Depth buffering and texture coords

1 files changed, 128 insertions, 74 deletions

diff --git a/src/swgfx.c b/src/swgfx.c
index 54f7432..c298fb4 100644
--- a/src/swgfx.c
+++ b/src/swgfx.c
@@ -22,7 +22,6 @@ Coordinate systems:
 static constexpr sgVec3 Up3 = (sgVec3){0,1,0};
 
 typedef struct sgViewport_t { int x0, y0, width, height; } sgViewport_t;
-typedef struct sgTri2       { sgVec2 p0, p1, p2; }         sgTri2;
 typedef struct sgAABB2      { sgVec2 pmin, pmax; }         sgAABB2;
 
 // Column-major math, column-major storage.
@@ -33,6 +32,7 @@ typedef struct sgMat4 {
 typedef struct swgfx {
   sgVec2i      dims;     // Colour buffer dimensions.
   sgPixel*     colour;   // Colour buffer.
+  R*           depth;    // Depth  buffer.
   sgViewport_t viewport;
   sgMat4       model;    // Model matrix.
   sgMat4       view;     // View matrix.
@@ -48,15 +48,25 @@ typedef struct swgfx {
   sgMat4       mvp;      // Model-view-projection matrix.
 } swgfx;
 
-static inline sgVec3 neg3(sgVec3 v) { return (sgVec3){-v.x, -v.y, -v.z}; }
+static inline R   rmin(R   a, R   b) { return (a <= b) ? a : b; }
+static inline R   rmax(R   a, R   b) { return (a >= b) ? a : b; }
+static inline int imin(int a, int b) { return (a <= b) ? a : b; }
+static inline int imax(int a, int b) { return (a >= b) ? a : b; }
 
-static inline sgVec3 sub3(sgVec3 a, sgVec3 b) {
-  return (sgVec3){a.x - b.x, a.y - b.y, a.z - b.z};
-}
+static inline sgVec2  min2(sgVec2 a, sgVec2 b) { return (sgVec2){.x = rmin(a.x, b.x), .y = rmin(a.y, b.y) }; }
+static inline sgVec2  max2(sgVec2 a, sgVec2 b) { return (sgVec2){.x = rmax(a.x, b.x), .y = rmax(a.y, b.y) }; }
+static inline sgVec2i min2i(sgVec2i a, sgVec2i b) { return (sgVec2i){.x = imin(a.x, b.x), .y = imin(a.y, b.y) }; }
+static inline sgVec2i max2i(sgVec2i a, sgVec2i b) { return (sgVec2i){.x = imax(a.x, b.x), .y = imax(a.y, b.y) }; }
 
-static inline R dot3(sgVec3 a, sgVec3 b) {
-  return a.x * b.x + a.y * b.y + a.z * b.z;
-}
+static inline sgVec2 add2(sgVec2 a, sgVec2 b) { return (sgVec2){a.x + b.x, a.y + b.y}; }
+static inline sgVec2 scale2(sgVec2 v, R s) { return (sgVec2){v.x * s, v.y * s}; }
+
+static inline sgVec3 neg3(sgVec3 v) { return (sgVec3){-v.x, -v.y, -v.z}; }
+static inline sgVec3 sub3(sgVec3 a, sgVec3 b) { return (sgVec3){a.x - b.x, a.y - b.y, a.z - b.z}; }
+static inline sgVec3 div3(sgVec3 a, sgVec3 b) { return (sgVec3){a.x / b.x, a.y / b.y, a.z / b.z}; }
+static inline R dot3(sgVec3 a, sgVec3 b) { return a.x * b.x + a.y * b.y + a.z * b.z; }
+static inline R normsq3(sgVec3 v) { return v.x * v.x + v.y * v.y + v.z * v.z; }
+static inline R norm3  (sgVec3 v) { return (R)sqrt(normsq3(v)); }
 
 static inline sgVec3 cross3(sgVec3 a, sgVec3 b) {
   return (sgVec3) {
@@ -65,18 +75,12 @@ static inline sgVec3 cross3(sgVec3 a, sgVec3 b) {
     a.x * b.y - a.y * b.x};
 }
 
-static inline R normsq3(sgVec3 v) { return v.x * v.x + v.y * v.y + v.z * v.z; }
-static inline R norm3  (sgVec3 v) { return (R)sqrt(normsq3(v)); }
-
 static inline sgVec3 normalize3(sgVec3 v) {
   const R n = norm3(v);
-  assert(n > 0);
-  return (sgVec3){v.x / n, v.y / n, v.z / n};
+  return (n > 0) ? (sgVec3){v.x / n, v.y / n, v.z / n} : (sgVec3){0, 0, 0};
 }
 
-static inline sgVec4 Vec4FromVec3(sgVec3 v, R w) {
-  return (sgVec4){v.x, v.y, v.z, w};
-}
+static inline sgVec4 Vec4FromVec3(sgVec3 v, R w) { return (sgVec4){v.x, v.y, v.z, w}; }
 
 static inline sgMat4 Mat4(
     R m00, R m01, R m02, R m03,   // v0.x v1.x v2.x v3.x
@@ -231,37 +235,39 @@ static inline sgMat4 Mat4Perspective(R fovy, R aspect, R near, R far) {
 }
 
 #ifndef _NDEBUG
-static bool InBounds(int width, int height, sgVec2i p) {
-  return (0 <= p.x) && (p.x < width) &&
-         (0 <= p.y) && (p.y < height);
+static bool InBounds(int width, int height, int x, int y) {
+  return (0 <= x) && (x < width) &&
+         (0 <= y) && (y < height);
 }
 #endif // _NDEBUG
 
-static inline sgPixel* Pixel(sgPixel* image, int width, int height, int x, int y) {
-  assert(InBounds(width, height, (sgVec2i){x,y}));
-  return image + (y * width) + x;
+static inline sgPixel* Pixel(swgfx* gfx, int x, int y) {
+  assert(gfx);
+  assert(gfx->colour);
+  assert(InBounds(gfx->dims.x, gfx->dims.y, x, y));
+  return gfx->colour + (y * gfx->dims.x) + x;
 }
 
-static inline R   rmin(R   a, R   b) { return (a <= b) ? a : b; }
-static inline R   rmax(R   a, R   b) { return (a >= b) ? a : b; }
-static inline int imin(int a, int b) { return (a <= b) ? a : b; }
-static inline int imax(int a, int b) { return (a >= b) ? a : b; }
-static inline sgVec2 min2(sgVec2 a, sgVec2 b) {
-  return (sgVec2){.x = rmin(a.x, b.x), .y = rmin(a.y, b.y) };
-}
-static inline sgVec2 max2(sgVec2 a, sgVec2 b) {
-  return (sgVec2){.x = rmax(a.x, b.x), .y = rmax(a.y, b.y) };
+static inline R* Depth(swgfx* gfx, int x, int y) {
+  assert(gfx);
+  assert(gfx->depth);
+  assert(InBounds(gfx->dims.x, gfx->dims.y, x, y));
+  return gfx->depth + (y * gfx->dims.x) + x;
 }
-static inline sgVec2i min2i(sgVec2i a, sgVec2i b) {
-  return (sgVec2i){.x = imin(a.x, b.x), .y = imin(a.y, b.y) };
+
+void SetPixel(swgfx* gfx, const sgVec2i p, sgPixel colour) {
+  assert(gfx);
+  *Pixel(gfx, p.x, p.y) = colour;
 }
-static inline sgVec2i max2i(sgVec2i a, sgVec2i b) {
-  return (sgVec2i){.x = imax(a.x, b.x), .y = imax(a.y, b.y) };
+
+void SetDepth(swgfx* gfx, const sgVec2i p, R depth) {
+  assert(gfx);
+  *Depth(gfx, p.x, p.y) = depth;
 }
 
-static inline sgAABB2 TriangleAabb2(const sgTri2 tri) {
-  return (sgAABB2){.pmin = min2(min2(tri.p0, tri.p1), tri.p2),
-                   .pmax = max2(max2(tri.p0, tri.p1), tri.p2)};
+static inline sgAABB2 TriangleAabb2(sgVec2 p0, sgVec2 p1, sgVec2 p2) {
+  return (sgAABB2){.pmin = min2(min2(p0, p1), p2),
+                   .pmax = max2(max2(p0, p1), p2)};
 }
 
 static inline sgVec2i Clip(const swgfx* gfx, const sgVec2i p) {
@@ -272,30 +278,45 @@ static inline sgVec2i Clip(const swgfx* gfx, const sgVec2i p) {
   return max2i(lower, min2i(upper, p));
 }
 
+static inline R BarycentricInterp1(sgVec3 bar, R a, R b, R c) {
+  return bar.x*a + bar.y*b + bar.z*c;
+}
+static inline sgVec2 BarycentricInterp2(sgVec3 bar, sgVec2 a, sgVec2 b, sgVec2 c) {
+  return add2(add2(scale2(a, bar.x), scale2(b, bar.y)), scale2(c, bar.z));
+}
+static inline sgVec2 PerspectiveInterp2(sgVec3 bar, sgVec3 depths, R z, sgVec2 a, sgVec2 b, sgVec2 c) {
+  return scale2(BarycentricInterp2(div3(bar, depths), a, b, c), z);
+}
+
 static inline R f(sgVec2 a, sgVec2 b, sgVec2 p) {
   return (a.y - b.y)*p.x + (b.x - a.x)*p.y + a.x*b.y - b.x*a.y;
 }
 
-static inline sgVec3 Barycentric(const sgTri2 tri, sgVec2 p) {
+static inline sgVec3 Barycentric(sgVec2 p0, sgVec2 p1, sgVec2 p2, sgVec2 p) {
   // There is no need to compute the third coordinate explicitly: a + b + c = 1.
   // But this results in a worse rasterization of the triangle along one of the edges.
   // It seems we can patch it with a small epsilon, though.
   // ---
   // Division by zero is only possible if the triangle has zero area.
   /*return (sgVec3){
-    f(tri.p1, tri.p2, p) / f(tri.p1, tri.p2, tri.p0),
-    f(tri.p2, tri.p0, p) / f(tri.p2, tri.p0, tri.p1),
-    f(tri.p0, tri.p1, p) / f(tri.p0, tri.p1, tri.p2)};*/
-  const R b = f(tri.p0, tri.p2, p) / f(tri.p0, tri.p2, tri.p1);
-  const R c = f(tri.p0, tri.p1, p) / f(tri.p0, tri.p1, tri.p2);
-  const R a = /*f(tri.p1, tri.p2, p) / f(tri.p1, tri.p2, tri.p0);*/1.f - b - c - (R)1e-7;
+    f(p1, p2, p) / f(p1, p2, p0),
+    f(p2, p0, p) / f(p2, p0, p1),
+    f(p0, p1, p) / f(p0, p1, p2)};*/
+  const R b = f(p0, p2, p) / f(p0, p2, p1);
+  const R c = f(p0, p1, p) / f(p0, p1, p2);
+  const R a = /*f(p1, p2, p) / f(p1, p2, p0);*/1.f - b - c - (R)1e-7;
   return (sgVec3){a,b,c};
 }
 
-static void DrawTriangle2(swgfx* gfx, const sgTri2* tri) {
+// Input triangle in screen space. Retains Z for depth testing and vertex
+// attribute interpolation.
+static void DrawTriangle2(swgfx* gfx, const sgTri3* const tri) {
   assert(gfx);
   assert(tri);
-  const sgAABB2 bbox = TriangleAabb2(*tri);
+  const sgVec2 p0 = (sgVec2){tri->p0.pos.x, tri->p0.pos.y};
+  const sgVec2 p1 = (sgVec2){tri->p1.pos.x, tri->p1.pos.y};
+  const sgVec2 p2 = (sgVec2){tri->p2.pos.x, tri->p2.pos.y};
+  const sgAABB2 bbox = TriangleAabb2(p0, p1, p2);
   // We consider (x,y) to be the pixel center.
   // Draw all pixels touched by the bounding box. TODO: Multi-sampling.
   sgVec2i pmin = (sgVec2i){(int)bbox.pmin.x, (int)bbox.pmin.y};
@@ -309,14 +330,23 @@ static void DrawTriangle2(swgfx* gfx, const sgTri2* tri) {
       const sgVec2 p = (sgVec2){(R)x, (R)y};
       // TODO: there is an incremental optimization to computing barycentric coordinates;
       //  read more about it.
-      const sgVec3 bar = Barycentric(*tri, p);
+      const sgVec3 bar = Barycentric(p0, p1, p2, p);
       // We need to check the third coordinate.
       //   a + b + c = 1
       //   So, e.g., if a >= 0 and b >= 0, then we have c <= 1, but we could also have c <= 0.
       //   In the case c <= 0, then point is outside the triangle.
       if ((bar.x >= 0) && (bar.y >= 0) && (bar.z >= 0)) {
-        const sgVec2i pi = (sgVec2i){(int)x, (int)y};
-        sgPixels(gfx, 1, &pi, (sgPixel){255, 255, 255, 255});
+        const R z = BarycentricInterp1(bar, tri->p0.pos.z, tri->p1.pos.z, tri->p2.pos.z);
+        R* depth = Depth(gfx, x, y);
+        if ((0.f <= z) && (z <= 1.f) && (z <= *depth)) {
+          *depth = z;
+          const sgVec3 depths = (sgVec3){tri->p0.pos.z, tri->p1.pos.z, tri->p2.pos.z};
+          const sgVec2  uv = PerspectiveInterp2(bar, depths, z, tri->p0.uv, tri->p1.uv, tri->p2.uv);
+          const int r = (int)(uv.x * 255.f);
+          const int g = (int)(uv.y * 255.f);
+          const sgVec2i pix = (sgVec2i){(int)x, (int)y};
+          SetPixel(gfx, pix, (sgPixel){r, g, 255, 255});
+        }
       }
     }
   }
@@ -327,17 +357,18 @@ static inline sgVec3 PerspDivide(sgVec4 v) {
 }
 
 // TODO: Compute a viewport matrix in sgViewport() instead.
-static inline sgVec2 ViewportTransform(sgViewport_t vp, sgVec3 ndc) {
-  return (sgVec2){
+static inline sgVec3 ViewportTransform(sgViewport_t vp, sgVec3 ndc) {
+  return (sgVec3){
     .x = (ndc.x+1.f) * ((R)vp.width/2.f)  + (R)vp.x0,
-    .y = (ndc.y+1.f) * ((R)vp.height/2.f) + (R)vp.y0};
+    .y = (ndc.y+1.f) * ((R)vp.height/2.f) + (R)vp.y0,
+    .z = ndc.z};
 }
 
-static inline sgVec2 ViewportToWindow(sgViewport_t vp, sgVec2 p) {
-  return (sgVec2){p.x, (R)vp.height - p.y};
+static inline sgVec3 ViewportToWindow(sgViewport_t vp, sgVec3 p) {
+  return (sgVec3){p.x, (R)vp.height - p.y, p.z};
 }
 
-static inline sgVec2 TransformPosition(const swgfx* gfx, sgVec3 p) {
+static inline sgVec3 TransformPosition(const swgfx* gfx, sgVec3 p) {
   assert(gfx);
   // Model to clip space.
   const sgVec4 p_clip = Mat4MulVec4(gfx->mvp, Vec4FromVec3(p, 1));
@@ -345,18 +376,24 @@ static inline sgVec2 TransformPosition(const swgfx* gfx, sgVec3 p) {
   // Perspective divide.
   const sgVec3 p_ndc = PerspDivide(p_clip);
   // TODO: Clip.
-  const sgVec2 p_vp = ViewportTransform(gfx->viewport, p_ndc);
+  const sgVec3 p_vp = ViewportTransform(gfx->viewport, p_ndc);
   return ViewportToWindow(gfx->viewport, p_vp);
 }
 
-static void DrawTriangle3(swgfx* gfx, const sgTri3* tri) {
+static void DrawTriangle3(swgfx* gfx, const sgTri3* const tri) {
   assert(gfx);
   assert(tri);
-  const sgVec2 p0 = TransformPosition(gfx, tri->p0);
-  const sgVec2 p1 = TransformPosition(gfx, tri->p1);
-  const sgVec2 p2 = TransformPosition(gfx, tri->p2);
-  const sgTri2 tri2 = (sgTri2){p0, p1, p2};
-  DrawTriangle2(gfx, &tri2);
+  const sgVec3 p0  = TransformPosition(gfx, tri->p0.pos);
+  const sgVec3 p1  = TransformPosition(gfx, tri->p1.pos);
+  const sgVec3 p2  = TransformPosition(gfx, tri->p2.pos);
+  const sgVec2 uv0 = tri->p0.uv;
+  const sgVec2 uv1 = tri->p1.uv;
+  const sgVec2 uv2 = tri->p2.uv;
+  const sgTri3 tri_screen = (sgTri3){
+    (sgVert3){p0, uv0},
+    (sgVert3){p1, uv1},
+    (sgVert3){p2, uv2}};
+  DrawTriangle2(gfx, &tri_screen);
 }
 
 #define is_pow2_or_0(X) ((X & (X - 1)) == 0)
@@ -389,6 +426,7 @@ static void* Alloc(void** ppMem, size_t count, size_t size) {
 size_t sgMem(int width, int height) {
   return Align(sizeof(swgfx)) +
          Align(width * height * sizeof(sgPixel)) +
+         Align(width * height * sizeof(R)) +
          (SG_ALIGN - 1); // To make room to align allocations within the buffer.
 }
 
@@ -397,6 +435,7 @@ swgfx* sgNew(int width, int height, void* mem) {
   swgfx* gfx  = SG_ALLOC(&aligned, 1, swgfx);
   gfx->dims   = (sgVec2i){width, height};
   gfx->colour = SG_ALLOC(&aligned, width * height, sgPixel);
+  gfx->depth  = SG_ALLOC(&aligned, width * height, R);
   return gfx;
 }
 
@@ -481,15 +520,17 @@ void sgViewport(swgfx* gfx, int x0, int y0, int width, int height) {
 
 void sgClear(swgfx* gfx) {
   assert(gfx);
-  memset(gfx->colour, 0, gfx->dims.x * gfx->dims.y * sizeof(sgPixel));
+  const int N = gfx->dims.x * gfx->dims.y;
+  memset(gfx->colour, 0, N * sizeof(*gfx->colour));
+  for (int i = 0; i < N; ++i) {
+      gfx->depth[i] = 1.0f;
+  }
 }
 
 void sgPixels(swgfx* gfx, size_t count, const sgVec2i* positions, sgPixel colour) {
   assert(gfx);
-#define XY(X,Y) Pixel(gfx->colour, gfx->dims.x, gfx->dims.y, X, Y)
   for (size_t i = 0; i < count; ++i) {
-    const sgVec2i p = positions[i];
-    *XY(p.x, p.y) = colour;
+    SetPixel(gfx, positions[i], colour);
   }
 }
 
@@ -507,7 +548,12 @@ void sgPixels(swgfx* gfx, size_t count, const sgVec2i* positions, sgPixel colour
 void sgTriangles2(swgfx* gfx, size_t count, const sgTri2* tris) {
   assert(gfx);
   for (size_t i = 0; i < count; ++i) {
-    DrawTriangle2(gfx, &tris[i]);
+    const sgTri3 tri3 = (sgTri3) {
+      .p0 = (sgVert3){.pos = (sgVec3){tris[i].p0.pos.x, tris[i].p0.pos.y, 0}, .uv = tris[i].p0.uv},
+      .p1 = (sgVert3){.pos = (sgVec3){tris[i].p1.pos.x, tris[i].p1.pos.y, 0}, .uv = tris[i].p1.uv},
+      .p2 = (sgVert3){.pos = (sgVec3){tris[i].p2.pos.x, tris[i].p2.pos.y, 0}, .uv = tris[i].p2.uv},
+    };
+    DrawTriangle2(gfx, &tri3);
   }
 }
 
@@ -520,10 +566,11 @@ void sgTriangles(swgfx* gfx, size_t count, const sgTri3* tris, const sgNormal*)
   }
 }
 
-void sgTrianglesIndexed(swgfx* gfx, size_t numIndices, const sgIdx* indices, const sgVec3* positions) {
+void sgTrianglesIndexed(swgfx* gfx, size_t numIndices, const sgIdx* indices, const sgVec3* positions, const sgVec2* texcoords) {
   assert(gfx);
   assert(indices);
   assert(positions);
+  assert(texcoords);
   for (size_t i = 0; i < numIndices; i+=3) {
     const sgIdx i0 = indices[i];
     const sgIdx i1 = indices[i+1];
@@ -531,21 +578,28 @@ void sgTrianglesIndexed(swgfx* gfx, size_t numIndices, const sgIdx* indices, con
     const sgVec3 p0 = positions[i0];
     const sgVec3 p1 = positions[i1];
     const sgVec3 p2 = positions[i2];
-    const sgTri3 tri = (sgTri3){p0, p1, p2};
+    const sgVec2 uv0 = texcoords[i0];
+    const sgVec2 uv1 = texcoords[i1];
+    const sgVec2 uv2 = texcoords[i2];
+    const sgTri3 tri = (sgTri3){
+      (sgVert3){p0, uv0},
+      (sgVert3){p1, uv1},
+      (sgVert3){p2, uv2}};
     DrawTriangle3(gfx, &tri);
   }
 }
 
-void sgTrianglesIndexedNonUniform(swgfx* gfx, size_t numTris, const sgTriIdx* tris, const sgVec3* positions) {
+void sgTrianglesIndexedNonUniform(swgfx* gfx, size_t numTris, const sgTriIdx* tris, const sgVec3* positions, const sgVec2* texcoords) {
   assert(gfx);
   assert(tris);
   assert(positions);
+  assert(texcoords);
   for (size_t t = 0; t < numTris; ++t) {
     const sgTriIdx* triIdx = &tris[t];
     const sgTri3 tri = (sgTri3){
-      positions[triIdx->v0.position],
-      positions[triIdx->v1.position],
-      positions[triIdx->v2.position]};
+      (sgVert3){positions[triIdx->v0.pos], texcoords[triIdx->v0.uv]},
+      (sgVert3){positions[triIdx->v1.pos], texcoords[triIdx->v1.uv]},
+      (sgVert3){positions[triIdx->v2.pos], texcoords[triIdx->v2.uv]}};
     DrawTriangle3(gfx, &tri);
   }
 }