/// <summary>
/// Convert 'source.Length' <see cref="float"/> values normalized into [0..1] from 'source' into 'dest' buffer of <see cref="byte"/> values.
/// The values gonna be scaled up into [0-255] and rounded.
/// Based on:
/// <see>
/// <cref>http://lolengine.net/blog/2011/3/20/understanding-fast-float-integer-conversions</cref>
/// </see>
/// </summary>
internal static void BulkConvertNormalizedFloatToByte(ReadOnlySpan <float> source, Span <byte> dest)
{
GuardAvx2(nameof(BulkConvertNormalizedFloatToByte));
DebugGuard.IsTrue((source.Length % Vector <float> .Count) == 0, nameof(source), "source.Length should be divisable by Vector<float>.Count!");
if (source.Length == 0)
{
return;
}
ref Vector <float> srcBase = ref Unsafe.As <float, Vector <float> >(ref source.DangerousGetPinnableReference());
internal static void BulkConvertNormalizedFloatToByteClampOverflows(ReadOnlySpan <float> source, Span <byte> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
#if SUPPORTS_EXTENDED_INTRINSICS
ExtendedIntrinsics.BulkConvertNormalizedFloatToByteClampOverflowsReduce(ref source, ref dest);
#else
BasicIntrinsics256.BulkConvertNormalizedFloatToByteClampOverflowsReduce(ref source, ref dest);
#endif
FallbackIntrinsics128.BulkConvertNormalizedFloatToByteClampOverflowsReduce(ref source, ref dest);
// Deal with the remainder:
if (source.Length > 0)
{
ConvertNormalizedFloatToByteRemainder(source, dest);
}
}
internal static void BulkConvertByteToNormalizedFloat(ReadOnlySpan <byte> source, Span <float> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
#if NETCOREAPP2_1
ExtendedIntrinsics.BulkConvertByteToNormalizedFloatReduce(ref source, ref dest);
#else
BasicIntrinsics256.BulkConvertByteToNormalizedFloatReduce(ref source, ref dest);
#endif
FallbackIntrinsics128.BulkConvertByteToNormalizedFloatReduce(ref source, ref dest);
// Deal with the remainder:
if (source.Length > 0)
{
ConverByteToNormalizedFloatRemainder(source, dest);
}
}
internal static void ByteToNormalizedFloat(ReadOnlySpan <byte> source, Span <float> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
#if SUPPORTS_EXTENDED_INTRINSICS
ExtendedIntrinsics.ByteToNormalizedFloatReduce(ref source, ref dest);
#else
BasicIntrinsics256.ByteToNormalizedFloatReduce(ref source, ref dest);
#endif
// Also deals with the remainder from previous conversions:
FallbackIntrinsics128.ByteToNormalizedFloatReduce(ref source, ref dest);
// Deal with the remainder:
if (source.Length > 0)
{
ConvertByteToNormalizedFloatRemainder(source, dest);
}
}
internal static void BulkConvertNormalizedFloatToByteClampOverflowsReduce(
ref ReadOnlySpan <float> source,
ref Span <byte> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
int remainder = ImageMaths.Modulo4(source.Length);
int adjustedCount = source.Length - remainder;
if (adjustedCount > 0)
{
BulkConvertNormalizedFloatToByteClampOverflows(
source.Slice(0, adjustedCount),
dest.Slice(0, adjustedCount));
source = source.Slice(adjustedCount);
dest = dest.Slice(adjustedCount);
}
}
internal static void NormalizedFloatToByteSaturateReduce(
ref ReadOnlySpan <float> source,
ref Span <byte> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
int remainder = Numerics.Modulo4(source.Length);
int adjustedCount = source.Length - remainder;
if (adjustedCount > 0)
{
NormalizedFloatToByteSaturate(
source.Slice(0, adjustedCount),
dest.Slice(0, adjustedCount));
source = source.Slice(adjustedCount);
dest = dest.Slice(adjustedCount);
}
}
/// <summary>
/// SIMD optimized bulk implementation of <see cref="IPixel.PackFromVector4(Vector4)"/>
/// that works only with `count` divisible by <see cref="Vector{UInt32}.Count"/>.
/// </summary>
/// <param name="sourceColors">The <see cref="Span{T}"/> to the source colors.</param>
/// <param name="destVectors">The <see cref="Span{T}"/> to the dstination vectors.</param>
/// <param name="count">The number of pixels to convert.</param>
/// <remarks>
/// Implementation adapted from:
/// <see>
/// <cref>http://stackoverflow.com/a/5362789</cref>
/// </see>
/// TODO: We can replace this implementation in the future using new Vector API-s:
/// <see>
/// <cref>https://github.com/dotnet/corefx/issues/15957</cref>
/// </see>
/// </remarks>
internal static void ToVector4SimdAligned(Span <Rgba32> sourceColors, Span <Vector4> destVectors, int count)
{
if (!Vector.IsHardwareAccelerated)
{
throw new InvalidOperationException(
"Rgba32.PixelOperations.ToVector4SimdAligned() should not be called when Vector.IsHardwareAccelerated == false!");
}
DebugGuard.IsTrue(
count % Vector <uint> .Count == 0,
nameof(count),
"Argument 'count' should divisible by Vector<uint>.Count!");
var bVec = new Vector <float>(256.0f / 255.0f);
var magicFloat = new Vector <float>(32768.0f);
var magicInt = new Vector <uint>(1191182336); // reinterpreded value of 32768.0f
var mask = new Vector <uint>(255);
int unpackedRawCount = count * 4;
ref uint sourceBase = ref Unsafe.As <Rgba32, uint>(ref sourceColors.DangerousGetPinnableReference());
internal static void BulkConvertByteToNormalizedFloatReduce(
ref ReadOnlySpan <byte> source,
ref Span <float> dest)
{
DebugGuard.IsTrue(source.Length == dest.Length, nameof(source), "Input spans must be of same length!");
if (!IsAvailable)
{
return;
}
int remainder = ImageMaths.ModuloP2(source.Length, Vector <byte> .Count);
int adjustedCount = source.Length - remainder;
if (adjustedCount > 0)
{
BulkConvertByteToNormalizedFloat(source.Slice(0, adjustedCount), dest.Slice(0, adjustedCount));
source = source.Slice(adjustedCount);
dest = dest.Slice(adjustedCount);
}
}
internal static void PackFromRgbPlanes(
Configuration configuration,
ReadOnlySpan <byte> redChannel,
ReadOnlySpan <byte> greenChannel,
ReadOnlySpan <byte> blueChannel,
Span <Rgba32> destination)
{
DebugGuard.IsTrue(greenChannel.Length == redChannel.Length, nameof(greenChannel), "Channels must be of same size!");
DebugGuard.IsTrue(blueChannel.Length == redChannel.Length, nameof(blueChannel), "Channels must be of same size!");
DebugGuard.IsTrue(destination.Length > redChannel.Length, nameof(destination), "'destination' span should not be shorter than the source channels!");
#if SUPPORTS_RUNTIME_INTRINSICS
if (Avx2.IsSupported)
{
HwIntrinsics.PackFromRgbPlanesAvx2Reduce(ref redChannel, ref greenChannel, ref blueChannel, ref destination);
}
else
#endif
{
PackFromRgbPlanesScalarBatchedReduce(ref redChannel, ref greenChannel, ref blueChannel, ref destination);
}
PackFromRgbPlanesRemainder(redChannel, greenChannel, blueChannel, destination);
}