static void FTransform(const uint8_t* src, const uint8_t* ref, int16_t* out) {
uint64_t out0, out1, out2, out3;
uint32_t in0, in1, in2, in3;
v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
v8i16 t0, t1, t2, t3;
v16u8 srcl0, srcl1, src0, src1;
const v8i16 mask0 = { 0, 4, 8, 12, 1, 5, 9, 13 };
const v8i16 mask1 = { 3, 7, 11, 15, 2, 6, 10, 14 };
const v8i16 mask2 = { 4, 0, 5, 1, 6, 2, 7, 3 };
const v8i16 mask3 = { 0, 4, 1, 5, 2, 6, 3, 7 };
const v8i16 cnst0 = { 2217, -5352, 2217, -5352, 2217, -5352, 2217, -5352 };
const v8i16 cnst1 = { 5352, 2217, 5352, 2217, 5352, 2217, 5352, 2217 };
LW4(src, BPS, in0, in1, in2, in3);
INSERT_W4_UB(in0, in1, in2, in3, src0);
LW4(ref, BPS, in0, in1, in2, in3);
INSERT_W4_UB(in0, in1, in2, in3, src1);
ILVRL_B2_UB(src0, src1, srcl0, srcl1);
HSUB_UB2_SH(srcl0, srcl1, t0, t1);
VSHF_H2_SH(t0, t1, t0, t1, mask0, mask1, t2, t3);
ADDSUB2(t2, t3, t0, t1);
t0 = SRLI_H(t0, 3);
VSHF_H2_SH(t0, t0, t1, t1, mask2, mask3, t3, t2);
tmp0 = __msa_hadd_s_w(t3, t3);
tmp2 = __msa_hsub_s_w(t3, t3);
FILL_W2_SW(1812, 937, tmp1, tmp3);
DPADD_SH2_SW(t2, t2, cnst0, cnst1, tmp3, tmp1);
SRAI_W2_SW(tmp1, tmp3, 9);
PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, t0, t1);
VSHF_H2_SH(t0, t1, t0, t1, mask0, mask1, t2, t3);
ADDSUB2(t2, t3, t0, t1);
VSHF_H2_SH(t0, t0, t1, t1, mask2, mask3, t3, t2);
tmp0 = __msa_hadd_s_w(t3, t3);
tmp2 = __msa_hsub_s_w(t3, t3);
ADDVI_W2_SW(tmp0, 7, tmp2, 7, tmp0, tmp2);
SRAI_W2_SW(tmp0, tmp2, 4);
FILL_W2_SW(12000, 51000, tmp1, tmp3);
DPADD_SH2_SW(t2, t2, cnst0, cnst1, tmp3, tmp1);
SRAI_W2_SW(tmp1, tmp3, 16);
UNPCK_R_SH_SW(t1, tmp4);
tmp5 = __msa_ceqi_w(tmp4, 0);
tmp4 = (v4i32)__msa_nor_v((v16u8)tmp5, (v16u8)tmp5);
tmp5 = __msa_fill_w(1);
tmp5 = (v4i32)__msa_and_v((v16u8)tmp5, (v16u8)tmp4);
tmp1 += tmp5;
PCKEV_H2_SH(tmp1, tmp0, tmp3, tmp2, t0, t1);
out0 = __msa_copy_s_d((v2i64)t0, 0);
out1 = __msa_copy_s_d((v2i64)t0, 1);
out2 = __msa_copy_s_d((v2i64)t1, 0);
out3 = __msa_copy_s_d((v2i64)t1, 1);
SD4(out0, out1, out2, out3, out, 8);
}
static void avc_sub4x4_dct_msa( uint8_t *p_src, int32_t i_src_stride,
uint8_t *p_ref, int32_t i_dst_stride,
int16_t *p_dst )
{
uint32_t i_src0, i_src1, i_src2, i_src3;
uint32_t i_ref0, i_ref1, i_ref2, i_ref3;
v16i8 src = { 0 };
v16i8 ref = { 0 };
v16u8 inp0, inp1;
v8i16 diff0, diff1, diff2, diff3;
v8i16 temp0, temp1, temp2, temp3;
LW4( p_src, i_src_stride, i_src0, i_src1, i_src2, i_src3 );
LW4( p_ref, i_dst_stride, i_ref0, i_ref1, i_ref2, i_ref3 );
INSERT_W4_SB( i_src0, i_src1, i_src2, i_src3, src );
INSERT_W4_SB( i_ref0, i_ref1, i_ref2, i_ref3, ref );
ILVRL_B2_UB( src, ref, inp0, inp1 );
HSUB_UB2_SH( inp0, inp1, diff0, diff2 );
diff1 = ( v8i16 ) __msa_ilvl_d( ( v2i64 ) diff0, ( v2i64 ) diff0 );
diff3 = ( v8i16 ) __msa_ilvl_d( ( v2i64 ) diff2, ( v2i64 ) diff2 );
BUTTERFLY_4( diff0, diff1, diff2, diff3, temp0, temp1, temp2, temp3 );
diff0 = temp0 + temp1;
diff1 = ( temp3 << 1 ) + temp2;
diff2 = temp0 - temp1;
diff3 = temp3 - ( temp2 << 1 );
TRANSPOSE4x4_SH_SH( diff0, diff1, diff2, diff3,
temp0, temp1, temp2, temp3 );
BUTTERFLY_4( temp0, temp1, temp2, temp3, diff0, diff1, diff2, diff3 );
temp0 = diff0 + diff1;
temp1 = ( diff3 << 1 ) + diff2;
temp2 = diff0 - diff1;
temp3 = diff3 - ( diff2 << 1 );
ILVR_D2_UB( temp1, temp0, temp3, temp2, inp0, inp1 );
ST_UB2( inp0, inp1, p_dst, 8 );
}
static int32_t subtract_sum4x4_msa( uint8_t *p_src, int32_t i_src_stride,
uint8_t *pred_ptr, int32_t i_pred_stride )
{
int16_t i_sum;
uint32_t i_src0, i_src1, i_src2, i_src3;
uint32_t i_pred0, i_pred1, i_pred2, i_pred3;
v16i8 src = { 0 };
v16i8 pred = { 0 };
v16u8 src_l0, src_l1;
v8i16 diff0, diff1;
LW4( p_src, i_src_stride, i_src0, i_src1, i_src2, i_src3 );
LW4( pred_ptr, i_pred_stride, i_pred0, i_pred1, i_pred2, i_pred3 );
INSERT_W4_SB( i_src0, i_src1, i_src2, i_src3, src );
INSERT_W4_SB( i_pred0, i_pred1, i_pred2, i_pred3, pred );
ILVRL_B2_UB( src, pred, src_l0, src_l1 );
HSUB_UB2_SH( src_l0, src_l1, diff0, diff1 );
i_sum = HADD_UH_U32( diff0 + diff1 );
return i_sum;
}
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;
}
}
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;
}
}
