yuyv to bgr

yuyv2bgr.c

#include <arm_neon.h>


void yuyv2bgr(const unsigned char *yuv, int len, unsigned char *bgr) {
  int i, j = 0;
  for(i = 0; i < len; i += 4) {
    const unsigned char *pyuv = yuv+i;
    unsigned char *pbgr = bgr+j;
    int r = (22987 * ((pyuv)[3] - 128)) >> 14;
    int g = (-5636 * ((pyuv)[1] - 128) - 11698 * ((pyuv)[3] - 128)) >> 14;
    int b = (29049 * ((pyuv)[1] - 128)) >> 14;
    (pbgr)[0] = (*(pyuv) + b);
    (pbgr)[1] = (*(pyuv) + g);
    (pbgr)[2] = (*(pyuv) + r);
    (pbgr)[3] = ((pyuv)[2] + b);
    (pbgr)[4] = ((pyuv)[2] + g);
    (pbgr)[5] = ((pyuv)[2] + r);
    j+= 6;
  }
}

void yuyv2bgr_neon(const unsigned char *yuv, int len, unsigned char *bgr) {
  int i, j = 0;
  int16x8_t const half = vdupq_n_s16(128);

  for(i = 0; i < len; i += 4*8) {
    const unsigned char *pyuv = yuv+i;
    unsigned char *pbgr = bgr+j;
    uint8x8x4_t yuvval  = vld4_u8(pyuv);
    int16x8_t yuv0 = (int16x8_t)vmovl_u8(yuvval.val[0]);
    int16x8_t yuv1 = vsubq_s16((int16x8_t)vmovl_u8(yuvval.val[1]), half);
    int16x8_t yuv2 = (int16x8_t)vmovl_u8(yuvval.val[2]);
    int16x8_t yuv3 = vsubq_s16((int16x8_t)vmovl_u8(yuvval.val[3]), half);

    //(29049 * ((pyuv)[1] - 128)) >> 14
    int32x4_t tB0 = vshrq_n_s32(vmull_n_s16(vget_low_s16(yuv1), 29049), 14);
    int32x4_t tB1 = vshrq_n_s32(vmull_n_s16(vget_high_s16(yuv1), 29049), 14);

    //(-5636 * ((pyuv)[1] - 128) - 11698 * ((pyuv)[3] - 128)) >> 14;
    int32x4_t tG0 = vshrq_n_s32(vaddq_s32(vmull_n_s16(vget_low_s16(yuv1), -5636), vmull_n_s16(vget_low_s16(yuv3), -11698)), 14);
    int32x4_t tG1 = vshrq_n_s32(vaddq_s32(vmull_n_s16(vget_high_s16(yuv1), -5636), vmull_n_s16(vget_high_s16(yuv3), -11698)), 14);

    //(22987 * ((pyuv)[3] - 128)) >> 14;
    int32x4_t tR0 = vshrq_n_s32(vmull_n_s16(vget_low_s16(yuv3), 22987), 14);
    int32x4_t tR1 = vshrq_n_s32(vmull_n_s16(vget_high_s16(yuv3), 22987), 14);

    int32x4_t yuv00 = vmovl_s16(vget_low_s16(yuv0));
    int32x4_t yuv01 = vmovl_s16(vget_high_s16(yuv0));

    int32x4_t yuv20 = vmovl_s16(vget_low_s16(yuv2));
    int32x4_t yuv21 = vmovl_s16(vget_high_s16(yuv2));
    uint8x8_t r0 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv00, tB0)), vqmovun_s32(vaddq_s32(yuv01, tB1))));
    uint8x8_t r1 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv00, tG0)), vqmovun_s32(vaddq_s32(yuv01, tG1))));
    uint8x8_t r2 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv00, tR0)), vqmovun_s32(vaddq_s32(yuv01, tR1))));

    uint8x8_t r3 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv20, tB0)), vqmovun_s32(vaddq_s32(yuv21, tB1))));
    uint8x8_t r4 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv20, tG0)), vqmovun_s32(vaddq_s32(yuv21, tG1))));
    uint8x8_t r5 = vmovn_u16(vcombine_u16(vqmovun_s32(vaddq_s32(yuv20, tR0)), vqmovun_s32(vaddq_s32(yuv21, tR1))));

    uint16x8x3_t pblock;

    pblock.val[0] = vaddw_u8(vshll_n_u8(r1, 8), r0);
    pblock.val[1] = vaddw_u8(vshll_n_u8(r3, 8), r2);
    pblock.val[2] = vaddw_u8(vshll_n_u8(r5, 8), r4);

    vst3q_u16((uint16_t *)pbgr, pblock);
    j += 6*8;
  }
}