Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

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 all 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
112 changes: 57 additions & 55 deletions 112 Include/opcode.h
View file
Open in desktop

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

2 changes: 2 additions & 0 deletions 2 Lib/opcode.py
View file
Open in desktop
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
2 changes: 2 additions & 0 deletions 2 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:`float`
operations by mutating the left operand in-place when it is safe to do so.
98 changes: 48 additions & 50 deletions 98 Python/ceval.c
View file
Open in desktop
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) \

@markshannon markshannon Feb 10, 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.

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.

@brandtbucher brandtbucher Feb 11, 2022

Copy link
Copy Markdown
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 48 Python/opcode_targets.h
View file
Open in desktop

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

Loading
Morty Proxy This is a proxified and sanitized view of the page, visit original site.