Switched to packing using f16s to maintain acceptable precision.

4 files changed, 110 insertions, 90 deletions

diff --git a/wgpu/src/quad/gradient.rs b/wgpu/src/quad/gradient.rs
index 1b45c9f4..6db37252 100644
--- a/wgpu/src/quad/gradient.rs
+++ b/wgpu/src/quad/gradient.rs
@@ -96,24 +96,26 @@ impl Pipeline {
                                 as u64,
                             step_mode: wgpu::VertexStepMode::Instance,
                             attributes: &wgpu::vertex_attr_array!(
-                                // Colors 1-4
+                                // Colors 1-2
                                 1 => Uint32x4,
-                                // Colors 5-8
+                                // Colors 3-4
                                 2 => Uint32x4,
-                                // Offsets 1-4
-                                3 => Float32x4,
-                                // Offsets 5-8
-                                4 => Float32x4,
+                                // Colors 5-6
+                                3 => Uint32x4,
+                                // Colors 7-8
+                                4 => Uint32x4,
+                                // Offsets 1-8
+                                5 => Uint32x4,
                                 // Direction
-                                5 => Float32x4,
-                                // Position & Scale
                                 6 => Float32x4,
-                                // Border color
+                                // Position & Scale
                                 7 => Float32x4,
-                                // Border radius
+                                // Border color
                                 8 => Float32x4,
+                                // Border radius
+                                9 => Float32x4,
                                 // Border width
-                                9 => Float32
+                                10 => Float32
                             ),
                         },
                     ],
diff --git a/wgpu/src/shader/quad.wgsl b/wgpu/src/shader/quad.wgsl
index 5a1237e6..fb402158 100644
--- a/wgpu/src/shader/quad.wgsl
+++ b/wgpu/src/shader/quad.wgsl
@@ -38,10 +38,11 @@ fn select_border_radius(radi: vec4<f32>, position: vec2<f32>, center: vec2<f32>)
     return rx;
 }
 
-fn unpack_u32(color: u32) -> vec4<f32> {
-    let u = unpack4x8unorm(color);
+fn unpack_u32(color: vec2<u32>) -> vec4<f32> {
+    let rg: vec2<f32> = unpack2x16float(color.x);
+    let ba: vec2<f32> = unpack2x16float(color.y);
 
-    return vec4<f32>(u.w, u.z, u.y, u.x);
+    return vec4<f32>(rg.y, rg.x, ba.y, ba.x);
 }
 
 struct SolidVertexInput {
@@ -148,26 +149,28 @@ struct GradientVertexInput {
     @location(0) v_pos: vec2<f32>,
     @location(1) colors_1: vec4<u32>,
     @location(2) colors_2: vec4<u32>,
-    @location(3) offsets_1: vec4<f32>,
-    @location(4) offsets_2: vec4<f32>,
-    @location(5) direction: vec4<f32>,
-    @location(6) position_and_scale: vec4<f32>,
-    @location(7) border_color: vec4<f32>,
-    @location(8) border_radius: vec4<f32>,
-    @location(9) border_width: f32,
+    @location(3) colors_3: vec4<u32>,
+    @location(4) colors_4: vec4<u32>,
+    @location(5) offsets: vec4<u32>,
+    @location(6) direction: vec4<f32>,
+    @location(7) position_and_scale: vec4<f32>,
+    @location(8) border_color: vec4<f32>,
+    @location(9) border_radius: vec4<f32>,
+    @location(10) border_width: f32,
 }
 
 struct GradientVertexOutput {
     @builtin(position) position: vec4<f32>,
     @location(1) colors_1: vec4<u32>,
     @location(2) colors_2: vec4<u32>,
-    @location(3) offsets_1: vec4<f32>,
-    @location(4) offsets_2: vec4<f32>,
-    @location(5) direction: vec4<f32>,
-    @location(6) position_and_scale: vec4<f32>,
-    @location(7) border_color: vec4<f32>,
-    @location(8) border_radius: vec4<f32>,
-    @location(9) border_width: f32,
+    @location(3) colors_3: vec4<u32>,
+    @location(4) colors_4: vec4<u32>,
+    @location(5) offsets: vec4<u32>,
+    @location(6) direction: vec4<f32>,
+    @location(7) position_and_scale: vec4<f32>,
+    @location(8) border_color: vec4<f32>,
+    @location(9) border_radius: vec4<f32>,
+    @location(10) border_width: f32,
 }
 
 @vertex
@@ -195,8 +198,9 @@ fn gradient_vs_main(input: GradientVertexInput) -> GradientVertexOutput {
     out.position = globals.transform * transform * vec4<f32>(input.v_pos, 0.0, 1.0);
     out.colors_1 = input.colors_1;
     out.colors_2 = input.colors_2;
-    out.offsets_1 = input.offsets_1;
-    out.offsets_2 = input.offsets_2;
+    out.colors_3 = input.colors_3;
+    out.colors_4 = input.colors_4;
+    out.offsets = input.offsets;
     out.direction = input.direction * globals.scale;
     out.position_and_scale = vec4<f32>(pos, scale);
     out.border_color = input.border_color;
@@ -262,25 +266,28 @@ fn gradient(
 @fragment
 fn gradient_fs_main(input: GradientVertexOutput) -> @location(0) vec4<f32> {
     let colors = array<vec4<f32>, 8>(
-        unpack_u32(input.colors_1.x),
-        unpack_u32(input.colors_1.y),
-        unpack_u32(input.colors_1.z),
-        unpack_u32(input.colors_1.w),
-        unpack_u32(input.colors_2.x),
-        unpack_u32(input.colors_2.y),
-        unpack_u32(input.colors_2.z),
-        unpack_u32(input.colors_2.w),
+        unpack_u32(input.colors_1.xy),
+        unpack_u32(input.colors_1.zw),
+        unpack_u32(input.colors_2.xy),
+        unpack_u32(input.colors_2.zw),
+        unpack_u32(input.colors_3.xy),
+        unpack_u32(input.colors_3.zw),
+        unpack_u32(input.colors_4.xy),
+        unpack_u32(input.colors_4.zw),
     );
 
+    let offsets_1: vec4<f32> = unpack_u32(input.offsets.xy);
+    let offsets_2: vec4<f32> = unpack_u32(input.offsets.zw);
+
     var offsets = array<f32, 8>(
-        input.offsets_1.x,
-        input.offsets_1.y,
-        input.offsets_1.z,
-        input.offsets_1.w,
-        input.offsets_2.x,
-        input.offsets_2.y,
-        input.offsets_2.z,
-        input.offsets_2.w,
+        offsets_1.x,
+        offsets_1.y,
+        offsets_1.z,
+        offsets_1.w,
+        offsets_2.x,
+        offsets_2.y,
+        offsets_2.z,
+        offsets_2.w,
     );
 
     //TODO could just pass this in to the shader but is probably more performant to just check it here
diff --git a/wgpu/src/shader/triangle.wgsl b/wgpu/src/shader/triangle.wgsl
index f1bb2733..9f512d14 100644
--- a/wgpu/src/shader/triangle.wgsl
+++ b/wgpu/src/shader/triangle.wgsl
@@ -4,10 +4,11 @@ struct Globals {
 
 @group(0) @binding(0) var<uniform> globals: Globals;
 
-fn unpack_u32(color: u32) -> vec4<f32> {
-    let u = unpack4x8unorm(color);
+fn unpack_u32(color: vec2<u32>) -> vec4<f32> {
+    let rg: vec2<f32> = unpack2x16float(color.x);
+    let ba: vec2<f32> = unpack2x16float(color.y);
 
-    return vec4<f32>(u.w, u.z, u.y, u.x);
+    return vec4<f32>(rg.y, rg.x, ba.y, ba.x);
 }
 
 struct SolidVertexInput {
@@ -35,34 +36,39 @@ fn solid_fs_main(input: SolidVertexOutput) -> @location(0) vec4<f32> {
     return input.color;
 }
 
+struct GradientVertexInput {
+    @location(0) v_pos: vec2<f32>,
+    @location(1) colors_1: vec4<u32>,
+    @location(2) colors_2: vec4<u32>,
+    @location(3) colors_3: vec4<u32>,
+    @location(4) colors_4: vec4<u32>,
+    @location(5) offsets: vec4<u32>,
+    @location(6) direction: vec4<f32>,
+}
+
 struct GradientVertexOutput {
     @builtin(position) position: vec4<f32>,
     @location(0) raw_position: vec2<f32>,
     @location(1) colors_1: vec4<u32>,
     @location(2) colors_2: vec4<u32>,
-    @location(3) offsets_1: vec4<f32>,
-    @location(4) offsets_2: vec4<f32>,
-    @location(5) direction: vec4<f32>,
+    @location(3) colors_3: vec4<u32>,
+    @location(4) colors_4: vec4<u32>,
+    @location(5) offsets: vec4<u32>,
+    @location(6) direction: vec4<f32>,
 }
 
 @vertex
-fn gradient_vs_main(
-    @location(0) input: vec2<f32>,
-    @location(1) colors_1: vec4<u32>,
-    @location(2) colors_2: vec4<u32>,
-    @location(3) offsets_1: vec4<f32>,
-    @location(4) offsets_2: vec4<f32>,
-    @location(5) direction: vec4<f32>,
-) -> GradientVertexOutput {
+fn gradient_vs_main(input: GradientVertexInput) -> GradientVertexOutput {
     var output: GradientVertexOutput;
 
-    output.position = globals.transform * vec4<f32>(input.xy, 0.0, 1.0);
-    output.raw_position = input;
-    output.colors_1 = colors_1;
-    output.colors_2 = colors_2;
-    output.offsets_1 = offsets_1;
-    output.offsets_2 = offsets_2;
-    output.direction = direction;
+    output.position = globals.transform * vec4<f32>(input.v_pos, 0.0, 1.0);
+    output.raw_position = input.v_pos;
+    output.colors_1 = input.colors_1;
+    output.colors_2 = input.colors_2;
+    output.colors_3 = input.colors_3;
+    output.colors_4 = input.colors_4;
+    output.offsets = input.offsets;
+    output.direction = input.direction;
 
     return output;
 }
@@ -123,25 +129,28 @@ fn gradient(
 @fragment
 fn gradient_fs_main(input: GradientVertexOutput) -> @location(0) vec4<f32> {
     let colors = array<vec4<f32>, 8>(
-        unpack_u32(input.colors_1.x),
-        unpack_u32(input.colors_1.y),
-        unpack_u32(input.colors_1.z),
-        unpack_u32(input.colors_1.w),
-        unpack_u32(input.colors_2.x),
-        unpack_u32(input.colors_2.y),
-        unpack_u32(input.colors_2.z),
-        unpack_u32(input.colors_2.w),
+        unpack_u32(input.colors_1.xy),
+        unpack_u32(input.colors_1.zw),
+        unpack_u32(input.colors_2.xy),
+        unpack_u32(input.colors_2.zw),
+        unpack_u32(input.colors_3.xy),
+        unpack_u32(input.colors_3.zw),
+        unpack_u32(input.colors_4.xy),
+        unpack_u32(input.colors_4.zw),
     );
 
+    let offsets_1: vec4<f32> = unpack_u32(input.offsets.xy);
+    let offsets_2: vec4<f32> = unpack_u32(input.offsets.zw);
+
     var offsets = array<f32, 8>(
-        input.offsets_1.x,
-        input.offsets_1.y,
-        input.offsets_1.z,
-        input.offsets_1.w,
-        input.offsets_2.x,
-        input.offsets_2.y,
-        input.offsets_2.z,
-        input.offsets_2.w,
+        offsets_1.x,
+        offsets_1.y,
+        offsets_1.z,
+        offsets_1.w,
+        offsets_2.x,
+        offsets_2.y,
+        offsets_2.z,
+        offsets_2.w,
     );
 
     var last_index = 7;
diff --git a/wgpu/src/triangle.rs b/wgpu/src/triangle.rs
index 3f633e14..3f3635cf 100644
--- a/wgpu/src/triangle.rs
+++ b/wgpu/src/triangle.rs
@@ -652,16 +652,18 @@ mod gradient {
                             attributes: &wgpu::vertex_attr_array!(
                                 // Position
                                 0 => Float32x2,
-                                // Colors 1-4
+                                // Colors 1-2
                                 1 => Uint32x4,
-                                // Colors 5-8,
+                                // Colors 3-4
                                 2 => Uint32x4,
-                                // Offsets 1-4
-                                3 => Float32x4,
-                                // Offsets 5-8
-                                4 => Float32x4,
+                                // Colors 5-6
+                                3 => Uint32x4,
+                                // Colors 7-8
+                                4 => Uint32x4,
+                                // Offsets
+                                5 => Uint32x4,
                                 // Direction
-                                5 => Float32x4
+                                6 => Float32x4
                             ),
                         }],
                     },