Opened 5 years ago

Last modified 5 years ago

#8948 new bug

Profiling report resolution too low

Reported by: tibbe Owned by:
Priority: normal Milestone:
Component: Profiling Version: 7.6.3
Keywords: Cc: hvr
Operating System: Unknown/Multiple Architecture: Unknown/Multiple
Type of failure: None/Unknown Test Case:
Blocked By: Blocking:
Related Tickets: Differential Rev(s):
Wiki Page:

Description

When running criterion benchmarks under profiling, the criterion internals usually dominate the report. That means that the code you actually care about ends up with very low percentages:

                                                            individual     inherited
COST CENTRE              MODULE             no.  entries  %time %alloc   %time %alloc
...
 Full                    Data.HashMap.Base  360   725180    0.2    0.5     0.4    0.9
  Leaf/neq               Data.HashMap.Base  364     5151    0.0    0.0     0.0    0.0
  BitmapIndexed/update   Data.HashMap.Base  362   227250    0.1    0.2     0.2    0.3
   BitmapIndexed/insert  Data.HashMap.Base  365    78376    0.0    0.0     0.0    0.0
   Leaf/neq              Data.HashMap.Base  363   132411    0.0    0.1     0.0    0.1
  BitmapIndexed/insert   Data.HashMap.Base  361   191597    0.0    0.1     0.0    0.1
 BitmapIndexed/update    Data.HashMap.Base  357     4545    0.0    0.0     0.0    0.0
  BitmapIndexed/insert   Data.HashMap.Base  359     3030    0.0    0.0     0.0    0.0
  Leaf/neq               Data.HashMap.Base  358     1515    0.0    0.0     0.0    0.0
 BitmapIndexed/insert    Data.HashMap.Base  356     1414    0.0    0.0     0.0    0.0
 Leaf/neq                Data.HashMap.Base  355      101    0.0    0.0     0.0    0.0
 Empty                   Data.HashMap.Base  354      101    0.0    0.0     0.0    0.0

If you're interested in comparing the relative time spent in these functions, that's not really possible as these numbers have too low accuracy.

Change History (3)

comment:1 Changed 5 years ago by tibbe

Using +RTS -p -P to output the raw ticks doesn't really help either, as several of the above functions have zero ticks (which is odd) and the rest have 1-29. Is it possible to sample more often?

comment:2 Changed 5 years ago by hvr

Cc: hvr added

comment:3 Changed 5 years ago by carter

Ive hit this same issue of criterion + profiling having zero information myself. Its definitely a frustrating issue.

Note: See TracTickets for help on using tickets.