Auto merge of servo#615 - kvark:instance, r=glennw

Instanced attributes

Closes servo#457

Performance-wise, I wasn't able to register a noticeable difference. Tested on https://github.com/servo/servo, full screen after the second page down, on `Mesa DRI Intel(R) HD Graphics 5500 (Broadwell GT2)` with resolution 2560x1440.

With the change, I got 4.5 ms mean GPU time in the first test, and 4.3 ms time in the second.
Without the change, I got 4.4 ms mean GPU time in the only test.

I suppose the difference is not visible since we are far from being VS-bound.

Note: this does not replace the optimization of having one large buffer in servo#456. Instanced attributes will benefit from it in the same way as UBOs.

<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/webrender/615)
<!-- Reviewable:end -->

Author: bors-servo

Cargo.lock

@@ -20,7 +20,7 @@ bit-set = "0.4"
 byteorder = "0.5"
 euclid = "0.10.3"
 fnv="1.0"
-gleam = "0.2.25"
+gleam = "0.2.28"
 lazy_static = "0.2"
 log = "0.3"
 num-traits = "0.1.32"

webrender/Cargo.toml

@@ -5,6 +5,11 @@
 
 #ifdef WR_VERTEX_SHADER
 
+in int aClipRenderTaskIndex;
+in int aClipLayerIndex;
+in int aClipDataIndex;
+in int aClipBaseTaskIndex;
+
 struct CacheClipInstance {
     int render_task_index;
     int layer_index;
@@ -15,14 +20,10 @@ struct CacheClipInstance {
 CacheClipInstance fetch_clip_item(int index) {
     CacheClipInstance cci;
 
-    int offset = index * 1;
-
-    ivec4 data0 = int_data[offset + 0];
-
-    cci.render_task_index = data0.x;
-    cci.layer_index = data0.y;
-    cci.data_index = data0.z;
-    cci.base_task_index = data0.w;
+    cci.render_task_index = aClipRenderTaskIndex;
+    cci.layer_index = aClipLayerIndex;
+    cci.data_index = aClipDataIndex;
+    cci.base_task_index = aClipBaseTaskIndex;
 
     return cci;
 }

webrender/res/clip_shared.glsl

@@ -9,8 +9,28 @@
 #define DIR_HORIZONTAL  0
 #define DIR_VERTICAL    1
 
+in int aBlurRenderTaskIndex;
+in int aBlurSourceTaskIndex;
+in int aBlurDirection;
+
+struct BlurCommand {
+    int task_id;
+    int src_task_id;
+    int dir;
+};
+
+BlurCommand fetch_blur() {
+    BlurCommand blur;
+
+    blur.task_id = aBlurRenderTaskIndex;
+    blur.src_task_id = aBlurSourceTaskIndex;
+    blur.dir = aBlurDirection;
+
+    return blur;
+}
+
 void main(void) {
-    BlurCommand cmd = fetch_blur(gl_InstanceID);
+    BlurCommand cmd = fetch_blur();
     RenderTaskData task = fetch_render_task(cmd.task_id);
     RenderTaskData src_task = fetch_render_task(cmd.src_task_id);
 

webrender/res/cs_blur.vs.glsl

@@ -4,7 +4,7 @@
  * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
 
 void main(void) {
-    CachePrimitiveInstance cpi = fetch_cache_instance(gl_InstanceID);
+    CachePrimitiveInstance cpi = fetch_cache_instance();
     RenderTaskData task = fetch_render_task(cpi.render_task_index);
     BoxShadow bs = fetch_boxshadow(cpi.specific_prim_index);
 

webrender/res/cs_box_shadow.vs.glsl

@@ -8,7 +8,7 @@
 // as text-shadow.
 
 void main(void) {
-    CachePrimitiveInstance cpi = fetch_cache_instance(gl_InstanceID);
+    CachePrimitiveInstance cpi = fetch_cache_instance();
     RenderTaskData task = fetch_render_task(cpi.render_task_index);
     TextRun text = fetch_text_run(cpi.specific_prim_index);
     Glyph glyph = fetch_glyph(cpi.sub_index);

webrender/res/cs_text_run.vs.glsl

@@ -65,6 +65,15 @@ uniform sampler2D sData64;
 uniform sampler2D sData128;
 uniform sampler2D sResourceRects;
 
+// Instanced attributes
+in int aGlobalPrimId;
+in int aPrimitiveAddress;
+in int aTaskIndex;
+in int aClipTaskIndex;
+in int aLayerIndex;
+in int aElementIndex;
+in ivec2 aUserData;
+
 #define get_fetch_uv(i, vpi)  ivec2(vpi * (i % (WR_MAX_VERTEX_TEXTURE_WIDTH/vpi)), i / (WR_MAX_VERTEX_TEXTURE_WIDTH/vpi))
 
 ivec2 get_fetch_uv_1(int index) {
@@ -90,10 +99,6 @@ struct Layer {
     vec4 screen_vertices[4];
 };
 
-layout(std140) uniform Data {
-    ivec4 int_data[WR_MAX_UBO_VECTORS];
-};
-
 Layer fetch_layer(int index) {
     Layer layer;
 
@@ -280,66 +285,38 @@ struct PrimitiveInstance {
     ivec2 user_data;
 };
 
-PrimitiveInstance fetch_instance(int index) {
+PrimitiveInstance fetch_prim_instance() {
     PrimitiveInstance pi;
 
-    int offset = index * 2;
-
-    ivec4 data0 = int_data[offset + 0];
-    ivec4 data1 = int_data[offset + 1];
-
-    pi.global_prim_index = data0.x;
-    pi.specific_prim_index = data0.y;
-    pi.render_task_index = data0.z;
-    pi.clip_task_index = data0.w;
-    pi.layer_index = data1.x;
-    pi.sub_index = data1.y;
-    pi.user_data = data1.zw;
+    pi.global_prim_index = aGlobalPrimId;
+    pi.specific_prim_index = aPrimitiveAddress;
+    pi.render_task_index = aTaskIndex;
+    pi.clip_task_index = aClipTaskIndex;
+    pi.layer_index = aLayerIndex;
+    pi.sub_index = aElementIndex;
+    pi.user_data = aUserData;
 
     return pi;
 }
 
-struct BlurCommand {
-    int task_id;
-    int src_task_id;
-    int dir;
-};
-
-BlurCommand fetch_blur(int index) {
-    BlurCommand blur;
-
-    int offset = index * 1;
-
-    ivec4 data0 = int_data[offset + 0];
-
-    blur.task_id = data0.x;
-    blur.src_task_id = data0.y;
-    blur.dir = data0.z;
-
-    return blur;
-}
-
 struct CachePrimitiveInstance {
     int global_prim_index;
     int specific_prim_index;
     int render_task_index;
     int sub_index;
-    ivec4 user_data;
+    ivec2 user_data;
 };
 
-CachePrimitiveInstance fetch_cache_instance(int index) {
+CachePrimitiveInstance fetch_cache_instance() {
     CachePrimitiveInstance cpi;
 
-    int offset = index * 2;
-
-    ivec4 data0 = int_data[offset + 0];
-    ivec4 data1 = int_data[offset + 1];
+    PrimitiveInstance pi = fetch_prim_instance();
 
-    cpi.global_prim_index = data0.x;
-    cpi.specific_prim_index = data0.y;
-    cpi.render_task_index = data0.z;
-    cpi.sub_index = data0.w;
-    cpi.user_data = data1;
+    cpi.global_prim_index = pi.global_prim_index;
+    cpi.specific_prim_index = pi.specific_prim_index;
+    cpi.render_task_index = pi.render_task_index;
+    cpi.sub_index = pi.sub_index;
+    cpi.user_data = pi.user_data;
 
     return cpi;
 }
@@ -357,11 +334,9 @@ struct Primitive {
     ivec2 user_data;
 };
 
-Primitive load_primitive(int index) {
+Primitive load_primitive_custom(PrimitiveInstance pi) {
     Primitive prim;
 
-    PrimitiveInstance pi = fetch_instance(index);
-
     prim.layer = fetch_layer(pi.layer_index);
     prim.tile = fetch_tile(pi.render_task_index);
     prim.clip_area = fetch_clip_area(pi.clip_task_index);
@@ -377,6 +352,13 @@ Primitive load_primitive(int index) {
     return prim;
 }
 
+Primitive load_primitive() {
+    PrimitiveInstance pi = fetch_prim_instance();
+
+    return load_primitive_custom(pi);
+}
+
+
 // Return the intersection of the plane (set up by "normal" and "point")
 // with the ray (set up by "ray_origin" and "ray_dir"),
 // writing the resulting scaler into "t".

webrender/res/prim_shared.glsl

@@ -4,7 +4,7 @@
  * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
 
 void main(void) {
-    Primitive prim = load_primitive(gl_InstanceID);
+    Primitive prim = load_primitive();
     Gradient gradient = fetch_gradient(prim.prim_index);
 
     VertexInfo vi = write_vertex(prim.local_rect,

webrender/res/ps_angle_gradient.vs.glsl

@@ -7,8 +7,8 @@ struct Blend {
     ivec4 src_id_target_id_op_amount;
 };
 
-Blend fetch_blend(int index) {
-    PrimitiveInstance pi = fetch_instance(index);
+Blend fetch_blend() {
+    PrimitiveInstance pi = fetch_prim_instance();
 
     Blend blend;
     blend.src_id_target_id_op_amount = ivec4(pi.user_data.x,
@@ -20,7 +20,7 @@ Blend fetch_blend(int index) {
 }
 
 void main(void) {
-    Blend blend = fetch_blend(gl_InstanceID);
+    Blend blend = fetch_blend();
     Tile src = fetch_tile(blend.src_id_target_id_op_amount.x);
     Tile dest = fetch_tile(blend.src_id_target_id_op_amount.y);
 

webrender/res/ps_blend.vs.glsl

@@ -4,7 +4,7 @@
  * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
 
 void main(void) {
-    Primitive prim = load_primitive(gl_InstanceID);
+    Primitive prim = load_primitive();
     Border border = fetch_border(prim.prim_index);
     int sub_part = prim.sub_index;