wave8_transpose_16_bf1()

void fl::detail::wave8_transpose_16_bf1	(	const u8	lanes[16],
		u8	W0,
		u8	W1,
		u8	output[16 *sizeof(Wave8Byte)] )

Definition at line 181 of file wave8.cpp.hpp.

                                                                            {
    u8 d_mask[8];
    u8 m0_mask[8];
    const u8 D_byte = W0 ^ W1;
    for (int p = 0; p < 8; ++p) {
        const int shift = 7 - p;
        d_mask[p] = ((D_byte >> shift) & 1) ? 0xFFu : 0x00u;
        m0_mask[p] = ((W0 >> shift) & 1) ? 0xFFu : 0x00u;
    }
    u8 cols[16];
    spread_transpose16_symbol(lanes, cols);
    for (int s = 0; s < 8; ++s) {
        const u8 col_lo = cols[2 * s + 0];
        const u8 col_hi = cols[2 * s + 1];
        for (int p = 0; p < 8; ++p) {
            output[s * 16 + p * 2 + 0] = m0_mask[p] ^ (col_lo & d_mask[p]);
            output[s * 16 + p * 2 + 1] = m0_mask[p] ^ (col_hi & d_mask[p]);
        }
    }
}

References fl::FL_NO_EXCEPT, spread_transpose16_symbol(), fl::W0, and fl::W1.

Referenced by fl::wave8Transpose_16_bf1().

Here is the call graph for this function:

Here is the caller graph for this function: