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

[Jay Foad via llvm-commits]

================
@@ -622,9 +622,11 @@ define double @global_agent_atomic_fmax_ret_f64__amdgpu_no_fine_grained_memory(p
 ; GFX12-NEXT:    s_or_b32 s0, vcc_lo, s0
 ; GFX12-NEXT:    s_wait_alu depctr_sa_sdst(0)
 ; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s0
+; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
 ; GFX12-NEXT:    s_cbranch_execnz .LBB6_1
 ; GFX12-NEXT:  ; %bb.2: ; %atomicrmw.end
 ; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s0
+; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
----------------
jayfoad wrote:

New delay between SALU instruction that writes to exec and VALU instruction that implicitly reads from exec.

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