aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorBen Chang <benc@nvidia.com>2017-06-24 12:17:14 +0000
committerLuca Barbato <lu_zero@gentoo.org>2017-06-24 12:35:52 +0000
commit7cb053e4ddf258e2dbf52ccc586548680742d758 (patch)
treef3b08d74bc31a0d935bf9a2b70d657e0c583fd8f
parent2e8d88ad5281ab923e1d9772eb99fdfb483435c2 (diff)
downloadffmpeg-7cb053e4ddf258e2dbf52ccc586548680742d758.tar.gz
nvenc: Minimize the surface allocation
The previous default sets the allocated surfaces to 32 unless it is user-overridden or the lookahead parameter is set. Change the surfaces calculation for default, B-frames and lookahead scenario.
-rw-r--r--libavcodec/nvenc.c45
-rw-r--r--libavcodec/nvenc_h264.c4
-rw-r--r--libavcodec/nvenc_hevc.c4
3 files changed, 44 insertions, 9 deletions
diff --git a/libavcodec/nvenc.c b/libavcodec/nvenc.c
index 7b30ad768e..884b344b96 100644
--- a/libavcodec/nvenc.c
+++ b/libavcodec/nvenc.c
@@ -874,6 +874,44 @@ static int nvenc_setup_codec_config(AVCodecContext *avctx)
return 0;
}
+static int nvenc_recalc_surfaces(AVCodecContext *avctx)
+{
+ NVENCContext *ctx = avctx->priv_data;
+ // default minimum of 4 surfaces
+ // multiply by 2 for number of NVENCs on gpu (hardcode to 2)
+ // another multiply by 2 to avoid blocking next PBB group
+ int nb_surfaces = FFMAX(4, ctx->config.frameIntervalP * 2 * 2);
+
+ // lookahead enabled
+ if (ctx->rc_lookahead > 0) {
+ // +1 is to account for lkd_bound calculation later
+ // +4 is to allow sufficient pipelining with lookahead
+ nb_surfaces = FFMAX(1, FFMAX(nb_surfaces, ctx->rc_lookahead + ctx->config.frameIntervalP + 1 + 4));
+ if (nb_surfaces > ctx->nb_surfaces && ctx->nb_surfaces > 0) {
+ av_log(avctx, AV_LOG_WARNING,
+ "Defined rc_lookahead requires more surfaces, "
+ "increasing used surfaces %d -> %d\n",
+ ctx->nb_surfaces, nb_surfaces);
+ }
+ ctx->nb_surfaces = FFMAX(nb_surfaces, ctx->nb_surfaces);
+ } else {
+ if (ctx->config.frameIntervalP > 1 &&
+ ctx->nb_surfaces < nb_surfaces && ctx->nb_surfaces > 0) {
+ av_log(avctx, AV_LOG_WARNING,
+ "Defined b-frame requires more surfaces, "
+ "increasing used surfaces %d -> %d\n",
+ ctx->nb_surfaces, nb_surfaces);
+ ctx->nb_surfaces = FFMAX(ctx->nb_surfaces, nb_surfaces);
+ } else if (ctx->nb_surfaces <= 0)
+ ctx->nb_surfaces = nb_surfaces;
+ // otherwise use user specified value
+ }
+
+ ctx->nb_surfaces = FFMAX(1, FFMIN(MAX_REGISTERED_FRAMES, ctx->nb_surfaces));
+ ctx->async_depth = FFMIN(ctx->async_depth, ctx->nb_surfaces - 1);
+ return 0;
+}
+
static int nvenc_setup_encoder(AVCodecContext *avctx)
{
NVENCContext *ctx = avctx->priv_data;
@@ -956,6 +994,8 @@ static int nvenc_setup_encoder(AVCodecContext *avctx)
ctx->initial_pts[0] = AV_NOPTS_VALUE;
ctx->initial_pts[1] = AV_NOPTS_VALUE;
+ nvenc_recalc_surfaces(avctx);
+
nvenc_setup_rate_control(avctx);
if (avctx->flags & AV_CODEC_FLAG_INTERLACED_DCT) {
@@ -1057,11 +1097,6 @@ static int nvenc_setup_surfaces(AVCodecContext *avctx)
NVENCContext *ctx = avctx->priv_data;
int i, ret;
- ctx->nb_surfaces = FFMAX(4 + avctx->max_b_frames,
- ctx->nb_surfaces);
- ctx->async_depth = FFMIN(ctx->async_depth, ctx->nb_surfaces - 1);
-
-
ctx->frames = av_mallocz_array(ctx->nb_surfaces, sizeof(*ctx->frames));
if (!ctx->frames)
return AVERROR(ENOMEM);
diff --git a/libavcodec/nvenc_h264.c b/libavcodec/nvenc_h264.c
index 1c9e07a776..bf983265df 100644
--- a/libavcodec/nvenc_h264.c
+++ b/libavcodec/nvenc_h264.c
@@ -72,14 +72,14 @@ static const AVOption options[] = {
{ "ll_2pass_quality", "Multi-pass optimized for image quality (only for low-latency presets)", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_QUALITY }, 0, 0, VE, "rc" },
{ "ll_2pass_size", "Multi-pass optimized for constant frame size (only for low-latency presets)", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP }, 0, 0, VE, "rc" },
{ "vbr_2pass", "Multi-pass variable bitrate mode", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_VBR }, 0, 0, VE, "rc" },
- { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 32 }, 0, INT_MAX, VE },
+ { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_REGISTERED_FRAMES, VE },
{ "device", "Select a specific NVENC device", OFFSET(device), AV_OPT_TYPE_INT, { .i64 = -1 }, -2, INT_MAX, VE, "device" },
{ "any", "Pick the first device available", 0, AV_OPT_TYPE_CONST, { .i64 = ANY_DEVICE }, 0, 0, VE, "device" },
{ "list", "List the available devices", 0, AV_OPT_TYPE_CONST, { .i64 = LIST_DEVICES }, 0, 0, VE, "device" },
{ "async_depth", "Delay frame output by the given amount of frames", OFFSET(async_depth), AV_OPT_TYPE_INT, { .i64 = INT_MAX }, 0, INT_MAX, VE },
{ "delay", "Delay frame output by the given amount of frames", OFFSET(async_depth), AV_OPT_TYPE_INT, { .i64 = INT_MAX }, 0, INT_MAX, VE },
#if NVENCAPI_MAJOR_VERSION >= 7
- { "rc-lookahead", "Number of frames to look ahead for rate-control", OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE },
+ { "rc-lookahead", "Number of frames to look ahead for rate-control", OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = 0 }, -1, INT_MAX, VE },
{ "no-scenecut", "When lookahead is enabled, set this to 1 to disable adaptive I-frame insertion at scene cuts", OFFSET(no_scenecut), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
{ "b_adapt", "When lookahead is enabled, set this to 0 to disable adaptive B-frame decision", OFFSET(b_adapt), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, VE },
{ "spatial-aq", "set to 1 to enable Spatial AQ", OFFSET(aq), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
diff --git a/libavcodec/nvenc_hevc.c b/libavcodec/nvenc_hevc.c
index a13db98356..caf7c4add9 100644
--- a/libavcodec/nvenc_hevc.c
+++ b/libavcodec/nvenc_hevc.c
@@ -72,14 +72,14 @@ static const AVOption options[] = {
{ "ll_2pass_quality", "Multi-pass optimized for image quality (only for low-latency presets)", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_QUALITY }, 0, 0, VE, "rc" },
{ "ll_2pass_size", "Multi-pass optimized for constant frame size (only for low-latency presets)", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP }, 0, 0, VE, "rc" },
{ "vbr_2pass", "Multi-pass variable bitrate mode", 0, AV_OPT_TYPE_CONST, { .i64 = NV_ENC_PARAMS_RC_2_PASS_VBR }, 0, 0, VE, "rc" },
- { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 32 }, 0, INT_MAX, VE },
+ { "surfaces", "Number of concurrent surfaces", OFFSET(nb_surfaces), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, MAX_REGISTERED_FRAMES, VE },
{ "device", "Select a specific NVENC device", OFFSET(device), AV_OPT_TYPE_INT, { .i64 = -1 }, -2, INT_MAX, VE, "device" },
{ "any", "Pick the first device available", 0, AV_OPT_TYPE_CONST, { .i64 = ANY_DEVICE }, 0, 0, VE, "device" },
{ "list", "List the available devices", 0, AV_OPT_TYPE_CONST, { .i64 = LIST_DEVICES }, 0, 0, VE, "device" },
{ "async_depth", "Delay frame output by the given amount of frames", OFFSET(async_depth), AV_OPT_TYPE_INT, { .i64 = INT_MAX }, 0, INT_MAX, VE },
{ "delay", "Delay frame output by the given amount of frames", OFFSET(async_depth), AV_OPT_TYPE_INT, { .i64 = INT_MAX }, 0, INT_MAX, VE },
#if NVENCAPI_MAJOR_VERSION >= 7
- { "rc-lookahead", "Number of frames to look ahead for rate-control", OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, INT_MAX, VE },
+ { "rc-lookahead", "Number of frames to look ahead for rate-control", OFFSET(rc_lookahead), AV_OPT_TYPE_INT, { .i64 = 0 }, -1, INT_MAX, VE },
{ "no-scenecut", "When lookahead is enabled, set this to 1 to disable adaptive I-frame insertion at scene cuts", OFFSET(no_scenecut), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
{ "spatial_aq", "set to 1 to enable Spatial AQ", OFFSET(aq), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
{ "zerolatency", "Set 1 to indicate zero latency operation (no reordering delay)", OFFSET(zerolatency), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },