int32_t vp8_block_error_msa(int16_t *coeff_ptr, int16_t *dq_coeff_ptr)
{
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, dq_coeff, coeff0, coeff1;
v4i32 diff0, diff1;
v2i64 err0 = { 0 };
v2i64 err1 = { 0 };
for (loop_cnt = 2; loop_cnt--;)
{
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
coeff_ptr += 8;
dq_coeff_ptr += 8;
}
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err = __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
return err;
}
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);
}
int32_t vp8_mbblock_error_msa(MACROBLOCK *mb, int32_t dc)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1;
v16u8 zero = { 0 };
v16u8 mask0 = (v16u8)__msa_ldi_b(255);
if (1 == dc)
{
mask0 = (v16u8)__msa_insve_w((v4i32)mask0, 0, (v4i32)zero);
}
for (loop_cnt = 0; loop_cnt < 8; loop_cnt++)
{
be = &mb->block[2 * loop_cnt];
bd = &mb->e_mbd.block[2 * loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[2 * loop_cnt + 1];
bd = &mb->e_mbd.block[2 * loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
int32_t vp8_mbuverror_msa(MACROBLOCK *mb)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1, err_dup0, err_dup1;
for (loop_cnt = 16; loop_cnt < 24; loop_cnt += 2)
{
be = &mb->block[loop_cnt];
bd = &mb->e_mbd.block[loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[loop_cnt + 1];
bd = &mb->e_mbd.block[loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
