void ff_vp8_h_loop_filter16_inner_msa(uint8_t *src, ptrdiff_t pitch,
int32_t e, int32_t i, int32_t h)
{
v16u8 mask, hev, flat;
v16u8 thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src - 4 + (8 * pitch), pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8) __msa_fill_b(h);
b_limit = (v16u8) __msa_fill_b(e);
limit = (v16u8) __msa_fill_b(i);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
src -= 2;
ST4x8_UB(tmp2, tmp3, src, pitch);
src += (8 * pitch);
ST4x8_UB(tmp4, tmp5, src, pitch);
}
void vp8_short_fdct8x4_msa(int16_t *input, int16_t *output, int32_t pitch) {
v8i16 in0, in1, in2, in3;
v8i16 temp0, temp1, tmp0, tmp1;
v8i16 const0, const1, const2;
v8i16 coeff = { 2217, 5352, -5352, 14500, 7500, 12000, 25000, 26000 };
v8i16 zero = { 0 };
v4i32 vec0_w, vec1_w, vec2_w, vec3_w;
LD_SH4(input, pitch / 2, in0, in1, in2, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
SLLI_4V(temp0, temp1, in1, in3, 3);
in0 = temp0 + temp1;
in2 = temp0 - temp1;
SET_DOTP_VALUES(coeff, 0, 1, 2, const1, const2);
temp0 = __msa_splati_h(coeff, 3);
vec1_w = (v4i32)__msa_ilvev_h(zero, temp0);
coeff = __msa_ilvl_h(zero, coeff);
vec3_w = __msa_splati_w((v4i32)coeff, 0);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 12);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
TRANSPOSE4x4_H(in0, in1, in2, in3, in0, in1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, temp0, temp1, in1, in3);
in0 = temp0 + temp1 + 7;
in2 = temp0 - temp1 + 7;
in0 >>= 4;
in2 >>= 4;
SPLATI_W2_SW(coeff, 2, vec3_w, vec1_w);
vec3_w += vec1_w;
vec1_w = __msa_splati_w((v4i32)coeff, 1);
const0 = RET_1_IF_NZERO_H(in3);
ILVRL_H2_SH(in3, in1, tmp1, tmp0);
vec0_w = vec1_w;
vec2_w = vec3_w;
DPADD_SH4_SW(tmp1, tmp0, tmp1, tmp0, const1, const1, const2, const2, vec0_w,
vec1_w, vec2_w, vec3_w);
SRA_4V(vec1_w, vec0_w, vec3_w, vec2_w, 16);
PCKEV_H2_SH(vec1_w, vec0_w, vec3_w, vec2_w, in1, in3);
in1 += const0;
PCKEV_D2_SH(in1, in0, in3, in2, temp0, temp1);
ST_SH2(temp0, temp1, output, 8);
PCKOD_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output + 16, 8);
}
void vp9_lpf_vertical_4_dual_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit0_ptr,
const uint8_t *limit0_ptr,
const uint8_t *thresh0_ptr,
const uint8_t *b_limit1_ptr,
const uint8_t *limit1_ptr,
const uint8_t *thresh1_ptr) {
v16u8 mask, hev, flat;
v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src - 4 + (8 * pitch), pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0,
hev, mask, flat);
VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
src -= 2;
ST4x8_UB(tmp2, tmp3, src, pitch);
src += (8 * pitch);
ST4x8_UB(tmp4, tmp5, src, pitch);
}
void vp9_lpf_vertical_4_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
const uint8_t *limit_ptr,
const uint8_t *thresh_ptr,
int32_t count) {
v16u8 mask, hev, flat, limit, thresh, b_limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v8i16 vec0, vec1, vec2, vec3;
(void)count;
LD_UB8((src - 4), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh_ptr);
b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
limit = (v16u8)__msa_fill_b(*limit_ptr);
TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3,
p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
ILVR_B2_SH(p0, p1, q1, q0, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec2, vec3);
src -= 2;
ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
src += 4 * pitch;
ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
}
int32_t vp8_block_error_msa(int16_t *coeff_ptr, int16_t *dq_coeff_ptr)
{
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, dq_coeff, coeff0, coeff1;
v4i32 diff0, diff1;
v2i64 err0 = { 0 };
v2i64 err1 = { 0 };
for (loop_cnt = 2; loop_cnt--;)
{
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
coeff_ptr += 8;
dq_coeff_ptr += 8;
}
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err = __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
return err;
}
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);
}
void vpx_lpf_vertical_8_dual_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit0,
const uint8_t *limit0,
const uint8_t *thresh0,
const uint8_t *b_limit1,
const uint8_t *limit1,
const uint8_t *thresh1) {
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p1_out, p0_out, q0_out, q1_out;
v16u8 flat, mask, hev, thresh, b_limit, limit;
v16u8 row4, row5, row6, row7, row12, row13, row14, row15;
v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
v16u8 zero = { 0 };
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
temp_src = src - 4;
LD_UB8(temp_src, pitch, p0, p1, p2, p3, row4, row5, row6, row7);
temp_src += (8 * pitch);
LD_UB8(temp_src, pitch, q3, q2, q1, q0, row12, row13, row14, row15);
/* transpose 16x8 matrix into 8x16 */
TRANSPOSE16x8_UB_UB(p0, p1, p2, p3, row4, row5, row6, row7,
q3, q2, q1, q0, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh0);
vec0 = (v8i16)__msa_fill_b(*thresh1);
thresh = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)thresh);
b_limit = (v16u8)__msa_fill_b(*b_limit0);
vec0 = (v8i16)__msa_fill_b(*b_limit1);
b_limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)b_limit);
limit = (v16u8)__msa_fill_b(*limit0);
vec0 = (v8i16)__msa_fill_b(*limit1);
limit = (v16u8)__msa_ilvr_d((v2i64)vec0, (v2i64)limit);
/* mask and hev */
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
/* flat4 */
VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
/* filter4 */
VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
if (__msa_test_bz_v(flat)) {
ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec2, vec3);
ILVL_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec4, vec5);
src -= 2;
ST4x8_UB(vec2, vec3, src, pitch);
src += 8 * pitch;
ST4x8_UB(vec4, vec5, src, pitch);
} else {
ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
q3_r);
VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
/* filter8 */
VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
/* convert 16 bit output data into 8 bit */
PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
p0_filt8_r, q0_filt8_r);
PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
q2_filt8_r);
/* store pixel values */
p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec3, vec4);
ILVL_B2_SH(p1, p2, q0, p0, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec6, vec7);
ILVRL_B2_SH(q2, q1, vec2, vec5);
src -= 3;
ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec2, 0, src + 4, pitch);
src += (4 * pitch);
ST4x4_UB(vec4, vec4, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec2, 4, src + 4, pitch);
src += (4 * pitch);
ST4x4_UB(vec6, vec6, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec5, 0, src + 4, pitch);
src += (4 * pitch);
ST4x4_UB(vec7, vec7, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec5, 4, src + 4, pitch);
}
}
void vpx_lpf_vertical_8_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
const uint8_t *limit_ptr,
const uint8_t *thresh_ptr) {
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p1_out, p0_out, q0_out, q1_out;
v16u8 flat, mask, hev, thresh, b_limit, limit;
v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
v16u8 zero = { 0 };
v8i16 vec0, vec1, vec2, vec3, vec4;
/* load vector elements */
LD_UB8(src - 4, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3,
p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh_ptr);
b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
limit = (v16u8)__msa_fill_b(*limit_ptr);
/* mask and hev */
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
/* flat4 */
VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
/* filter4 */
VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
flat = (v16u8)__msa_ilvr_d((v2i64)zero, (v2i64)flat);
if (__msa_test_bz_v(flat)) {
/* Store 4 pixels p1-_q1 */
ILVR_B2_SH(p0_out, p1_out, q1_out, q0_out, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec2, vec3);
src -= 2;
ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
src += 4 * pitch;
ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
} else {
ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r,
q3_r);
VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
/* convert 16 bit output data into 8 bit */
PCKEV_B4_SH(p2_filt8_r, p2_filt8_r, p1_filt8_r, p1_filt8_r, p0_filt8_r,
p0_filt8_r, q0_filt8_r, q0_filt8_r, p2_filt8_r, p1_filt8_r,
p0_filt8_r, q0_filt8_r);
PCKEV_B2_SH(q1_filt8_r, q1_filt8_r, q2_filt8_r, q2_filt8_r, q1_filt8_r,
q2_filt8_r);
/* store pixel values */
p2 = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
p1 = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
p0 = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
q0 = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
q1 = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
q2 = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
/* Store 6 pixels p2-_q2 */
ILVR_B2_SH(p1, p2, q0, p0, vec0, vec1);
ILVRL_H2_SH(vec1, vec0, vec2, vec3);
vec4 = (v8i16)__msa_ilvr_b((v16i8)q2, (v16i8)q1);
src -= 3;
ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec4, 0, src + 4, pitch);
src += (4 * pitch);
ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
ST2x4_UB(vec4, 4, src + 4, pitch);
}
}
int32_t vp8_mbblock_error_msa(MACROBLOCK *mb, int32_t dc)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1;
v16u8 zero = { 0 };
v16u8 mask0 = (v16u8)__msa_ldi_b(255);
if (1 == dc)
{
mask0 = (v16u8)__msa_insve_w((v4i32)mask0, 0, (v4i32)zero);
}
for (loop_cnt = 0; loop_cnt < 8; loop_cnt++)
{
be = &mb->block[2 * loop_cnt];
bd = &mb->e_mbd.block[2 * loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[2 * loop_cnt + 1];
bd = &mb->e_mbd.block[2 * loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
diff0 = (v4i32)__msa_bmnz_v(zero, (v16u8)diff0, mask0);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err0 += __msa_splati_d(err0, 1);
err1 += __msa_splati_d(err1, 1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
int32_t vp8_mbuverror_msa(MACROBLOCK *mb)
{
BLOCK *be;
BLOCKD *bd;
int16_t *coeff_ptr, *dq_coeff_ptr;
int32_t err = 0;
uint32_t loop_cnt;
v8i16 coeff, coeff0, coeff1, coeff2, coeff3, coeff4;
v8i16 dq_coeff, dq_coeff2, dq_coeff3, dq_coeff4;
v4i32 diff0, diff1;
v2i64 err0, err1, err_dup0, err_dup1;
for (loop_cnt = 16; loop_cnt < 24; loop_cnt += 2)
{
be = &mb->block[loop_cnt];
bd = &mb->e_mbd.block[loop_cnt];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff = LD_SH(coeff_ptr);
dq_coeff = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff2 = LD_SH(coeff_ptr);
dq_coeff2 = LD_SH(dq_coeff_ptr);
be = &mb->block[loop_cnt + 1];
bd = &mb->e_mbd.block[loop_cnt + 1];
coeff_ptr = be->coeff;
dq_coeff_ptr = bd->dqcoeff;
coeff3 = LD_SH(coeff_ptr);
dq_coeff3 = LD_SH(dq_coeff_ptr);
coeff_ptr += 8;
dq_coeff_ptr += 8;
coeff4 = LD_SH(coeff_ptr);
dq_coeff4 = LD_SH(dq_coeff_ptr);
ILVRL_H2_SH(coeff, dq_coeff, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff2, dq_coeff2, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
ILVRL_H2_SH(coeff3, dq_coeff3, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DOTP_SW2_SD(diff0, diff1, diff0, diff1, err0, err1);
ILVRL_H2_SH(coeff4, dq_coeff4, coeff0, coeff1);
HSUB_UH2_SW(coeff0, coeff1, diff0, diff1);
DPADD_SD2_SD(diff0, diff1, err0, err1);
err_dup0 = __msa_splati_d(err0, 1);
err_dup1 = __msa_splati_d(err1, 1);
ADD2(err0, err_dup0, err1, err_dup1, err0, err1);
err += __msa_copy_s_d(err0, 0);
err += __msa_copy_s_d(err1, 0);
}
return err;
}
static void mbloop_filter_vertical_edge_uv_msa(uint8_t *src_u, uint8_t *src_v,
int32_t pitch,
const uint8_t b_limit_in,
const uint8_t limit_in,
const uint8_t thresh_in) {
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
b_limit = (v16u8)__msa_fill_b(b_limit_in);
limit = (v16u8)__msa_fill_b(limit_in);
thresh = (v16u8)__msa_fill_b(thresh_in);
LD_UB8(src_u - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src_v - 4, pitch, row8, row9, row10, row11, row12, row13, row14,
row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
row9, row10, row11, row12, row13, row14, row15, p3, p2,
p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
ILVRL_B2_SH(q2, q1, tmp2, tmp5);
src_u -= 3;
VP8_ST6x1_UB(tmp3, 0, tmp2, 0, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 1, tmp2, 1, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 2, tmp2, 2, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 3, tmp2, 3, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 0, tmp2, 4, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 1, tmp2, 5, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 2, tmp2, 6, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 3, tmp2, 7, src_u, 4);
src_v -= 3;
VP8_ST6x1_UB(tmp6, 0, tmp5, 0, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 1, tmp5, 1, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 2, tmp5, 2, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 3, tmp5, 3, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 0, tmp5, 4, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 1, tmp5, 5, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 2, tmp5, 6, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 3, tmp5, 7, src_v, 4);
}
void ff_vp8_h_loop_filter16_msa(uint8_t *src, ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
temp_src = src - 4;
LD_UB8(temp_src, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
temp_src += (8 * pitch);
LD_UB8(temp_src, pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
ILVRL_B2_SH(q2, q1, tmp2, tmp5);
temp_src = src - 3;
VP8_ST6x1_UB(tmp3, 0, tmp2, 0, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 1, tmp2, 1, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 2, tmp2, 2, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 3, tmp2, 3, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 0, tmp2, 4, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 1, tmp2, 5, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 2, tmp2, 6, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 3, tmp2, 7, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 0, tmp5, 0, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 1, tmp5, 1, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 2, tmp5, 2, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 3, tmp5, 3, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 0, tmp5, 4, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 1, tmp5, 5, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 2, tmp5, 6, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 3, tmp5, 7, temp_src, 4);
}