IO performance regression in 7.2 vs 7.0

The following program:

main = interact id

runs about a third slower in 7.2.1 compared to 7.0.3 (and current 7.3 is about the same as 7.2.1).

With 7.2.1:

$ ls -l stuff                             
-rw-rw-r-- 1 simonmar GHC 128464904 2011-10-07 14:59 stuff
$ ./cat <stuff >/dev/null +RTS -s      
   8,357,007,912 bytes allocated in the heap
     793,616,736 bytes copied during GC
          83,984 bytes maximum residency (1 sample(s))
          24,936 bytes maximum slop
               1 MB total memory in use (0 MB lost due to fragmentation)

                                    Tot time (elapsed)  Avg pause  Max pause
  Gen  0     15966 colls,     0 par    0.94s    0.94s     0.0001s    0.0001s
  Gen  1         1 colls,     0 par    0.00s    0.00s     0.0004s    0.0004s

  INIT    time    0.00s  (  0.00s elapsed)
  MUT     time    5.71s  (  5.71s elapsed)
  GC      time    0.94s  (  0.94s elapsed)
  EXIT    time    0.00s  (  0.00s elapsed)
  Total   time    6.65s  (  6.65s elapsed)

and with 7.0.3:

$ ./cat <stuff >/dev/null +RTS -s      
./cat +RTS -s 
   5,299,570,680 bytes allocated in the heap
     250,350,336 bytes copied during GC
          53,232 bytes maximum residency (1 sample(s))
          30,776 bytes maximum slop
               1 MB total memory in use (0 MB lost due to fragmentation)

  Generation 0: 10086 collections,     0 parallel,  0.50s,  0.50s elapsed
  Generation 1:     1 collections,     0 parallel,  0.00s,  0.00s elapsed

  INIT  time    0.00s  (  0.00s elapsed)
  MUT   time    4.38s  (  4.38s elapsed)
  GC    time    0.50s  (  0.50s elapsed)
  EXIT  time    0.00s  (  0.00s elapsed)
  Total time    4.88s  (  4.88s elapsed)

I strongly suspect this has to do with the changes to the encoding machinery in 7.2.1.

In binary mode, the slowdown is even worse. The program:

import System.IO
main = do
  hSetBinaryMode stdin True
  hSetBinaryMode stdout True
  getContents >>= putStr

With 7.0.3 this runs in 3.35s, with 7.2.1 it takes 5.49s, a slowdown of 63%.

After fixing this, we need to add a performance regression test.

Trac metadata

Trac field	Value
Version	7.2.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	high
Resolution	Unresolved
Component	libraries/base
Test case
Differential revisions
BlockedBy
Related
Blocking
CC	batterseapower
Operating system
Architecture