10 files changed, 203 insertions, 52 deletions

diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index d7c0080fa..2afa0916d 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -659,6 +659,10 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr,
     }
 }
 
+void CopyBlock(VAddr dest_addr, VAddr src_addr, size_t size) {
+    CopyBlock(*Core::CurrentProcess(), dest_addr, src_addr, size);
+}
+
 boost::optional<PAddr> TryVirtualToPhysicalAddress(const VAddr addr) {
     if (addr == 0) {
         return 0;
diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp
index f6a88f031..2eaece298 100644
--- a/src/video_core/command_processor.cpp
+++ b/src/video_core/command_processor.cpp
@@ -24,10 +24,7 @@ namespace Tegra {
 
 enum class BufferMethods {
     BindObject = 0,
-    SetGraphMacroCode = 0x45,
-    SetGraphMacroCodeArg = 0x46,
-    SetGraphMacroEntry = 0x47,
-    CountBufferMethods = 0x100,
+    CountBufferMethods = 0x40,
 };
 
 void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params) {
@@ -36,28 +33,6 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params)
                   "{:08X} remaining params {}",
                   method, subchannel, value, remaining_params);
 
-    if (method == static_cast<u32>(BufferMethods::SetGraphMacroEntry)) {
-        // Prepare to upload a new macro, reset the upload counter.
-        NGLOG_DEBUG(HW_GPU, "Uploading GPU macro {:08X}", value);
-        current_macro_entry = value;
-        current_macro_code.clear();
-        return;
-    }
-
-    if (method == static_cast<u32>(BufferMethods::SetGraphMacroCodeArg)) {
-        // Append a new code word to the current macro.
-        current_macro_code.push_back(value);
-
-        // There are no more params remaining, submit the code to the 3D engine.
-        if (remaining_params == 0) {
-            maxwell_3d->SubmitMacroCode(current_macro_entry, std::move(current_macro_code));
-            current_macro_entry = InvalidGraphMacroEntry;
-            current_macro_code.clear();
-        }
-
-        return;
-    }
-
     if (method == static_cast<u32>(BufferMethods::BindObject)) {
         // Bind the current subchannel to the desired engine id.
         NGLOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", subchannel, value);
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index 7aab163dc..9019f2504 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -2,12 +2,71 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "core/memory.h"
 #include "video_core/engines/fermi_2d.h"
+#include "video_core/textures/decoders.h"
 
 namespace Tegra {
 namespace Engines {
 
-void Fermi2D::WriteReg(u32 method, u32 value) {}
+Fermi2D::Fermi2D(MemoryManager& memory_manager) : memory_manager(memory_manager) {}
+
+void Fermi2D::WriteReg(u32 method, u32 value) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
+               "Invalid Fermi2D register, increase the size of the Regs structure");
+
+    regs.reg_array[method] = value;
+
+    switch (method) {
+    case FERMI2D_REG_INDEX(trigger): {
+        HandleSurfaceCopy();
+        break;
+    }
+    }
+}
+
+void Fermi2D::HandleSurfaceCopy() {
+    NGLOG_WARNING(HW_GPU, "Requested a surface copy with operation {}",
+                  static_cast<u32>(regs.operation));
+
+    const GPUVAddr source = regs.src.Address();
+    const GPUVAddr dest = regs.dst.Address();
+
+    // TODO(Subv): Only same-format and same-size copies are allowed for now.
+    ASSERT(regs.src.format == regs.dst.format);
+    ASSERT(regs.src.width * regs.src.height == regs.dst.width * regs.dst.height);
+
+    // TODO(Subv): Only raw copies are implemented.
+    ASSERT(regs.operation == Regs::Operation::SrcCopy);
+
+    const VAddr source_cpu = *memory_manager.GpuToCpuAddress(source);
+    const VAddr dest_cpu = *memory_manager.GpuToCpuAddress(dest);
+
+    u32 src_bytes_per_pixel = RenderTargetBytesPerPixel(regs.src.format);
+    u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format);
+
+    if (regs.src.linear == regs.dst.linear) {
+        // If the input layout and the output layout are the same, just perform a raw copy.
+        Memory::CopyBlock(dest_cpu, source_cpu,
+                          src_bytes_per_pixel * regs.dst.width * regs.dst.height);
+        return;
+    }
+
+    u8* src_buffer = Memory::GetPointer(source_cpu);
+    u8* dst_buffer = Memory::GetPointer(dest_cpu);
+
+    if (!regs.src.linear && regs.dst.linear) {
+        // If the input is tiled and the output is linear, deswizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
+                                  dst_bytes_per_pixel, src_buffer, dst_buffer, true,
+                                  regs.src.block_height);
+    } else {
+        // If the input is linear and the output is tiled, swizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
+                                  dst_bytes_per_pixel, dst_buffer, src_buffer, false,
+                                  regs.dst.block_height);
+    }
+}
 
 } // namespace Engines
 } // namespace Tegra
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index 8967ddede..0c5b413cc 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -4,19 +4,106 @@
 
 #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 {
 
+#define FERMI2D_REG_INDEX(field_name)                                                              \
+    (offsetof(Tegra::Engines::Fermi2D::Regs, field_name) / sizeof(u32))
+
 class Fermi2D final {
 public:
-    Fermi2D() = default;
+    explicit Fermi2D(MemoryManager& memory_manager);
     ~Fermi2D() = default;
 
     /// Write the value to the register identified by method.
     void WriteReg(u32 method, u32 value);
+
+    struct Regs {
+        static constexpr size_t NUM_REGS = 0x258;
+
+        struct Surface {
+            RenderTargetFormat format;
+            BitField<0, 1, u32> linear;
+            union {
+                BitField<0, 4, u32> block_depth;
+                BitField<4, 4, u32> block_height;
+                BitField<8, 4, u32> block_width;
+            };
+            u32 depth;
+            u32 layer;
+            u32 pitch;
+            u32 width;
+            u32 height;
+            u32 address_high;
+            u32 address_low;
+
+            GPUVAddr Address() const {
+                return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                             address_low);
+            }
+        };
+        static_assert(sizeof(Surface) == 0x28, "Surface has incorrect size");
+
+        enum class Operation : u32 {
+            SrcCopyAnd = 0,
+            ROPAnd = 1,
+            Blend = 2,
+            SrcCopy = 3,
+            ROP = 4,
+            SrcCopyPremult = 5,
+            BlendPremult = 6,
+        };
+
+        union {
+            struct {
+                INSERT_PADDING_WORDS(0x80);
+
+                Surface dst;
+
+                INSERT_PADDING_WORDS(2);
+
+                Surface src;
+
+                INSERT_PADDING_WORDS(0x15);
+
+                Operation operation;
+
+                INSERT_PADDING_WORDS(0x9);
+
+                // TODO(Subv): This is only a guess.
+                u32 trigger;
+
+                INSERT_PADDING_WORDS(0x1A3);
+            };
+            std::array<u32, NUM_REGS> reg_array;
+        };
+    } regs{};
+
+    MemoryManager& memory_manager;
+
+private:
+    /// Performs the copy from the source surface to the destination surface as configured in the
+    /// registers.
+    void HandleSurfaceCopy();
 };
 
+#define ASSERT_REG_POSITION(field_name, position)                                                  \
+    static_assert(offsetof(Fermi2D::Regs, field_name) == position * 4,                             \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_REG_POSITION(dst, 0x80);
+ASSERT_REG_POSITION(src, 0x8C);
+ASSERT_REG_POSITION(operation, 0xAB);
+ASSERT_REG_POSITION(trigger, 0xB5);
+#undef ASSERT_REG_POSITION
+
 } // namespace Engines
 } // namespace Tegra
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 2acbb9cd6..4306b894f 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -22,10 +22,6 @@ constexpr u32 MacroRegistersStart = 0xE00;
 Maxwell3D::Maxwell3D(MemoryManager& memory_manager)
     : memory_manager(memory_manager), macro_interpreter(*this) {}
 
-void Maxwell3D::SubmitMacroCode(u32 entry, std::vector<u32> code) {
-    uploaded_macros[entry * 2 + MacroRegistersStart] = std::move(code);
-}
-
 void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
     auto macro_code = uploaded_macros.find(method);
     // The requested macro must have been uploaded already.
@@ -37,9 +33,6 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
 }
 
 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
@@ -68,6 +61,9 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
         return;
     }
 
+    ASSERT_MSG(method < Regs::NUM_REGS,
+               "Invalid Maxwell3D register, increase the size of the Regs structure");
+
     if (debug_context) {
         debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr);
     }
@@ -75,6 +71,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     regs.reg_array[method] = value;
 
     switch (method) {
+    case MAXWELL3D_REG_INDEX(macros.data): {
+        ProcessMacroUpload(value);
+        break;
+    }
     case MAXWELL3D_REG_INDEX(code_address.code_address_high):
     case MAXWELL3D_REG_INDEX(code_address.code_address_low): {
         // Note: For some reason games (like Puyo Puyo Tetris) seem to write 0 to the CODE_ADDRESS
@@ -141,6 +141,12 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     }
 }
 
+void Maxwell3D::ProcessMacroUpload(u32 data) {
+    // Store the uploaded macro code to interpret them when they're called.
+    auto& macro = uploaded_macros[regs.macros.entry * 2 + MacroRegistersStart];
+    macro.push_back(data);
+}
+
 void Maxwell3D::ProcessQueryGet() {
     GPUVAddr sequence_address = regs.query.QueryAddress();
     // Since the sequence address is given as a GPU VAddr, we have to convert it to an application
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index a022665eb..5cf62fb01 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -31,7 +31,7 @@ public:
     /// Register structure of the Maxwell3D engine.
     /// TODO(Subv): This structure will need to be made bigger as more registers are discovered.
     struct Regs {
-        static constexpr size_t NUM_REGS = 0xE36;
+        static constexpr size_t NUM_REGS = 0xE00;
 
         static constexpr size_t NumRenderTargets = 8;
         static constexpr size_t NumViewports = 16;
@@ -322,7 +322,15 @@ public:
 
         union {
             struct {
-                INSERT_PADDING_WORDS(0x200);
+                INSERT_PADDING_WORDS(0x45);
+
+                struct {
+                    INSERT_PADDING_WORDS(1);
+                    u32 data;
+                    u32 entry;
+                } macros;
+
+                INSERT_PADDING_WORDS(0x1B8);
 
                 struct {
                     u32 address_high;
@@ -605,7 +613,7 @@ public:
                     u32 size[MaxShaderStage];
                 } tex_info_buffers;
 
-                INSERT_PADDING_WORDS(0x102);
+                INSERT_PADDING_WORDS(0xCC);
             };
             std::array<u32, NUM_REGS> reg_array;
         };
@@ -637,9 +645,6 @@ public:
     /// 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::FullTextureInfo> GetStageTextures(Regs::ShaderStage stage) const;
 
@@ -670,6 +675,9 @@ private:
      */
     void CallMacroMethod(u32 method, std::vector<u32> parameters);
 
+    /// Handles writes to the macro uploading registers.
+    void ProcessMacroUpload(u32 data);
+
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
 
@@ -687,6 +695,7 @@ private:
     static_assert(offsetof(Maxwell3D::Regs, field_name) == position * 4,                           \
                   "Field " #field_name " has invalid position")
 
+ASSERT_REG_POSITION(macros, 0x45);
 ASSERT_REG_POSITION(rt, 0x200);
 ASSERT_REG_POSITION(viewport_transform[0], 0x280);
 ASSERT_REG_POSITION(viewport, 0x300);
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 9463cd5d6..9eb143918 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -12,7 +12,7 @@ namespace Tegra {
 GPU::GPU() {
     memory_manager = std::make_unique<MemoryManager>();
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(*memory_manager);
-    fermi_2d = std::make_unique<Engines::Fermi2D>();
+    fermi_2d = std::make_unique<Engines::Fermi2D>(*memory_manager);
     maxwell_compute = std::make_unique<Engines::MaxwellCompute>();
 }
 
@@ -22,4 +22,16 @@ const Tegra::Engines::Maxwell3D& GPU::Get3DEngine() const {
     return *maxwell_3d;
 }
 
+u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
+    ASSERT(format != RenderTargetFormat::NONE);
+
+    switch (format) {
+    case RenderTargetFormat::RGBA8_UNORM:
+    case RenderTargetFormat::RGB10_A2_UNORM:
+        return 4;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented render target format %u", static_cast<u32>(format));
+    }
+}
+
 } // namespace Tegra
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 2888daedc..f168a5171 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -21,6 +21,9 @@ enum class RenderTargetFormat : u32 {
     RGBA8_SRGB = 0xD6,
 };
 
+/// Returns the number of bytes per pixel of each rendertarget format.
+u32 RenderTargetBytesPerPixel(RenderTargetFormat format);
+
 class DebugContext;
 
 /**
@@ -86,8 +89,6 @@ public:
     }
 
 private:
-    static constexpr u32 InvalidGraphMacroEntry = 0xFFFFFFFF;
-
     /// Writes a single register in the engine bound to the specified subchannel
     void WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params);
 
@@ -100,11 +101,6 @@ private:
     std::unique_ptr<Engines::Fermi2D> fermi_2d;
     /// Compute engine
     std::unique_ptr<Engines::MaxwellCompute> maxwell_compute;
-
-    /// Entry of the macro that is currently being uploaded
-    u32 current_macro_entry = InvalidGraphMacroEntry;
-    /// Code being uploaded for the current macro
-    std::vector<u32> current_macro_code;
 };
 
 } // namespace Tegra
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 9c3ae875c..8b39b2bdf 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -27,9 +27,8 @@ static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel,
     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) {
+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) {
diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h
index a700911cf..2562c4b06 100644
--- a/src/video_core/textures/decoders.h
+++ b/src/video_core/textures/decoders.h
@@ -17,6 +17,10 @@ namespace Texture {
 std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height,
                                  u32 block_height = TICEntry::DefaultBlockHeight);
 
+/// Copies texture data from a buffer and performs swizzling/unswizzling as necessary.
+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);
+
 /**
  * Decodes an unswizzled texture into a A8R8G8B8 texture.
  */