Lines Matching refs:half
20 // it is done for float32 by default) and (N-1) half-width iteration plus one
21 // final full-width iteration. It is expected that half-width integer
28 // rep_t-sized bitmask with lower half of bits set to ones
131 // rep_t-sized iterations may be slower than the corresponding half-width
134 // NB: Using half-width iterations increases computation errors due to
138 // Starting with (n-1) half-width iterations
146 // Use 16-bit initial estimation in case we are using half-width iterations
179 // For half-width iterations, b_hw is used instead of b.
219 // For initial half-width iterations, U = 2^-HW
267 // <--- higher half ---><--- lower half --->
282 // Just like the case of half-width iterations but with possibility
292 // Then like for the half-width iterations:
302 // u_n for different precisions (with N-1 half-width iterations):