- User Since
- Feb 28 2017, 2:27 PM (51 w, 1 h)
Sun, Feb 11
Wonderful. Does the test suite pass with EXTRA_HC_OPTS=-dstg-lint? If so, should this be enabled by default?
Tue, Feb 6
Actually fix threads006
Add PROF_SPIN guards to CMM
Sorry for the spam @bgamari I don't actually have a windows box to test these on.
- Fixed reported avg/max pause time;
- Add TODO comment in mut_user_time_until.
The end of line whitespace cleanup was unintentional, but stand I by it.
- Remove counters from RTSStats;
- Update Note;
- Do counter increments in StgMiscClosures.cmm;
- Remove REVEIWERS comments.
Sun, Feb 4
I like the idea of having a local counter with an atomic update to keep the counter values accurate. Maybe a future change?
Yes, there's enough here already.
Tue, Jan 23
Thank you for your help!
Correct entered/exit counts.
Mon, Jan 22
Jan 16 2018
The diff has changed too much for these inline comment to be useful, but nevertheless.
Squashing this into D4302, with the promised Note
- Remove unnecessary fields from RTSSummaryStats
- Add init times to RTSStats
- Absorbed D4301.
- Added --internal-counters flag
- Added a Note
- Added several more counters.
- Several comments marked REVEIWERS where I'm unsure
Okay, I'll make the changes.
Jan 15 2018
I think I need to remove all conditional compilation in RTSStatsSummary to facilitate adding it to GHC.Stats, will that be ok?
Some thoughts on spin lock counters:
I am a bit concerned that the spin counter can grossly undercount spins. I think in the worst case we can undercount by a factor of (n_capabilites -1). Highlighting the worst case is one of the primary purposes of these counters!
They should be documented somewhere. Whether this be in a Note or the users guide, I think it's important that their semantics are defined somewhere besides the source code.
Jan 14 2018
Huge refactoring. (Almost) No long lines.
Jan 12 2018
Jan 11 2018
bytes_allocated -> bytes allocated
Jan 10 2018
Remove accidental inclusion from next diff.
Jan 7 2018
Looks great, thanks!
Jan 1 2018
Reworking See D3836
Reworking. See D3836
I'm going to rework this. We can get most of the benefit by reorganising DriverPipeline a bit to allow GhcMake to parallelise hscIncrementalCompile and hscGenHardCode.
Nov 29 2017
Nov 28 2017
EDIT: Oh, I see you've already got a note to do this.
Nov 21 2017
To continue in the tradition of bikeshedding likely trivialities: ghc environment files and perhaps ghc version?
There is GHC_PACKAGE_PATH.
I may be paranoid, but I am a little concerned that two ghc's may run with the same command line arguments at the same time. It seems easy to include the environment and pwd in the hash?
Nov 17 2017
This seems great to me, but then I don't understand why -O0 suppressed this originally.
Nov 15 2017
Adding -rtsopts to the compilation command and -G1 to the RTS options of the invocation command gives a much larger sampleValue discrepancy using the instrumentation here.
Nov 12 2017
This is wrong, see the ticket and D4184
This fixes the issue on the ticket, but may very well break something else. Relying on CI to validate.
Nov 8 2017
Let's see how this looks on gipedia.
Nov 7 2017
Gipedia diff here.
Nov 6 2017
Simplify a bit, and add more comments
Please hold off on merging this, I think I can simplify it somewhat.
Nov 5 2017