static void avc_dct4x4dc_msa( int16_t *p_src, int16_t *p_dst,
int32_t i_src_stride )
{
v8i16 src0, src1, src2, src3, ver_res0, ver_res1, ver_res2, ver_res3;
v4i32 src0_r, src1_r, src2_r, src3_r, tmp0, tmp1, tmp2, tmp3;
v4i32 hor_res0, hor_res1, hor_res2, hor_res3;
v4i32 ver_res0_r, ver_res1_r, ver_res2_r, ver_res3_r;
LD_SH4( p_src, i_src_stride, src0, src1, src2, src3 );
UNPCK_R_SH_SW( src0, src0_r );
UNPCK_R_SH_SW( src1, src1_r );
UNPCK_R_SH_SW( src2, src2_r );
UNPCK_R_SH_SW( src3, src3_r );
BUTTERFLY_4( src0_r, src2_r, src3_r, src1_r,
tmp0, tmp3, tmp2, tmp1 );
BUTTERFLY_4( tmp0, tmp1, tmp2, tmp3,
hor_res0, hor_res3, hor_res2, hor_res1 );
TRANSPOSE4x4_SW_SW( hor_res0, hor_res1, hor_res2, hor_res3,
hor_res0, hor_res1, hor_res2, hor_res3 );
BUTTERFLY_4( hor_res0, hor_res2, hor_res3, hor_res1,
tmp0, tmp3, tmp2, tmp1 );
BUTTERFLY_4( tmp0, tmp1, tmp2, tmp3,
ver_res0_r, ver_res3_r, ver_res2_r, ver_res1_r );
SRARI_W4_SW( ver_res0_r, ver_res1_r, ver_res2_r, ver_res3_r, 1 );
PCKEV_H4_SH( ver_res0_r, ver_res0_r, ver_res1_r, ver_res1_r,
ver_res2_r, ver_res2_r, ver_res3_r, ver_res3_r,
ver_res0, ver_res1, ver_res2, ver_res3 );
PCKOD_D2_SH( ver_res1, ver_res0, ver_res3, ver_res2, ver_res0, ver_res2 );
ST_SH2( ver_res0, ver_res2, p_dst, 8 );
}
void vp8_short_fdct8x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0, in1, in2, in3;
v8i16 temp0, temp1, tmp0, tmp1;
v8i16 const0, const1, const2;
v8i16 coeff = { 2217, 5352, -5352, 14500, 7500, 12000, 25000, 26000 };
v8i16 zero = { 0 };
v4i32 vec0_w, vec1_w, vec2_w, vec3_w;
LD_SH4(input, pitch / 2, in0, in1, in2, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
SLLI_4V(temp0, temp1, in1, in3, 3);
in0 = temp0 + temp1;
in2 = temp0 - temp1;
SET_DOTP_VALUES(coeff, 0, 1, 2, const1, const2);
temp0 = __msa_splati_h(coeff, 3);
vec1_w = (v4i32)__msa_ilvev_h(zero, temp0);
coeff = __msa_ilvl_h(zero, coeff);
vec3_w = __msa_splati_w((v4i32)coeff, 0);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 12);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
in0 = temp0 + temp1 + 7;
in2 = temp0 - temp1 + 7;
in0 >>= 4;
in2 >>= 4;
SPLATI_W2_SW(coeff, 2, vec3_w, vec1_w);
vec3_w += vec1_w;
vec1_w = __msa_splati_w((v4i32)coeff, 1);
const0 = RET_1_IF_NZERO_H(in3);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 16);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
in1 += const0;
PCKEV_D2_SH(in1, in0, in3, in2, temp0, temp1);
ST_SH2(temp0, temp1, output, 8);
PCKOD_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output + 16, 8);
}