void vpx_fdct4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, src_stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 vec, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
vec = __msa_ceqi_h(in0, 0);
vec = vec ^ 255;
vec = mask & vec;
in0 += vec;
}
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
void vp9_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 temp, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
temp = __msa_ceqi_h(in0, 0);
temp = (v8i16)__msa_xori_b((v16u8)temp, 255);
temp = mask & temp;
in0 += temp;
}
switch (tx_type) {
case DCT_DCT:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_DCT:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case DCT_ADST:
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
case ADST_ADST:
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
}
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
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;
}
}
SLDI_B2_SH(hz_out2, hz_out4, hz_out0, hz_out2, hz_out1, hz_out3, 8);
filt = LD_SH(filter_vert);
SPLATI_H4_SH(filt, 0, 1, 2, 3, filt_vt0, filt_vt1, filt_vt2, filt_vt3);
ILVEV_B2_SH(hz_out0, hz_out1, hz_out2, hz_out3, out0, out1);
out2 = (v8i16)__msa_ilvev_b((v16i8)hz_out5, (v16i8)hz_out4);
for (loop_cnt = (height >> 2); loop_cnt--;) {
LD_SB4(src, src_stride, src7, src8, src9, src10);
XORI_B4_128_SB(src7, src8, src9, src10);
src += (4 * src_stride);
hz_out7 = HORIZ_8TAP_FILT(src7, src8, mask0, mask1, mask2, mask3,
filt_hz0, filt_hz1, filt_hz2, filt_hz3);
hz_out6 = (v8i16)__msa_sldi_b((v16i8)hz_out7, (v16i8)hz_out5, 8);
out3 = (v8i16)__msa_ilvev_b((v16i8)hz_out7, (v16i8)hz_out6);
tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vt0, filt_vt1,
filt_vt2, filt_vt3);
hz_out9 = HORIZ_8TAP_FILT(src9, src10, mask0, mask1, mask2, mask3,
filt_hz0, filt_hz1, filt_hz2, filt_hz3);
hz_out8 = (v8i16)__msa_sldi_b((v16i8)hz_out9, (v16i8)hz_out7, 8);
out4 = (v8i16)__msa_ilvev_b((v16i8)hz_out9, (v16i8)hz_out8);
tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vt0, filt_vt1,
filt_vt2, filt_vt3);
SRARI_H2_SH(tmp0, tmp1, FILTER_BITS);
SAT_SH2_SH(tmp0, tmp1, 7);
out = PCKEV_XORI128_UB(tmp0, tmp1);
ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride);
dst += (4 * dst_stride);
