Thank you. On Fri, Oct 17, 2014 at 9:48 AM, Juergen Sauermann < juergen.sauerm...@t-online.de> wrote:
> Hi David, > > you may see a non-zero startup cost even though an operation shows cost 0. > This is because startup cost is averaged over all monadic or all dyadic > operations. > > The reason for zero startup cost on the products is most likely due to a > reorg of the counter numbers. > I forgot to update *ScalarBenchmark**.**apl*; fixed in *SVN **489*. > > In general the *OP *and *STAT* columns in *ScalarBenchmark**.**apl* > should match the *]PSTAT* command, e.g if: > > * ]pstat 38* > > *╔═════════════════╦════════════╤══════════╤══════════╤══════════╤══════════╗* > *║ A f.g B ║ 0 │ 0 │ 0 │ 0 │ > 0 ║* > > *╚═════════════════╩════════════╧══════════╧══════════╧══════════╧══════════╝* > > then the *STAT* number for *f.g* should be *38* in > *ScalarBenchmark,apl.* > > /// Jürgen > > > > On 10/17/2014 05:58 PM, David B. Lamkins wrote: > > I'm seeing zero start-up costs for inner and outer products when running > ScalarBenchmark.apl. > > ===================== Mat1_IRC +.× Mat1_IRC =============================== > > Benchmarking start-up cost for Mat1_IRC +.× Mat1_IRC ... > Length Sequ Cycles Para Cycles Linear Sequ Linear Para > ====== =========== =========== =========== =========== > 25 0 0 0 0 > 25 0 0 0 0 > 25 0 0 0 0 > 25 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 4 0 0 0 0 > 4 0 0 0 0 > 4 0 0 0 0 > 1 0 0 0 0 > > regression line sequential: 0 + 0×N cycles > regression line parallel: 0 + 0×N cycles > > ===================== Vec1_IRC ∘.× Vec1_IRC =============================== > > Benchmarking start-up cost for Vec1_IRC ∘.× Vec1_IRC ... > Length Sequ Cycles Para Cycles Linear Sequ Linear Para > ====== =========== =========== =========== =========== > 25 0 0 0 0 > 25 0 0 0 0 > 25 0 0 0 0 > 25 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 16 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 9 0 0 0 0 > 4 0 0 0 0 > 4 0 0 0 0 > 4 0 0 0 0 > 1 0 0 0 0 > > regression line sequential: 0 + 0×N cycles > regression line parallel: 0 + 0×N cycles > > > But then in the summary section -- just above ]PSTAT -- I see: > > -------------- Mat1_IRC +.× Mat1_IRC -------------- > average sequential startup cost: 359 cycles > average parallel startup cost: 832 cycles > per item cost sequential: 0 cycles > per item cost parallel: 0 cycles > parallel break-even length: not reached > > -------------- Vec1_IRC ∘.× Vec1_IRC -------------- > average sequential startup cost: 359 cycles > average parallel startup cost: 832 cycles > per item cost sequential: 0 cycles > per item cost parallel: 0 cycles > parallel break-even length: not reached > > > Here the startup costs are nonzero, but the per-item costs are all zero. > > This doesn't look right... Or am I missing something? > > In case it might shed some additional light, here's the final section of > the ]PSTAT output. The rest looks reasonable except for epsilon-underbar, > which reports all zeroes. > > ╔═════════════════╦════════════╤══════════╤══════════╤══════════╤══════════╗ > ║ Function ║ │ N │ ⌀ VLEN │ ⌀ cycles │ cyc÷VLEN ║ > ╟─────────────────╫────────────┼──────────┼──────────┼──────────┼──────────╢ > ║ f B overhead ║ 18446744003448130869 │ 283 │ 1993 │ > 34818579233229 │ 17466187239 ║ > ║ A f B overhead ║ 18446743954621671206 │ 1114 │ 84 │ > 1447585256996 │ 17221844259 ║ > ║ scalar B ║ 130198460 │ 283 │ 3873 │ 460065 │ 118 ║ > ║ A scalar B ║ 91680403 │ 1114 │ 949 │ 82298 │ 86 ║ > ║ clone B ║ 233950109373 │ 75391125 │ 131 │ 3103 │ 23 ║ > ║ A f.g B ║ 911702656227 │ 40046 │ 163 │ 22766385 │ 139671 ║ > ║ A ∘.g B ║ 9809803882 │ 121 │ 1000000 │ 81072759 │ 81 ║ > ║ A ⍴ B ║ 9071 │ 3 │ 27 │ 3023 │ 111 ║ > ║ PrintBuffer(B) ║ 135760049 │ 1168 │ 25 │ 116232 │ 4649 ║ > ╚═════════════════╩════════════╧══════════╧══════════╧══════════╧══════════╝ > > > > > -- "The secret to creativity is knowing how to hide your sources." Albert Einstein http://soundcloud.com/davidlamkins http://reverbnation.com/lamkins http://reverbnation.com/lcw http://lamkins-guitar.com/ http://lamkins.net/ http://successful-lisp.com/
