3D: change viewport handling to match hardware. finally fixes #18

2 files changed, 27 insertions, 30 deletions

diff --git a/src/GPU3D.cpp b/src/GPU3D.cpp
index a746189..25d0476 100644
--- a/src/GPU3D.cpp
+++ b/src/GPU3D.cpp
@@ -203,7 +203,7 @@ s32 TexMatrix[16];
 s32 ClipMatrix[16];
 bool ClipMatrixDirty;
 
-s32 Viewport[4];
+u32 Viewport[6];
 
 s32 ProjMatrixStack[16];
 s32 PosMatrixStack[31][16];
@@ -723,7 +723,6 @@ void SubmitPolygon()
     // clipping
 
     nverts = ClipPolygon<true>(clippedvertices, nverts, clipstart);
-
     if (nverts == 0)
     {
         LastStripPolygon = NULL;
@@ -803,24 +802,12 @@ void SubmitPolygon()
             // the transform for Z uses the normalized W values
             // W normalization is applied to separate polygons, even within strips
 
-            posX = (((s64)(vtx->Position[0] + w) * Viewport[2]) / (((s64)w) << 1)) + Viewport[0];
-            posY = (((s64)(-vtx->Position[1] + w) * Viewport[3]) / (((s64)w) << 1)) + Viewport[1];
-
-            //if (FlushAttributes & 0x2) posZ = w;
-            //else                       posZ = (((s64)vtx->Position[2] * 0x800000) / w) + 0x7FFEFF;
+            posX = (((s64)(vtx->Position[0] + w) * Viewport[4]) / (((s64)w) << 1)) + Viewport[0];
+            posY = (((s64)(-vtx->Position[1] + w) * Viewport[5]) / (((s64)w) << 1)) + Viewport[3];
         }
 
-        if      (posX < 0)        posX = 0;
-        else if (posX > 256)      posX = 256;
-        if      (posY < 0)        posY = 0;
-        else if (posY > 192)      posY = 192;
-        //if      (posZ < 0)        posZ = 0;
-        //else if (posZ > 0xFFFFFF) posZ = 0xFFFFFF;
-
-        vtx->FinalPosition[0] = posX;
-        vtx->FinalPosition[1] = posY;
-        //vtx->FinalPosition[2] = posZ;
-        //vtx->FinalPosition[3] = w;
+        vtx->FinalPosition[0] = posX & 0x1FF;
+        vtx->FinalPosition[1] = posY & 0xFF;
 
         vtx->FinalColor[0] = vtx->Color[0] >> 12;
         if (vtx->FinalColor[0]) vtx->FinalColor[0] = ((vtx->FinalColor[0] << 4) + 0xF);
@@ -1711,10 +1698,12 @@ void ExecuteCommand()
 
         case 0x60: // viewport x1,y1,x2,y2
             // note: viewport Y coordinates are upside-down
-            Viewport[0] = ExecParams[0] & 0xFF;
-            Viewport[1] = 191 - (ExecParams[0] >> 24);
-            Viewport[2] = ((ExecParams[0] >> 16) & 0xFF) - Viewport[0] + 1;
-            Viewport[3] = (191 - ((ExecParams[0] >> 8) & 0xFF)) - Viewport[1] + 1;
+            Viewport[0] = ExecParams[0] & 0xFF;                             // x0
+            Viewport[1] = (191 - ((ExecParams[0] >> 8) & 0xFF)) & 0xFF;     // y0
+            Viewport[2] = (ExecParams[0] >> 16) & 0xFF;                     // x1
+            Viewport[3] = (191 - (ExecParams[0] >> 24)) & 0xFF;             // y1
+            Viewport[4] = (Viewport[2] - Viewport[0] + 1) & 0x1FF;          // width
+            Viewport[5] = (Viewport[1] - Viewport[3] + 1) & 0xFF;           // height
             break;
 
         case 0x70: // box test
diff --git a/src/GPU3D_Soft.cpp b/src/GPU3D_Soft.cpp
index ab7f517..b12ed8c 100644
--- a/src/GPU3D_Soft.cpp
+++ b/src/GPU3D_Soft.cpp
@@ -1049,10 +1049,12 @@ void RenderPolygonScanline(RendererPolygon* rp, s32 y)
     Interpolator interpX(xstart, xend+1, wl, wr, 8);
 
     if (x < 0) x = 0;
+    s32 xlimit;
 
     // part 1: left edge
     edge = yedge | 0x1;
-    for (; x < xstart+l_edgelen; x++)
+    xlimit = xstart+l_edgelen; if (xlimit > 256) xlimit = 256;
+    for (; x < xlimit; x++)
     {
         u32 pixeladdr = FirstPixelOffset + (y*ScanlineWidth) + x;
         u32 attr = (polygon->Attr & 0x3F008000);
@@ -1180,8 +1182,9 @@ void RenderPolygonScanline(RendererPolygon* rp, s32 y)
 
     // part 2: polygon inside
     edge = yedge;
-    if (wireframe && !edge) x = xend-r_edgelen+1;
-    else for (; x <= xend-r_edgelen; x++)
+    xlimit = xend-r_edgelen+1; if (xlimit > 256) xlimit = 256;
+    if (wireframe && !edge) x = xlimit;
+    else for (; x < xlimit; x++)
     {
         u32 pixeladdr = FirstPixelOffset + (y*ScanlineWidth) + x;
         u32 attr = (polygon->Attr & 0x3F008000);
@@ -1268,11 +1271,10 @@ void RenderPolygonScanline(RendererPolygon* rp, s32 y)
         AttrBuffer[pixeladdr] = attr;
     }
 
-    if (xend > 255) xend = 255;
-
     // part 3: right edge
     edge = yedge | 0x2;
-    for (; x <= xend; x++)
+    xlimit = xend+1; if (xlimit > 256) xlimit = 256;
+    for (; x < xlimit; x++)
     {
         u32 pixeladdr = FirstPixelOffset + (y*ScanlineWidth) + x;
         u32 attr = (polygon->Attr & 0x3F008000);
@@ -1679,23 +1681,29 @@ void RenderPolygons(bool threaded, Polygon* polygons, int npolys)
     // TODO: Y-sorting for translucent polygons
     // TODO: all sorting should be done in GPU3D.cpp
 
+    // polygons with ybottom>192 aren't rendered at all
+
     int j = 0;
     for (int i = 0; i < npolys; i++)
     {
         if (polygons[i].Translucent) continue;
+
+        if (polygons[i].YBottom > 192) continue;
         SetupPolygon(&PolygonList[j++], &polygons[i]);
     }
     for (int i = 0; i < npolys; i++)
     {
         if (!polygons[i].Translucent) continue;
+
+        if (polygons[i].YBottom > 192) continue;
         SetupPolygon(&PolygonList[j++], &polygons[i]);
     }
 
-    RenderScanline(0, npolys);
+    RenderScanline(0, j);
 
     for (s32 y = 1; y < 192; y++)
     {
-        RenderScanline(y, npolys);
+        RenderScanline(y, j);
         ScanlineFinalPass(y-1);
 
         if (threaded)