[llvm] AMDGPU: VALU data fast-forwarding needs no s_delay (PR #205481)

[Jay Foad via llvm-commits]

================
@@ -1838,7 +1846,7 @@ define double @buffer_fat_ptr_agent_atomic_fmax_ret_f64__amdgpu_no_fine_grained_
 ; GFX12-NEXT:    v_max_num_f64_e32 v[0:1], v[4:5], v[4:5]
 ; GFX12-NEXT:    s_wait_storecnt 0x0
 ; GFX12-NEXT:    v_max_num_f64_e32 v[2:3], v[0:1], v[6:7]
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
----------------
jayfoad wrote:

New delay between `v_mov_b32_e32 v2, v4` and `v_mov_b32_e32 v3, v5`. We do not want this. I suspect it is due to the implicit superreg use operands added by copyPhysReg. 

https://github.com/llvm/llvm-project/pull/205481