NVDEC (H264): Use separate contexts per channel and decode frames in DTS order (#2671)

* Use separate NVDEC contexts per channel (for FFMPEG)

* Remove NVDEC -> VIC frame override hack

* Add missing bottom_field_pic_order_in_frame_present_flag

* Make FFMPEG logging static

* nit: Remove empty lines

* New FFMPEG decoding approach -- call h264_decode_frame directly, trim surface cache to reduce memory usage

* Fix case

* Silence warnings

* PR feedback

* Per-decoder rather than per-codec ownership of surfaces on the cache
This commit is contained in:
gdkchan 2021-09-28 19:43:40 -03:00 committed by GitHub
parent 0d23504e30
commit f4f496cb48
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
18 changed files with 358 additions and 200 deletions

View file

@ -21,7 +21,7 @@ namespace Ryujinx.Graphics.Nvdec.Image
public uint ChromaOffset;
public int Width;
public int Height;
public CodecId CodecId;
public IDecoder Owner;
public ISurface Surface;
}
@ -34,104 +34,110 @@ namespace Ryujinx.Graphics.Nvdec.Image
_gmm = gmm;
}
public ISurface Get(IDecoder decoder, CodecId codecId, uint lumaOffset, uint chromaOffset, int width, int height)
public ISurface Get(IDecoder decoder, uint lumaOffset, uint chromaOffset, int width, int height)
{
ISurface surface = null;
// Try to find a compatible surface with same parameters, and same offsets.
for (int i = 0; i < MaxItems; i++)
lock (_pool)
{
ref CacheItem item = ref _pool[i];
ISurface surface = null;
if (item.LumaOffset == lumaOffset &&
item.ChromaOffset == chromaOffset &&
item.CodecId == codecId &&
item.Width == width &&
item.Height == height)
{
item.ReferenceCount++;
surface = item.Surface;
MoveToFront(i);
break;
}
}
// If we failed to find a perfect match, now ignore the offsets.
// Search backwards to replace the oldest compatible surface,
// this avoids thrashing frquently used surfaces.
// Now we need to ensure that the surface is not in use, as we'll change the data.
if (surface == null)
{
for (int i = MaxItems - 1; i >= 0; i--)
// Try to find a compatible surface with same parameters, and same offsets.
for (int i = 0; i < MaxItems; i++)
{
ref CacheItem item = ref _pool[i];
if (item.ReferenceCount == 0 && item.CodecId == codecId && item.Width == width && item.Height == height)
if (item.LumaOffset == lumaOffset &&
item.ChromaOffset == chromaOffset &&
item.Owner == decoder &&
item.Width == width &&
item.Height == height)
{
item.ReferenceCount = 1;
item.LumaOffset = lumaOffset;
item.ChromaOffset = chromaOffset;
item.ReferenceCount++;
surface = item.Surface;
MoveToFront(i);
break;
}
}
// If we failed to find a perfect match, now ignore the offsets.
// Search backwards to replace the oldest compatible surface,
// this avoids thrashing frequently used surfaces.
// Now we need to ensure that the surface is not in use, as we'll change the data.
if (surface == null)
{
for (int i = MaxItems - 1; i >= 0; i--)
{
ref CacheItem item = ref _pool[i];
if (item.ReferenceCount == 0 && item.Owner == decoder && item.Width == width && item.Height == height)
{
item.ReferenceCount = 1;
item.LumaOffset = lumaOffset;
item.ChromaOffset = chromaOffset;
surface = item.Surface;
if ((lumaOffset | chromaOffset) != 0)
{
SurfaceReader.Read(_gmm, surface, lumaOffset, chromaOffset);
}
MoveToFront(i);
break;
}
}
}
// If everything else failed, we try to create a new surface,
// and insert it on the pool. We replace the oldest item on the
// pool to avoid thrashing frequently used surfaces.
// If even the oldest item is in use, that means that the entire pool
// is in use, in that case we throw as there's no place to insert
// the new surface.
if (surface == null)
{
if (_pool[MaxItems - 1].ReferenceCount == 0)
{
surface = decoder.CreateSurface(width, height);
if ((lumaOffset | chromaOffset) != 0)
{
SurfaceReader.Read(_gmm, surface, lumaOffset, chromaOffset);
}
MoveToFront(i);
break;
MoveToFront(MaxItems - 1);
ref CacheItem item = ref _pool[0];
item.Surface?.Dispose();
item.ReferenceCount = 1;
item.LumaOffset = lumaOffset;
item.ChromaOffset = chromaOffset;
item.Width = width;
item.Height = height;
item.Owner = decoder;
item.Surface = surface;
}
}
}
// If everything else failed, we try to create a new surface,
// and insert it on the pool. We replace the oldest item on the
// pool to avoid thrashing frequently used surfaces.
// If even the oldest item is in use, that means that the entire pool
// is in use, in that case we throw as there's no place to insert
// the new surface.
if (surface == null)
{
if (_pool[MaxItems - 1].ReferenceCount == 0)
{
surface = decoder.CreateSurface(width, height);
if ((lumaOffset | chromaOffset) != 0)
else
{
SurfaceReader.Read(_gmm, surface, lumaOffset, chromaOffset);
throw new InvalidOperationException("No free slot on the surface pool.");
}
}
MoveToFront(MaxItems - 1);
ref CacheItem item = ref _pool[0];
item.Surface?.Dispose();
item.ReferenceCount = 1;
item.LumaOffset = lumaOffset;
item.ChromaOffset = chromaOffset;
item.Width = width;
item.Height = height;
item.CodecId = codecId;
item.Surface = surface;
}
else
{
throw new InvalidOperationException("No free slot on the surface pool.");
}
return surface;
}
return surface;
}
public void Put(ISurface surface)
{
for (int i = 0; i < MaxItems; i++)
lock (_pool)
{
ref CacheItem item = ref _pool[i];
if (item.Surface == surface)
for (int i = 0; i < MaxItems; i++)
{
item.ReferenceCount--;
Debug.Assert(item.ReferenceCount >= 0);
break;
ref CacheItem item = ref _pool[i];
if (item.Surface == surface)
{
item.ReferenceCount--;
Debug.Assert(item.ReferenceCount >= 0);
break;
}
}
}
}
@ -147,5 +153,22 @@ namespace Ryujinx.Graphics.Nvdec.Image
_pool[0] = temp;
}
}
public void Trim()
{
lock (_pool)
{
for (int i = 0; i < MaxItems; i++)
{
ref CacheItem item = ref _pool[i];
if (item.ReferenceCount == 0)
{
item.Surface?.Dispose();
item = default;
}
}
}
}
}
}