Merge lp:~alan-griffiths/mir/partial-fix-for-1532202 into lp:mir

[Alan Griffiths]

android: mga::HWCFallbackGLRenderer::render() - Don't use a gl program to do nothing

android: mga::HWCFallbackGLRenderer::render() - Don't use a gl program to do nothing

This isn't a complete solution to the linked bug - the hack mentioned in lp:1532202/comments/12 gets better performance than this. But it gives a noticeable improvement and I'm at EOD.

I'll look further tomorrow, and also try to figure if there's a way to automate testing.

[Kevin DuBois (kdub)]

Clearing the contents appropriately will correct artifacts present in the buffer from previous renders, so we need some way to know whether a clear is needed or not.
So maybe something like:

if (renderlist.empty() && clear) return;
(where clear is a variable that gets set after the first clear, and unset when the renderables are drawn)
would work.

Still thinking what could be the root cause. If the swapping (ie, producing gpu traffic) is slowing us down, maybe its a bus traffic jam issue. (and if thats the case, maybe the n7 wouldn't have the problem)

[Kevin DuBois (kdub)]

review: Needs Fixing

[PS Jenkins bot (ps-jenkins)]

FAILED: Continuous integration, rev:3238
http://jenkins.qa.ubuntu.com/job/mir-ci/5988/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-vivid-i386-build/5499
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-vivid-amd64-build/4406
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-vivid-touch/5455/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/312
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/312/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/312
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/312/artifact/work/output/*zip*/output.zip
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-vivid-armhf/5452/console

Click here to trigger a rebuild:
http://s-jenkins.ubuntu-ci:8080/job/mir-ci/5988/rebuild

review: Needs Fixing (continuous-integration)

[PS Jenkins bot (ps-jenkins)]

FAILED: Continuous integration, rev:3238
http://jenkins.qa.ubuntu.com/job/mir-ci/5990/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-vivid-i386-build/5501
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-vivid-amd64-build/4408
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-vivid-touch/5457/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/314
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/314/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/314
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/314/artifact/work/output/*zip*/output.zip
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-vivid-armhf/5454/console

Click here to trigger a rebuild:
http://s-jenkins.ubuntu-ci:8080/job/mir-ci/5990/rebuild

review: Needs Fixing (continuous-integration)

[PS Jenkins bot (ps-jenkins)]

FAILED: Continuous integration, rev:3238
http://jenkins.qa.ubuntu.com/job/mir-ci/5993/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-android-vivid-i386-build/5504
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-clang-vivid-amd64-build/4411
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-vivid-touch/5460/console
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/317
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-amd64-ci/317/artifact/work/output/*zip*/output.zip
    SUCCESS: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/317
        deb: http://jenkins.qa.ubuntu.com/job/mir-xenial-i386-ci/317/artifact/work/output/*zip*/output.zip
    FAILURE: http://jenkins.qa.ubuntu.com/job/mir-mediumtests-builder-vivid-armhf/5457/console

Click here to trigger a rebuild:
http://s-jenkins.ubuntu-ci:8080/job/mir-ci/5993/rebuild

review: Needs Fixing (continuous-integration)

[Daniel van Vugt (vanvugt)]

Seems like a partial fix indeed. I'm hesitant to call it a hack because that's really the style of all graphics optimizations.

Although if you're skipping most of the rendering strain, and particularly skipping the call to context.swap_buffers() then you should put the return at the top of the function and so skip the clears too. Because those clears are now unused (their effects are never seen on screen).

Even better, avoid calling render() with an empty list. Because DefaultDisplayBufferCompositor::composite() should be calling renderer->suspend() instead.

review: Needs Fixing

Unmerged revisions

3238. By Alan Griffiths on 2016-01-12

Don't use a gl program to do nothing

=== modified file 'src/platforms/android/server/hwc_fallback_gl_renderer.cpp'
--- src/platforms/android/server/hwc_fallback_gl_renderer.cpp	2015-10-06 02:30:52 +0000
+++ src/platforms/android/server/hwc_fallback_gl_renderer.cpp	2016-01-12 17:57:14 +0000
@@ -102,8 +102,6 @@
 void mga::HWCFallbackGLRenderer::render(
     RenderableList const& renderlist, geom::Displacement offset, SwappingGLContext const& context) const
 {
-    glUseProgram(*program);
-
     /* NOTE: some HWC implementations rely on the framebuffer target layer
      * being cleared to transparent black. eg, in mixed-mode composition,
      * krillin actually arranges the fb_target in the topmost level of its
@@ -113,6 +111,10 @@
     glClearColor(0.0, 0.0, 0.0, 0.0);
     glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
     glClear(GL_COLOR_BUFFER_BIT);
+
+    if (renderlist.empty()) return;
+
+    glUseProgram(*program);
     glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
     glEnableVertexAttribArray(position_attr);
     glEnableVertexAttribArray(texcoord_attr);