Eliminate allocation in StreamAsIStream.Read

[bgrainger]

Description

The native/managed interop layer will allocate a byte array to marshal a native buffer to the StreamDescriptor.Read delegate. By using Span<byte> and the corresponding overload of Stream.Read, we can read directly into the caller's buffer when CManagedStream::Read calls this delegate.

Sample improvement by running a benchmark to create a BitmapImage with a MemoryStream StreamSource containing a 79KB PNG:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19044.1706 (21H2)
Intel Core i7-10875H CPU 2.30GHz, 1 CPU, 16 logical and 8 physical cores
.NET SDK=7.0.100-preview.4.22252.9
  [Host]     : .NET 6.0.5 (6.0.522.21309), X64 RyuJIT
  Job-FFRIND : .NET 6.0.5 (6.0.522.21309), X64 RyuJIT

Runtime=.NET 6.0  Toolchain=net6.0-windows

Method	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
Before	3.623 ms	0.0702 ms	0.0656 ms	984.3750	984.3750	984.3750	82 KB
After	3.619 ms	0.1386 ms	0.4088 ms	960.9375	960.9375	960.9375	3 KB

Customer Impact

Unnecessary allocation and memory copy between native/managed buffers.

Regression

No

Testing

Tested in real-world application via incorporation into Faithlife.Wpf.

Risk

Minimal.

Microsoft Reviewers: Open in CodeFlow

[lindexi]

Although only 4 milliseconds have been reduced, it is important to reduce 79KB of memory allocated.

[bgrainger]

Actually only 4 microseconds faster, which is well within the margin of error.

I'm not claiming that this PR is any faster; the primary goal is to eliminate managed allocations (which were previously equal in size to the input size of the image) and GCs.

[miloush]

No regression when the array is bigger than int? Doesn't AsSpan throw with negative numbers?

[bgrainger]

There's no regression because the old code cast the length to an int, too: https://github.com/dotnet/wpf/pull/6632/files#diff-43d84464891cb611646041f77d33aaf1c13800873ec0b7ff899f891985500d60L304

[bgrainger]

@dipeshmsft Could this PR be considered for the 8.x release series? It feels like a small change that has measurable allocation/GC benefits.

[pchaurasia14]

@bgrainger - Unfortunately, .NET8 RC candidates are locked in and we won't be able to take this in for .NET8. However, we'll consider this PR for upcoming CTP runs and have it merged asap.

[rchauhan18]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[rchauhan18]

Thanks @bgrainger for the contribution!