Use 4x4 matrices for transforms

Sources/vger/vger.metal

@@ -58,7 +58,7 @@ kernel void vger_prune(uint gid [[thread_position_in_grid]],
 vertex VertexOut vger_vertex(uint vid [[vertex_id]],
                              uint iid [[instance_id]],
                              const device vgerPrim* prims,
-                             const device float3x3* xforms,
+                             const device float4x4* xforms,
                              constant float2& viewSize) {
 
     device auto& prim = prims[iid];
@@ -67,13 +67,11 @@ vertex VertexOut vger_vertex(uint vid [[vertex_id]],
     out.primIndex = iid;
     out.t = float2(prim.texBounds[vid & 1].x, prim.texBounds[vid >> 1].y);
 
-    auto q = xforms[prim.xform] * float3(float2(prim.quadBounds[vid & 1].x,
-                                                prim.quadBounds[vid >> 1].y),
-                                         1.0);
+    out.position = xforms[prim.xform] * float4(prim.quadBounds[vid & 1].x,
+                                               prim.quadBounds[vid >> 1].y,
+                                               0.0,
+                                               1.0);
 
-    auto p = float2{q.x/q.z, q.y/q.z};
-    out.position = float4(2.0 * p / viewSize - 1.0, 0, 1);
-
     return out;
 }
 

Sources/vger/vger.mm

@@ -44,15 +44,14 @@
         scene.cvs = GPUVec<float2>(device);
         scene.cvs.buffer.label = [NSString stringWithFormat:@"cv buffer scene %d", i];
 
-        scene.xforms = GPUVec<float3x3>(device);
+        scene.xforms = GPUVec<float4x4>(device);
         scene.xforms.buffer.label = [NSString stringWithFormat:@"xform buffer scene %d", i];
 
         scene.paints = GPUVec<vgerPaint>(device);
         scene.paints.buffer.label = [NSString stringWithFormat:@"paints buffer scene %d", i];
 
         scenes[i] = scene;
     }
-    txStack.push_back(matrix_identity_float3x3);
 
     auto desc = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:MTLPixelFormatRGBA8Unorm width:1 height:1 mipmapped:NO];
     nullTexture = [device newTextureWithDescriptor:desc];
@@ -80,6 +79,15 @@ void vgerDelete(vgerContext vg) {
     this->devicePxRatio = devicePxRatio;
     computedGlyphBounds = false;
 
+    // Add view transform to matrix stack.
+    txStack.clear();
+    auto M = matrix_identity_float4x4;
+    M.columns[0].x = 2.0 / windowWidth;
+    M.columns[1].y = 2.0 / windowHeight;
+    M.columns[3].x = -1;
+    M.columns[3].y = -1;
+    txStack.push_back(M);
+
     // Prune the text cache.
     for(auto it = std::begin(textCache); it != std::end(textCache);) {
         if (it->second.lastFrame != currentFrame) {
@@ -319,9 +327,9 @@ size_t vgerPrimCount(vgerContext vg) {
     return vg->primCount();
 }
 
-static float averageScale(const float3x3& M)
+static float averageScale(const float4x4& M, float2 windowSize)
 {
-    return 0.5f * (length(M.columns[0].xy) + length(M.columns[1].xy));
+    return 0.5f * (length(M.columns[0].xy) * windowSize.x/2 + length(M.columns[1].xy) * windowSize.y/2);
 }
 
 static float2 alignOffset(CTLineRef line, int align) {
@@ -501,7 +509,7 @@ void vgerText(vgerContext vg, const char* str, float4 color, int align) {
 
     } else {
 
-        auto scale = averageScale(txStack.back()) * devicePxRatio;
+        auto scale = averageScale(txStack.back(), windowSize) * devicePxRatio;
         auto key = TextLayoutKey{std::string(str), scale, align};
 
         if(renderCachedText(key, paint, xform)) {
@@ -592,7 +600,7 @@ void vgerTextBox(vgerContext vg, const char* str, float breakRowWidth, float4 co
     }
 
     auto paint = vgerColorPaint(this, color);
-    auto scale = averageScale(txStack.back()) * devicePxRatio;
+    auto scale = averageScale(txStack.back(), windowSize) * devicePxRatio;
     auto key = TextLayoutKey{std::string(str), scale, align, breakRowWidth};
     auto xform = addxform(txStack.back());
 
@@ -840,13 +848,20 @@ static bool isValid(matrix_float3x3 M) {
            isValid(M.columns[2]);
 }
 
+static bool isValid(matrix_float4x4 M) {
+    return isValid(M.columns[0]) &&
+           isValid(M.columns[1]) &&
+           isValid(M.columns[2]) &&
+           isValid(M.columns[3]);
+}
+
 void vgerTranslate(vgerContext vg, float2 t) {
     if(!isValid(t)) {
         fprintf(stderr, "vgerTranslate: bad translation: (%f, %f)\n", t.x, t.y);
         return;
     }
-    auto M = matrix_identity_float3x3;
-    M.columns[2] = vector3(t, 1);
+    auto M = matrix_identity_float4x4;
+    M.columns[3] = float4{t.x, t.y, 0, 1};
 
     assert(!vg->txStack.empty());
     auto A = vg->txStack.back();
@@ -865,7 +880,7 @@ void vgerScale(vgerContext vg, float2 s) {
         fprintf(stderr, "vgerScale: bad scale: (%f, %f)\n", s.x, s.y);
         return;
     }
-    auto M = matrix_identity_float3x3;
+    auto M = matrix_identity_float4x4;
     M.columns[0].x = s.x;
     M.columns[1].y = s.y;
 
@@ -881,7 +896,7 @@ void vgerScale(vgerContext vg, float2 s) {
 }
 
 void vgerRotate(vgerContext vg, float theta) {
-    auto M = matrix_identity_float3x3;
+    auto M = matrix_identity_float4x4;
     M.columns[0].x = cosf(theta);
     M.columns[0].y = sinf(theta);
     M.columns[1].x = - M.columns[0].y;
@@ -894,8 +909,9 @@ void vgerRotate(vgerContext vg, float theta) {
 /// Transforms a point according to the current transformation.
 float2 vgerTransform(vgerContext vg, float2 p) {
     auto& M = vg->txStack.back();
-    auto q = matrix_multiply(M, float3{p.x,p.y,1.0});
-    return {q.x/q.z, q.y/q.z};
+    auto q = matrix_multiply(M, float4{p.x,p.y,0.0,1.0});
+    return { (q.x/q.w + 1) * vg->windowSize.x / 2,
+             (q.y/q.w + 1) * vg->windowSize.y / 2};
 }
 
 simd_float3x2 vgerCurrentTransform(vgerContext vg) {

Sources/vger/vgerScene.h

@@ -56,7 +56,7 @@ struct GPUVec {
 struct vgerScene {
     GPUVec<vgerPrim>  prims[VGER_MAX_LAYERS];
     GPUVec<float2>    cvs;
-    GPUVec<float3x3>  xforms;
+    GPUVec<float4x4>  xforms;
     GPUVec<vgerPaint> paints;
 
     void clear() {

Sources/vger/vger_private.h

@@ -65,7 +65,7 @@ struct vger {
     vgerRenderer* glowRenderer;
 
     /// Transform matrix stack.
-    std::vector<float3x3> txStack;
+    std::vector<float4x4> txStack;
 
     /// Number of buffers.
     int maxBuffers = 1;
@@ -138,7 +138,7 @@ struct vger {
         scenes[currentScene].cvs.append(p);
     }
 
-    uint32_t addxform(const matrix_float3x3& M) {
+    uint32_t addxform(const matrix_float4x4& M) {
         uint32_t idx = (uint32_t) scenes[currentScene].xforms.count;
         scenes[currentScene].xforms.append(M);
         return idx;