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);
}
static void intra_predict_tm_4x4_msa(const uint8_t *src_top_ptr,
const uint8_t *src_left,
uint8_t *dst, int32_t dst_stride) {
uint32_t val;
uint8_t top_left = src_top_ptr[-1];
v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };
v16u8 src0, src1, src2, src3;
v8u16 src_top_left, vec0, vec1, vec2, vec3;
src_top_left = (v8u16)__msa_fill_h(top_left);
val = LW(src_top_ptr);
src_top = (v16i8)__msa_insert_w((v4i32)src_top, 0, val);
src_left0 = __msa_fill_b(src_left[0]);
src_left1 = __msa_fill_b(src_left[1]);
src_left2 = __msa_fill_b(src_left[2]);
src_left3 = __msa_fill_b(src_left[3]);
ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
src_left3, src_top, src0, src1, src2, src3);
HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
ST4x4_UB(tmp0, tmp1, 0, 2, 0, 2, dst, dst_stride);
}
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);
}
static void loop_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) {
uint8_t *temp_src_u, *temp_src_v;
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;
v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
thresh = (v16u8)__msa_fill_b(thresh_in);
limit = (v16u8)__msa_fill_b(limit_in);
b_limit = (v16u8)__msa_fill_b(b_limit_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_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ILVR_B2_SW(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SW(tmp1, tmp0, tmp2, tmp3);
tmp0 = (v4i32)__msa_ilvl_b((v16i8)p0, (v16i8)p1);
tmp1 = (v4i32)__msa_ilvl_b((v16i8)q1, (v16i8)q0);
ILVRL_H2_SW(tmp1, tmp0, tmp4, tmp5);
temp_src_u = src_u - 2;
ST4x4_UB(tmp2, tmp2, 0, 1, 2, 3, temp_src_u, pitch);
temp_src_u += 4 * pitch;
ST4x4_UB(tmp3, tmp3, 0, 1, 2, 3, temp_src_u, pitch);
temp_src_v = src_v - 2;
ST4x4_UB(tmp4, tmp4, 0, 1, 2, 3, temp_src_v, pitch);
temp_src_v += 4 * pitch;
ST4x4_UB(tmp5, tmp5, 0, 1, 2, 3, temp_src_v, pitch);
}
static void hevc_addblk_4x4_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint32_t dst0, dst1, dst2, dst3;
v8i16 dst_r0, dst_l0, in0, in1;
v4i32 dst_vec = { 0 };
v16u8 zeros = { 0 };
LD_SH2(coeffs, 8, in0, in1);
LW4(dst, stride, dst0, dst1, dst2, dst3);
INSERT_W4_SW(dst0, dst1, dst2, dst3, dst_vec);
ILVRL_B2_SH(zeros, dst_vec, dst_r0, dst_l0);
ADD2(dst_r0, in0, dst_l0, in1, dst_r0, dst_l0);
CLIP_SH2_0_255(dst_r0, dst_l0);
dst_vec = (v4i32) __msa_pckev_b((v16i8) dst_l0, (v16i8) dst_r0);
ST4x4_UB(dst_vec, dst_vec, 0, 1, 2, 3, dst, stride);
}
static void avc_idct4x4_addblk_dc_msa( uint8_t *p_dst, int16_t *p_src,
int32_t i_dst_stride )
{
int16_t i_dc;
uint32_t i_src0, i_src1, i_src2, i_src3;
v16u8 pred = { 0 };
v16i8 out;
v8i16 input_dc, pred_r, pred_l;
i_dc = ( p_src[0] + 32 ) >> 6;
input_dc = __msa_fill_h( i_dc );
p_src[ 0 ] = 0;
LW4( p_dst, i_dst_stride, i_src0, i_src1, i_src2, i_src3 );
INSERT_W4_UB( i_src0, i_src1, i_src2, i_src3, pred );
UNPCK_UB_SH( pred, pred_r, pred_l );
pred_r += input_dc;
pred_l += input_dc;
CLIP_SH2_0_255( pred_r, pred_l );
out = __msa_pckev_b( ( v16i8 ) pred_l, ( v16i8 ) pred_r );
ST4x4_UB( out, out, 0, 1, 2, 3, p_dst, i_dst_stride );
}
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);
}
}
