A note to this one: both the array containing the texture regions as well as the data pointer are accessed asynchronously. So, do not put them on the stack as you probably get access violations or weird errors.

By Simon Oehrl on 2020-07-01T18:50:17 (imported from GitLab)

For a synchronized update, this can be used:

TArray<FColor> ColorData;
// Init Data

// Create texture if it doesn't exist
FTexture2DMipMap& Mip = Texture->PlatformData->Mips[0];
void* TextureData = Mip.BulkData.Lock( LOCK_READ_WRITE );
FMemory::Memcpy( TextureData, ColorData.GetData(), (X * Y * sizeof(FColor))); // can just loop through values as well 
Mip.BulkData.Unlock( );
Texture->UpdateResource();

By David Gilbert on 2020-07-02T09:24:04 (imported from GitLab)

Yes, that's a nice way to do this. However, but this explicitly introduces synchronization between CPU/GPU introducing quite some overhead.

By Simon Oehrl on 2020-07-02T09:28:21 (imported from GitLab)

Yeah. The code I ended up with in the end is based on the usage I found in the engine itself - pretty sure it could be improved:

FUpdateTextureRegion2D* Region = new FUpdateTextureRegion2D(
		InitialTextureMarker.X, 
		InitialTextureMarker.Y, 
		InitialTextureMarker.X,
		InitialTextureMarker.Y, Line.Num(), 1);	

	// Copy for now - no need to do that generally
	TArray<FVector4>* TextureData = new TArray<FVector4>(RawLineData);

        // Moving the raw arraw data instead of a copy is a lot quicker, if your data structure allows that:
	TArray<FVector4>* TextureData = new TArray<FVector4>(MoveTemp(RawLineData));


	
	PositionTexture->UpdateTextureRegions(0, 1, Region, TextureWidth * sizeof(FVector4), sizeof(FVector4), (uint8*)TextureData->GetData(),
		[](auto InTextureData, auto InRegions)
	{
		delete InTextureData;
		delete InRegions;
	});

By David Gilbert on 2020-07-02T09:54:17 (imported from GitLab)