Optimize particles vertex data

Use uchar instead of float for vertex data that doesn't need float.
Continue using floats in shaders. Also remove animY2, which is same as
animY1. These changes reduce memory usage especially when the amount
of particles increases.

Testing on windows, memory reductions with emitters/trailemitter
example were:
- OpenGL: 82.7 MB -> 76.5 MB
- Vulkan: 130.8 MB -> 126.3 MB
- D3D11: 143.7 MB -> 135.8 MB

Task-number: QTBUG-88124
Change-Id: I8f8dcb3845323b0e69fb99b5bff830cd0f151a47
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Reviewed-by: Tomi Korpipää <tomi.korpipaa@qt.io>

14 files changed, 51 insertions, 50 deletions

diff --git a/src/particles/qquickimageparticle.cpp b/src/particles/qquickimageparticle.cpp
index 7ba26c4e54..1331032e7a 100644
--- a/src/particles/qquickimageparticle.cpp
+++ b/src/particles/qquickimageparticle.cpp
@@ -1067,8 +1067,8 @@ void QQuickImageParticle::createEngine()
 }
 
 static QSGGeometry::Attribute SimpleParticle_Attributes[] = {
-    QSGGeometry::Attribute::create(0, 2, QSGGeometry::FloatType, true),             // Position
-    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),             // Data
+    QSGGeometry::Attribute::create(0, 2, QSGGeometry::FloatType, true),      // Position
+    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),            // Data
     QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType)             // Vectors
 };
 
@@ -1080,9 +1080,9 @@ static QSGGeometry::AttributeSet SimpleParticle_AttributeSet =
 };
 
 static QSGGeometry::Attribute ColoredParticle_Attributes[] = {
-    QSGGeometry::Attribute::create(0, 2, QSGGeometry::FloatType, true),             // Position
-    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),             // Data
-    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),             // Vectors
+    QSGGeometry::Attribute::create(0, 2, QSGGeometry::FloatType, true),      // Position
+    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),            // Data
+    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),            // Vectors
     QSGGeometry::Attribute::create(3, 4, QSGGeometry::UnsignedByteType),     // Colors
 };
 
@@ -1094,36 +1094,36 @@ static QSGGeometry::AttributeSet ColoredParticle_AttributeSet =
 };
 
 static QSGGeometry::Attribute DeformableParticle_Attributes[] = {
-    QSGGeometry::Attribute::create(0, 4, QSGGeometry::FloatType),             // Position & TexCoord
-    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),             // Data
-    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),             // Vectors
+    QSGGeometry::Attribute::create(0, 4, QSGGeometry::FloatType),            // Position & Rotation
+    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),            // Data
+    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),            // Vectors
     QSGGeometry::Attribute::create(3, 4, QSGGeometry::UnsignedByteType),     // Colors
-    QSGGeometry::Attribute::create(4, 4, QSGGeometry::FloatType),             // DeformationVectors
-    QSGGeometry::Attribute::create(5, 3, QSGGeometry::FloatType),             // Rotation
+    QSGGeometry::Attribute::create(4, 4, QSGGeometry::FloatType),            // DeformationVectors
+    QSGGeometry::Attribute::create(5, 4, QSGGeometry::UnsignedByteType),     // TexCoord & autoRotate
 };
 
 static QSGGeometry::AttributeSet DeformableParticle_AttributeSet =
 {
     6, // Attribute Count
-    (4 + 4 + 4 + 4 + 3) * sizeof(float) + 4 * sizeof(uchar),
+    (4 + 4 + 4 + 4) * sizeof(float) + (4 + 4) * sizeof(uchar),
     DeformableParticle_Attributes
 };
 
 static QSGGeometry::Attribute SpriteParticle_Attributes[] = {
-    QSGGeometry::Attribute::create(0, 4, QSGGeometry::FloatType),       // Position & TexCoord
-    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),             // Data
-    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),             // Vectors
+    QSGGeometry::Attribute::create(0, 4, QSGGeometry::FloatType),            // Position & Rotation
+    QSGGeometry::Attribute::create(1, 4, QSGGeometry::FloatType),            // Data
+    QSGGeometry::Attribute::create(2, 4, QSGGeometry::FloatType),            // Vectors
     QSGGeometry::Attribute::create(3, 4, QSGGeometry::UnsignedByteType),     // Colors
-    QSGGeometry::Attribute::create(4, 4, QSGGeometry::FloatType),             // DeformationVectors
-    QSGGeometry::Attribute::create(5, 3, QSGGeometry::FloatType),             // Rotation
-    QSGGeometry::Attribute::create(6, 3, QSGGeometry::FloatType),             // Anim Data
-    QSGGeometry::Attribute::create(7, 4, QSGGeometry::FloatType)              // Anim Pos
+    QSGGeometry::Attribute::create(4, 4, QSGGeometry::FloatType),            // DeformationVectors
+    QSGGeometry::Attribute::create(5, 4, QSGGeometry::UnsignedByteType),     // TexCoord & autoRotate
+    QSGGeometry::Attribute::create(6, 3, QSGGeometry::FloatType),            // Anim Data
+    QSGGeometry::Attribute::create(7, 3, QSGGeometry::FloatType)             // Anim Pos
 };
 
 static QSGGeometry::AttributeSet SpriteParticle_AttributeSet =
 {
     8, // Attribute Count
-    (4 + 4 + 4 + 4 + 3 + 3 + 4) * sizeof(float) + 4 * sizeof(uchar),
+    (4 + 4 + 4 + 4 + 3 + 3) * sizeof(float) + (4 + 4) * sizeof(uchar),
     SpriteParticle_Attributes
 };
 
@@ -1622,7 +1622,6 @@ void QQuickImageParticle::spritesUpdate(qreal time)
                 spriteVertices[i].animX1 = x1;
                 spriteVertices[i].animY1 = y;
                 spriteVertices[i].animX2 = x2;
-                spriteVertices[i].animY2 = y;
                 spriteVertices[i].animW = w;
                 spriteVertices[i].animH = h;
                 spriteVertices[i].animProgress = progress;
@@ -1677,7 +1676,7 @@ void QQuickImageParticle::initialize(int gIdx, int pIdx)
 
     float rotation;
     float rotationVelocity;
-    float autoRotate;
+    uchar autoRotate;
     switch (perfLevel){//Fall-through is intended on all of them
         case Sprites:
             // Initial Sprite State
@@ -1749,7 +1748,7 @@ void QQuickImageParticle::initialize(int gIdx, int pIdx)
                         (m_rotation + (m_rotationVariation - 2*QRandomGenerator::global()->bounded(m_rotationVariation)) ) * CONV;
                 rotationVelocity =
                         (m_rotationVelocity + (m_rotationVelocityVariation - 2*QRandomGenerator::global()->bounded(m_rotationVelocityVariation)) ) * CONV;
-                autoRotate = m_autoRotation?1.0:0.0;
+                autoRotate = m_autoRotation ? 1 : 0;
                 if (datum->rotationOwner == this) {
                     datum->rotation = rotation;
                     datum->rotationVelocity = rotationVelocity;
diff --git a/src/particles/qquickimageparticle_p.h b/src/particles/qquickimageparticle_p.h
index 5399bc068c..894d31c300 100644
--- a/src/particles/qquickimageparticle_p.h
+++ b/src/particles/qquickimageparticle_p.h
@@ -99,8 +99,8 @@ struct ColoredVertex {
 struct DeformableVertex {
     float x;
     float y;
-    float tx;
-    float ty;
+    float rotation;
+    float rotationVelocity;
     float t;
     float lifeSpan;
     float size;
@@ -114,16 +114,17 @@ struct DeformableVertex {
     float xy;
     float yx;
     float yy;
-    float rotation;
-    float rotationVelocity;
-    float autoRotate;//Assumed that GPUs prefer floats to bools
+    uchar tx;
+    uchar ty;
+    uchar autoRotate;
+    uchar _padding; // feel free to use
 };
 
 struct SpriteVertex {
     float x;
     float y;
-    float tx;
-    float ty;
+    float rotation;
+    float rotationVelocity;
     float t;
     float lifeSpan;
     float size;
@@ -137,16 +138,16 @@ struct SpriteVertex {
     float xy;
     float yx;
     float yy;
-    float rotation;
-    float rotationVelocity;
-    float autoRotate;//Assumed that GPUs prefer floats to bools
+    uchar tx;
+    uchar ty;
+    uchar autoRotate;
+    uchar _padding; // feel free to use
     float animW;
     float animH;
     float animProgress;
     float animX1;
     float animY1;
     float animX2;
-    float animY2;
 };
 
 template <typename Vertex>
@@ -433,18 +434,19 @@ private:
     template<class Vertex>
     void initTexCoords(Vertex* v, int count){
         Vertex* end = v + count;
+        // Vertex coords between (0.0, 0.0) and (1.0, 1.0)
         while (v < end){
             v[0].tx = 0;
             v[0].ty = 0;
 
-            v[1].tx = 1;
+            v[1].tx = 255;
             v[1].ty = 0;
 
             v[2].tx = 0;
-            v[2].ty = 1;
+            v[2].ty = 255;
 
-            v[3].tx = 1;
-            v[3].ty = 1;
+            v[3].tx = 255;
+            v[3].ty = 255;
 
             v += 4;
         }
diff --git a/src/particles/qquickparticlesystem_p.h b/src/particles/qquickparticlesystem_p.h
index 8b14b74131..d991257452 100644
--- a/src/particles/qquickparticlesystem_p.h
+++ b/src/particles/qquickparticlesystem_p.h
@@ -301,7 +301,7 @@ public:
     float yy;
     float rotation;
     float rotationVelocity;
-    float autoRotate;//Assume that GPUs prefer floats to bools
+    uchar autoRotate; // Basically a bool
     //Used by ImageParticle Sprite mode
     float animIdx;
     float frameDuration;
diff --git a/src/particles/shaders_ng/imageparticle.vert b/src/particles/shaders_ng/imageparticle.vert
index 870139452b..124071e48e 100644
--- a/src/particles/shaders_ng/imageparticle.vert
+++ b/src/particles/shaders_ng/imageparticle.vert
@@ -4,7 +4,7 @@ layout(location = 1) in vec4 vData; // x = time, y = lifeSpan, z = size, w = end
 layout(location = 2) in vec4 vVec;  // x,y = constant velocity,  z,w = acceleration
 
 #if defined(DEFORM)
-layout(location = 0) in vec4 vPosTex;
+layout(location = 0) in vec4 vPosRot; //x = x, y = y,  z = radians of rotation, w = rotation velocity
 #else
 layout(location = 0) in vec2 vPos;
 #endif
@@ -15,12 +15,12 @@ layout(location = 3) in vec4 vColor;
 
 #if defined(DEFORM)
 layout(location = 4) in vec4 vDeformVec; // x,y x unit vector; z,w = y unit vector
-layout(location = 5) in vec3 vRotation;  // x = radians of rotation, y = rotation velocity, z = bool autoRotate
+layout(location = 5) in vec3 vTex;  // x = tx, y = ty, z = bool autoRotate
 #endif
 
 #if defined(SPRITE)
 layout(location = 6) in vec3 vAnimData; // w,h(premultiplied of anim), interpolation progress
-layout(location = 7) in vec4 vAnimPos;  // x,y, x,y (two frames for interpolation)
+layout(location = 7) in vec3 vAnimPos;  // x, y, x2 (two frames for interpolation)
 #endif
 
 #if defined(TABLE)
@@ -55,7 +55,7 @@ void main()
     float t = (ubuf.timestamp - vData.x) / vData.y;
     if (t < 0. || t > 1.) {
 #if defined(DEFORM)
-        gl_Position = ubuf.matrix * vec4(vPosTex.x, vPosTex.y, 0., 1.);
+        gl_Position = ubuf.matrix * vec4(vPosRot.x, vPosRot.y, 0., 1.);
 #else
         gl_PointSize = 0.;
 #endif
@@ -64,13 +64,13 @@ void main()
         tt.y = vAnimData.z;
 
         // Calculate frame location in texture
-        fTexS.xy = vAnimPos.xy + vPosTex.zw * vAnimData.xy;
+        fTexS.xy = vAnimPos.xy + vTex.xy * vAnimData.xy;
 
         // Next frame is also passed, for interpolation
-        fTexS.zw = vAnimPos.zw + vPosTex.zw * vAnimData.xy;
+        fTexS.zw = vAnimPos.zy + vTex.xy * vAnimData.xy;
 
 #elif defined(DEFORM)
-        fTex = vPosTex.zw;
+        fTex = vTex.xy;
 #endif
         float currentSize = mix(vData.z, vData.w, t * t);
         float fade = 1.;
@@ -89,7 +89,7 @@ void main()
 
         if (currentSize <= 0.) {
 #if defined(DEFORM)
-            gl_Position = ubuf.matrix * vec4(vPosTex.x, vPosTex.y, 0., 1.);
+            gl_Position = ubuf.matrix * vec4(vPosRot.x, vPosRot.y, 0., 1.);
 #else
             gl_PointSize = 0.;
 #endif
@@ -99,14 +99,14 @@ void main()
 
             vec2 pos;
 #if defined(DEFORM)
-            float rotation = vRotation.x + vRotation.y * t * vData.y;
-            if (vRotation.z == 1.0) {
+            float rotation = vPosRot.z + vPosRot.w * t * vData.y;
+            if (vTex.z > 0.) {
                 vec2 curVel = vVec.zw * t * vData.y + vVec.xy;
                 if (length(curVel) > 0.)
                     rotation += atan(curVel.y, curVel.x);
             }
             vec2 trigCalcs = vec2(cos(rotation), sin(rotation));
-            vec4 deform = vDeformVec * currentSize * (vPosTex.zzww - 0.5);
+            vec4 deform = vDeformVec * currentSize * (vTex.xxyy - 0.5);
             vec4 rotatedDeform = deform.xxzz * trigCalcs.xyxy;
             rotatedDeform = rotatedDeform + (deform.yyww * trigCalcs.yxyx * vec4(-1.,1.,-1.,1.));
             /* The readable version:
@@ -119,7 +119,7 @@ void main()
             yRotatedDeform.x = trigCalcs.x*yDeform.x - trigCalcs.y*yDeform.y;
             yRotatedDeform.y = trigCalcs.y*yDeform.x + trigCalcs.x*yDeform.y;
             */
-            pos = vPosTex.xy
+            pos = vPosRot.xy
                   + rotatedDeform.xy
                   + rotatedDeform.zw
                   + vVec.xy * t * vData.y         // apply velocity
diff --git a/src/particles/shaders_ng/imageparticle_colored.frag.qsb b/src/particles/shaders_ng/imageparticle_colored.frag.qsb
index 72e517c0ee..84ba3262eb 100644
--- a/src/particles/shaders_ng/imageparticle_colored.frag.qsb
+++ b/src/particles/shaders_ng/imageparticle_colored.frag.qsb
diff --git a/src/particles/shaders_ng/imageparticle_colored.vert.qsb b/src/particles/shaders_ng/imageparticle_colored.vert.qsb
index a18b222b61..293cc7ea4e 100644
--- a/src/particles/shaders_ng/imageparticle_colored.vert.qsb
+++ b/src/particles/shaders_ng/imageparticle_colored.vert.qsb
diff --git a/src/particles/shaders_ng/imageparticle_deformed.frag.qsb b/src/particles/shaders_ng/imageparticle_deformed.frag.qsb
index 3107cfeddd..02179080fa 100644
--- a/src/particles/shaders_ng/imageparticle_deformed.frag.qsb
+++ b/src/particles/shaders_ng/imageparticle_deformed.frag.qsb
diff --git a/src/particles/shaders_ng/imageparticle_deformed.vert.qsb b/src/particles/shaders_ng/imageparticle_deformed.vert.qsb
index b4a701e9fb..d67cb043bb 100644
--- a/src/particles/shaders_ng/imageparticle_deformed.vert.qsb
+++ b/src/particles/shaders_ng/imageparticle_deformed.vert.qsb
diff --git a/src/particles/shaders_ng/imageparticle_simple.frag.qsb b/src/particles/shaders_ng/imageparticle_simple.frag.qsb
index 79e81fa479..e781a1d7a8 100644
--- a/src/particles/shaders_ng/imageparticle_simple.frag.qsb
+++ b/src/particles/shaders_ng/imageparticle_simple.frag.qsb
diff --git a/src/particles/shaders_ng/imageparticle_simple.vert.qsb b/src/particles/shaders_ng/imageparticle_simple.vert.qsb
index 4e1d4c0423..a2e0df027b 100644
--- a/src/particles/shaders_ng/imageparticle_simple.vert.qsb
+++ b/src/particles/shaders_ng/imageparticle_simple.vert.qsb
diff --git a/src/particles/shaders_ng/imageparticle_sprite.frag.qsb b/src/particles/shaders_ng/imageparticle_sprite.frag.qsb
index 05858b1e1e..b720360a1b 100644
--- a/src/particles/shaders_ng/imageparticle_sprite.frag.qsb
+++ b/src/particles/shaders_ng/imageparticle_sprite.frag.qsb
diff --git a/src/particles/shaders_ng/imageparticle_sprite.vert.qsb b/src/particles/shaders_ng/imageparticle_sprite.vert.qsb
index 4054a0268e..284f610f51 100644
--- a/src/particles/shaders_ng/imageparticle_sprite.vert.qsb
+++ b/src/particles/shaders_ng/imageparticle_sprite.vert.qsb
diff --git a/src/particles/shaders_ng/imageparticle_tabled.frag.qsb b/src/particles/shaders_ng/imageparticle_tabled.frag.qsb
index 35da4df2f4..c4ca41814b 100644
--- a/src/particles/shaders_ng/imageparticle_tabled.frag.qsb
+++ b/src/particles/shaders_ng/imageparticle_tabled.frag.qsb
diff --git a/src/particles/shaders_ng/imageparticle_tabled.vert.qsb b/src/particles/shaders_ng/imageparticle_tabled.vert.qsb
index c8be30f414..91955cb521 100644
--- a/src/particles/shaders_ng/imageparticle_tabled.vert.qsb
+++ b/src/particles/shaders_ng/imageparticle_tabled.vert.qsb