Skip to content

bpo-46372: Try to mutate the LHS during fast float ops #30594

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
brandtbucher wants to merge 18 commits into from
Closed

bpo-46372: Try to mutate the LHS during fast float ops #30594

Show file tree
Hide file tree
Changes from 6 commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
cf64e13
Perform specialized ops in-place when possible
brandtbucher Jan 11, 2022
5e49a13
Break out common logic
brandtbucher Jan 12, 2022
f7187d9
Clean things up a bit
brandtbucher Jan 12, 2022
dd19cec
Clean up macro
brandtbucher Jan 12, 2022
78cc5c4
Simplify the "small" int case
brandtbucher Jan 13, 2022
218b10b
blurb add
brandtbucher Jan 14, 2022
505fd8e
Catch up with main
brandtbucher Jan 18, 2022
6b4cc6d
Apply first round of code revew comments
brandtbucher Jan 18, 2022
f21a813
Fix 32-bit Windows errrors
brandtbucher Jan 19, 2022
d6a69d5
Perform the inplace checks when specializing
brandtbucher Jan 19, 2022
f11cf6d
Remove duplicate code
brandtbucher Jan 19, 2022
b6f041a
Simplify things further
brandtbucher Jan 19, 2022
eb586b1
Remove int and refcount == 2 optimizations
brandtbucher Jan 19, 2022
def9c64
Catch up with main
brandtbucher Jan 21, 2022
6e262ba
Revert int changes
brandtbucher Jan 24, 2022
37fab55
Catch up with main
brandtbucher Jan 28, 2022
04f186c
Add the "adaptive superinstructions" back
brandtbucher Feb 1, 2022
e0d0042
Catch up with main
brandtbucher Feb 1, 2022
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
2 changes: 2 additions & 0 deletions Misc/NEWS.d/next/Core and Builtins/2022-01-13-22-50-49.bpo-46372.b00Vdn.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
Improve the performance of specialized :class:`int` and :class:`float`
operations by mutating the left operand in-place when it is safe to do so.
164 changes: 75 additions & 89 deletions Python/ceval.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -863,6 +863,75 @@ static const binaryfunc binary_ops[] = {
[NB_INPLACE_XOR] = PyNumber_InPlaceXor,
};

#define NEXT_OP_STORES(O) \
Comment thread
brandtbucher marked this conversation as resolved.
Outdated
((_Py_OPCODE(*next_instr) == STORE_FAST || \
_Py_OPCODE(*next_instr) == STORE_FAST__LOAD_FAST) && \
GETLOCAL(_Py_OPARG(*next_instr)) == (O)) \

#define BINARY_OP_FAST_INT(OP) \
do { \
PyObject *lhs = SECOND(); \
PyObject *rhs = TOP(); \
DEOPT_IF(!PyLong_CheckExact(lhs), BINARY_OP); \
DEOPT_IF(!PyLong_CheckExact(rhs), BINARY_OP); \
DEOPT_IF(1 < Py_ABS(Py_SIZE(lhs)), BINARY_OP); \
DEOPT_IF(1 < Py_ABS(Py_SIZE(rhs)), BINARY_OP); \
STAT_INC(BINARY_OP, hit); \
stwodigits l = Py_SIZE(lhs) * ((PyLongObject *)lhs)->ob_digit[0]; \
stwodigits r = Py_SIZE(rhs) * ((PyLongObject *)rhs)->ob_digit[0]; \
stwodigits i = l OP r; \
Py_DECREF(rhs); \
STACK_SHRINK(1); \
if (-_PY_NSMALLNEGINTS <= i && i < _PY_NSMALLPOSINTS) { \
Py_DECREF(lhs); \
PyLongObject *res = &_PyLong_SMALL_INTS[_PY_NSMALLNEGINTS + i]; \
Py_INCREF(res); \
SET_TOP((PyObject *)res); \
DISPATCH(); \
} \
bool inplace = NEXT_OP_STORES(lhs); \
if (Py_ABS(i) < PyLong_BASE && Py_REFCNT(lhs) == inplace + 1) { \
assert(lhs < (PyObject *)&_PyLong_SMALL_INTS[0] || \
Comment thread
brandtbucher marked this conversation as resolved.
Outdated
lhs >= (PyObject *)&_PyLong_SMALL_INTS[_PY_NSMALLNEGINTS + \
_PY_NSMALLPOSINTS]); \
((PyLongObject *)lhs)->ob_digit[0] = Py_ABS(i); \
Comment thread
brandtbucher marked this conversation as resolved.
Outdated
Py_SET_SIZE(lhs, i < 0 ? -1 : 1); \
DISPATCH(); \
} \
Py_DECREF(lhs); \
PyObject *res = PyLong_FromLong(i); \
SET_TOP(res); \
if (res == NULL) { \
goto error; \
} \
DISPATCH(); \
} while (0)

#define BINARY_OP_FAST_FLOAT(OP) \
do { \
PyObject *lhs = SECOND(); \
PyObject *rhs = TOP(); \
DEOPT_IF(!PyFloat_CheckExact(lhs), BINARY_OP); \
DEOPT_IF(!PyFloat_CheckExact(rhs), BINARY_OP); \
STAT_INC(BINARY_OP, hit); \
double l = PyFloat_AS_DOUBLE(lhs); \
double r = PyFloat_AS_DOUBLE(rhs); \
double d = l OP r; \
Py_DECREF(rhs); \
STACK_SHRINK(1); \
bool inplace = NEXT_OP_STORES(lhs); \
if (Py_REFCNT(lhs) == inplace + 1) { \
PyFloat_AS_DOUBLE(lhs) = d; \

@gvanrossum gvanrossum Jan 31, 2022

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Presumably (see PEP 674) Victor won't like this use of a macro as a target. :-)

Nevertheless, this whole macro is so nice and straightforward compared to ints!

DISPATCH(); \
} \
Py_DECREF(lhs); \
Comment thread
brandtbucher marked this conversation as resolved.
Outdated
PyObject *res = PyFloat_FromDouble(d); \
SET_TOP(res); \
if (res == NULL) { \
goto error; \
} \
DISPATCH(); \
} while (0)

// PEP 634: Structural Pattern Matching

Expand Down Expand Up @@ -2054,74 +2123,19 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
}

TARGET(BINARY_OP_MULTIPLY_INT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyLong_CheckExact(left), BINARY_OP);
DEOPT_IF(!PyLong_CheckExact(right), BINARY_OP);
STAT_INC(BINARY_OP, hit);
PyObject *prod = _PyLong_Multiply((PyLongObject *)left, (PyLongObject *)right);
SET_SECOND(prod);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (prod == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_INT(*);
}

TARGET(BINARY_OP_MULTIPLY_FLOAT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyFloat_CheckExact(left), BINARY_OP);
DEOPT_IF(!PyFloat_CheckExact(right), BINARY_OP);
STAT_INC(BINARY_OP, hit);
double dprod = ((PyFloatObject *)left)->ob_fval *
((PyFloatObject *)right)->ob_fval;
PyObject *prod = PyFloat_FromDouble(dprod);
SET_SECOND(prod);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (prod == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_FLOAT(*);
}

TARGET(BINARY_OP_SUBTRACT_INT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyLong_CheckExact(left), BINARY_OP);
DEOPT_IF(!PyLong_CheckExact(right), BINARY_OP);
STAT_INC(BINARY_OP, hit);
PyObject *sub = _PyLong_Subtract((PyLongObject *)left, (PyLongObject *)right);
SET_SECOND(sub);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (sub == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_INT(-);
}

TARGET(BINARY_OP_SUBTRACT_FLOAT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyFloat_CheckExact(left), BINARY_OP);
DEOPT_IF(!PyFloat_CheckExact(right), BINARY_OP);
STAT_INC(BINARY_OP, hit);
double dsub = ((PyFloatObject *)left)->ob_fval - ((PyFloatObject *)right)->ob_fval;
PyObject *sub = PyFloat_FromDouble(dsub);
SET_SECOND(sub);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (sub == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_FLOAT(-);
}

TARGET(BINARY_OP_ADD_UNICODE) {
Expand Down Expand Up @@ -2170,39 +2184,11 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
}

TARGET(BINARY_OP_ADD_FLOAT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyFloat_CheckExact(left), BINARY_OP);
DEOPT_IF(Py_TYPE(right) != Py_TYPE(left), BINARY_OP);
STAT_INC(BINARY_OP, hit);
double dsum = ((PyFloatObject *)left)->ob_fval +
((PyFloatObject *)right)->ob_fval;
PyObject *sum = PyFloat_FromDouble(dsum);
SET_SECOND(sum);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (sum == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_FLOAT(+);
}

TARGET(BINARY_OP_ADD_INT) {
PyObject *left = SECOND();
PyObject *right = TOP();
DEOPT_IF(!PyLong_CheckExact(left), BINARY_OP);
DEOPT_IF(Py_TYPE(right) != Py_TYPE(left), BINARY_OP);
STAT_INC(BINARY_OP, hit);
PyObject *sum = _PyLong_Add((PyLongObject *)left, (PyLongObject *)right);
SET_SECOND(sum);
Py_DECREF(right);
Py_DECREF(left);
STACK_SHRINK(1);
if (sum == NULL) {
goto error;
}
DISPATCH();
BINARY_OP_FAST_INT(+);
}

TARGET(BINARY_SUBSCR) {
Expand Down
12 changes: 9 additions & 3 deletions Python/specialize.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1621,7 +1621,9 @@ _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
_Py_OPARG(*instr));
goto success;
}
if (PyLong_CheckExact(lhs)) {
if (PyLong_CheckExact(lhs) &&
Py_ABS(Py_SIZE(lhs)) < 2 && Py_ABS(Py_SIZE(rhs)) < 2)
{
*instr = _Py_MAKECODEUNIT(BINARY_OP_ADD_INT, _Py_OPARG(*instr));
goto success;
}
Expand All @@ -1637,7 +1639,9 @@ _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
SPECIALIZATION_FAIL(BINARY_OP, SPEC_FAIL_DIFFERENT_TYPES);
goto failure;
}
if (PyLong_CheckExact(lhs)) {
if (PyLong_CheckExact(lhs) &&
Py_ABS(Py_SIZE(lhs)) < 2 && Py_ABS(Py_SIZE(rhs)) < 2)
{
*instr = _Py_MAKECODEUNIT(BINARY_OP_MULTIPLY_INT,
_Py_OPARG(*instr));
goto success;
Expand All @@ -1654,7 +1658,9 @@ _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
SPECIALIZATION_FAIL(BINARY_OP, SPEC_FAIL_DIFFERENT_TYPES);
goto failure;
}
if (PyLong_CheckExact(lhs)) {
if (PyLong_CheckExact(lhs) &&
Py_ABS(Py_SIZE(lhs)) < 2 && Py_ABS(Py_SIZE(rhs)) < 2)
{
*instr = _Py_MAKECODEUNIT(BINARY_OP_SUBTRACT_INT,
_Py_OPARG(*instr));
goto success;
Expand Down