static void intra_predict_dc_32x32_msa(const uint8_t *src_top,
const uint8_t *src_left, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
v16u8 top0, top1, left0, left1, out;
v8u16 sum_h, sum_top0, sum_top1, sum_left0, sum_left1;
v4u32 sum_w;
v2u64 sum_d;
LD_UB2(src_top, 16, top0, top1);
LD_UB2(src_left, 16, left0, left1);
HADD_UB2_UH(top0, top1, sum_top0, sum_top1);
HADD_UB2_UH(left0, left1, sum_left0, sum_left1);
sum_h = sum_top0 + sum_top1;
sum_h += sum_left0 + sum_left1;
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 6);
out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
for (row = 16; row--;) {
ST_UB2(out, out, dst, 16);
dst += dst_stride;
ST_UB2(out, out, dst, 16);
dst += dst_stride;
}
}
static void intra_predict_tm_16x16_msa(const uint8_t *src_top_ptr,
const uint8_t *src_left,
uint8_t *dst, int32_t dst_stride) {
uint8_t top_left = src_top_ptr[-1];
uint32_t loop_cnt;
v16i8 src_top, src_left0, src_left1, src_left2, src_left3;
v8u16 src_top_left, res_r, res_l;
src_top = LD_SB(src_top_ptr);
src_top_left = (v8u16)__msa_fill_h(top_left);
for (loop_cnt = 4; loop_cnt--;) {
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]);
src_left += 4;
ILVRL_B2_UH(src_left0, src_top, res_r, res_l);
HADD_UB2_UH(res_r, res_l, res_r, res_l);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
SAT_UH2_UH(res_r, res_l, 7);
PCKEV_ST_SB(res_r, res_l, dst);
dst += dst_stride;
ILVRL_B2_UH(src_left1, src_top, res_r, res_l);
HADD_UB2_UH(res_r, res_l, res_r, res_l);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
SAT_UH2_UH(res_r, res_l, 7);
PCKEV_ST_SB(res_r, res_l, dst);
dst += dst_stride;
ILVRL_B2_UH(src_left2, src_top, res_r, res_l);
HADD_UB2_UH(res_r, res_l, res_r, res_l);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
SAT_UH2_UH(res_r, res_l, 7);
PCKEV_ST_SB(res_r, res_l, dst);
dst += dst_stride;
ILVRL_B2_UH(src_left3, src_top, res_r, res_l);
HADD_UB2_UH(res_r, res_l, res_r, res_l);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, res_r, res_l);
SAT_UH2_UH(res_r, res_l, 7);
PCKEV_ST_SB(res_r, res_l, dst);
dst += dst_stride;
}
}
static void intra_predict_dc_tl_32x32_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
v16u8 data0, data1, out;
v8u16 sum_h, sum_data0, sum_data1;
v4u32 sum_w;
v2u64 sum_d;
LD_UB2(src, 16, data0, data1);
HADD_UB2_UH(data0, data1, sum_data0, sum_data1);
sum_h = sum_data0 + sum_data1;
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
for (row = 16; row--;) {
ST_UB2(out, out, dst, 16);
dst += dst_stride;
ST_UB2(out, out, dst, 16);
dst += dst_stride;
}
}
static int32_t avc_coeff_last64_msa( int16_t *p_src )
{
uint32_t u_res;
v8i16 src0, src1, src2, src3, src4, src5, src6, src7;
v8i16 tmp_h0, tmp_h1, tmp_h2, tmp_h3, tmp_h4, tmp_h5, tmp_h6, tmp_h7;
v16u8 tmp0, tmp1, tmp2, tmp3;
v8u16 vec0, vec1, vec2, vec3;
v4i32 out0;
v16u8 mask = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
LD_SH8( p_src, 8, src0, src1, src2, src3, src4, src5, src6, src7 );
tmp_h0 = __msa_ceqi_h( src0, 0 );
tmp_h1 = __msa_ceqi_h( src1, 0 );
tmp_h2 = __msa_ceqi_h( src2, 0 );
tmp_h3 = __msa_ceqi_h( src3, 0 );
tmp_h4 = __msa_ceqi_h( src4, 0 );
tmp_h5 = __msa_ceqi_h( src5, 0 );
tmp_h6 = __msa_ceqi_h( src6, 0 );
tmp_h7 = __msa_ceqi_h( src7, 0 );
PCKEV_B4_UB( tmp_h1, tmp_h0, tmp_h3, tmp_h2, tmp_h5, tmp_h4, tmp_h7, tmp_h6,
tmp0, tmp1, tmp2, tmp3 );
tmp0 = tmp0 & mask;
tmp1 = tmp1 & mask;
tmp2 = tmp2 & mask;
tmp3 = tmp3 & mask;
HADD_UB4_UH( tmp0, tmp1, tmp2, tmp3, vec0, vec1, vec2, vec3 );
PCKEV_B2_UB( vec1, vec0, vec3, vec2, tmp0, tmp1 );
HADD_UB2_UH( tmp0, tmp1, vec0, vec1 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) vec1, ( v16i8 ) vec0 );
vec0 = __msa_hadd_u_h( tmp0, tmp0 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) vec0, ( v16i8 ) vec0 );
out0 = ( v4i32 ) __msa_nloc_d( ( v2i64 ) tmp0 );
u_res = __msa_copy_u_w( out0, 0 );
return ( 63 - u_res );
}
static void intra_predict_dc_16x16_msa(const uint8_t *src_top,
const uint8_t *src_left, uint8_t *dst,
int32_t dst_stride) {
v16u8 top, left, out;
v8u16 sum_h, sum_top, sum_left;
v4u32 sum_w;
v2u64 sum_d;
top = LD_UB(src_top);
left = LD_UB(src_left);
HADD_UB2_UH(top, left, sum_top, sum_left);
sum_h = sum_top + sum_left;
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
dst += (8 * dst_stride);
ST_UB8(out, out, out, out, out, out, out, out, dst, dst_stride);
}