From: Bruno Haible <bruno@cl...>  20040830 14:10:33

Sam wrote: > You said that on some rare hardware GMP is faster than CLN/CLISP. > Would it make sense to link CLISP against GMP? Here are benchmarks of CLN, built on a PowerPC machine a) with GMP b) without GMP (i.e. with the same lowlevel routines as in clisp, namely for PowerPC the portable C version). N = 100 Function CLN1.1.8 CLN1.1.8 no gmp gmp4.1.2 multiplication 0.0000069 0.0000066 division 0.0000047 0.0000047 isqrt 0.0000024 0.0000024 gcd 0.0000163 0.0000163 N = 1000 Function CLN1.1.8 CLN1.1.8 no gmp gmp4.1.2 multiplication 0.000197 0.000192 division 0.000179 0.000179 isqrt 0.000021 0.000021 gcd 0.000603 0.000600 N = 10000 Function CLN1.1.8 CLN1.1.8 no gmp gmp4.1.2 multiplication 0.0082 0.0081 division 0.0156 0.0155 isqrt 0.0009 0.0010 gcd 0.0449 0.0445 N = 100000 Function CLN1.1.8 CLN1.1.8 no gmp gmp4.1.2 multiplication 0.087 0.087 division 0.27 0.27 isqrt 0.132 0.133 gcd 3.28 3.26 You see that the timing differences are minimal. Bruno 