Merge 3268f94f4b into 5e8beb3ab7

why did i optimize swapped polygons?
they're not that common.

BUILD.md

@@ -47,7 +47,7 @@
      1. Install Qt: `pacman -S <prefix>-{qt6-base,qt6-svg,qt6-multimedia,qt6-svg,qt6-tools}`
      2. Set up the build directory with `cmake -B build`
    * Static builds (without DLLs, standalone executable)
-     1. Install Qt: `pacman -S <prefi>-qt5-static`  
+     1. Install Qt: `pacman -S <prefix>-qt5-static`  
         (Note: As of writing, the `qt6-static` package does not work.)
      2. Set up the build directory with `cmake -B build -DBUILD_STATIC=ON -DUSE_QT6=OFF -DCMAKE_PREFIX_PATH=$MSYSTEM_PREFIX/qt5-static`
 7. Compile: `cmake --build build`

src/GPU3D_Soft.cpp

@@ -774,8 +774,8 @@ void SoftRenderer::RenderShadowMaskScanline(const GPU3D& gpu3d, RendererPolygon*
         interp_start = &rp->SlopeR.Interp;
         interp_end = &rp->SlopeL.Interp;
 
-        rp->SlopeR.EdgeParams<true>(&l_edgelen, &l_edgecov);
+        rp->SlopeR.EdgeParams(&l_edgelen, &l_edgecov, true);
-        rp->SlopeL.EdgeParams<true>(&r_edgelen, &r_edgecov);
+        rp->SlopeL.EdgeParams(&r_edgelen, &r_edgecov, true);
 
         std::swap(xstart, xend);
         std::swap(wl, wr);
@@ -806,8 +806,8 @@ void SoftRenderer::RenderShadowMaskScanline(const GPU3D& gpu3d, RendererPolygon*
         interp_start = &rp->SlopeL.Interp;
         interp_end = &rp->SlopeR.Interp;
 
-        rp->SlopeL.EdgeParams<false>(&l_edgelen, &l_edgecov);
+        rp->SlopeL.EdgeParams(&l_edgelen, &l_edgecov, false);
-        rp->SlopeR.EdgeParams<false>(&r_edgelen, &r_edgecov);
+        rp->SlopeR.EdgeParams(&r_edgelen, &r_edgecov, false);
         
         // CHECKME: edge fill rules for unswapped opaque shadow mask polygons
         if ((gpu3d.RenderDispCnt & ((1<<4)|(1<<5))) || ((polyalpha < 31) && (gpu3d.RenderDispCnt & (1<<3))) || wireframe)
@@ -1003,8 +1003,8 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
         interp_start = &rp->SlopeR.Interp;
         interp_end = &rp->SlopeL.Interp;
 
-        rp->SlopeR.EdgeParams<true>(&l_edgelen, &l_edgecov);
+        rp->SlopeR.EdgeParams(&l_edgelen, &l_edgecov, true);
-        rp->SlopeL.EdgeParams<true>(&r_edgelen, &r_edgecov);
+        rp->SlopeL.EdgeParams(&r_edgelen, &r_edgecov, true);
 
         std::swap(xstart, xend);
         std::swap(wl, wr);
@@ -1041,8 +1041,8 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
         interp_start = &rp->SlopeL.Interp;
         interp_end = &rp->SlopeR.Interp;
 
-        rp->SlopeL.EdgeParams<false>(&l_edgelen, &l_edgecov);
+        rp->SlopeL.EdgeParams(&l_edgelen, &l_edgecov, false);
-        rp->SlopeR.EdgeParams<false>(&r_edgelen, &r_edgecov);
+        rp->SlopeR.EdgeParams(&r_edgelen, &r_edgecov, false);
 
         // edge fill rules for unswapped opaque edges:
         // * right edge is filled if slope > 1
@@ -1096,18 +1096,11 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
     if (x < 0) x = 0;
     s32 xlimit;
 
-    s32 xcov = 0;
-
     // part 1: left edge
     edge = yedge | 0x1;
     xlimit = xstart+l_edgelen;
     if (xlimit > xend+1) xlimit = xend+1;
     if (xlimit > 256) xlimit = 256;
-    if (l_edgecov & (1<<31))
-    {
-        xcov = (l_edgecov >> 12) & 0x3FF;
-        if (xcov == 0x3FF) xcov = 0;
-    }
 
     if (!l_filledge) x = xlimit;
     else
@@ -1166,14 +1159,12 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
                 // anti-aliasing: all edges are rendered
 
                 // calculate coverage
-                s32 cov = l_edgecov;
+                attr |= (l_edgecov >> 5) << 8;
-                if (cov & (1<<31))
+                if (x < xlimit-1)
                 {
-                    cov = xcov >> 5;
+                    l_edgecov += rp->SlopeL.XCov_Incr;
-                    if (cov > 31) cov = 31;
+                    if (l_edgecov > 0x3FF) l_edgecov = 0x3FF;
-                    xcov += (l_edgecov & 0x3FF);
                 }
-                attr |= (cov << 8);
 
                 // push old pixel down if needed
                 if (pixeladdr < BufferSize)
@@ -1292,11 +1283,6 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
     edge = yedge | 0x2;
     xlimit = xend+1;
     if (xlimit > 256) xlimit = 256;
-    if (r_edgecov & (1<<31))
-    {
-        xcov = (r_edgecov >> 12) & 0x3FF;
-        if (xcov == 0x3FF) xcov = 0;
-    }
 
     if (r_filledge)
     for (; x < xlimit; x++)
@@ -1354,14 +1340,12 @@ void SoftRenderer::RenderPolygonScanline(const GPU& gpu, RendererPolygon* rp, s3
                 // anti-aliasing: all edges are rendered
 
                 // calculate coverage
-                s32 cov = r_edgecov;
+                attr |= (r_edgecov >> 5) << 8;
-                if (cov & (1<<31))
+                if (x < xlimit-1)
                 {
-                    cov = 0x1F - (xcov >> 5);
+                    r_edgecov -= rp->SlopeR.XCov_Incr;
-                    if (cov < 0) cov = 0;
+                    if (r_edgecov < 0) r_edgecov = 0;
-                    xcov += (r_edgecov & 0x3FF);
                 }
-                attr |= (cov << 8);
 
                 // push old pixel down if needed
                 if (pixeladdr < BufferSize)

src/GPU3D_Soft.h

@@ -238,6 +238,7 @@ private:
             dx = 0;
 
             this->x0 = x0;
+            this->y0 = y0;
             this->xmin = x0;
             this->xmax = x0;
 
@@ -247,7 +248,7 @@ private:
             Interp.Setup(0, 0, 0, 0);
             Interp.SetX(0);
 
-            xcov_incr = 0;
+            XCov_Incr = 0;
 
             return x0;
         }
@@ -255,6 +256,7 @@ private:
         constexpr s32 Setup(s32 x0, s32 x1, s32 y0, s32 y1, s32 w0, s32 w1, s32 y)
         {
             this->x0 = x0;
+            this->y0 = y0;
             this->y = y;
 
             if (x1 > x0)
@@ -322,7 +324,7 @@ private:
             Interp.SetX(y);
 
             // used for calculating AA coverage
-            if (XMajor) xcov_incr = (ylen << 10) / xlen;
+            if (XMajor) XCov_Incr = ((ylen << 10) / xlen) & 0x3FF;
 
             return x;
         }
@@ -348,18 +350,14 @@ private:
             return ret;
         }
         
-        template<bool swapped>
+        constexpr void EdgeParams_XMajor(s32* length, s32* coverage, bool swapped) const
-        constexpr void EdgeParams_XMajor(s32* length, s32* coverage) const
         {
             // only do length calc for right side when swapped as it's
             // only needed for aa calcs, as actual line spans are broken
-            if constexpr (!swapped || side)
+            if (side ^ Negative)
-            {
+                *length = (dx >> 18) - ((dx-Increment) >> 18);
-                if (side ^ Negative)
+            else
-                    *length = (dx >> 18) - ((dx-Increment) >> 18);
+                *length = ((dx+Increment) >> 18) - (dx >> 18);
-                else
-                    *length = ((dx+Increment) >> 18) - (dx >> 18);
-            }
 
             // for X-major edges, we return the coverage
             // for the first pixel, and the increment for
@@ -369,13 +367,20 @@ private:
             if (side)     startx = startx - *length + 1;
 
             s32 startcov = (((startx << 10) + 0x1FF) * ylen) / xlen;
-            *coverage = (1<<31) | ((startcov & 0x3FF) << 12) | (xcov_incr & 0x3FF);
 
-            if constexpr (swapped) *length = 1;
+            // fix the y value for negative slopes
+            s32 ycoord = Negative ? (ylen << 10) - startcov >> 10 : startcov >> 10;
+            // if yvalue is not equal to actual y value, invert coverage value
+            startcov &= 0x3FF;
+            if (ycoord != y - y0) startcov = 0x3FF - startcov;
+            if (side ^ swapped) startcov = 0x3FF - startcov;
+
+            *coverage = startcov;
+
+            if (swapped) *length = 1;
         }
 
-        template<bool swapped>
+        constexpr void EdgeParams_YMajor(s32* length, s32* coverage, bool swapped) const
-        constexpr void EdgeParams_YMajor(s32* length, s32* coverage) const
         {
             *length = 1;
 
@@ -383,51 +388,39 @@ private:
             {
                 // for some reason vertical edges' aa values
                 // are inverted too when the edges are swapped
-                if constexpr (swapped)
+                *coverage = swapped ? 0 : 31 << 5;
-                    *coverage = 0;
-                else
-                    *coverage = 31;
             }
             else
             {
                 s32 cov = ((dx >> 9) + (Increment >> 10)) >> 4;
                 if ((cov >> 5) != (dx >> 18)) cov = 31;
                 cov &= 0x1F;
-                if constexpr (swapped)
+                if (side ^ !Negative ^ swapped) cov = 0x1F - cov;
-                {
-                    if (side ^ Negative) cov = 0x1F - cov;
-                }
-                else
-                {
-                    if (!(side ^ Negative)) cov = 0x1F - cov;
-                }
 
-                *coverage = cov;
+                // shift left 5 just to make it align with xmajor coverage values
+                *coverage = cov << 5;
             }
         }
         
-        template<bool swapped>
+        constexpr void EdgeParams(s32* length, s32* coverage, bool swapped) const
-        constexpr void EdgeParams(s32* length, s32* coverage) const
         {
             if (XMajor)
-                return EdgeParams_XMajor<swapped>(length, coverage);
+                return EdgeParams_XMajor(length, coverage, swapped);
             else
-                return EdgeParams_YMajor<swapped>(length, coverage);
+                return EdgeParams_YMajor(length, coverage, swapped);
         }
 
         s32 Increment;
         bool Negative;
         bool XMajor;
         Interpolator<1> Interp;
+        s32 XCov_Incr;
 
     private:
-        s32 x0, xmin, xmax;
+        s32 x0, y0, xmin, xmax;
         s32 xlen, ylen;
         s32 dx;
         s32 y;
-
-        s32 xcov_incr;
-        s32 ycoverage, ycov_incr;
     };
 
     u32 AlphaBlend(const GPU3D& gpu3d, u32 srccolor, u32 dstcolor, u32 alpha) const noexcept;