static int32_t avc_coeff_last16_msa( int16_t *p_src )
{
uint32_t u_res;
v8i16 src0, src1;
v8u16 tmp_h0;
v16u8 tmp0;
v8i16 out0, out1;
v16i8 res0;
v16u8 mask = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
LD_SH2( p_src, 8, src0, src1 );
out0 = __msa_ceqi_h( src0, 0 );
out1 = __msa_ceqi_h( src1, 0 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) out1, ( v16i8 ) out0 );
tmp0 = tmp0 & mask;
tmp_h0 = __msa_hadd_u_h( tmp0, tmp0 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) tmp_h0, ( v16i8 ) tmp_h0 );
tmp_h0 = __msa_hadd_u_h( tmp0, tmp0 );
tmp0 = ( v16u8 ) __msa_pckev_b( ( v16i8 ) tmp_h0, ( v16i8 ) tmp_h0 );
tmp_h0 = __msa_hadd_u_h( tmp0, tmp0 );
res0 = __msa_pckev_b( ( v16i8 ) tmp_h0, ( v16i8 ) tmp_h0 );
out0 = __msa_nloc_h( ( v8i16 ) res0 );
u_res = __msa_copy_u_h( out0, 0 );
return ( 15 - u_res );
}
static WEBP_INLINE void DCMode16x16(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
int DC;
v16u8 out;
if (top != NULL && left != NULL) {
const v16u8 rtop = LD_UB(top);
const v8u16 dctop = __msa_hadd_u_h(rtop, rtop);
const v16u8 rleft = LD_UB(left);
const v8u16 dcleft = __msa_hadd_u_h(rleft, rleft);
const v8u16 dctemp = dctop + dcleft;
DC = HADD_UH_U32(dctemp);
DC = (DC + 16) >> 5;
} else if (left != NULL) { // left but no top
static void intra_predict_dc_8x8_msa(uint8_t *src_top, uint8_t *src_left,
int32_t src_stride_left,
uint8_t *dst, int32_t dst_stride,
uint8_t is_above, uint8_t is_left)
{
uint32_t row, addition = 0;
uint64_t out;
v16u8 src_above, store;
v8u16 sum_above;
v4u32 sum_top;
v2u64 sum;
if (is_left && is_above)
{
src_above = LD_UB(src_top);
sum_above = __msa_hadd_u_h(src_above, src_above);
sum_top = __msa_hadd_u_w(sum_above, sum_above);
sum = __msa_hadd_u_d(sum_top, sum_top);
addition = __msa_copy_u_w((v4i32)sum, 0);
for (row = 0; row < 8; ++row)
{
addition += src_left[row * src_stride_left];
}
addition = (addition + 8) >> 4;
store = (v16u8)__msa_fill_b(addition);
}
static void intra_predict_dc_8x8_msa(const uint8_t *src_top,
const uint8_t *src_left, uint8_t *dst,
int32_t dst_stride) {
uint64_t val0, val1;
v16i8 store;
v16u8 src = { 0 };
v8u16 sum_h;
v4u32 sum_w;
v2u64 sum_d;
val0 = LD(src_top);
val1 = LD(src_left);
INSERT_D2_UB(val0, val1, src);
sum_h = __msa_hadd_u_h(src, src);
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, 4);
store = __msa_splati_b((v16i8)sum_w, 0);
val0 = __msa_copy_u_d((v2i64)store, 0);
SD4(val0, val0, val0, val0, dst, dst_stride);
dst += (4 * dst_stride);
SD4(val0, val0, val0, val0, dst, dst_stride);
}
static void intra_predict_dc_tl_4x4_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint32_t val0;
v16i8 store, data = { 0 };
v8u16 sum_h;
v4u32 sum_w;
val0 = LW(src);
data = (v16i8)__msa_insert_w((v4i32)data, 0, val0);
sum_h = __msa_hadd_u_h((v16u8)data, (v16u8)data);
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_w, 2);
store = __msa_splati_b((v16i8)sum_w, 0);
val0 = __msa_copy_u_w((v4i32)store, 0);
SW4(val0, val0, val0, val0, dst, dst_stride);
}
static void intra_predict_dc_tl_16x16_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
v16u8 data, out;
v8u16 sum_h;
v4u32 sum_w;
v2u64 sum_d;
data = LD_UB(src);
sum_h = __msa_hadd_u_h(data, data);
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, 4);
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);
}
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 );
}
uint32_t vp10_avg_4x4_msa(const uint8_t *src, int32_t src_stride) {
uint32_t sum_out;
uint32_t src0, src1, src2, src3;
v16u8 vec = { 0 };
v8u16 sum0;
v4u32 sum1;
v2u64 sum2;
LW4(src, src_stride, src0, src1, src2, src3);
INSERT_W4_UB(src0, src1, src2, src3, vec);
sum0 = __msa_hadd_u_h(vec, vec);
sum1 = __msa_hadd_u_w(sum0, sum0);
sum0 = (v8u16)__msa_pckev_h((v8i16)sum1, (v8i16)sum1);
sum1 = __msa_hadd_u_w(sum0, sum0);
sum2 = __msa_hadd_u_d(sum1, sum1);
sum1 = (v4u32)__msa_srari_w((v4i32)sum2, 4);
sum_out = __msa_copy_u_w((v4i32)sum1, 0);
return sum_out;
}
static void intra_predict_dc_4x4_msa(const uint8_t *src_top,
const uint8_t *src_left, uint8_t *dst,
int32_t dst_stride) {
uint32_t val0, val1;
v16i8 store, src = { 0 };
v8u16 sum_h;
v4u32 sum_w;
v2u64 sum_d;
val0 = LW(src_top);
val1 = LW(src_left);
INSERT_W2_SB(val0, val1, src);
sum_h = __msa_hadd_u_h((v16u8)src, (v16u8)src);
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
store = __msa_splati_b((v16i8)sum_w, 0);
val0 = __msa_copy_u_w((v4i32)store, 0);
SW4(val0, val0, val0, val0, dst, dst_stride);
}
static void intra_predict_dc_tl_8x8_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint64_t val0;
v16i8 store;
v16u8 data = { 0 };
v8u16 sum_h;
v4u32 sum_w;
v2u64 sum_d;
val0 = LD(src);
data = (v16u8)__msa_insert_d((v2i64)data, 0, val0);
sum_h = __msa_hadd_u_h(data, data);
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 3);
store = __msa_splati_b((v16i8)sum_w, 0);
val0 = __msa_copy_u_d((v2i64)store, 0);
SD4(val0, val0, val0, val0, dst, dst_stride);
dst += (4 * dst_stride);
SD4(val0, val0, val0, val0, dst, dst_stride);
}
static WEBP_INLINE void DCMode16x16(uint8_t* dst, const uint8_t* left,
const uint8_t* top) {
int DC;
v16u8 out;
if (top != NULL && left != NULL) {
const v16u8 rtop = LD_UB(top);
const v8u16 dctop = __msa_hadd_u_h(rtop, rtop);
const v16u8 rleft = LD_UB(left);
const v8u16 dcleft = __msa_hadd_u_h(rleft, rleft);
const v8u16 dctemp = dctop + dcleft;
DC = HADD_UH_U32(dctemp);
DC = (DC + 16) >> 5;
} else if (left != NULL) { // left but no top
const v16u8 rleft = LD_UB(left);
const v8u16 dcleft = __msa_hadd_u_h(rleft, rleft);
DC = HADD_UH_U32(dcleft);
DC = (DC + DC + 16) >> 5;
} else if (top != NULL) { // top but no left
const v16u8 rtop = LD_UB(top);
const v8u16 dctop = __msa_hadd_u_h(rtop, rtop);
DC = HADD_UH_U32(dctop);
DC = (DC + DC + 16) >> 5;
} else { // no top, no left, nothing.
DC = 0x80;
}
out = (v16u8)__msa_fill_b(DC);
STORE16x16(out, dst);
}
static void Intra16Preds(uint8_t* dst,
