static void temporal_filter_apply_16size_msa(uint8_t *frame1_ptr,
uint32_t stride,
uint8_t *frame2_ptr,
int32_t strength_in,
int32_t filter_wt_in,
uint32_t *acc, uint16_t *cnt)
{
uint32_t row;
v16i8 frame1_0_b, frame1_1_b, frame2_0_b, frame2_1_b;
v16u8 frame_l, frame_h;
v16i8 zero = { 0 };
v8i16 frame2_0_h, frame2_1_h, mod0_h, mod1_h;
v8i16 diff0, diff1, cnt0, cnt1;
v4i32 const3, const16, filter_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frame2_0, frame2_1, frame2_2, frame2_3;
v4i32 acc0, acc1, acc2, acc3;
filter_wt = __msa_fill_w(filter_wt_in);
strength = __msa_fill_w(strength_in);
const3 = __msa_ldi_w(3);
const16 = __msa_ldi_w(16);
for (row = 8; row--;)
{
frame1_0_b = LD_SB(frame1_ptr);
frame2_0_b = LD_SB(frame2_ptr);
frame1_ptr += stride;
frame2_ptr += 16;
frame1_1_b = LD_SB(frame1_ptr);
frame2_1_b = LD_SB(frame2_ptr);
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame1_0_b, frame2_0_b, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h)
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
ILVRL_B2_SH(zero, frame2_0_b, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame1_1_b, frame2_1_b, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame2_1_b, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
frame1_ptr += stride;
frame2_ptr += 16;
}
}
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 temporal_filter_apply_8size_msa(uint8_t *frame1_ptr,
uint32_t stride,
uint8_t *frame2_ptr,
int32_t strength_in,
int32_t filter_wt_in,
uint32_t *acc, uint16_t *cnt)
{
uint32_t row;
uint64_t f0, f1, f2, f3, f4, f5, f6, f7;
v16i8 frame1 = { 0 };
v16i8 frame2 = { 0 };
v16i8 frame3 = { 0 };
v16i8 frame4 = { 0 };
v16u8 frame_l, frame_h;
v8i16 frame2_0_h, frame2_1_h, mod0_h, mod1_h;
v8i16 diff0, diff1, cnt0, cnt1;
v4i32 const3, const16;
v4i32 filter_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frame2_0, frame2_1, frame2_2, frame2_3;
v4i32 acc0, acc1, acc2, acc3;
filter_wt = __msa_fill_w(filter_wt_in);
strength = __msa_fill_w(strength_in);
const3 = __msa_ldi_w(3);
const16 = __msa_ldi_w(16);
for (row = 2; row--;)
{
LD2(frame1_ptr, stride, f0, f1);
frame1_ptr += (2 * stride);
LD2(frame2_ptr, 8, f2, f3);
frame2_ptr += 16;
LD2(frame1_ptr, stride, f4, f5);
frame1_ptr += (2 * stride);
LD2(frame2_ptr, 8, f6, f7);
frame2_ptr += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
INSERT_D2_SB(f0, f1, frame1);
INSERT_D2_SB(f2, f3, frame2);
INSERT_D2_SB(f4, f5, frame3);
INSERT_D2_SB(f6, f7, frame4);
ILVRL_B2_UB(frame1, frame2, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame2, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame3, frame4, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame4, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 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 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;
}
