Skip to main content

tiled_dot

hermes_simd

Function tiled_dot 

Source 
pub fn tiled_dot<T, Arch, Align, const TILE_M: usize>(
    a: &SimdView<'_, T, Arch, Align>,
    b: &SimdView<'_, T, Arch, Align>,
) -> Result<T, SimdError>
where
    Arch: SimdArch + SimdKernel<T>,
    Align: Alignment,
    T: NumericElement,
Expand description

Compute the dot product of two slices using TILE_M independent vector accumulators.

The inner loop processes TILE_M * LANE_COUNT elements per iteration, holding TILE_M accumulator registers simultaneously to saturate FMA throughput.