1 files changed, 223 insertions, 55 deletions

diff --git a/src/swgfx.c b/src/swgfx.c
index 772a691..3b9ce29 100644
--- a/src/swgfx.c
+++ b/src/swgfx.c
@@ -6,6 +6,11 @@ Matrices:
 Coordinate systems:
   - Right-handed.
   - NDC in [-1, +1].
+  - Viewport goes up and to the right.
+  - Window goes down and to the right.
+  - (x,y) is the center of a pixel.
+    - Top-left:     (x - 1/2, y - 1/2)
+    - Bottom-right: (x + 1/2, y + 1/2)
 */
 #include <swgfx.h>
 
@@ -15,7 +20,7 @@ Coordinate systems:
 #include <stdlib.h>
 #include <string.h>
 
-static const sgVec3 Up3 = (sgVec3){0,1,0};
+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;
@@ -30,8 +35,18 @@ typedef struct swgfx {
   sgVec2i      dims;     // Colour buffer dimensions.
   sgPixel*     colour;   // Colour buffer.
   sgViewport_t viewport;
+  sgMat4       model;    // Model matrix.
   sgMat4       view;     // View matrix.
   sgMat4       proj;     // Projection matrix.
+  // Pre-multiplied matrices.
+  // The model matrix changes once per object, more frequently than view or
+  // projection. View and projection are expected to change infrequently, maybe
+  // once per frame.
+  // Make it so that changing the model matrix only requires one matrix
+  // multiplication (mvp = model * viewProj) and not two (mvp = model * view * projection)
+  // before rendering the model's triangles.
+  sgMat4       viewProj; // View-projection matrix.
+  sgMat4       mvp;      // Model-view-projection matrix.
 } swgfx;
 
 static inline sgVec3 neg3(sgVec3 v) { return (sgVec3){-v.x, -v.y, -v.z}; }
@@ -40,6 +55,10 @@ static inline sgVec3 sub3(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 sgVec3 cross3(sgVec3 a, sgVec3 b) {
   return (sgVec3) {
     a.y * b.z - a.z * b.y,
@@ -48,8 +67,7 @@ static inline sgVec3 cross3(sgVec3 a, sgVec3 b) {
 }
 
 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 sqrt(normsq3(v)); }
+static inline R norm3  (sgVec3 v) { return (R)sqrt(normsq3(v)); }
 
 static inline sgVec3 normalize3(sgVec3 v) {
   const R n = norm3(v);
@@ -57,6 +75,10 @@ static inline sgVec3 normalize3(sgVec3 v) {
   return (sgVec3){v.x / n, v.y / n, v.z / n};
 }
 
+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
     R m10, R m11, R m12, R m13,   // v0.y v1.y v2.y v3.y
@@ -78,6 +100,10 @@ static inline sgMat4 Mat4FromVec3(sgVec3 right, sgVec3 up, sgVec3 forward, sgVec
 }
 
 static inline R Mat4At(sgMat4 m, int row, int col) { return m.val[col][row]; }
+static inline sgVec3 Mat4v0(sgMat4 m) { return *((sgVec3*)m.val[0]); }
+static inline sgVec3 Mat4v1(sgMat4 m) { return *((sgVec3*)m.val[1]); }
+static inline sgVec3 Mat4v2(sgMat4 m) { return *((sgVec3*)m.val[2]); }
+static inline sgVec3 Mat4v3(sgMat4 m) { return *((sgVec3*)m.val[3]); }
 
 static inline sgMat4 Mat4Mul(sgMat4 A, sgMat4 B) {
   R m00 = Mat4At(A, 0, 0) * Mat4At(B, 0, 0) +
@@ -162,6 +188,31 @@ static inline sgVec3 Mat4MulVec3(sgMat4 m, sgVec3 v, R w) {
     .z = Mat4At(m, 2, 0) * v.x + Mat4At(m, 2, 1) * v.y + Mat4At(m, 2, 2) * v.z + Mat4At(m, 2, 3) * w};
 }
 
+static inline sgVec4 Mat4MulVec4(sgMat4 m, sgVec4 v) {
+  sgVec4 u;
+  u.x = Mat4At(m, 0, 0) * v.x + Mat4At(m, 0, 1) * v.y +
+        Mat4At(m, 0, 2) * v.z + Mat4At(m, 0, 3) * v.w;
+  u.y = Mat4At(m, 1, 0) * v.x + Mat4At(m, 1, 1) * v.y +
+        Mat4At(m, 1, 2) * v.z + Mat4At(m, 1, 3) * v.w;
+  u.z = Mat4At(m, 2, 0) * v.x + Mat4At(m, 2, 1) * v.y +
+        Mat4At(m, 2, 2) * v.z + Mat4At(m, 2, 3) * v.w;
+  u.w = Mat4At(m, 3, 0) * v.x + Mat4At(m, 3, 1) * v.y +
+        Mat4At(m, 3, 2) * v.z + Mat4At(m, 3, 3) * v.w;
+  return u;
+}
+
+static inline sgMat4 Mat4InverseTransform(sgMat4 m) {
+  const sgVec3 r = Mat4v0(m);
+  const sgVec3 u = Mat4v1(m);
+  const sgVec3 f = Mat4v2(m);
+  const sgVec3 t = Mat4v3(m);
+  return Mat4(
+      r.x, r.y, r.z, -dot3(r, t),
+      u.x, u.y, u.z, -dot3(u, t),
+      f.x, f.y, f.z, -dot3(f, t),
+      0.f, 0.f, 0.f, 1.f);
+}
+
 static inline sgMat4 Mat4Look(sgVec3 position, sgVec3 forward, sgVec3 up) {
   const sgVec3 right = normalize3(cross3(forward, up));
   up                 = normalize3(cross3(right, forward));
@@ -169,43 +220,59 @@ static inline sgMat4 Mat4Look(sgVec3 position, sgVec3 forward, sgVec3 up) {
 }
 
 static inline sgMat4 Mat4Perspective(R fovy, R aspect, R near, R far) {
-  R f = tan(fovy / 2.0);
+  R f = (R)tan(fovy / 2.0);
   assert(f > 0.0);
-  f = 1.0 / f;
+  f = 1.f / f;
   const R a = near - far;
   return Mat4(
-    f / aspect, 0, 0, 0,
-    0, f, 0, 0,
-    0, 0, (far + near) / a, (2 * far * near / a),
-    0, 0, -1, 0);
+    f / aspect, 0,                 0,                    0,
+    0,          f,                 0,                    0,
+    0,          0,  (far + near) / a, (2 * far * near / a),
+    0,          0,                 -1,                  0);
 }
 
-static inline sgPixel* PixelRow(sgPixel* image, int width, int y) {
-  return image + (y * width);
+#ifndef _NDEBUG
+static bool InBounds(int width, int height, sgVec2i p) {
+  return (0 <= p.x) && (p.x < width) &&
+         (0 <= p.y) && (p.y < height);
 }
+#endif // _NDEBUG
 
-static inline sgPixel* Pixel(sgPixel* image, int width, int x, int y) {
+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;
 }
 
-#define XY(X,Y) Pixel(gfx->colour, gfx->dims.x, X, Y)
-
-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 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 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 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 sgVec2i Clip(const swgfx* gfx, const sgVec2i p) {
+  assert(gfx);
+  constexpr sgVec2i lower = (sgVec2i){0,0};
+  const     sgVec2i upper = (sgVec2i){gfx->viewport.width  - 1,
+                                      gfx->viewport.height - 1};
+  return max2i(lower, min2i(upper, p));
+}
+
 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;
 }
@@ -222,10 +289,77 @@ static inline sgVec3 Barycentric(const sgTri2 tri, sgVec2 p) {
     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 - b - c - 1e-7;
+  const R a = /*f(tri.p1, tri.p2, p) / f(tri.p1, tri.p2, tri.p0);*/1.f - b - c - (R)1e-7;
   return (sgVec3){a,b,c};
 }
 
+static void DrawTriangle2(swgfx* gfx, const sgTri2* tri) {
+  assert(gfx);
+  assert(tri);
+  const sgAABB2 bbox = TriangleAabb2(*tri);
+  // 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};
+  sgVec2i pmax = (sgVec2i){(int)(bbox.pmax.x + 0.5f), (int)(bbox.pmax.y + 0.5f)};
+  // Clip to screen space.
+  pmin = Clip(gfx, pmin);
+  pmax = Clip(gfx, pmax);
+  // Draw.
+  for   (int y = pmin.y; y <= pmax.y; ++y) {
+    for (int x = pmin.x; x <= pmax.x; ++x) {
+      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);
+      // 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});
+      }
+    }
+  }
+}
+
+static inline sgVec3 PerspDivide(sgVec4 v) {
+  return (sgVec3){v.x / v.w, v.y / v.w, v.z / v.w};
+}
+
+// TODO: Compute a viewport matrix in sgViewport() instead.
+static inline sgVec2 ViewportTransform(sgViewport_t vp, sgVec3 ndc) {
+  return (sgVec2){
+    .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};
+}
+
+static inline sgVec2 ViewportToWindow(sgViewport_t vp, sgVec2 p) {
+  return (sgVec2){p.x, (R)vp.height - p.y};
+}
+
+static inline sgVec2 TransformPosition(const swgfx* gfx, sgVec3 p) {
+  assert(gfx);
+  // Model to clip space.
+  const sgVec4 p_clip = Mat4MulVec4(gfx->mvp, Vec4FromVec3(p, 1));
+  // TODO: Backface culling.
+  // Perspective divide.
+  const sgVec3 p_ndc = PerspDivide(p_clip);
+  // TODO: Clip.
+  const sgVec2 p_vp = ViewportTransform(gfx->viewport, p_ndc);
+  return ViewportToWindow(gfx->viewport, p_vp);
+}
+
+static void DrawTriangle3(swgfx* gfx, const sgTri3* 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);
+}
+
 #define is_pow2_or_0(X) ((X & (X - 1)) == 0)
 
 static size_t align(size_t size) {
@@ -295,14 +429,44 @@ void sgPresent(swgfx* gfx, sgVec2i dimensions, sgPixel* screen) {
   }
 }
 
-void sgCam(swgfx* gfx, sgVec3 position, sgVec3 forward) {
+static void sgUpdateViewProjection(swgfx* gfx) {
+  assert(gfx);
+  gfx->viewProj = Mat4Mul(gfx->proj, gfx->view);
+}
+
+static void sgUpdateMvp(swgfx* gfx) {
+  assert(gfx);
+  gfx->mvp = Mat4Mul(gfx->viewProj, gfx->model);
+}
+
+void sgModelId(swgfx* gfx) {
   assert(gfx);
-  gfx->view = Mat4Look(position, forward, Up3);
+  sgModel(gfx,
+    (sgVec3){0,0,0},
+    (sgVec3){1, 0, 0},
+    (sgVec3){0, 1, 0},
+    (sgVec3){0, 0, 1});
+}
+
+void sgModel(swgfx* gfx, sgVec3 position, sgVec3 right, sgVec3 up, sgVec3 forward) {
+  assert(gfx);
+  gfx->model = Mat4FromVec3(right, up, forward, position);
+  sgUpdateMvp(gfx);
+}
+
+void sgView(swgfx* gfx, sgVec3 position, sgVec3 forward) {
+  assert(gfx);
+  const sgMat4 camera = Mat4Look(position, forward, Up3);
+  gfx->view = Mat4InverseTransform(camera);
+  sgUpdateViewProjection(gfx);
+  sgUpdateMvp(gfx);
 }
 
 void sgPerspective(swgfx* gfx, R fovy, R aspect, R near, R far) {
   assert(gfx);
   gfx->proj = Mat4Perspective(fovy, aspect, near, far);
+  sgUpdateViewProjection(gfx);
+  sgUpdateMvp(gfx);
 }
 
 void sgViewport(swgfx* gfx, int x0, int y0, int width, int height) {
@@ -317,34 +481,13 @@ void sgClear(swgfx* gfx) {
 
 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;
   }
 }
 
-static void DrawTriangle2(swgfx* gfx, const sgTri2* tri) {
-  assert(gfx);
-  assert(tri);
-  const sgAABB2 bbox = TriangleAabb2(*tri);
-  for   (int y = bbox.pmin.y; y <= bbox.pmax.y; ++y) {
-    for (int x = bbox.pmin.x; x <= bbox.pmax.x; ++x) {
-      const sgVec2 p = (sgVec2){x, y};
-      // TODO: there is an incremental optimization to computing barycentric coordinates;
-      // read more about it.
-      const sgVec3 bar = Barycentric(*tri, 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});
-      }
-    }
-  }
-}
-
 // TODO: DrawTriangle3 with clipping. Leave DrawTriangle2 to not clip for
 //       performance; assume that 2D triangles are within bounds.
 // TODO: If the triangle is out of bounds, skip entirely.
@@ -365,26 +508,51 @@ void sgTriangles2(swgfx* gfx, size_t count, const sgTri2* tris) {
 
 void sgTriangles(swgfx* gfx, size_t count, const sgTri3* tris, const sgNormal*) {
   assert(gfx);
+  assert(tris);
   for (size_t i = 0; i < count; ++i) {
-    // Ignore projection matrix for now. Rasterize 2D triangles.
-    const sgTri3* tri3 = &tris[i];
-    const sgTri2 tri2 = (sgTri2) {
-      .p0 = (sgVec2){tri3->p0.x, tri3->p0.y},
-      .p1 = (sgVec2){tri3->p1.x, tri3->p1.y},
-      .p2 = (sgVec2){tri3->p2.x, tri3->p2.y},
-    };
-    DrawTriangle2(gfx, &tri2);
+    const sgTri3* tri = &tris[i];
+    DrawTriangle3(gfx, tri);
+  }
+}
+
+void sgTrianglesIndexed(swgfx* gfx, size_t numIndices, const sgIdx* indices, const sgVec3* positions) {
+  assert(gfx);
+  assert(indices);
+  assert(positions);
+  for (size_t i = 0; i < numIndices; i+=3) {
+    const sgIdx i0 = indices[i];
+    const sgIdx i1 = indices[i+1];
+    const sgIdx i2 = indices[i+2];
+    const sgVec3 p0 = positions[i0];
+    const sgVec3 p1 = positions[i1];
+    const sgVec3 p2 = positions[i2];
+    const sgTri3 tri = (sgTri3){p0, p1, p2};
+    DrawTriangle3(gfx, &tri);
+  }
+}
+
+void sgTrianglesIndexedNonUniform(swgfx* gfx, size_t numTris, const sgTriIdx* tris, const sgVec3* positions) {
+  assert(gfx);
+  assert(tris);
+  assert(positions);
+  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]};
+    DrawTriangle3(gfx, &tri);
   }
 }
 
-static inline void AssertViewportWithinBuffer(swgfx* gfx) {
+static bool ViewportWithinBuffer(swgfx* gfx) {
   assert(gfx);
   const sgViewport_t vp = gfx->viewport;
-  assert((vp.x0 + vp.width)  <= gfx->dims.x);
-  assert((vp.y0 + vp.height) <= gfx->dims.y);
+  return ((vp.x0 + vp.width)  <= gfx->dims.x) &&
+         ((vp.y0 + vp.height) <= gfx->dims.y);
 }
 
 void sgCheck(swgfx* gfx) {
   assert(gfx);
-  AssertViewportWithinBuffer(gfx);
+  assert(ViewportWithinBuffer(gfx));
 }