Richard Henderson <r...@twiddle.net> writes: I'm not 100% sure how to interpret the speed results, since they're not really in the same units as before. But comparing to mpn_mul_2 they're not encouraging. Am I doing something wrong with the testing or did the 1.64 cyc/limb I got for addmul_8 not really come over to this test? But at least it pases make check... On systems where we cannot reach a cycle counter, and where the clock drops for an idle system, speed is not reliable.
Passing -p10000000 might help. parma$ tune/speed -p10000000 -C -s 1-40 mpn_mul_basecase mpn_sqr_basecaseclock_gettime is 1.000ns accurate overhead 51.51 cycles, precision 10000000 units of 1.00e-09 secs, CPU freq 1694.10 MHz mpn_mul_basecase mpn_sqr_basecase 1 249.1149 #10.9920 2 17.9868 #8.4933 3 18.6519 #8.4938 4 18.4860 #11.2413 5 21.5837 #12.9901 6 21.4839 #15.1557 7 24.4103 #15.9892 8 25.4805 #17.4870 9 29.2007 #17.7648 10 34.2352 #19.0867 11 35.1140 #19.6235 12 36.5579 #20.6528 13 39.5108 #21.4463 14 40.8981 #25.1966 15 43.8365 #26.2477 16 45.5297 #26.6692 17 48.7875 #27.6847 18 50.1309 #27.7579 19 53.3292 #28.9786 20 54.9094 #29.0294 21 58.3364 #30.7874 22 59.6392 #30.8415 23 62.9972 #33.1067 24 64.4941 #32.7251 25 68.0297 #34.8538 26 69.2953 #34.4750 27 72.7605 #37.2694 28 74.1959 #36.3672 29 77.8029 #39.0739 30 79.0416 #38.3379 31 82.5836 #41.6473 32 83.9695 #41.1573 33 87.6311 #43.5729 34 88.8488 #42.8273 35 92.4461 #46.3504 36 93.7906 #46.0767 37 97.4999 #48.3971 38 98.6904 #47.9918 39 102.3349 #51.2183 40 103.6475 #50.2492 -- Torbjörn _______________________________________________ gmp-devel mailing list gmp-devel@gmplib.org http://gmplib.org/mailman/listinfo/gmp-devel