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

[Jay Foad via llvm-commits]

================
@@ -31,6 +31,7 @@ define amdgpu_gfx void @basic_test(i32 %x, i32 inreg %c, ptr addrspace(1) %ptr)
 ; DAGISEL-NEXT:    s_mov_b32 s1, good_callee at abs32@hi
 ; DAGISEL-NEXT:    s_mov_b32 s0, good_callee at abs32@lo
 ; DAGISEL-NEXT:    s_wait_alu depctr_sa_sdst(0)
+; DAGISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
----------------
jayfoad wrote:

I don't think this is wanted. I guess s_swappc_b64 has an implicit use of v1 indicating that the callee uses v1?

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