void vpx_fdct4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, src_stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 vec, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
vec = __msa_ceqi_h(in0, 0);
vec = vec ^ 255;
vec = mask & vec;
in0 += vec;
}
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 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);
}
void vp9_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 temp, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
temp = __msa_ceqi_h(in0, 0);
temp = (v8i16)__msa_xori_b((v16u8)temp, 255);
temp = mask & temp;
in0 += temp;
}
switch (tx_type) {
case DCT_DCT:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
}
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
