[ruby-cvs:69232] shyouhei:r62051 (trunk): move ADD_PC around to optimize PC manipluiations

[shyouhei at ruby-lang.org]

shyouhei	2018-01-26 15:30:58 +0900 (Fri, 26 Jan 2018)

  New Revision: 62051

  https://svn.ruby-lang.org/cgi-bin/viewvc.cgi?view=revision&revision=62051

  Log:
    move ADD_PC around to optimize PC manipluiations
    
    This commit introduces new attribute handles_flame and if that is
    _not_ the case, places ADD_PC right after INC_SP.  This improves
    locality of PC manipulations to prevents unnecessary register spill-
    outs. As a result, it reduces the size of vm_exec_core from 32,688
    bytes to 32,384 bytes on my machine.
    
    Speedup is very faint, but certain.
    
    -----------------------------------------------------------
    benchmark results:
    minimum results in each 3 measurements.
    Execution time (sec)
    name    before  after
    so_ackermann     0.476  0.464
    so_array         0.742  0.728
    so_binary_trees  5.493  5.466
    so_concatenate   3.619  3.395
    so_count_words   0.190  0.184
    so_exception     0.249  0.239
    so_fannkuch      0.994  0.953
    so_fasta         1.369  1.374
    so_k_nucleotide  1.111  1.111
    so_lists         0.470  0.481
    so_mandelbrot    2.059  2.050
    so_matrix        0.466  0.465
    so_meteor_contest        2.712  2.781
    so_nbody         1.154  1.204
    so_nested_loop   0.852  0.846
    so_nsieve        1.636  1.623
    so_nsieve_bits   2.073  2.039
    so_object        0.616  0.584
    so_partial_sums  1.464  1.481
    so_pidigits      1.075  1.082
    so_random        0.321  0.317
    so_reverse_complement    0.555  0.558
    so_sieve         0.495  0.490
    so_spectralnorm  1.634  1.627
    
    Speedup ratio: compare with the result of `before' (greater is better)
    name    after
    so_ackermann    1.025
    so_array        1.019
    so_binary_trees 1.005
    so_concatenate  1.066
    so_count_words  1.030
    so_exception    1.040
    so_fannkuch     1.043
    so_fasta        0.996
    so_k_nucleotide 1.000
    so_lists        0.978
    so_mandelbrot   1.004
    so_matrix       1.001
    so_meteor_contest       0.975
    so_nbody        0.959
    so_nested_loop  1.007
    so_nsieve       1.008
    so_nsieve_bits  1.017
    so_object       1.056
    so_partial_sums 0.989
    so_pidigits     0.994
    so_random       1.014
    so_reverse_complement   0.996
    so_sieve        1.010
    so_spectralnorm 1.004

  Modified files:
    trunk/insns.def
    trunk/tool/ruby_vm/models/bare_instructions.rb
    trunk/tool/ruby_vm/views/_insn_entry.erb