Recherche avancée

Sur d’autres sites (9533)

• Reduce HLS latency from +30 seconds

  4 juin 2014, par Rick

  Ubuntu 12.04

  nginx 1.2.4

  avconv -version

  avconv version 0.8.10-4:0.8.10-0ubuntu0.12.04.1, Copyright (c) 2000-2013 the Libav developers
  
    built on Feb  6 2014 20:56:59 with gcc 4.6.3
  
  avconv 0.8.10-4:0.8.10-0ubuntu0.12.04.1
  
  libavutil    51. 22. 2 / 51. 22. 2
  
  libavcodec   53. 35. 0 / 53. 35. 0
  
  libavformat  53. 21. 1 / 53. 21. 1
  
  libavdevice  53.  2. 0 / 53.  2. 0
  
  libavfilter   2. 15. 0 /  2. 15. 0
  
  libswscale    2.  1. 0 /  2.  1. 0
  
  libpostproc  52.  0. 0 / 52.  0. 0

  I’m using avconv and nginx to create an HLS stream but right now my latency is regularly well over 30 seconds. After much reading I am aware that HLS has built in latency and that 10s is expected and even preferred but 30s seems quite extreme.

  I’ve seen a lot of discussion on the nginx-rtmp google group, this thread in particular had a lot of suggestions. I have attempted to reduce solve my issue by reducing the hls_fragment and the hls_playlist_length but they haven’t had a significant effect.

  nginx.conf :

  #user  nobody;
  
  worker_processes  1;
  
  
  
  error_log  logs/error.log debug;
  
  
  
  events {
  
      worker_connections  1024;
  
  }
  
  
  
  http {
  
      include       mime.types;
  
      default_type  application/octet-stream;
  
  
  
      sendfile        on;
  
      keepalive_timeout  65;
  
  
  
      server {
  
          listen       8888;
  
          server_name  localhost;
  
  
  
          add_header 'Access-Control-Allow-Origin' "*";
  
  
  
          location /hls {
  
              types {
  
                  application/vnd.apple.mpegurl m3u8;
  
                  video/mp2t ts;
  
              }
  
              root /tmp;
  
          }
  
  
  
          # rtmp stat
  
          location /stat {
  
              rtmp_stat all;
  
              rtmp_stat_stylesheet stat.xsl;
  
          }
  
          location /stat.xsl {
  
              # you can move stat.xsl to a different location
  
              root /usr/build/nginx-rtmp-module;
  
          }
  
  
  
          # rtmp control
  
          location /control {
  
              rtmp_control all;
  
          }
  
  
  
          error_page   500 502 503 504  /50x.html;
  
          location = /50x.html {
  
              root   html;
  
          }
  
      }
  
  }
  
  
  
  rtmp {
  
      server {
  
          listen 1935;
  
          ping 30s;
  
          notify_method get;
  
  
  
          application myapp {
  
              live on;
  
  
  
              hls on;
  
              hls_path /tmp/hls;
  
              hls_base_url http://x.x.x.x:8888/hls/;
  
              hls_sync 2ms;
  
              hls_fragment 2s;
  
  
  
              #hls_variant _low BANDWIDTH=160000;
  
              #hls_variant _mid BANDWIDTH=320000;
  
              #hls_variant _hi BANDWIDTH=640000;
  
          }
  
      }
  
  }

  avconv command :

  avconv -r 30 -y -f image2pipe -codec:v mjpeg -i - -f flv -codec:v libx264 -profile:v baseline -preset ultrafast -tune zerolatency -an -f flv rtmp://127.0.0.1:1935/myapp/mystream

• Bit-field badness

  30 janvier 2010, par Mans — Compilers, Optimisation

  Consider the following C code which is based on an real-world situation.

  struct bf1_31 
      unsigned a:1 ;
      unsigned b:31 ;
   ;
  void func(struct bf1_31 *p, int n, int a)
  
  
  
      int i = 0 ;
  
      do 
  
          if (p[i].a)
  
              p[i].b += a ;
  
       while (++i < n) ;
  
  
  

  How would we best write this in ARM assembler ? This is how I would do it :
  

  func :
          ldr     r3,  [r0], #4
          tst     r3,  #1
          add     r3,  r3,  r2,  lsl #1
          strne   r3,  [r0, #-4]
          subs    r1,  r1,  #1
          bgt     func
          bx      lr
  

  The add instruction is unconditional to avoid a dependency on the comparison. Unrolling the loop would mask the latency of the ldr instruction as well, but that is outside the scope of this experiment.

  Now compile this code with gcc -march=armv5te -O3 and watch in horror :

  func :
          push    r4
          mov     ip, #0
          mov     r4, r2
  loop :
          ldrb    r3, [r0]
          add     ip, ip, #1
          tst     r3, #1
          ldrne   r3, [r0]
          andne   r2, r3, #1
          addne   r3, r4, r3, lsr #1
          orrne   r2, r2, r3, lsl #1
          strne   r2, [r0]
          cmp     ip, r1
          add     r0, r0, #4
          blt     loop
          pop     r4
          bx      lr
  

  This is nothing short of awful :

  • The same value is loaded from memory twice.
  • A complicated mask/shift/or operation is used where a simple shifted add would suffice.
  • Write-back addressing is not used.
  • The loop control counts up and compares instead of counting down.
  • Useless mov in the prologue ; swapping the roles or r2 and r4 would avoid this.
  • Using lr in place of r4 would allow the return to be done with pop {pc}, saving one instruction (ignoring for the moment that no callee-saved registers are needed at all).

  Even for this trivial function the gcc-generated code is more than twice the optimal size and slower by approximately the same factor.

  The main issue I wanted to illustrate is the poor handling of bit-fields by gcc. When accessing bitfields from memory, gcc issues a separate load for each field even when they are contained in the same aligned memory word. Although each load after the first will most likely hit L1 cache, this is still bad for several reasons :

  • Loads have typically two or three cycles result latency compared to one cycle for data processing instructions. Any bit-field can be extracted from a register with two shifts, and on ARM the second of these can generally be achieved using a shifted second operand to a following instruction. The ARMv6T2 instruction set also adds the SBFX and UBFX instructions for extracting any signed or unsigned bit-field in one cycle.
  • Most CPUs have more data processing units than load/store units. It is thus more likely for an ALU instruction than a load/store to issue without delay on a superscalar processor.
  • Redundant memory accesses can trigger early flushing of store buffers rendering these less efficient.

  No gcc bashing is complete without a comparison with another compiler, so without further ado, here is the ARM RVCT output (armcc --cpu 5te -O3) :

  func :
          mov     r3, #0
          push    r4, lr
  loop :
          ldr     ip, [r0, r3, lsl #2]
          tst     ip, #1
          addne   ip, ip, r2, lsl #1
          strne   ip, [r0, r3, lsl #2]
          add     r3, r3, #1
          cmp     r3, r1
          blt     loop
          pop     r4, pc
  

  This is much better, the core loop using only one instruction more than my version. The loop control is counting up, but at least this register is reused as offset for the memory accesses. More remarkable is the push/pop of two registers that are never used. I had not expected to see this from RVCT.

  Even the best compilers are still no match for a human.

• Anomalie #1993 : Modifications perdues lors d’un retour en arrière (entité no-store)

  27 janvier 2011, par Redmine Admin

  L’entité no-store est livrée à deux endroits : ecrire/inc/headers.php:167 : header("Cache-Control : no-store, no-cache, must-revalidate") ; ecrire/public/balises.php:862 : .’ ?php header("Cache-Control : no-store, no-cache, must-revalidate") ; ?’ Suffit-il de la supprimer là (...)