77ed4414d6
The result of the xor operation is unsigned. If coeff was negative, this results in an unsigned value - INT_MIN. Change-Id: I1f1edeaa6de1f4c68b848e8a82a666d390b749f0
323 lines
12 KiB
C
323 lines
12 KiB
C
/*
|
|
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license
|
|
* that can be found in the LICENSE file in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#include "./vpx_dsp_rtcd.h"
|
|
#include "vpx_dsp/quantize.h"
|
|
#include "vpx_mem/vpx_mem.h"
|
|
|
|
void vpx_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs, int skip_block,
|
|
const int16_t *round_ptr, const int16_t quant,
|
|
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
|
|
const int16_t dequant_ptr, uint16_t *eob_ptr) {
|
|
const int rc = 0;
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
int tmp, eob = -1;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
|
|
tmp = (tmp * quant) >> 16;
|
|
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr;
|
|
if (tmp) eob = 0;
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
|
|
#if CONFIG_VP9_HIGHBITDEPTH
|
|
void vpx_highbd_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
|
|
int skip_block, const int16_t *round_ptr,
|
|
const int16_t quant, tran_low_t *qcoeff_ptr,
|
|
tran_low_t *dqcoeff_ptr, const int16_t dequant_ptr,
|
|
uint16_t *eob_ptr) {
|
|
int eob = -1;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
const int coeff = coeff_ptr[0];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
const int64_t tmp = abs_coeff + round_ptr[0];
|
|
const int abs_qcoeff = (int)((tmp * quant) >> 16);
|
|
qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
|
|
dqcoeff_ptr[0] = qcoeff_ptr[0] * dequant_ptr;
|
|
if (abs_qcoeff) eob = 0;
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
#endif
|
|
|
|
void vpx_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
|
|
const int16_t *round_ptr, const int16_t quant,
|
|
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
|
|
const int16_t dequant_ptr, uint16_t *eob_ptr) {
|
|
const int n_coeffs = 1024;
|
|
const int rc = 0;
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
int tmp, eob = -1;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
tmp = clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1),
|
|
INT16_MIN, INT16_MAX);
|
|
tmp = (tmp * quant) >> 15;
|
|
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr / 2;
|
|
if (tmp) eob = 0;
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
|
|
#if CONFIG_VP9_HIGHBITDEPTH
|
|
void vpx_highbd_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
|
|
const int16_t *round_ptr, const int16_t quant,
|
|
tran_low_t *qcoeff_ptr,
|
|
tran_low_t *dqcoeff_ptr,
|
|
const int16_t dequant_ptr,
|
|
uint16_t *eob_ptr) {
|
|
const int n_coeffs = 1024;
|
|
int eob = -1;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
const int coeff = coeff_ptr[0];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
const int64_t tmp = abs_coeff + ROUND_POWER_OF_TWO(round_ptr[0], 1);
|
|
const int abs_qcoeff = (int)((tmp * quant) >> 15);
|
|
qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
|
|
dqcoeff_ptr[0] = qcoeff_ptr[0] * dequant_ptr / 2;
|
|
if (abs_qcoeff) eob = 0;
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
#endif
|
|
|
|
void vpx_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
|
|
int skip_block, 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 i, non_zero_count = (int)n_coeffs, eob = -1;
|
|
const int zbins[2] = { zbin_ptr[0], zbin_ptr[1] };
|
|
const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
|
|
(void)iscan;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
// Pre-scan pass
|
|
for (i = (int)n_coeffs - 1; i >= 0; i--) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
|
|
if (coeff < zbins[rc != 0] && coeff > nzbins[rc != 0])
|
|
non_zero_count--;
|
|
else
|
|
break;
|
|
}
|
|
|
|
// Quantization pass: All coefficients with index >= zero_flag are
|
|
// skippable. Note: zero_flag can be zero.
|
|
for (i = 0; i < non_zero_count; i++) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
|
|
if (abs_coeff >= zbins[rc != 0]) {
|
|
int tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
|
|
tmp = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
|
|
quant_shift_ptr[rc != 0]) >>
|
|
16; // quantization
|
|
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
|
|
|
|
if (tmp) eob = i;
|
|
}
|
|
}
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
|
|
#if CONFIG_VP9_HIGHBITDEPTH
|
|
void vpx_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
|
|
int skip_block, 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 i, non_zero_count = (int)n_coeffs, eob = -1;
|
|
const int zbins[2] = { zbin_ptr[0], zbin_ptr[1] };
|
|
const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
|
|
(void)iscan;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
// Pre-scan pass
|
|
for (i = (int)n_coeffs - 1; i >= 0; i--) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
|
|
if (coeff < zbins[rc != 0] && coeff > nzbins[rc != 0])
|
|
non_zero_count--;
|
|
else
|
|
break;
|
|
}
|
|
|
|
// Quantization pass: All coefficients with index >= zero_flag are
|
|
// skippable. Note: zero_flag can be zero.
|
|
for (i = 0; i < non_zero_count; i++) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
|
|
if (abs_coeff >= zbins[rc != 0]) {
|
|
const int64_t tmp1 = abs_coeff + round_ptr[rc != 0];
|
|
const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
|
|
const uint32_t abs_qcoeff =
|
|
(uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 16);
|
|
// Restoring the sign triggers unsigned overflow warnings with negative
|
|
// values because the result of the xor operation is unsigned.
|
|
qcoeff_ptr[rc] = (tran_low_t)(abs_qcoeff ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
|
|
if (abs_qcoeff) eob = i;
|
|
}
|
|
}
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
#endif
|
|
|
|
void vpx_quantize_b_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
|
|
int skip_block, 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) {
|
|
const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0], 1),
|
|
ROUND_POWER_OF_TWO(zbin_ptr[1], 1) };
|
|
const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
|
|
|
|
int idx = 0;
|
|
int idx_arr[1024];
|
|
int i, eob = -1;
|
|
(void)iscan;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
// Pre-scan pass
|
|
for (i = 0; i < n_coeffs; i++) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
|
|
// If the coefficient is out of the base ZBIN range, keep it for
|
|
// quantization.
|
|
if (coeff >= zbins[rc != 0] || coeff <= nzbins[rc != 0])
|
|
idx_arr[idx++] = i;
|
|
}
|
|
|
|
// Quantization pass: only process the coefficients selected in
|
|
// pre-scan pass. Note: idx can be zero.
|
|
for (i = 0; i < idx; i++) {
|
|
const int rc = scan[idx_arr[i]];
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
int tmp;
|
|
int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
|
|
abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
|
|
tmp = ((((abs_coeff * quant_ptr[rc != 0]) >> 16) + abs_coeff) *
|
|
quant_shift_ptr[rc != 0]) >>
|
|
15;
|
|
|
|
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
|
|
|
|
if (tmp) eob = idx_arr[i];
|
|
}
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
|
|
#if CONFIG_VP9_HIGHBITDEPTH
|
|
void vpx_highbd_quantize_b_32x32_c(
|
|
const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block,
|
|
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) {
|
|
const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0], 1),
|
|
ROUND_POWER_OF_TWO(zbin_ptr[1], 1) };
|
|
const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
|
|
|
|
int idx = 0;
|
|
int idx_arr[1024];
|
|
int i, eob = -1;
|
|
(void)iscan;
|
|
|
|
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
|
|
memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
|
|
|
|
if (!skip_block) {
|
|
// Pre-scan pass
|
|
for (i = 0; i < n_coeffs; i++) {
|
|
const int rc = scan[i];
|
|
const int coeff = coeff_ptr[rc];
|
|
|
|
// If the coefficient is out of the base ZBIN range, keep it for
|
|
// quantization.
|
|
if (coeff >= zbins[rc != 0] || coeff <= nzbins[rc != 0])
|
|
idx_arr[idx++] = i;
|
|
}
|
|
|
|
// Quantization pass: only process the coefficients selected in
|
|
// pre-scan pass. Note: idx can be zero.
|
|
for (i = 0; i < idx; i++) {
|
|
const int rc = scan[idx_arr[i]];
|
|
const int coeff = coeff_ptr[rc];
|
|
const int coeff_sign = (coeff >> 31);
|
|
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
|
|
const int64_t tmp1 =
|
|
abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
|
|
const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
|
|
const uint32_t abs_qcoeff =
|
|
(uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> 15);
|
|
qcoeff_ptr[rc] = (tran_low_t)(abs_qcoeff ^ coeff_sign) - coeff_sign;
|
|
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
|
|
if (abs_qcoeff) eob = idx_arr[i];
|
|
}
|
|
}
|
|
*eob_ptr = eob + 1;
|
|
}
|
|
#endif
|