[Trac #749] X performance regression between build225 and build226

[Zarro Boogs per Child]

#749: X performance regression between build225 and build226
--------------------+-------------------------------------------------------
 Reporter:  ajax    |       Owner:  ajax     
     Type:  defect  |      Status:  new      
 Priority:  normal  |   Milestone:  Untriaged
Component:  distro  |    Keywords:           
--------------------+-------------------------------------------------------
 {{{
 1: build225.cooked
 2: build226.cooked

     1              2           Operation
 --------   -----------------   -----------------
 410000.0   412000.0 (  1.00)   Dot
 405000.0   408000.0 (  1.01)   1x1 rectangle
 155000.0   159000.0 (  1.03)   10x10 rectangle
   5810.0    16800.0 (  2.89)   100x100 rectangle
    268.0     1030.0 (  3.84)   500x500 rectangle
  10400.0     3660.0 (  0.35)   Char in 80-char aa line (Charter 10)
   5380.0     1390.0 (  0.26)   Char in 30-char aa line (Charter 24)
  11000.0     3110.0 (  0.28)   Char in 80-char aa line (Courier 12)
  10700.0     4540.0 (  0.42)   Char in 80-char rgb line (Charter 10)
   5540.0     1820.0 (  0.33)   Char in 30-char rgb line (Charter 24)
  10800.0     4030.0 (  0.37)   Char in 80-char rgb line (Courier 12)
  19500.0    19900.0 (  1.02)   Copy 10x10 from window to window
    607.0      805.0 (  1.33)   Copy 100x100 from window to window
     25.8       34.3 (  1.33)   Copy 500x500 from window to window
  29800.0    30000.0 (  1.01)   Copy 10x10 from pixmap to window
   3140.0     3630.0 (  1.16)   Copy 100x100 from pixmap to window
    176.0      212.0 (  1.20)   Copy 500x500 from pixmap to window
  24600.0    24100.0 (  0.98)   Copy 10x10 from window to pixmap
    600.0     1190.0 (  1.98)   Copy 100x100 from window to pixmap
     24.9       51.4 (  2.06)   Copy 500x500 from window to pixmap
  43700.0    42900.0 (  0.98)   Copy 10x10 from pixmap to pixmap
   2870.0     3060.0 (  1.07)   Copy 100x100 from pixmap to pixmap
    123.0      132.0 (  1.07)   Copy 500x500 from pixmap to pixmap
  26000.0    25600.0 (  0.98)   PutImage 10x10 square
    886.0      863.0 (  0.97)   PutImage 100x100 square
     41.8       42.5 (  1.02)   PutImage 500x500 square
  20700.0    22400.0 (  1.08)   ShmPutImage 10x10 square
   2960.0     3450.0 (  1.17)   ShmPutImage 100x100 square
    175.0      210.0 (  1.20)   ShmPutImage 500x500 square
   1600.0     1580.0 (  0.99)   GetImage 10x10 square
    383.0      383.0 (  1.00)   GetImage 100x100 square
     17.4       17.5 (  1.01)   GetImage 500x500 square
 769000.0   753000.0 (  0.98)   X protocol NoOperation
 }}}

 Many things got faster, but text got much much worse.  The relevant change
 here was enabling MMX.  The MMX solid and blit routines are fairly simple,
 but the composite code is a pile of inlining that's probably going wrong
 due to -Os.

-- 
Ticket URL: <http://dev.laptop.org/ticket/749>
One Laptop Per Child <http://laptop.org/>