Using FORMAT_RF properly

diegzumillo

I wrote this piece of code in order to send some floats to a shader:

data = [ float(x),float(y),float(z)]
	var byte_array = PoolByteArray(data)
	var img = Image.new()
	img.create_from_data(3, 1, false, Image.FORMAT_RF , byte_array)
	var texture = ImageTexture.new()
	texture.create_from_image(img, 0)

But I get the error:

Expected data size of 12 bytes in Image::create(), got instead 3 bytes

So clearly I am missing some step.

cybereality

I believe you want to use FORMAT_RGBF for 3 components. FORMAT_RF is for 1 float.

diegzumillo

cybereality I want 1 float. My example is misleading because I used x, y and z. But really, the actual array will have an arbitrary number of floats and RF will be more convenient to use.

cybereality

Okay, I understand. A PoolByteArray stores data on individual bytes. A floating-point number is 4 bytes, so 3 should be 12 bytes total. You get the error message because the Image create_from_data() line is correct, but is expecting a byte array of 12 (4 bytes * 3), but that byte_array must not be initialized correctly. So the problem is on the first line. You likely need to use the function var2bytes(). So it may look like this:

var data_array = [x, y, z]
var byte_array = var2bytes(data_array)
# rest of code should work

diegzumillo

cybereality Oooh! we're getting warm here! now the error message is

Expected data size of 12 bytes in Image::create(), got instead 32 bytes.

Still not working but for the first time in a while I think we're on the right path.

diegzumillo

Ah, ok. So according to this:
https://docs.godotengine.org/en/stable/classes/class_float.html
Godot uses 64 bit for floats. It's really a double. Format-RF expects 32 bits though, so I need to somehow cast that to 32 bit.

cybereality

Okay, this is a hack, but it looks like it works.

func _ready():
	var data_array = [ 1.17, 2.05, 3.14 ]
	var img = Image.new()
	img.create(data_array.size(), 1, false, Image.FORMAT_RF)
	img.lock()
	for i in range(data_array.size()):
		img.set_pixel(i, 0, Color(data_array[i], 0.0, 0.0))
	img.unlock()
	var texture = ImageTexture.new()
	texture.create_from_image(img, 0)
	# test to make sure it works
	img.lock()
	for i in range(data_array.size()):
		print("pixel ", i, ": ", img.get_pixel(i, 0))
	img.unlock()
	print("texture width: ", texture.get_width(), ", height: ", texture.get_height())

diegzumillo

cybereality Good lord 😅
I'm marking this as the best answer, but I hope to find something less hacky. I tried something similar but using Vector2, because I know each component is stored as 32 bit floats.

Right now I am trying to understand what exactly is var2bytes doing. Because when I print out each byte it's always something weird and unintuitive. I think it's converting to byte some extra information I don't need. So once I figure out what it is exactly I might be able to exclude some elements from it or something. But that's for tomorrow's brain. Today's brain is tired.

I really appreciate the help, by the way! thank you.

cybereality

I did find this, which is along the lines of what you are saying, but I couldn't get it to work and it's even more of a hack than my solution.

https://godotengine.org/qa/102927/float-to-poolbytearray