3 files changed, 230 insertions, 150 deletions

diff --git a/src/core/hle/service/gsp.cpp b/src/core/hle/service/gsp.cpp
index e241b31c8..08e65612e 100644
--- a/src/core/hle/service/gsp.cpp
+++ b/src/core/hle/service/gsp.cpp
@@ -173,11 +173,11 @@ void ExecuteCommand(const Command& command) {
     case CommandId::SET_COMMAND_LIST_LAST:
     {
         auto& params = command.set_command_list_last;
-        WriteGPURegister(GPU::Regs::CommandProcessor + 2, params.address >> 3);
-        WriteGPURegister(GPU::Regs::CommandProcessor, params.size >> 3);
+        WriteGPURegister(GPU_REG_INDEX(command_processor_config.address), params.address >> 3);
+        WriteGPURegister(GPU_REG_INDEX(command_processor_config.size), params.size >> 3);
 
         // TODO: Not sure if we are supposed to always write this .. seems to trigger processing though
-        WriteGPURegister(GPU::Regs::CommandProcessor + 4, 1);
+        WriteGPURegister(GPU_REG_INDEX(command_processor_config.trigger), 1);
 
         // TODO: Move this to GPU
         // TODO: Not sure what units the size is measured in
@@ -193,15 +193,15 @@ void ExecuteCommand(const Command& command) {
     case CommandId::SET_MEMORY_FILL:
     {
         auto& params = command.memory_fill;
-        WriteGPURegister(GPU::Regs::MemoryFill, params.start1 >> 3);
-        WriteGPURegister(GPU::Regs::MemoryFill + 1, params.end1 >> 3);
-        WriteGPURegister(GPU::Regs::MemoryFill + 2, params.end1 - params.start1);
-        WriteGPURegister(GPU::Regs::MemoryFill + 3, params.value1);
-
-        WriteGPURegister(GPU::Regs::MemoryFill + 4, params.start2 >> 3);
-        WriteGPURegister(GPU::Regs::MemoryFill + 5, params.end2 >> 3);
-        WriteGPURegister(GPU::Regs::MemoryFill + 6, params.end2 - params.start2);
-        WriteGPURegister(GPU::Regs::MemoryFill + 7, params.value2);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].address_start), params.start1 >> 3);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].address_end), params.end1 >> 3);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].size), params.end1 - params.start1);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[0].value), params.value1);
+
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].address_start), params.start2 >> 3);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].address_end), params.end2 >> 3);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].size), params.end2 - params.start2);
+        WriteGPURegister(GPU_REG_INDEX(memory_fill_config[1].value), params.value2);
         break;
     }
 
@@ -220,15 +220,15 @@ void ExecuteCommand(const Command& command) {
     case CommandId::SET_TEXTURE_COPY:
     {
         auto& params = command.image_copy;
-        WriteGPURegister(GPU::Regs::DisplayTransfer, params.in_buffer_address >> 3);
-        WriteGPURegister(GPU::Regs::DisplayTransfer + 1, params.out_buffer_address >> 3);
-        WriteGPURegister(GPU::Regs::DisplayTransfer + 3, params.in_buffer_size);
-        WriteGPURegister(GPU::Regs::DisplayTransfer + 2, params.out_buffer_size);
-        WriteGPURegister(GPU::Regs::DisplayTransfer + 4, params.flags);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_address), params.in_buffer_address >> 3);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_address), params.out_buffer_address >> 3);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.input_size), params.in_buffer_size);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.output_size), params.out_buffer_size);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.flags), params.flags);
 
         // TODO: Should this only be ORed with 1 for texture copies?
         // trigger transfer
-        WriteGPURegister(GPU::Regs::DisplayTransfer + 6, 1);
+        WriteGPURegister(GPU_REG_INDEX(display_transfer_config.trigger), 1);
         break;
     }
 
diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp
index d94c2329b..fd40f8ac0 100644
--- a/src/core/hw/gpu.cpp
+++ b/src/core/hw/gpu.cpp
@@ -19,7 +19,7 @@
 
 namespace GPU {
 
-RegisterSet<u32, Regs> g_regs;
+Regs g_regs;
 
 u32 g_cur_line = 0;         ///< Current vertical screen line
 u64 g_last_line_ticks = 0;  ///< CPU tick count from last vertical screen line
@@ -32,8 +32,8 @@ void SetFramebufferLocation(const FramebufferLocation mode) {
     switch (mode) {
     case FRAMEBUFFER_LOCATION_FCRAM:
     {
-        auto& framebuffer_top = g_regs.Get<Regs::FramebufferTop>();
-        auto& framebuffer_sub = g_regs.Get<Regs::FramebufferBottom>();
+        auto& framebuffer_top = g_regs.framebuffer_config[0];
+        auto& framebuffer_sub = g_regs.framebuffer_config[1];
 
         framebuffer_top.address_left1  = PADDR_TOP_LEFT_FRAME1;
         framebuffer_top.address_left2  = PADDR_TOP_LEFT_FRAME2;
@@ -48,8 +48,8 @@ void SetFramebufferLocation(const FramebufferLocation mode) {
 
     case FRAMEBUFFER_LOCATION_VRAM:
     {
-        auto& framebuffer_top = g_regs.Get<Regs::FramebufferTop>();
-        auto& framebuffer_sub = g_regs.Get<Regs::FramebufferBottom>();
+        auto& framebuffer_top = g_regs.framebuffer_config[0];
+        auto& framebuffer_sub = g_regs.framebuffer_config[1];
 
         framebuffer_top.address_left1  = PADDR_VRAM_TOP_LEFT_FRAME1;
         framebuffer_top.address_left2  = PADDR_VRAM_TOP_LEFT_FRAME2;
@@ -107,13 +107,12 @@ inline void Read(T &var, const u32 raw_addr) {
     int index = addr / 4;
 
     // Reads other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= Regs::NumIds || !std::is_same<T,u32>::value)
-    {
+    if (index >= Regs::NumIds() || !std::is_same<T,u32>::value) {
         ERROR_LOG(GPU, "unknown Read%d @ 0x%08X", sizeof(var) * 8, addr);
         return;
     }
 
-    var = g_regs[static_cast<Regs::Id>(addr / 4)];
+    var = g_regs[addr / 4];
 }
 
 template <typename T>
@@ -122,22 +121,22 @@ inline void Write(u32 addr, const T data) {
     int index = addr / 4;
 
     // Writes other than u32 are untested, so I'd rather have them abort than silently fail
-    if (index >= Regs::NumIds || !std::is_same<T,u32>::value)
-    {
+    if (index >= Regs::NumIds() || !std::is_same<T,u32>::value) {
         ERROR_LOG(GPU, "unknown Write%d 0x%08X @ 0x%08X", sizeof(data) * 8, data, addr);
         return;
     }
 
-    g_regs[static_cast<Regs::Id>(index)] = data;
+    g_regs[index] = data;
 
-    switch (static_cast<Regs::Id>(index)) {
+    switch (index) {
 
     // Memory fills are triggered once the fill value is written.
     // NOTE: This is not verified.
-    case Regs::MemoryFill + 3:
-    case Regs::MemoryFill + 7:
+    case GPU_REG_INDEX_WORKAROUND(memory_fill_config[0].value, 0x00004 + 0x3):
+    case GPU_REG_INDEX_WORKAROUND(memory_fill_config[1].value, 0x00008 + 0x3):
     {
-        const auto& config = g_regs.Get<Regs::MemoryFill>(static_cast<Regs::Id>(index - 3));
+        const bool is_second_filler = (index != GPU_REG_INDEX(memory_fill_config[0].value));
+        const auto& config = g_regs.memory_fill_config[is_second_filler];
 
         // TODO: Not sure if this check should be done at GSP level instead
         if (config.address_start) {
@@ -152,9 +151,9 @@ inline void Write(u32 addr, const T data) {
         break;
     }
 
-    case Regs::DisplayTransfer + 6:
+    case GPU_REG_INDEX(display_transfer_config.trigger):
     {
-        const auto& config = g_regs.Get<Regs::DisplayTransfer>();
+        const auto& config = g_regs.display_transfer_config;
         if (config.trigger & 1) {
             u8* source_pointer = Memory::GetPointer(config.GetPhysicalInputAddress());
             u8* dest_pointer = Memory::GetPointer(config.GetPhysicalOutputAddress());
@@ -221,13 +220,13 @@ inline void Write(u32 addr, const T data) {
         break;
     }
 
-    case Regs::CommandProcessor + 4:
+    case GPU_REG_INDEX(command_processor_config.trigger):
     {
-        const auto& config = g_regs.Get<Regs::CommandProcessor>();
+        const auto& config = g_regs.command_processor_config;
         if (config.trigger & 1)
         {
-            // u32* buffer = (u32*)Memory::GetPointer(config.address << 3);
-            ERROR_LOG(GPU, "Beginning 0x%08x bytes of commands from address 0x%08x", config.size, config.address << 3);
+            // u32* buffer = (u32*)Memory::GetPointer(config.GetPhysicalAddress());
+            ERROR_LOG(GPU, "Beginning 0x%08x bytes of commands from address 0x%08x", config.size, config.GetPhysicalAddress());
             // TODO: Process command list!
         }
         break;
@@ -252,7 +251,7 @@ template void Write<u8>(u32 addr, const u8 data);
 
 /// Update hardware
 void Update() {
-    auto& framebuffer_top = g_regs.Get<Regs::FramebufferTop>();
+    auto& framebuffer_top = g_regs.framebuffer_config[0];
     u64 current_ticks = Core::g_app_core->GetTicks();
 
     // Synchronize line...
@@ -280,8 +279,8 @@ void Init() {
 //    SetFramebufferLocation(FRAMEBUFFER_LOCATION_FCRAM);
     SetFramebufferLocation(FRAMEBUFFER_LOCATION_VRAM);
 
-    auto& framebuffer_top = g_regs.Get<Regs::FramebufferTop>();
-    auto& framebuffer_sub = g_regs.Get<Regs::FramebufferBottom>();
+    auto& framebuffer_top = g_regs.framebuffer_config[0];
+    auto& framebuffer_sub = g_regs.framebuffer_config[1];
     // TODO: Width should be 240 instead?
     framebuffer_top.width = 480;
     framebuffer_top.height = 400;
diff --git a/src/core/hw/gpu.h b/src/core/hw/gpu.h
index 42f18a0e7..3065da891 100644
--- a/src/core/hw/gpu.h
+++ b/src/core/hw/gpu.h
@@ -4,32 +4,57 @@
 
 #pragma once
 
+#include <cstddef>
+
 #include "common/common_types.h"
 #include "common/bit_field.h"
-#include "common/register_set.h"
 
 namespace GPU {
 
 static const u32 kFrameCycles   = 268123480 / 60;   ///< 268MHz / 60 frames per second
 static const u32 kFrameTicks    = kFrameCycles / 3; ///< Approximate number of instructions/frame
 
+// Returns index corresponding to the Regs member labeled by field_name
+// TODO: Due to Visual studio bug 209229, offsetof does not return constant expressions
+//       when used with array elements (e.g. GPU_REG_INDEX(memory_fill_config[0])).
+//       For details cf. https://connect.microsoft.com/VisualStudio/feedback/details/209229/offsetof-does-not-produce-a-constant-expression-for-array-members
+//       Hopefully, this will be fixed sometime in the future.
+//       For lack of better alternatives, we currently hardcode the offsets when constant
+//       expressions are needed via GPU_REG_INDEX_WORKAROUND (on sane compilers, static_asserts
+//       will then make sure the offsets indeed match the automatically calculated ones).
+#define GPU_REG_INDEX(field_name) (offsetof(GPU::Regs, field_name) / sizeof(u32))
+#if defined(_MSC_VER)
+#define GPU_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) (backup_workaround_index)
+#else
+// NOTE: Yeah, hacking in a static_assert here just to workaround the lacking MSVC compiler
+//       really is this annoying. This macro just forwards its first argument to GPU_REG_INDEX
+//       and then performs a (no-op) cast to size_t iff the second argument matches the expected
+//       field offset. Otherwise, the compiler will fail to compile this code.
+#define GPU_REG_INDEX_WORKAROUND(field_name, backup_workaround_index) \
+    ((typename std::enable_if<backup_workaround_index == GPU_REG_INDEX(field_name), size_t>::type)GPU_REG_INDEX(field_name))
+#endif
+
 // MMIO region 0x1EFxxxxx
 struct Regs {
-    enum Id : u32 {
-        MemoryFill                = 0x00004, // + 5,6,7; second block at 8-11
-
-        FramebufferTop            = 0x00117, // + 11a,11b,11c,11d(?),11e...126
-        FramebufferBottom         = 0x00157, // + 15a,15b,15c,15d(?),15e...166
-
-        DisplayTransfer           = 0x00300, // + 301,302,303,304,305,306
-
-        CommandProcessor          = 0x00638, // + 63a,63c
 
-        NumIds                    = 0x01000
-    };
-
-    template<Id id>
-    struct Struct;
+// helper macro to properly align structure members.
+// Calling INSERT_PADDING_WORDS will add a new member variable with a name like "pad121",
+// depending on the current source line to make sure variable names are unique.
+#define INSERT_PADDING_WORDS_HELPER1(x, y) x ## y
+#define INSERT_PADDING_WORDS_HELPER2(x, y) INSERT_PADDING_WORDS_HELPER1(x, y)
+#define INSERT_PADDING_WORDS(num_words) u32 INSERT_PADDING_WORDS_HELPER2(pad, __LINE__)[(num_words)];
+
+// helper macro to make sure the defined structures are of the expected size.
+#if defined(_MSC_VER)
+// TODO: MSVC does not support using sizeof() on non-static data members even though this
+//       is technically allowed since C++11. This macro should be enabled once MSVC adds
+//       support for that.
+#define ASSERT_MEMBER_SIZE(name, size_in_bytes)
+#else
+#define ASSERT_MEMBER_SIZE(name, size_in_bytes)  \
+    static_assert(sizeof(name) == size_in_bytes, \
+                  "Structure size and register block length don't match");
+#endif
 
     enum class FramebufferFormat : u32 {
         RGBA8  = 0,
@@ -38,135 +63,191 @@ struct Regs {
         RGB5A1 = 3,
         RGBA4  = 4,
     };
-};
 
-template<>
-struct Regs::Struct<Regs::MemoryFill> {
-    u32 address_start;
-    u32 address_end; // ?
-    u32 size;
-    u32 value; // ?
+    INSERT_PADDING_WORDS(0x4);
 
-    inline u32 GetStartAddress() const {
-        return address_start * 8;
-    }
+    struct {
+        u32 address_start;
+        u32 address_end; // ?
+        u32 size;
+        u32 value; // ?
 
-    inline u32 GetEndAddress() const {
-        return address_end * 8;
-    }
-};
-static_assert(sizeof(Regs::Struct<Regs::MemoryFill>) == 0x10, "Structure size and register block length don't match");
+        inline u32 GetStartAddress() const {
+            return DecodeAddressRegister(address_start);
+        }
 
-template<>
-struct Regs::Struct<Regs::FramebufferTop> {
-    using Format = Regs::FramebufferFormat;
+        inline u32 GetEndAddress() const {
+            return DecodeAddressRegister(address_end);
+        }
+    } memory_fill_config[2];
+    ASSERT_MEMBER_SIZE(memory_fill_config[0], 0x10);
 
-    union {
-        u32 size;
+    INSERT_PADDING_WORDS(0x10b);
 
-        BitField< 0, 16, u32> width;
-        BitField<16, 16, u32> height;
-    };
+    struct {
+        using Format = Regs::FramebufferFormat;
 
-    u32 pad0[2];
+        union {
+            u32 size;
 
-    u32 address_left1;
-    u32 address_left2;
+            BitField< 0, 16, u32> width;
+            BitField<16, 16, u32> height;
+        };
 
-    union {
-        u32 format;
+        INSERT_PADDING_WORDS(0x2);
 
-        BitField< 0, 3, Format> color_format;
-    };
+        u32 address_left1;
+        u32 address_left2;
 
-    u32 pad1;
+        union {
+            u32 format;
 
-    union {
-        u32 active_fb;
+            BitField< 0, 3, Format> color_format;
+        };
 
-        // 0: Use parameters ending with "1"
-        // 1: Use parameters ending with "2"
-        BitField<0, 1, u32> second_fb_active;
-    };
+        INSERT_PADDING_WORDS(0x1);
 
-    u32 pad2[5];
+        union {
+            u32 active_fb;
 
-    // Distance between two pixel rows, in bytes
-    u32 stride;
+            // 0: Use parameters ending with "1"
+            // 1: Use parameters ending with "2"
+            BitField<0, 1, u32> second_fb_active;
+        };
 
-    u32 address_right1;
-    u32 address_right2;
-};
+        INSERT_PADDING_WORDS(0x5);
 
-template<>
-struct Regs::Struct<Regs::FramebufferBottom> : public Regs::Struct<Regs::FramebufferTop> {
-};
-static_assert(sizeof(Regs::Struct<Regs::FramebufferTop>) == 0x40, "Structure size and register block length don't match");
+        // Distance between two pixel rows, in bytes
+        u32 stride;
 
-template<>
-struct Regs::Struct<Regs::DisplayTransfer> {
-    using Format = Regs::FramebufferFormat;
+        u32 address_right1;
+        u32 address_right2;
 
-    u32 input_address;
-    u32 output_address;
+        INSERT_PADDING_WORDS(0x30);
+    } framebuffer_config[2];
+    ASSERT_MEMBER_SIZE(framebuffer_config[0], 0x100);
 
-    inline u32 GetPhysicalInputAddress() const {
-        return input_address * 8;
-    }
+    INSERT_PADDING_WORDS(0x169);
 
-    inline u32 GetPhysicalOutputAddress() const {
-        return output_address * 8;
-    }
+    struct {
+        using Format = Regs::FramebufferFormat;
 
-    union {
-        u32 output_size;
+        u32 input_address;
+        u32 output_address;
 
-        BitField< 0, 16, u32> output_width;
-        BitField<16, 16, u32> output_height;
-    };
+        inline u32 GetPhysicalInputAddress() const {
+            return DecodeAddressRegister(input_address);
+        }
 
-    union {
-        u32 input_size;
+        inline u32 GetPhysicalOutputAddress() const {
+            return DecodeAddressRegister(output_address);
+        }
 
-        BitField< 0, 16, u32> input_width;
-        BitField<16, 16, u32> input_height;
-    };
+        union {
+            u32 output_size;
 
-    union {
-        u32 flags;
+            BitField< 0, 16, u32> output_width;
+            BitField<16, 16, u32> output_height;
+        };
 
-        BitField< 0, 1, u32> flip_data;        // flips input data horizontally (TODO) if true
-        BitField< 8, 3, Format> input_format;
-        BitField<12, 3, Format> output_format;
-        BitField<16, 1, u32> output_tiled;     // stores output in a tiled format
-    };
+        union {
+            u32 input_size;
 
-    u32 unknown;
+            BitField< 0, 16, u32> input_width;
+            BitField<16, 16, u32> input_height;
+        };
 
-    // it seems that writing to this field triggers the display transfer
-    u32 trigger;
-};
-static_assert(sizeof(Regs::Struct<Regs::DisplayTransfer>) == 0x1C, "Structure size and register block length don't match");
+        union {
+            u32 flags;
 
-template<>
-struct Regs::Struct<Regs::CommandProcessor> {
-    // command list size
-    u32 size;
+            BitField< 0, 1, u32> flip_data;        // flips input data horizontally (TODO) if true
+            BitField< 8, 3, Format> input_format;
+            BitField<12, 3, Format> output_format;
+            BitField<16, 1, u32> output_tiled;     // stores output in a tiled format
+        };
 
-    u32 pad0;
+        INSERT_PADDING_WORDS(0x1);
 
-    // command list address
-    u32 address;
+        // it seems that writing to this field triggers the display transfer
+        u32 trigger;
+    } display_transfer_config;
+    ASSERT_MEMBER_SIZE(display_transfer_config, 0x1c);
 
-    u32 pad1;
+    INSERT_PADDING_WORDS(0x331);
 
-    // it seems that writing to this field triggers command list processing
-    u32 trigger;
-};
-static_assert(sizeof(Regs::Struct<Regs::CommandProcessor>) == 0x14, "Structure size and register block length don't match");
+    struct {
+        // command list size
+        u32 size;
+
+        INSERT_PADDING_WORDS(0x1);
+
+        // command list address
+        u32 address;
+
+        INSERT_PADDING_WORDS(0x1);
+
+        // it seems that writing to this field triggers command list processing
+        u32 trigger;
 
+        inline u32 GetPhysicalAddress() const {
+            return DecodeAddressRegister(address);
+        }
+    } command_processor_config;
+    ASSERT_MEMBER_SIZE(command_processor_config, 0x14);
 
-extern RegisterSet<u32, Regs> g_regs;
+    INSERT_PADDING_WORDS(0x9c3);
+
+#undef INSERT_PADDING_WORDS_HELPER1
+#undef INSERT_PADDING_WORDS_HELPER2
+#undef INSERT_PADDING_WORDS
+
+    static inline int NumIds() {
+        return sizeof(Regs) / sizeof(u32);
+    }
+
+    u32& operator [] (int index) const {
+        u32* content = (u32*)this;
+        return content[index];
+    }
+
+    u32& operator [] (int index) {
+        u32* content = (u32*)this;
+        return content[index];
+    }
+
+private:
+    /*
+     * Most physical addresses which GPU registers refer to are 8-byte aligned.
+     * This function should be used to get the address from a raw register value.
+     */
+    static inline u32 DecodeAddressRegister(u32 register_value) {
+        return register_value * 8;
+    }
+};
+static_assert(std::is_standard_layout<Regs>::value, "Structure does not use standard layout");
+
+// TODO: MSVC does not support using offsetof() on non-static data members even though this
+//       is technically allowed since C++11. This macro should be enabled once MSVC adds
+//       support for that.
+#ifndef _MSC_VER
+#define ASSERT_REG_POSITION(field_name, position)             \
+    static_assert(offsetof(Regs, field_name) == position * 4, \
+                  "Field "#field_name" has invalid position")
+
+ASSERT_REG_POSITION(memory_fill_config[0],    0x00004);
+ASSERT_REG_POSITION(memory_fill_config[1],    0x00008);
+ASSERT_REG_POSITION(framebuffer_config[0],    0x00117);
+ASSERT_REG_POSITION(framebuffer_config[1],    0x00157);
+ASSERT_REG_POSITION(display_transfer_config,  0x00300);
+ASSERT_REG_POSITION(command_processor_config, 0x00638);
+
+#undef ASSERT_REG_POSITION
+#endif // !defined(_MSC_VER)
+
+// The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value anyway.
+static_assert(sizeof(Regs) == 0x1000 * sizeof(u32), "Invalid total size of register set");
+
+extern Regs g_regs;
 
 enum {
     TOP_ASPECT_X        = 0x5,