO: Evaluate if access to these members should be hidden behind a public interface.
+    std::array<BreakPoint, (int)Event::NumEvents> breakpoints;
+    Event active_breakpoint;
+    bool at_breakpoint = false;
+
+private:
+    /**
+     * Private default constructor to make sure people always construct this through Construct()
+     * instead.
+     */
+    DebugContext() = default;
+
+    /// Mutex protecting current breakpoint state and the observer list.
+    std::mutex breakpoint_mutex;
+
+    /// Used by OnEvent to wait for resumption.
+    std::condition_variable resume_from_breakpoint;
+
+    /// List of registered observers
+    std::list<BreakPointObserver*> breakpoint_observers;
+};
+
+} // namespace Tegra
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 4d9745e48..986165c6d 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -2,8 +2,13 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <cinttypes>
 #include "common/assert.h"
+#include "core/core.h"
+#include "video_core/debug_utils/debug_utils.h"
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/textures/decoders.h"
+#include "video_core/textures/texture.h"
 
 namespace Tegra {
 namespace Engines {
@@ -46,6 +51,8 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Maxwell3D register, increase the size of the Regs structure");
 
+    auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
+
     // It is an error to write to a register other than the current macro's ARG register before it
     // has finished execution.
     if (executing_macro != 0) {
@@ -72,6 +79,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
         return;
     }
 
+    if (debug_context) {
+        debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr);
+    }
+
     regs.reg_array[method] = value;
 
 #define MAXWELL3D_REG_INDEX(field_name) (offsetof(Regs, field_name) / sizeof(u32))
@@ -137,6 +148,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     }
 
 #undef MAXWELL3D_REG_INDEX
+
+    if (debug_context) {
+        debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandProcessed, nullptr);
+    }
 }
 
 void Maxwell3D::ProcessQueryGet() {
@@ -160,6 +175,15 @@ void Maxwell3D::ProcessQueryGet() {
 
 void Maxwell3D::DrawArrays() {
     LOG_WARNING(HW_GPU, "Game requested a DrawArrays, ignoring");
+    auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
+
+    if (debug_context) {
+        debug_context->OnEvent(Tegra::DebugContext::Event::IncomingPrimitiveBatch, nullptr);
+    }
+
+    if (debug_context) {
+        debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr);
+    }
 }
 
 void Maxwell3D::BindTextureInfoBuffer(const std::vector<u32>& parameters) {
@@ -270,5 +294,50 @@ void Maxwell3D::ProcessCBData(u32 value) {
     regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
 }
 
+std::vector<Texture::TICEntry> Maxwell3D::GetStageTextures(Regs::ShaderStage stage) {
+    std::vector<Texture::TICEntry> textures;
+
+    auto& fragment_shader = state.shader_stages[static_cast<size_t>(stage)];
+    auto& tex_info_buffer = fragment_shader.const_buffers[regs.tex_cb_index];
+    ASSERT(tex_info_buffer.enabled && tex_info_buffer.address != 0);
+
+    GPUVAddr tic_base_address = regs.tic.TICAddress();
+
+    GPUVAddr tex_info_buffer_end = tex_info_buffer.address + tex_info_buffer.size;
+
+    // Offset into the texture constbuffer where the texture info begins.
+    static constexpr size_t TextureInfoOffset = 0x20;
+
+    for (GPUVAddr current_texture = tex_info_buffer.address + TextureInfoOffset;
+         current_texture < tex_info_buffer_end; current_texture += 4) {
+
+        Texture::TextureHandle tex_info{
+            Memory::Read32(memory_manager.PhysicalToVirtualAddress(current_texture))};
+
+        if (tex_info.tic_id != 0 || tex_info.tsc_id != 0) {
+            GPUVAddr tic_address_gpu =
+                tic_base_address + tex_info.tic_id * sizeof(Texture::TICEntry);
+            VAddr tic_address_cpu = memory_manager.PhysicalToVirtualAddress(tic_address_gpu);
+
+            Texture::TICEntry tic_entry;
+            Memory::ReadBlock(tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry));
+
+            auto r_type = tic_entry.r_type.Value();
+            auto g_type = tic_entry.g_type.Value();
+            auto b_type = tic_entry.b_type.Value();
+            auto a_type = tic_entry.a_type.Value();
+
+            // TODO(Subv): Different data types for separate components are not supported
+            ASSERT(r_type == g_type && r_type == b_type && r_type == a_type);
+
+            auto format = tic_entry.format.Value();
+
+            textures.push_back(tic_entry);
+        }
+    }
+
+    return textures;
+}
+
 } // namespace Engines
 } // namespace Tegra
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 545d7ff35..441cc0c19 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -12,6 +12,7 @@
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "video_core/memory_manager.h"
+#include "video_core/textures/texture.h"
 
 namespace Tegra {
 namespace Engines {
@@ -21,12 +22,6 @@ public:
     explicit Maxwell3D(MemoryManager& memory_manager);
     ~Maxwell3D() = default;
 
-    /// Write the value to the register identified by method.
-    void WriteReg(u32 method, u32 value, u32 remaining_params);
-
-    /// Uploads the code for a GPU macro program associated with the specified entry.
-    void SubmitMacroCode(u32 entry, std::vector<u32> code);
-
     /// Register structure of the Maxwell3D engine.
     /// TODO(Subv): This structure will need to be made bigger as more registers are discovered.
     struct Regs {
@@ -430,6 +425,15 @@ public:
 
     State state{};
 
+    /// Write the value to the register identified by method.
+    void WriteReg(u32 method, u32 value, u32 remaining_params);
+
+    /// Uploads the code for a GPU macro program associated with the specified entry.
+    void SubmitMacroCode(u32 entry, std::vector<u32> code);
+
+    /// Returns a list of enabled textures for the specified shader stage.
+    std::vector<Texture::TICEntry> GetStageTextures(Regs::ShaderStage stage);
+
 private:
     MemoryManager& memory_manager;
 
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index c384d236e..9463cd5d6 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -18,4 +18,8 @@ GPU::GPU() {
 
 GPU::~GPU() = default;
 
+const Tegra::Engines::Maxwell3D& GPU::Get3DEngine() const {
+    return *maxwell_3d;
+}
+
 } // namespace Tegra
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 206b3e05e..8183b12e9 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -13,6 +13,12 @@
 
 namespace Tegra {
 
+enum class RenderTargetFormat {
+    RGBA8_UNORM = 0xD5,
+};
+
+class DebugContext;
+
 /**
  * Struct describing framebuffer configuration
  */
@@ -66,6 +72,9 @@ public:
     /// Processes a command list stored at the specified address in GPU memory.
     void ProcessCommandList(GPUVAddr address, u32 size);
 
+    /// Returns a reference to the Maxwell3D GPU engine.
+    const Engines::Maxwell3D& Get3DEngine() const;
+
     std::unique_ptr<MemoryManager> memory_manager;
 
     Engines::Maxwell3D& Maxwell3D() {
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
new file mode 100644
index 000000000..2e87281eb
--- /dev/null
+++ b/src/video_core/textures/decoders.cpp
@@ -0,0 +1,105 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include "common/assert.h"
+#include "video_core/textures/decoders.h"
+#include "video_core/textures/texture.h"
+
+namespace Tegra {
+namespace Texture {
+
+/**
+ * Calculates the offset of an (x, y) position within a swizzled texture.
+ * Taken from the Tegra X1 TRM.
+ */
+static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel, u32 block_height) {
+    u32 image_width_in_gobs = image_width * bytes_per_pixel / 64;
+    u32 GOB_address = 0 + (y / (8 * block_height)) * 512 * block_height * image_width_in_gobs +
+                      (x * bytes_per_pixel / 64) * 512 * block_height +
+                      (y % (8 * block_height) / 8) * 512;
+    x *= bytes_per_pixel;
+    u32 address = GOB_address + ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
+                  (y % 2) * 16 + (x % 16);
+
+    return address;
+}
+
+static void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
+                             u8* swizzled_data, u8* unswizzled_data, bool unswizzle,
+                             u32 block_height) {
+    u8* data_ptrs[2];
+    for (unsigned y = 0; y < height; ++y) {
+        for (unsigned x = 0; x < width; ++x) {
+            u32 swizzle_offset = GetSwizzleOffset(x, y, width, bytes_per_pixel, block_height);
+            u32 pixel_index = (x + y * width) * out_bytes_per_pixel;
+
+            data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
+            data_ptrs[!unswizzle] = &unswizzled_data[pixel_index];
+
+            std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
+        }
+    }
+}
+
+u32 BytesPerPixel(TextureFormat format) {
+    switch (format) {
+    case TextureFormat::DXT1:
+        // In this case a 'pixel' actually refers to a 4x4 tile.
+        return 8;
+    case TextureFormat::A8R8G8B8:
+        return 4;
+    default:
+        UNIMPLEMENTED_MSG("Format not implemented");
+        break;
+    }
+}
+
+std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height) {
+    u8* data = Memory::GetPointer(address);
+    u32 bytes_per_pixel = BytesPerPixel(format);
+
+    static constexpr u32 DefaultBlockHeight = 16;
+
+    std::vector<u8> unswizzled_data(width * height * bytes_per_pixel);
+
+    switch (format) {
+    case TextureFormat::DXT1:
+        // In the DXT1 format, each 4x4 tile is swizzled instead of just individual pixel values.
+        CopySwizzledData(width / 4, height / 4, bytes_per_pixel, bytes_per_pixel, data,
+                         unswizzled_data.data(), true, DefaultBlockHeight);
+        break;
+    case TextureFormat::A8R8G8B8:
+        CopySwizzledData(width, height, bytes_per_pixel, bytes_per_pixel, data,
+                         unswizzled_data.data(), true, DefaultBlockHeight);
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Format not implemented");
+        break;
+    }
+
+    return unswizzled_data;
+}
+
+std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width,
+                              u32 height) {
+    std::vector<u8> rgba_data;
+
+    // TODO(Subv): Implement.
+    switch (format) {
+    case TextureFormat::DXT1:
+    case TextureFormat::A8R8G8B8:
+        // TODO(Subv): For the time being just forward the same data without any decoding.
+        rgba_data = texture_data;
+        break;
+    default:
+        UNIMPLEMENTED_MSG("Format not implemented");
+        break;
+    }
+
+    return rgba_data;
+}
+
+} // namespace Texture
+} // namespace Tegra
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
new file mode 100644
index 000000000..0c21694ff
--- /dev/null
+++ b/src/video_core/textures/decoders.h
@@ -0,0 +1,26 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/textures/texture.h"
+
+namespace Tegra {
+namespace Texture {
+
+/**
+ * Unswizzles a swizzled texture without changing its format.
+ */
+std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height);
+
+/**
+ * Decodes an unswizzled texture into a A8R8G8B8 texture.
+ */
+std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat format, u32 width,
+                              u32 height);
+
+} // namespace Texture
+} // namespace Tegra
diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h
new file mode 100644
index 000000000..d969bcdd9
--- /dev/null
+++ b/src/video_core/textures/texture.h
@@ -0,0 +1,61 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/bit_field.h"
+#include "common/common_funcs.h"
+#include "common/common_types.h"
+#include "video_core/memory_manager.h"
+
+namespace Tegra {
+namespace Texture {
+
+enum class TextureFormat : u32 {
+    A8R8G8B8 = 8,
+    DXT1 = 0x24,
+};
+
+union TextureHandle {
+    u32 raw;
+    BitField<0, 20, u32> tic_id;
+    BitField<20, 12, u32> tsc_id;
+};
+
+struct TICEntry {
+    union {
+        u32 raw;
+        BitField<0, 7, TextureFormat> format;
+        BitField<7, 3, u32> r_type;
+        BitField<10, 3, u32> g_type;
+        BitField<13, 3, u32> b_type;
+        BitField<16, 3, u32> a_type;
+    };
+    u32 address_low;
+    u16 address_high;
+    INSERT_PADDING_BYTES(6);
+    u16 width_minus_1;
+    INSERT_PADDING_BYTES(2);
+    u16 height_minus_1;
+    INSERT_PADDING_BYTES(10);
+
+    GPUVAddr Address() const {
+        return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) | address_low);
+    }
+
+    u32 Width() const {
+        return width_minus_1 + 1;
+    }
+
+    u32 Height() const {
+        return height_minus_1 + 1;
+    }
+};
+static_assert(sizeof(TICEntry) == 0x20, "TICEntry has wrong size");
+
+/// Returns the number of bytes per pixel of the input texture format.
+u32 BytesPerPixel(TextureFormat format);
+
+} // namespace Texture
+} // namespace Tegra