pub unsafe fn scale_c_row(row: u32, alpha: f32, scratch: u8)Expand description
Broadcast alpha into f[scratch], then scale the PS register pair for row.
Convenience wrapper: calls broadcast_ps(alpha, scratch) then
fmul_ps_row(row, scratch). The broadcast is repeated on every call;
when scaling multiple rows with the same alpha, call broadcast_ps
once and fmul_ps_row for each row instead.
Pass PS_SCRATCH_DEFAULT (28) for scratch when the C tile has at
most 14 rows. For full 16-row tiles, see the module-level doc for the
spill/restore pattern.
ยงSafety
Call tensor_wait(TensorEvent::Fma) before this function; the tensor
co-processor must have finished writing the FP register file.