From 36d22d82aa202bb199967e9512281e9a53db42c9 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 7 Apr 2024 21:33:14 +0200 Subject: Adding upstream version 115.7.0esr. Signed-off-by: Daniel Baumann --- third_party/aom/aom_dsp/x86/quantize_sse2.c | 147 ++++++++++++++++++++++++++++ 1 file changed, 147 insertions(+) create mode 100644 third_party/aom/aom_dsp/x86/quantize_sse2.c (limited to 'third_party/aom/aom_dsp/x86/quantize_sse2.c') diff --git a/third_party/aom/aom_dsp/x86/quantize_sse2.c b/third_party/aom/aom_dsp/x86/quantize_sse2.c new file mode 100644 index 0000000000..d3de6e24db --- /dev/null +++ b/third_party/aom/aom_dsp/x86/quantize_sse2.c @@ -0,0 +1,147 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include +#include +#include + +#include "config/aom_dsp_rtcd.h" + +#include "aom/aom_integer.h" +#include "aom_dsp/x86/quantize_x86.h" + +static INLINE __m128i load_coefficients(const tran_low_t *coeff_ptr) { + assert(sizeof(tran_low_t) == 4); + + return _mm_setr_epi16((int16_t)coeff_ptr[0], (int16_t)coeff_ptr[1], + (int16_t)coeff_ptr[2], (int16_t)coeff_ptr[3], + (int16_t)coeff_ptr[4], (int16_t)coeff_ptr[5], + (int16_t)coeff_ptr[6], (int16_t)coeff_ptr[7]); +} + +static INLINE void store_coefficients(__m128i coeff_vals, + tran_low_t *coeff_ptr) { + assert(sizeof(tran_low_t) == 4); + + __m128i one = _mm_set1_epi16(1); + __m128i coeff_vals_hi = _mm_mulhi_epi16(coeff_vals, one); + __m128i coeff_vals_lo = _mm_mullo_epi16(coeff_vals, one); + __m128i coeff_vals_1 = _mm_unpacklo_epi16(coeff_vals_lo, coeff_vals_hi); + __m128i coeff_vals_2 = _mm_unpackhi_epi16(coeff_vals_lo, coeff_vals_hi); + _mm_store_si128((__m128i *)(coeff_ptr), coeff_vals_1); + _mm_store_si128((__m128i *)(coeff_ptr + 4), coeff_vals_2); +} + +void aom_quantize_b_sse2(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_ptr, + const int16_t *iscan_ptr) { + const __m128i zero = _mm_setzero_si128(); + int index = 16; + + __m128i zbin, round, quant, dequant, shift; + __m128i coeff0, coeff1, coeff0_sign, coeff1_sign; + __m128i qcoeff0, qcoeff1; + __m128i cmp_mask0, cmp_mask1; + __m128i eob, eob0; + + (void)scan_ptr; + + // Setup global values. + load_b_values(zbin_ptr, &zbin, round_ptr, &round, quant_ptr, &quant, + dequant_ptr, &dequant, quant_shift_ptr, &shift); + + // Do DC and first 15 AC. + coeff0 = load_coefficients(coeff_ptr); + coeff1 = load_coefficients(coeff_ptr + 8); + + // Poor man's abs(). + coeff0_sign = _mm_srai_epi16(coeff0, 15); + coeff1_sign = _mm_srai_epi16(coeff1, 15); + qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign); + qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign); + + cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin); + zbin = _mm_unpackhi_epi64(zbin, zbin); // Switch DC to AC + cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin); + + calculate_qcoeff(&qcoeff0, round, quant, shift); + + round = _mm_unpackhi_epi64(round, round); + quant = _mm_unpackhi_epi64(quant, quant); + shift = _mm_unpackhi_epi64(shift, shift); + + calculate_qcoeff(&qcoeff1, round, quant, shift); + + // Reinsert signs + qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign); + qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign); + + // Mask out zbin threshold coeffs + qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0); + qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1); + + store_coefficients(qcoeff0, qcoeff_ptr); + store_coefficients(qcoeff1, qcoeff_ptr + 8); + + coeff0 = calculate_dqcoeff(qcoeff0, dequant); + dequant = _mm_unpackhi_epi64(dequant, dequant); + coeff1 = calculate_dqcoeff(qcoeff1, dequant); + + store_coefficients(coeff0, dqcoeff_ptr); + store_coefficients(coeff1, dqcoeff_ptr + 8); + + eob = + scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, 0, zero); + + // AC only loop. + while (index < n_coeffs) { + coeff0 = load_coefficients(coeff_ptr + index); + coeff1 = load_coefficients(coeff_ptr + index + 8); + + coeff0_sign = _mm_srai_epi16(coeff0, 15); + coeff1_sign = _mm_srai_epi16(coeff1, 15); + qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign); + qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign); + + cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin); + cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin); + + calculate_qcoeff(&qcoeff0, round, quant, shift); + calculate_qcoeff(&qcoeff1, round, quant, shift); + + qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign); + qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign); + + qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0); + qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1); + + store_coefficients(qcoeff0, qcoeff_ptr + index); + store_coefficients(qcoeff1, qcoeff_ptr + index + 8); + + coeff0 = calculate_dqcoeff(qcoeff0, dequant); + coeff1 = calculate_dqcoeff(qcoeff1, dequant); + + store_coefficients(coeff0, dqcoeff_ptr + index); + store_coefficients(coeff1, dqcoeff_ptr + index + 8); + + eob0 = scan_for_eob(&coeff0, &coeff1, cmp_mask0, cmp_mask1, iscan_ptr, + index, zero); + eob = _mm_max_epi16(eob, eob0); + + index += 16; + } + + *eob_ptr = accumulate_eob(eob); +} -- cgit v1.2.3