Fix GX_NRM_NBT and GX_NRM_NBT3 vertex attribute handling

[Ahmad8864]

Fixes the crash path for #93. Aurora was crashing on GX_NRM_NBT / GX_NRM_NBT3 vertex attributes in a few places: comp_cnt_count only handled GX_NRM_XYZ, the VAT-A decoder ignored the bit-31 escape and collapsed NBT3 into NBT, and the shader generator tried to emit a nonexistent fetch_f32_9. All three are fixed here.

The follow-up commit addresses @magcius's review: NBT3 indexed mode now reads three separate indices per vertex from the DL buffer (one each for N, B, T) instead of treating it as a single-index NBT, and GX_TG_BINRM / GX_TG_TANGENT are supported as texgen sources, the binormal and tangent slices of GX_VA_NRM are loaded on-demand when a sampled texgen consumes them. The latest commit implements emboss bump mapping (GX_TG_BUMP0–7), offsetting the source texcoord by the light direction projected onto the binormal and tangent.

All 180 existing tests still pass, though I don't have an NBT test ROM to visually verify the output.

[dbalatoni13]

Thanks a lot! I will test it.

[magcius]

Even without implementing bump, you should still be able to add support for GX_TG_BINRM / GX_TG_TANGENT as texgen sources. I don't think this implements NBT3 indexing correctly either (you need to read three indices from the DL buffer).

[Ahmad8864]

Even without implementing bump, you should still be able to add support for GX_TG_BINRM / GX_TG_TANGENT as texgen sources. I don't think this implements NBT3 indexing correctly either (you need to read three indices from the DL buffer).

I'll take a look at adding texgen support and revisit the NBT3 indexing, thanks

[Ahmad8864]

@magcius @dbalatoni13 Sorry for the delay, addressed the NBT3 review feedback and added binormal/tangent texgen support. Would you have a chance to test?

[dbalatoni13]

Looks correct to me graphically

[Ahmad8864]

Hi @encounter does this PR look good to you?

[encounter]

It does, thanks for the implementation!