Support BC5 DDS encoding

[pepperoni505]

BC5 is currently not supported by Pillow. There is already some code in the repo that can read BC5 files. I would have made a PR for this but I'm not familiar with how this project is laid out.

[radarhere]

Hi. You may be interested to know that there is a PR in review that adds support for saving uncompressed DDS textures - #5402.

Where is the code that can read BC5 files? I look at https://github.com/python-pillow/Pillow/blob/master/src/PIL/DdsImagePlugin.py and I can only see code to read uncompressed textures, DXT1, DXT3, DXT5 and DX10? If you can't see it, attaching an BC5 file here that you can open with Pillow would be another way of answering my question.

[pepperoni505]

Hello. Sorry, I should have been more specific. I was talking about code in the following file: src/libImaging/BcnDecode.c. I don't know how easy it would be to add it into src/PIL/DdsImagePlugin.py.

[pepperoni505]

I tried modifying src/PIL/DdsImagePlugin.py to accept BC5 files, but it doesn't fully work. Here is a diff:

diff --git a/src/PIL/DdsImagePlugin.py b/src/PIL/DdsImagePlugin.py
index 1a7fe003..83c6c16d 100644
--- a/src/PIL/DdsImagePlugin.py
+++ b/src/PIL/DdsImagePlugin.py
@@ -97,6 +97,9 @@ DXT5_FOURCC = 0x35545844
 DXGI_FORMAT_R8G8B8A8_TYPELESS = 27
 DXGI_FORMAT_R8G8B8A8_UNORM = 28
 DXGI_FORMAT_R8G8B8A8_UNORM_SRGB = 29
+DXGI_FORMAT_BC5_TYPELESS = 82
+DXGI_FORMAT_BC5_UNORM = 83
+DXGI_FORMAT_BC5_SNORM = 84
 DXGI_FORMAT_BC7_TYPELESS = 97
 DXGI_FORMAT_BC7_UNORM = 98
 DXGI_FORMAT_BC7_UNORM_SRGB = 99
@@ -150,12 +153,18 @@ class DdsImageFile(ImageFile.ImageFile):
             elif fourcc == b"DXT5":
                 self.pixel_format = "DXT5"
                 n = 3
+            elif fourcc == b"BC5S":
+                self.pixel_format = "BC5"
+                n = 5
             elif fourcc == b"DX10":
                 data_start += 20
                 # ignoring flags which pertain to volume textures and cubemaps
                 dxt10 = BytesIO(self.fp.read(20))
                 dxgi_format, dimension = struct.unpack("<II", dxt10.read(8))
-                if dxgi_format in (DXGI_FORMAT_BC7_TYPELESS, DXGI_FORMAT_BC7_UNORM):
+                if dxgi_format in (DXGI_FORMAT_BC5_TYPELESS, DXGI_FORMAT_BC5_UNORM):
+                    self.pixel_format = "BC5"   
+                    n = 5
+                elif dxgi_format in (DXGI_FORMAT_BC7_TYPELESS, DXGI_FORMAT_BC7_UNORM):
                     self.pixel_format = "BC7"
                     n = 7
                 elif dxgi_format == DXGI_FORMAT_BC7_UNORM_SRGB:

I have the following sample code used to convert this to a .PNG image:

from PIL import Image

image = Image.open("C:/Users/pepperoni505/Downloads/test.dds")
image.save("C:/Users/pepperoni505/Downloads/test.png", 'png')

That generates this PNG image. As you can see, it's just an empty image.

[radarhere]

Just because I'm thinking ahead - if you're interested in Pillow because able to read these files before being able to write them, would you be able to generate a BC5 image that is smaller in file size, that you're happy for us to include in our test suite and distribute under the Pillow license?

[pepperoni505]

Yeah, I can send another image in a few hours. Any ideal size?

[radarhere]

128px by 128px?

[radarhere]

Ok, I have a fix ready - it turns out the alpha channel was just set to zero.
If it's not inconvenient to change my size request, 256 by 256.

[pepperoni505]

Here's a 256x256 BC5_UNORM image along with a PNG. Thanks for making the PR! Is it possible to support the other BC5 types?

[radarhere]

Thanks. I've incorporated the DDS file into the PR. I've left your PNG alone - for some reason, it has slightly different colors?

The easiest way to add the other formats... are you able to generate sample TYPELESS and SNORM files, for inclusion in Pillow, same as before?

[pepperoni505]

I managed to get a SNORM file but was unable to get a TYPELESS file.
BC5_SNORM.zip

[radarhere]

In trying to get Pillow to read the SNORM file, I came up with this.

Is that how your image should look?

[pepperoni505]

Yeah, that looks right. When I generated the image there were some weird artifacts as well

[radarhere]

Thanks. I've added SNORM to the PR.