static int TTransform(const uint8_t* in, const uint16_t* w) {
int sum;
uint32_t in0_m, in1_m, in2_m, in3_m;
v16i8 src0;
v8i16 in0, in1, tmp0, tmp1, tmp2, tmp3;
v4i32 dst0, dst1;
const v16i8 zero = { 0 };
const v8i16 mask0 = { 0, 1, 2, 3, 8, 9, 10, 11 };
const v8i16 mask1 = { 4, 5, 6, 7, 12, 13, 14, 15 };
const v8i16 mask2 = { 0, 4, 8, 12, 1, 5, 9, 13 };
const v8i16 mask3 = { 3, 7, 11, 15, 2, 6, 10, 14 };
LW4(in, BPS, in0_m, in1_m, in2_m, in3_m);
INSERT_W4_SB(in0_m, in1_m, in2_m, in3_m, src0);
ILVRL_B2_SH(zero, src0, tmp0, tmp1);
VSHF_H2_SH(tmp0, tmp1, tmp0, tmp1, mask2, mask3, in0, in1);
ADDSUB2(in0, in1, tmp0, tmp1);
VSHF_H2_SH(tmp0, tmp1, tmp0, tmp1, mask0, mask1, tmp2, tmp3);
ADDSUB2(tmp2, tmp3, tmp0, tmp1);
VSHF_H2_SH(tmp0, tmp1, tmp0, tmp1, mask2, mask3, in0, in1);
ADDSUB2(in0, in1, tmp0, tmp1);
VSHF_H2_SH(tmp0, tmp1, tmp0, tmp1, mask0, mask1, tmp2, tmp3);
ADDSUB2(tmp2, tmp3, tmp0, tmp1);
tmp0 = __msa_add_a_h(tmp0, (v8i16)zero);
tmp1 = __msa_add_a_h(tmp1, (v8i16)zero);
LD_SH2(w, 8, tmp2, tmp3);
DOTP_SH2_SW(tmp0, tmp1, tmp2, tmp3, dst0, dst1);
dst0 = dst0 + dst1;
sum = HADD_SW_S32(dst0);
return sum;
}
static void CollectHistogram(const uint8_t* ref, const uint8_t* pred,
int start_block, int end_block,
VP8Histogram* const histo) {
int j;
int distribution[MAX_COEFF_THRESH + 1] = { 0 };
for (j = start_block; j < end_block; ++j) {
int16_t out[16];
VP8FTransform(ref + VP8DspScan[j], pred + VP8DspScan[j], out);
{
int k;
v8i16 coeff0, coeff1;
const v8i16 zero = { 0 };
const v8i16 max_coeff_thr = __msa_ldi_h(MAX_COEFF_THRESH);
LD_SH2(&out[0], 8, coeff0, coeff1);
coeff0 = __msa_add_a_h(coeff0, zero);
coeff1 = __msa_add_a_h(coeff1, zero);
SRAI_H2_SH(coeff0, coeff1, 3);
coeff0 = __msa_min_s_h(coeff0, max_coeff_thr);
coeff1 = __msa_min_s_h(coeff1, max_coeff_thr);
ST_SH2(coeff0, coeff1, &out[0], 8);
for (k = 0; k < 16; ++k) {
++distribution[out[k]];
}
}
}
VP8SetHistogramData(distribution, histo);
}
