Refactor aom_quantize_b
Use one helper function to remove copy paste of
aom_quantize_b_c, aom_quantize_b_32x32_c, aom_quantize_b_64x64_c.
And remove redundant av1_quantize_b_c, av1_highbd_quantize_b_c.
BUG=aomedia:442
Change-Id: I3e7923a64a16af223c3e2a42dc27c0f28a9b134b
diff --git a/aom_dsp/quantize.c b/aom_dsp/quantize.c
index 1f8a4e5..3e541a3 100644
--- a/aom_dsp/quantize.c
+++ b/aom_dsp/quantize.c
@@ -195,67 +195,6 @@
#endif // CONFIG_TX64X64
#endif
-void aom_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_HIGHBITDEPTH
void aom_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
@@ -316,136 +255,6 @@
}
#endif
-void aom_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_TX64X64
-void aom_quantize_b_64x64_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], 2),
- ROUND_POWER_OF_TWO(zbin_ptr[1], 2) };
- const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
-
- int idx = 0;
- int idx_arr[4096];
- 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], 2);
- tmp = clamp(abs_coeff, INT16_MIN, INT16_MAX);
- tmp = tmp * wt;
- tmp32 = ((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
- quant_shift_ptr[rc != 0]) >>
- (14 + 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) / 4;
-
- if (tmp32) eob = idx_arr[i];
- }
- }
- *eob_ptr = eob + 1;
-}
-#endif // CONFIG_TX64X64
-
#if CONFIG_HIGHBITDEPTH
void aom_highbd_quantize_b_32x32_c(
const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block,
@@ -724,16 +533,22 @@
#endif // CONFIG_TX64X64
#endif
-void aom_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] };
+static void quantize_b_helper_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,
+#if CONFIG_AOM_QM
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
+#endif
+ const int log_scale) {
+ const int zbins[2] = { ROUND_POWER_OF_TWO(zbin_ptr[0], log_scale),
+ ROUND_POWER_OF_TWO(zbin_ptr[1], log_scale) };
const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
+ int i, non_zero_count = (int)n_coeffs, eob = -1;
(void)iscan;
memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
@@ -743,10 +558,20 @@
// Pre-scan pass
for (i = (int)n_coeffs - 1; i >= 0; i--) {
const int rc = scan[i];
+#if CONFIG_AOM_QM
+ const qm_val_t wt = qm_ptr[rc];
+ const int coeff = coeff_ptr[rc] * wt;
+#else
const int coeff = coeff_ptr[rc];
+#endif // CONFIG_AOM_QM
- if (coeff < zbins[rc != 0] && coeff > nzbins[rc != 0])
+#if CONFIG_AOM_QM
+ if (coeff < (zbins[rc != 0] << AOM_QM_BITS) &&
+ coeff > (nzbins[rc != 0] << AOM_QM_BITS))
non_zero_count--;
+#else
+ if (coeff < zbins[rc != 0] && coeff > nzbins[rc != 0]) non_zero_count--;
+#endif // CONFIG_AOM_QM
else
break;
}
@@ -758,22 +583,67 @@
const int coeff = coeff_ptr[rc];
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+ int tmp32;
+#if CONFIG_AOM_QM
+ const qm_val_t wt = qm_ptr[rc];
+ if (abs_coeff * wt >= (zbins[rc != 0] << AOM_QM_BITS)) {
+#else
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
+#endif // CONFIG_AOM_QM
+ int64_t tmp =
+ clamp(abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale),
+ INT16_MIN, INT16_MAX);
+#if CONFIG_AOM_QM
+ const qm_val_t wt = qm_ptr[rc];
+ tmp *= wt;
+ tmp32 = (int)(((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+ quant_shift_ptr[rc != 0]) >>
+ (16 - log_scale + AOM_QM_BITS)); // quantization
+#else
+ tmp32 = (int)(((((tmp * quant_ptr[rc != 0]) >> 16) + tmp) *
+ quant_shift_ptr[rc != 0]) >>
+ (16 - log_scale)); // quantization
+#endif // CONFIG_AOM_QM
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
+#if CONFIG_AOM_QM
+ dequant =
+ (dequant_ptr[rc != 0] * iqm_ptr[rc] + (1 << (AOM_QM_BITS - 1))) >>
+ AOM_QM_BITS;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant / (1 << log_scale);
+#else
+ dqcoeff_ptr[rc] =
+ qcoeff_ptr[rc] * dequant_ptr[rc != 0] / (1 << log_scale);
+#endif // CONFIG_AOM_QM
- if (tmp) eob = i;
+ if (tmp32) eob = i;
}
}
}
*eob_ptr = eob + 1;
}
+void aom_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
+#if CONFIG_AOM_QM
+ ,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
+#endif
+ ) {
+ quantize_b_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
+ dequant_ptr, eob_ptr, scan, iscan,
+#if CONFIG_AOM_QM
+ qm_ptr, iqm_ptr,
+#endif
+ 0);
+}
+
#if CONFIG_HIGHBITDEPTH
void aom_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
@@ -831,52 +701,19 @@
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;
+ const int16_t *scan, const int16_t *iscan
+#if CONFIG_AOM_QM
+ ,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
+#endif
+ ) {
+ quantize_b_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
+ dequant_ptr, eob_ptr, scan, iscan,
+#if CONFIG_AOM_QM
+ qm_ptr, iqm_ptr,
+#endif
+ 1);
}
#if CONFIG_TX64X64
@@ -886,52 +723,19 @@
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], 2),
- ROUND_POWER_OF_TWO(zbin_ptr[1], 2) };
- const int nzbins[2] = { zbins[0] * -1, zbins[1] * -1 };
-
- int idx = 0;
- int idx_arr[4096];
- 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], 2);
- abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
- tmp = ((((abs_coeff * quant_ptr[rc != 0]) >> 16) + abs_coeff) *
- quant_shift_ptr[rc != 0]) >>
- 14;
-
- qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 4;
-
- if (tmp) eob = idx_arr[i];
- }
- }
- *eob_ptr = eob + 1;
+ const int16_t *scan, const int16_t *iscan
+#if CONFIG_AOM_QM
+ ,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
+#endif
+ ) {
+ quantize_b_helper_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
+ dequant_ptr, eob_ptr, scan, iscan,
+#if CONFIG_AOM_QM
+ qm_ptr, iqm_ptr,
+#endif
+ 2);
}
#endif // CONFIG_TX64X64
diff --git a/av1/encoder/av1_quantize.c b/av1/encoder/av1_quantize.c
index 64d9055..f4bdde9 100644
--- a/av1/encoder/av1_quantize.c
+++ b/av1/encoder/av1_quantize.c
@@ -897,13 +897,43 @@
const qm_val_t *iqm_ptr = qparam->iqmatrix;
#endif // CONFIG_AOM_QM
- av1_highbd_quantize_b(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
- p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
- pd->dequant, eob_ptr, sc->scan, sc->iscan,
+ switch (qparam->log_scale) {
+ case 0:
+ aom_highbd_quantize_b(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round,
+ p->quant, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
+ pd->dequant, eob_ptr, sc->scan, sc->iscan
#if CONFIG_AOM_QM
- qm_ptr, iqm_ptr,
+ ,
+ qm_ptr, iqm_ptr
#endif
- qparam->log_scale);
+ );
+ break;
+ case 1:
+ aom_highbd_quantize_b_32x32(coeff_ptr, n_coeffs, skip_block, p->zbin,
+ p->round, p->quant, p->quant_shift,
+ qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
+ sc->scan, sc->iscan
+#if CONFIG_AOM_QM
+ ,
+ qm_ptr, iqm_ptr
+#endif
+ );
+ break;
+#if CONFIG_TX64X64
+ case 2:
+ aom_highbd_quantize_b_64x64(coeff_ptr, n_coeffs, skip_block, p->zbin,
+ p->round, p->quant, p->quant_shift,
+ qcoeff_ptr, dqcoeff_ptr, pd->dequant, eob_ptr,
+ sc->scan, sc->iscan
+#if CONFIG_AOM_QM
+ ,
+ qm_ptr, iqm_ptr
+#endif
+ );
+ break;
+#endif // CONFIG_TX64X64
+ default: assert(0);
+ }
}
#if CONFIG_HIGHBITDEPTH
@@ -1537,113 +1567,6 @@
#endif // CONFIG_HIGHBITDEPTH
-void av1_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,
-#if CONFIG_AOM_QM
- const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
-#endif
- int log_scale) {
- int i, non_zero_count = (int)n_coeffs, eob = -1;
- int zbins[2] = { zbin_ptr[0], zbin_ptr[1] };
- int round[2] = { round_ptr[0], round_ptr[1] };
- int nzbins[2];
- int scale = 1;
- int shift = 16;
- (void)iscan;
-
- if (log_scale > 0) {
- zbins[0] = ROUND_POWER_OF_TWO(zbin_ptr[0], log_scale);
- zbins[1] = ROUND_POWER_OF_TWO(zbin_ptr[1], log_scale);
- round[0] = ROUND_POWER_OF_TWO(round_ptr[0], log_scale);
- round[1] = ROUND_POWER_OF_TWO(round_ptr[1], log_scale);
- scale = 1 << log_scale;
- shift = 16 - log_scale;
- }
-
- nzbins[0] = zbins[0] * -1;
- nzbins[1] = zbins[1] * -1;
-
- 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];
-#if CONFIG_AOM_QM
- const qm_val_t wt = qm_ptr[rc];
- const qm_val_t iwt = iqm_ptr[rc];
- const int dequant =
- (dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
- AOM_QM_BITS;
-#endif
- const int coeff = coeff_ptr[rc];
- const int coeff_sign = (coeff >> 31);
- const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-#if CONFIG_AOM_QM
- if (abs_coeff * wt >= (zbins[rc != 0] << AOM_QM_BITS)) {
-#else
-
- if (abs_coeff >= zbins[rc != 0]) {
-#endif
- const int64_t tmp1 = abs_coeff + round[rc != 0];
- const int64_t tmp2 = ((tmp1 * quant_ptr[rc != 0]) >> 16) + tmp1;
-#if CONFIG_AOM_QM
- const uint32_t abs_qcoeff = (uint32_t)(
- (tmp2 * wt * quant_shift_ptr[rc != 0]) >> (AOM_QM_BITS + shift));
- qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
- dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / scale;
-#else
- const uint32_t abs_qcoeff =
- (uint32_t)((tmp2 * quant_shift_ptr[rc != 0]) >> shift);
- qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / scale;
-#endif // CONFIG_AOM_QM
- if (abs_qcoeff) eob = i;
- }
- }
- }
- *eob_ptr = eob + 1;
-}
-
-#if CONFIG_HIGHBITDEPTH
-void av1_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,
-#if CONFIG_AOM_QM
- const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr,
-#endif // CONFIG_AOM_QM
- int log_scale) {
- av1_quantize_b_c(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr,
- dequant_ptr, eob_ptr, scan, iscan,
-#if CONFIG_AOM_QM
- qm_ptr, iqm_ptr,
-#endif // CONFIG_AOM_QM
- log_scale);
-}
-#endif // CONFIG_HIGHBITDEPTH
-
static void invert_quant(int16_t *quant, int16_t *shift, int d) {
uint32_t t;
int l, m;
