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.

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 18 commits into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
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
112 changes: 57 additions & 55 deletions Include/opcode.h

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 2 additions & 0 deletions Lib/opcode.py
Original file line number Diff line number Diff line change
Expand Up @@ -232,6 +232,8 @@ def jabs_op(name, op):
"BINARY_OP_ADD_FLOAT",
"BINARY_OP_ADD_UNICODE",
"BINARY_OP_INPLACE_ADD_UNICODE",
"BINARY_OP_INPLACE_ADD_FLOAT",
"BINARY_OP_INPLACE_SUBTRACT_FLOAT",
"BINARY_OP_MULTIPLY_INT",
"BINARY_OP_MULTIPLY_FLOAT",
"BINARY_OP_SUBTRACT_INT",
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
Improve the performance of specialized :class:`float`
operations by mutating the left operand in-place when it is safe to do so.
98 changes: 48 additions & 50 deletions Python/ceval.c
Original file line number Diff line number Diff line change
Expand Up @@ -857,6 +857,43 @@ static const binaryfunc binary_ops[] = {
[NB_INPLACE_XOR] = PyNumber_InPlaceXor,
};

// NOTE: INPLACE must be a compile-time constant to avoid runtime branching!
#define BINARY_OP_FAST_FLOAT(OP, INPLACE) \
Copy link
Member

Choose a reason for hiding this comment

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

I'm not a fan of giant macros and I don't like the hidden compile time control flow.
Given that the inplace and non-inplace forms specialized forms differ in a fundamental way, I'd like that to be explicit in the instructions themselves.

Maybe have two different macros, one for the inplace form and one for the non-inplace form.
You should be able to factor out some of the common code into helper macros.
That might be clearer. Up to you.

Copy link
Member Author

Choose a reason for hiding this comment

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

Makes sense. I'll go ahead and split it up.

do { \
assert(cframe.use_tracing == 0); \
PyObject *lhs = SECOND(); \
PyObject *rhs = TOP(); \
DEOPT_IF(!PyFloat_CheckExact(lhs), BINARY_OP); \
DEOPT_IF(!PyFloat_CheckExact(rhs), BINARY_OP); \
if (INPLACE) { \
assert(_Py_OPCODE(*next_instr) == STORE_FAST || \
_Py_OPCODE(*next_instr) == STORE_FAST__LOAD_FAST); \
DEOPT_IF(GETLOCAL(_Py_OPARG(*next_instr)) != lhs, BINARY_OP); \
} \
STAT_INC(BINARY_OP, hit); \
double l = PyFloat_AS_DOUBLE(lhs); \
double r = PyFloat_AS_DOUBLE(rhs); \
double d = l OP r; \
STACK_SHRINK(1); \
Py_DECREF(rhs); \
if (Py_REFCNT(lhs) == 1 + !!(INPLACE)) { \
if (INPLACE) { \
Py_DECREF(lhs); \
STACK_SHRINK(1); \
next_instr++; \
} \
PyFloat_AS_DOUBLE(lhs) = d; \
NOTRACE_DISPATCH(); \
} \
Py_DECREF(lhs); \
PyObject *res = PyFloat_FromDouble(d); \
SET_TOP(res); \
if (res == NULL) { \
goto error; \
} \
NOTRACE_DISPATCH(); \
} while (0)


// PEP 634: Structural Pattern Matching

Expand Down Expand Up @@ -1986,23 +2023,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
}

TARGET(BINARY_OP_MULTIPLY_FLOAT) {
assert(cframe.use_tracing == 0);
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;
}
NOTRACE_DISPATCH();
BINARY_OP_FAST_FLOAT(*, false);
}

TARGET(BINARY_OP_SUBTRACT_INT) {
Expand All @@ -2024,22 +2045,11 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
}

TARGET(BINARY_OP_SUBTRACT_FLOAT) {
assert(cframe.use_tracing == 0);
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;
}
NOTRACE_DISPATCH();
BINARY_OP_FAST_FLOAT(-, false);
}

TARGET(BINARY_OP_INPLACE_SUBTRACT_FLOAT) {
BINARY_OP_FAST_FLOAT(-, true);
}

TARGET(BINARY_OP_ADD_UNICODE) {
Expand Down Expand Up @@ -2090,23 +2100,11 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
}

TARGET(BINARY_OP_ADD_FLOAT) {
assert(cframe.use_tracing == 0);
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;
}
NOTRACE_DISPATCH();
BINARY_OP_FAST_FLOAT(+, false);
}

TARGET(BINARY_OP_INPLACE_ADD_FLOAT) {
BINARY_OP_FAST_FLOAT(+, true);
}

TARGET(BINARY_OP_ADD_INT) {
Expand Down
48 changes: 24 additions & 24 deletions Python/opcode_targets.h

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading