Re: Data types (was: Re: Access the neighbors of an element)

[Paul Kienzle]

Still more timing data on the difference between using unsigned
ops and casting to double when calculating sums and products of
unsigned integers.

If anybody is using them, then it is probably worthwhile to
implement them using pure integer operations rather than
converting them to double and back, especially since it allows
us to support 64 bit integer ops correctly.

- Paul

Type    Op      DBLSAT  SAT  NOSAT (for MIPS)
CHAR    ADD      1.764 1.008 0.532
SHORT   ADD      1.427 1.006 0.529
LONG    ADD      1.777 0.629 0.416
LLONG   ADD      2.458 0.630 0.416
CHAR    MULT     1.490 1.489 1.322
SHORT   MULT     1.336 2.834 1.298
LONG    MULT     1.728 2.214 1.411
LLONG   MULT     2.061 2.218 2.229

Type    Op      DBLSAT  SAT  NOSAT (for PPC)
CHAR    ADD      3.119 0.731 0.598
SHORT   ADD      3.322 0.674 0.520
LONG    ADD      4.507 0.420 0.318
LLONG   ADD     26.995 1.293 0.823
CHAR    MULT     2.699 0.785 0.672
SHORT   MULT     2.647 0.975 0.772
LONG    MULT     2.640 1.176 0.622
LLONG   MULT    19.662 1.863 1.631

Type    Op      DBLSAT  SAT  NOSAT (for X86 32)
CHAR    ADD     11.115 1.480 0.911
SHORT   ADD      9.955 1.594 1.336
LONG    ADD     19.034 1.381 0.863
LLONG   ADD     42.416 2.722 2.602
CHAR    MULT     5.343 2.311 1.045
SHORT   MULT     4.192 1.856 1.452
LONG    MULT    10.320 1.842 1.011
LLONG   MULT    29.523 6.033 5.220 (4.243 for LITTLEENDIAN_32 SAT)