10000 gh-112354: Add executor for less-taken branch by gvanrossum · Pull Request #112902 · python/cpython · GitHub
[go: up one dir, main page]
Skip to content

gh-112354: Add executor for less-taken branch #112902

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 41 commits into from
Closed

gh-112354: Add executor for less-taken branch #112902

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
41 commits
Select commit Hold shift + click to select a range
36feeb1
Skip ENTER_EXECUTOR as deopt target (use vm_data)
gvanrossum Nov 24, 2023
d12533b
Add an array of 'extras' to UOpExecutor
gvanrossum Nov 23, 2023
f21f2d8
Count side exits per uop loc and print if >= 10
gvanrossum Nov 24, 2023
8463965
Add _PyOptimizer_Anywhere (not yet used)
gvanrossum Nov 25, 2023
6403752
Only jump in ENTER_EXECUTOR if overwriting JUMP_BACKWARD
gvanrossum Nov 25, 2023
329dead
Assert base opcode in _Py_Specialize_ForIter
gvanrossum Nov 25, 2023
f1998c0
Disable curses tests in --fast-ci mode (make test)
gvanrossum Nov 25, 2023
b0944e6
Improve (?) check for executor recursion
gvanrossum Nov 25, 2023
26b5f89
Only generate extra executors for branches
gvanrossum Nov 25, 2023
649581c
Fix Uop -> UOp
gvanrossum Dec 1, 2023
835bf13
WIP
gvanrossum Dec 1, 2023
256b156
Fix where next_instr points upon E_E avoidance
gvanrossum Dec 2, 2023
75c7c32
Allow executors with oparg >= 256
gvanrossum Dec 2, 2023
a94c7f1
Don't try to optimize with default optimizer
gvanrossum Dec 2, 2023
747a3f0
Use separate 'counters' and 'executors' arrays
gvanrossum Dec 11, 2023
682cf5a
Jump directly to side-exit executors
gvanrossum Dec 12, 2023
359c6fc
Remove progress check; clean up the rest a big
gvanrossum Dec 13, 2023
ca6ed3a
Ensure array of executor pointers is 64-bit aligned
gvanrossum Dec 13, 2023
e2a26b5
Check at least two uops; further cleanup
gvanrossum Dec 13, 2023
38c7aab
Move exit_trace up, since it is smaller
gvanrossum Dec 13, 2023
0f64231
Use configured threshold and exp. backoff for counter
gvanrossum Dec 13, 2023
83297df
Add API to access sub-interpreters
gvanrossum Dec 13, 2023
d065a94
Move optimizer/executor tests to new file test_capi/test_opt.py
gvanrossum Dec 13, 2023
0f71a03
Merge branch 'test-opt' into uops-extras
gvanrossum Dec 13, 2023
075ab91
In _PyOptimizer_Unanchored, assert not ENTER_EXECUTOR, accept JUMP_BA…
gvanrossum Dec 13, 2023
c54daef
Call DISPATCH() directly from exit_trace
gvanrossum Dec 13, 2023
934a115
Correct comment on deoptimize
gvanrossum Dec 13, 2023
52c49eb
Merge branch 'main' into uops-extras
gvanrossum Dec 13, 2023
8f5e623
Remove enter_tier_one label
gvanrossum Dec 13, 2023
10b98f1
Add test
gvanrossum Dec 13, 2023
1450ca6
Fix memory leak
gvanrossum Dec 13, 2023
dcde4d3
Clear sub-executors array upon dealloc
gvanrossum Dec 13, 2023
15df63f
Add blurb
gvanrossum Dec 13, 2023
c786418
Avoid redundant stack frame saves/restores
gvanrossum Dec 13, 2023
ee0734b
Revert "Disable curses tests in --fast-ci mode (make test)"
gvanrossum Dec 14, 2023
655a841
Merge branch 'main' into uops-extras
gvanrossum Dec 14, 2023
32e36fa
Merge branch 'main' into uops-extras
gvanrossum Dec 14, 2023
f5b317a
Fix compiler warning about int/Py_ssize_t
gvanrossum Dec 14, 2023
4804a3c
Be less casual about incref/decref current executor
gvanrossum Dec 16, 2023
46c7d26
Slightly nicer way to handle refcounts
gvanrossum Dec 16, 2023
b991279
Silence compiler warning
gvanrossum Dec 17, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Use separate 'counters' and 'executors' arrays 
(The latter as yet unused.)
  • Loading branch information
@gvanrossum
gvanrossum committed Dec 13, 2023
commit 747a3f07ef1624161ce7b56fea98a9a644e8bae4
4 changes: 3 additions & 1 deletion Include/internal/pycore_uops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,9 @@ typedef struct {

typedef struct {
_PyExecutorObject base;
uintptr_t *extra; // An array of uintptr_t of size base.ob_base.ob_size
// Auxiliary arrays, allocated after trace[base.ob_size]
uint16_t *counters; // An array of counters
_PyExecutorObject **executors; // An array of executors
_PyUOpInstruction trace[1];
} _PyUOpExecutorObject;

Expand Down
4 changes: 2 additions & 2 deletions Python/ceval.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1108,7 +1108,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
_PyFrame_SetStackPointer(frame, stack_pointer);
// Increment side exit counter for this uop
int pc = next_uop - 1 - current_executor->trace;
uintptr_t *pcounter = current_executor->extra + pc;
uint16_t *pcounter = current_executor->counters + pc;
*pcounter += 1;
if (*pcounter == 16 && // TODO: use resume_threshold
tstate->interp->optimizer != &_PyOptimizer_Default &&
Expand All @@ -1120,7 +1120,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
DPRINTF(2, "--> %s @ %d in %p has %d side exits\n",
_PyUOpName(uopcode), pc, current_executor, (int)(*pcounter));
DPRINTF(2, " T1: %s\n", _PyOpcode_OpName[opcode]);
// The counter will cycle around in 2**64 executions :-)
// The counter will cycle around once the 16 bits overflow
int optimized = _PyOptimizer_Anywhere(frame, src, dest, stack_pointer);
if (optimized < 0) {
goto error_tier_two;
Expand Down
8 changes: 5 additions & 3 deletions Python/optimizer.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -403,7 +403,7 @@ PyTypeObject _PyUOpExecutor_Type = {
PyVarObject_HEAD_INIT(&PyType_Type, 0)
.tp_name = "uop_executor",
.tp_basicsize = sizeof(_PyUOpExecutorObject) - sizeof(_PyUOpInstruction),
.tp_itemsize = sizeof(_PyUOpInstruction) + sizeof(uintptr_t),
.tp_itemsize = sizeof(_PyUOpInstruction) + sizeof(uint16_t) + sizeof(_PyExecutorObject *),
.tp_flags = Py_TPFLAGS_DEFAULT | Py_TPFLAGS_DISALLOW_INSTANTIATION,
.tp_dealloc = (destructor)uop_dealloc,
.tp_as_sequence = &uop_as_sequence,
Expand Down Expand Up @@ -849,8 +849,10 @@ make_executor_from_uops(_PyUOpInstruction *buffer, _PyBloomFilter *dependencies)
if (executor == NULL) {
return NULL;
}
executor->extra = (uintptr_t *)(executor->trace + length);
memset(executor->extra, 0, sizeof(uintptr_t) * length);
executor->counters = (uint16_t *)(&executor->trace[length]);
memset(executor->counters, 0, sizeof(uint16_t) * length);
executor->executors = (_PyExecutorObject **)(&executor->counters[length]);
memset(executor->executors, 0, sizeof(_PyExecutorObject *) * length);
int dest = length - 1;
/* Scan backwards, so that we see the destinations of jumps before the jumps themselves. */
for (int i = _Py_UOP_MAX_TRACE_LENGTH-1; i >= 0; i--) {
Expand Down
0