[Milhouse]

New OpenELEC Gotham build: #0301
(Supercedes previous build)

Code:

# uname -a
Linux rpi512 3.13.5 #1 PREEMPT Sat Mar 1 17:34:11 GMT 2014 armv6l GNU/Linux

# vcgencmd version
Feb 28 2014 16:43:10
Copyright (c) 2012 Broadcom
version a9ba6435869933cdfd1ad5631b6b9a97affd8806 (clean) (release)

# lsb_release
OpenELEC (Milhouse) - Version: devel-20140301173301-r17837-g599837b

Based on tip of XBMC master (86a5608, changelog) and tip of OpenELEC master (599837b, changelog) with the following modifications:

• Includes newclock3 commits (except for 60c02d0 which I've replaced with a static spinner to avoid hammering the GUI)
  
• Includes kernel 3.13.y-next commits (see below for details on enabling FIQ_FSM patch)
  
• Excludes the OpenELEC fernetmenta patches (due to conflict with newclock3)
  
• Default setting for "Show RSS Feed" changed to disabled
  
• Disabled "Total Duration" in Confluence (see build #0221 for details)
  
• ALSA is enabled and any suitable hardware should be available in XBMC Audio Settings
  
• Hifiberry fix: CONFIG_DMA_BCM2708=m
  
• Includes PR4235: Move to SQLite 3080301
  
• Includes PR4275: Use mmap to accelerate SQLite file accesses
  
• Includes PR4290: Fix memory leakage in python bindings
  
• Includes PR4296: DVD various fixes related to libdvdnav issues and skip to menu
  

Build Highlights:

1. newclock3: [omxplayer] Allow small audio packets to be concatenated to make better use of audio fifo
   
   Some audio codecs produce small packets which causes a high overhead when submitting to GPU, and doesn't make full use of GPU side buffering. TrueHD in particular can produce packets with 40 samples (so 1200 packets per second) which causes very high overhead.
   
   What this aims to do is to concatenate audio packets until they approach the ideal audio packet size, and then deal with the awkwardness of concatenated planar formats.
   
   Quoting popcornmix:
   

   When not using passthough, this has quite a significant performance benefit for me. I can reduce arm clock significantly before a stream starts buffering.
   It also starts to make TrueHD audio usable (some TrueHD streams are playing correctly with overclock whereas before it was impossible).
   
   You can get test TrueHD streams here:
   http://www.demo-world.eu/trailers/high-d...ailers.php
   
   There is a chance this could cause audio glitches or sync problems, but it's behaving for my test streams.
   
   TrueHD is still problematic. In a standalone test AC3 takes about 17% of the (stock frequency) ARM's CPU. TrueHD takes about 57%, so it's much more expensive.
   We've got Ben to look at optimising the codec. It will always be worse than AC3/DTS, but in conjunction with this patch it may become feasible to play TrueHD on an overclocked Pi.

   ---

2. newclock3: [cec] Experimental: Add suppression and auto-repeat to CEC.
   
   Probably not usable yet, but should be harmless without the advancedsettings.
   
   Code:

   <cecsuppress> # brief time during which subsequent presses will be ignored (ms)
<cecdelay>    # brief time until remote starts repeating (ms)
<cecrepeat>   # brief interval remote repats at (ms)
<cecrelease>  # brief time before remote is considered released (ms)
   
   

---

Additional Testing Notes:

1. Testers should try adding the following entry to their advancedsettings.xml:
   
   Code:

   <advancedsettings>
  <video>
    <defaultplayer>dvdplayer</defaultplayer>
    <defaultdvdplayer>dvdplayer</defaultdvdplayer>
  </video>
</advancedsettings>
   and report if it is better/worse than omxplayer. You can still play files with omxplayer using the context-menu "Play using... OMXPlayer".
   

   ---

2. The following settings are no longer required in config.txt and should be removed:
   
   Code:

   no_hdmi_resample=1
hdmi_stream_channels=1
   no_resample_audio is now a default, and hdmi_stream_channels is switched based on audio content. For the time being when using passthrough, 2.0 speaker layout should continue to be used (you will still get 5.1 with AC3/DTS).
   

   ---

3. To enable the FIQ_FSM patch, add the following options to the end of your current line in /flash/cmdline.txt:
   
   Code:

   dwc_otg.fiq_enable=1 dwc_otg.fiq_fsm_enable=1 dwc_otg.fiq_fsm_mask=0x3
   The FIQ_FSM announcement thread is here. There are currently no known problems with this patch, and it needs testing and will be the default in the future. Enabling the cmdline.txt options can improve network performance as well as USB devices. To determine if FIQ_FSM is enabled or not, run:
   
   Code:

   dmesg | grep FSM
   and the line:
   
   Code:

   [    1.524680] dwc_otg: FIQ split-transaction FSM enabled
   will confirm if FIQ_FSM is enabled or disabled.