[llvm] [AMDGPU] Rewrite GFX12 SGPR hazard handling to dedicated pass (PR #118750)

[Jay Foad via llvm-commits]

================
@@ -14918,9 +14915,10 @@ define <2 x bfloat> @flat_agent_atomic_fmax_ret_v2bf16__offset12b_neg__amdgpu_no
 ; GFX12-NEXT:    v_bfe_u32 v7, v3, 16, 1
 ; GFX12-NEXT:    v_or_b32_e32 v9, 0x400000, v3
 ; GFX12-NEXT:    v_cmp_u_f32_e64 s0, v3, v3
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
----------------
jayfoad wrote:

Example where s_delay_alu is removed

https://github.com/llvm/llvm-project/pull/118750