
   These are some (very approximate) timings produced by the test program,
   running on my 486 DX33 with a Tseng ET4000 graphics card.

   All times are the number of operations per second, without accounting
   for the overhead of looping and random number generation. Except for 
   the clear(), these were all writing to video memory. Using regular
   memory instead should (especially with the djgpp version) result in a 
   big speed increase.


   <operation>          <djgpp>     <borland>


   putpixel():          116395         66049

   hline():              57581         34494

   vline():              41065         29133

   line():                7683          7332

   rect():               15808         12761

   rectfill():            2305          2023

   triangle():            1158           932

   circle():              1830          1542

   circlefill():          1115           679

   textout(12 chars):     1802          1736

   fmul():              356984        133386

   fdiv():              316115         32975

   fsqrt():              30204          9188

   fsin():             3018275        338547

   fcos():             2553903        345610

   ftan():             3018248        342044

   fasin():             101579         59814

   facos():             101271         60580

   fatan():             102739         66887

   fatan2():             79366         21607

   clear(320x200 bitmap):  477           243

   blit(320x200 and
   three 64x32 bitmaps):   153           151

   16x16 sprite:
      opaque:            28867         20464
      word aligned:      35227         25400
      masked:            13310          9852

   32x32 sprite:
      opaque:             9312          7467
      word aligned:      11590         10443
      masked:             3797          2974
