Assorted PS3 texture export/conversion fixes

[hydnhntr]

I finally found the time to wrap this and some other issues that cropped up along the way.

Reversed channels

In the ‘Swap channels for DDS’ block, each pixel’s bytes were first reversed ([A, R, G, B] → [B, G, R, A]), before the individual channels were extracted using the InR/InG/InB/InA remap offsets. The remapping offsets describe the pixel’s original, un-reversed byte layout, so applying them after the reversal incorrectly read each channel.

	BCED00393/projects/gt5p/arcade/JP/image/gt5p/common/hot.dds
Before
After

Linear row pitch

Uncompressed linear layout textures can have their rows padded to a hardware aligned pitch (assuming 1024 bytes), wider than the actual texture, eg

GTHD_Logo_H.dds (420×120) has a pitch of 2048 bytes per row, but only 1680 bytes of actual pixel data.

	NPUA80019/project-ps3/image/gthd/GTHD_Logo_H.dds
Before
After

DDS mip-count

Previosuly when extracting BCED00393/character/ma__0051 (5 textures, each with up to 10 mips) it’d fault with Offset and length were out of bounds....

Now all 5 textures convert but only the base level (0), as mip export isn’t a feature.

Does this need to be flipped vertically?

BufferId reference

PGLUCellTextureInfo.Read() never sets BufferId (unlike the PS4/PSP readers) and stays as 0. TextureSet3 resolved each texture’s buffer via Buffers[(int)textureInfo.BufferId], so with a multi-buffer container any texture after the first would still read the first (0) and fault.

BCED00393/crs/race.txs (6 textures) used to throw (Specified argument was out of the range of valid values.).

Is this supposed to be noise?

Testing

I picked samples from GTHD and GT5P which have a combo of features to validate these changes against.

2D

Index	Format	Layout	Dimensions	Mips	Pitch bytes	Row bytes
NPUA80019/project-ps3/image/component/dialog_head2.png
–	A8R8G8B8	Linear	2×20	1	8	8
NPUA80019/project-ps3/image/component/numberinput_up.png
–	A8R8G8B8	Linear	11×8	1	64	44
NPUA80019/project-ps3/image/arcade/common/popup_item_button.png
–	A8R8G8B8	Linear	155×24	1	1024	620
BCED00393/projects/gt5p/race/GB/image/race_display/ps3/GT4_meter_texture_point.png
–	A8R8G8B8	Linear	9×9	1	64	36
BCED00393/projects/component/image/component/scr2_2.png Smallest dimensions (12px square)
–	A8R8G8B8	Linear	6×2	1	32	24
NPUA80019/project-ps3/image/gthd/GTHD_Logo_H.dds Channel and pitch bugs
–	A8R8G8B8	Linear	420×120	1	2048	1680
BCED00393/projects/gt5p/arcade/JP/image/gt5p/common/hot.dds Channel bug
–	A8R8G8B8	Linear	68×72	1	272	272
NPUA80019/project-ps3/image/guide/notice/JP/notice_0_480.dds
–	DXT1	Linear	640×480	1	–	–
NPUA80019/project-ps3/image/option/white.dds
–	DXT1	Swizzled	4×4	1	–	–
NPUA80019/project-ps3/image/component/ps3/btn_sys1_1.dds
–	DXT3	Linear	32×56	1	–	–
NPUA80019/project-ps3/image/gthd/manual/HD_Ctl_Menu.dds
–	DXT3	Linear	1648×748	1	–	–
NPUA80019/project-ps3/image/component/ps3/dot.dds
–	DXT5	Linear	12×12	1	–	–
NPUA80019/project-ps3/image/quick/line_dot_q.dds
–	DXT5	Linear	4×100	1	–	–
NPUA80019/project-ps3/image/gthd/icon_ranking.dds
–	DXT5	Linear	116×88	1	–	–
NPUA80019/project-ps3/image/gthd/panel_nomal.dds
–	DXT5	Linear	336×336	1	–	–
NPUA80019/project-ps3/image/guide/direction/JP/direction_07_1080.dds
–	DXT5	Linear	1648×748	1	–	–
BCED00393/character/ma__0051 Mip-chain example
0	A8R8G8B8	Swizzled	256×256	9	–	–
1	DXT5	Swizzled	256×256	9	–	–
2	DXT5	Swizzled	128×256	9	–	–
3	DXT5	Swizzled	512×512	10	–	–
4	DXT5	Swizzled	256×256	9	–	–
BCED00393/crs/race.txs Only D8R8G8B8 example (textures 1–3)
0	A8R8G8B8	Swizzled	128×64	1	–	–
1	D8R8G8B8	Swizzled	128×128	1	–	–
2	D8R8G8B8	Swizzled	256×256	1	–	–
3	D8R8G8B8	Swizzled	64×64	1	–	–
4	A8R8G8B8	Swizzled	128×128	1	–	–
5	A8R8G8B8	Swizzled	128×128	1	–	–

Cubemap

Format	Layout	Dimensions	Mips
BCED00393/scene/texset/tx0001
A8R8G8B8	Swizzled	128×128	1
NPUA80019/projects-ps3/trial/US/image/gthd/colorchipenv.dds Cubemap with mip levels and exhibited the DDS mip-count bug
A8R8G8B8	Swizzled	128×128	8

[Nenkai]

I don't completely remember why I added endian swap in the first place, but it must've been a good reason

Namely I believed that the pixel components in the pixel buffers are in big-endian according to PPC in the first place, regardless of remap

If you can, I would make sure that it works correctly across the majority of textures, namely i'd check GTHD, GT5P/5 and track textures

[hydnhntr]

Edit: outdated, see PR body for latest changes.

---

Small update after looking at GTHD Concept’s UI textures:

Format	Output	Asset
DXT45		NPUA80019/project-ps3/image/component/ps3/cursor_chrome.dds
A8R8G8B8		NPUA80019/project-ps3/image/cursor/cursor_chrome.dds

cursors.zip

---

The uncompressed A8R8G8B8 version is currently ignoring the pitch, and includes the padding in the output.

I’m unsure how to handle this, whether to rejig the data inside CreateDDSData() or whether Pfimage can be passed a stride.

PDTools/PDTools.Files/Textures/PS3/PGLUCellTextureInfo.cs

Line 440 in 5bb714c

var i = Image.LoadPixelData<Bgra32>(dds.Data, (int)Utils.MiscUtils.AlignValue((uint)dds.Width, 4), dds.Height);

I also needed to remove the manual alignment above, as it was resulting in a mismatch in the expected length of data.

[hydnhntr]

@Nenkai 11 months later and finally some updates ready for review