1 Tips for performance tuning on a specific architecture:
3 1a. Choose the optimal digit size (intDsize). This is fundamental. On 32-bit
4 platforms intDsize=32 is best. On 64-bit platforms intDsize=64 may be
5 better, especially if there is a 64x64-bit multiplication in hardware.
7 1b. Alternatively, tune GMP. When GMP is used, CLN's digit size (intDsize) has
8 to match GMP's limb size (sizeof(mp_limb_t)). There is nothing to do at the
9 CLN side: The configure script will take care of intDsize automatically.
11 2. The break-even points between several algorithms for the same task
12 have to be determined experimentally, in the order given below:
15 cl_DS_mul.cc karatsuba_threshold
16 cl_DS_mul.cc function cl_fftm_suitable
18 cl_DS_div.cc function cl_recip_suitable
20 cl_2DS_recip.cc recip2adic_threshold
22 cl_2DS_div.cc function cl_recip_suitable
24 cl_DS_sqrt.cc function cl_recipsqrt_suitable
25 cl_LF_sqrt.cc "if (len > ...)"
27 cl_I_gcd.cc cl_gcd_double_threshold
28 binary->decimal conversion:
29 cl_I_to_digits.cc cl_digits_div_threshold
31 cl_LF_pi.cc best of 4 algorithms
33 cl_F_expx.cc factor limit_slope of isqrt(d)
34 cl_R_exp.cc inside function exp
35 cl_R_ln.cc inside function ln
37 cl_LF_eulerconst.cc function compute_eulerconst
39 cl_F_sinx.cc factor limit_slope of isqrt(d)
40 cl_R_sin.cc inside function sin
41 cl_R_cos.cc inside function cos
42 cl_R_cossin.cc inside function cl_cos_sin
43 cl_F_sinhx.cc factor limit_slope of isqrt(d)
44 cl_R_sinh.cc inside function sinh
45 cl_R_cosh.cc inside function cosh
46 cl_R_coshsinh.cc inside function cl_cosh_sinh
47 cl_F_atanx.cc factor limit_slope of isqrt(d)
48 cl_F_atanx.cc inside function atanx
49 cl_F_atanhx.cc factor limit_slope of isqrt(d)
50 cl_F_atanhx.cc inside function atanhx