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_walsh4x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0_h, in1_h, in2_h, in3_h;
v4i32 in0_w, in1_w, in2_w, in3_w, temp0, temp1, temp2, temp3;
LD_SH4(input, pitch / 2, in0_h, in1_h, in2_h, in3_h);
TRANSPOSE4x4_SH_SH(in0_h, in1_h, in2_h, in3_h, in0_h, in1_h, in2_h, in3_h);
UNPCK_R_SH_SW(in0_h, in0_w);
UNPCK_R_SH_SW(in1_h, in1_w);
UNPCK_R_SH_SW(in2_h, in2_w);
UNPCK_R_SH_SW(in3_h, in3_w);
BUTTERFLY_4(in0_w, in1_w, in3_w, in2_w, temp0, temp3, temp2, temp1);
SLLI_4V(temp0, temp1, temp2, temp3, 2);
BUTTERFLY_4(temp0, temp1, temp2, temp3, in0_w, in1_w, in2_w, in3_w);
temp0 = RET_1_IF_NZERO_W(temp0);
in0_w += temp0;
TRANSPOSE4x4_SW_SW(in0_w, in1_w, in2_w, in3_w, in0_w, in1_w, in2_w, in3_w);
BUTTERFLY_4(in0_w, in1_w, in3_w, in2_w, temp0, temp3, temp2, temp1);
BUTTERFLY_4(temp0, temp1, temp2, temp3, in0_w, in1_w, in2_w, in3_w);
in0_w += RET_1_IF_NEG_W(in0_w);
in1_w += RET_1_IF_NEG_W(in1_w);
in2_w += RET_1_IF_NEG_W(in2_w);
in3_w += RET_1_IF_NEG_W(in3_w);
ADD4(in0_w, 3, in1_w, 3, in2_w, 3, in3_w, 3, in0_w, in1_w, in2_w, in3_w);
SRA_4V(in0_w, in1_w, in2_w, in3_w, 3);
PCKEV_H2_SH(in1_w, in0_w, in3_w, in2_w, in0_h, in1_h);
ST_SH2(in0_h, in1_h, output, 8);
}
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 vp9_fwht4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4;
LD_SH4(input, src_stride, in0, in1, in2, in3);
in0 += in1;
in3 -= in2;
in4 = (in0 - in3) >> 1;
SUB2(in4, in1, in4, in2, in1, in2);
in0 -= in2;
in3 += in1;
TRANSPOSE4x4_SH_SH(in0, in2, in3, in1, in0, in2, in3, in1);
in0 += in2;
in1 -= in3;
in4 = (in0 - in1) >> 1;
SUB2(in4, in2, in4, in3, in2, in3);
in0 -= in3;
in1 += in2;
SLLI_4V(in0, in1, in2, in3, 2);
TRANSPOSE4x4_SH_SH(in0, in3, in1, in2, in0, in3, in1, in2);
ST4x2_UB(in0, output, 4);
ST4x2_UB(in3, output + 4, 4);
ST4x2_UB(in1, output + 8, 4);
ST4x2_UB(in2, output + 12, 4);
}
static void avc_idct4x4dc_msa( int16_t *p_src, int32_t i_src_stride,
int16_t *p_dst, int32_t i_dst_stride )
{
v8i16 src0, src1, src2, src3;
v4i32 src0_r, src1_r, src2_r, src3_r;
v4i32 hres0, hres1, hres2, hres3;
v8i16 vres0, vres1, vres2, vres3;
v4i32 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v2i64 res0, res1;
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, vec0, vec3, vec2, vec1 );
BUTTERFLY_4( vec0, vec1, vec2, vec3, hres0, hres3, hres2, hres1 );
TRANSPOSE4x4_SW_SW( hres0, hres1, hres2, hres3,
hres0, hres1, hres2, hres3 );
BUTTERFLY_4( hres0, hres2, hres3, hres1, vec0, vec3, vec2, vec1 );
BUTTERFLY_4( vec0, vec1, vec2, vec3, vec4, vec7, vec6, vec5 );
PCKEV_H4_SH( vec4, vec4, vec5, vec5, vec6, vec6, vec7, vec7,
vres0, vres1, vres2, vres3 );
PCKOD_D2_SD( vres1, vres0, vres3, vres2, res0, res1 );
ST8x4_UB( res0, res1, p_dst, i_dst_stride * 2 );
}
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);
}
static void fdct8x32_1d_column_load_butterfly(const int16_t *input,
int32_t src_stride,
int16_t *temp_buff) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 step0, step1, step2, step3;
v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
v8i16 step0_1, step1_1, step2_1, step3_1;
/* 1st and 2nd set */
LD_SH4(input, src_stride, in0, in1, in2, in3);
LD_SH4(input + (28 * src_stride), src_stride, in4, in5, in6, in7);
LD_SH4(input + (4 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
LD_SH4(input + (24 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
step0, step1, step2, step3, in4, in5, in6, in7);
BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
ST_SH4(step0, step1, step2, step3, temp_buff, 8);
ST_SH4(in4, in5, in6, in7, temp_buff + (28 * 8), 8);
ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (4 * 8), 8);
ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (24 * 8), 8);
/* 3rd and 4th set */
LD_SH4(input + (8 * src_stride), src_stride, in0, in1, in2, in3);
LD_SH4(input + (20 * src_stride), src_stride, in4, in5, in6, in7);
LD_SH4(input + (12 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
LD_SH4(input + (16 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
step0, step1, step2, step3, in4, in5, in6, in7);
BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
ST_SH4(step0, step1, step2, step3, temp_buff + (8 * 8), 8);
ST_SH4(in4, in5, in6, in7, temp_buff + (20 * 8), 8);
ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (12 * 8), 8);
ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (15 * 8) + 8, 8);
}
void vp8_short_fdct4x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0, in1, in2, in3;
v8i16 temp0, temp1;
v8i16 const0, const1;
v8i16 coeff = { 2217, 5352, -5352, 14500, 7500, 12000, 25000, 26000 };
v4i32 out0, out1, out2, out3;
v8i16 zero = { 0 };
LD_SH4(input, pitch / 2, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(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, const0, const1);
temp0 = __msa_ilvr_h(in3, in1);
in1 = __msa_splati_h(coeff, 3);
out0 = (v4i32)__msa_ilvev_h(zero, in1);
coeff = __msa_ilvl_h(zero, coeff);
out1 = __msa_splati_w((v4i32)coeff, 0);
DPADD_SH2_SW(temp0, temp0, const0, const1, out0, out1);
out0 >>= 12;
out1 >>= 12;
PCKEV_H2_SH(out0, out0, out1, out1, in1, in3);
TRANSPOSE4x4_SH_SH(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;
ILVR_H2_SW(zero, in0, zero, in2, out0, out2);
temp1 = RET_1_IF_NZERO_H(in3);
ILVR_H2_SH(zero, temp1, in3, in1, temp1, temp0);
SPLATI_W2_SW(coeff, 2, out3, out1);
out3 += out1;
out1 = __msa_splati_w((v4i32)coeff, 1);
DPADD_SH2_SW(temp0, temp0, const0, const1, out1, out3);
out1 >>= 16;
out3 >>= 16;
out1 += (v4i32)temp1;
PCKEV_H2_SH(out1, out0, out3, out2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
static void fdct8x32_1d_column_even_store(int16_t *input, int16_t *temp) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v8i16 temp0, temp1;
/* fdct even */
LD_SH4(input, 8, in0, in1, in2, in3);
LD_SH4(input + 96, 8, in12, in13, in14, in15);
BUTTERFLY_8(in0, in1, in2, in3, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, in12, in13, in14, in15);
LD_SH4(input + 32, 8, in4, in5, in6, in7);
LD_SH4(input + 64, 8, in8, in9, in10, in11);
BUTTERFLY_8(in4, in5, in6, in7, in8, in9, in10, in11,
vec4, vec5, vec6, vec7, in8, in9, in10, in11);
/* Stage 3 */
ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp);
ST_SH(temp1, temp + 512);
DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 256);
ST_SH(temp1, temp + 768);
SUB4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, vec7, vec6, vec5, vec4);
DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 128);
ST_SH(temp1, temp + 896);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 640);
ST_SH(temp1, temp + 384);
DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
ADD2(in0, in1, in2, in3, vec0, vec7);
DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 64);
ST_SH(temp1, temp + 960);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 576);
ST_SH(temp1, temp + 448);
SUB2(in9, vec2, in14, vec5, vec2, vec5);
DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 320);
ST_SH(temp1, temp + 704);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
ST_SH(temp0, temp + 192);
ST_SH(temp1, temp + 832);
}
static void fdct8x32_1d_column_odd_store(int16_t *input, int16_t *temp_ptr) {
v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
in20 = LD_SH(input + 32);
in21 = LD_SH(input + 40);
in26 = LD_SH(input + 80);
in27 = LD_SH(input + 88);
DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
in18 = LD_SH(input + 16);
in19 = LD_SH(input + 24);
in28 = LD_SH(input + 96);
in29 = LD_SH(input + 104);
vec4 = in19 - in20;
ST_SH(vec4, input + 32);
vec4 = in18 - in21;
ST_SH(vec4, input + 40);
vec4 = in29 - in26;
ST_SH(vec4, input + 80);
vec4 = in28 - in27;
ST_SH(vec4, input + 88);
in21 = in18 + in21;
in20 = in19 + in20;
in27 = in28 + in27;
in26 = in29 + in26;
LD_SH4(input + 48, 8, in22, in23, in24, in25);
DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
in16 = LD_SH(input);
in17 = LD_SH(input + 8);
in30 = LD_SH(input + 112);
in31 = LD_SH(input + 120);
vec4 = in17 - in22;
ST_SH(vec4, input + 16);
vec4 = in16 - in23;
ST_SH(vec4, input + 24);
vec4 = in31 - in24;
ST_SH(vec4, input + 96);
vec4 = in30 - in25;
ST_SH(vec4, input + 104);
ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
ADD2(in27, in26, in25, in24, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr);
ST_SH(vec4, temp_ptr + 960);
SUB2(in27, in26, in25, in24, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 448);
ST_SH(vec4, temp_ptr + 512);
SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
SUB2(in26, in27, in24, in25, in23, in20);
DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec4, temp_ptr + 704);
ST_SH(vec5, temp_ptr + 256);
ADD2(in26, in27, in24, in25, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec4, temp_ptr + 192);
ST_SH(vec5, temp_ptr + 768);
LD_SH4(input + 16, 8, in22, in23, in20, in21);
LD_SH4(input + 80, 8, in26, in27, in24, in25);
in16 = in20;
in17 = in21;
DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
ADD2(in28, in29, in31, in30, in16, in19);
DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 832);
ST_SH(vec4, temp_ptr + 128);
SUB2(in28, in29, in31, in30, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 320);
ST_SH(vec4, temp_ptr + 640);
ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
SUB2(in29, in28, in30, in31, in16, in19);
DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 576);
ST_SH(vec4, temp_ptr + 384);
ADD2(in29, in28, in30, in31, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
ST_SH(vec5, temp_ptr + 64);
ST_SH(vec4, temp_ptr + 896);
}
static void hevc_addblk_32x32_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t loop_cnt;
uint8_t *temp_dst = dst;
v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
v8i16 dst_r0, dst_l0, dst_r1, dst_l1, dst_r2, dst_l2, dst_r3, dst_l3;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
for (loop_cnt = 14; loop_cnt--;) {
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
}
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
}
static void hevc_idct_8x32_column_msa(int16_t *coeffs, uint8_t buf_pitch,
uint8_t round)
{
uint8_t i;
const int16_t *filter_ptr0 = >32x32_cnst0[0];
const int16_t *filter_ptr1 = >32x32_cnst1[0];
const int16_t *filter_ptr2 = >32x32_cnst2[0];
const int16_t *filter_ptr3 = >8x8_cnst[0];
int16_t *src0 = (coeffs + buf_pitch);
int16_t *src1 = (coeffs + 2 * buf_pitch);
int16_t *src2 = (coeffs + 4 * buf_pitch);
int16_t *src3 = (coeffs);
int32_t cnst0, cnst1;
int32_t tmp_buf[8 * 32 + 15];
int32_t *tmp_buf_ptr = tmp_buf + 15;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r;
v8i16 src0_l, src1_l, src2_l, src3_l, src4_l, src5_l, src6_l, src7_l;
v8i16 filt0, filter0, filter1, filter2, filter3;
v4i32 sum0_r, sum0_l, sum1_r, sum1_l, tmp0_r, tmp0_l, tmp1_r, tmp1_l;
/* Align pointer to 64 byte boundary */
tmp_buf_ptr = (int32_t *)(((uintptr_t) tmp_buf_ptr) & ~(uintptr_t) 63);
/* process coeff 4, 12, 20, 28 */
LD_SH4(src2, 8 * buf_pitch, in0, in1, in2, in3);
ILVR_H2_SH(in1, in0, in3, in2, src0_r, src1_r);
ILVL_H2_SH(in1, in0, in3, in2, src0_l, src1_l);
LD_SH2(src3, 16 * buf_pitch, in4, in6);
LD_SH2((src3 + 8 * buf_pitch), 16 * buf_pitch, in5, in7);
ILVR_H2_SH(in6, in4, in7, in5, src2_r, src3_r);
ILVL_H2_SH(in6, in4, in7, in5, src2_l, src3_l);
/* loop for all columns of constants */
for (i = 0; i < 2; i++) {
/* processing single column of constants */
cnst0 = LW(filter_ptr2);
cnst1 = LW(filter_ptr2 + 2);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH2_SW(src1_r, src1_l, filter1, filter1, sum0_r, sum0_l);
ST_SW2(sum0_r, sum0_l, (tmp_buf_ptr + 2 * i * 8), 4);
/* processing single column of constants */
cnst0 = LW(filter_ptr2 + 4);
cnst1 = LW(filter_ptr2 + 6);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH2_SW(src1_r, src1_l, filter1, filter1, sum0_r, sum0_l);
ST_SW2(sum0_r, sum0_l, (tmp_buf_ptr + (2 * i + 1) * 8), 4);
filter_ptr2 += 8;
}
/* process coeff 0, 8, 16, 24 */
/* loop for all columns of constants */
for (i = 0; i < 2; i++) {
/* processing first column of filter constants */
cnst0 = LW(filter_ptr3);
cnst1 = LW(filter_ptr3 + 2);
filter0 = (v8i16) __msa_fill_w(cnst0);
filter1 = (v8i16) __msa_fill_w(cnst1);
DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l, filter0, filter0, filter1,
filter1, sum0_r, sum0_l, tmp1_r, tmp1_l);
sum1_r = sum0_r - tmp1_r;
sum1_l = sum0_l - tmp1_l;
sum0_r = sum0_r + tmp1_r;
sum0_l = sum0_l + tmp1_l;
HEVC_EVEN16_CALC(tmp_buf_ptr, sum0_r, sum0_l, i, (7 - i));
HEVC_EVEN16_CALC(tmp_buf_ptr, sum1_r, sum1_l, (3 - i), (4 + i));
filter_ptr3 += 8;
}
/* process coeff 2 6 10 14 18 22 26 30 */
LD_SH8(src1, 4 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_r, src1_r, src2_r, src3_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_l, src1_l, src2_l, src3_l);
/* loop for all columns of constants */
for (i = 0; i < 8; i++) {
/* processing single column of constants */
filt0 = LD_SH(filter_ptr1);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
tmp1_r = tmp0_r;
tmp1_l = tmp0_l;
tmp0_r += sum0_r;
tmp0_l += sum0_l;
ST_SW2(tmp0_r, tmp0_l, (tmp_buf_ptr + i * 8), 4);
tmp1_r -= sum0_r;
tmp1_l -= sum0_l;
ST_SW2(tmp1_r, tmp1_l, (tmp_buf_ptr + (15 - i) * 8), 4);
filter_ptr1 += 8;
}
/* process coeff 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 */
LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
src0 += 16 * buf_pitch;
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_r, src1_r, src2_r, src3_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src0_l, src1_l, src2_l, src3_l);
LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src4_r, src5_r, src6_r, src7_r);
ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
src4_l, src5_l, src6_l, src7_l);
/* loop for all columns of filter constants */
for (i = 0; i < 16; i++) {
/* processing single column of constants */
filt0 = LD_SH(filter_ptr0);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
tmp1_r = sum0_r;
tmp1_l = sum0_l;
filt0 = LD_SH(filter_ptr0 + 8);
SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
DOTP_SH2_SW(src4_r, src4_l, filter0, filter0, sum0_r, sum0_l);
DPADD_SH4_SW(src5_r, src5_l, src6_r, src6_l, filter1, filter1, filter2,
filter2, sum0_r, sum0_l, sum0_r, sum0_l);
DPADD_SH2_SW(src7_r, src7_l, filter3, filter3, sum0_r, sum0_l);
sum0_r += tmp1_r;
sum0_l += tmp1_l;
LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
tmp1_r = tmp0_r;
tmp1_l = tmp0_l;
tmp0_r += sum0_r;
tmp0_l += sum0_l;
sum1_r = __msa_fill_w(round);
SRAR_W2_SW(tmp0_r, tmp0_l, sum1_r);
SAT_SW2_SW(tmp0_r, tmp0_l, 15);
in0 = __msa_pckev_h((v8i16) tmp0_l, (v8i16) tmp0_r);
ST_SH(in0, (coeffs + i * buf_pitch));
tmp1_r -= sum0_r;
tmp1_l -= sum0_l;
SRAR_W2_SW(tmp1_r, tmp1_l, sum1_r);
SAT_SW2_SW(tmp1_r, tmp1_l, 15);
in0 = __msa_pckev_h((v8i16) tmp1_l, (v8i16) tmp1_r);
ST_SH(in0, (coeffs + (31 - i) * buf_pitch));
filter_ptr0 += 16;
}
}
static int32_t avc_quant_8x8_msa( int16_t *p_dct, uint16_t *p_mf,
uint16_t *p_bias )
{
int32_t non_zero = 0;
v8i16 dct0, dct1, dct2, dct3;
v8i16 zero = { 0 };
v8i16 dct0_mask, dct1_mask, dct2_mask, dct3_mask;
v8i16 dct_h0, dct_h1, dct_h2, dct_h3, mf_h0, mf_h1, mf_h2, mf_h3;
v8i16 bias_h0, bias_h1, bias_h2, bias_h3;
v4i32 dct_w0, dct_w1, dct_w2, dct_w3, dct_w4, dct_w5, dct_w6, dct_w7;
v4i32 dct_signed_w0, dct_signed_w1, dct_signed_w2, dct_signed_w3;
v4i32 dct_signed_w4, dct_signed_w5, dct_signed_w6, dct_signed_w7;
v4i32 mf_vec0, mf_vec1, mf_vec2, mf_vec3;
v4i32 mf_vec4, mf_vec5, mf_vec6, mf_vec7;
v4i32 bias0, bias1, bias2, bias3, bias4, bias5, bias6, bias7;
LD_SH4( p_dct, 8, dct0, dct1, dct2, dct3 );
dct0_mask = __msa_clei_s_h( dct0, 0 );
dct1_mask = __msa_clei_s_h( dct1, 0 );
dct2_mask = __msa_clei_s_h( dct2, 0 );
dct3_mask = __msa_clei_s_h( dct3, 0 );
UNPCK_SH_SW( dct0, dct_signed_w0, dct_signed_w1 );
UNPCK_SH_SW( dct1, dct_signed_w2, dct_signed_w3 );
UNPCK_SH_SW( dct2, dct_signed_w4, dct_signed_w5 );
UNPCK_SH_SW( dct3, dct_signed_w6, dct_signed_w7 );
LD_SH4( p_bias, 8, bias_h0, bias_h1, bias_h2, bias_h3 );
ILVR_H4_SW( zero, bias_h0, zero, bias_h1, zero, bias_h2, zero, bias_h3,
bias0, bias2, bias4, bias6 );
ILVL_H4_SW( zero, bias_h0, zero, bias_h1, zero, bias_h2, zero, bias_h3,
bias1, bias3, bias5, bias7 );
LD_SH4( p_mf, 8, mf_h0, mf_h1, mf_h2, mf_h3 );
ILVR_H4_SW( zero, mf_h0, zero, mf_h1, zero, mf_h2, zero, mf_h3,
mf_vec0, mf_vec2, mf_vec4, mf_vec6 );
ILVL_H4_SW( zero, mf_h0, zero, mf_h1, zero, mf_h2, zero, mf_h3,
mf_vec1, mf_vec3, mf_vec5, mf_vec7 );
dct_w0 = __msa_add_a_w( dct_signed_w0, bias0 );
dct_w1 = __msa_add_a_w( dct_signed_w1, bias1 );
dct_w2 = __msa_add_a_w( dct_signed_w2, bias2 );
dct_w3 = __msa_add_a_w( dct_signed_w3, bias3 );
dct_w4 = __msa_add_a_w( dct_signed_w4, bias4 );
dct_w5 = __msa_add_a_w( dct_signed_w5, bias5 );
dct_w6 = __msa_add_a_w( dct_signed_w6, bias6 );
dct_w7 = __msa_add_a_w( dct_signed_w7, bias7 );
dct_w0 *= mf_vec0;
dct_w1 *= mf_vec1;
dct_w2 *= mf_vec2;
dct_w3 *= mf_vec3;
dct_w4 *= mf_vec4;
dct_w5 *= mf_vec5;
dct_w6 *= mf_vec6;
dct_w7 *= mf_vec7;
SRA_4V( dct_w0, dct_w1, dct_w2, dct_w3, 16 );
SRA_4V( dct_w4, dct_w5, dct_w6, dct_w7, 16 );
PCKEV_H4_SH( dct_w1, dct_w0, dct_w3, dct_w2, dct_w5, dct_w4, dct_w7, dct_w6,
dct_h0, dct_h1, dct_h2, dct_h3 );
SUB4( zero, dct_h0, zero, dct_h1, zero, dct_h2, zero, dct_h3,
dct0, dct1, dct2, dct3 );
dct0 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h0,
( v16u8 ) dct0, ( v16u8 ) dct0_mask );
dct1 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h1,
( v16u8 ) dct1, ( v16u8 ) dct1_mask );
dct2 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h2,
( v16u8 ) dct2, ( v16u8 ) dct2_mask );
dct3 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h3,
( v16u8 ) dct3, ( v16u8 ) dct3_mask );
non_zero = HADD_SW_S32( ( v4u32 )( dct_h0 + dct_h1 + dct_h2 + dct_h3 ) );
ST_SH4( dct0, dct1, dct2, dct3, p_dct, 8 );
LD_SH4( p_dct + 32, 8, dct0, dct1, dct2, dct3 );
dct0_mask = __msa_clei_s_h( dct0, 0 );
dct1_mask = __msa_clei_s_h( dct1, 0 );
dct2_mask = __msa_clei_s_h( dct2, 0 );
dct3_mask = __msa_clei_s_h( dct3, 0 );
UNPCK_SH_SW( dct0, dct_signed_w0, dct_signed_w1 );
UNPCK_SH_SW( dct1, dct_signed_w2, dct_signed_w3 );
UNPCK_SH_SW( dct2, dct_signed_w4, dct_signed_w5 );
UNPCK_SH_SW( dct3, dct_signed_w6, dct_signed_w7 );
LD_SH4( p_bias + 32, 8, bias_h0, bias_h1, bias_h2, bias_h3 );
ILVR_H4_SW( zero, bias_h0, zero, bias_h1, zero, bias_h2, zero, bias_h3,
bias0, bias2, bias4, bias6 );
ILVL_H4_SW( zero, bias_h0, zero, bias_h1, zero, bias_h2, zero, bias_h3,
bias1, bias3, bias5, bias7 );
LD_SH4( p_mf + 32, 8, mf_h0, mf_h1, mf_h2, mf_h3 );
ILVR_H4_SW( zero, mf_h0, zero, mf_h1, zero, mf_h2, zero, mf_h3,
mf_vec0, mf_vec2, mf_vec4, mf_vec6 );
ILVL_H4_SW( zero, mf_h0, zero, mf_h1, zero, mf_h2, zero, mf_h3,
mf_vec1, mf_vec3, mf_vec5, mf_vec7 );
dct_w0 = __msa_add_a_w( dct_signed_w0, bias0 );
dct_w1 = __msa_add_a_w( dct_signed_w1, bias1 );
dct_w2 = __msa_add_a_w( dct_signed_w2, bias2 );
dct_w3 = __msa_add_a_w( dct_signed_w3, bias3 );
dct_w4 = __msa_add_a_w( dct_signed_w4, bias4 );
dct_w5 = __msa_add_a_w( dct_signed_w5, bias5 );
dct_w6 = __msa_add_a_w( dct_signed_w6, bias6 );
dct_w7 = __msa_add_a_w( dct_signed_w7, bias7 );
dct_w0 *= mf_vec0;
dct_w1 *= mf_vec1;
dct_w2 *= mf_vec2;
dct_w3 *= mf_vec3;
dct_w4 *= mf_vec4;
dct_w5 *= mf_vec5;
dct_w6 *= mf_vec6;
dct_w7 *= mf_vec7;
SRA_4V( dct_w0, dct_w1, dct_w2, dct_w3, 16 );
SRA_4V( dct_w4, dct_w5, dct_w6, dct_w7, 16 );
PCKEV_H2_SH( dct_w1, dct_w0, dct_w3, dct_w2, dct_h0, dct_h1 );
PCKEV_H2_SH( dct_w5, dct_w4, dct_w7, dct_w6, dct_h2, dct_h3 );
SUB4( zero, dct_h0, zero, dct_h1, zero, dct_h2, zero, dct_h3,
dct0, dct1, dct2, dct3 );
dct0 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h0,
( v16u8 ) dct0, ( v16u8 ) dct0_mask );
dct1 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h1,
( v16u8 ) dct1, ( v16u8 ) dct1_mask );
dct2 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h2,
( v16u8 ) dct2, ( v16u8 ) dct2_mask );
dct3 = ( v8i16 ) __msa_bmnz_v( ( v16u8 ) dct_h3,
( v16u8 ) dct3, ( v16u8 ) dct3_mask );
non_zero += HADD_SW_S32( ( v4u32 ) ( dct_h0 + dct_h1 + dct_h2 + dct_h3 ) );
ST_SH4( dct0, dct1, dct2, dct3, p_dct + 32, 8 );
return !!non_zero;
}