void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
int32_t src_stride) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
-cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
v8i16 coeff2 = {
-cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64, 0, 0, 0, 0
};
LD_SH16(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7, in8, in9,
in10, in11, in12, in13, in14, in15);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in8, in9, in10, in11, 2);
SLLI_4V(in12, in13, in14, in15, 2);
ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
tmp_ptr += 16;
/* stp 1 */
ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
cnst4 = __msa_splati_h(coeff, 0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
cnst5 = __msa_splati_h(coeff, 1);
cnst5 = __msa_ilvev_h(cnst5, cnst4);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
stp24 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
stp23 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
/* stp2 */
BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
stp26 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
cnst0 = __msa_splati_h(coeff, 4);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp21 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
ILVRL_H2_SH(in15, in8, vec1, vec0);
SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr);
cnst0 = __msa_splati_h(coeff2, 0);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 224);
ILVRL_H2_SH(in14, in9, vec1, vec0);
SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 128);
cnst1 = __msa_splati_h(coeff2, 2);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 96);
SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
cnst1 = __msa_splati_h(coeff, 3);
cnst1 = __msa_ilvev_h(cnst0, cnst1);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
/* stp4 */
ADD2(stp34, stp25, stp33, stp22, in13, in10);
ILVRL_H2_SH(in13, in10, vec1, vec0);
SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 64);
cnst0 = __msa_splati_h(coeff2, 1);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 160);
SUB2(stp34, stp25, stp33, stp22, in12, in11);
ILVRL_H2_SH(in12, in11, vec1, vec0);
SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 192);
cnst1 = __msa_splati_h(coeff2, 3);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 32);
}
static void fdct8x32_1d_row_odd_rd(int16_t *temp, int16_t *interm_ptr,
int16_t *out) {
v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
v8i16 in24, in25, in26, in27, in28, in29, in30, in31;
v8i16 vec4, vec5;
in20 = LD_SH(temp + 32);
in21 = LD_SH(temp + 40);
in26 = LD_SH(temp + 80);
in27 = LD_SH(temp + 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);
FDCT_POSTPROC_2V_NEG_H(in20, in21);
FDCT_POSTPROC_2V_NEG_H(in26, in27);
in18 = LD_SH(temp + 16);
in19 = LD_SH(temp + 24);
in28 = LD_SH(temp + 96);
in29 = LD_SH(temp + 104);
FDCT_POSTPROC_2V_NEG_H(in18, in19);
FDCT_POSTPROC_2V_NEG_H(in28, in29);
vec4 = in19 - in20;
ST_SH(vec4, interm_ptr + 32);
vec4 = in18 - in21;
ST_SH(vec4, interm_ptr + 88);
vec4 = in29 - in26;
ST_SH(vec4, interm_ptr + 64);
vec4 = in28 - in27;
ST_SH(vec4, interm_ptr + 56);
ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
in22 = LD_SH(temp + 48);
in23 = LD_SH(temp + 56);
in24 = LD_SH(temp + 64);
in25 = LD_SH(temp + 72);
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);
FDCT_POSTPROC_2V_NEG_H(in22, in23);
FDCT_POSTPROC_2V_NEG_H(in24, in25);
in16 = LD_SH(temp);
in17 = LD_SH(temp + 8);
in30 = LD_SH(temp + 112);
in31 = LD_SH(temp + 120);
FDCT_POSTPROC_2V_NEG_H(in16, in17);
FDCT_POSTPROC_2V_NEG_H(in30, in31);
vec4 = in17 - in22;
ST_SH(vec4, interm_ptr + 40);
vec4 = in30 - in25;
ST_SH(vec4, interm_ptr + 48);
vec4 = in31 - in24;
ST_SH(vec4, interm_ptr + 72);
vec4 = in16 - in23;
ST_SH(vec4, interm_ptr + 80);
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);
ST_SH(vec5, out);
ST_SH(vec4, out + 120);
SUB2(in27, in26, in25, in24, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
ST_SH(vec5, out + 112);
ST_SH(vec4, out + 8);
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);
ST_SH(vec4, out + 16);
ST_SH(vec5, out + 104);
ADD2(in26, in27, in24, in25, in22, in21);
DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
ST_SH(vec4, out + 24);
ST_SH(vec5, out + 96);
in20 = LD_SH(interm_ptr + 32);
in21 = LD_SH(interm_ptr + 88);
in27 = LD_SH(interm_ptr + 56);
in26 = LD_SH(interm_ptr + 64);
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);
in22 = LD_SH(interm_ptr + 40);
in25 = LD_SH(interm_ptr + 48);
in24 = LD_SH(interm_ptr + 72);
in23 = LD_SH(interm_ptr + 80);
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);
in16 = in28 + in29;
in19 = in31 + in30;
DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
ST_SH(vec5, out + 32);
ST_SH(vec4, out + 88);
SUB2(in28, in29, in31, in30, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
ST_SH(vec5, out + 40);
ST_SH(vec4, out + 80);
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);
ST_SH(vec5, out + 72);
ST_SH(vec4, out + 48);
ADD2(in29, in28, in30, in31, in17, in18);
DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
ST_SH(vec4, out + 56);
ST_SH(vec5, out + 64);
}
static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0,
int16_t *out) {
int16_t *out_ptr = out + 128;
v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
v4i32 k0, k1, k2, k3;
LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15);
LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7);
LD_SW2(const0 + 4 * 19, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 21);
MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
tp0 = LD_SH(int_buf + 4 * 8);
tp1 = LD_SH(int_buf + 5 * 8);
tp3 = LD_SH(int_buf + 10 * 8);
tp2 = LD_SH(int_buf + 14 * 8);
LD_SW2(const0 + 4 * 22, 4, k0, k1);
k2 = LD_SW(const0 + 4 * 24);
MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
out4 = -out4;
ST_SH(out4, (out + 3 * 16));
ST_SH(out5, (out_ptr + 4 * 16));
h1 = LD_SH(int_buf + 9 * 8);
h3 = LD_SH(int_buf + 12 * 8);
MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
out13 = -out13;
ST_SH(out12, (out + 2 * 16));
ST_SH(out13, (out_ptr + 5 * 16));
tp0 = LD_SH(int_buf);
tp1 = LD_SH(int_buf + 8);
tp2 = LD_SH(int_buf + 2 * 8);
tp3 = LD_SH(int_buf + 6 * 8);
BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
out1 = -out1;
ST_SH(out0, (out));
ST_SH(out1, (out_ptr + 7 * 16));
h0 = LD_SH(int_buf + 8 * 8);
h2 = LD_SH(int_buf + 13 * 8);
BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
out8 = -out8;
ST_SH(out8, (out + 16));
ST_SH(out9, (out_ptr + 6 * 16));
/* stage 4 */
LD_SW2(const0 + 4 * 25, 4, k0, k1);
LD_SW2(const0 + 4 * 27, 4, k2, k3);
MADD_SHORT(h10, h11, k1, k2, out2, out3);
ST_SH(out2, (out + 7 * 16));
ST_SH(out3, (out_ptr));
MADD_SHORT(out6, out7, k0, k3, out6, out7);
ST_SH(out6, (out + 4 * 16));
ST_SH(out7, (out_ptr + 3 * 16));
MADD_SHORT(out10, out11, k0, k3, out10, out11);
ST_SH(out10, (out + 6 * 16));
ST_SH(out11, (out_ptr + 16));
MADD_SHORT(out14, out15, k1, k2, out14, out15);
ST_SH(out14, (out + 5 * 16));
ST_SH(out15, (out_ptr + 2 * 16));
}
static void fdct8x32_1d_row_even_rd(int16_t *temp, int16_t *out) {
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, temp0, temp1;
/* fdct32 even */
/* stage 2 */
LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
in8, in9, in10, in11, in12, in13, in14, in15);
FDCT_POSTPROC_2V_NEG_H(vec0, vec1);
FDCT_POSTPROC_2V_NEG_H(vec2, vec3);
FDCT_POSTPROC_2V_NEG_H(vec4, vec5);
FDCT_POSTPROC_2V_NEG_H(vec6, vec7);
FDCT_POSTPROC_2V_NEG_H(in8, in9);
FDCT_POSTPROC_2V_NEG_H(in10, in11);
FDCT_POSTPROC_2V_NEG_H(in12, in13);
FDCT_POSTPROC_2V_NEG_H(in14, in15);
/* Stage 3 */
ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
temp0 = in0 + in3;
in0 = in0 - in3;
in3 = in1 + in2;
in1 = in1 - in2;
DOTP_CONST_PAIR(temp0, in3, cospi_16_64, cospi_16_64, temp1, temp0);
ST_SH(temp0, out);
ST_SH(temp1, out + 8);
DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
ST_SH(temp0, out + 16);
ST_SH(temp1, out + 24);
SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
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);
ST_SH(temp0, out + 32);
ST_SH(temp1, out + 56);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
ST_SH(temp0, out + 40);
ST_SH(temp1, out + 48);
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);
ST_SH(temp0, out + 64);
ST_SH(temp1, out + 120);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
ST_SH(temp0, out + 72);
ST_SH(temp1, out + 112);
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);
ST_SH(temp0, out + 80);
ST_SH(temp1, out + 104);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
ST_SH(temp0, out + 96);
ST_SH(temp1, out + 88);
}
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_row_even_4x(int16_t *input, int16_t *interm_ptr,
int16_t *out) {
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;
v4i32 vec0_l, vec1_l, vec2_l, vec3_l, vec4_l, vec5_l, vec6_l, vec7_l;
v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec4_r, vec5_r, vec6_r, vec7_r;
v4i32 tmp0_w, tmp1_w, tmp2_w, tmp3_w;
/* fdct32 even */
/* stage 2 */
LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8(input + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, interm_ptr, 8);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, interm_ptr + 64, 8);
/* Stage 3 */
UNPCK_SH_SW(vec0, vec0_l, vec0_r);
UNPCK_SH_SW(vec1, vec1_l, vec1_r);
UNPCK_SH_SW(vec2, vec2_l, vec2_r);
UNPCK_SH_SW(vec3, vec3_l, vec3_r);
UNPCK_SH_SW(vec4, vec4_l, vec4_r);
UNPCK_SH_SW(vec5, vec5_l, vec5_r);
UNPCK_SH_SW(vec6, vec6_l, vec6_r);
UNPCK_SH_SW(vec7, vec7_l, vec7_r);
ADD4(vec0_r, vec7_r, vec1_r, vec6_r, vec2_r, vec5_r, vec3_r, vec4_r,
tmp0_w, tmp1_w, tmp2_w, tmp3_w);
BUTTERFLY_4(tmp0_w, tmp1_w, tmp2_w, tmp3_w, vec4_r, vec6_r, vec7_r, vec5_r);
ADD4(vec0_l, vec7_l, vec1_l, vec6_l, vec2_l, vec5_l, vec3_l, vec4_l,
vec0_r, vec1_r, vec2_r, vec3_r);
tmp3_w = vec0_r + vec3_r;
vec0_r = vec0_r - vec3_r;
vec3_r = vec1_r + vec2_r;
vec1_r = vec1_r - vec2_r;
DOTP_CONST_PAIR_W(vec4_r, vec6_r, tmp3_w, vec3_r, cospi_16_64,
cospi_16_64, vec4_r, tmp3_w, vec6_r, vec3_r);
FDCT32_POSTPROC_NEG_W(vec4_r);
FDCT32_POSTPROC_NEG_W(tmp3_w);
FDCT32_POSTPROC_NEG_W(vec6_r);
FDCT32_POSTPROC_NEG_W(vec3_r);
PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
ST_SH2(vec5, vec4, out, 8);
DOTP_CONST_PAIR_W(vec5_r, vec7_r, vec0_r, vec1_r, cospi_24_64,
cospi_8_64, vec4_r, tmp3_w, vec6_r, vec3_r);
FDCT32_POSTPROC_NEG_W(vec4_r);
FDCT32_POSTPROC_NEG_W(tmp3_w);
FDCT32_POSTPROC_NEG_W(vec6_r);
FDCT32_POSTPROC_NEG_W(vec3_r);
PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
ST_SH2(vec5, vec4, out + 16, 8);
LD_SH8(interm_ptr, 8, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 32);
ST_SH(in5, out + 56);
SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 40);
ST_SH(in5, out + 48);
LD_SH8(interm_ptr + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 64);
ST_SH(in5, out + 120);
SUB2(in0, in1, in2, in3, in0, in2);
DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 72);
ST_SH(in5, out + 112);
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, in5, in4);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 80);
ST_SH(in5, out + 104);
ADD2(in3, in2, in0, in1, vec3, vec4);
DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, in4, in5);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
ST_SH(in4, out + 96);
ST_SH(in5, out + 88);
}
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_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;
}
}