Refactor av1_quantize_fp_c functions.
Use one helper function to remove copy paste of
av1_quantize_fp_c, av1_quantize_fp_32x32_c, av1_quantize_fp_64x64_c.
BUG=aomedia:442
Change-Id: I749bb27d7cfeb4040ff5e0f394f691a44377efad
diff --git a/av1/encoder/av1_quantize.c b/av1/encoder/av1_quantize.c
index beb5d7f..f9a30d1 100644
--- a/av1/encoder/av1_quantize.c
+++ b/av1/encoder/av1_quantize.c
@@ -438,6 +438,138 @@
*eob_ptr = 0;
}
+static void quantize_fp_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
+ int log_scale) {
+ int i, eob = -1;
+ // TODO(jingning) Decide the need of these arguments after the
+ // quantization process is completed.
+ (void)zbin_ptr;
+ (void)quant_shift_ptr;
+ (void)iscan;
+#if CONFIG_AOM_QM
+ const qm_val_t *qm_ptr = qparam->qmatrix;
+ const qm_val_t *iqm_ptr = qparam->iqmatrix;
+#endif // CONFIG_AOM_QM
+
+ memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
+ memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
+
+ if (!skip_block) {
+ // Quantization pass: All coefficients with index >= zero_flag are
+ // skippable. Note: zero_flag can be zero.
+ for (i = 0; i < n_coeffs; i++) {
+ const int rc = scan[i];
+ const int coeff = coeff_ptr[rc];
+#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_sign = (coeff >> 31);
+ int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
+ int tmp32 = 0;
+#if CONFIG_AOM_QM
+ if (abs_coeff * wt >=
+ (dequant_ptr[rc != 0] << (AOM_QM_BITS - (1 + log_scale)))) {
+#else
+ if (abs_coeff >= (dequant_ptr[rc != 0] >> (1 + log_scale))) {
+#endif
+ abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], log_scale);
+ abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
+#if CONFIG_AOM_QM
+ tmp32 = (int)((abs_coeff * wt * quant_ptr[rc != 0]) >>
+ ((16 - log_scale) + AOM_QM_BITS));
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant / (1 << log_scale);
+#else
+ tmp32 = (int)((abs_coeff * quant_ptr[rc != 0]) >> (16 - log_scale));
+ qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
+ dqcoeff_ptr[rc] =
+ qcoeff_ptr[rc] * dequant_ptr[rc != 0] / (1 << log_scale);
+#endif
+ }
+
+ if (tmp32) eob = i;
+ }
+ }
+ *eob_ptr = eob + 1;
+}
+
+void av1_quantize_fp_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_fp_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);
+}
+
+void av1_quantize_fp_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
+#if CONFIG_AOM_QM
+ ,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
+#endif
+ ) {
+ quantize_fp_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
+void av1_quantize_fp_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
+#if CONFIG_AOM_QM
+ ,
+ const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
+#endif
+ ) {
+ quantize_fp_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
+
void av1_quantize_fp_facade(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
const MACROBLOCK_PLANE *p, tran_low_t *qcoeff_ptr,
const MACROBLOCKD_PLANE *pd,
@@ -455,14 +587,14 @@
if (n_coeffs < 16) {
// TODO(jingning): Need SIMD implementation for smaller block size
// quantization.
- av1_quantize_fp_c(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp,
- p->quant_fp, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
- pd->dequant, eob_ptr, sc->scan, sc->iscan
+ quantize_fp_helper_c(coeff_ptr, n_coeffs, skip_block, p->zbin,
+ p->round_fp, p->quant_fp, 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);
} else {
av1_quantize_fp(coeff_ptr, n_coeffs, skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff_ptr, dqcoeff_ptr,
@@ -1348,62 +1480,6 @@
#endif // CONFIG_NEW_QUANT
#endif // CONFIG_HIGHBITDEPTH
-void av1_quantize_fp_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 i, eob = -1;
- // TODO(jingning) Decide the need of these arguments after the
- // quantization process is completed.
- (void)zbin_ptr;
- (void)quant_shift_ptr;
- (void)iscan;
-
- memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
- memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
-
- if (!skip_block) {
- // Quantization pass: All coefficients with index >= zero_flag are
- // skippable. Note: zero_flag can be zero.
- for (i = 0; i < n_coeffs; i++) {
- const int rc = scan[i];
- const int coeff = coeff_ptr[rc];
-#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_sign = (coeff >> 31);
- const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
- int64_t tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
- int tmp32;
-#if CONFIG_AOM_QM
- tmp32 = (int)((tmp * wt * quant_ptr[rc != 0]) >> (16 + AOM_QM_BITS));
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
-#else
- tmp32 = (int)((tmp * quant_ptr[rc != 0]) >> 16);
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
-#endif
-
- if (tmp32) eob = i;
- }
- }
- *eob_ptr = eob + 1;
-}
-
#if CONFIG_HIGHBITDEPTH
void av1_highbd_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t count,
int skip_block, const int16_t *zbin_ptr,
@@ -1465,132 +1541,6 @@
#endif // CONFIG_HIGHBITDEPTH
-// TODO(jingning) Refactor this file and combine functions with similar
-// operations.
-void av1_quantize_fp_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
-#if CONFIG_AOM_QM
- ,
- const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
-#endif
- ) {
- int i, eob = -1;
- (void)zbin_ptr;
- (void)quant_shift_ptr;
- (void)iscan;
-
- memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
- memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
-
- if (!skip_block) {
- for (i = 0; i < n_coeffs; i++) {
- const int rc = scan[i];
- const int coeff = coeff_ptr[rc];
-#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;
- int64_t tmp = 0;
-#endif
- const int coeff_sign = (coeff >> 31);
- int tmp32 = 0;
- int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
-#if CONFIG_AOM_QM
- if (abs_coeff * wt >= (dequant_ptr[rc != 0] << (AOM_QM_BITS - 2))) {
-#else
- if (abs_coeff >= (dequant_ptr[rc != 0] >> 2)) {
-#endif
- abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
- abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
-#if CONFIG_AOM_QM
- tmp = abs_coeff * wt;
- tmp32 = (int)(tmp * quant_ptr[rc != 0]) >> (AOM_QM_BITS + 15);
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / 2;
-#else
- tmp32 = (abs_coeff * quant_ptr[rc != 0]) >> 15;
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant_ptr[rc != 0]) / 2;
-#endif
- }
-
- if (tmp32) eob = i;
- }
- }
- *eob_ptr = eob + 1;
-}
-
-#if CONFIG_TX64X64
-void av1_quantize_fp_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
-#if CONFIG_AOM_QM
- ,
- const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
-#endif
- ) {
- int i, eob = -1;
- (void)zbin_ptr;
- (void)quant_shift_ptr;
- (void)iscan;
-
- memset(qcoeff_ptr, 0, n_coeffs * sizeof(*qcoeff_ptr));
- memset(dqcoeff_ptr, 0, n_coeffs * sizeof(*dqcoeff_ptr));
-
- if (!skip_block) {
- for (i = 0; i < n_coeffs; i++) {
- const int rc = scan[i];
- const int coeff = coeff_ptr[rc];
-#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;
- int64_t tmp = 0;
-#endif
- const int coeff_sign = (coeff >> 31);
- int tmp32 = 0;
- int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
-
-#if CONFIG_AOM_QM
- if (abs_coeff * wt >= (dequant_ptr[rc != 0] << (AOM_QM_BITS - 3))) {
-#else
- if (abs_coeff >= (dequant_ptr[rc != 0] >> 3)) {
-#endif
- abs_coeff += ROUND_POWER_OF_TWO(round_ptr[rc != 0], 2);
- abs_coeff = clamp(abs_coeff, INT16_MIN, INT16_MAX);
-#if CONFIG_AOM_QM
- tmp = abs_coeff * wt;
- tmp32 = (int)(tmp * quant_ptr[rc != 0]) >> (AOM_QM_BITS + 14);
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant) / 4;
-#else
- tmp32 = (abs_coeff * quant_ptr[rc != 0]) >> 14;
- qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
- dqcoeff_ptr[rc] = (qcoeff_ptr[rc] * dequant_ptr[rc != 0]) / 4;
-#endif
- }
-
- if (tmp32) eob = i;
- }
- }
- *eob_ptr = eob + 1;
-}
-#endif // CONFIG_TX64X64
-
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,
