bm-20241025-azure-x86_64-brandtbucher-justin_no_externs-3.14.0a1+-5791853-pystats-gc_traversal-vs-base.md

Execution counts

Execution counts for Tier 1 instructions.

The "miss ratio" column shows the percentage of times the instruction executed that it deoptimized. When this happens, the base unspecialized instruction is not counted.

Name	Base Count	Head Count	Change
LOAD_FAST	1,740	1,740	0.0%
STORE_FAST	1,140	1,140	0.0%
LOAD_CONST	780	780	0.0%
PUSH_NULL	540	540	0.0%
POP_TOP	480	480	0.0%
LOAD_GLOBAL_MODULE	480	480	0.0%
RESUME_CHECK	480	480	0.0%
LOAD_ATTR_MODULE	360	360	0.0%
RETURN_VALUE	300	300	0.0%
LOAD_FAST_LOAD_FAST	300	300	0.0%
CALL_NON_PY_GENERAL	300	300	0.0%
CALL_PY_EXACT_ARGS	300	300	0.0%
CALL	260	260	0.0%
LOAD_ATTR	260	260	0.0%
LOAD_ATTR_INSTANCE_VALUE	240	240	0.0%
GET_ITER	180	180	0.0%
BUILD_LIST	180	180	0.0%
RETURN_CONST	180	180	0.0%
CALL_BUILTIN_CLASS	180	180	0.0%
FOR_ITER_RANGE	180	180	0.0%
LOAD_GLOBAL_BUILTIN	180	180	0.0%
BUILD_TUPLE	120	120	0.0%
CALL_FUNCTION_EX	120	120	0.0%
JUMP_BACKWARD	120	120	0.0%
LOAD_DEREF	120	120	0.0%
LOAD_GLOBAL	120	120	0.0%
POP_JUMP_IF_FALSE	120	120	0.0%
POP_JUMP_IF_NOT_NONE	120	120	0.0%
CALL_BUILTIN_FAST_WITH_KEYWORDS	120	120	0.0%
LOAD_ATTR_METHOD_NO_DICT	120	120	0.0%
LOAD_ATTR_METHOD_WITH_VALUES	120	120	0.0%
TO_BOOL	100	100	0.0%
BINARY_OP_ADD_FLOAT	60	60	0.0%
MAKE_FUNCTION	60	60	0.0%
NOP	60	60	0.0%
BINARY_OP	60	60	0.0%
CALL_INTRINSIC_1	60	60	0.0%
COPY_FREE_VARS	60	60	0.0%
FOR_ITER	60	60	0.0%
IS_OP	60	60	0.0%
JUMP_FORWARD	60	60	0.0%
LIST_EXTEND	60	60	0.0%
MAKE_CELL	60	60	0.0%
POP_JUMP_IF_TRUE	60	60	0.0%
SET_FUNCTION_ATTRIBUTE	60	60	0.0%
STORE_DEREF	60	60	0.0%
STORE_FAST_STORE_FAST	60	60	0.0%
BINARY_OP_SUBTRACT_FLOAT	60	60	0.0%
BINARY_SUBSCR_TUPLE_INT	60	60	0.0%
CALL_METHOD_DESCRIPTOR_NOARGS	60	60	0.0%
CALL_METHOD_DESCRIPTOR_O	60	60	0.0%
CALL_PY_GENERAL	60	60	0.0%
COMPARE_OP_INT	60	60	0.0%
TO_BOOL_BOOL	60	60	0.0%
UNPACK_SEQUENCE_TWO_TUPLE	60	60	0.0%
ENTER_EXECUTOR	60	60	0.0%
BINARY_SUBSCR	20	20	0.0%
UNPACK_SEQUENCE	20	20	0.0%

Pair counts

Pair counts for top 100 opcode pairs

Pairs of specialized operations that deoptimize and are then followed by the corresponding unspecialized instruction are not counted as pairs.

Not included in comparative output.

Predecessor/Successor Pairs

Top 5 predecessors and successors of each Tier 1 opcode.

This does not include the unspecialized instructions that occur after a specialized instruction deoptimizes.

Not included in comparative output.

Specialization stats

Specialization stats by family

BINARY_OP

specialization stats for BINARY_OP family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
deferred ⓘ Lists the number of "deferred" (i.e. not specialized) instructions executed.	60	1.5%	60	1.5%	0.0%
hit ⓘ Specialized instructions that complete.	3,780	96.9%	3,780	96.9%	0.0%
miss ⓘ Specialized instructions that deopt.	60	1.5%	60	1.5%	0.0%

BINARY_SUBSCR

specialization stats for BINARY_SUBSCR family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	60	75.0%	60	75.0%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	20	100.0%	20	100.0%	0.0%
Failure	0	0.0%	0	0.0%

CALL

specialization stats for CALL family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	64,380	99.6%	64,380	99.6%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	260	100.0%	260	100.0%	0.0%
Failure	0	0.0%	0	0.0%

COMPARE_OP

specialization stats for COMPARE_OP family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	1,920	100.0%	1,920	100.0%	0.0%

FOR_ITER

specialization stats for FOR_ITER family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
deferred ⓘ Lists the number of "deferred" (i.e. not specialized) instructions executed.	60	25.0%	60	25.0%	0.0%
hit ⓘ Specialized instructions that complete.	180	75.0%	180	75.0%	0.0%

LOAD_ATTR

specialization stats for LOAD_ATTR family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
deferred ⓘ Lists the number of "deferred" (i.e. not specialized) instructions executed.	60	5.5%	60	5.5%	0.0%
hit ⓘ Specialized instructions that complete.	840	76.4%	840	76.4%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	180	90.0%	180	90.0%	0.0%
Failure	20	10.0%	20	10.0%	0.0%

LOAD_GLOBAL

specialization stats for LOAD_GLOBAL family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	660	84.6%	660	84.6%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	120	100.0%	120	100.0%	0.0%
Failure	0	0.0%	0	0.0%

STORE_SUBSCR

specialization stats for STORE_SUBSCR family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	29,970,000	100.0%	29,970,000	100.0%	0.0%

TO_BOOL

specialization stats for TO_BOOL family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
deferred ⓘ Lists the number of "deferred" (i.e. not specialized) instructions executed.	60	37.5%	60	37.5%	0.0%
hit ⓘ Specialized instructions that complete.	60	37.5%	60	37.5%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	20	50.0%	20	50.0%	0.0%
Failure	20	50.0%	20	50.0%	0.0%

Failure kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
sequence	20	100.0%	20	100.0%	0.0%

UNPACK_SEQUENCE

specialization stats for UNPACK_SEQUENCE family

Kind	Base Count	Base Ratio	Head Count	Head Ratio	Change
hit ⓘ Specialized instructions that complete.	60	75.0%	60	75.0%	0.0%

Success	Base Count	Base Ratio	Head Count	Head Ratio	Change
Success	20	100.0%	20	100.0%	0.0%
Failure	0	0.0%	0	0.0%

Specialization effectiveness

specialization effectiveness

All entries are execution counts. Should add up to the total number of Tier 1 instructions executed.

Instructions	Base Count	Base Ratio	Head Count	Head Ratio	Change
Basic ⓘ Instructions that are not and cannot be specialized, e.g. LOAD_FAST.	7,320	61.9%	7,320	61.9%	0.0%
Not specialized ⓘ Instructions that could be specialized but aren't, e.g. LOAD_ATTR, BINARY_SLICE.	900	7.6%	900	7.6%	0.0%
Specialized hits ⓘ Specialized instructions, e.g. LOAD_ATTR_MODULE that complete.	3,540	29.9%	3,540	29.9%	0.0%
Specialized misses ⓘ Specialized instructions, e.g. LOAD_ATTR_MODULE that deopt.	60	0.5%	60	0.5%	0.0%

Deferred by instruction

Breakdown of deferred (not specialized) instruction counts by family

Name	Base Count	Base Ratio	Head Count	Head Ratio	Change
TO_BOOL	60	25.0%	60	25.0%	0.0%
BINARY_OP	60	25.0%	60	25.0%	0.0%
FOR_ITER	60	25.0%	60	25.0%	0.0%
LOAD_ATTR	60	25.0%	60	25.0%	0.0%
BINARY_SLICE	0	0.0%	0	0.0%
STORE_SLICE	0	0.0%	0	0.0%
CACHE	0	0.0%	0	0.0%
BINARY_SUBSCR	0	0.0%	0	0.0%
GET_ITER	0	0.0%	0	0.0%
MAKE_FUNCTION	0	0.0%	0	0.0%

Misses by instruction

Breakdown of misses (specialized deopts) instruction counts by family

Name	Base Count	Base Ratio	Head Count	Head Ratio	Change
BINARY_OP_ADD_FLOAT	60	100.0%	60	100.0%	0.0%
CACHE	0	0.0%	0	0.0%
GET_ITER	0	0.0%	0	0.0%
MAKE_FUNCTION	0	0.0%	0	0.0%
NOP	0	0.0%	0	0.0%
POP_TOP	0	0.0%	0	0.0%
PUSH_NULL	0	0.0%	0	0.0%
RETURN_VALUE	0	0.0%	0	0.0%
BUILD_LIST	0	0.0%	0	0.0%
BUILD_TUPLE	0	0.0%	0	0.0%

Call stats

Inlined calls and frame stats

This shows what fraction of calls to Python functions are inlined (i.e. not having a call at the C level) and for those that are not, where the call comes from. The various categories overlap.

Also includes the count of frame objects created.

	Base Count	Base Ratio	Head Count	Head Ratio	Change
Calls to PyEval_EvalDefault	60	12.5%	60	12.5%	0.0%
Calls to Python functions inlined	420	87.5%	420	87.5%	0.0%
Calls via PyEval_EvalFrame (total)	60	12.5%	60	12.5%	0.0%
Calls via PyEval_EvalFrame (vector)	60	12.5%	60	12.5%	0.0%
Calls via PyEval_EvalFrame (generator)	0	0.0%	0	0.0%
Calls via PyEval_EvalFrame (legacy)	0	0.0%	0	0.0%
Calls via PyEval_EvalFrame (function vectorcall)	60	12.5%	60	12.5%	0.0%
Calls via PyEval_EvalFrame (build class)	0	0.0%	0	0.0%
Calls via PyEval_EvalFrame (slot)	0	0.0%	0	0.0%
Calls via PyEval_EvalFrame (function ex)	60	12.5%	60	12.5%	0.0%
Calls via PyEval_EvalFrame (api)	0	0.0%	0	0.0%
Calls via PyEval_EvalFrame (method)	0	0.0%	0	0.0%
Frame objects created	0	0.0%	0	0.0%
Frames pushed	480	100.0%	480	100.0%	0.0%

Object stats

Allocations, frees and dict materializatons

Below, "allocations" means "allocations that are not from a freelist". Total allocations = "Allocations from freelist" + "Allocations".

"Inline values" is the number of values arrays inlined into objects.

The cache hit/miss numbers are for the MRO cache, split into dunder and other names.

	Base Count	Base Ratio	Head Count	Head Ratio	Change
Method cache collisions	21		16		-23.8%
Method cache misses	29		24		-17.2%
Method cache hits	191		196		2.6%
Immortal increfs	30,312,429	28.3%	30,312,424	28.3%	-0.0%
Immortal decrefs	57,304,742	37.9%	57,304,737	37.9%	-0.0%
Mortal increfs	76,910,969	71.7%	76,910,964	71.7%	-0.0%
Mortal decrefs	93,791,316	62.1%	93,791,311	62.1%	-0.0%
Allocations from freelist	65,300	0.4%	65,300	0.4%	0.0%
Frees to freelist	69,380		69,380		0.0%
Allocations	16,935,880	99.6%	16,935,880	99.6%	0.0%
Allocations to 512 bytes	16,879,660	99.3%	16,879,660	99.3%	0.0%
Allocations to 4 kbytes	26,940	0.2%	26,940	0.2%	0.0%
Allocations over 4 kbytes	29,280	0.2%	29,280	0.2%	0.0%
Frees	16,942,161		16,942,161		0.0%
Inline values	0		0
Interpreter mortal increfs	3,960	0.0%	3,960	0.0%	0.0%
Interpreter mortal decrefs	5,260	0.0%	5,260	0.0%	0.0%
Interpreter immortal increfs	2,220	0.0%	2,220	0.0%	0.0%
Interpreter immortal decrefs	1,620	0.0%	1,620	0.0%	0.0%
Materialize dict (on request)	0		0
Materialize dict (new key)	0		0
Materialize dict (too big)	0		0
Materialize dict (str subclass)	0		0
Method cache dunder hits	0		0
Method cache dunder misses	0		0

GC stats

GC collections and effectiveness

Collected/visits gives some measure of efficiency.

Generation	Base Collections	Base Objects collected	Base Object visits	Head Collections	Head Objects collected	Head Object visits
0	0	0	0	0	0	0
1	0	0	0	0	0	0
2	3,840	160	4,500,842,200	3,840	160	4,500,842,200

Optimization (Tier 2) stats

statistics about the Tier 2 optimizer

	Base Count	Base Ratio	Head Count	Head Ratio	Change
Optimization attempts ⓘ The number of times a potential trace is identified. Specifically, this occurs in the JUMP BACKWARD instruction when the counter reaches a threshold.	60		60		0.0%
Traces created ⓘ The number of traces that were successfully created.	60	100.0%	60	100.0%	0.0%
Trace stack overflow ⓘ A trace is truncated because it would require more than 5 stack frames.	0	0.0%	0	0.0%
Trace stack underflow ⓘ A potential trace is abandoned because it pops more frames than it pushes.	0	0.0%	0	0.0%
Trace too long ⓘ A trace is truncated because it is longer than the instruction buffer.	0	0.0%	0	0.0%
Trace too short ⓘ A potential trace is abandoced because it it too short.	0	0.0%	0	0.0%
Inner loop found ⓘ A trace is truncated because it has an inner loop	0	0.0%	0	0.0%
Recursive call ⓘ A trace is truncated because it has a recursive call.	0	0.0%	0	0.0%
Low confidence ⓘ A trace is abandoned because the likelihood of the jump to top being taken is too low.	0	0.0%	0	0.0%
Executors invalidated ⓘ The number of executors that were invalidated due to watched dictionary changes.	0	0.0%	0	0.0%
Traces executed ⓘ The number of traces that were executed	179,940		179,940		0.0%
Uops executed ⓘ The total number of uops (micro-operations) that were executed	391,807,440	217,743.4%	391,807,440	217,743.4%	0.0%

	Base Count	Base Ratio	Head Count	Head Ratio	Change
Optimizer attempts ⓘ The number of times the trace optimizer (_Py_uop_analyze_and_optimize) was run.	60		60		0.0%
Optimizer successes ⓘ The number of traces that were successfully optimized.	60	100.0%	60	100.0%	0.0%
Optimizer no memory ⓘ The number of optimizations that failed due to no memory.	0	0.0%	0	0.0%
Remove globals builtins changed ⓘ The builtins changed during optimization	0	0.0%	0	0.0%
Remove globals incorrect keys ⓘ The keys in the globals dictionary aren't what was expected	0	0.0%	0	0.0%

Trace length histogram

trace length histogram

Range	Base Count	Base Ratio	Head Count	Head Ratio	Change
<= 1	0	0.0%	0	0.0%
<= 2	0	0.0%	0	0.0%
<= 4	0	0.0%	0	0.0%
<= 8	0	0.0%	0	0.0%
<= 16	0	0.0%	0	0.0%
<= 32	0	0.0%	0	0.0%
<= 64	0	0.0%	0	0.0%
<= 128	60	100.0%	60	100.0%	0.0%

Optimized trace length histogram

optimized trace length histogram

Range	Base Count	Base Ratio	Head Count	Head Ratio	Change
<= 1	0	0.0%	0	0.0%
<= 2	0	0.0%	0	0.0%
<= 4	0	0.0%	0	0.0%
<= 8	0	0.0%	0	0.0%
<= 16	0	0.0%	0	0.0%
<= 32	0	0.0%	0	0.0%
<= 64	60	100.0%	60	100.0%	0.0%

Trace run length histogram

trace run length histogram

Range	Base Count	Base Ratio	Head Count	Head Ratio	Change
<= 1	0	0.0%	0	0.0%

Uop execution stats

uop execution stats

Name	Base Count	Head Count	Change
_SET_IP	30,159,240	30,159,240	0.0%
_CHECK_VALIDITY	30,159,240	30,159,240	0.0%
_CHECK_PERIODIC	30,099,300	30,099,300	0.0%
_GUARD_NOT_EXHAUSTED_RANGE	30,091,860	30,091,860	0.0%
_ITER_CHECK_RANGE	30,091,860	30,091,860	0.0%
_MAKE_WARM	30,091,860	30,091,860	0.0%
_ITER_NEXT_RANGE	30,031,800	30,031,800	0.0%
_LOAD_FAST_3	30,029,940	30,029,940	0.0%
_STORE_FAST_4	29,971,860	29,971,860	0.0%
_STORE_SUBSCR_LIST_INT	29,970,000	29,970,000	0.0%
_LOAD_FAST_1	29,970,000	29,970,000	0.0%
_LOAD_FAST_4	29,970,000	29,970,000	0.0%
_JUMP_TO_TOP	29,911,920	29,911,920	0.0%
_EXIT_TRACE	179,940	179,940	0.0%
_START_EXECUTOR	179,940	179,940	0.0%
_LOAD_FAST_2	121,740	121,740	0.0%
_POP_TOP	61,800	61,800	0.0%
_CHECK_FUNCTION	61,800	61,800	0.0%
_LOAD_CONST_INLINE_BORROW	61,800	61,800	0.0%
_STORE_FAST_2	61,800	61,800	0.0%
_GET_ITER	59,940	59,940	0.0%
_BUILD_LIST	59,940	59,940	0.0%
_BINARY_OP	59,940	59,940	0.0%
_CALL_BUILTIN_CLASS	59,940	59,940	0.0%
_CHECK_VALIDITY_AND_SET_IP	59,940	59,940	0.0%
_LOAD_CONST_INLINE_BORROW_WITH_NULL	59,940	59,940	0.0%
_STORE_FAST_1	59,940	59,940	0.0%
_STORE_FAST_3	59,940	59,940	0.0%
_PUSH_NULL	7,440	7,440	0.0%
_LOAD_CONST_INLINE	7,440	7,440	0.0%
_CALL_BUILTIN_FAST_WITH_KEYWORDS	3,720	3,720	0.0%
_CALL_NON_PY_GENERAL	3,720	3,720	0.0%
_CHECK_IS_NOT_PY_CALLABLE	3,720	3,720	0.0%
_LOAD_FAST_6	3,720	3,720	0.0%
_BINARY_OP_ADD_FLOAT	1,860	1,860	0.0%
_BINARY_OP_SUBTRACT_FLOAT	1,860	1,860	0.0%
_COMPARE_OP_INT	1,860	1,860	0.0%
_GUARD_BOTH_FLOAT	1,860	1,860	0.0%
_GUARD_IS_NOT_NONE_POP	1,860	1,860	0.0%
_GUARD_IS_TRUE_POP	1,860	1,860	0.0%
_GUARD_NOS_FLOAT	1,860	1,860	0.0%
_GUARD_NOS_INT	1,860	1,860	0.0%
_LOAD_FAST_5	1,860	1,860	0.0%
_STORE_FAST_5	1,860	1,860	0.0%
_STORE_FAST_6	1,860	1,860	0.0%

Pair counts

Pair counts for top 100 Non-JIT uop pairs

Pairs of specialized operations that deoptimize and are then followed by the corresponding unspecialized instruction are not counted as pairs.

Not included in comparative output.

Unsupported opcodes

unsupported opcodes

Optimizer errored out with opcode

Optimization stopped after encountering this opcode

Rare events

Counts of rare/unlikely events

Event	Base Count	Head Count	Change
set class ⓘ Setting an object's class, obj.__class__ = ...	0	0
set bases ⓘ Setting the bases of a class, cls.__bases__ = ...	0	0
set eval frame func ⓘ Setting the PEP 523 frame eval function _PyInterpreterState_SetFrameEvalFunc()	0	0
builtin dict ⓘ Modifying the builtins, __builtins__.__dict__[var] = ...	0	0
func modification ⓘ Modifying a function, e.g. func.__defaults__ = ..., etc.	0	0
watched dict modification ⓘ A watched dict has been modified	0	0
watched globals modification ⓘ A watched globals() dict has been modified	0	0

Meta stats

Meta statistics

	Base Count	Head Count	Change
Number of data files	20	20	0.0%

---

Stats gathered on: 2024-10-26