bench: Don't pause/resume timer

[chfast]

Most of the benchmark cases don't have init memory at all - this is handled by a separate benchmark loop where init_memory() is not used at all.

For the remaining cases (mul256, eli_interpreter) the init memory is very small (max 105 bytes for an eli_interpreter input). It is not worth to pause and resume the timer as

PauseTiming()/ResumeTiming() are relatively heavyweight, and so their use should generally be avoided within each benchmark iteration, if possible.

fizzy/execute/blake2b/512_bytes_rounds_1_mean                     -0.0212         -0.0213            92            90            92            90
fizzy/execute/blake2b/512_bytes_rounds_16_mean                    -0.0179         -0.0179          1378          1354          1378          1354
fizzy/execute/ecpairing/onepoint_mean                             -0.0114         -0.0114        495862        490227        495865        490232
fizzy/execute/keccak256/512_bytes_rounds_1_mean                   +0.0041         +0.0040           109           110           109           110
fizzy/execute/keccak256/512_bytes_rounds_16_mean                  +0.0027         +0.0027          1605          1609          1605          1609
fizzy/execute/memset/256_bytes_mean                               -0.0623         -0.0631             8             8             8             8
fizzy/execute/memset/60000_bytes_mean                             -0.0150         -0.0150          1682          1657          1682          1657
fizzy/execute/mul256_opt0/input0_mean                             +0.0129         +0.0127            29            29            29            29
fizzy/execute/mul256_opt0/input1_mean                             +0.0121         +0.0119            29            29            29            29
fizzy/execute/sha1/512_bytes_rounds_1_mean                        -0.0259         -0.0260            99            97            99            97
fizzy/execute/sha1/512_bytes_rounds_16_mean                       -0.0108         -0.0108          1360          1345          1360          1345
fizzy/execute/sha256/512_bytes_rounds_1_mean                      -0.0108         -0.0109            98            97            98            97
fizzy/execute/sha256/512_bytes_rounds_16_mean                     -0.0089         -0.0089          1351          1339          1351          1339
fizzy/execute/micro/eli_interpreter/halt_mean                     -0.8269         -0.8277             1             0             1             0
fizzy/execute/micro/eli_interpreter/exec105_mean                  -0.0803         -0.0815             6             5             6             5
fizzy/execute/micro/factorial/10_mean                             -0.4272         -0.4293             1             1             1             1
fizzy/execute/micro/factorial/20_mean                             -0.2747         -0.2766             2             1             2             1
fizzy/execute/micro/fibonacci/24_mean                             -0.0064         -0.0064         10202         10136         10202         10136
fizzy/execute/micro/host_adler32/1_mean                           -0.7817         -0.7834             1             0             1             0
fizzy/execute/micro/host_adler32/100_mean                         -0.0625         -0.0632             7             6             7             6
fizzy/execute/micro/host_adler32/1000_mean                        +0.0036         +0.0035            63            63            63            63
fizzy/execute/micro/spinner/1_mean                                -0.9216         -0.9219             0             0             0             0
fizzy/execute/micro/spinner/1000_mean                             -0.0497         -0.0502            11            11            11            11

[axic]

Will this not make it harder to compare 0.2.0 against 0.1.0? Or did we introduce this pausing since 0.1.0?

[chfast]

Will this not make it harder to compare 0.2.0 against 0.1.0? Or did we introduce this pausing since 0.1.0?

I will need to backport bench changes to 0.1 before comparison.

[axic]

Assuming you create 0.1.1 with that backport?

[chfast]

Assuming you create 0.1.1 with that backport?

Yes, can be done like that when other release is ready. Or do it locally and never go back to 0.1.

[codecov]

Codecov Report

Merging #383 into master will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #383   +/-   ##
=======================================
  Coverage   99.32%   99.32%           
=======================================
  Files          42       42           
  Lines       12808    12814    +6     
=======================================
+ Hits        12722    12728    +6     
  Misses         86       86