Merge 1670c806f1 into 10798c3464

make sure we dont forget that this is still implemented wrong.

src/DSi.cpp

@@ -3173,4 +3173,9 @@ void DSi::ARM7IOWrite32(u32 addr, u32 val)
     return NDS::ARM7IOWrite32(addr, val);
 }
 
+bool DSi::GetSCFGRasterBit() const
+{
+    return SCFG_EXT[0] & (1<<2);
+}
+
 }

src/DSi.h

@@ -129,6 +129,8 @@ public:
     void ARM7IOWrite16(u32 addr, u16 val) override;
     void ARM7IOWrite32(u32 addr, u32 val) override;
 
+    bool GetSCFGRasterBit() const override;
+
 public:
     DSi(DSiArgs&& args) noexcept;
     ~DSi() noexcept override;

src/GPU3D.cpp

@@ -182,6 +182,8 @@ void GPU3D::ResetRenderingState() noexcept
 
     RenderClearAttr1 = 0x3F000000;
     RenderClearAttr2 = 0x00007FFF;
+
+    RenderRasterRev = false; // CHECKME: when should this be reset?
 }
 
 void GPU3D::Reset() noexcept
@@ -354,6 +356,8 @@ void GPU3D::DoSavestate(Savestate* file) noexcept
 
     file->Var32(&RenderClearAttr1);
     file->Var32(&RenderClearAttr2);
+    
+    file->Bool32(&RenderRasterRev);
 
     file->Var16(&RenderXPos);
 
@@ -553,6 +557,7 @@ void GPU3D::DoSavestate(Savestate* file) noexcept
     file->Var32(&TexParam);
     file->Var32(&TexPalette);
     RenderFrameIdentical = false;
+    ForceRerender = false;
     if (softRenderer && softRenderer->IsThreaded())
     {
         softRenderer->EnableRenderThread();
@@ -1216,7 +1221,7 @@ void GPU3D::SubmitPolygon() noexcept
 
     u32 texfmt = (TexParam >> 26) & 0x7;
     u32 polyalpha = (CurPolygonAttr >> 16) & 0x1F;
-    poly->Translucent = ((texfmt == 1 || texfmt == 6) && !(CurPolygonAttr & 0x10)) || (polyalpha > 0 && polyalpha < 31);
+    poly->Translucent = (texfmt == 1 || texfmt == 6) || (polyalpha > 0 && polyalpha < 31);
 
     poly->IsShadowMask = ((CurPolygonAttr & 0x3F000030) == 0x00000030);
     poly->IsShadow = ((CurPolygonAttr & 0x30) == 0x30) && !poly->IsShadowMask;
@@ -2459,18 +2464,30 @@ void GPU3D::VBlank() noexcept
 
                 RenderNumPolygons = NumPolygons;
                 RenderFrameIdentical = false;
+                ForceRerender = 0;
             }
             else
             {
-                RenderFrameIdentical = RenderDispCnt == DispCnt
-                    && RenderAlphaRef == AlphaRef
-                    && RenderClearAttr1 == ClearAttr1
-                    && RenderClearAttr2 == ClearAttr2
-                    && RenderFogColor == FogColor
-                    && RenderFogOffset == FogOffset * 0x200
-                    && memcmp(RenderEdgeTable, EdgeTable, 8*2) == 0
-                    && memcmp(RenderFogDensityTable + 1, FogDensityTable, 32) == 0
-                    && memcmp(RenderToonTable, ToonTable, 32*2) == 0;
+                
+                if (ForceRerender == true)
+                {
+                    RenderFrameIdentical = false;
+                    ForceRerender = false;
+                    DontRerenderLoop = true;
+                }
+                else
+                {
+                    RenderFrameIdentical = RenderDispCnt == DispCnt
+                        && RenderAlphaRef == AlphaRef
+                        && RenderClearAttr1 == ClearAttr1
+                        && RenderClearAttr2 == ClearAttr2
+                        && RenderFogColor == FogColor
+                        && RenderFogOffset == FogOffset * 0x200
+                        && memcmp(RenderEdgeTable, EdgeTable, 8*2) == 0
+                        && memcmp(RenderFogDensityTable + 1, FogDensityTable, 32) == 0
+                        && memcmp(RenderToonTable, ToonTable, 32*2) == 0
+                        && RenderRasterRev == NDS.GetSCFGRasterBit();
+                }
             }
 
             RenderDispCnt = DispCnt;
@@ -2488,6 +2505,9 @@ void GPU3D::VBlank() noexcept
 
             RenderClearAttr1 = ClearAttr1;
             RenderClearAttr2 = ClearAttr2;
+            // NOTE: this is not latched. it updates the *moment* you set it. it does not cancel the current scanline. it just causes pain and suffering.
+            // in less dramatic terms: annoying to emulate properly, so we're just gonna pretend it's latched for now, ok?
+            RenderRasterRev = NDS.GetSCFGRasterBit();
         }
 
         if (FlushRequest)

src/GPU3D.h

@@ -271,6 +271,8 @@ public:
     u32 RenderClearAttr1 = 0;
     u32 RenderClearAttr2 = 0;
 
+    bool RenderRasterRev = false;
+
     bool RenderFrameIdentical = false; // not part of the hardware state, don't serialize
 
     bool AbortFrame = false;
@@ -323,6 +325,10 @@ public:
     std::array<Polygon*,2048> RenderPolygonRAM {};
     u32 RenderNumPolygons = 0;
 
+    // used to fix stencil buffer's frame-to-frame persistence not working properly under extreme misuse of shadow masks/shadows
+    bool ForceRerender = false;
+    bool DontRerenderLoop = false;
+
     u32 FlushRequest = 0;
     u32 FlushAttributes = 0;
     u32 ScrolledLine[256]; // not part of the hardware state, don't serialize

src/GPU3D_Soft.h

@@ -26,6 +26,18 @@
 
 namespace melonDS
 {
+
+enum EdgeFlags
+{
+    EF_None      = (0),
+    EF_AnyEdge   = (0xF),
+
+    EF_TopXMajor = (1<<0),
+    EF_BotXMajor = (1<<1),
+    EF_LYMajor   = (1<<2),
+    EF_RYMajor   = (1<<3),
+};
+
 class SoftRenderer : public Renderer3D
 {
 public:
@@ -455,19 +467,20 @@ private:
     };
 
     RendererPolygon PolygonList[2048];
-    void TextureLookup(const GPU& gpu, u32 texparam, u32 texpal, s16 s, s16 t, u16* color, u8* alpha) const;
+    template <bool colorcorrect> inline void ColorConv(const u16 color, u8* r, u8* g, u8* b) const;
+    void TextureLookup(const GPU& gpu, u32 texparam, u32 texpal, s16 s, s16 t, u8* tr, u8* tg, u8* tb, u8* alpha) const;
     u32 RenderPixel(const GPU& gpu, const Polygon* polygon, u8 vr, u8 vg, u8 vb, s16 s, s16 t) const;
     void PlotTranslucentPixel(const GPU3D& gpu3d, u32 pixeladdr, u32 color, u32 z, u32 polyattr, u32 shadow);
     void SetupPolygonLeftEdge(RendererPolygon* rp, s32 y) const;
     void SetupPolygonRightEdge(RendererPolygon* rp, s32 y) const;
     void SetupPolygon(RendererPolygon* rp, Polygon* polygon) const;
     void RenderShadowMaskScanline(const GPU3D& gpu3d, RendererPolygon* rp, s32 y);
-    void RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s32 y);
-    void RenderScanline(const GPU& gpu, s32 y, int npolys);
+    void RenderPolygonScanline(GPU& gpu, RendererPolygon* rp, s32 y);
+    void RenderScanline(GPU& gpu, s32 y, int npolys);
     u32 CalculateFogDensity(const GPU3D& gpu3d, u32 pixeladdr) const;
     void ScanlineFinalPass(const GPU3D& gpu3d, s32 y);
     void ClearBuffers(const GPU& gpu);
-    void RenderPolygons(const GPU& gpu, bool threaded, Polygon** polygons, int npolys);
+    void RenderPolygons(GPU& gpu, bool threaded, Polygon** polygons, int npolys);
 
     void RenderThreadFunc(GPU& gpu);
 
@@ -487,7 +500,7 @@ private:
     u32 AttrBuffer[BufferSize * 2];
 
     // attribute buffer:
-    // bit0-3: edge flags (left/right/top/bottom)
+    // bit0-3: edge flags (top xmajor/bottom xmajor/left ymajor/right ymajor)
     // bit4: backfacing flag
     // bit8-12: antialiasing alpha
     // bit15: fog enable
@@ -496,7 +509,8 @@ private:
     // bit24-29: polygon ID for opaque pixels
 
     u8 StencilBuffer[256*2];
-    bool PrevIsShadowMask;
+    bool ShadowRendered[2];
+    bool StencilCleared;
 
     bool Enabled;
 

src/NDS.cpp

@@ -4334,4 +4334,9 @@ void NDS::ARM7IOWrite32(u32 addr, u32 val)
     Log(LogLevel::Debug, "unknown ARM7 IO write32 %08X %08X %08X\n", addr, val, ARM7.R[15]);
 }
 
+bool NDS::GetSCFGRasterBit() const
+{
+    return false;
+}
+
 }

src/NDS.h

@@ -471,6 +471,8 @@ public: // TODO: Encapsulate the rest of these members
     void SetJITArgs(std::optional<JITArgs> args) noexcept {}
 #endif
 
+    virtual bool GetSCFGRasterBit() const;
+   
 private:
     void InitTimings();
     u32 SchedListMask;