type usage fixes.

This commit is contained in:
IRainman 2025-12-29 18:48:28 +03:00
parent 147cf3b4c7
commit 351e4d30ab

View File

@ -141,8 +141,8 @@ compute_float(am_pow_t q, am_mant_t w) noexcept {
// branchless approach: value128 product = compute_product(q, w); but in
// practice, we can win big with the compute_product_approximation if its
// additional branch is easily predicted. Which is best is data specific.
auto const upperbit = static_cast<am_bits_t>(product.high >> 63);
am_bits_t const shift = upperbit + 64 - binary::mantissa_explicit_bits() - 3;
auto const upperbit = product.high >> 63;
auto const shift = upperbit + 64 - binary::mantissa_explicit_bits() - 3;
// Shift right the mantissa to the correct position
answer.mantissa = product.high >> shift;