Formed in 2009, the Archive Team (not to be confused with the archive.org Archive-It Team) is a rogue archivist collective dedicated to saving copies of rapidly dying or deleted websites for the sake of history and digital heritage. The group is 100% composed of volunteers and interested parties, and has expanded into a large amount of related projects for saving online and digital history.
gh-115178: Add Counts of UOp Sequences to pystats #115181
Conversation
Make optimization_stats.opcode an array of pointers to UOpStats objects, instead of a member array. Add a *next_stats[512] field to the UOpStats struct Add _init_pystats function to initialize these new member structs Call _init_pystats inside _PyConfig_Write
Add print_uop_sequence function to specialize.c to output stored chains of uops
Sets the maximum depth of UOP sequences to track. Defaults to 2.
JeffersGlass
requested review from
sethmlarson,
barneygale,
a team,
FFY00,
ethanfurman,
berkerpeksag,
erlend-aasland,
pablogsal,
kumaraditya303,
iritkatriel,
ericsnowcurrently and
markshannon
as code owners
JeffersGlass
requested review from
ericvsmith,
gpshead,
pganssle,
abalkin,
vsajip and
Fidget-Spinner
as code owners
|
Once again this seems to have automatically requested reviews from a bunch of Code Owners... not sure if that's due to the merge from Main or from touching |
I think your merge from main caught a bunch of unrelated changes (look at the diff of the PR in the files tab). |
This usually does the trick: $ git switch <feature-branch>
$ # hack away
$ git commit
$ git switch main
$ git pull # <= make sure main is up to date
$ git switch -
$ git merge --no-ff main -m "Sync with main" |
erlend-aasland
removed request for
iritkatriel,
ericsnowcurrently,
markshannon,
gvanrossum,
gpshead,
pganssle,
abalkin,
vsajip and
Fidget-Spinner
|
@JeffersGlass, here's a recipe for fixing this: $ git switch uop-sequence-count
$ git revert 03db7a5e950639cbbaea38ad4521f55ab39b1495
$ git switch main
$ git pull
$ git switch -
$ git merge --no-ff mainSee also: |
This reverts commit 03db7a5.
|
Thank you both! @erlend-aasland your fix recipe worked a treat - I will be more careful about merging from main/pushing in the future. |
|
A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated. Once you have made the requested changes, please leave a comment on this pull request containing the phrase |
|
Thank you for the corrections. I have made the requested changes; please review again. |
Adds the ability to track pairs, triples, and longer sequences of UOps to the
pystatsstatistics.Currently, this only counts these statistics for the non-JIT tier 2 interpreter. I intend to include tracking in the JIT in a follow-up PR.
To try it out, run:
./configure --enable-pystats ./make mkdir /tmp/py_stats PYTHON_UOPS=1 ./python -X pystats -c $'x = 0\nfor i in range(1000):\n x+=i'The output file at
/tmp/py_stats/???.txtwill contain a new section with UOp Sequence counts:Tools/scripts/summarize_stats.pyhas also been expanded to encompass these new stats, with a new section:Counts for top 100 UOp Sequences of Length 2
Finally, a new environment variable,
PYTHONSTATS_UOPDEPTHcan be set to any integer to record sequences of length longer than two. For example:PYTHONSTATS_UOPDEPTH=4 PYTHON_UOPS=1 ./python -X pystats -c 'x = 0; for i in range(1000): x+=iYields:
Counts for top 100 UOp Sequences of Length 2
Counts for top 100 UOp Sequences of Length 3
Counts for top 100 UOp Sequences of Length 4
pystats #115178EDIT: I see this has automatically requested a review from a huge number of people. This is my first time submitting a PR to CPython - please do forgive me if I've goofed this process up somehow.
📚 Documentation preview 📚: https://cpython-previews--115181.org.readthedocs.build/