Allocate dv_costs when required
The structure 'dv_costs' is used to hold the cost of signalling
the dv in the bitstream in case of intraBC mode. This patch
allocates 'dv_costs' only when intraBC mode needs to be evaluated.
As 'dv_costs' is not used during first pass, the code related
to its allocation is cleaned up from fp_prepare_enc_workers().
For AVIF image encode with speed = 9,
HEAP Memory reduction(%)
Resolution threads=1 threads=4
640x360 3.25 2.64
768x512 2.06 2.28
832x480 2.14 1.84
1280x720 1.03 0.92
HEAP memory reduction was measured using the following command.
$valgrind --tool=massif ./avifenc ...
Change-Id: If40abe0bde968edc067b56143e9320078f57dad9
diff --git a/av1/encoder/encoder.h b/av1/encoder/encoder.h
index 6279c66..ad38eb5 100644
--- a/av1/encoder/encoder.h
+++ b/av1/encoder/encoder.h
@@ -27,6 +27,7 @@
#include "av1/common/blockd.h"
#include "av1/common/entropymode.h"
#include "av1/common/enums.h"
+#include "av1/common/reconintra.h"
#include "av1/common/resize.h"
#include "av1/common/thread_common.h"
#include "av1/common/timing.h"
@@ -3731,6 +3732,12 @@
(cpi->compressor_stage == ENCODE_STAGE) && cpi->ppi->lap_enabled));
}
+// Decide whether 'dv_costs' need to be allocated/stored during the encoding.
+static AOM_INLINE bool av1_need_dv_costs(const AV1_COMP *const cpi) {
+ return !cpi->sf.rt_sf.use_nonrd_pick_mode &&
+ av1_allow_intrabc(&cpi->common) && !is_stat_generation_stage(cpi);
+}
+
/*!\endcond */
/*!\brief Check if the current stage has statistics
*
diff --git a/av1/encoder/encoder_alloc.h b/av1/encoder/encoder_alloc.h
index 85278f6..89784c5 100644
--- a/av1/encoder/encoder_alloc.h
+++ b/av1/encoder/encoder_alloc.h
@@ -74,13 +74,6 @@
(MvCosts *)aom_calloc(1, sizeof(MvCosts)));
}
- if (cpi->td.mb.dv_costs) {
- aom_free(cpi->td.mb.dv_costs);
- cpi->td.mb.dv_costs = NULL;
- }
- CHECK_MEM_ERROR(cm, cpi->td.mb.dv_costs,
- (IntraBCMVCosts *)aom_malloc(sizeof(*cpi->td.mb.dv_costs)));
-
av1_setup_shared_coeff_buffer(cm->seq_params, &cpi->td.shared_coeff_buf,
cm->error);
av1_setup_sms_tree(cpi, &cpi->td);
diff --git a/av1/encoder/ethread.c b/av1/encoder/ethread.c
index 09acf1a..b95457a 100644
--- a/av1/encoder/ethread.c
+++ b/av1/encoder/ethread.c
@@ -1297,7 +1297,8 @@
memcpy(thread_data->td->mb.mv_costs, cpi->td.mb.mv_costs,
sizeof(MvCosts));
}
- if (cpi->sf.intra_sf.dv_cost_upd_level != INTERNAL_COST_UPD_OFF) {
+ if ((cpi->sf.intra_sf.dv_cost_upd_level != INTERNAL_COST_UPD_OFF) &&
+ av1_need_dv_costs(cpi)) {
CHECK_MEM_ERROR(cm, thread_data->td->mb.dv_costs,
(IntraBCMVCosts *)aom_malloc(sizeof(IntraBCMVCosts)));
memcpy(thread_data->td->mb.dv_costs, cpi->td.mb.dv_costs,
@@ -1375,12 +1376,6 @@
memcpy(thread_data->td->mb.mv_costs, cpi->td.mb.mv_costs,
sizeof(MvCosts));
}
- if (cpi->sf.intra_sf.dv_cost_upd_level != INTERNAL_COST_UPD_OFF) {
- CHECK_MEM_ERROR(cm, thread_data->td->mb.dv_costs,
- (IntraBCMVCosts *)aom_malloc(sizeof(IntraBCMVCosts)));
- memcpy(thread_data->td->mb.dv_costs, cpi->td.mb.dv_costs,
- sizeof(IntraBCMVCosts));
- }
}
av1_alloc_mb_data(cm, &thread_data->td->mb,
@@ -1662,9 +1657,7 @@
if (cpi->sf.inter_sf.mv_cost_upd_level != INTERNAL_COST_UPD_OFF) {
aom_free(thread_data->td->mb.mv_costs);
}
- if (cpi->sf.intra_sf.dv_cost_upd_level != INTERNAL_COST_UPD_OFF) {
- aom_free(thread_data->td->mb.dv_costs);
- }
+ assert(!thread_data->td->mb.dv_costs);
}
av1_dealloc_mb_data(cm, &thread_data->td->mb);
}
diff --git a/av1/encoder/rd.c b/av1/encoder/rd.c
index 7ada9b9..17c7960 100644
--- a/av1/encoder/rd.c
+++ b/av1/encoder/rd.c
@@ -749,8 +749,12 @@
av1_fill_mode_rates(cm, &x->mode_costs, cm->fc);
// Frame level dv cost update
- if (!use_nonrd_pick_mode && av1_allow_intrabc(cm) &&
- !is_stat_generation_stage(cpi)) {
+ if (av1_need_dv_costs(cpi)) {
+ if (cpi->td.mb.dv_costs == NULL) {
+ CHECK_MEM_ERROR(
+ cm, cpi->td.mb.dv_costs,
+ (IntraBCMVCosts *)aom_malloc(sizeof(*cpi->td.mb.dv_costs)));
+ }
av1_fill_dv_costs(&cm->fc->ndvc, x->dv_costs);
}
}
