Module simd

Expand description

SIMD-optimized kernels for tensor operations.

Provides portable SIMD implementations via macerator with automatic dispatch to the best available instruction set:

Enable with the simd feature flag (enabled by default).

Modules§

abs_inplace_f32
add_inplace_f32
add_shared_row_inplace_f32
bool_and_inplace_u8
bool_and_u8
bool_not_inplace_u8
bool_not_u8
bool_or_inplace_u8
bool_or_u8
bool_xor_inplace_u8
bool_xor_u8
cmp_f32
cmp_scalar_f32
div_inplace_f32
div_shared_row_inplace_f32
mask_fill_f32: Conditional fill: out[i] = if mask[i] != 0 { fill_value } else { tensor[i] }
mask_fill_f64: Conditional fill for f64.
mask_fill_i64: Conditional fill for i64.
mask_fill_u8: Conditional fill for u8 (bool tensors).
mask_where_f32: Conditional select: out[i] = if mask[i] != 0 { value[i] } else { tensor[i] }
mask_where_f64: Conditional select for f64.
mask_where_i64: Conditional select for i64.
mask_where_u8: Conditional select for u8 (bool tensors).
mul_inplace_f32
mul_shared_row_inplace_f32
recip_inplace_f32
sub_inplace_f32
sub_shared_row_inplace_f32