Index of /anton/lvas/effizienz-abgaben/2010w/gruppe103/shortest-path

[ICO]NameLast modifiedSizeDescription

[PARENTDIR]Parent Directory  -  
[   ]Makefile2011-01-26 14:44 1.5K 
[DIR]img/2011-01-26 14:44 -  
[   ]input-bench-bigendian2011-01-26 14:44 448K 
[   ]input-bench-littleendian2011-01-26 14:44 448K 
[   ]input-ref-bigendian2011-01-26 14:44 90K 
[   ]input-ref-littleendian2011-01-26 14:44 90K 
[   ]output-bench2011-01-26 14:44 709K 
[   ]output-ref2011-01-26 14:44 169K 
[TXT]shortest-path.c2011-01-26 14:44 154K 

**Code style**

* Using tabs for indentation in source code (width = 2)

**Initial virtual cycle count**

    42761449 -O0
    181112540 -Os
    147486690 -O2
    144582390 -O3

**Changelog (newest on top)**

** Current -O optimisations **
di
SHA:
caching calculations of optimize_rewrite.
From 131800000 to 78788200 

SHA:
shortest-path.c:4075 : #define HASH_SIZE 512
a hashmap is used for fast access. bigger maps (512 instead of 256, more is not much faster) seems to lead to sigificant shorter linked lists.
From 135500000 to 131800000

SHA:
Deferred declaration of jcost 
From 138482000 to 138321240

Tried to inline printinst at shortest-path.c:4075
From 138482000 to 1396... cycles
Worsening possibly due to increased code size or strange optimisations only doable within functions

SHA:
Inlined ss_cost via macro
From 146757520 to 138482000 cycles

SHA:
Inlined transitions()
From 149510310 to 146757520 cycles

SHA:
Removed unnecessary ss_cost pointer
From 152035090 to 149510310 cycles

SHA:
reordering in transitions
From 154123780 to 152035090 cycles 

SHA:
-O compile with vanilla source
154123780 cycles

** Old -O3 optimisations **

SHA: ff64b7c0bb911d9b3d0054de24a635831668de90
Removed unnecessary ss_cost pointer
From 141876560 to 128666630 cycles

SHA: eccc116c017d6949147e2be2d26cc1475930d9f1
Reordering in inner loop of transitions()
From 144835200 to 141445610 cycles

SHA:
Pulled array indexing out of loop
From 128442880 to 127602780 cycles