mirror of
https://repo.dactyloidae.xyz/Dactyloidae/UXP.git
synced 2026-10-06 15:27:30 +09:00
Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918
This commit is contained in:
trav90
Roy Tam
parent
e329bc4331
commit
2ab8fcbfd1
370 changed files with 56185 additions and 32026 deletions
234
third_party/aom/aom_dsp/x86/inv_txfm_sse2.c
vendored
234
third_party/aom/aom_dsp/x86/inv_txfm_sse2.c
vendored
|
|
@ -3498,237 +3498,3 @@ void idct32_8col(__m128i *in0, __m128i *in1) {
|
|||
in1[14] = _mm_sub_epi16(stp1_1, stp1_30);
|
||||
in1[15] = _mm_sub_epi16(stp1_0, stp1_31);
|
||||
}
|
||||
|
||||
#if CONFIG_HIGHBITDEPTH
|
||||
static INLINE __m128i clamp_high_sse2(__m128i value, int bd) {
|
||||
__m128i ubounded, retval;
|
||||
const __m128i zero = _mm_set1_epi16(0);
|
||||
const __m128i one = _mm_set1_epi16(1);
|
||||
const __m128i max = _mm_subs_epi16(_mm_slli_epi16(one, bd), one);
|
||||
ubounded = _mm_cmpgt_epi16(value, max);
|
||||
retval = _mm_andnot_si128(ubounded, value);
|
||||
ubounded = _mm_and_si128(ubounded, max);
|
||||
retval = _mm_or_si128(retval, ubounded);
|
||||
retval = _mm_and_si128(retval, _mm_cmpgt_epi16(retval, zero));
|
||||
return retval;
|
||||
}
|
||||
|
||||
void aom_highbd_idct4x4_16_add_sse2(const tran_low_t *input, uint8_t *dest8,
|
||||
int stride, int bd) {
|
||||
tran_low_t out[4 * 4];
|
||||
tran_low_t *outptr = out;
|
||||
int i, j;
|
||||
__m128i inptr[4];
|
||||
__m128i sign_bits[2];
|
||||
__m128i temp_mm, min_input, max_input;
|
||||
int test;
|
||||
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
|
||||
int optimised_cols = 0;
|
||||
const __m128i zero = _mm_set1_epi16(0);
|
||||
const __m128i eight = _mm_set1_epi16(8);
|
||||
const __m128i max = _mm_set1_epi16(12043);
|
||||
const __m128i min = _mm_set1_epi16(-12043);
|
||||
// Load input into __m128i
|
||||
inptr[0] = _mm_loadu_si128((const __m128i *)input);
|
||||
inptr[1] = _mm_loadu_si128((const __m128i *)(input + 4));
|
||||
inptr[2] = _mm_loadu_si128((const __m128i *)(input + 8));
|
||||
inptr[3] = _mm_loadu_si128((const __m128i *)(input + 12));
|
||||
|
||||
// Pack to 16 bits
|
||||
inptr[0] = _mm_packs_epi32(inptr[0], inptr[1]);
|
||||
inptr[1] = _mm_packs_epi32(inptr[2], inptr[3]);
|
||||
|
||||
max_input = _mm_max_epi16(inptr[0], inptr[1]);
|
||||
min_input = _mm_min_epi16(inptr[0], inptr[1]);
|
||||
max_input = _mm_cmpgt_epi16(max_input, max);
|
||||
min_input = _mm_cmplt_epi16(min_input, min);
|
||||
temp_mm = _mm_or_si128(max_input, min_input);
|
||||
test = _mm_movemask_epi8(temp_mm);
|
||||
|
||||
if (!test) {
|
||||
// Do the row transform
|
||||
aom_idct4_sse2(inptr);
|
||||
|
||||
// Check the min & max values
|
||||
max_input = _mm_max_epi16(inptr[0], inptr[1]);
|
||||
min_input = _mm_min_epi16(inptr[0], inptr[1]);
|
||||
max_input = _mm_cmpgt_epi16(max_input, max);
|
||||
min_input = _mm_cmplt_epi16(min_input, min);
|
||||
temp_mm = _mm_or_si128(max_input, min_input);
|
||||
test = _mm_movemask_epi8(temp_mm);
|
||||
|
||||
if (test) {
|
||||
array_transpose_4x4(inptr);
|
||||
sign_bits[0] = _mm_cmplt_epi16(inptr[0], zero);
|
||||
sign_bits[1] = _mm_cmplt_epi16(inptr[1], zero);
|
||||
inptr[3] = _mm_unpackhi_epi16(inptr[1], sign_bits[1]);
|
||||
inptr[2] = _mm_unpacklo_epi16(inptr[1], sign_bits[1]);
|
||||
inptr[1] = _mm_unpackhi_epi16(inptr[0], sign_bits[0]);
|
||||
inptr[0] = _mm_unpacklo_epi16(inptr[0], sign_bits[0]);
|
||||
_mm_storeu_si128((__m128i *)outptr, inptr[0]);
|
||||
_mm_storeu_si128((__m128i *)(outptr + 4), inptr[1]);
|
||||
_mm_storeu_si128((__m128i *)(outptr + 8), inptr[2]);
|
||||
_mm_storeu_si128((__m128i *)(outptr + 12), inptr[3]);
|
||||
} else {
|
||||
// Set to use the optimised transform for the column
|
||||
optimised_cols = 1;
|
||||
}
|
||||
} else {
|
||||
// Run the un-optimised row transform
|
||||
for (i = 0; i < 4; ++i) {
|
||||
aom_highbd_idct4_c(input, outptr, bd);
|
||||
input += 4;
|
||||
outptr += 4;
|
||||
}
|
||||
}
|
||||
|
||||
if (optimised_cols) {
|
||||
aom_idct4_sse2(inptr);
|
||||
|
||||
// Final round and shift
|
||||
inptr[0] = _mm_add_epi16(inptr[0], eight);
|
||||
inptr[1] = _mm_add_epi16(inptr[1], eight);
|
||||
|
||||
inptr[0] = _mm_srai_epi16(inptr[0], 4);
|
||||
inptr[1] = _mm_srai_epi16(inptr[1], 4);
|
||||
|
||||
// Reconstruction and Store
|
||||
{
|
||||
__m128i d0 = _mm_loadl_epi64((const __m128i *)dest);
|
||||
__m128i d2 = _mm_loadl_epi64((const __m128i *)(dest + stride * 2));
|
||||
d0 = _mm_unpacklo_epi64(
|
||||
d0, _mm_loadl_epi64((const __m128i *)(dest + stride)));
|
||||
d2 = _mm_unpacklo_epi64(
|
||||
d2, _mm_loadl_epi64((const __m128i *)(dest + stride * 3)));
|
||||
d0 = clamp_high_sse2(_mm_adds_epi16(d0, inptr[0]), bd);
|
||||
d2 = clamp_high_sse2(_mm_adds_epi16(d2, inptr[1]), bd);
|
||||
// store input0
|
||||
_mm_storel_epi64((__m128i *)dest, d0);
|
||||
// store input1
|
||||
d0 = _mm_srli_si128(d0, 8);
|
||||
_mm_storel_epi64((__m128i *)(dest + stride), d0);
|
||||
// store input2
|
||||
_mm_storel_epi64((__m128i *)(dest + stride * 2), d2);
|
||||
// store input3
|
||||
d2 = _mm_srli_si128(d2, 8);
|
||||
_mm_storel_epi64((__m128i *)(dest + stride * 3), d2);
|
||||
}
|
||||
} else {
|
||||
// Run the un-optimised column transform
|
||||
tran_low_t temp_in[4], temp_out[4];
|
||||
// Columns
|
||||
for (i = 0; i < 4; ++i) {
|
||||
for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i];
|
||||
aom_highbd_idct4_c(temp_in, temp_out, bd);
|
||||
for (j = 0; j < 4; ++j) {
|
||||
dest[j * stride + i] = highbd_clip_pixel_add(
|
||||
dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 4), bd);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void aom_highbd_idct8x8_10_add_sse2(const tran_low_t *input, uint8_t *dest8,
|
||||
int stride, int bd) {
|
||||
tran_low_t out[8 * 8] = { 0 };
|
||||
tran_low_t *outptr = out;
|
||||
int i, j, test;
|
||||
__m128i inptr[8];
|
||||
__m128i min_input, max_input, temp1, temp2, sign_bits;
|
||||
uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
|
||||
const __m128i zero = _mm_set1_epi16(0);
|
||||
const __m128i sixteen = _mm_set1_epi16(16);
|
||||
const __m128i max = _mm_set1_epi16(6201);
|
||||
const __m128i min = _mm_set1_epi16(-6201);
|
||||
int optimised_cols = 0;
|
||||
|
||||
// Load input into __m128i & pack to 16 bits
|
||||
for (i = 0; i < 8; i++) {
|
||||
temp1 = _mm_loadu_si128((const __m128i *)(input + 8 * i));
|
||||
temp2 = _mm_loadu_si128((const __m128i *)(input + 8 * i + 4));
|
||||
inptr[i] = _mm_packs_epi32(temp1, temp2);
|
||||
}
|
||||
|
||||
// Find the min & max for the row transform
|
||||
// only first 4 row has non-zero coefs
|
||||
max_input = _mm_max_epi16(inptr[0], inptr[1]);
|
||||
min_input = _mm_min_epi16(inptr[0], inptr[1]);
|
||||
for (i = 2; i < 4; i++) {
|
||||
max_input = _mm_max_epi16(max_input, inptr[i]);
|
||||
min_input = _mm_min_epi16(min_input, inptr[i]);
|
||||
}
|
||||
max_input = _mm_cmpgt_epi16(max_input, max);
|
||||
min_input = _mm_cmplt_epi16(min_input, min);
|
||||
temp1 = _mm_or_si128(max_input, min_input);
|
||||
test = _mm_movemask_epi8(temp1);
|
||||
|
||||
if (!test) {
|
||||
// Do the row transform
|
||||
aom_idct8_sse2(inptr);
|
||||
|
||||
// Find the min & max for the column transform
|
||||
// N.B. Only first 4 cols contain non-zero coeffs
|
||||
max_input = _mm_max_epi16(inptr[0], inptr[1]);
|
||||
min_input = _mm_min_epi16(inptr[0], inptr[1]);
|
||||
for (i = 2; i < 8; i++) {
|
||||
max_input = _mm_max_epi16(max_input, inptr[i]);
|
||||
min_input = _mm_min_epi16(min_input, inptr[i]);
|
||||
}
|
||||
max_input = _mm_cmpgt_epi16(max_input, max);
|
||||
min_input = _mm_cmplt_epi16(min_input, min);
|
||||
temp1 = _mm_or_si128(max_input, min_input);
|
||||
test = _mm_movemask_epi8(temp1);
|
||||
|
||||
if (test) {
|
||||
// Use fact only first 4 rows contain non-zero coeffs
|
||||
array_transpose_4X8(inptr, inptr);
|
||||
for (i = 0; i < 4; i++) {
|
||||
sign_bits = _mm_cmplt_epi16(inptr[i], zero);
|
||||
temp1 = _mm_unpackhi_epi16(inptr[i], sign_bits);
|
||||
temp2 = _mm_unpacklo_epi16(inptr[i], sign_bits);
|
||||
_mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i + 1)), temp1);
|
||||
_mm_storeu_si128((__m128i *)(outptr + 4 * (2 * i)), temp2);
|
||||
}
|
||||
} else {
|
||||
// Set to use the optimised transform for the column
|
||||
optimised_cols = 1;
|
||||
}
|
||||
} else {
|
||||
// Run the un-optimised row transform
|
||||
for (i = 0; i < 4; ++i) {
|
||||
aom_highbd_idct8_c(input, outptr, bd);
|
||||
input += 8;
|
||||
outptr += 8;
|
||||
}
|
||||
}
|
||||
|
||||
if (optimised_cols) {
|
||||
aom_idct8_sse2(inptr);
|
||||
|
||||
// Final round & shift and Reconstruction and Store
|
||||
{
|
||||
__m128i d[8];
|
||||
for (i = 0; i < 8; i++) {
|
||||
inptr[i] = _mm_add_epi16(inptr[i], sixteen);
|
||||
d[i] = _mm_loadu_si128((const __m128i *)(dest + stride * i));
|
||||
inptr[i] = _mm_srai_epi16(inptr[i], 5);
|
||||
d[i] = clamp_high_sse2(_mm_adds_epi16(d[i], inptr[i]), bd);
|
||||
// Store
|
||||
_mm_storeu_si128((__m128i *)(dest + stride * i), d[i]);
|
||||
}
|
||||
}
|
||||
} else {
|
||||
// Run the un-optimised column transform
|
||||
tran_low_t temp_in[8], temp_out[8];
|
||||
for (i = 0; i < 8; ++i) {
|
||||
for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
|
||||
aom_highbd_idct8_c(temp_in, temp_out, bd);
|
||||
for (j = 0; j < 8; ++j) {
|
||||
dest[j * stride + i] = highbd_clip_pixel_add(
|
||||
dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif // CONFIG_HIGHBITDEPTH
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue