Skip to content

GH-100485: Convert from Fast2Sum to 2Sum #100836

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

Merged
merged 4 commits into from
Jan 8, 2023
Merged

GH-100485: Convert from Fast2Sum to 2Sum #100836

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0c6c2e1
fasttwosum root
rhettinger Jan 8, 2023
15eb16c
twosum() based
rhettinger Jan 8, 2023
8941d47
Clean-up
rhettinger Jan 8, 2023
42c0bfe
Convert dl_zero() from a function to a constant.
rhettinger Jan 8, 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
29 changes: 16 additions & 13 deletions Modules/mathmodule.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2847,7 +2847,7 @@ based on ideas from three sources:

The double length routines allow for quite a bit of instruction
level parallelism. On a 3.22 Ghz Apple M1 Max, the incremental
cost of increasing the input vector size by one is 6.25 nsec.
cost of increasing the input vector size by one is 6.0 nsec.

dl_zero() returns an extended precision zero
dl_split() exactly splits a double into two half precision components.
Expand All @@ -2860,22 +2860,25 @@ dl_to_d() converts from extended precision to double precision.

typedef struct{ double hi; double lo; } DoubleLength;

static const DoubleLength dl_zero = {0.0, 0.0};

static inline DoubleLength
dl_zero()
twosum(double a, double b)
{
return (DoubleLength) {0.0, 0.0};
double s = a + b;
double ap = s - b;
double bp = s - a;
double da = a - ap;
double db = b - bp;
double t = da + db;
return (DoubleLength) {s, t};
}

static inline DoubleLength
dl_add(DoubleLength total, double x)
{
double s = total.hi + x;
double c = total.lo;
if (fabs(total.hi) >= fabs(x)) {
c += (total.hi - s) + x;
} else {
c += (x - s) + total.hi;
}
return (DoubleLength) {s, c};
DoubleLength s = twosum(total.hi, x);
return (DoubleLength) {s.hi, total.lo + s.lo};
}

static inline DoubleLength
Expand Down Expand Up @@ -2941,7 +2944,7 @@ math_sumprod_impl(PyObject *module, PyObject *p, PyObject *q)
bool int_path_enabled = true, int_total_in_use = false;
bool flt_path_enabled = true, flt_total_in_use = false;
long int_total = 0;
DoubleLength flt_total = dl_zero();
DoubleLength flt_total = dl_zero;

p_it = PyObject_GetIter(p);
if (p_it == NULL) {
Expand Down Expand Up @@ -3101,7 +3104,7 @@ math_sumprod_impl(PyObject *module, PyObject *p, PyObject *q)
Py_SETREF(total, new_total);
new_total = NULL;
Py_CLEAR(term_i);
flt_total = dl_zero();
flt_total = dl_zero;
flt_total_in_use = false;
}
}
Expand Down