[Bug-apl] ScalarBenchmark for inner and outer products

[David B. Lamkins]

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 ║
╚═════════════════╩════════════╧══════════╧══════════╧══════════╧══════════╝