Y2R: Rework conversion process, enabling support for all formats

5 files changed, 695 insertions, 163 deletions

diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
index bbc28516..ea5533dc 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -96,6 +96,7 @@ set(SRCS
             hw/gpu.cpp
             hw/hw.cpp
             hw/lcd.cpp
+            hw/y2r.cpp
             loader/3dsx.cpp
             loader/elf.cpp
             loader/loader.cpp
@@ -209,6 +210,7 @@ set(HEADERS
             hw/gpu.h
             hw/hw.h
             hw/lcd.h
+            hw/y2r.h
             loader/3dsx.h
             loader/elf.h
             loader/loader.h
diff --git a/src/core/hle/service/y2r_u.cpp b/src/core/hle/service/y2r_u.cpp
index 73a0899d..17cb4f0f 100644
--- a/src/core/hle/service/y2r_u.cpp
+++ b/src/core/hle/service/y2r_u.cpp
@@ -9,8 +9,8 @@
 #include "core/hle/hle.h"
 #include "core/hle/kernel/event.h"
 #include "core/hle/service/y2r_u.h"
+#include "core/hw/y2r.h"
 #include "core/mem_map.h"
-#include "core/memory.h"
 
 #include "video_core/utils.h"
 #include "video_core/video_core.h"
@@ -20,47 +20,6 @@
 
 namespace Y2R_U {
 
-enum class InputFormat : u8 {
-    /// 8-bit input, with YUV components in separate planes and using 4:2:2 subsampling.
-    YUV422_Indiv8 = 0,
-    /// 8-bit input, with YUV components in separate planes and using 4:2:0 subsampling.
-    YUV420_Indiv8 = 1,
-
-    YUV422_INDIV_16 = 2,
-    YUV420_INDIV_16 = 3,
-    YUV422_BATCH = 4,
-};
-
-enum class OutputFormat : u8 {
-    Rgb32 = 0,
-    Rgb24 = 1,
-    Rgb16_555 = 2,
-    Rgb16_565 = 3,
-};
-
-enum class Rotation : u8 {
-    None = 0,
-    Clockwise_90 = 1,
-    Clockwise_180 = 2,
-    Clockwise_270 = 3,
-};
-
-enum class BlockAlignment : u8 {
-    /// Image is output in linear format suitable for use as a framebuffer.
-    Linear = 0,
-    /// Image is output in tiled PICA format, suitable for use as a texture.
-    Block8x8 = 1,
-};
-
-enum class StandardCoefficient : u8 {
-    ITU_Rec601 = 0,
-    ITU_Rec709 = 1,
-    ITU_Rec601_Scaling = 2,
-    ITU_Rec709_Scaling = 3,
-};
-
-static Kernel::SharedPtr<Kernel::Event> completion_event;
-
 struct ConversionParameters {
     InputFormat input_format;
     OutputFormat output_format;
@@ -74,28 +33,60 @@ struct ConversionParameters {
 };
 static_assert(sizeof(ConversionParameters) == 12, "ConversionParameters struct has incorrect size");
 
-struct ConversionBuffer {
-    VAddr address;
-    u32 image_size;
-    u16 transfer_unit;
-    u16 stride;
-};
+static Kernel::SharedPtr<Kernel::Event> completion_event;
+static ConversionConfiguration conversion;
 
-struct ConversionData {
-    ConversionParameters params;
-    /// Input parameters for the Y (luma) plane
-    ConversionBuffer src_Y;
-    /// Output parameters for the conversion results
-    ConversionBuffer dst;
+static const CoefficientSet standard_coefficients[4] = {
+    {{ 0x100, 0x166, 0xB6, 0x58, 0x1C5, -0x166F, 0x10EE, -0x1C5B }}, // ITU_Rec601
+    {{ 0x100, 0x193, 0x77, 0x2F, 0x1DB, -0x1933,  0xA7C, -0x1D51 }}, // ITU_Rec709
+    {{ 0x12A, 0x198, 0xD0, 0x64, 0x204, -0x1BDE, 0x10F2, -0x229B }}, // ITU_Rec601_Scaling
+    {{ 0x12A, 0x1CA, 0x88, 0x36, 0x21C, -0x1F04,  0x99C, -0x2421 }}, // ITU_Rec709_Scaling
 };
 
-static ConversionData conversion;
+ResultCode ConversionConfiguration::SetInputLineWidth(u16 width) {
+    if (width == 0 || width > 1024 || width % 8 != 0) {
+        return ResultCode(ErrorDescription::OutOfRange, ErrorModule::CAM,
+            ErrorSummary::InvalidArgument, ErrorLevel::Usage); // 0xE0E053FD
+    }
+
+    // Note: The hardware uses the register value 0 to represent a width of 1024, so for a width of
+    // 1024 the `camera` module would set the value 0 here, but we don't need to emulate this
+    // internal detail.
+    this->input_line_width = width;
+    return RESULT_SUCCESS;
+}
+
+ResultCode ConversionConfiguration::SetInputLines(u16 lines) {
+    if (lines == 0 || lines > 1024) {
+        return ResultCode(ErrorDescription::OutOfRange, ErrorModule::CAM,
+            ErrorSummary::InvalidArgument, ErrorLevel::Usage); // 0xE0E053FD
+    }
+
+    // Note: In what appears to be a bug, the `camera` module does not set the hardware register at
+    // all if `lines` is 1024, so the conversion uses the last value that was set. The intention
+    // was probably to set it to 0 like in SetInputLineWidth.
+    if (lines != 1024) {
+        this->input_lines = lines;
+    }
+    return RESULT_SUCCESS;
+}
+
+ResultCode ConversionConfiguration::SetStandardCoefficient(StandardCoefficient standard_coefficient) {
+    size_t index = static_cast<size_t>(standard_coefficient);
+    if (index >= 4) {
+        return ResultCode(ErrorDescription::InvalidEnumValue, ErrorModule::CAM,
+            ErrorSummary::InvalidArgument, ErrorLevel::Usage); // 0xE0E053ED
+    }
+
+    std::memcpy(coefficients.data(), standard_coefficients[index].data(), sizeof(coefficients));
+    return RESULT_SUCCESS;
+}
 
 static void SetInputFormat(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.input_format = static_cast<InputFormat>(cmd_buff[1]);
-    LOG_DEBUG(Service_Y2R, "called input_format=%u", conversion.params.input_format);
+    conversion.input_format = static_cast<InputFormat>(cmd_buff[1]);
+    LOG_DEBUG(Service_Y2R, "called input_format=%hhu", conversion.input_format);
 
     cmd_buff[1] = RESULT_SUCCESS.raw;
 }
@@ -103,8 +94,8 @@ static void SetInputFormat(Service::Interface* self) {
 static void SetOutputFormat(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.output_format = static_cast<OutputFormat>(cmd_buff[1]);
-    LOG_DEBUG(Service_Y2R, "called output_format=%u", conversion.params.output_format);
+    conversion.output_format = static_cast<OutputFormat>(cmd_buff[1]);
+    LOG_DEBUG(Service_Y2R, "called output_format=%hhu", conversion.output_format);
 
     cmd_buff[1] = RESULT_SUCCESS.raw;
 }
@@ -112,8 +103,8 @@ static void SetOutputFormat(Service::Interface* self) {
 static void SetRotation(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.rotation = static_cast<Rotation>(cmd_buff[1]);
-    LOG_DEBUG(Service_Y2R, "called rotation=%u", conversion.params.rotation);
+    conversion.rotation = static_cast<Rotation>(cmd_buff[1]);
+    LOG_DEBUG(Service_Y2R, "called rotation=%hhu", conversion.rotation);
 
     cmd_buff[1] = RESULT_SUCCESS.raw;
 }
@@ -121,10 +112,18 @@ static void SetRotation(Service::Interface* self) {
 static void SetBlockAlignment(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.block_alignment = static_cast<BlockAlignment>(cmd_buff[1]);
-    LOG_DEBUG(Service_Y2R, "called alignment=%u", conversion.params.block_alignment);
+    conversion.block_alignment = static_cast<BlockAlignment>(cmd_buff[1]);
+    LOG_DEBUG(Service_Y2R, "called alignment=%hhu", conversion.block_alignment);
+
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
+
+static void SetTransferEndInterrupt(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
 
+    cmd_buff[0] = 0x000D0040;
     cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "(STUBBED) called");
 }
 
 /**
@@ -147,11 +146,56 @@ static void SetSendingY(Service::Interface* self) {
     conversion.src_Y.address = cmd_buff[1];
     conversion.src_Y.image_size = cmd_buff[2];
     conversion.src_Y.transfer_unit = cmd_buff[3];
-    conversion.src_Y.stride = cmd_buff[4];
+    conversion.src_Y.gap = cmd_buff[4];
     u32 src_process_handle = cmd_buff[6];
     LOG_DEBUG(Service_Y2R, "called image_size=0x%08X, transfer_unit=%hu, transfer_stride=%hu, "
         "src_process_handle=0x%08X", conversion.src_Y.image_size,
-        conversion.src_Y.transfer_unit, conversion.src_Y.stride, src_process_handle);
+        conversion.src_Y.transfer_unit, conversion.src_Y.gap, src_process_handle);
+
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
+
+static void SetSendingU(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    conversion.src_U.address = cmd_buff[1];
+    conversion.src_U.image_size = cmd_buff[2];
+    conversion.src_U.transfer_unit = cmd_buff[3];
+    conversion.src_U.gap = cmd_buff[4];
+    u32 src_process_handle = cmd_buff[6];
+    LOG_DEBUG(Service_Y2R, "called image_size=0x%08X, transfer_unit=%hu, transfer_stride=%hu, "
+        "src_process_handle=0x%08X", conversion.src_U.image_size,
+        conversion.src_U.transfer_unit, conversion.src_U.gap, src_process_handle);
+
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
+
+static void SetSendingV(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    conversion.src_V.address = cmd_buff[1];
+    conversion.src_V.image_size = cmd_buff[2];
+    conversion.src_V.transfer_unit = cmd_buff[3];
+    conversion.src_V.gap = cmd_buff[4];
+    u32 src_process_handle = cmd_buff[6];
+    LOG_DEBUG(Service_Y2R, "called image_size=0x%08X, transfer_unit=%hu, transfer_stride=%hu, "
+        "src_process_handle=0x%08X", conversion.src_V.image_size,
+        conversion.src_V.transfer_unit, conversion.src_V.gap, src_process_handle);
+
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
+
+static void SetSendingYUYV(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    conversion.src_YUYV.address = cmd_buff[1];
+    conversion.src_YUYV.image_size = cmd_buff[2];
+    conversion.src_YUYV.transfer_unit = cmd_buff[3];
+    conversion.src_YUYV.gap = cmd_buff[4];
+    u32 src_process_handle = cmd_buff[6];
+    LOG_DEBUG(Service_Y2R, "called image_size=0x%08X, transfer_unit=%hu, transfer_stride=%hu, "
+        "src_process_handle=0x%08X", conversion.src_YUYV.image_size,
+        conversion.src_YUYV.transfer_unit, conversion.src_YUYV.gap, src_process_handle);
 
     cmd_buff[1] = RESULT_SUCCESS.raw;
 }
@@ -162,11 +206,11 @@ static void SetReceiving(Service::Interface* self) {
     conversion.dst.address = cmd_buff[1];
     conversion.dst.image_size = cmd_buff[2];
     conversion.dst.transfer_unit = cmd_buff[3];
-    conversion.dst.stride = cmd_buff[4];
+    conversion.dst.gap = cmd_buff[4];
     u32 dst_process_handle = cmd_buff[6];
     LOG_DEBUG(Service_Y2R, "called image_size=0x%08X, transfer_unit=%hu, transfer_stride=%hu, "
         "dst_process_handle=0x%08X", conversion.dst.image_size,
-        conversion.dst.transfer_unit, conversion.dst.stride,
+        conversion.dst.transfer_unit, conversion.dst.gap,
         dst_process_handle);
 
     cmd_buff[1] = RESULT_SUCCESS.raw;
@@ -175,107 +219,54 @@ static void SetReceiving(Service::Interface* self) {
 static void SetInputLineWidth(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.input_line_width = cmd_buff[1];
-    LOG_DEBUG(Service_Y2R, "input_line_width=%u", conversion.params.input_line_width);
-
-    cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "called input_line_width=%u", cmd_buff[1]);
+    cmd_buff[1] = conversion.SetInputLineWidth(cmd_buff[1]).raw;
 }
 
 static void SetInputLines(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    conversion.params.input_lines = cmd_buff[1];
-    LOG_DEBUG(Service_Y2R, "input_line_number=%u", conversion.params.input_lines);
-
-    cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "called input_line_number=%u", cmd_buff[1]);
+    cmd_buff[1] = conversion.SetInputLines(cmd_buff[1]).raw;
 }
 
-static void StartConversion(Service::Interface* self) {
+static void SetCoefficient(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    const ConversionParameters& params = conversion.params;
-
-    const u8* srcY_buffer = Memory::GetPointer(conversion.src_Y.address);
-    u8* dst_buffer = Memory::GetPointer(conversion.dst.address);
-
-    // TODO: support color and other kinds of conversions
-    ASSERT(params.input_format == InputFormat::YUV422_Indiv8
-        || params.input_format == InputFormat::YUV420_Indiv8);
-    ASSERT(params.output_format == OutputFormat::Rgb24);
-    ASSERT(params.rotation == Rotation::None);
-    const int bpp = 3;
-
-    switch (params.block_alignment) {
-    case BlockAlignment::Linear:
-    {
-        const size_t input_lines = params.input_lines;
-        const size_t input_line_width = params.input_line_width;
-        const size_t srcY_stride = conversion.src_Y.stride;
-        const size_t dst_stride = conversion.dst.stride;
-
-        size_t srcY_offset = 0;
-        size_t dst_offset = 0;
-
-        for (size_t line = 0; line < input_lines; ++line) {
-            for (size_t i = 0; i < input_line_width; ++i) {
-                u8 Y = srcY_buffer[srcY_offset];
-                dst_buffer[dst_offset + 0] = Y;
-                dst_buffer[dst_offset + 1] = Y;
-                dst_buffer[dst_offset + 2] = Y;
-
-                srcY_offset += 1;
-                dst_offset += bpp;
-            }
-            srcY_offset += srcY_stride;
-            dst_offset += dst_stride;
-        }
-        break;
-    }
-    case BlockAlignment::Block8x8:
-    {
-        const size_t input_lines = params.input_lines;
-        const size_t input_line_width = params.input_line_width;
-        const size_t srcY_stride = conversion.src_Y.stride;
-        const size_t dst_transfer_unit = conversion.dst.transfer_unit;
-        const size_t dst_stride = conversion.dst.stride;
-
-        size_t srcY_offset = 0;
-        size_t dst_tile_line_offs = 0;
+    const u16* coefficients = reinterpret_cast<const u16*>(&cmd_buff[1]);
+    std::memcpy(conversion.coefficients.data(), coefficients, sizeof(CoefficientSet));
+    LOG_DEBUG(Service_Y2R, "called coefficients=[%hX, %hX, %hX, %hX, %hX, %hX, %hX, %hX]",
+            coefficients[0], coefficients[1], coefficients[2], coefficients[3],
+            coefficients[4], coefficients[5], coefficients[6], coefficients[7]);
 
-        const size_t tile_size = 8 * 8 * bpp;
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
 
-        for (size_t line = 0; line < input_lines;) {
-            size_t max_line = line + 8;
+static void SetStandardCoefficient(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
 
-            for (; line < max_line; ++line) {
-                for (size_t x = 0; x < input_line_width; ++x) {
-                    size_t tile_x = x / 8;
+    LOG_DEBUG(Service_Y2R, "called standard_coefficient=%u", cmd_buff[1]);
 
-                    size_t dst_tile_offs = dst_tile_line_offs + tile_x * tile_size;
-                    size_t tile_i = VideoCore::MortonInterleave((u32)x, (u32)line);
+    cmd_buff[1] = conversion.SetStandardCoefficient((StandardCoefficient)cmd_buff[1]).raw;
+}
 
-                    size_t dst_offset = dst_tile_offs + tile_i * bpp;
+static void SetAlpha(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
 
-                    u8 Y = srcY_buffer[srcY_offset];
-                    dst_buffer[dst_offset + 0] = Y;
-                    dst_buffer[dst_offset + 1] = Y;
-                    dst_buffer[dst_offset + 2] = Y;
+    conversion.alpha = cmd_buff[1];
+    LOG_DEBUG(Service_Y2R, "called alpha=%hu", conversion.alpha);
 
-                    srcY_offset += 1;
-                }
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+}
 
-                srcY_offset += srcY_stride;
-            }
+static void StartConversion(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
 
-            dst_tile_line_offs += dst_transfer_unit + dst_stride;
-        }
-        break;
-    }
-    }
+    HW::Y2R::PerformConversion(conversion);
 
-    // dst_image_size would seem to be perfect for this, but it doesn't include the stride :(
-    u32 total_output_size = params.input_lines *
-        (conversion.dst.transfer_unit + conversion.dst.stride);
+    // dst_image_size would seem to be perfect for this, but it doesn't include the gap :(
+    u32 total_output_size = conversion.input_lines *
+        (conversion.dst.transfer_unit + conversion.dst.gap);
     VideoCore::g_renderer->hw_rasterizer->NotifyFlush(
         Memory::VirtualToPhysicalAddress(conversion.dst.address), total_output_size);
 
@@ -285,6 +276,14 @@ static void StartConversion(Service::Interface* self) {
     cmd_buff[1] = RESULT_SUCCESS.raw;
 }
 
+static void StopConversion(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    cmd_buff[0] = 0x00270040;
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "called");
+}
+
 /**
  * Y2R_U::IsBusyConversion service function
  *  Outputs:
@@ -306,15 +305,31 @@ static void SetConversionParams(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
     auto params = reinterpret_cast<const ConversionParameters*>(&cmd_buff[1]);
-    conversion.params = *params;
-
-    cmd_buff[0] = 0x00290000; // TODO verify
-    cmd_buff[1] = RESULT_SUCCESS.raw;
     LOG_DEBUG(Service_Y2R,
         "called input_format=%hhu output_format=%hhu rotation=%hhu block_alignment=%hhu "
-        "input_line_width=%hX input_lines=%hu standard_coefficient=%hhu reserved=%hhu alpha=%hX",
+        "input_line_width=%hu input_lines=%hu standard_coefficient=%hhu "
+        "reserved=%hhu alpha=%hX",
         params->input_format, params->output_format, params->rotation, params->block_alignment,
-        params->input_line_width, params->input_lines, params->standard_coefficient);
+        params->input_line_width, params->input_lines, params->standard_coefficient,
+        params->reserved, params->alpha);
+
+    ResultCode result = RESULT_SUCCESS;
+
+    conversion.input_format = params->input_format;
+    conversion.output_format = params->output_format;
+    conversion.rotation = params->rotation;
+    conversion.block_alignment = params->block_alignment;
+    result = conversion.SetInputLineWidth(params->input_line_width);
+    if (result.IsError()) goto cleanup;
+    result = conversion.SetInputLines(params->input_lines);
+    if (result.IsError()) goto cleanup;
+    result = conversion.SetStandardCoefficient(params->standard_coefficient);
+    if (result.IsError()) goto cleanup;
+    conversion.alpha = params->alpha;
+
+cleanup:
+    cmd_buff[0] = 0x00290040; // TODO verify
+    cmd_buff[1] = result.raw;
 }
 
 static void PingProcess(Service::Interface* self) {
@@ -325,28 +340,63 @@ static void PingProcess(Service::Interface* self) {
     LOG_WARNING(Service_Y2R, "(STUBBED) called");
 }
 
+static void DriverInitialize(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    conversion.input_format = InputFormat::YUV422_Indiv8;
+    conversion.output_format = OutputFormat::RGBA8;
+    conversion.rotation = Rotation::None;
+    conversion.block_alignment = BlockAlignment::Linear;
+    conversion.coefficients.fill(0);
+    conversion.SetInputLineWidth(1024);
+    conversion.SetInputLines(1024);
+    conversion.alpha = 0;
+
+    ConversionBuffer zero_buffer = {};
+    conversion.src_Y = zero_buffer;
+    conversion.src_U = zero_buffer;
+    conversion.src_V = zero_buffer;
+    conversion.dst = zero_buffer;
+
+    completion_event->Clear();
+
+    cmd_buff[0] = 0x002B0040;
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "called");
+}
+
+static void DriverFinalize(Service::Interface* self) {
+    u32* cmd_buff = Kernel::GetCommandBuffer();
+
+    cmd_buff[0] = 0x002C0040;
+    cmd_buff[1] = RESULT_SUCCESS.raw;
+    LOG_DEBUG(Service_Y2R, "called");
+}
+
 const Interface::FunctionInfo FunctionTable[] = {
     {0x00010040, SetInputFormat,          "SetInputFormat"},
     {0x00030040, SetOutputFormat,         "SetOutputFormat"},
     {0x00050040, SetRotation,             "SetRotation"},
     {0x00070040, SetBlockAlignment,       "SetBlockAlignment"},
-    {0x000D0040, nullptr,                 "SetTransferEndInterrupt"},
+    {0x000D0040, SetTransferEndInterrupt, "SetTransferEndInterrupt"},
     {0x000F0000, GetTransferEndEvent,     "GetTransferEndEvent"},
     {0x00100102, SetSendingY,             "SetSendingY"},
-    {0x00110102, nullptr,                 "SetSendingU"},
-    {0x00120102, nullptr,                 "SetSendingV"},
+    {0x00110102, SetSendingU,             "SetSendingU"},
+    {0x00120102, SetSendingV,             "SetSendingV"},
+    {0x00130102, SetSendingYUYV,          "SetSendingYUYV"},
     {0x00180102, SetReceiving,            "SetReceiving"},
     {0x001A0040, SetInputLineWidth,       "SetInputLineWidth"},
     {0x001C0040, SetInputLines,           "SetInputLines"},
-    {0x00200040, nullptr,                 "SetStandardCoefficient"},
-    {0x00220040, nullptr,                 "SetAlpha"},
+    {0x001E0100, SetCoefficient,          "SetCoefficient"},
+    {0x00200040, SetStandardCoefficient,  "SetStandardCoefficient"},
+    {0x00220040, SetAlpha,                "SetAlpha"},
     {0x00260000, StartConversion,         "StartConversion"},
-    {0x00270000, nullptr,                 "StopConversion"},
+    {0x00270000, StopConversion,          "StopConversion"},
     {0x00280000, IsBusyConversion,        "IsBusyConversion"},
     {0x002901C0, SetConversionParams,     "SetConversionParams"},
     {0x002A0000, PingProcess,             "PingProcess"},
-    {0x002B0000, nullptr,                 "DriverInitialize"},
-    {0x002C0000, nullptr,                 "DriverFinalize"},
+    {0x002B0000, DriverInitialize,        "DriverInitialize"},
+    {0x002C0000, DriverFinalize,          "DriverFinalize"},
 };
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
diff --git a/src/core/hle/service/y2r_u.h b/src/core/hle/service/y2r_u.h
index 171aecfd..7df47fcb 100644
--- a/src/core/hle/service/y2r_u.h
+++ b/src/core/hle/service/y2r_u.h
@@ -4,6 +4,10 @@
 
 #pragma once
 
+#include <array>
+
+#include "common/common_types.h"
+
 #include "core/hle/service/service.h"
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -11,6 +15,98 @@
 
 namespace Y2R_U {
 
+enum class InputFormat : u8 {
+    /// 8-bit input, with YUV components in separate planes and 4:2:2 subsampling.
+    YUV422_Indiv8 = 0,
+    /// 8-bit input, with YUV components in separate planes and 4:2:0 subsampling.
+    YUV420_Indiv8 = 1,
+
+    /// 16-bit input (only LSB used), with YUV components in separate planes and 4:2:2 subsampling.
+    YUV422_Indiv16 = 2,
+    /// 16-bit input (only LSB used), with YUV components in separate planes and 4:2:0 subsampling.
+    YUV420_Indiv16 = 3,
+
+    /// 8-bit input, with a single interleaved stream in YUYV format and 4:2:2 subsampling.
+    YUYV422_Interleaved = 4,
+};
+
+enum class OutputFormat : u8 {
+    RGBA8 = 0,
+    RGB8 = 1,
+    RGB5A1 = 2,
+    RGB565 = 3,
+};
+
+enum class Rotation : u8 {
+    None = 0,
+    Clockwise_90 = 1,
+    Clockwise_180 = 2,
+    Clockwise_270 = 3,
+};
+
+enum class BlockAlignment : u8 {
+    /// Image is output in linear format suitable for use as a framebuffer.
+    Linear = 0,
+    /// Image is output in tiled PICA format, suitable for use as a texture.
+    Block8x8 = 1,
+};
+
+enum class StandardCoefficient : u8 {
+    /// ITU Rec. BT.601 primaries, with PC ranges.
+    ITU_Rec601 = 0,
+    /// ITU Rec. BT.709 primaries, with PC ranges.
+    ITU_Rec709 = 1,
+    /// ITU Rec. BT.601 primaries, with TV ranges.
+    ITU_Rec601_Scaling = 2,
+    /// ITU Rec. BT.709 primaries, with TV ranges.
+    ITU_Rec709_Scaling = 3,
+};
+
+/**
+ * A set of coefficients configuring the RGB to YUV conversion. Coefficients 0-4 are unsigned 2.8
+ * fixed pointer numbers representing entries on the conversion matrix, while coefficient 5-7 are
+ * signed 11.5 fixed point numbers added as offsets to the RGB result.
+ *
+ * The overall conversion process formula is:
+ * ```
+ * R = trunc((c_0 * Y           + c_1 * V) + c_5 + 0.75)
+ * G = trunc((c_0 * Y - c_3 * U - c_2 * V) + c_6 + 0.75)
+ * B = trunc((c_0 * Y + c_4 * U          ) + c_7 + 0.75)
+ * ```
+ */
+using CoefficientSet = std::array<s16, 8>;
+
+struct ConversionBuffer {
+    /// Current reading/writing address of this buffer.
+    VAddr address;
+    /// Remaining amount of bytes to be DMAed, does not include the inter-trasfer gap.
+    u32 image_size;
+    /// Size of a single DMA transfer.
+    u16 transfer_unit;
+    /// Amount of bytes to be skipped between copying each `transfer_unit` bytes.
+    u16 gap;
+};
+
+struct ConversionConfiguration {
+    InputFormat input_format;
+    OutputFormat output_format;
+    Rotation rotation;
+    BlockAlignment block_alignment;
+    u16 input_line_width;
+    u16 input_lines;
+    CoefficientSet coefficients;
+    u16 alpha;
+
+    /// Input parameters for the Y (luma) plane
+    ConversionBuffer src_Y, src_U, src_V, src_YUYV;
+    /// Output parameters for the conversion results
+    ConversionBuffer dst;
+
+    ResultCode SetInputLineWidth(u16 width);
+    ResultCode SetInputLines(u16 lines);
+    ResultCode SetStandardCoefficient(StandardCoefficient standard_coefficient);
+};
+
 class Interface : public Service::Interface {
 public:
     Interface();
diff --git a/src/core/hw/y2r.cpp b/src/core/hw/y2r.cpp
new file mode 100644
index 00000000..5b7fb39e
--- /dev/null
+++ b/src/core/hw/y2r.cpp
@@ -0,0 +1,369 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <numeric>
+
+#include "common/assert.h"
+#include "common/color.h"
+#include "common/common_types.h"
+#include "common/math_util.h"
+#include "common/vector_math.h"
+
+#include "core/hle/service/y2r_u.h"
+#include "core/memory.h"
+
+namespace HW {
+namespace Y2R {
+
+using namespace Y2R_U;
+
+static const size_t MAX_TILES = 1024 / 8;
+static const size_t TILE_SIZE = 8 * 8;
+using ImageTile = std::array<u32, TILE_SIZE>;
+
+/// Converts a image strip from the source YUV format into individual 8x8 RGB32 tiles.
+static void ConvertYUVToRGB(InputFormat input_format,
+        const u8* input_Y, const u8* input_U, const u8* input_V, ImageTile output[],
+        unsigned int width, unsigned int height, const CoefficientSet& coefficients) {
+
+    for (unsigned int y = 0; y < height; ++y) {
+        for (unsigned int x = 0; x < width; ++x) {
+            s32 Y, U, V;
+            switch (input_format) {
+            case InputFormat::YUV422_Indiv8:
+            case InputFormat::YUV422_Indiv16:
+                Y = input_Y[y * width + x];
+                U = input_U[(y * width + x) / 2];
+                V = input_V[(y * width + x) / 2];
+                break;
+            case InputFormat::YUV420_Indiv8:
+            case InputFormat::YUV420_Indiv16:
+                Y = input_Y[y * width + x];
+                U = input_U[((y / 2) * width + x) / 2];
+                V = input_V[((y / 2) * width + x) / 2];
+                break;
+            case InputFormat::YUYV422_Interleaved:
+                Y = input_Y[(y * width + x) * 2];
+                U = input_Y[(y * width + (x / 2) * 2) * 2 + 1];
+                V = input_Y[(y * width + (x / 2) * 2) * 2 + 3];
+                break;
+            }
+
+            // This conversion process is bit-exact with hardware, as far as could be tested.
+            auto& c = coefficients;
+            s32 cY = c[0]*Y;
+
+            s32 r = cY          + c[1]*V;
+            s32 g = cY - c[3]*U - c[2]*V;
+            s32 b = cY + c[4]*U;
+
+            const s32 rounding_offset = 0x18;
+            r = (r >> 3) + c[5] + rounding_offset;
+            g = (g >> 3) + c[6] + rounding_offset;
+            b = (b >> 3) + c[7] + rounding_offset;
+
+            unsigned int tile = x / 8;
+            unsigned int tile_x = x % 8;
+            u32* out = &output[tile][y * 8 + tile_x];
+
+            using MathUtil::Clamp;
+            *out = ((u32)Clamp(r >> 5, 0, 0xFF) << 24) |
+                   ((u32)Clamp(g >> 5, 0, 0xFF) << 16) |
+                   ((u32)Clamp(b >> 5, 0, 0xFF) << 8);
+        }
+    }
+}
+
+/// Simulates an incoming CDMA transfer. The N parameter is used to automatically convert 16-bit formats to 8-bit.
+template <size_t N>
+static void ReceiveData(u8* output, ConversionBuffer& buf, size_t amount_of_data) {
+    const u8* input = Memory::GetPointer(buf.address);
+
+    size_t output_unit = buf.transfer_unit / N;
+    ASSERT(amount_of_data % output_unit == 0);
+
+    while (amount_of_data > 0) {
+        for (size_t i = 0; i < output_unit; ++i) {
+            output[i] = input[i * N];
+        }
+
+        output += output_unit;
+        input += buf.transfer_unit + buf.gap;
+
+        buf.address += buf.transfer_unit + buf.gap;
+        buf.image_size -= buf.transfer_unit;
+        amount_of_data -= output_unit;
+    }
+}
+
+/// Convert intermediate RGB32 format to the final output format while simulating an outgoing CDMA transfer.
+static void SendData(const u32* input, ConversionBuffer& buf, int amount_of_data,
+        OutputFormat output_format, u8 alpha) {
+
+    u8* output = Memory::GetPointer(buf.address);
+
+    while (amount_of_data > 0) {
+        u8* unit_end = output + buf.transfer_unit;
+        while (output < unit_end) {
+            u32 color = *input++;
+            Math::Vec4<u8> col_vec{
+                (color >> 24) & 0xFF, (color >> 16) & 0xFF, (color >>  8) & 0xFF, alpha,
+            };
+
+            switch (output_format) {
+            case OutputFormat::RGBA8:
+                Color::EncodeRGBA8(col_vec, output);
+                output += 4;
+                break;
+            case OutputFormat::RGB8:
+                Color::EncodeRGB8(col_vec, output);
+                output += 3;
+                break;
+            case OutputFormat::RGB5A1:
+                Color::EncodeRGB5A1(col_vec, output);
+                output += 2;
+                break;
+            case OutputFormat::RGB565:
+                Color::EncodeRGB565(col_vec, output);
+                output += 2;
+                break;
+            }
+
+            amount_of_data -= 1;
+        }
+
+        output += buf.gap;
+        buf.address += buf.transfer_unit + buf.gap;
+        buf.image_size -= buf.transfer_unit;
+    }
+}
+
+static const u8 linear_lut[64] = {
+     0,  1,  2,  3,  4,  5,  6,  7,
+     8,  9, 10, 11, 12, 13, 14, 15,
+    16, 17, 18, 19, 20, 21, 22, 23,
+    24, 25, 26, 27, 28, 29, 30, 31,
+    32, 33, 34, 35, 36, 37, 38, 39,
+    40, 41, 42, 43, 44, 45, 46, 47,
+    48, 49, 50, 51, 52, 53, 54, 55,
+    56, 57, 58, 59, 60, 61, 62, 63,
+};
+
+static const u8 morton_lut[64] = {
+     0,  1,  4,  5, 16, 17, 20, 21,
+     2,  3,  6,  7, 18, 19, 22, 23,
+     8,  9, 12, 13, 24, 25, 28, 29,
+    10, 11, 14, 15, 26, 27, 30, 31,
+    32, 33, 36, 37, 48, 49, 52, 53,
+    34, 35, 38, 39, 50, 51, 54, 55,
+    40, 41, 44, 45, 56, 57, 60, 61,
+    42, 43, 46, 47, 58, 59, 62, 63,
+};
+
+static void RotateTile0(const ImageTile& input, ImageTile& output, int height, const u8 out_map[64]) {
+    for (int i = 0; i < height * 8; ++i) {
+        output[out_map[i]] = input[i];
+    }
+}
+
+static void RotateTile90(const ImageTile& input, ImageTile& output, int height, const u8 out_map[64]) {
+    int out_i = 0;
+    for (int x = 0; x < 8; ++x) {
+        for (int y = height - 1; y >= 0; --y) {
+            output[out_map[out_i++]] = input[y * 8 + x];
+        }
+    }
+}
+
+static void RotateTile180(const ImageTile& input, ImageTile& output, int height, const u8 out_map[64]) {
+    int out_i = 0;
+    for (int i = height * 8 - 1; i >= 0; --i) {
+        output[out_map[out_i++]] = input[i];
+    }
+}
+
+static void RotateTile270(const ImageTile& input, ImageTile& output, int height, const u8 out_map[64]) {
+    int out_i = 0;
+    for (int x = 8-1; x >= 0; --x) {
+        for (int y = 0; y < height; ++y) {
+            output[out_map[out_i++]] = input[y * 8 + x];
+        }
+    }
+}
+
+static void WriteTileToOutput(u32* output, const ImageTile& tile, int height, int line_stride) {
+    for (int y = 0; y < height; ++y) {
+        for (int x = 0; x < 8; ++x) {
+            output[y * line_stride + x] = tile[y * 8 + x];
+        }
+    }
+}
+
+/**
+ * Performs a Y2R colorspace conversion.
+ *
+ * The Y2R hardware implements hardware-accelerated YUV to RGB colorspace conversions. It is most
+ * commonly used for video playback or to display camera input to the screen.
+ *
+ * The conversion process is quite configurable, and can be divided in distinct steps. From
+ * observation, it appears that the hardware buffers a single 8-pixel tall strip of image data
+ * internally and converts it in one go before writing to the output and loading the next strip.
+ *
+ * The steps taken to convert one strip of image data are:
+ *
+ * - The hardware receives data via CDMA (http://3dbrew.org/wiki/Corelink_DMA_Engines), which is
+ *   presumably stored in one or more internal buffers. This process can be done in several separate
+ *   transfers, as long as they don't exceed the size of the internal image buffer. This allows
+ *   flexibility in input strides.
+ * - The input data is decoded into a YUV tuple. Several formats are suported, see the `InputFormat`
+ *   enum.
+ * - The YUV tuple is converted, using fixed point calculations, to RGB. This step can be configured
+ *   using a set of coefficients to support different colorspace standards. See `CoefficientSet`.
+ * - The strip can be optionally rotated 90, 180 or 270 degrees. Since each strip is processed
+ *   independently, this notably rotates each *strip*, not the entire image. This means that for 90
+ *   or 270 degree rotations, the output will be in terms of several 8 x height images, and for any
+ *   non-zero rotation the strips will have to be re-arranged so that the parts of the image will
+ *   not be shuffled together. This limitation makes this a feature of somewhat dubious utility. 90
+ *   or 270 degree rotations in images with non-even height don't seem to work properly.
+ * - The data is converted to the output RGB format. See the `OutputFormat` enum.
+ * - The data can be output either linearly line-by-line or in the swizzled 8x8 tile format used by
+ *   the PICA. This is decided by the `BlockAlignment` enum. If 8x8 alignment is used, then the
+ *   image must have a height divisible by 8. The image width must always be divisible by 8.
+ * - The final data is then CDMAed out to main memory and the next image strip is processed. This
+ *   offers the same flexibility as the input stage.
+ *
+ * In this implementation, to avoid the combinatorial explosion of parameter combinations, common
+ * intermediate formats are used and where possible tables or parameters are used instead of
+ * diverging code paths to keep the amount of branches in check. Some steps are also merged to
+ * increase efficiency.
+ *
+ * Output for all valid settings combinations matches hardware, however output in some edge-cases
+ * differs:
+ *
+ * - `Block8x8` alignment with non-mod8 height produces different garbage patterns on the last
+ *   strip, especially when combined with rotation.
+ * - Hardware, when using `Linear` alignment with a non-even height and 90 or 270 degree rotation
+ *   produces misaligned output on the last strip. This implmentation produces output with the
+ *   correct "expected" alignment.
+ *
+ * Hardware behaves strangely (doesn't fire the completion interrupt, for example) in these cases,
+ * so they are believed to be invalid configurations anyway.
+ */
+void PerformConversion(ConversionConfiguration& cvt) {
+    ASSERT(cvt.input_line_width % 8 == 0);
+    ASSERT(cvt.block_alignment != BlockAlignment::Block8x8 || cvt.input_lines % 8 == 0);
+    // Tiles per row
+    size_t num_tiles = cvt.input_line_width / 8;
+    ASSERT(num_tiles < MAX_TILES);
+
+    // Buffer used as a CDMA source/target.
+    std::unique_ptr<u8[]> data_buffer(new u8[cvt.input_line_width * 8 * 4]);
+    // Intermediate storage for decoded 8x8 image tiles. Always stored as RGB32.
+    std::unique_ptr<ImageTile[]> tiles(new ImageTile[num_tiles]);
+    ImageTile tmp_tile;
+
+    // LUT used to remap writes to a tile. Used to allow linear or swizzled output without
+    // requiring two different code paths.
+    const u8* tile_remap;
+    switch (cvt.block_alignment) {
+    case BlockAlignment::Linear:
+        tile_remap = linear_lut; break;
+    case BlockAlignment::Block8x8:
+        tile_remap = morton_lut; break;
+    }
+
+    for (unsigned int y = 0; y < cvt.input_lines; y += 8) {
+        unsigned int row_height = std::min(cvt.input_lines - y, 8u);
+
+        // Total size in pixels of incoming data required for this strip.
+        const size_t row_data_size = row_height * cvt.input_line_width;
+
+        u8* input_Y = data_buffer.get();
+        u8* input_U = input_Y + 8 * cvt.input_line_width;
+        u8* input_V = input_U + 8 * cvt.input_line_width / 2;
+
+        switch (cvt.input_format) {
+        case InputFormat::YUV422_Indiv8:
+            ReceiveData<1>(input_Y, cvt.src_Y, row_data_size);
+            ReceiveData<1>(input_U, cvt.src_U, row_data_size / 2);
+            ReceiveData<1>(input_V, cvt.src_V, row_data_size / 2);
+            break;
+        case InputFormat::YUV420_Indiv8:
+            ReceiveData<1>(input_Y, cvt.src_Y, row_data_size);
+            ReceiveData<1>(input_U, cvt.src_U, row_data_size / 4);
+            ReceiveData<1>(input_V, cvt.src_V, row_data_size / 4);
+            break;
+        case InputFormat::YUV422_Indiv16:
+            ReceiveData<2>(input_Y, cvt.src_Y, row_data_size);
+            ReceiveData<2>(input_U, cvt.src_U, row_data_size / 2);
+            ReceiveData<2>(input_V, cvt.src_V, row_data_size / 2);
+            break;
+        case InputFormat::YUV420_Indiv16:
+            ReceiveData<2>(input_Y, cvt.src_Y, row_data_size);
+            ReceiveData<2>(input_U, cvt.src_U, row_data_size / 4);
+            ReceiveData<2>(input_V, cvt.src_V, row_data_size / 4);
+            break;
+        case InputFormat::YUYV422_Interleaved:
+            input_U = nullptr;
+            input_V = nullptr;
+            ReceiveData<1>(input_Y, cvt.src_YUYV, row_data_size * 2);
+            break;
+        }
+
+        // Note(yuriks): If additional optimization is required, input_format can be moved to a
+        // template parameter, so that its dispatch can be moved to outside the inner loop.
+        ConvertYUVToRGB(cvt.input_format, input_Y, input_U, input_V, tiles.get(),
+                cvt.input_line_width, row_height, cvt.coefficients);
+
+        u32* output_buffer = reinterpret_cast<u32*>(data_buffer.get());
+
+        for (int i = 0; i < num_tiles; ++i) {
+            int image_strip_width, output_stride;
+
+            switch (cvt.rotation) {
+            case Rotation::None:
+                RotateTile0(tiles[i], tmp_tile, row_height, tile_remap);
+                image_strip_width = cvt.input_line_width;
+                output_stride = 8;
+                break;
+            case Rotation::Clockwise_90:
+                RotateTile90(tiles[i], tmp_tile, row_height, tile_remap);
+                image_strip_width = 8;
+                output_stride = 8 * row_height;
+                break;
+            case Rotation::Clockwise_180:
+                // For 180 and 270 degree rotations we also invert the order of tiles in the strip,
+                // since the rotates are done individually on each tile.
+                RotateTile180(tiles[num_tiles - i - 1], tmp_tile, row_height, tile_remap);
+                image_strip_width = cvt.input_line_width;
+                output_stride = 8;
+                break;
+            case Rotation::Clockwise_270:
+                RotateTile270(tiles[num_tiles - i - 1], tmp_tile, row_height, tile_remap);
+                image_strip_width = 8;
+                output_stride = 8 * row_height;
+                break;
+            }
+
+            switch (cvt.block_alignment) {
+            case BlockAlignment::Linear:
+                WriteTileToOutput(output_buffer, tmp_tile, row_height, image_strip_width);
+                output_buffer += output_stride;
+                break;
+            case BlockAlignment::Block8x8:
+                WriteTileToOutput(output_buffer, tmp_tile, 8, 8);
+                output_buffer += TILE_SIZE;
+                break;
+            }
+        }
+
+        // Note(yuriks): If additional optimization is required, output_format can be moved to a
+        // template parameter, so that its dispatch can be moved to outside the inner loop.
+        SendData(reinterpret_cast<u32*>(data_buffer.get()), cvt.dst, (int)row_data_size, cvt.output_format, (u8)cvt.alpha);
+    }
+}
+
+}
+}
diff --git a/src/core/hw/y2r.h b/src/core/hw/y2r.h
new file mode 100644
index 00000000..729e1eee
--- /dev/null
+++ b/src/core/hw/y2r.h
@@ -0,0 +1,15 @@
+// Copyright 2015 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+namespace Y2R_U {
+    struct ConversionConfiguration;
+}
+
+namespace HW {
+namespace Y2R {
+
+void PerformConversion(Y2R_U::ConversionConfiguration& cvt);
+
+}
+}