[juliotec]

clear_tex already leaves BC images out of vkCmdClearColorImage, but not ASTC
ones. On a GPU that samples ASTC natively the image stays in ASTC blocks, and
vkCmdClearColorImage does not take a compressed format, and the region path only
converts the rectangle to BC blocks, so the clear was invalid and some drivers
hang on it. A colour written over ASTC blocks would be garbage anyway.

What changed (gpu/vk/vk_xfer.cpp, clear_tex): a clear of an image whose format
is ASTC is skipped with a one-time warning (warn 14). Decoded ASTC textures
(RGBA8) and every other format are cleared as before.

How tested: release build; Zelda OoT on Radeon RX 9070 XT (Windows,
Adrenalin).

Assisted-by: GitHub Copilot

[Joey-Magnus]

CI never ran here. I merged this with #2719, #2741, #2742 and #2743 onto current main. I compiled vk_xfer.cpp with clang++ using the back end's flags and ran tests/run_all.py with the Vulkan headers. None of the failures come from this PR, and test_bc_raw_targets.py and test_vulkan_clear_mip.py pass.

The guard is right. vkCmdClearColorImage must not be given a compressed format, and with astc_native() the image keeps the ASTC VkFormat, so the whole-image branch at gpu/vk/vk_xfer.cpp:340 would do exactly that. BC3-kept ASTC still goes down the raw path as before.

Two notes:

• AMD and NVIDIA desktop drivers don't expose textureCompressionASTC_LDR, so on the RX 9070 XT you tested, this code never runs. Only Intel iGPUs and similar would reach it. The "some drivers hang" claim would be good to back with a log or a validation-layer message from such a device, or the description should say it is from the spec.
• The early return also skips texture_written(s.tex) at :404, so the texture is not marked GPU-written. Since nothing was written, that seems right, but please confirm it was intended.