[avr-libc-dev] libm: Speed or IEEE 754 functionality? Both!

[Dmitry K.]

Hi.

In two right columns time of operations after addition of
the IEEE 754 rules is shown. After little changes speed has
increased even, sometimes - considerably. Measurements are
lead on files of random numbers (on 500 pieces). Files are
picked up in view of features of this or that function.

Tuning/testing in progress...

              gcc/fp-bit      1.4.2          1.5
File          avr2  avr4    avr2  avr4    avr2  avr4
----------------------------------------------------
t_add.c        959   890     126   126     113   110
t_cmp.c        515   486      57    57      42    42
t_div.c       1458  1418     483   483     467   466
t_fix.c        369   358     189   189      68    68
t_float.c      841   745     264   264      59    59
t_frexp.c       -1    -1      34    33      34    33
t_ldexp.c       -1    -1      49    49      39    39
t_modf.c        -1    -1     775   775     395   391
t_mul.c       2179  2077     364   133     347   125
t1_acos.h       -1    -1    6527  4565    4886  2887
t1_asin.h       -1    -1    6380  4417    4758  2768
t1_atan.h       -1    -1    5513  3087    4852  2414
t1_ceil.h       -1    -1     285   285     118   118
t1_cos.h        -1    -1    6250  3827    3515  1804
t1_cosh.h       -1    -1    5545  3825    4858  2882
t1_exp.h        -1    -1    4586  2908    4262  2327
t1_floor.h      -1    -1     274   274     117   117
t1_log10.h      -1    -1    7277  4603    4820  2481
t1_log.h        -1    -1    6783  4111    4468  2348
t1_sin.h        -1    -1    6377  3963    3334  1641
t1_sinh.h       -1    -1    5500  3785    4773  2833
t1_sqrt.h       -1    -1    1815  1815     711   706
t1_tan.h        -1    -1    5881  3738    3843  1916
t1_tanh.h       -1    -1    5659  3992    4944  3022
t1_trunc.h      -1    -1      -1    -1     114   114
t2_atan2.h      -1    -1    6248  3825    5323  2872
t2_fmod.h       -1    -1    1545  1321     127   127
t2_pow.h        -1    -1   12336  7777    9201  4950

Dmitry.