void vp8_loop_filter_simple_vertical_edge_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr) {
uint8_t *temp_src;
v16u8 p1, p0, q1, q0;
v16u8 mask, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1;
b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
temp_src = src - 2;
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);
TRANSPOSE16x4_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
row9, row10, row11, row12, row13, row14, row15, p1, p0,
q0, q1);
VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
ILVRL_B2_SH(q0, p0, tmp1, tmp0);
src -= 1;
ST2x4_UB(tmp1, 0, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp1, 4, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp0, 0, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp0, 4, src, pitch);
src += 4 * pitch;
}
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);
}
static int32_t sum_u8src_16width_msa(uint8_t *src, int32_t stride)
{
uint32_t sum = 0;
v16u8 in0, in1, in2, in3, in4, in5, in6, in7;
v16u8 in8, in9, in10, in11, in12, in13, in14, in15;
LD_UB8(src, stride, in0, in1, in2, in3, in4, in5, in6, in7);
src += (8 * stride);
LD_UB8(src, stride, in8, in9, in10, in11, in12, in13, in14, in15);
HADD_UB4_UB(in0, in1, in2, in3, in0, in1, in2, in3);
HADD_UB4_UB(in4, in5, in6, in7, in4, in5, in6, in7);
HADD_UB4_UB(in8, in9, in10, in11, in8, in9, in10, in11);
HADD_UB4_UB(in12, in13, in14, in15, in12, in13, in14, in15);
sum = HADD_UH_U32(in0);
sum += HADD_UH_U32(in1);
sum += HADD_UH_U32(in2);
sum += HADD_UH_U32(in3);
sum += HADD_UH_U32(in4);
sum += HADD_UH_U32(in5);
sum += HADD_UH_U32(in6);
sum += HADD_UH_U32(in7);
sum += HADD_UH_U32(in8);
sum += HADD_UH_U32(in9);
sum += HADD_UH_U32(in10);
sum += HADD_UH_U32(in11);
sum += HADD_UH_U32(in12);
sum += HADD_UH_U32(in13);
sum += HADD_UH_U32(in14);
sum += HADD_UH_U32(in15);
return sum;
}
static void copy_width8_msa(const uint8_t *src, int32_t src_stride,
uint8_t *dst, int32_t dst_stride, int32_t height) {
int32_t cnt;
uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
if (0 == height % 12) {
for (cnt = (height / 12); cnt--;) {
LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
src += (8 * src_stride);
out0 = __msa_copy_u_d((v2i64)src0, 0);
out1 = __msa_copy_u_d((v2i64)src1, 0);
out2 = __msa_copy_u_d((v2i64)src2, 0);
out3 = __msa_copy_u_d((v2i64)src3, 0);
out4 = __msa_copy_u_d((v2i64)src4, 0);
out5 = __msa_copy_u_d((v2i64)src5, 0);
out6 = __msa_copy_u_d((v2i64)src6, 0);
out7 = __msa_copy_u_d((v2i64)src7, 0);
SD4(out0, out1, out2, out3, dst, dst_stride);
dst += (4 * dst_stride);
SD4(out4, out5, out6, out7, dst, dst_stride);
dst += (4 * dst_stride);
LD_UB4(src, src_stride, src0, src1, src2, src3);
src += (4 * src_stride);
out0 = __msa_copy_u_d((v2i64)src0, 0);
out1 = __msa_copy_u_d((v2i64)src1, 0);
out2 = __msa_copy_u_d((v2i64)src2, 0);
out3 = __msa_copy_u_d((v2i64)src3, 0);
SD4(out0, out1, out2, out3, dst, dst_stride);
dst += (4 * dst_stride);
}
} else if (0 == height % 8) {
for (cnt = height >> 3; cnt--;) {
LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
src += (8 * src_stride);
out0 = __msa_copy_u_d((v2i64)src0, 0);
out1 = __msa_copy_u_d((v2i64)src1, 0);
out2 = __msa_copy_u_d((v2i64)src2, 0);
out3 = __msa_copy_u_d((v2i64)src3, 0);
out4 = __msa_copy_u_d((v2i64)src4, 0);
out5 = __msa_copy_u_d((v2i64)src5, 0);
out6 = __msa_copy_u_d((v2i64)src6, 0);
out7 = __msa_copy_u_d((v2i64)src7, 0);
SD4(out0, out1, out2, out3, dst, dst_stride);
dst += (4 * dst_stride);
SD4(out4, out5, out6, out7, dst, dst_stride);
dst += (4 * dst_stride);
}
} else if (0 == height % 4) {
void ff_vp8_v_loop_filter8uv_msa(uint8_t *src_u, uint8_t *src_v,
ptrdiff_t pitch, int b_limit_in, int limit_in,
int thresh_in)
{
uint8_t *temp_src;
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
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_u - (pitch << 2);
LD_UB8(temp_src, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
temp_src = src_v - (pitch << 2);
LD_UB8(temp_src, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
/* rht 8 element of p3 are u pixel and left 8 element of p3 are v pixel */
ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, 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);
p2_d = __msa_copy_u_d((v2i64) p2, 0);
p1_d = __msa_copy_u_d((v2i64) p1, 0);
p0_d = __msa_copy_u_d((v2i64) p0, 0);
q0_d = __msa_copy_u_d((v2i64) q0, 0);
q1_d = __msa_copy_u_d((v2i64) q1, 0);
q2_d = __msa_copy_u_d((v2i64) q2, 0);
src_u -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_u, pitch);
src_u += 4 * pitch;
SD(q1_d, src_u);
src_u += pitch;
SD(q2_d, src_u);
p2_d = __msa_copy_u_d((v2i64) p2, 1);
p1_d = __msa_copy_u_d((v2i64) p1, 1);
p0_d = __msa_copy_u_d((v2i64) p0, 1);
q0_d = __msa_copy_u_d((v2i64) q0, 1);
q1_d = __msa_copy_u_d((v2i64) q1, 1);
q2_d = __msa_copy_u_d((v2i64) q2, 1);
src_v -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_v, pitch);
src_v += 4 * pitch;
SD(q1_d, src_v);
src_v += pitch;
SD(q2_d, src_v);
}
static void mbloop_filter_horizontal_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;
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
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_u - (pitch << 2);
LD_UB8(temp_src, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
temp_src = src_v - (pitch << 2);
LD_UB8(temp_src, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, 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);
p2_d = __msa_copy_u_d((v2i64)p2, 0);
p1_d = __msa_copy_u_d((v2i64)p1, 0);
p0_d = __msa_copy_u_d((v2i64)p0, 0);
q0_d = __msa_copy_u_d((v2i64)q0, 0);
q1_d = __msa_copy_u_d((v2i64)q1, 0);
q2_d = __msa_copy_u_d((v2i64)q2, 0);
src_u -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_u, pitch);
src_u += 4 * pitch;
SD(q1_d, src_u);
src_u += pitch;
SD(q2_d, src_u);
p2_d = __msa_copy_u_d((v2i64)p2, 1);
p1_d = __msa_copy_u_d((v2i64)p1, 1);
p0_d = __msa_copy_u_d((v2i64)p0, 1);
q0_d = __msa_copy_u_d((v2i64)q0, 1);
q1_d = __msa_copy_u_d((v2i64)q1, 1);
q2_d = __msa_copy_u_d((v2i64)q2, 1);
src_v -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_v, pitch);
src_v += 4 * pitch;
SD(q1_d, src_v);
src_v += pitch;
SD(q2_d, src_v);
}
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);
}
static void loop_filter_horizontal_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;
LD_UB8((src - 4 * pitch), pitch, 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);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
}
void vp9_lpf_horizontal_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) {
uint64_t p1_d, p0_d, q0_d, q1_d;
v16u8 mask, hev, flat, thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p1_out, p0_out, q0_out, q1_out;
(void)count;
/* load vector elements */
LD_UB8((src - 4 * pitch), 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);
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_out, p0_out, q0_out, q1_out);
p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch);
}
void vp9_lpf_horizontal_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, thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
/* load vector elements */
LD_UB8((src - 4 * pitch), pitch, 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);
ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), 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);
}
static void loop_filter_horizontal_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) {
uint64_t p1_d, p0_d, q0_d, q1_d;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
thresh = (v16u8)__msa_fill_b(thresh_in);
limit = (v16u8)__msa_fill_b(limit_in);
b_limit = (v16u8)__msa_fill_b(b_limit_in);
src_u = src_u - (pitch << 2);
LD_UB8(src_u, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
src_u += (5 * pitch);
src_v = src_v - (pitch << 2);
LD_UB8(src_v, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
src_v += (5 * pitch);
/* right 8 element of p3 are u pixel and
left 8 element of p3 are v pixel */
ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, 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);
p1_d = __msa_copy_u_d((v2i64)p1, 0);
p0_d = __msa_copy_u_d((v2i64)p0, 0);
q0_d = __msa_copy_u_d((v2i64)q0, 0);
q1_d = __msa_copy_u_d((v2i64)q1, 0);
SD4(q1_d, q0_d, p0_d, p1_d, src_u, (-pitch));
p1_d = __msa_copy_u_d((v2i64)p1, 1);
p0_d = __msa_copy_u_d((v2i64)p0, 1);
q0_d = __msa_copy_u_d((v2i64)q0, 1);
q1_d = __msa_copy_u_d((v2i64)q1, 1);
SD4(q1_d, q0_d, p0_d, p1_d, src_v, (-pitch));
}
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);
}
void ff_vp8_v_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;
/* load vector elements */
LD_UB8((src - 4 * pitch), pitch, 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);
ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
}
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 mbloop_filter_horizontal_edge_y_msa(uint8_t *src, int32_t pitch,
const uint8_t b_limit_in,
const uint8_t limit_in,
const uint8_t thresh_in) {
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
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 - (pitch << 2);
LD_UB8(temp_src, pitch, 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);
temp_src = src - 3 * pitch;
ST_UB4(p2, p1, p0, q0, temp_src, pitch);
temp_src += (4 * pitch);
ST_UB2(q1, q2, temp_src, pitch);
}
void ff_vp8_v_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;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
/* load vector elements */
temp_src = src - (pitch << 2);
LD_UB8(temp_src, pitch, 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);
/* store vector elements */
temp_src = src - 3 * pitch;
ST_UB4(p2, p1, p0, q0, temp_src, pitch);
temp_src += (4 * pitch);
ST_UB2(q1, q2, temp_src, pitch);
}
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);
}
void vpx_lpf_horizontal_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) {
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
v16u8 flat, mask, hev, tmp, thresh, b_limit, limit;
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 };
/* load vector elements */
LD_UB8(src - (4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh0);
tmp = (v16u8)__msa_fill_b(*thresh1);
thresh = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)thresh);
b_limit = (v16u8)__msa_fill_b(*b_limit0);
tmp = (v16u8)__msa_fill_b(*b_limit1);
b_limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)b_limit);
limit = (v16u8)__msa_fill_b(*limit0);
tmp = (v16u8)__msa_fill_b(*limit1);
limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)limit);
/* mask and hev */
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
if (__msa_test_bz_v(flat)) {
ST_UB4(p1_out, p0_out, q0_out, q1_out, (src - 2 * pitch), 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);
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_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
src -= 3 * pitch;
ST_UB4(p2_out, p1_out, p0_out, q0_out, src, pitch);
src += (4 * pitch);
ST_UB2(q1_out, q2_out, src, pitch);
src += (2 * pitch);
}
}
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);
}
}
void vpx_lpf_horizontal_8_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr,
const uint8_t *limit_ptr,
const uint8_t *thresh_ptr) {
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
v16u8 mask, hev, flat, thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
v8i16 p2_filter8, p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8;
v8u16 p3_r, p2_r, p1_r, p0_r, q3_r, q2_r, q1_r, q0_r;
v16i8 zero = { 0 };
/* load vector elements */
LD_UB8((src - 4 * pitch), 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);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
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)) {
p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), 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_filter8,
p1_filter8, p0_filter8, q0_filter8, q1_filter8, q2_filter8);
/* convert 16 bit output data into 8 bit */
PCKEV_B4_SH(zero, p2_filter8, zero, p1_filter8, zero, p0_filter8,
zero, q0_filter8, p2_filter8, p1_filter8, p0_filter8,
q0_filter8);
PCKEV_B2_SH(zero, q1_filter8, zero, q2_filter8, q1_filter8, q2_filter8);
/* store pixel values */
p2_out = __msa_bmnz_v(p2, (v16u8)p2_filter8, flat);
p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filter8, flat);
p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filter8, flat);
q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filter8, flat);
q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filter8, flat);
q2_out = __msa_bmnz_v(q2, (v16u8)q2_filter8, flat);
p2_d = __msa_copy_u_d((v2i64)p2_out, 0);
p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
q2_d = __msa_copy_u_d((v2i64)q2_out, 0);
src -= 3 * pitch;
SD4(p2_d, p1_d, p0_d, q0_d, src, pitch);
src += (4 * pitch);
SD(q1_d, src);
src += pitch;
SD(q2_d, src);
}
}
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);
}
