[clang-tools-extra] [compiler-rt] [llvm] [AMDGPU] Enable WMMA256bInsts + Wave32 for gfx1200/gfx1201 + SISchedule + TargetParser + Virtual FP4/MXFP4 (PR #201810)

[via cfe-commits]

clearnature wrote:

@krzysz00 Thanks for reviewing. I replied to the same concerns above — please see my response to @searlmc1 for the full explanation.

Quick summary:

1. **Virtual FP4 types** — These are LLVM infrastructure headers (like APFloat), NOT backend emulation. They define E2M1/E3M0 numerical format representations used by the WMMA/SWMMAC lowering. No runtime simulation logic.

2. **Chinese docs** — Development notes. Community can decide whether to keep or drop.

3. **compiler-rt fix** — Build blocker on Ubuntu 25.04 (GCC 15.2.0). Happy to split into separate PR.

4. **AI policy** — I confirmed policy compliance in my earlier reply. All code was personally written, reviewed, and tested on real RX 9060 XT hardware (4842/4853 AMDGPU tests pass). AI used only for formatting.

I am prepared to iterate on any part of this PR.

https://github.com/llvm/llvm-project/pull/201810