Split code paths for quantize_b_adaptive
Change-Id: Ib56a8ac04345313b5dc0efb475bd8dd520ebd9c3
diff --git a/aom_dsp/quantize.c b/aom_dsp/quantize.c
index 5a28698..4acbf20 100644
--- a/aom_dsp/quantize.c
+++ b/aom_dsp/quantize.c
@@ -73,6 +73,7 @@
}
*eob_ptr = eob + 1;
}
+
void quantize_b_helper_c(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,
@@ -252,23 +253,89 @@
/* These functions should only be called when quantisation matrices
are not used. */
+void aom_quantize_b_adaptive_c(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, const int16_t *iscan) {
+ quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+ NULL, NULL, 0);
+}
+
+void aom_quantize_b_32x32_adaptive_c(
+ 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, const int16_t *iscan) {
+ quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+ NULL, NULL, 1);
+}
+
+void aom_quantize_b_64x64_adaptive_c(
+ 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, const int16_t *iscan) {
+ quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
+ NULL, NULL, 2);
+}
+
+void aom_highbd_quantize_b_adaptive_c(
+ 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, const int16_t *iscan) {
+ highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+ iscan, NULL, NULL, 0);
+}
+
+void aom_highbd_quantize_b_32x32_adaptive_c(
+ 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, const int16_t *iscan) {
+ highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+ iscan, NULL, NULL, 1);
+}
+
+void aom_highbd_quantize_b_64x64_adaptive_c(
+ 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, const int16_t *iscan) {
+ highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
+ quant_ptr, quant_shift_ptr, qcoeff_ptr,
+ dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
+ iscan, NULL, NULL, 2);
+}
+
void aom_quantize_b_c(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, const int16_t *iscan) {
-#if ADAPTIVE_QUANT_B
- quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
- NULL, NULL, 0);
-
-#else
quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr, quant_ptr,
quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr,
eob_ptr, scan, iscan, NULL, NULL, 0);
-#endif
}
void aom_quantize_b_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
@@ -278,16 +345,9 @@
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 ADAPTIVE_QUANT_B
- quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
- NULL, NULL, 1);
-#else
quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr, quant_ptr,
quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr,
eob_ptr, scan, iscan, NULL, NULL, 1);
-#endif
}
void aom_quantize_b_64x64_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
@@ -297,16 +357,9 @@
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 ADAPTIVE_QUANT_B
- quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
- NULL, NULL, 2);
-#else
quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr, quant_ptr,
quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr,
eob_ptr, scan, iscan, NULL, NULL, 2);
-#endif
}
void aom_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
@@ -316,17 +369,10 @@
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 ADAPTIVE_QUANT_B
- highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
- iscan, NULL, NULL, 0);
-#else
highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
quant_ptr, quant_shift_ptr, qcoeff_ptr,
dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
NULL, NULL, 0);
-#endif
}
void aom_highbd_quantize_b_32x32_c(
@@ -335,17 +381,10 @@
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 ADAPTIVE_QUANT_B
- highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
- iscan, NULL, NULL, 1);
-#else
highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
quant_ptr, quant_shift_ptr, qcoeff_ptr,
dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
NULL, NULL, 1);
-#endif
}
void aom_highbd_quantize_b_64x64_c(
@@ -354,15 +393,8 @@
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 ADAPTIVE_QUANT_B
- highbd_quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
- quant_ptr, quant_shift_ptr, qcoeff_ptr,
- dqcoeff_ptr, dequant_ptr, eob_ptr, scan,
- iscan, NULL, NULL, 2);
-#else
highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, zbin_ptr, round_ptr,
quant_ptr, quant_shift_ptr, qcoeff_ptr,
dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan,
NULL, NULL, 2);
-#endif
}
diff --git a/aom_dsp/quantize.h b/aom_dsp/quantize.h
index df7f306..3f2112a 100644
--- a/aom_dsp/quantize.h
+++ b/aom_dsp/quantize.h
@@ -22,6 +22,66 @@
extern "C" {
#endif
+void quantize_b_adaptive_helper_c(
+ 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, const int16_t *iscan, const qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr, const int log_scale);
+
+void aom_quantize_b_adaptive_c(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, const int16_t *iscan);
+
+void aom_quantize_b_32x32_adaptive_c(
+ 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, const int16_t *iscan);
+
+void aom_quantize_b_64x64_adaptive_c(
+ 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, const int16_t *iscan);
+
+void highbd_quantize_b_adaptive_helper_c(
+ 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, const int16_t *iscan, const qm_val_t *qm_ptr,
+ const qm_val_t *iqm_ptr, const int log_scale);
+
+void aom_highbd_quantize_b_adaptive_c(
+ 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, const int16_t *iscan);
+
+void aom_highbd_quantize_b_32x32_adaptive_c(
+ 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, const int16_t *iscan);
+
+void aom_highbd_quantize_b_64x64_adaptive_c(
+ 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, const int16_t *iscan);
+
void quantize_b_helper_c(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,
diff --git a/av1/encoder/av1_quantize.c b/av1/encoder/av1_quantize.c
index 9e48358..10136f7 100644
--- a/av1/encoder/av1_quantize.c
+++ b/av1/encoder/av1_quantize.c
@@ -275,37 +275,44 @@
const SCAN_ORDER *sc, const QUANT_PARAM *qparam) {
const qm_val_t *qm_ptr = qparam->qmatrix;
const qm_val_t *iqm_ptr = qparam->iqmatrix;
+#if ADAPTIVE_QUANT_B
+ // TODO(sarahparker) These quantize_b optimizations need SIMD
+ // implementations
+ if (qm_ptr != NULL && iqm_ptr != NULL) {
+ quantize_b_adaptive_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
+ p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
+ dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
+ sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+ } else {
+ switch (qparam->log_scale) {
+ case 0:
+ aom_quantize_b_adaptive_c(coeff_ptr, n_coeffs, p->zbin_QTX,
+ p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr,
+ p->dequant_QTX, eob_ptr, sc->scan, sc->iscan);
+ break;
+ case 1:
+ aom_quantize_b_32x32_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ break;
+ case 2:
+ aom_quantize_b_64x64_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ break;
+ default: assert(0);
+ }
+ }
+#else
if (qm_ptr != NULL && iqm_ptr != NULL) {
quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
} else {
-#if ADAPTIVE_QUANT_B
- // TODO(sarahparker) These quantize_b optimizations need SIMD
- // implementations
- switch (qparam->log_scale) {
- case 0:
- aom_quantize_b_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
- p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
- dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
- sc->iscan);
- break;
- case 1:
- aom_quantize_b_32x32_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
- p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
- dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
- sc->iscan);
- break;
- case 2:
- aom_quantize_b_64x64_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
- p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
- dqcoeff_ptr, p->dequant_QTX, eob_ptr, sc->scan,
- sc->iscan);
- break;
- default: assert(0);
- }
-#else
switch (qparam->log_scale) {
case 0:
aom_quantize_b(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
@@ -327,8 +334,8 @@
break;
default: assert(0);
}
-#endif
}
+#endif
}
static void quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
@@ -417,6 +424,45 @@
const QUANT_PARAM *qparam) {
const qm_val_t *qm_ptr = qparam->qmatrix;
const qm_val_t *iqm_ptr = qparam->iqmatrix;
+#if ADAPTIVE_QUANT_B
+ if (qm_ptr != NULL && iqm_ptr != NULL) {
+ highbd_quantize_b_adaptive_helper_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX, eob_ptr,
+ sc->scan, sc->iscan, qm_ptr, iqm_ptr, qparam->log_scale);
+ } else {
+ switch (qparam->log_scale) {
+ case 0:
+ if (LIKELY(n_coeffs >= 8)) {
+ aom_highbd_quantize_b_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ } else {
+ // TODO(luoyi): Need SIMD (e.g. sse2) for smaller block size
+ // quantization
+ aom_highbd_quantize_b_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ }
+ break;
+ case 1:
+ aom_highbd_quantize_b_32x32_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ break;
+ case 2:
+ aom_highbd_quantize_b_64x64_adaptive_c(
+ coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX, p->quant_QTX,
+ p->quant_shift_QTX, qcoeff_ptr, dqcoeff_ptr, p->dequant_QTX,
+ eob_ptr, sc->scan, sc->iscan);
+ break;
+ default: assert(0);
+ }
+ }
+#else
if (qm_ptr != NULL && iqm_ptr != NULL) {
highbd_quantize_b_helper_c(coeff_ptr, n_coeffs, p->zbin_QTX, p->round_QTX,
p->quant_QTX, p->quant_shift_QTX, qcoeff_ptr,
@@ -454,6 +500,7 @@
default: assert(0);
}
}
+#endif
}
static INLINE void highbd_quantize_dc(