Port folder renaming changes from AOM
Manually cherry-picked commits:
ceef058 libvpx->libaom part2
3d26d91 libvpx -> libaom
cfea7dd vp10/ -> av1/
3a8eff7 Fix a build issue for a test
bf4202e Rename vpx to aom
Change-Id: I1b0eb5a40796e3aaf41c58984b4229a439a597dc
diff --git a/aom_dsp/quantize.c b/aom_dsp/quantize.c
new file mode 100644
index 0000000..c901aa0
--- /dev/null
+++ b/aom_dsp/quantize.c
@@ -0,0 +1,686 @@
+/*
+ * 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 "aom_dsp/quantize.h"
+#include "aom_mem/vpx_mem.h"
+
+#if CONFIG_AOM_QM
+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 qm_val_t *qm_ptr, const qm_val_t *iqm_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;
+ int64_t tmp, eob = -1;
+ int32_t tmp32;
+ int dequant =
+ (dequant_ptr * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+
+ 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);
+ tmp32 = (int32_t)((tmp * qm_ptr[rc] * quant) >> (16 + AOM_QM_BITS));
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
+ if (tmp32) 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, const qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr) {
+ int eob = -1;
+ int dequant =
+ (dequant_ptr * iqm_ptr[0] + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+
+ 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 uint32_t abs_qcoeff =
+ (uint32_t)((tmp * qm_ptr[0] * quant) >> (16 + AOM_QM_BITS));
+ qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dqcoeff_ptr[0] = qcoeff_ptr[0] * dequant;
+ 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 qm_val_t *qm_ptr, const qm_val_t *iqm_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;
+ int64_t tmp, eob = -1;
+ int32_t tmp32;
+ int dequant;
+
+ 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);
+ tmp32 = (int32_t)((tmp * qm_ptr[rc] * quant) >> (15 + AOM_QM_BITS));
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dequant =
+ (dequant_ptr * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+ dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / 2;
+ if (tmp32) 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 qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr) {
+ const int n_coeffs = 1024;
+ int eob = -1;
+ int dequant;
+
+ 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 uint32_t abs_qcoeff =
+ (uint32_t)((tmp * qm_ptr[0] * quant) >> (15 + AOM_QM_BITS));
+ qcoeff_ptr[0] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dequant =
+ (dequant_ptr * iqm_ptr[0] + (1 << (AOM_QM_BITS - 1))) >> AOM_QM_BITS;
+ dqcoeff_ptr[0] = (qcoeff_ptr[0] * dequant) / 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, const qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr) {
+ 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 qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc] * wt;
+
+ if (coeff < (zbins[rc != 0] << AOM_QM_BITS) &&
+ coeff > (nzbins[rc != 0] << AOM_QM_BITS))
+ 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 qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc];
+ const int coeff_sign = (coeff >> 31);
+ const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+ int dequant;
+
+ if (abs_coeff * wt >= (zbins[rc != 0] << AOM_QM_BITS)) {
+ int32_t tmp32;
+ int64_t tmp =
+ clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
+ tmp = tmp * wt;
+ tmp32 = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+ quant_shift_ptr[rc != 0]) >>
+ (16 + AOM_QM_BITS); // quantization
+ dequant =
+ (dequant_ptr[rc != 0] * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >>
+ AOM_QM_BITS;
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
+
+ if (tmp32) 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,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr) {
+ 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 };
+ int dequant;
+ (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 qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc] * wt;
+
+ if (coeff < (zbins[rc != 0] << AOM_QM_BITS) &&
+ coeff > (nzbins[rc != 0] << AOM_QM_BITS))
+ 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 qm_val_t wt = qm_ptr[rc];
+ const int coeff_sign = (coeff >> 31);
+ const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+
+ if (abs_coeff * wt >= (zbins[rc != 0] << AOM_QM_BITS)) {
+ const int64_t tmp1 = abs_coeff + round_ptr[rc != 0];
+ const int64_t tmpw = tmp1 * wt;
+ const int64_t tmp2 = ((tmpw * quant_ptr[rc != 0]) >> 16) + tmpw;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> (16 + AOM_QM_BITS));
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dequant =
+ (dequant_ptr[rc != 0] * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >>
+ AOM_QM_BITS;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
+ 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 qm_val_t *qm_ptr, const qm_val_t *iqm_ptr) {
+ 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;
+ int dequant;
+ (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 qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc] * wt;
+
+ // If the coefficient is out of the base ZBIN range, keep it for
+ // quantization.
+ if (coeff >= (zbins[rc != 0] << AOM_QM_BITS) ||
+ coeff <= (nzbins[rc != 0] << AOM_QM_BITS))
+ 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 qm_val_t wt = qm_ptr[rc];
+ int64_t tmp;
+ int tmp32;
+ int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+ abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
+ tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
+ tmp = tmp * wt;
+ tmp32 = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+ quant_shift_ptr[rc != 0]) >>
+ (15 + AOM_QM_BITS);
+
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dequant =
+ (dequant_ptr[rc != 0] * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >>
+ AOM_QM_BITS;
+ dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / 2;
+
+ if (tmp32) 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 qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr) {
+ 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;
+ int dequant;
+ (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 qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc] * wt;
+
+ // If the coefficient is out of the base ZBIN range, keep it for
+ // quantization.
+ if (coeff >= (zbins[rc != 0] << AOM_QM_BITS) ||
+ coeff <= (nzbins[rc != 0] << AOM_QM_BITS))
+ 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 qm_val_t wt = qm_ptr[rc];
+ 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 tmpw = tmp1 * wt;
+ const int64_t tmp2 = ((tmpw * quant_ptr[rc != 0]) >> 16) + tmpw;
+ const uint32_t abs_qcoeff =
+ (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> (15 + AOM_QM_BITS));
+ qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
+ dequant =
+ (dequant_ptr[rc != 0] * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >>
+ AOM_QM_BITS;
+ dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / 2;
+ if (abs_qcoeff) eob = idx_arr[i];
+ }
+ }
+ *eob_ptr = eob + 1;
+}
+#endif
+#else
+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);
+ 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
+#endif
