static void filter_by_weight8x8_msa(const uint8_t *src_ptr, int32_t src_stride,
uint8_t *dst_ptr, int32_t dst_stride,
int32_t src_weight) {
int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight;
int32_t row;
uint64_t src0_d, src1_d, dst0_d, dst1_d;
v16i8 src0 = { 0 };
v16i8 src1 = { 0 };
v16i8 dst0 = { 0 };
v16i8 dst1 = { 0 };
v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l;
src_wt = __msa_fill_h(src_weight);
dst_wt = __msa_fill_h(dst_weight);
for (row = 2; row--;) {
LD2(src_ptr, src_stride, src0_d, src1_d);
src_ptr += (2 * src_stride);
LD2(dst_ptr, dst_stride, dst0_d, dst1_d);
INSERT_D2_SB(src0_d, src1_d, src0);
INSERT_D2_SB(dst0_d, dst1_d, dst0);
LD2(src_ptr, src_stride, src0_d, src1_d);
src_ptr += (2 * src_stride);
LD2((dst_ptr + 2 * dst_stride), dst_stride, dst0_d, dst1_d);
INSERT_D2_SB(src0_d, src1_d, src1);
INSERT_D2_SB(dst0_d, dst1_d, dst1);
UNPCK_UB_SH(src0, src_r, src_l);
UNPCK_UB_SH(dst0, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
dst0 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r);
ST8x2_UB(dst0, dst_ptr, dst_stride);
dst_ptr += (2 * dst_stride);
UNPCK_UB_SH(src1, src_r, src_l);
UNPCK_UB_SH(dst1, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
dst1 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r);
ST8x2_UB(dst1, dst_ptr, dst_stride);
dst_ptr += (2 * dst_stride);
}
}
static void filter_by_weight16x16_msa(const uint8_t *src_ptr,
int32_t src_stride, uint8_t *dst_ptr,
int32_t dst_stride, int32_t src_weight) {
int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight;
int32_t row;
v16i8 src0, src1, src2, src3, dst0, dst1, dst2, dst3;
v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l;
src_wt = __msa_fill_h(src_weight);
dst_wt = __msa_fill_h(dst_weight);
for (row = 4; row--;) {
LD_SB4(src_ptr, src_stride, src0, src1, src2, src3);
src_ptr += (4 * src_stride);
LD_SB4(dst_ptr, dst_stride, dst0, dst1, dst2, dst3);
UNPCK_UB_SH(src0, src_r, src_l);
UNPCK_UB_SH(dst0, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src1, src_r, src_l);
UNPCK_UB_SH(dst1, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src2, src_r, src_l);
UNPCK_UB_SH(dst2, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
UNPCK_UB_SH(src3, src_r, src_l);
UNPCK_UB_SH(dst3, dst_r, dst_l);
res_h_r = (src_r * src_wt);
res_h_r += (dst_r * dst_wt);
res_h_l = (src_l * src_wt);
res_h_l += (dst_l * dst_wt);
SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION);
PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr);
dst_ptr += dst_stride;
}
}
static void hevc_addblk_8x8_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t *temp_dst = dst;
uint64_t dst0, dst1, dst2, dst3;
v2i64 dst_vec0 = { 0 };
v2i64 dst_vec1 = { 0 };
v8i16 dst_r0, dst_l0, dst_r1, dst_l1;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v16u8 zeros = { 0 };
LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
temp_dst += (4 * stride);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
ILVRL_B2_SH(zeros, dst_vec0, dst_r0, dst_l0);
ILVRL_B2_SH(zeros, dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in0, dst_l0, in1, dst_r1, in2, dst_l1, in3,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, dst, stride);
dst += (4 * stride);
LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
INSERT_D2_SD(dst0, dst1, dst_vec0);
INSERT_D2_SD(dst2, dst3, dst_vec1);
UNPCK_UB_SH(dst_vec0, dst_r0, dst_l0);
UNPCK_UB_SH(dst_vec1, dst_r1, dst_l1);
ADD4(dst_r0, in4, dst_l0, in5, dst_r1, in6, dst_l1, in7,
dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
ST8x4_UB(dst_r0, dst_r1, 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
yuv_abgr_convert_msa (JSAMPROW p_in_y, JSAMPROW p_in_cb, JSAMPROW p_in_cr,
JSAMPROW p_rgb, JDIMENSION out_width)
{
int y, cb, cr;
unsigned int col, num_cols_mul_16 = out_width >> 4;
unsigned int remaining_wd = out_width & 0xF;
v16i8 alpha = __msa_ldi_b(0xFF);
v16i8 const_128 = __msa_ldi_b(128);
v16u8 out0, out1, out2, out3, input_y = {0};
v16i8 input_cb, input_cr, out_rgb0, out_rgb1, out_ab0, out_ab1;
v8i16 y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1;
v4i32 cb_w0, cb_w1, cb_w2, cb_w3, cr_w0, cr_w1, cr_w2, cr_w3, zero = {0};
v16i8 out_r0, out_g0, out_b0;
for (col = num_cols_mul_16; col--;) {
input_y = LD_UB(p_in_y);
input_cb = LD_SB(p_in_cb);
input_cr = LD_SB(p_in_cr);
p_in_y += 16;
p_in_cb += 16;
p_in_cr += 16;
input_cb -= const_128;
input_cr -= const_128;
UNPCK_UB_SH(input_y, y_h0, y_h1);
UNPCK_SB_SH(input_cb, cb_h0, cb_h1);
UNPCK_SB_SH(input_cr, cr_h0, cr_h1);
CALC_G4_FRM_YUV(y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1, out_g0);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
UNPCK_SH_SW(cr_h1, cr_w2, cr_w3);
CALC_R4_FRM_YUV(y_h0, y_h1, cr_w0, cr_w1, cr_w2, cr_w3, out_r0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cb_h1, cb_w2, cb_w3);
CALC_B4_FRM_YUV(y_h0, y_h1, cb_w0, cb_w1, cb_w2, cb_w3, out_b0);
ILVRL_B2_SB(out_r0, out_g0, out_rgb0, out_rgb1);
ILVRL_B2_SB(out_b0, alpha, out_ab0, out_ab1);
ILVRL_H2_UB(out_rgb0, out_ab0, out0, out1);
ILVRL_H2_UB(out_rgb1, out_ab1, out2, out3);
ST_UB4(out0, out1, out2, out3, p_rgb, 16);
p_rgb += 16 * 4;
}
if (remaining_wd >= 8) {
uint64_t in_y, in_cb, in_cr;
v16i8 input_cbcr = {0};
in_y = LD(p_in_y);
in_cb = LD(p_in_cb);
in_cr = LD(p_in_cr);
p_in_y += 8;
p_in_cb += 8;
p_in_cr += 8;
input_y = (v16u8) __msa_insert_d((v2i64) input_y, 0, in_y);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 0, in_cb);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 1, in_cr);
input_cbcr -= const_128;
y_h0 = (v8i16) __msa_ilvr_b((v16i8) zero, (v16i8) input_y);
UNPCK_SB_SH(input_cbcr, cb_h0, cr_h0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
CALC_R2_FRM_YUV(y_h0, cr_w0, cr_w1, out_r0);
CALC_G2_FRM_YUV(y_h0, cb_h0, cr_h0, out_g0);
CALC_B2_FRM_YUV(y_h0, cb_w0, cb_w1, out_b0);
out_rgb0 = (v16i8) __msa_ilvr_b((v16i8) out_r0, (v16i8) out_g0);
out_ab0 = (v16i8) __msa_ilvr_b((v16i8) out_b0, alpha);
ILVRL_H2_UB(out_rgb0, out_ab0, out0, out1);
ST_UB2(out0, out1, p_rgb, 16);
p_rgb += 16 * 2;
remaining_wd -= 8;
}
for (col = 0; col < remaining_wd; col++) {
y = (int) (p_in_y[col]);
cb = (int) (p_in_cb[col]) - 128;
cr = (int) (p_in_cr[col]) - 128;
p_rgb[0] = 0xFF;
p_rgb[1] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_77200 * cb, 16));
p_rgb[2] = clip_pixel(y + ROUND_POWER_OF_TWO(((-FIX_0_34414) * cb -
FIX_0_71414 * cr), 16));
p_rgb[3] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_40200 * cr, 16));
p_rgb += 4;
}
}
void
yuv_bgr_convert_msa (JSAMPROW p_in_y, JSAMPROW p_in_cb, JSAMPROW p_in_cr,
JSAMPROW p_rgb, JDIMENSION out_width)
{
int32_t y, cb, cr;
uint32_t col, num_cols_mul_16 = out_width >> 4;
uint32_t remaining_wd = out_width & 0xF;
v16u8 mask_rgb0 = {0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19, 8, 9, 20, 10};
v16u8 mask_rgb1 = {11, 21, 12, 13, 22, 14, 15, 23, 0, 1, 24, 2, 3, 25, 4, 5};
v16u8 mask_rgb2 = {26, 6, 7, 27, 8, 9, 28, 10, 11, 29, 12, 13, 30, 14, 15, 31};
v16u8 tmp0, tmp1, out0, out1, out2, input_y = {0};
v16i8 input_cb, input_cr, out_rgb0, out_rgb1, const_128 = __msa_ldi_b(128);
v8i16 y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1;
v4i32 cb_w0, cb_w1, cb_w2, cb_w3, cr_w0, cr_w1, cr_w2, cr_w3, zero = {0};
v16i8 out_r0, out_g0, out_b0;
for (col = num_cols_mul_16; col--;) {
input_y = LD_UB(p_in_y);
input_cb = LD_SB(p_in_cb);
input_cr = LD_SB(p_in_cr);
p_in_y += 16;
p_in_cb += 16;
p_in_cr += 16;
input_cb -= const_128;
input_cr -= const_128;
UNPCK_UB_SH(input_y, y_h0, y_h1);
UNPCK_SB_SH(input_cb, cb_h0, cb_h1);
UNPCK_SB_SH(input_cr, cr_h0, cr_h1);
CALC_G4_FRM_YUV(y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1, out_g0);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
UNPCK_SH_SW(cr_h1, cr_w2, cr_w3);
CALC_R4_FRM_YUV(y_h0, y_h1, cr_w0, cr_w1, cr_w2, cr_w3, out_r0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cb_h1, cb_w2, cb_w3);
CALC_B4_FRM_YUV(y_h0, y_h1, cb_w0, cb_w1, cb_w2, cb_w3, out_b0);
ILVRL_B2_SB(out_g0, out_b0, out_rgb0, out_rgb1);
VSHF_B2_UB(out_rgb0, out_r0, out_rgb0, out_r0, mask_rgb0, mask_rgb1,
out0, tmp0);
VSHF_B2_UB(out_rgb1, out_r0, out_rgb1, out_r0, mask_rgb1, mask_rgb2,
tmp1, out2);
out1 = (v16u8) __msa_sldi_b((v16i8) zero, (v16i8) tmp1, 8);
out1 = (v16u8) __msa_pckev_d((v2i64) out1, (v2i64) tmp0);
ST_UB(out0, p_rgb);
p_rgb += 16;
ST_UB(out1, p_rgb);
p_rgb += 16;
ST_UB(out2, p_rgb);
p_rgb += 16;
}
if (remaining_wd >= 8) {
uint64_t in_y, in_cb, in_cr;
v16i8 input_cbcr = {0};
in_y = LD(p_in_y);
in_cb = LD(p_in_cb);
in_cr = LD(p_in_cr);
p_in_y += 8;
p_in_cb += 8;
p_in_cr += 8;
input_y = (v16u8) __msa_insert_d((v2i64) input_y, 0, in_y);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 0, in_cb);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 1, in_cr);
input_cbcr -= const_128;
y_h0 = (v8i16) __msa_ilvr_b((v16i8) zero, (v16i8) input_y);
UNPCK_SB_SH(input_cbcr, cb_h0, cr_h0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
CALC_R2_FRM_YUV(y_h0, cr_w0, cr_w1, out_r0);
CALC_G2_FRM_YUV(y_h0, cb_h0, cr_h0, out_g0);
CALC_B2_FRM_YUV(y_h0, cb_w0, cb_w1, out_b0);
out_rgb0 = (v16i8) __msa_ilvr_b((v16i8) out_g0, (v16i8) out_b0);
VSHF_B2_UB(out_rgb0, out_r0, out_rgb0, out_r0, mask_rgb0, mask_rgb1,
out0, out1);
ST_UB(out0, p_rgb);
p_rgb += 16;
ST8x1_UB(out1, p_rgb);
p_rgb += 8;
remaining_wd -= 8;
}
for (col = 0; col < remaining_wd; col++) {
y = (int) (p_in_y[col]);
cb = (int) (p_in_cb[col]) - 128;
cr = (int) (p_in_cr[col]) - 128;
/* Range-limiting is essential due to noise introduced by DCT losses. */
p_rgb[0] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_77200 * cb, 16));
p_rgb[1] = clip_pixel(y + ROUND_POWER_OF_TWO(((-FIX_0_34414) * cb -
FIX_0_71414 * cr), 16));
p_rgb[2] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_40200 * cr, 16));
p_rgb += 3;
}
}
static void temporal_filter_apply_16size_msa(uint8_t *frame1_ptr,
uint32_t stride,
uint8_t *frame2_ptr,
int32_t strength_in,
int32_t filter_wt_in,
uint32_t *acc, uint16_t *cnt)
{
uint32_t row;
v16i8 frame1_0_b, frame1_1_b, frame2_0_b, frame2_1_b;
v16u8 frame_l, frame_h;
v16i8 zero = { 0 };
v8i16 frame2_0_h, frame2_1_h, mod0_h, mod1_h;
v8i16 diff0, diff1, cnt0, cnt1;
v4i32 const3, const16, filter_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frame2_0, frame2_1, frame2_2, frame2_3;
v4i32 acc0, acc1, acc2, acc3;
filter_wt = __msa_fill_w(filter_wt_in);
strength = __msa_fill_w(strength_in);
const3 = __msa_ldi_w(3);
const16 = __msa_ldi_w(16);
for (row = 8; row--;)
{
frame1_0_b = LD_SB(frame1_ptr);
frame2_0_b = LD_SB(frame2_ptr);
frame1_ptr += stride;
frame2_ptr += 16;
frame1_1_b = LD_SB(frame1_ptr);
frame2_1_b = LD_SB(frame2_ptr);
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame1_0_b, frame2_0_b, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h)
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
ILVRL_B2_SH(zero, frame2_0_b, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame1_1_b, frame2_1_b, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame2_1_b, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
frame1_ptr += stride;
frame2_ptr += 16;
}
}
static void temporal_filter_apply_8size_msa(uint8_t *frame1_ptr,
uint32_t stride,
uint8_t *frame2_ptr,
int32_t strength_in,
int32_t filter_wt_in,
uint32_t *acc, uint16_t *cnt)
{
uint32_t row;
uint64_t f0, f1, f2, f3, f4, f5, f6, f7;
v16i8 frame1 = { 0 };
v16i8 frame2 = { 0 };
v16i8 frame3 = { 0 };
v16i8 frame4 = { 0 };
v16u8 frame_l, frame_h;
v8i16 frame2_0_h, frame2_1_h, mod0_h, mod1_h;
v8i16 diff0, diff1, cnt0, cnt1;
v4i32 const3, const16;
v4i32 filter_wt, strength;
v4i32 mod0_w, mod1_w, mod2_w, mod3_w;
v4i32 diff0_r, diff0_l, diff1_r, diff1_l;
v4i32 frame2_0, frame2_1, frame2_2, frame2_3;
v4i32 acc0, acc1, acc2, acc3;
filter_wt = __msa_fill_w(filter_wt_in);
strength = __msa_fill_w(strength_in);
const3 = __msa_ldi_w(3);
const16 = __msa_ldi_w(16);
for (row = 2; row--;)
{
LD2(frame1_ptr, stride, f0, f1);
frame1_ptr += (2 * stride);
LD2(frame2_ptr, 8, f2, f3);
frame2_ptr += 16;
LD2(frame1_ptr, stride, f4, f5);
frame1_ptr += (2 * stride);
LD2(frame2_ptr, 8, f6, f7);
frame2_ptr += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
INSERT_D2_SB(f0, f1, frame1);
INSERT_D2_SB(f2, f3, frame2);
INSERT_D2_SB(f4, f5, frame3);
INSERT_D2_SB(f6, f7, frame4);
ILVRL_B2_UB(frame1, frame2, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame2, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
LD_SW2(acc, 4, acc0, acc1);
LD_SW2(acc + 8, 4, acc2, acc3);
LD_SH2(cnt, 8, cnt0, cnt1);
ILVRL_B2_UB(frame3, frame4, frame_l, frame_h);
HSUB_UB2_SH(frame_l, frame_h, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, const3, mod1_w, const3, mod2_w, const3, mod3_w, const3,
mod0_w, mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < const16);
diff0_l = (mod1_w < const16);
diff1_r = (mod2_w < const16);
diff1_l = (mod3_w < const16);
SUB4(const16, mod0_w, const16, mod1_w, const16, mod2_w, const16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
mod2_w = diff1_r & mod2_w;
mod3_w = diff1_l & mod3_w;
MUL4(mod0_w, filter_wt, mod1_w, filter_wt, mod2_w, filter_wt, mod3_w,
filter_wt, mod0_w, mod1_w, mod2_w, mod3_w);
PCKEV_H2_SH(mod1_w, mod0_w, mod3_w, mod2_w, mod0_h, mod1_h);
ADD2(mod0_h, cnt0, mod1_h, cnt1, mod0_h, mod1_h);
ST_SH2(mod0_h, mod1_h, cnt, 8);
cnt += 16;
UNPCK_UB_SH(frame4, frame2_0_h, frame2_1_h);
UNPCK_SH_SW(frame2_0_h, frame2_0, frame2_1);
UNPCK_SH_SW(frame2_1_h, frame2_2, frame2_3);
MUL4(mod0_w, frame2_0, mod1_w, frame2_1, mod2_w, frame2_2, mod3_w,
frame2_3, mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
ST_SW2(mod2_w, mod3_w, acc + 8, 4);
acc += 16;
}
}
static void hevc_addblk_32x32_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t loop_cnt;
uint8_t *temp_dst = dst;
v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
v8i16 dst_r0, dst_l0, dst_r1, dst_l1, dst_r2, dst_l2, dst_r3, dst_l3;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
for (loop_cnt = 14; loop_cnt--;) {
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
}
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
}
