#@climbs rocks Would you be able to help

1 messages ยท Page 1 of 1 (latest)

solar skiff
#

Sorry for the delayed response, I took a look at repro'ing this morning on 5.7 and it seems to be working:

#

We haven't touched the texture ops much since 5.6, so it should be very similar

#

aha, it seems that

float4 Color = (0, 0, 0, 1);
doesn't play nice. It gives this warning and the same behavior you're seeing

comma expression used where a constructor list may have been intended [-Wcomma-in-init]

I think if you use float4 Color = float4(0, 0, 0, 1); it will probably work

hasty mantle
hasty mantle
#

I have one more thought on this topic, in the image I can see the PCGTextureData as points when debugging and the texture is packed into $Color etc. But I can't force this to readback to CPU, which would be nice.

#

At the moment I'm doing some cringe things in the texture processing kernel to create points representing the texels, because I can read points back to CPU

#

Or just a way to get the actual texture resource and copy it over into a new asset would shorten this whole journey ๐Ÿ˜…

hasty mantle
#

I was trying to use "samplepointlocal" on pcgtexturedata but I don't think this was working either

#

Cc @solar skiff

solar skiff
#

I have one more thought on this topic, in the image I can see the PCGTextureData as points when debugging and the texture is packed into $Color etc. But I can't force this to readback to CPU, which would be nice.

Yeah, this is unfortunate but currently the compute graph export doesn't do a CPU initialization of the texture, so using a kernel to write texels into points (like you've already done) is the recommended way for now. We'd also like to add a texture export node so you can bake your texture into an asset for use in the graph or anywhere. Both are scoped out of 5.7 currently crying (but they're highly requested so I may take a look anyways)

I was trying to use "samplepointlocal" on pcgtexturedata but I don't think this was working either

Yeah, it will fail because internally the data has not read any color data back to the CPU, and it won't do that on demand without being re-initialized (which isn't exposed in the graph but could be done in C++ I think)

hasty mantle