Stop using `glBlitFramebuffer()` for scaling tasks.

[pcwalton]

The driver is allowed to read outside of the source rect, and this can cause
incorrect results.

This patch makes preserve-3d.png test have some minor rounding differences in
its colors from the previous implementation, so I regenerated the reference
image.

Closes #3088.

r? @gw3583

---

This change is 

[gw3583]

Looks good to me, we should do a try run for safety though.

[gw3583]

CI warnings error:

error: unused import: `ScalingInfo`
  --> webrender/src/renderer.rs:74:53
   |
74 | use tiling::{Frame, RenderTarget, RenderTargetKind, ScalingInfo, TextureCacheRenderTarget};
   |                                                     ^^^^^^^^^^^
   |
   = note: `-D unused-imports` implied by `-D warnings`

[pcwalton]

@gw3583 OK, I fixed that issue. I still can't push to try yet because my hg access was disabled for inactivity, though.

[nical]

try push: https://treeherder.mozilla.org/#/jobs?repo=try&revision=b15f823f8cb73908e770d973b90316a5b4689d6b

[kvark]

Code looks good, minus a few nits. The try push is all red.

Reviewed 7 of 7 files at r1.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @pcwalton)

---

webrender/res/cs_scale.glsl, line 49 at r1 (raw file):

    vec2 uv0 = src_rect.p0 / texture_size;
    vec2 uv1 = (src_rect.p0 + src_rect.size) / texture_size;
    vUv.xy = mix(uv0, uv1, aPosition.xy);

nit: could just do vUv.xy = (src_rect.p0 + src_rect.size * aPosition.xy) / texture_size;

---

webrender/src/renderer.rs, line 2864 at r1 (raw file):

        }

        match source {

nit: let shader = match source {...}; then shader.bind(...);

[pcwalton]

I don't understand what's going on on try.

[pcwalton]

Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @pcwalton)

---

webrender/res/cs_scale.glsl, line 49 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

nit: could just do vUv.xy = (src_rect.p0 + src_rect.size * aPosition.xy) / texture_size;

This code was copied from cs_blur.glsl. I'll rewrite the corresponding code in cs_blur.glsl when I get around to updating that file too.

[pcwalton]

Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @kvark and @pcwalton)

---

webrender/src/renderer.rs, line 2864 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

nit: let shader = match source {...}; then shader.bind(...);

It won't borrow check that way because of the lack of NLL (self.draw_instanced_batch is &mut self). Do you still want me to make the change with an extra {} level to work around lack of NLL or is that too ugly?

[kvark]

Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @kvark)

---

webrender/src/renderer.rs, line 2864 at r1 (raw file):

Previously, pcwalton (Patrick Walton) wrote…

It won't borrow check that way because of the lack of NLL (self.draw_instanced_batch is &mut self). Do you still want me to make the change with an extra {} level to work around lack of NLL or is that too ugly?

I see, nvm then :) we'll clean this up later

[pcwalton]

Kicked off a try build: https://treeherder.mozilla.org/#/jobs?repo=try&revision=2ef0e6a9b321a1197d5eeee9aac514b4ab960f85

I verified that the browser works, so I'm not sure why the previous build failed.

[pcwalton]

Doh, I didn't realize new shaders have to be explicitly deinitialized now. I added a comment so that others won't be as confused as I was in the future and kicked off a new try build: https://treeherder.mozilla.org/#/jobs?repo=try&revision=7c4bc7e1093d447d091c8531d398df7f66b616cb

[gw3583]

Looks like a lot of orange on that last try build - is that the correct try run?

[pcwalton]

@gw3583 This is the most recent try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=bc356ef3f3f4b15b324607b374e285052b95c0d2&selectedJob=200912030

[gw3583]

@bors-servo r+ 🚀

[bors-servo]

📌 Commit a559a71 has been approved by gw3583

[bors-servo]

⌛ Testing commit a559a71 with merge fa2214b...

[bors-servo]

💔 Test failed - status-taskcluster

[gw3583]

CI failures:

REFTEST INFO | 308 passing, 3 failing
Reftests with unexpected results:
	reftests/boxshadow/inset-border-radius.yaml == reftests/boxshadow/inset-border-radius.png
	reftests/filters/filter-blur.yaml == reftests/filters/filter-blur.png
	reftests/snap/preserve-3d.yaml == reftests/snap/preserve-3d.png

[pcwalton]

The test failures look like imperceptible differences.

[pcwalton]

@bors-servo: r=gw3583

[bors-servo]

📌 Commit 3c50dc4 has been approved by gw3583

[bors-servo]

⌛ Testing commit 3c50dc4 with merge 43ffae1...

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: gw3583
Pushing 43ffae1 to master...