FPMT: Add av1_compute_num_fp_contexts()
Added av1_compute_num_fp_contexts() to
calculate the max number of frame parallel
encodes possible for the current config. Also
added is_fp_config() which decides whether
frame parallel encode is valid.
Change-Id: I1de7b56a59894b73a88543ea4d1947e1ce163f5a
diff --git a/av1/av1_cx_iface.c b/av1/av1_cx_iface.c
index 6a0bf16..702a16e 100644
--- a/av1/av1_cx_iface.c
+++ b/av1/av1_cx_iface.c
@@ -2388,6 +2388,15 @@
if (res != AOM_CODEC_OK) {
return res;
}
+ if (i == 0) {
+ // Calculate the maximum number of frames that can be encoded in
+ // parallel
+ priv->ppi->num_fp_contexts = av1_compute_num_fp_contexts(
+ priv->ppi, &priv->ppi->parallel_cpi[i]->oxcf,
+ av1_compute_num_enc_workers(
+ priv->ppi->parallel_cpi[i],
+ priv->ppi->parallel_cpi[i]->oxcf.max_threads));
+ }
}
priv->ppi->cpi = priv->ppi->parallel_cpi[0];
#else
diff --git a/av1/encoder/ethread.c b/av1/encoder/ethread.c
index 7e4f11d..203810a 100644
--- a/av1/encoder/ethread.c
+++ b/av1/encoder/ethread.c
@@ -730,6 +730,50 @@
}
#if CONFIG_FRAME_PARALLEL_ENCODE
+static int AOM_INLINE is_fp_config(AV1_PRIMARY *ppi, AV1EncoderConfig *oxcf) {
+ // TODO(Mufaddal, Aasaipriya): Test and enable multi-tile, resize and vbr
+ // config.
+ if (oxcf->rc_cfg.mode != AOM_Q) {
+ return 0;
+ }
+ if (ppi->use_svc) {
+ return 0;
+ }
+ if (oxcf->tile_cfg.tile_columns > 0 || oxcf->tile_cfg.tile_rows > 0) {
+ return 0;
+ }
+ if (oxcf->dec_model_cfg.timing_info_present) {
+ return 0;
+ }
+ if (oxcf->mode != GOOD) {
+ return 0;
+ }
+ if (oxcf->tool_cfg.error_resilient_mode) {
+ return 0;
+ }
+ if (oxcf->resize_cfg.resize_mode) {
+ return 0;
+ }
+
+ return 1;
+}
+
+int av1_compute_num_fp_contexts(AV1_PRIMARY *ppi, AV1EncoderConfig *oxcf,
+ int max_num_enc_workers) {
+ if (!is_fp_config(ppi, oxcf)) {
+ return 1;
+ }
+
+ // A parallel frame encode must have at least half the theoretical limit of
+ // max enc workers. TODO(Mufaddal) : Tune this value based on empirical
+ // analysis.
+ int workers_per_frame = (max_num_enc_workers + 1) / 2;
+ int max_threads = oxcf->max_threads;
+ int num_fp_contexts = max_threads / workers_per_frame;
+
+ return AOMMIN(num_fp_contexts, MAX_PARALLEL_FRAMES);
+}
+
// Prepare level 1 workers. This function is only called for
// parallel_frame_count > 1. This function populates the mt_info structure of
// frame level contexts appropriately by dividing the total number of available
@@ -1023,7 +1067,7 @@
}
// Computes the number of workers for encoding stage (row/tile multi-threading)
-static AOM_INLINE int compute_num_enc_workers(AV1_COMP *cpi, int max_workers) {
+int av1_compute_num_enc_workers(AV1_COMP *cpi, int max_workers) {
if (max_workers <= 1) return 1;
if (cpi->oxcf.row_mt)
return compute_num_enc_row_mt_workers(&cpi->common, max_workers);
@@ -2279,7 +2323,7 @@
// found to improve speed. Hence the thread assignment for single-pass encode
// is kept based on compute_num_enc_workers().
if (cpi->oxcf.pass < AOM_RC_SECOND_PASS)
- return (compute_num_enc_workers(cpi, cpi->oxcf.max_threads));
+ return (av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads));
if (cpi->oxcf.max_threads <= 1) return 1;
@@ -2292,22 +2336,22 @@
// Computes num_workers for tpl multi-threading.
static AOM_INLINE int compute_num_tpl_workers(AV1_COMP *cpi) {
- return compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ return av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
}
// Computes num_workers for loop filter multi-threading.
static AOM_INLINE int compute_num_lf_workers(AV1_COMP *cpi) {
- return compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ return av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
}
// Computes num_workers for cdef multi-threading.
static AOM_INLINE int compute_num_cdef_workers(AV1_COMP *cpi) {
- return compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ return av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
}
// Computes num_workers for loop-restoration multi-threading.
static AOM_INLINE int compute_num_lr_workers(AV1_COMP *cpi) {
- return compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ return av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
}
// Computes num_workers for pack bitstream multi-threading.
@@ -2323,13 +2367,14 @@
if (cpi->oxcf.pass >= AOM_RC_SECOND_PASS)
num_mod_workers = 0;
else
- num_mod_workers = compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ num_mod_workers =
+ av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
break;
case MOD_TF: num_mod_workers = compute_num_tf_workers(cpi); break;
case MOD_TPL: num_mod_workers = compute_num_tpl_workers(cpi); break;
case MOD_GME: num_mod_workers = 1; break;
case MOD_ENC:
- num_mod_workers = compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
+ num_mod_workers = av1_compute_num_enc_workers(cpi, cpi->oxcf.max_threads);
break;
case MOD_LPF: num_mod_workers = compute_num_lf_workers(cpi); break;
case MOD_CDEF_SEARCH:
diff --git a/av1/encoder/ethread.h b/av1/encoder/ethread.h
index 6541110..59fb043 100644
--- a/av1/encoder/ethread.h
+++ b/av1/encoder/ethread.h
@@ -88,6 +88,9 @@
void av1_init_tile_thread_data(AV1_PRIMARY *ppi, int is_first_pass);
+int av1_compute_num_fp_contexts(AV1_PRIMARY *ppi, AV1EncoderConfig *oxcf,
+ int max_num_enc_workers);
+
void av1_cdef_mse_calc_frame_mt(AV1_COMMON *cm, MultiThreadInfo *mt_info,
CdefSearchCtx *cdef_search_ctx);
@@ -100,6 +103,8 @@
unsigned int *max_tile_size, uint32_t *const obu_header_size,
uint8_t **tile_data_start, const int num_workers);
+int av1_compute_num_enc_workers(AV1_COMP *cpi, int max_workers);
+
#ifdef __cplusplus
} // extern "C"
#endif
