void av1_highbd_quantize_fp_avx2(
const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan, int log_scale) {
(void)scan;
(void)zbin_ptr;
(void)quant_shift_ptr;
const unsigned int step = 8;
__m256i qp[3], coeff;
init_qp(round_ptr, quant_ptr, dequant_ptr, log_scale, qp);
coeff = _mm256_loadu_si256((const __m256i *)coeff_ptr);
__m256i eob = _mm256_setzero_si256();
quantize(qp, &coeff, iscan, log_scale, qcoeff_ptr, dqcoeff_ptr, &eob);
coeff_ptr += step;
qcoeff_ptr += step;
dqcoeff_ptr += step;
iscan += step;
n_coeffs -= step;
update_qp(qp);
while (n_coeffs > 0) {
coeff = _mm256_loadu_si256((const __m256i *)coeff_ptr);
quantize(qp, &coeff, iscan, log_scale, qcoeff_ptr, dqcoeff_ptr, &eob);
coeff_ptr += step;
qcoeff_ptr += step;
dqcoeff_ptr += step;
iscan += step;
n_coeffs -= step;
}
{
__m256i eob_s;
eob_s = _mm256_shuffle_epi32(eob, 0xe);
eob = _mm256_max_epi16(eob, eob_s);
eob_s = _mm256_shufflelo_epi16(eob, 0xe);
eob = _mm256_max_epi16(eob, eob_s);
eob_s = _mm256_shufflelo_epi16(eob, 1);
eob = _mm256_max_epi16(eob, eob_s);
const __m128i final_eob = _mm_max_epi16(_mm256_castsi256_si128(eob),
_mm256_extractf128_si256(eob, 1));
*eob_ptr = _mm_extract_epi16(final_eob, 0);
}
}
static void satd_8bit_4x4_dual_avx2(
const pred_buffer preds, const kvz_pixel * const orig, unsigned num_modes, unsigned *satds_out)
{
__m256i original = _mm256_broadcastsi128_si256(_mm_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)orig)));
__m256i pred = _mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)preds[0]));
pred = _mm256_inserti128_si256(pred, _mm_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)preds[1])), 1);
__m256i diff_lo = _mm256_sub_epi16(pred, original);
original = _mm256_broadcastsi128_si256(_mm_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)(orig + 8))));
pred = _mm256_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)(preds[0] + 8)));
pred = _mm256_inserti128_si256(pred, _mm_cvtepu8_epi16(_mm_loadl_epi64((__m128i*)(preds[1] + 8))), 1);
__m256i diff_hi = _mm256_sub_epi16(pred, original);
//Hor
__m256i row0 = _mm256_hadd_epi16(diff_lo, diff_hi);
__m256i row1 = _mm256_hsub_epi16(diff_lo, diff_hi);
__m256i row2 = _mm256_hadd_epi16(row0, row1);
__m256i row3 = _mm256_hsub_epi16(row0, row1);
//Ver
row0 = _mm256_hadd_epi16(row2, row3);
row1 = _mm256_hsub_epi16(row2, row3);
row2 = _mm256_hadd_epi16(row0, row1);
row3 = _mm256_hsub_epi16(row0, row1);
//Abs and sum
row2 = _mm256_abs_epi16(row2);
row3 = _mm256_abs_epi16(row3);
row3 = _mm256_add_epi16(row2, row3);
row3 = _mm256_add_epi16(row3, _mm256_shuffle_epi32(row3, KVZ_PERMUTE(2, 3, 0, 1) ));
row3 = _mm256_add_epi16(row3, _mm256_shuffle_epi32(row3, KVZ_PERMUTE(1, 0, 1, 0) ));
row3 = _mm256_add_epi16(row3, _mm256_shufflelo_epi16(row3, KVZ_PERMUTE(1, 0, 1, 0) ));
unsigned sum1 = _mm_extract_epi16(_mm256_castsi256_si128(row3), 0);
sum1 = (sum1 + 1) >> 1;
unsigned sum2 = _mm_extract_epi16(_mm256_extracti128_si256(row3, 1), 0);
sum2 = (sum2 + 1) >> 1;
satds_out[0] = sum1;
satds_out[1] = sum2;
}
void
test8bit (void)
{
l1 = _mm256_mpsadbw_epu8 (l2, l3, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_alignr_epi8 (l2, l3, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_blend_epi32 (i1, i1, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_blend_epi32 (l2, l3, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_blend_epi16(l2, l3, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_permute2x128_si256 (l2, l3, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
e1 = _mm256_permute4x64_pd (e2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_permute4x64_epi64 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_shuffle_epi32 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_shufflehi_epi16 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_shufflelo_epi16 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_slli_si256 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_srli_si256 (l2, 256); /* { dg-error "the last argument must be an 8-bit immediate" } */
}
static INLINE void hor_transform_row_dual_avx2(__m256i* row){
__m256i mask_pos = _mm256_set1_epi16(1);
__m256i mask_neg = _mm256_set1_epi16(-1);
__m256i sign_mask = _mm256_unpacklo_epi64(mask_pos, mask_neg);
__m256i temp = _mm256_shuffle_epi32(*row, KVZ_PERMUTE(2, 3, 0, 1));
*row = _mm256_sign_epi16(*row, sign_mask);
*row = _mm256_add_epi16(*row, temp);
sign_mask = _mm256_unpacklo_epi32(mask_pos, mask_neg);
temp = _mm256_shuffle_epi32(*row, KVZ_PERMUTE(1, 0, 3, 2));
*row = _mm256_sign_epi16(*row, sign_mask);
*row = _mm256_add_epi16(*row, temp);
sign_mask = _mm256_unpacklo_epi16(mask_pos, mask_neg);
temp = _mm256_shufflelo_epi16(*row, KVZ_PERMUTE(1,0,3,2));
temp = _mm256_shufflehi_epi16(temp, KVZ_PERMUTE(1,0,3,2));
*row = _mm256_sign_epi16(*row, sign_mask);
*row = _mm256_add_epi16(*row, temp);
}
__m256i test_mm256_shufflelo_epi16(__m256i a) {
// CHECK: shufflevector <16 x i16> %{{.*}}, <16 x i16> undef, <16 x i32> <i32 3, i32 0, i32 1, i32 1, i32 4, i32 5, i32 6, i32 7, i32 11, i32 8, i32 9, i32 9, i32 12, i32 13, i32 14, i32 15>
return _mm256_shufflelo_epi16(a, 83);
}