Use 3x3 PMV matrices rather than 4x4 in the GL2 engine

QGraphicsView based applications will set a new transform for every
item before it's painted. This leads to lots of updates to the PMV
matrix. So switching to a 3x3 rather than a 4x4 gives us less data
to pass to GL for each QGraphicsItem which gets rendered. It also
means the vertex shader is more efficient. However, this patch only
gives a maximum 2.5% speed improvement on the SGX, which seems to be
only due to the faster vertex shader rather than the reduced amount
of data we pass to GL.

Reviewed-By: Kim

3 files changed, 62 insertions, 64 deletions

diff --git a/src/opengl/gl2paintengineex/qglengineshadersource_p.h b/src/opengl/gl2paintengineex/qglengineshadersource_p.h
index 2407979..46de124 100644
--- a/src/opengl/gl2paintengineex/qglengineshadersource_p.h
+++ b/src/opengl/gl2paintengineex/qglengineshadersource_p.h
@@ -99,12 +99,15 @@ static const char* const qglslMainWithTexCoordsAndOpacityVertexShader = "\
             opacity = opacityArray; \
     }";
 
+// NOTE: We let GL do the perspective correction so texture lookups in the fragment
+//       shader are also perspective corrected.
 static const char* const qglslPositionOnlyVertexShader = "\
-    attribute highp vec4    vertexCoordsArray;\
-    uniform   highp mat4    pmvMatrix;\
+    attribute highp vec2    vertexCoordsArray;\
+    uniform   highp mat3    pmvMatrix;\
     void setPosition(void)\
     {\
-            gl_Position = pmvMatrix * vertexCoordsArray;\
+        vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+        gl_Position = vec4(transformedPos.xy, 0.0, transformedPos.z); \
     }";
 
 static const char* const qglslUntransformedPositionVertexShader = "\
@@ -116,20 +119,19 @@ static const char* const qglslUntransformedPositionVertexShader = "\
 
 // Pattern Brush - This assumes the texture size is 8x8 and thus, the inverted size is 0.125
 static const char* const qglslPositionWithPatternBrushVertexShader = "\
-    attribute highp   vec4  vertexCoordsArray; \
-    uniform   highp   mat4  pmvMatrix; \
+    attribute highp   vec2  vertexCoordsArray; \
+    uniform   highp   mat3  pmvMatrix; \
     uniform   mediump vec2  halfViewportSize; \
     uniform   highp   vec2  invertedTextureSize; \
     uniform   highp   mat3  brushTransform; \
     varying   highp   vec2  patternTexCoords; \
     void setPosition(void) { \
-            gl_Position = pmvMatrix * vertexCoordsArray;\
-            gl_Position.xy = gl_Position.xy / gl_Position.w; \
+            vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+            gl_Position.xy = transformedPos.xy / transformedPos.z; \
             mediump vec2 viewportCoords = (gl_Position.xy + 1.0) * halfViewportSize; \
-            mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1); \
+            mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1.0); \
             mediump float invertedHTexCoordsZ = 1.0 / hTexCoords.z; \
-            gl_Position.xy = gl_Position.xy * invertedHTexCoordsZ; \
-            gl_Position.w = invertedHTexCoordsZ; \
+            gl_Position = vec4(gl_Position.xy * invertedHTexCoordsZ, 0.0, invertedHTexCoordsZ); \
             patternTexCoords.xy = (hTexCoords.xy * 0.125) * invertedHTexCoordsZ; \
     }";
 
@@ -147,20 +149,19 @@ static const char* const qglslPatternBrushSrcFragmentShader = "\
 
 // Linear Gradient Brush
 static const char* const qglslPositionWithLinearGradientBrushVertexShader = "\
-    attribute highp   vec4  vertexCoordsArray; \
-    uniform   highp   mat4  pmvMatrix; \
+    attribute highp   vec2  vertexCoordsArray; \
+    uniform   highp   mat3  pmvMatrix; \
     uniform   mediump vec2  halfViewportSize; \
     uniform   highp   vec3  linearData; \
     uniform   highp   mat3  brushTransform; \
     varying   mediump float index; \
     void setPosition() { \
-        gl_Position = pmvMatrix * vertexCoordsArray;\
-        gl_Position.xy = gl_Position.xy / gl_Position.w; \
+        vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+        gl_Position.xy = transformedPos.xy / transformedPos.z; \
         mediump vec2 viewportCoords = (gl_Position.xy + 1.0) * halfViewportSize; \
         mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1); \
         mediump float invertedHTexCoordsZ = 1.0 / hTexCoords.z; \
-        gl_Position.xy = gl_Position.xy * invertedHTexCoordsZ; \
-        gl_Position.w = invertedHTexCoordsZ; \
+        gl_Position = vec4(gl_Position.xy * invertedHTexCoordsZ, 0.0, invertedHTexCoordsZ); \
         index = (dot(linearData.xy, hTexCoords.xy) * linearData.z) * invertedHTexCoordsZ; \
     }";
 
@@ -178,20 +179,19 @@ static const char* const qglslLinearGradientBrushSrcFragmentShader = "\
 
 // Conical Gradient Brush
 static const char* const qglslPositionWithConicalGradientBrushVertexShader = "\
-    attribute highp   vec4  vertexCoordsArray;\
-    uniform   highp   mat4  pmvMatrix;\
+    attribute highp   vec2  vertexCoordsArray;\
+    uniform   highp   mat3  pmvMatrix;\
     uniform   mediump vec2  halfViewportSize; \
     uniform   highp   mat3  brushTransform; \
     varying   highp   vec2  A; \
     void setPosition(void)\
     {\
-        gl_Position = pmvMatrix * vertexCoordsArray;\
-        gl_Position.xy = gl_Position.xy / gl_Position.w; \
+        vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+        gl_Position.xy = transformedPos.xy / transformedPos.z; \
         mediump vec2  viewportCoords = (gl_Position.xy + 1.0) * halfViewportSize; \
         mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1); \
         mediump float invertedHTexCoordsZ = 1.0 / hTexCoords.z; \
-        gl_Position.xy = gl_Position.xy * invertedHTexCoordsZ; \
-        gl_Position.w = invertedHTexCoordsZ; \
+        gl_Position = vec4(gl_Position.xy * invertedHTexCoordsZ, 0.0, invertedHTexCoordsZ); \
         A = hTexCoords.xy * invertedHTexCoordsZ; \
     }";
 
@@ -215,8 +215,8 @@ static const char* const qglslConicalGradientBrushSrcFragmentShader = "\n\
 
 // Radial Gradient Brush
 static const char* const qglslPositionWithRadialGradientBrushVertexShader = "\
-    attribute highp   vec4 vertexCoordsArray;\
-    uniform   highp   mat4 pmvMatrix;\
+    attribute highp   vec2 vertexCoordsArray;\
+    uniform   highp   mat3 pmvMatrix;\
     uniform   mediump vec2 halfViewportSize; \
     uniform   highp   mat3 brushTransform; \
     uniform   highp   vec2 fmp; \
@@ -224,13 +224,12 @@ static const char* const qglslPositionWithRadialGradientBrushVertexShader = "\
     varying   highp   vec2  A; \
     void setPosition(void) \
     {\
-        gl_Position = pmvMatrix * vertexCoordsArray;\
-        gl_Position.xy = gl_Position.xy / gl_Position.w; \
+        vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+        gl_Position.xy = transformedPos.xy / transformedPos.z; \
         mediump vec2 viewportCoords = (gl_Position.xy + 1.0) * halfViewportSize; \
         mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1); \
         mediump float invertedHTexCoordsZ = 1.0 / hTexCoords.z; \
-        gl_Position.xy = gl_Position.xy * invertedHTexCoordsZ; \
-        gl_Position.w = invertedHTexCoordsZ; \
+        gl_Position = vec4(gl_Position.xy * invertedHTexCoordsZ, 0.0, invertedHTexCoordsZ); \
         A = hTexCoords.xy * invertedHTexCoordsZ; \
         b = 2.0 * dot(A, fmp); \
     }";
@@ -253,20 +252,19 @@ static const char* const qglslRadialGradientBrushSrcFragmentShader = "\
 
 // Texture Brush
 static const char* const qglslPositionWithTextureBrushVertexShader = "\
-    attribute highp   vec4  vertexCoordsArray; \
-    uniform   highp   mat4  pmvMatrix; \
+    attribute highp   vec2 vertexCoordsArray; \
+    uniform   highp   mat3  pmvMatrix; \
     uniform   mediump vec2  halfViewportSize; \
     uniform   highp   vec2  invertedTextureSize; \
     uniform   highp   mat3  brushTransform; \
     varying   highp   vec2  textureCoords; \
     void setPosition(void) { \
-            gl_Position = pmvMatrix * vertexCoordsArray;\
-            gl_Position.xy = gl_Position.xy / gl_Position.w; \
+            vec3 transformedPos = pmvMatrix * vec3(vertexCoordsArray.xy, 1.0); \
+            gl_Position.xy = transformedPos.xy / transformedPos.z; \
             mediump vec2 viewportCoords = (gl_Position.xy + 1.0) * halfViewportSize; \
             mediump vec3 hTexCoords = brushTransform * vec3(viewportCoords, 1); \
             mediump float invertedHTexCoordsZ = 1.0 / hTexCoords.z; \
-            gl_Position.xy = gl_Position.xy * invertedHTexCoordsZ; \
-            gl_Position.w = invertedHTexCoordsZ; \
+            gl_Position = vec4(gl_Position.xy * invertedHTexCoordsZ, 0.0, invertedHTexCoordsZ); \
             textureCoords.xy = (hTexCoords.xy * invertedTextureSize) * gl_Position.w; \
     }";
 
diff --git a/src/opengl/gl2paintengineex/qpaintengineex_opengl2.cpp b/src/opengl/gl2paintengineex/qpaintengineex_opengl2.cpp
index 8ca2fd4..f52ed92 100644
--- a/src/opengl/gl2paintengineex/qpaintengineex_opengl2.cpp
+++ b/src/opengl/gl2paintengineex/qpaintengineex_opengl2.cpp
@@ -583,37 +583,37 @@ void QGL2PaintEngineExPrivate::updateMatrix()
 {
 //     qDebug("QGL2PaintEngineExPrivate::updateMatrix()");
 
-    // We set up the 4x4 transformation matrix on the vertex shaders to
-    // be the equivalent of glOrtho(0, w, h, 0, -1, 1) * transform:
-    //
-    // | 2/width     0     0 -1 |   | m11  m21  0   dx |
-    // |   0    -2/height  0  1 |   | m12  m22  0   dy |
-    // |   0         0    -1  0 | * |  0    0   1   0  |
-    // |   0         0     0  1 |   | m13  m23  0  m33 |
-    //
-    // We expand out the multiplication to save the cost of a full 4x4
-    // matrix multiplication as most of the components are trivial.
     const QTransform& transform = q->state()->matrix;
 
-    qreal wfactor = 2.0 / width;
-    qreal hfactor = -2.0 / height;
-
-    pmvMatrix[0][0] = wfactor * transform.m11() - transform.m13();
-    pmvMatrix[0][1] = hfactor * transform.m12() + transform.m13();
-    pmvMatrix[0][2] = 0.0;
-    pmvMatrix[0][3] = transform.m13();
-    pmvMatrix[1][0] = wfactor * transform.m21() - transform.m23();
-    pmvMatrix[1][1] = hfactor * transform.m22() + transform.m23();
-    pmvMatrix[1][2] = 0.0;
-    pmvMatrix[1][3] = transform.m23();
-    pmvMatrix[2][0] = 0.0;
-    pmvMatrix[2][1] = 0.0;
-    pmvMatrix[2][2] = -1.0;
-    pmvMatrix[2][3] = 0.0;
-    pmvMatrix[3][0] = wfactor * transform.dx() - transform.m33();
-    pmvMatrix[3][1] = hfactor * transform.dy() + transform.m33();
-    pmvMatrix[3][2] = 0.0;
-    pmvMatrix[3][3] = transform.m33();
+    // The projection matrix converts from Qt's coordinate system to GL's coordinate system
+    //    * GL's viewport is 2x2, Qt's is width x height
+    //    * GL has +y -> -y going from bottom -> top, Qt is the other way round
+    //    * GL has [0,0] in the center, Qt has it in the top-left
+    //
+    // This results in the Projection matrix below, which is multiplied by the painter's
+    // transformation matrix, as shown below:
+    //
+    //                Projection Matrix                      Painter Transform
+    // ------------------------------------------------   ------------------------
+    // | 2.0 / width  |      0.0      |     -1.0      |   |  m11  |  m21  |  dx  |
+    // |     0.0      | -2.0 / height |      1.0      | * |  m12  |  m22  |  dy  |
+    // |     0.0      |      0.0      |      1.0      |   |  m13  |  m23  |  m33 |
+    // ------------------------------------------------   ------------------------
+    //
+    // NOTE: The resultant matrix is also transposed, as GL expects column-major matracies
+
+    const GLfloat wfactor = 2.0f / width;
+    const GLfloat hfactor = -2.0f / height;
+
+    pmvMatrix[0][0] = (wfactor * transform.m11())  - transform.m13();
+    pmvMatrix[1][0] = (wfactor * transform.m21())  - transform.m23();
+    pmvMatrix[2][0] = (wfactor * transform.dx() )  - transform.m33();
+    pmvMatrix[0][1] = (hfactor * transform.m12())  + transform.m13();
+    pmvMatrix[1][1] = (hfactor * transform.m22())  + transform.m23();
+    pmvMatrix[2][1] = (hfactor * transform.dy() )  + transform.m33();
+    pmvMatrix[0][2] = transform.m13();
+    pmvMatrix[1][2] = transform.m23();
+    pmvMatrix[2][2] = transform.m33();
 
     // 1/10000 == 0.0001, so we have good enough res to cover curves
     // that span the entire widget...
diff --git a/src/opengl/gl2paintengineex/qpaintengineex_opengl2_p.h b/src/opengl/gl2paintengineex/qpaintengineex_opengl2_p.h
index 33ce24d..f1ec6e6 100644
--- a/src/opengl/gl2paintengineex/qpaintengineex_opengl2_p.h
+++ b/src/opengl/gl2paintengineex/qpaintengineex_opengl2_p.h
@@ -254,7 +254,7 @@ public:
     GLfloat staticVertexCoordinateArray[8];
     GLfloat staticTextureCoordinateArray[8];
 
-    GLfloat pmvMatrix[4][4];
+    GLfloat pmvMatrix[3][3];
 
     QGLEngineShaderManager* shaderManager;