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 FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
uint64_t out0, out1, out2, out3;
uint32_t in0, in1, in2, in3;
v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
v8i16 t0, t1, t2, t3;
v16u8 srcl0, srcl1, src0, src1;
const v8i16 mask0 = { 0, 4, 8, 12, 1, 5, 9, 13 };
const v8i16 mask1 = { 3, 7, 11, 15, 2, 6, 10, 14 };
const v8i16 mask2 = { 4, 0, 5, 1, 6, 2, 7, 3 };
const v8i16 mask3 = { 0, 4, 1, 5, 2, 6, 3, 7 };
const v8i16 cnst0 = { 2217, -5352, 2217, -5352, 2217, -5352, 2217, -5352 };
const v8i16 cnst1 = { 5352, 2217, 5352, 2217, 5352, 2217, 5352, 2217 };
LW4(src, BPS, in0, in1, in2, in3);
INSERT_W4_UB(in0, in1, in2, in3, src0);
LW4(ref, BPS, in0, in1, in2, in3);
INSERT_W4_UB(in0, in1, in2, in3, src1);
ILVRL_B2_UB(src0, src1, srcl0, srcl1);
HSUB_UB2_SH(srcl0, srcl1, t0, t1);
VSHF_H2_SH(t0, t1, t0, t1, mask0, mask1, t2, t3);
ADDSUB2(t2, t3, t0, t1);
t0 = SRLI_H(t0, 3);
VSHF_H2_SH(t0, t0, t1, t1, mask2, mask3, t3, t2);
tmp0 = __msa_hadd_s_w(t3, t3);
tmp2 = __msa_hsub_s_w(t3, t3);
FILL_W2_SW(1812, 937, tmp1, tmp3);
DPADD_SH2_SW(t2, t2, cnst0, cnst1, tmp3, tmp1);
SRAI_W2_SW(tmp1, tmp3, 9);
PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, t0, t1);
VSHF_H2_SH(t0, t1, t0, t1, mask0, mask1, t2, t3);
ADDSUB2(t2, t3, t0, t1);
VSHF_H2_SH(t0, t0, t1, t1, mask2, mask3, t3, t2);
tmp0 = __msa_hadd_s_w(t3, t3);
tmp2 = __msa_hsub_s_w(t3, t3);
ADDVI_W2_SW(tmp0, 7, tmp2, 7, tmp0, tmp2);
SRAI_W2_SW(tmp0, tmp2, 4);
FILL_W2_SW(12000, 51000, tmp1, tmp3);
DPADD_SH2_SW(t2, t2, cnst0, cnst1, tmp3, tmp1);
SRAI_W2_SW(tmp1, tmp3, 16);
UNPCK_R_SH_SW(t1, tmp4);
tmp5 = __msa_ceqi_w(tmp4, 0);
tmp4 = (v4i32)__msa_nor_v((v16u8)tmp5, (v16u8)tmp5);
tmp5 = __msa_fill_w(1);
tmp5 = (v4i32)__msa_and_v((v16u8)tmp5, (v16u8)tmp4);
tmp1 += tmp5;
PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, t0, t1);
out0 = __msa_copy_s_d((v2i64)t0, 0);
out1 = __msa_copy_s_d((v2i64)t0, 1);
out2 = __msa_copy_s_d((v2i64)t1, 0);
out3 = __msa_copy_s_d((v2i64)t1, 1);
SD4(out0, out1, out2, out3, out, 8);
}
static void avc_zigzag_scan_4x4_frame_msa( int16_t pi_dct[16],
int16_t pi_level[16] )
{
v8i16 src0, src1;
v8i16 mask0 = { 0, 4, 1, 2, 5, 8, 12, 9 };
v8i16 mask1 = { 6, 3, 7, 10, 13, 14, 11, 15 };
LD_SH2( pi_dct, 8, src0, src1 );
VSHF_H2_SH( src0, src1, src0, src1, mask0, mask1, mask0, mask1 );
ST_SH2( mask0, mask1, pi_level, 8 );
}
static void FTransformWHT(const int16_t* in, int16_t* out) {
v8i16 in0 = { 0 };
v8i16 in1 = { 0 };
v8i16 tmp0, tmp1, tmp2, tmp3;
v8i16 out0, out1;
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 };
in0 = __msa_insert_h(in0, 0, in[ 0]);
in0 = __msa_insert_h(in0, 1, in[ 64]);
in0 = __msa_insert_h(in0, 2, in[128]);
in0 = __msa_insert_h(in0, 3, in[192]);
in0 = __msa_insert_h(in0, 4, in[ 16]);
in0 = __msa_insert_h(in0, 5, in[ 80]);
in0 = __msa_insert_h(in0, 6, in[144]);
in0 = __msa_insert_h(in0, 7, in[208]);
in1 = __msa_insert_h(in1, 0, in[ 48]);
in1 = __msa_insert_h(in1, 1, in[112]);
in1 = __msa_insert_h(in1, 2, in[176]);
in1 = __msa_insert_h(in1, 3, in[240]);
in1 = __msa_insert_h(in1, 4, in[ 32]);
in1 = __msa_insert_h(in1, 5, in[ 96]);
in1 = __msa_insert_h(in1, 6, in[160]);
in1 = __msa_insert_h(in1, 7, in[224]);
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, out0, out1);
SRAI_H2_SH(out0, out1, 1);
ST_SH2(out0, out1, out, 8);
}
