Skip to content

Faster _fmpz_vec_scalar_divexact_ui #2371

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 7 commits into from
Oct 31, 2025
Merged

Faster _fmpz_vec_scalar_divexact_ui #2371

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
93a3bbd
Faster _fmpz_vec_scalar_divexact for single-word divisors
fredrik-johansson Jul 18, 2025
57fcdc5
Try to fix _gr_vec_divexact test code
fredrik-johansson Jul 18, 2025
7d6daf2
fmpz_mpoly_scalar_divexact_fmpz
fredrik-johansson Jul 18, 2025
31ccb70
Fix fmpz_mpoly_scalar_divexact_fmpz test
fredrik-johansson Jul 18, 2025
b3990ee
Profiling code; fmpz safety
fredrik-johansson Jul 19, 2025
a11d019
Slightly faster n_binvert
fredrik-johansson Jul 22, 2025
3197a77
Missing underscore
fredrik-johansson Jul 22, 2025
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
33 changes: 24 additions & 9 deletions src/fmpz_vec/scalar_divexact.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,16 +21,15 @@
/* Helper functions that could go in longlong.h, ulong_extras.h
and mpn_extras.h as soon as they find a use elsewhere. */

/* The table is slightly faster but requires GMP internals. */
#define USE BINVERT_TABLE 0

/* Tuned for Zen 3; with assembly code we would not need GMP. */
#define DIVEXACT_1_ODD_GMP_CUTOFF 35
#define DIVEXACT_1_EVEN_GMP_CUTOFF 25

/* The table is slightly faster but requires GMP internals. */
#define USE BINVERT_TABLE 0

#if USE_BINVERT_TABLE
extern const unsigned char __gmp_binvert_limb_table[128];
#endif

/* Invert n mod 2^FLINT_BITS */
static ulong n_binvert(ulong n)
Expand All @@ -39,13 +38,8 @@ static ulong n_binvert(ulong n)

FLINT_ASSERT(n % 2);

#if USE_BINVERT_TABLE
r = __gmp_binvert_limb_table[(n / 2) & 0x7F];
#else
r = (((n + 2) & 4) << 1) ^ n; /* 4 bits */
r = 2 * r - r * r * n; /* 8 bits */
#endif

r = 2 * r - r * r * n; /* 16 bits */
r = 2 * r - r * r * n; /* 32 bits */
#if FLINT_BITS == 64
Expand All @@ -54,6 +48,27 @@ static ulong n_binvert(ulong n)

return r;
}
#else

/* Use Hurchalla's algorithm https://arxiv.org/abs/2204.04342 */
static ulong n_binvert(ulong a)
{
ulong r, y;

r = (3 * a) ^ 2; /* 5 bits */
y = 1 - a * r;
r = r * (1 + y); /* 10 bits */
y *= y;
r = r * (1 + y); /* 20 bits */
y *= y;
r = r * (1 + y); /* 40 bits */
#if FLINT_BITS == 64
y *= y;
r = r * (1 + y); /* 80 bits */
#endif
return r;
}
#endif

FLINT_FORCE_INLINE
ulong n_mulhi(ulong a, ulong b)
Expand Down
Loading