849555 - WebGLContext with NON power of two size is crashing some SGX530 drivers

Status: RESOLVED WONTFIX

(Core :: Graphics: CanvasWebGL, defect)

Description

[Ivaylo Dimitrov]

Opening http://www.randelshofer.ch/webgl/rubikscube/ results in http://tinypic.com/view.php?pic=smsfn5&s=6 . I played with different WebGL pages, and found that if canvas size of http://learningwebgl.com/lessons/lesson05/index.html is changed(by downloading the page locally) from 500x500 to 512x512, the content plays without problem. The other "good" combinations are 256x256, 256x512, 512x256. In general, if the canvas size is not power of 2, it either results in distorted canvas rendering or device hang.

Comment 1

[Oleg Romashin (:romaxa)]

Could be that non pow2 canvas causing non-pow2 FB/GLStreamingBuffer creation, which is not very friendly to some drivers...
IIUC that works fine with Non POW2 GLContext size and SW rendering (no GLStreaming buffer involved)

Comment 2

[Ivaylo Dimitrov]

Yes, with SW rendering(no GLStreaming buffer involved) those pages work with non ^2 canvas size, albeit with much lower fps.

Comment 3

[Kelsey Gilbert [:jgilbert]]

Can we get an about:support for the affected systems?

Comment 4

[Ivaylo Dimitrov]

  Application Basics

        Name
        Firefox

        Version
        22.0a1

        User Agent
        Mozilla/5.0 (X11; Linux armv7l; rv:22.0) Gecko/20130312 Firefox/22.0

        Build Configuration

          about:buildconfig

  Extensions

        Name

        Version

        Enabled

        ID

  Important Modified Preferences

      Name

      Value

        browser.cache.disk.capacity
        2048

        browser.cache.disk.smart_size_cached_value
        102400

        browser.cache.disk.smart_size.first_run
        false

        browser.cache.disk.smart_size.use_old_max
        false

        browser.places.importBookmarksHTML
        false

        browser.places.smartBookmarksVersion
        4

        browser.startup.homepage_override.buildID
        20130312133817

        browser.startup.homepage_override.mstone
        22.0a1

        dom.max_script_run_time
        0

        extensions.lastAppVersion
        22.0a1

        javascript.options.ion.content
        false

        javascript.options.ion.parallel_compilation
        false

        javascript.options.methodjit.chrome
        false

        javascript.options.methodjit.content
        false

        network.cookie.prefsMigrated
        true

        places.history.expiration.transient_current_max_pages
        3638

        plugin.disable_full_page_plugin_for_types
        application/pdf

        privacy.sanitize.migrateFx3Prefs
        true

  Graphics

        GPU Accelerated Windows
        0/1 Basic Blocked for your graphics driver version. Try updating your graphics driver to version <Anything with EXT_texture_from_pixmap support> or newer.

        WebGL Renderer
        Imagination Technologies -- PowerVR SGX 530

        AzureCanvasBackend
        cairo

        AzureContentBackend
        none

        AzureFallbackCanvasBackend
        none

  JavaScript

        Incremental GC
        true

  Accessibility

        Activated
        false

        Prevent Accessibility
        0

  Library Versions

        Expected minimum version

        Version in use

        NSPR
        4.9.6 Beta
        4.9.6 Beta

        NSS
        3.14.3.0 Basic ECC Beta
        3.14.3.0 Basic ECC Beta

        NSSSMIME
        3.14.3.0 Basic ECC Beta
        3.14.3.0 Basic ECC Beta

        NSSSSL
        3.14.3.0 Basic ECC Beta
        3.14.3.0 Basic ECC Beta

        NSSUTIL
        3.14.3.0 Beta
        3.14.3.0 Beta

Comment 5

[Ivaylo Dimitrov]

Now the bug 843015 is fixed, I am able to pass webgl conformance tests https://www.khronos.org/registry/webgl/conformance-suites/1.0.1/webgl-conformance-tests.html with SW rendering (GLStreamingBuffers disabled) with no(serious) issues. Could I have some support/help on what and where needs to be done, in order to have that bug resolved?

Comment 6

[Kelsey Gilbert [:jgilbert]]

How are you going about disabling streaming buffers? Just flipping the force-readback pref?

Comment 7

[Oleg Romashin (:romaxa)]

I think that is going to be the same as SW mode, but with additional upload of SW buffer into top level GL context, and will be 2x slower...
IIUC from last measurement simple GLScene with flipping colors 512x512 has ~15FPS difference SW mode comparing HW mode.

Comment 8

[Kelsey Gilbert [:jgilbert]]

Attachment: Disable EGLImage sharing on SGX530

Comment 9

[Kelsey Gilbert [:jgilbert]]

Attachment: Alternative: 'downscale' to nearest POT, but add FSAA

Here's an alternative.
One option is to fail to give anything NPOT on this hardware, which is spec-conformant. However, the spec guarantees that we fail 'width' and 'height' small, never upscaling them. We can, however, give back a smaller width and height, but boost the drawingBufferWidth/Height a la FSAA.

We probably want this patch instead, unless it looks unusably awful.

Comment 10

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 748263 [details] [diff] [review]
Alternative: 'downscale' to nearest POT, but add FSAA

Making some more changes.

Comment 11

[Kelsey Gilbert [:jgilbert]]

This hardware is from 2005 and should not be on the internet.