Issue metadata
Sign in to add a comment
|
Regression: MP4/H264//AVC/GIFV broken for specific videos on old intel GPU (3rd gen)
Reported by
colmj.or...@gmail.com,
May 6 2018
|
||||||||||||||||||||||
Issue descriptionUserAgent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/67.0.3396.30 Safari/537.36 Example URL: https://i.imgur.com/hTtH3pM.gifv Steps to reproduce the problem: 1. Use Linux on old laptop with 3rd gen Intel GMA 3150 or equivalent - requiring software (ffpmpeg) video decode. (Problem does not happen on my newer PC stick so it is hardware specific.) 2. Look at the MP4 gifv URL given above in Chrome 3. Only about one frame every five seconds is played What is the expected behavior? The linked GIFV should play reasonably smoothly just like many other larger GIFV/MP4s still play smnoothly on this platform, such as: https://i.imgur.com/1JCAtcY.gifv What went wrong? Only about one frame every five seconds is played. The default GPU compositor seems to hang on the GPU for seconds at a time for certain MP4s like this one. (Running chrome with GPU compositor disabled fixes the problem.) See attached Renderer trace and performance profile while problem occurs (separate runs). The problem seems to be specific to MP4/H264 - Certain youtube videos are broken when forced to play MP4 but OK when playing VP8/VP9. Problem still exists on latest chrome canary 68.0.3418.2 with new profile. The problem also occurs on a clean install of ubuntu 16.04.2. I realise you guys probaly don't want to spend time supporting old hardware and apparently don't even have the hardware to reproduce the problem. If you like you could just blacklist the GPU compositor on this generation GPU so that MP4 still works out of the box. Did this work before? Yes 60.0.3112.113 (broken in 61.0.3163.79) Is it a problem with Flash or HTML5? HTML5 Does this work in other browsers? Yes Chrome version: 67.0.3396.30 Channel: beta OS Version: Ubuntu 18.04 4.15.0-20-generic Flash Version: N/A Contents of chrome://gpu: Graphics Feature Status Canvas: Hardware accelerated CheckerImaging: Enabled Flash: Hardware accelerated Flash Stage3D: Hardware accelerated Flash Stage3D Baseline profile: Hardware accelerated Compositing: Hardware accelerated Multiple Raster Threads: Disabled Native GpuMemoryBuffers: Software only. Hardware acceleration disabled Rasterization: Software only. Hardware acceleration disabled Surface Synchronization: Enabled Video Decode: Unavailable Viz Service Display Compositor: Disabled WebGL: Hardware accelerated WebGL2: Hardware accelerated Driver Bug Workarounds adjust_src_dst_region_for_blitframebuffer clear_uniforms_before_first_program_use count_all_in_varyings_packing decode_encode_srgb_for_generatemipmap disable_framebuffer_cmaa disable_post_sub_buffers_for_onscreen_surfaces msaa_is_slow scalarize_vec_and_mat_constructor_args Problems Detected Accelerated video decode is unavailable on Linux: 137247 Disabled Features: accelerated_video_decode Clear uniforms before first program use on all platforms: 124764, 349137 Applied Workarounds: clear_uniforms_before_first_program_use Mesa drivers in Linux handle varyings without static use incorrectly: 333885 Applied Workarounds: count_all_in_varyings_packing Disable partial swaps on Mesa drivers (detected with GL_RENDERER): 339493 Applied Workarounds: disable_post_sub_buffers_for_onscreen_surfaces Always rewrite vec/mat constructors to be consistent: 398694 Applied Workarounds: scalarize_vec_and_mat_constructor_args On Intel GPUs MSAA performance is not acceptable for GPU rasterization: 527565 Applied Workarounds: msaa_is_slow Timer queries crash on Intel GPUs on Linux: 540543, 576991 Applied Workarounds: disable(GL_ARB_timer_query), disable(GL_EXT_timer_query) Use GL_INTEL_framebuffer_CMAA on ChromeOS: 535198 Applied Workarounds: disable_framebuffer_cmaa Disable partial swaps on Mesa drivers (detected with GL_VERSION): 339493 Applied Workarounds: disable_post_sub_buffers_for_onscreen_surfaces Decode and encode before generateMipmap for srgb format textures on os except macosx: 634519 Applied Workarounds: decode_encode_srgb_for_generatemipmap adjust src/dst region if blitting pixels outside read framebuffer on Linux Intel: 664740 Applied Workarounds: adjust_src_dst_region_for_blitframebuffer Disable KHR_blend_equation_advanced until cc shaders are updated: 661715 Applied Workarounds: disable(GL_KHR_blend_equation_advanced), disable(GL_KHR_blend_equation_advanced_coherent) Don't expose disjoint_timer_query extensions to WebGL: 808744 Raster is using a single thread. Disabled Features: multiple_raster_threads Native GpuMemoryBuffers have been disabled, either via about:flags or command line. Disabled Features: native_gpu_memory_buffers Viz service display compositor is not enabled by default. Disabled Features: viz_display_compositor Version Information Data exported 2018-05-06T13:48:14.398Z Chrome version Chrome/67.0.3396.30 Operating system Linux 4.15.0-20-generic Software rendering list URL https://chromium.googlesource.com/chromium/src/+/e2e233129ba8f46eef20ca735042858bfe04dc0d/gpu/config/software_rendering_list.json Driver bug list URL https://chromium.googlesource.com/chromium/src/+/e2e233129ba8f46eef20ca735042858bfe04dc0d/gpu/config/gpu_driver_bug_list.json ANGLE commit id f7d4f25ecb44 2D graphics backend Skia/67 937dcf426d268625252cdae8393f2270d16bb353- Command Line /usr/bin/google-chrome-beta --flag-switches-begin --flag-switches-end Driver Information Initialization time 443 In-process GPU false Passthrough Command Decoder false Direct Composition false Supports overlays false Sandboxed true GPU0 VENDOR = 0x8086, DEVICE= 0xa011 *ACTIVE* Optimus false AMD switchable false Driver vendor Mesa Driver version 18.0.0 Driver date Pixel shader version 1.20 Vertex shader version 1.20 Max. MSAA samples 0 Machine model name Machine model version GL_VENDOR Intel Open Source Technology Center GL_RENDERER Mesa DRI Intel(R) Pineview M GL_VERSION 1.4 Mesa 18.0.0-rc5 GL_EXTENSIONS GL_3DFX_texture_compression_FXT1 GL_AMD_shader_trinary_minmax GL_ANGLE_texture_compression_dxt3 GL_ANGLE_texture_compression_dxt5 GL_APPLE_object_purgeable GL_APPLE_packed_pixels GL_ARB_ES2_compatibility GL_ARB_clear_buffer_object GL_ARB_compressed_texture_pixel_storage GL_ARB_copy_buffer GL_ARB_debug_output GL_ARB_depth_texture GL_ARB_draw_buffers GL_ARB_draw_elements_base_vertex GL_ARB_explicit_attrib_location GL_ARB_explicit_uniform_location GL_ARB_fragment_program GL_ARB_framebuffer_object GL_ARB_get_program_binary GL_ARB_get_texture_sub_image GL_ARB_half_float_pixel GL_ARB_internalformat_query GL_ARB_invalidate_subdata GL_ARB_map_buffer_alignment GL_ARB_map_buffer_range GL_ARB_multi_bind GL_ARB_multisample GL_ARB_multitexture GL_ARB_pixel_buffer_object GL_ARB_point_parameters GL_ARB_point_sprite GL_ARB_program_interface_query GL_ARB_provoking_vertex GL_ARB_robustness GL_ARB_sampler_objects GL_ARB_separate_shader_objects GL_ARB_shader_objects GL_ARB_shading_language_100 GL_ARB_shadow GL_ARB_sync GL_ARB_texture_border_clamp GL_ARB_texture_compression GL_ARB_texture_cube_map GL_ARB_texture_env_add GL_ARB_texture_env_combine GL_ARB_texture_env_crossbar GL_ARB_texture_env_dot3 GL_ARB_texture_mirrored_repeat GL_ARB_texture_non_power_of_two GL_ARB_texture_rectangle GL_ARB_texture_storage GL_ARB_transpose_matrix GL_ARB_vertex_array_object GL_ARB_vertex_attrib_binding GL_ARB_vertex_buffer_object GL_ARB_vertex_program GL_ARB_vertex_shader GL_ARB_window_pos GL_ATI_blend_equation_separate GL_ATI_draw_buffers GL_ATI_separate_stencil GL_ATI_texture_env_combine3 GL_EXT_abgr GL_EXT_bgra GL_EXT_blend_color GL_EXT_blend_equation_separate GL_EXT_blend_func_separate GL_EXT_blend_minmax GL_EXT_blend_subtract GL_EXT_compiled_vertex_array GL_EXT_copy_texture GL_EXT_draw_range_elements GL_EXT_fog_coord GL_EXT_framebuffer_blit GL_EXT_framebuffer_object GL_EXT_gpu_program_parameters GL_EXT_multi_draw_arrays GL_EXT_packed_depth_stencil GL_EXT_packed_pixels GL_EXT_pixel_buffer_object GL_EXT_point_parameters GL_EXT_polygon_offset GL_EXT_provoking_vertex GL_EXT_rescale_normal GL_EXT_secondary_color GL_EXT_separate_specular_color GL_EXT_shadow_funcs GL_EXT_stencil_two_side GL_EXT_stencil_wrap GL_EXT_subtexture GL_EXT_texture GL_EXT_texture3D GL_EXT_texture_compression_dxt1 GL_EXT_texture_compression_s3tc GL_EXT_texture_cube_map GL_EXT_texture_edge_clamp GL_EXT_texture_env_add GL_EXT_texture_env_combine GL_EXT_texture_env_dot3 GL_EXT_texture_filter_anisotropic GL_EXT_texture_lod_bias GL_EXT_texture_object GL_EXT_texture_rectangle GL_EXT_texture_sRGB GL_EXT_texture_sRGB_decode GL_EXT_vertex_array GL_IBM_multimode_draw_arrays GL_IBM_rasterpos_clip GL_IBM_texture_mirrored_repeat GL_INGR_blend_func_separate GL_KHR_context_flush_control GL_KHR_debug GL_KHR_no_error GL_MESA_pack_invert GL_MESA_window_pos GL_MESA_ycbcr_texture GL_NV_blend_square GL_NV_light_max_exponent GL_NV_packed_depth_stencil GL_NV_texgen_reflection GL_NV_texture_env_combine4 GL_NV_texture_rectangle GL_OES_EGL_image GL_OES_read_format GL_S3_s3tc GL_SGIS_generate_mipmap GL_SGIS_texture_border_clamp GL_SGIS_texture_edge_clamp GL_SGIS_texture_lod GL_SUN_multi_draw_arrays Disabled Extensions GL_ARB_timer_query GL_EXT_timer_query GL_KHR_blend_equation_advanced GL_KHR_blend_equation_advanced_coherent Disabled WebGL Extensions EXT_disjoint_timer_query EXT_disjoint_timer_query_webgl2 Window system binding vendor SGI Window system binding version 1.4 Window system binding extensions GLX_ARB_create_context GLX_ARB_create_context_profile GLX_ARB_fbconfig_float GLX_ARB_framebuffer_sRGB GLX_ARB_multisample GLX_EXT_create_context_es_profile GLX_EXT_create_context_es2_profile GLX_EXT_fbconfig_packed_float GLX_EXT_framebuffer_sRGB GLX_EXT_import_context GLX_EXT_libglvnd GLX_EXT_texture_from_pixmap GLX_EXT_visual_info GLX_EXT_visual_rating GLX_MESA_copy_sub_buffer GLX_OML_swap_method GLX_SGI_make_current_read GLX_SGI_swap_control GLX_SGIS_multisample GLX_SGIX_fbconfig GLX_SGIX_pbuffer GLX_SGIX_visual_select_group GLX_INTEL_swap_event Window manager GNOME Shell XDG_CURRENT_DESKTOP GNOME GDMSESSION gnome Compositing manager Yes Direct rendering Yes Reset notification strategy 0x8261 GPU process crash count 0 System visual ID 32 RGBA visual ID 155 Compositor Information Tile Update Mode One-copy Partial Raster Enabled GpuMemoryBuffers Status ATC Software only ATCIA Software only DXT1 Software only DXT5 Software only ETC1 Software only R_8 Software only R_16 Software only RG_88 Software only BGR_565 Software only RGBA_4444 Software only RGBX_8888 Software only RGBA_8888 Software only BGRX_8888 Software only BGRX_1010102 Software only RGBX_1010102 Software only BGRA_8888 Software only RGBA_F16 Software only YVU_420 Software only YUV_420_BIPLANAR Software only UYVY_422 Software only Display(s) Information Info Display[9680100370939970] bounds=[0,0 1024x600], workarea=[0,0 1024x600], scale=1, external. Color space information {primaries:INVALID, transfer:INVALID, matrix:INVALID, range:INVALID} Bits per color component 8 Bits per pixel 24 Video Acceleration Information
,
May 8 2018
Hmm, that's strange. I would expect neither to play smoothly if it's a compositor issue. +hubbe, libearto, gpu
,
May 10 2018
Unable to reproduce the issue on chrome reported version 67.0.3396.30 using Ubuntu 17.10 7th generation laptop, as TE team doesn't have Linux Intel 3rd generation laptop to check and confirm the issue, hence adding TE-Hardware-Dependency component to this issue. Thanks!
,
May 11 2018
+piman, thomasanderson The trace clearly shows the display compositor blocked for ~2seconds at a time. Could you try grabbing a trace with gpu_decoder and gpu.service enabled to show the exact sequence of GL commands.
,
May 12 2018
Thanks for looking into this. As this old GPU only has mpeg2 decode, it's doing software decode for this issue so I added media trace as well as the two you requested. For comparison, I've also added two more traces of: - the same gifv in Chrome 60 where it works OK - from Chrome 67 for a different mp4 gifv (linked in the original report) that still works OK. Let me know if there's any other traces that would help.
,
May 14 2018
We need to kill GPU acceleration on these devices. They don't support GL2 (only 1.4+extensions) which is the minimum requirement for Chrome.
,
May 14 2018
The GPU does support GL 2.1 if I set two driconf parameters. One of the extensions this enables, occlusion_query, is actually just a stub but chromium doesn't use that with linux on intel anyway: https://chromiumcodereview.appspot.com/10381137/ Just for the record, the bug still happens with GL 2.1.
,
May 14 2018
I don't know if we are going to be able to fix it. A call to the GL driver is taking surprisingly long, and I don't know to what extent we can work around it. Your best bet may be to try to figure out what caused the regression, using the steps described at https://www.chromium.org/developers/bisect-builds-py to identify a range of CLs. It may be that we enabled a path that tickles the driver in the wrong way (different upload path? HDR? not sure...), and that we can disable as a workaround.
,
May 14 2018
Fair enough. I had tentatively looked at bisect but the need to add the h264 support to chromium makes it awkward. I don't suppose there's a full repository of chrome builds somewhere?
,
May 14 2018
Unfortunately, I don't think official builds are accessible publicly to allow for bisection.
,
May 14 2018
You should just be able to test with a webm like this: http://storage.googleapis.com/dalecurtis/buck720.webm
,
May 15 2018
Well I had been under the impression until now that it was just an h264 issue and indeed your VP9, like all the VP9s on youtube I'd tried, plays OK. After testing some more it turns out that some webm files do exhibit the problem such as this VP8 version of your video: http://dl3.webmfiles.org/big-buck-bunny_trailer.webm And then I found a VP9 which breaks too: https://base-n.de/webm/out9.webm So all formats are susceptible and yes that will make bisect possible. In general I guess it's looking more like an X/driver issue because for a while my machine got into a state where even videos that normally play OK were exhibiting the problem. In this state, every video I tried in all formats showed the problem even after restarting the browser. (chrome 60/gpu and 61+/soft-compositing still played OK though.) Only re-logging in to restart X got me back to the normal state where certain videos are always broken and others always good. I'll try to do the bisect over the next few days.
,
May 17 2018
The VP8 and VP9 videos in the previous message started to fail at different versions so there's two bisects but they both seem to point to colorspace related changes. The VP8 bisect came up with this commit list: https://chromium.googlesource.com/chromium/src/+log/837dc664cbbf86a9248212a02f45f25b26d080e2..f9064b1d8b6e1c8335da4635fff2f750b3b3d0d5 one of which is about colorspace for VP8 so that's the likely issue: https://chromium.googlesource.com/chromium/src/+/f9064b1d8b6e1c8335da4635fff2f750b3b3d0d5 especially as the VP9 bisect: https://chromium.googlesource.com/chromium/src/+log/75a35d664b7ee32f32f8856280434be4274cae7f..a606972d0f1f6c28af88f6f23b272bb0422988d1 also includes a colorspace change: https://chromium.googlesource.com/chromium/src/+/81dae6b3d8825169ce471ed190d7761731a3dae0 Both commits refer to this bug report: https://bugs.chromium.org/p/chromium/issues/detail?id=754986 Both of these changes are more recent than the h264 regression but the referred bug report is open since August so I suppose it's likely another colorspace related change that has broken some of my mp4 videos.
,
May 17 2018
=>hubbe for investigation then.
,
May 17 2018
Do you have an ICC profile installed? If so, could you attach it to this bug? If you go into about:flags and under "force color profile" select sRGB, does make any difference? I looked at the working and non-working example videos, and none of them are actually using color space information. This leads me to think that the actual pixel values are causing a problem somehow. One possibility is that the videos in question are causing a division by zero, or a pow() on a negative number in the shader code and the gpu or driver isn't handling it well. My color space changes could potentially make such a problem happen more often.
,
May 17 2018
I don't have an ICC profile installed. When I check the "device color profiles" tab in gnome settings, it says "unable to detect any device that can be color managed" if that's relevant - I guess this gpu/driver doesn't support color management? Forcing the color profile to sRGB in about flags didn't make any difference.
,
May 17 2018
A little pull-back on that: I installed gnome-color-manager and restarted X, and now the laptop screen shows up in color profile settings. Still no change to the broken videos though, with or without forced sRGB.
,
May 17 2018
+ccameron for ICC profiles
,
May 17 2018
As far as I can tell, both of the color space changes here would eliminate some color transformation steps in the fragment shader code. What's less clear is why less work would take more time.
,
May 18 2018
This GPU only has partial fragment shader support and anyway I've always been running it in GL 1.4 mode where it doesn't advertise fragment shader extension. I've also tested it previously in GL 2.1 mode with fragment shader support enabled (and double-checked just now) and it doesn't cure the problem. (I should perhaps have mentioned it during the earlier openGL discussion but I assumed it was just a 3D feature). Do your changes maybe require the fragment shader where previously it allowed software fallback? Some details on the level of fragment shader support (from https://dri.freedesktop.org/wiki/IntelGL/): "The 915-class hardware doesn't support many of the features of GLSL, notably dynamic branching. However, GLSL is quite useful from a programmer's perspective even if some features of the language aren't available. So, we've added the "fragment_shader" driconf option that enables limited GLSL program execution support. Among the features that don't work are if statements, non-unrolled for loops, subroutines, dFdx(), dFdy(), fwidth(), backface color, and noise(). Some of these should be fixed in the future, including flattening if statements to work, inlining more subroutines so they work, and backface color support. Additionally, the 915-class hardware has no support for occlusion queries, which are a requirement for GL 1.5. There doesn't appear to be a way to emulate them that would be useful for application developers. However, most applications using GLSL shaders require GL 2.0, as the GL 2.0 entrypoints are simpler and very nice to use. To support running those applications, a debug driconf option called "stub_occlusion_query" is provided that enables nonconformant ARB_occlusion_query support, that in combination with the ARB_fragment_shader support results in the driver reporting GL 2.0 support."
,
Jul 26
I don't think we have anything this old around for testing, so the best we can do is probably to blacklist GL compositing on this device. This is essentially equivalent to going into settings->advanced and turning off "Use hardware acceleration when available".
,
Jul 26
That's OK with me - I've been disabling GPU compositing via the command line flag anyway and I haven't noticed any other side-effect.
,
Oct 24
piman, regarding #6: If we nee GL2, why don't we just require that when we open the GL device?
,
Oct 24
@#23: we used to need GL1.4 on very early chromebook devices (which supported the set of extensions we needed at the time), so that's why we didn't apply teh restriction strictly. But yes, we should now, as those devices have been out of support for a long time.
,
Oct 24
I don't suppose there is a bug# for that?
,
Oct 24
As far as I'm concerned it'd be OK if you close with WontFix - the command line fix is adequate for me. There probably isn't anyone else still running linux on a 8 year old netbook anyway. In any case, the check for GL 2.0 alone wouldn't always be enough to catch the problem as you would also have to check for the limited shader that my card has in 2.0 mode.
,
Oct 24
In that case: WontFix |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by susan.boorgula@chromium.org
, May 6 2018