Difference between revisions of "User:Le Forgeron/vault/Compilation"

From POV-Wiki
Jump to navigation Jump to search
(Created page with "Tested on Ubuntu 13.04, 64 bits system (x86_64) My preference is to not be bothered with io-restrictions, but your mileage may vary. == gcc (Gnu) == * version 4.7.3 # ./con...")
 
 
(9 intermediate revisions by the same user not shown)
Line 1: Line 1:
−
Tested on Ubuntu 13.04, 64 bits system (x86_64)
+
Tested on Ubuntu 13.04, 64 bits system (x86_64), with boost 1.49.
  
 
My preference is to not be bothered with io-restrictions, but your mileage may vary.
 
My preference is to not be bothered with io-restrictions, but your mileage may vary.
 +
 +
== Ubuntu 13.10 & libboost 1.53 ==
 +
 +
Add LIBS="-lboost_system -lboost_thread"
  
 
== gcc (Gnu) ==
 
== gcc (Gnu) ==
−
* version 4.7.3
 
  
 +
=== Ubuntu 13.04, g++ version 4.7.3 ===
 
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions
 
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions
 +
 +
=== Ubuntu 13.10, g++ version 4.8.1 ===
 +
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread"
 +
 +
=== same result ===
  
 
   I/O restrictions:          disabled
 
   I/O restrictions:          disabled
Line 20: Line 29:
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -fno-enforce-eh-specs -s -O3 -ffast-math -march=native -pthread
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -fno-enforce-eh-specs -s -O3 -ffast-math -march=native -pthread
  
 +
== clang (LLVM) ==
 +
* version 3.2-1
 +
=== ubuntu 13.04 ===
 +
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=clang++ CC=clang
 +
=== ubuntu 13.10 ===
 +
* clang 3.2 has a bug in its distributed packages
 +
* clang 3.3 is ok
  
 +
====Fix to be able to find stddef.h (and other include files)====
  
 +
# cd /usr/lib/clang/3.2/
 +
# sudo ln -s /usr/lib/llvm-3.2/lib/clang/3.2/include
 +
==== configuration ====
 +
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread" CXX=clang++ CC=clang
  
−
== clang (LLVM) ==
+
=== same result ===
−
* version 3.2-1
 
−
 
 
−
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=clang++ CC=clang
 
  
 
   I/O restrictions:          disabled
 
   I/O restrictions:          disabled
Line 39: Line 57:
 
   Compiler version:    clang++ 4.2.1
 
   Compiler version:    clang++ 4.2.1
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -O3 -ffast-math -march=native -pthread
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -O3 -ffast-math -march=native -pthread
−
 
−
 
  
 
== icc (Intel) ==
 
== icc (Intel) ==
Line 46: Line 62:
  
 
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=icpc CC=icc CPPFLAGS=-I/usr/include/x86_64-linux-gnu/c++/4.7
 
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=icpc CC=icc CPPFLAGS=-I/usr/include/x86_64-linux-gnu/c++/4.7
 +
# ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread" CXX=icpc CC=icc CPPFLAGS=-I/usr/include/x86_64-linux-gnu/c++/4.8
  
  
Line 59: Line 76:
 
   Compiler version:    icpc 13.0.1
 
   Compiler version:    icpc 13.0.1
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -s -O3 -ip -xHost -pthread
 
   Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -s -O3 -ip -xHost -pthread
 +
 +
== Which one to choose ==
 +
 +
Remember that your mileage might vary.
 +
 +
Done on i7 980X (6 cores + HT), non-overclocked.
 +
 +
Personal conclusions:
 +
# Icc is best.
 +
# Clang is better than Gcc as long as no HT core are available.
 +
# Gcc is better than Clang when HT cores are available.
 +
# an HT core is worth only 0.2 to 0.25 a real core.
 +
 +
* X-axis: number of thread (-WT option of povray)
 +
* Y-axis: relative performance (the higher the better)
 +
 +
[[File:LeForgeron_i7_980_povperf.png]]

Latest revision as of 19:24, 22 October 2013

Tested on Ubuntu 13.04, 64 bits system (x86_64), with boost 1.49.

My preference is to not be bothered with io-restrictions, but your mileage may vary.

Ubuntu 13.10 & libboost 1.53

Add LIBS="-lboost_system -lboost_thread"

gcc (Gnu)

Ubuntu 13.04, g++ version 4.7.3

  1. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions

Ubuntu 13.10, g++ version 4.8.1

  1. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread"

same result

  I/O restrictions:          disabled
  X Window display:          enabled (using SDL)
  Supported image formats:   gif tga iff ppm pgm hdr png jpeg tiff openexr
  Unsupported image formats: -

Compilation settings:
  Build architecture:  x86_64-unknown-linux-gnu
  Built/Optimized for: x86_64-unknown-linux-gnu (using -march=native)
  Compiler vendor:     gnu
  Compiler version:    g++ 4.7
  Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -fno-enforce-eh-specs -s -O3 -ffast-math -march=native -pthread

clang (LLVM)

  • version 3.2-1

ubuntu 13.04

  1. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=clang++ CC=clang

ubuntu 13.10

  • clang 3.2 has a bug in its distributed packages
  • clang 3.3 is ok

Fix to be able to find stddef.h (and other include files)

  1. cd /usr/lib/clang/3.2/
  2. sudo ln -s /usr/lib/llvm-3.2/lib/clang/3.2/include

configuration

  1. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread" CXX=clang++ CC=clang

same result

  I/O restrictions:          disabled
  X Window display:          enabled (using SDL)
  Supported image formats:   gif tga iff ppm pgm hdr png jpeg tiff openexr
  Unsupported image formats: -

Compilation settings:
  Build architecture:  x86_64-unknown-linux-gnu
  Built/Optimized for: x86_64-unknown-linux-gnu (using -march=native)
  Compiler vendor:     gnu
  Compiler version:    clang++ 4.2.1
  Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -O3 -ffast-math -march=native -pthread

icc (Intel)

  • version 13.0.1
  1. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions CXX=icpc CC=icc CPPFLAGS=-I/usr/include/x86_64-linux-gnu/c++/4.7
  2. ./configure COMPILED_BY="Jérôme Grimbert <[email protected]>" --disable-io-restrictions LIBS="-lboost_system -lboost_thread" CXX=icpc CC=icc CPPFLAGS=-I/usr/include/x86_64-linux-gnu/c++/4.8


  I/O restrictions:          disabled
  X Window display:          enabled (using SDL)
  Supported image formats:   gif tga iff ppm pgm hdr png jpeg tiff openexr
  Unsupported image formats: -

Compilation settings:
  Build architecture:  x86_64-unknown-linux-gnu
  Built/Optimized for: x86_64-unknown-linux-gnu (using -xHost)
  Compiler vendor:     intel
  Compiler version:    icpc 13.0.1
  Compiler flags:      -pipe -Wno-multichar -Wno-write-strings -s -O3 -ip -xHost -pthread

Which one to choose

Remember that your mileage might vary.

Done on i7 980X (6 cores + HT), non-overclocked.

Personal conclusions:

  1. Icc is best.
  2. Clang is better than Gcc as long as no HT core are available.
  3. Gcc is better than Clang when HT cores are available.
  4. an HT core is worth only 0.2 to 0.25 a real core.
  • X-axis: number of thread (-WT option of povray)
  • Y-axis: relative performance (the higher the better)

LeForgeron i7 980 povperf.png

Retrieved from "https://wiki.povray.org/content?title=User:Le_Forgeron/vault/Compilation&oldid=8176"