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 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_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;
}
}
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;
}
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;
}
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 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 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);
}