Skip to content

gh-109329: Support for basic pystats for Tier 2#109913

Merged
brandtbucher merged 17 commits intopython:mainfrom
mdboom:tier2-stats
Oct 4, 2023
Merged

gh-109329: Support for basic pystats for Tier 2#109913
brandtbucher merged 17 commits intopython:mainfrom
mdboom:tier2-stats

Conversation

@mdboom
Copy link
Contributor

@mdboom mdboom commented Sep 26, 2023 •
edited
Loading

This just implements a set of tier 2 pystats mentioned in #109329.

This implements:

  • Total micro-ops executed
  • Total number of traces started
  • Total number of traces created
  • Optimization attempts
  • Per uop execution counts, like we have for tier 1 instructions.
  • A histogram of uops executed per trace
  • Trace too long
  • Unsupported opcode (maybe even with counters for each offender)
  • Inner loop found
  • Too many frame pushes (currently capped at a depth of 5)
  • Too many frame pops (if we return from the original frame)
  • Recursive function call

It also fixes a "bug" where specialization "hits" in the running Tier 2 interpreter would count against the Tier 1 stats.

Not implemented (since it will require reworking of DEOPT_IF calls):

  • Exit reason counts: polymorphism vs branch mis-prediction

Example output (for nbody benchmark):

Optimization (Tier 2) stats

statistics about the Tier 2 optimizer

Overall stats

overall stats
Count Ratio
Optimization attempts 5
Traces created 5 100.0%
Traces executed 1,600,070
Uops executed 65,202,621 40
Trace stack overflow 0
Trace stack underflow 0
Trace too long 5
Inner loop found 0
Recursive call 0

Trace length histogram

Range Count Ratio
<= 4 0 0.0%
<= 8 0 0.0%
<= 16 0 0.0%
<= 32 0 0.0%
<= 64 5 100.0%

Optimized trace length histogram

Range Count Ratio
<= 4 0 0.0%
<= 8 0 0.0%
<= 16 0 0.0%
<= 32 0 0.0%
<= 64 5 100.0%

Trace run length histogram

Range Count Ratio
<= 4 0 0.0%
<= 8 0 0.0%
<= 16 0 0.0%
<= 32 0 0.0%
<= 64 1,600,070 100.0%

Uop stats

uop stats
Uop Count Self Cumulative
STORE_FAST 14,900,598 22.9% 22.9%
_SET_IP 14,100,583 21.6% 44.5%
LOAD_FAST 10,100,513 15.5% 60.0%
_GUARD_BOTH_FLOAT 3,700,274 5.7% 65.6%
_BINARY_OP_SUBTRACT_FLOAT 2,900,138 4.4% 70.1%
UNPACK_SEQUENCE_TUPLE 2,400,092 3.7% 73.8%
UNPACK_SEQUENCE_LIST 2,400,092 3.7% 77.5%
_POP_JUMP_IF_TRUE 1,700,049 2.6% 80.1%
_EXIT_TRACE 1,600,070 2.5% 82.5%
_ITER_CHECK_LIST 1,600,065 2.5% 85.0%
_IS_ITER_EXHAUSTED_LIST 1,600,065 2.5% 87.4%
COPY 1,599,948 2.5% 89.9%
_ITER_NEXT_LIST 1,400,053 2.1% 92.0%
UNPACK_SEQUENCE_TWO_TUPLE 1,000,039 1.5% 93.6%
LOAD_CONST 800,001 1.2% 94.8%
SWAP 799,974 1.2% 96.0%
BINARY_SUBSCR_LIST_INT 799,974 1.2% 97.2%
_BINARY_OP_MULTIPLY_FLOAT 400,095 0.6% 97.9%
_BINARY_OP_ADD_FLOAT 400,041 0.6% 98.5%
STORE_SUBSCR_LIST_INT 399,987 0.6% 99.1%
POP_TOP 200,017 0.3% 99.4%
_ITER_CHECK_RANGE 99,984 0.2% 99.5%
_IS_ITER_EXHAUSTED_RANGE 99,984 0.2% 99.7%
_ITER_NEXT_RANGE 99,979 0.2% 99.8%
GET_ITER 99,979 0.2% 100.0%
BINARY_OP 27 0.0% 100.0%

Unsupported opcodes

unsupported opcodes
Opcode Count

Loading
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@brandtbucher brandtbucher brandtbucher approved these changes

Assignees

@brandtbucher brandtbucher

Labels

skip news

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@mdboom @brandtbucher