void test_vminQs32 (void)
{
int32x4_t out_int32x4_t;
int32x4_t arg0_int32x4_t;
int32x4_t arg1_int32x4_t;
out_int32x4_t = vminq_s32 (arg0_int32x4_t, arg1_int32x4_t);
}
static inline void yuv2rgb_4x2(const uint8_t *y1, const uint8_t *y2, const uint8_t *u, const uint8_t *v, int16_t *r1, int16_t *g1, int16_t *b1, int16_t *r2, int16_t *g2, int16_t *b2){
int32x4_t ry1;
int32x4_t ry2;
int32x4_t rvug;
int32x4_t rvr;
int32x4_t rub;
int32x4_t rr1,rg1,rb1,rr2,rg2,rb2;
int32x4_t max;
LOAD_Y_PREMULTS(0)
LOAD_Y_PREMULTS(1)
LOAD_Y_PREMULTS(2)
LOAD_Y_PREMULTS(3)
LOAD_UV_PREMULTS(0)
LOAD_UV_PREMULTS(1)
max=vld1q_s32(yuvmax);
/*the following does not work */
//max=vdupq_n_s32(255);
rr1=vaddq_s32(ry1,rvr);
rr2=vaddq_s32(ry2,rvr);
rg1=vaddq_s32(ry1,rvug);
rg2=vaddq_s32(ry2,rvug);
rb1=vaddq_s32(ry1,rub);
rb2=vaddq_s32(ry2,rub);
rr1=vminq_s32(vabsq_s32(rr1),max);
rr2=vminq_s32(vabsq_s32(rr2),max);
rg1=vminq_s32(vabsq_s32(rg1),max);
rg2=vminq_s32(vabsq_s32(rg2),max);
rb1=vminq_s32(vabsq_s32(rb1),max);
rb2=vminq_s32(vabsq_s32(rb2),max);
vst1_s16(r1,vqshrn_n_s32(rr1,13));
vst1_s16(r2,vqshrn_n_s32(rr2,13));
vst1_s16(g1,vqshrn_n_s32(rg1,13));
vst1_s16(g2,vqshrn_n_s32(rg2,13));
vst1_s16(b1,vqshrn_n_s32(rb1,13));
vst1_s16(b2,vqshrn_n_s32(rb2,13));
}
void ne10_img_vresize_linear_neon (const int** src, unsigned char* dst, const short* beta, int width)
{
const int *S0 = src[0], *S1 = src[1];
int32x4_t qS0_0123, qS0_4567, qS1_0123, qS1_4567;
int32x4_t qT_0123, qT_4567;
int16x4_t dT_0123, dT_4567;
uint16x8_t qT_01234567;
uint8x8_t dT_01234567, dDst_01234567;
int32x2_t dBeta;
dBeta = vset_lane_s32 ( (int) (beta[0]), dBeta, 0);
dBeta = vset_lane_s32 ( (int) (beta[1]), dBeta, 1);
int32x4_t qDelta, qMin, qMax;
qDelta = vdupq_n_s32 (DELTA);
qMin = vdupq_n_s32 (0);
qMax = vdupq_n_s32 (255);
int x = 0;
for (; x <= width - 8; x += 8)
{
qS0_0123 = vld1q_s32 (&S0[x]);
qS0_4567 = vld1q_s32 (&S0[x + 4]);
qS1_0123 = vld1q_s32 (&S1[x]);
qS1_4567 = vld1q_s32 (&S1[x + 4]);
qT_0123 = vmulq_lane_s32 (qS0_0123, dBeta, 0);
qT_4567 = vmulq_lane_s32 (qS0_4567, dBeta, 0);
qT_0123 = vmlaq_lane_s32 (qT_0123, qS1_0123, dBeta, 1);
qT_4567 = vmlaq_lane_s32 (qT_4567, qS1_4567, dBeta, 1);
qT_0123 = vaddq_s32 (qT_0123, qDelta);
qT_4567 = vaddq_s32 (qT_4567, qDelta);
qT_0123 = vshrq_n_s32 (qT_0123, BITS);
qT_4567 = vshrq_n_s32 (qT_4567, BITS);
qT_0123 = vmaxq_s32 (qT_0123, qMin);
qT_4567 = vmaxq_s32 (qT_4567, qMin);
qT_0123 = vminq_s32 (qT_0123, qMax);
qT_4567 = vminq_s32 (qT_4567, qMax);
dT_0123 = vmovn_s32 (qT_0123);
dT_4567 = vmovn_s32 (qT_4567);
qT_01234567 = vreinterpretq_u16_s16 (vcombine_s16 (dT_0123, dT_4567));
dT_01234567 = vmovn_u16 (qT_01234567);
vst1_u8 (&dst[x], dT_01234567);
}
if (x < width)
{
uint8x8_t dMask;
dMask = vld1_u8 ( (uint8_t *) (&ne10_img_vresize_linear_mask_residual_table[ (width - x - 1)]));
dDst_01234567 = vld1_u8 (&dst[x]);
qS0_0123 = vld1q_s32 (&S0[x]);
qS0_4567 = vld1q_s32 (&S0[x + 4]);
qS1_0123 = vld1q_s32 (&S1[x]);
qS1_4567 = vld1q_s32 (&S1[x + 4]);
qT_0123 = vmulq_lane_s32 (qS0_0123, dBeta, 0);
qT_4567 = vmulq_lane_s32 (qS0_4567, dBeta, 0);
qT_0123 = vmlaq_lane_s32 (qT_0123, qS1_0123, dBeta, 1);
qT_4567 = vmlaq_lane_s32 (qT_4567, qS1_4567, dBeta, 1);
qT_0123 = vaddq_s32 (qT_0123, qDelta);
qT_4567 = vaddq_s32 (qT_4567, qDelta);
qT_0123 = vshrq_n_s32 (qT_0123, BITS);
qT_4567 = vshrq_n_s32 (qT_4567, BITS);
qT_0123 = vmaxq_s32 (qT_0123, qMin);
qT_4567 = vmaxq_s32 (qT_4567, qMin);
qT_0123 = vminq_s32 (qT_0123, qMax);
qT_4567 = vminq_s32 (qT_4567, qMax);
dT_0123 = vmovn_s32 (qT_0123);
dT_4567 = vmovn_s32 (qT_4567);
qT_01234567 = vreinterpretq_u16_s16 (vcombine_s16 (dT_0123, dT_4567));
dT_01234567 = vmovn_u16 (qT_01234567);
dMask = vbsl_u8 (dMask, dT_01234567, dDst_01234567);
vst1_u8 (&dst[x], dMask);
}
}
inline int32x4_t vminq(const int32x4_t & v0, const int32x4_t & v1) { return vminq_s32(v0, v1); }
int32x4_t
test_vminq_s32 (int32x4_t __a, int32x4_t __b)
{
return vminq_s32(__a, __b);
}
