void fdct16x8_1d_row(int16_t *input, int16_t *output) {
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;
LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
in10, in11, in12, in13, in14, in15);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
SRA_4V(in0, in1, in2, in3, 2);
SRA_4V(in4, in5, in6, in7, 2);
SRA_4V(in8, in9, in10, in11, 2);
SRA_4V(in12, in13, in14, in15, 2);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
tmp5, in5, tmp6, in6, tmp7, in7);
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
}
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);
}
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);
}
static WEBP_INLINE void ITransformOne(const uint8_t* ref, const int16_t* in,
uint8_t* dst) {
v8i16 input0, input1;
v4i32 in0, in1, in2, in3, hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3;
v4i32 res0, res1, res2, res3;
v16i8 dest0, dest1, dest2, dest3;
const v16i8 zero = { 0 };
LD_SH2(in, 8, input0, input1);
UNPCK_SH_SW(input0, in0, in1);
UNPCK_SH_SW(input1, in2, in3);
IDCT_1D_W(in0, in1, in2, in3, hz0, hz1, hz2, hz3);
TRANSPOSE4x4_SW_SW(hz0, hz1, hz2, hz3, hz0, hz1, hz2, hz3);
IDCT_1D_W(hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3);
SRARI_W4_SW(vt0, vt1, vt2, vt3, 3);
TRANSPOSE4x4_SW_SW(vt0, vt1, vt2, vt3, vt0, vt1, vt2, vt3);
LD_SB4(ref, BPS, dest0, dest1, dest2, dest3);
ILVR_B4_SW(zero, dest0, zero, dest1, zero, dest2, zero, dest3,
res0, res1, res2, res3);
ILVR_H4_SW(zero, res0, zero, res1, zero, res2, zero, res3,
res0, res1, res2, res3);
ADD4(res0, vt0, res1, vt1, res2, vt2, res3, vt3, res0, res1, res2, res3);
CLIP_SW4_0_255(res0, res1, res2, res3);
PCKEV_B2_SW(res0, res1, res2, res3, vt0, vt1);
res0 = (v4i32)__msa_pckev_b((v16i8)vt0, (v16i8)vt1);
ST4x4_UB(res0, res0, 3, 2, 1, 0, dst, BPS);
}
void vpx_fdct8x8_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v4i32 vec_w;
LD_SH8(input, stride, in0, in1, in2, in3, in4, in5, in6, in7);
ADD4(in0, in1, in2, in3, in4, in5, in6, in7, in0, in2, in4, in6);
ADD2(in0, in2, in4, in6, in0, in4);
vec_w = __msa_hadd_s_w(in0, in0);
vec_w += __msa_hadd_s_w(in4, in4);
out[0] = HADD_SW_S32(vec_w);
out[1] = 0;
}
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 WEBP_INLINE void TrueMotion16x16(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
if (left != NULL) {
if (top != NULL) {
int j;
v8i16 d1, d2;
const v16i8 zero = { 0 };
const v8i16 TL = (v8i16)__msa_fill_h(left[-1]);
const v16u8 T = LD_UB(top);
ILVRL_B2_SH(zero, T, d1, d2);
SUB2(d1, TL, d2, TL, d1, d2);
for (j = 0; j < 16; j += 4) {
v16i8 t0, t1, t2, t3;
v8i16 r0, r1, r2, r3, r4, r5, r6, r7;
const v8i16 L0 = (v8i16)__msa_fill_h(left[j + 0]);
const v8i16 L1 = (v8i16)__msa_fill_h(left[j + 1]);
const v8i16 L2 = (v8i16)__msa_fill_h(left[j + 2]);
const v8i16 L3 = (v8i16)__msa_fill_h(left[j + 3]);
ADD4(d1, L0, d1, L1, d1, L2, d1, L3, r0, r1, r2, r3);
ADD4(d2, L0, d2, L1, d2, L2, d2, L3, r4, r5, r6, r7);
CLIP_SH4_0_255(r0, r1, r2, r3);
CLIP_SH4_0_255(r4, r5, r6, r7);
PCKEV_B4_SB(r4, r0, r5, r1, r6, r2, r7, r3, t0, t1, t2, t3);
ST_SB4(t0, t1, t2, t3, dst, BPS);
dst += 4 * BPS;
}
} else {
HorizontalPred16x16(dst, left);
}
} else {
if (top != NULL) {
VerticalPred16x16(dst, top);
} else {
const v16u8 out = (v16u8)__msa_fill_b(0x81);
STORE16x16(out, dst);
}
}
}
static void hevc_addblk_8x8_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t *temp_dst = dst;
uint64_t dst0, dst1, dst2, dst3;
v2i64 dst_vec0 = { 0 };
v2i64 dst_vec1 = { 0 };
v8i16 dst_r0, dst_l0, dst_r1, dst_l1;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v16u8 zeros = { 0 };
LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
temp_dst += (4 * stride);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
ILVRL_B2_SH(zeros, dst_vec0, dst_r0, dst_l0);
ILVRL_B2_SH(zeros, dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in0, dst_l0, in1, dst_r1, in2, dst_l1, in3,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, dst, stride);
dst += (4 * stride);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
UNPCK_UB_SH(dst_vec0, dst_r0, dst_l0);
UNPCK_UB_SH(dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in4, dst_l0, in5, dst_r1, in6, dst_l1, in7,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, dst, stride);
}
static WEBP_INLINE void TM4(uint8_t* dst, const uint8_t* top) {
const v16i8 zero = { 0 };
const v8i16 TL = (v8i16)__msa_fill_h(top[-1]);
const v8i16 L0 = (v8i16)__msa_fill_h(top[-2]);
const v8i16 L1 = (v8i16)__msa_fill_h(top[-3]);
const v8i16 L2 = (v8i16)__msa_fill_h(top[-4]);
const v8i16 L3 = (v8i16)__msa_fill_h(top[-5]);
const v16u8 T1 = LD_UB(top);
const v8i16 T = (v8i16)__msa_ilvr_b(zero, (v16i8)T1);
const v8i16 d = T - TL;
v8i16 r0, r1, r2, r3;
ADD4(d, L0, d, L1, d, L2, d, L3, r0, r1, r2, r3);
CLIP_SH4_0_255(r0, r1, r2, r3);
PCKEV_ST4x4_UB(r0, r1, r2, r3, dst, BPS);
}
uint32_t vp10_avg_8x8_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
v8u16 sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7;
v4u32 sum = { 0 };
LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
HADD_UB4_UH(src0, src1, src2, src3, sum0, sum1, sum2, sum3);
HADD_UB4_UH(src4, src5, src6, src7, sum4, sum5, sum6, sum7);
ADD4(sum0, sum1, sum2, sum3, sum4, sum5, sum6, sum7, sum0, sum2, sum4, sum6);
ADD2(sum0, sum2, sum4, sum6, sum0, sum4);
sum0 += sum4;
sum = __msa_hadd_u_w(sum0, sum0);
sum0 = (v8u16)__msa_pckev_h((v8i16)sum, (v8i16)sum);
sum = __msa_hadd_u_w(sum0, sum0);
sum = (v4u32)__msa_srari_w((v4i32)sum, 6);
sum_out = __msa_copy_u_w((v4i32)sum, 0);
return sum_out;
}
void vpx_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
int sum, i;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v4i32 vec_w = { 0 };
for (i = 0; i < 4; ++i) {
LD_SH2(input, 8, in0, in1);
input += stride;
LD_SH2(input, 8, in2, in3);
input += stride;
LD_SH2(input, 8, in4, in5);
input += stride;
LD_SH2(input, 8, in6, in7);
input += stride;
ADD4(in0, in1, in2, in3, in4, in5, in6, in7, in0, in2, in4, in6);
ADD2(in0, in2, in4, in6, in0, in4);
vec_w += __msa_hadd_s_w(in0, in0);
vec_w += __msa_hadd_s_w(in4, in4);
}
sum = HADD_SW_S32(vec_w);
out[0] = (int16_t)(sum >> 1);
}
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 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 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);
}
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_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 avc_idct8_addblk_msa( uint8_t *p_dst, int16_t *p_src,
int32_t i_dst_stride )
{
v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
v8i16 vec0, vec1, vec2, vec3;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 res0, res1, res2, res3, res4, res5, res6, res7;
v4i32 tmp0_r, tmp1_r, tmp2_r, tmp3_r, tmp4_r, tmp5_r, tmp6_r, tmp7_r;
v4i32 tmp0_l, tmp1_l, tmp2_l, tmp3_l, tmp4_l, tmp5_l, tmp6_l, tmp7_l;
v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec0_l, vec1_l, vec2_l, vec3_l;
v4i32 res0_r, res1_r, res2_r, res3_r, res4_r, res5_r, res6_r, res7_r;
v4i32 res0_l, res1_l, res2_l, res3_l, res4_l, res5_l, res6_l, res7_l;
v16i8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
v16i8 zeros = { 0 };
p_src[ 0 ] += 32;
LD_SH8( p_src, 8, src0, src1, src2, src3, src4, src5, src6, src7 );
vec0 = src0 + src4;
vec1 = src0 - src4;
vec2 = src2 >> 1;
vec2 = vec2 - src6;
vec3 = src6 >> 1;
vec3 = src2 + vec3;
BUTTERFLY_4( vec0, vec1, vec2, vec3, tmp0, tmp1, tmp2, tmp3 );
vec0 = src7 >> 1;
vec0 = src5 - vec0 - src3 - src7;
vec1 = src3 >> 1;
vec1 = src1 - vec1 + src7 - src3;
vec2 = src5 >> 1;
vec2 = vec2 - src1 + src7 + src5;
vec3 = src1 >> 1;
vec3 = vec3 + src3 + src5 + src1;
tmp4 = vec3 >> 2;
tmp4 += vec0;
tmp5 = vec2 >> 2;
tmp5 += vec1;
tmp6 = vec1 >> 2;
tmp6 -= vec2;
tmp7 = vec0 >> 2;
tmp7 = vec3 - tmp7;
BUTTERFLY_8( tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
res0, res1, res2, res3, res4, res5, res6, res7 );
TRANSPOSE8x8_SH_SH( res0, res1, res2, res3, res4, res5, res6, res7,
res0, res1, res2, res3, res4, res5, res6, res7 );
UNPCK_SH_SW( res0, tmp0_r, tmp0_l );
UNPCK_SH_SW( res1, tmp1_r, tmp1_l );
UNPCK_SH_SW( res2, tmp2_r, tmp2_l );
UNPCK_SH_SW( res3, tmp3_r, tmp3_l );
UNPCK_SH_SW( res4, tmp4_r, tmp4_l );
UNPCK_SH_SW( res5, tmp5_r, tmp5_l );
UNPCK_SH_SW( res6, tmp6_r, tmp6_l );
UNPCK_SH_SW( res7, tmp7_r, tmp7_l );
BUTTERFLY_4( tmp0_r, tmp0_l, tmp4_l, tmp4_r,
vec0_r, vec0_l, vec1_l, vec1_r );
vec2_r = tmp2_r >> 1;
vec2_l = tmp2_l >> 1;
vec2_r -= tmp6_r;
vec2_l -= tmp6_l;
vec3_r = tmp6_r >> 1;
vec3_l = tmp6_l >> 1;
vec3_r += tmp2_r;
vec3_l += tmp2_l;
BUTTERFLY_4( vec0_r, vec1_r, vec2_r, vec3_r,
tmp0_r, tmp2_r, tmp4_r, tmp6_r );
BUTTERFLY_4( vec0_l, vec1_l, vec2_l, vec3_l,
tmp0_l, tmp2_l, tmp4_l, tmp6_l );
vec0_r = tmp7_r >> 1;
vec0_l = tmp7_l >> 1;
vec0_r = tmp5_r - vec0_r - tmp3_r - tmp7_r;
vec0_l = tmp5_l - vec0_l - tmp3_l - tmp7_l;
vec1_r = tmp3_r >> 1;
vec1_l = tmp3_l >> 1;
vec1_r = tmp1_r - vec1_r + tmp7_r - tmp3_r;
vec1_l = tmp1_l - vec1_l + tmp7_l - tmp3_l;
vec2_r = tmp5_r >> 1;
vec2_l = tmp5_l >> 1;
vec2_r = vec2_r - tmp1_r + tmp7_r + tmp5_r;
vec2_l = vec2_l - tmp1_l + tmp7_l + tmp5_l;
vec3_r = tmp1_r >> 1;
vec3_l = tmp1_l >> 1;
vec3_r = vec3_r + tmp3_r + tmp5_r + tmp1_r;
vec3_l = vec3_l + tmp3_l + tmp5_l + tmp1_l;
tmp1_r = vec3_r >> 2;
tmp1_l = vec3_l >> 2;
tmp1_r += vec0_r;
tmp1_l += vec0_l;
tmp3_r = vec2_r >> 2;
tmp3_l = vec2_l >> 2;
tmp3_r += vec1_r;
tmp3_l += vec1_l;
tmp5_r = vec1_r >> 2;
tmp5_l = vec1_l >> 2;
tmp5_r -= vec2_r;
tmp5_l -= vec2_l;
tmp7_r = vec0_r >> 2;
tmp7_l = vec0_l >> 2;
tmp7_r = vec3_r - tmp7_r;
tmp7_l = vec3_l - tmp7_l;
BUTTERFLY_4( tmp0_r, tmp0_l, tmp7_l, tmp7_r,
res0_r, res0_l, res7_l, res7_r );
BUTTERFLY_4( tmp2_r, tmp2_l, tmp5_l, tmp5_r,
res1_r, res1_l, res6_l, res6_r );
BUTTERFLY_4( tmp4_r, tmp4_l, tmp3_l, tmp3_r,
res2_r, res2_l, res5_l, res5_r );
BUTTERFLY_4( tmp6_r, tmp6_l, tmp1_l, tmp1_r,
res3_r, res3_l, res4_l, res4_r );
SRA_4V( res0_r, res0_l, res1_r, res1_l, 6 );
SRA_4V( res2_r, res2_l, res3_r, res3_l, 6 );
SRA_4V( res4_r, res4_l, res5_r, res5_l, 6 );
SRA_4V( res6_r, res6_l, res7_r, res7_l, 6 );
PCKEV_H4_SH( res0_l, res0_r, res1_l, res1_r, res2_l, res2_r, res3_l, res3_r,
res0, res1, res2, res3 );
PCKEV_H4_SH( res4_l, res4_r, res5_l, res5_r, res6_l, res6_r, res7_l, res7_r,
res4, res5, res6, res7 );
LD_SB8( p_dst, i_dst_stride,
dst0, dst1, dst2, dst3,
dst4, dst5, dst6, dst7 );
ILVR_B4_SH( zeros, dst0, zeros, dst1, zeros, dst2, zeros, dst3,
tmp0, tmp1, tmp2, tmp3 );
ILVR_B4_SH( zeros, dst4, zeros, dst5, zeros, dst6, zeros, dst7,
tmp4, tmp5, tmp6, tmp7 );
ADD4( res0, tmp0, res1, tmp1, res2, tmp2, res3, tmp3,
res0, res1, res2, res3 );
ADD4( res4, tmp4, res5, tmp5, res6, tmp6, res7, tmp7,
res4, res5, res6, res7 );
CLIP_SH4_0_255( res0, res1, res2, res3 );
CLIP_SH4_0_255( res4, res5, res6, res7 );
PCKEV_B4_SB( res1, res0, res3, res2, res5, res4, res7, res6,
dst0, dst1, dst2, dst3 );
ST8x4_UB( dst0, dst1, p_dst, i_dst_stride );
p_dst += ( 4 * i_dst_stride );
ST8x4_UB( dst2, dst3, p_dst, i_dst_stride );
}
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;
}
}
