inline times

Suresh Jagannathan suresh@research.nj.nec.com
Thu, 11 Feb 1999 17:42:28 -0500


Here are times taken on Vaikuntam with recursive inlining,
inlining continuations, and 3 rounds of simplification,
followed by inlining at a specified threshold.  The second
set of numbers are with the non-recursive small-inliner
currently used in the compiler.  The last table is the
percentage difference in times using the 0 threshold numbers
from the recursive inliner.  I'll gather size and chunk
statistics in the next run.

===================================

(Recursive inlining -- 3 rounds + threshold)

                      0      50
            fft    73.2    76.8
     barnes-hut    34.7    31.5
matrix-multiply    16.8    24.4
   knuth-bendix    17.6    17.5
        lexgen     56.5    59.2
   count-graphs    37.5    27.1
          life    126.7   131.5
         logic    101.1   104.1
     mandelbrot    52.7    52.7
         mlyacc    40.2    41.3
        nucleic    97.5    93.5
         simple    38.2    39.5
            tsp    70.8    67.5

(Current Appel-style inliner)

            fft    78.0    
     barnes-hut    58.8    
matrix-multiply    28.9    
   knuth-bendix    15.6    
         lexgen    95.8    
   count-graphs    25.7    
           life   129.3   
          logic   114.3   
     mandelbrot    51.4   
         mlyacc    47.3
         simple    37.7
         tsp      102.8


Ratios (Recursive/Current)

            fft    .93
     barnes-hut    .59
matrix-multiply    .58
   knuth-bendix   1.14
         lexgen    .58
   count-graphs   1.45
           life    .97
          logic    .88
     mandelbrot   1.02
         mlyacc    .86
         simple   1.02
            tsp    .68