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

[Jay Foad via llvm-commits]

================
@@ -3143,6 +3143,7 @@ define amdgpu_ps void @insertelement_s_v16i16_v_s(ptr addrspace(4) inreg %ptr, i
 ; GFX11-NEXT:    v_dual_mov_b32 v4, s12 :: v_dual_mov_b32 v7, s15
 ; GFX11-NEXT:    v_mov_b32_e32 v6, s14
 ; GFX11-NEXT:    v_mov_b32_e32 v8, 0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
----------------
jayfoad wrote:

This is not wanted. I guess it's because v_movreld has an implicit use operand for its result registers, but AMDGPUInsertDelayAlu should ignore that.

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