7 files changed, 76 insertions, 61 deletions

diff --git a/src/citra_qt/debugger/graphics_framebuffer.cpp b/src/citra_qt/debugger/graphics_framebuffer.cpp
index 2985cadb1..d621d7204 100644
--- a/src/citra_qt/debugger/graphics_framebuffer.cpp
+++ b/src/citra_qt/debugger/graphics_framebuffer.cpp
@@ -176,7 +176,7 @@ void GraphicsFramebufferWidget::OnUpdate()
     {
         // TODO: Store a reference to the registers in the debug context instead of accessing them directly...
 
-        auto framebuffer = Pica::registers.framebuffer;
+        const auto& framebuffer = Pica::registers.framebuffer;
 
         framebuffer_address = framebuffer.GetColorBufferPhysicalAddress();
         framebuffer_width = framebuffer.GetWidth();
@@ -189,7 +189,7 @@ void GraphicsFramebufferWidget::OnUpdate()
 
     case Source::DepthBuffer:
     {
-        auto framebuffer = Pica::registers.framebuffer;
+        const auto& framebuffer = Pica::registers.framebuffer;
 
         framebuffer_address = framebuffer.GetDepthBufferPhysicalAddress();
         framebuffer_width = framebuffer.GetWidth();
@@ -294,4 +294,4 @@ u32 GraphicsFramebufferWidget::BytesPerPixel(GraphicsFramebufferWidget::Format f
         case Format::D16:
             return 2;
     }
-}
\ No newline at end of file
+}
diff --git a/src/citra_qt/debugger/graphics_framebuffer.h b/src/citra_qt/debugger/graphics_framebuffer.h
index dff91d131..4cb396ffe 100644
--- a/src/citra_qt/debugger/graphics_framebuffer.h
+++ b/src/citra_qt/debugger/graphics_framebuffer.h
@@ -20,9 +20,9 @@ class GraphicsFramebufferWidget : public BreakPointObserverDock {
     using Event = Pica::DebugContext::Event;
 
     enum class Source {
-        PicaTarget = 0,
-        DepthBuffer = 1,
-        Custom = 2,
+        PicaTarget   = 0,
+        DepthBuffer  = 1,
+        Custom       = 2,
 
         // TODO: Add GPU framebuffer sources!
     };
diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp
index b7102b874..424ce2ca7 100644
--- a/src/core/hw/gpu.cpp
+++ b/src/core/hw/gpu.cpp
@@ -81,9 +81,9 @@ inline void Write(u32 addr, const T data) {
             if (config.fill_24bit) {
                 // fill with 24-bit values
                 for (u8* ptr = start; ptr < end; ptr += 3) {
-                    ptr[0] = config.value_24bit_r;
+                    ptr[0] = config.value_24bit_b;
                     ptr[1] = config.value_24bit_g;
-                    ptr[2] = config.value_24bit_b;
+                    ptr[2] = config.value_24bit_r;
                 }
             } else if (config.fill_32bit) {
                 // fill with 32-bit values
diff --git a/src/core/hw/gpu.h b/src/core/hw/gpu.h
index 5ca4a5450..737b1e968 100644
--- a/src/core/hw/gpu.h
+++ b/src/core/hw/gpu.h
@@ -100,10 +100,10 @@ struct Regs {
             // Set to 1 upon completion.
             BitField<0, 1, u32> finished;
 
-            // If both of these bits are unset, then it will fill the memory with a 16 bit value
-            // 1: fill with 24-bit wide values
+            // 0: fill with 16- or 32-bit wide values; 1: fill with 24-bit wide values
             BitField<8, 1, u32> fill_24bit;
-            // 1: fill with 32-bit wide values
+
+            // 0: fill with 16-bit wide values; 1: fill with 32-bit wide values
             BitField<9, 1, u32> fill_32bit;
         };
 
diff --git a/src/video_core/color.h b/src/video_core/color.h
index 35b56efc0..14ade74f2 100644
--- a/src/video_core/color.h
+++ b/src/video_core/color.h
@@ -105,7 +105,7 @@ inline const Math::Vec4<u8> DecodeRGBA4(const u8* bytes) {
  * @param bytes Pointer to encoded source value
  * @return Depth value as an u32
  */
-inline const u32 DecodeD16(const u8* bytes) {
+inline u32 DecodeD16(const u8* bytes) {
     return *reinterpret_cast<const u16_le*>(bytes);
 }
 
@@ -114,7 +114,7 @@ inline const u32 DecodeD16(const u8* bytes) {
  * @param bytes Pointer to encoded source value
  * @return Depth value as an u32
  */
-inline const u32 DecodeD24(const u8* bytes) {
+inline u32 DecodeD24(const u8* bytes) {
     return (bytes[2] << 16) | (bytes[1] << 8) | bytes[0];
 }
 
@@ -181,8 +181,8 @@ inline void EncodeRGBA4(const Math::Vec4<u8>& color, u8* bytes) {
 }
 
 /**
- * Encode a depth value as D16 format
- * @param value Source depth value to encode
+ * Encode a 16 bit depth value as D16 format
+ * @param value 16 bit source depth value to encode
  * @param bytes Pointer where to store the encoded value
  */
 inline void EncodeD16(u32 value, u8* bytes) {
@@ -190,8 +190,8 @@ inline void EncodeD16(u32 value, u8* bytes) {
 }
 
 /**
- * Encode a depth value as D24 format
- * @param value Source depth value to encode
+ * Encode a 24 bit depth value as D24 format
+ * @param value 24 bit source depth value to encode
  * @param bytes Pointer where to store the encoded value
  */
 inline void EncodeD24(u32 value, u8* bytes) {
@@ -201,9 +201,9 @@ inline void EncodeD24(u32 value, u8* bytes) {
 }
 
 /**
- * Encode depth and stencil values as D24S8 format
- * @param depth Source depth values to encode
- * @param stencil Source stencil value to encode
+ * Encode a 24 bit depth and 8 bit stencil values as D24S8 format
+ * @param depth 24 bit source depth value to encode
+ * @param stencil 8 bit source stencil value to encode
  * @param bytes Pointer where to store the encoded value
  */
 inline void EncodeD24S8(u32 depth, u8 stencil, u8* bytes) {
diff --git a/src/video_core/pica.h b/src/video_core/pica.h
index 6549693f5..fe20cd77d 100644
--- a/src/video_core/pica.h
+++ b/src/video_core/pica.h
@@ -393,7 +393,15 @@ struct Regs {
             BitField< 8, 8, u32> ref;
         } alpha_test;
 
-        INSERT_PADDING_WORDS(0x2);
+        union {
+            BitField< 0, 1, u32> stencil_test_enable;
+            BitField< 4, 3, CompareFunc> stencil_test_func;
+            BitField< 8, 8, u32> stencil_replacement_value;
+            BitField<16, 8, u32> stencil_reference_value;
+            BitField<24, 8, u32> stencil_mask;
+        } stencil_test;
+
+        INSERT_PADDING_WORDS(0x1);
 
         union {
             BitField< 0, 1, u32> depth_test_enable;
@@ -408,6 +416,30 @@ struct Regs {
         INSERT_PADDING_WORDS(0x8);
     } output_merger;
 
+    enum DepthFormat : u32 {
+        D16    = 0,
+
+        D24    = 2,
+        D24S8  = 3
+    };
+
+    /*
+     * Returns the number of bytes in the specified depth format
+     */
+    static u32 BytesPerDepthPixel(DepthFormat format) {
+        switch (format) {
+        case DepthFormat::D16:
+            return 2;
+        case DepthFormat::D24:
+            return 3;
+        case DepthFormat::D24S8:
+            return 4;
+        default:
+            LOG_CRITICAL(HW_GPU, "Unknown depth format %u", format);
+            UNIMPLEMENTED();
+        }
+    }
+
     struct {
         // Components are laid out in reverse byte order, most significant bits first.
         enum ColorFormat : u32 {
@@ -418,16 +450,9 @@ struct Regs {
             RGBA4    = 4,
         };
 
-        enum DepthFormat : u32 {
-            D16     = 0,
-        
-            D24     = 2,
-            D24S8   = 3
-        };
-
         INSERT_PADDING_WORDS(0x6);
 
-        u32 depth_format;
+        DepthFormat depth_format;
         BitField<16, 3, u32> color_format;
 
         INSERT_PADDING_WORDS(0x4);
diff --git a/src/video_core/rasterizer.cpp b/src/video_core/rasterizer.cpp
index dc32128c6..dd46f0ec3 100644
--- a/src/video_core/rasterizer.cpp
+++ b/src/video_core/rasterizer.cpp
@@ -100,24 +100,19 @@ static u32 GetDepth(int x, int y) {
     y = (registers.framebuffer.height - y);
     
     const u32 coarse_y = y & ~7;
+    u32 bytes_per_pixel = Pica::Regs::BytesPerDepthPixel(registers.framebuffer.depth_format);
+    u32 stride = registers.framebuffer.width * bytes_per_pixel;
+
+    u32 src_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
+    u8* src_pixel = depth_buffer + src_offset;
 
     switch (registers.framebuffer.depth_format) {
-        case registers.framebuffer.D16:
-        {
-            u32 stride = registers.framebuffer.width * 2;
-            return Color::DecodeD16(depth_buffer + VideoCore::GetMortonOffset(x, y, 2) + coarse_y * stride);
-        }
-        case registers.framebuffer.D24:
-        {
-            u32 stride = registers.framebuffer.width * 3;
-            u8* address = depth_buffer + VideoCore::GetMortonOffset(x, y, 3) + coarse_y * stride;
-            return Color::DecodeD24(address);
-        }
-        case registers.framebuffer.D24S8:
-        {
-            u32 stride = registers.framebuffer.width * 4;
-            return Color::DecodeD24S8(depth_buffer + VideoCore::GetMortonOffset(x, y, 4) + coarse_y * stride).x;
-        }
+        case Pica::Regs::DepthFormat::D16:
+            return Color::DecodeD16(src_pixel);
+        case Pica::Regs::DepthFormat::D24:
+            return Color::DecodeD24(src_pixel);
+        case Pica::Regs::DepthFormat::D24S8:
+            return Color::DecodeD24S8(src_pixel).x;
         default:
             LOG_CRITICAL(HW_GPU, "Unimplemented depth format %u", registers.framebuffer.depth_format);
             UNIMPLEMENTED();
@@ -132,28 +127,23 @@ static void SetDepth(int x, int y, u32 value) {
     y = (registers.framebuffer.height - y);
 
     const u32 coarse_y = y & ~7;
+    u32 bytes_per_pixel = Pica::Regs::BytesPerDepthPixel(registers.framebuffer.depth_format);
+    u32 stride = registers.framebuffer.width * bytes_per_pixel;
+
+    u32 dst_offset = VideoCore::GetMortonOffset(x, y, bytes_per_pixel) + coarse_y * stride;
+    u8* dst_pixel = depth_buffer + dst_offset;
 
     switch (registers.framebuffer.depth_format) {
-        case registers.framebuffer.D16:
-        {
-            u32 stride = registers.framebuffer.width * 2;
-            Color::EncodeD16(value, depth_buffer + VideoCore::GetMortonOffset(x, y, 2) + coarse_y * stride);
+        case Pica::Regs::DepthFormat::D16:
+            Color::EncodeD16(value, dst_pixel);
             break;
-        }
-        case registers.framebuffer.D24:
-        {
-            u32 stride = registers.framebuffer.width * 3;
-            u8* address = depth_buffer + VideoCore::GetMortonOffset(x, y, 3) + coarse_y * stride;
-            Color::EncodeD24(value, address);
+        case Pica::Regs::DepthFormat::D24:
+            Color::EncodeD24(value, dst_pixel);
             break;
-        }
-        case registers.framebuffer.D24S8:
-        {
-            u32 stride = registers.framebuffer.width * 4;
+        case Pica::Regs::DepthFormat::D24S8:
             // TODO(Subv): Implement the stencil buffer
-            Color::EncodeD24S8(value, 0, depth_buffer + VideoCore::GetMortonOffset(x, y, 4) + coarse_y * stride);
+            Color::EncodeD24S8(value, 0, dst_pixel);
             break;
-        }
         default:
             LOG_CRITICAL(HW_GPU, "Unimplemented depth format %u", registers.framebuffer.depth_format);
             UNIMPLEMENTED();