int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y,
FLOAT c, FLOAT s)
{
BLASLONG i, j;
FLOAT *px, *py;
FLOAT tp0, tp1, tp2, tp3, tp4, tp5, tp6, tp7;
FLOAT fx0, fx1, fx2, fx3, fy0, fy1, fy2, fy3;
v2f64 x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, y6, y7;
v2f64 out0, out1, out2, out3, out4, out5, out6, out7;
v2f64 out8, out9, out10, out11, out12, out13, out14, out15, c0, s0;
if (n <= 0) return (0);
px = x;
py = y;
if ((1 == inc_x) && (1 == inc_y))
{
if ((0 == c) && (0 == s))
{
v2f64 zero = {0, 0};
zero = (v2f64) __msa_insert_d((v2i64) zero, 0, 0.0);
zero = (v2f64) __msa_insert_d((v2i64) zero, 1, 0.0);
/* process 4 floats */
for (j = (n >> 2); j--;)
{
ST_DP(zero, px);
ST_DP(zero, py);
px += 2;
py += 2;
ST_DP(zero, px);
ST_DP(zero, py);
px += 2;
py += 2;
}
if (n & 2)
{
ST_DP(zero, px);
ST_DP(zero, py);
px += 2;
py += 2;
}
if (n & 1)
{
px[0] = 0;
py[0] = 0;
}
}
static void intra_predict_tm_8x8_msa(const uint8_t *src_top_ptr,
const uint8_t *src_left,
uint8_t *dst, int32_t dst_stride) {
uint64_t val;
uint8_t top_left = src_top_ptr[-1];
uint32_t loop_cnt;
v16i8 src_left0, src_left1, src_left2, src_left3, tmp0, tmp1, src_top = { 0 };
v8u16 src_top_left, vec0, vec1, vec2, vec3;
v16u8 src0, src1, src2, src3;
val = LD(src_top_ptr);
src_top = (v16i8)__msa_insert_d((v2i64)src_top, 0, val);
src_top_left = (v8u16)__msa_fill_h(top_left);
for (loop_cnt = 2; loop_cnt--;) {
src_left0 = __msa_fill_b(src_left[0]);
src_left1 = __msa_fill_b(src_left[1]);
src_left2 = __msa_fill_b(src_left[2]);
src_left3 = __msa_fill_b(src_left[3]);
src_left += 4;
ILVR_B4_UB(src_left0, src_top, src_left1, src_top, src_left2, src_top,
src_left3, src_top, src0, src1, src2, src3);
HADD_UB4_UH(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec0, vec1);
IPRED_SUBS_UH2_UH(src_top_left, src_top_left, vec2, vec3);
SAT_UH4_UH(vec0, vec1, vec2, vec3, 7);
PCKEV_B2_SB(vec1, vec0, vec3, vec2, tmp0, tmp1);
ST8x4_UB(tmp0, tmp1, dst, dst_stride);
dst += (4 * dst_stride);
}
}
static WEBP_INLINE void VE4(uint8_t* dst, const uint8_t* top) { // vertical
const uint64_t val_m = LD(top - 1);
const v16u8 A = (v16u8)__msa_insert_d((v2i64)A, 0, val_m);
const v16u8 B = SLDI_UB(A, A, 1);
const v16u8 C = SLDI_UB(A, A, 2);
const v16u8 AC = __msa_ave_u_b(A, C);
const v16u8 B2 = __msa_ave_u_b(B, B);
const v16u8 R = __msa_aver_u_b(AC, B2);
const uint32_t out = __msa_copy_s_w((v4i32)R, 0);
SW4(out, out, out, out, dst, BPS);
}
static WEBP_INLINE void LD4(uint8_t* dst, const uint8_t* top) {
const uint64_t val_m = LD(top);
const v16u8 A = (v16u8)__msa_insert_d((v2i64)A, 0, val_m);
const v16u8 B = SLDI_UB(A, A, 1);
const v16u8 C1 = SLDI_UB(A, A, 2);
const v16u8 C = (v16u8)__msa_insert_b((v16i8)C1, 6, top[7]);
const v16u8 AC = __msa_ave_u_b(A, C);
const v16u8 B2 = __msa_ave_u_b(B, B);
const v16u8 R0 = __msa_aver_u_b(AC, B2);
const v16u8 R1 = SLDI_UB(R0, R0, 1);
const v16u8 R2 = SLDI_UB(R1, R1, 1);
const v16u8 R3 = SLDI_UB(R2, R2, 1);
const uint32_t val0 = __msa_copy_s_w((v4i32)R0, 0);
const uint32_t val1 = __msa_copy_s_w((v4i32)R1, 0);
const uint32_t val2 = __msa_copy_s_w((v4i32)R2, 0);
const uint32_t val3 = __msa_copy_s_w((v4i32)R3, 0);
SW4(val0, val1, val2, val3, dst, BPS);
}
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);
}
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;
}
}
