[llvm] [AMDGPU] Don't fix the scavenge slot at offset 0 (PR #79136)

[Diana Picus via llvm-commits]

================
@@ -10315,8 +10315,8 @@ define amdgpu_kernel void @test_limited_sgpr(ptr addrspace(1) %out, ptr addrspac
 ; GFX6-NEXT:    s_mov_b64 exec, s[6:7]
 ; GFX6-NEXT:    s_mov_b64 s[6:7], exec
 ; GFX6-NEXT:    s_mov_b64 exec, 0xff
-; GFX6-NEXT:    s_mov_b32 s34, 0x84800
 ; GFX6-NEXT:    buffer_store_dword v4, off, s[40:43], 0
+; GFX6-NEXT:    s_mov_b32 s34, 0x84800
----------------
rovka wrote:

I'm not sure why the scheduling changes here, but it seems pretty harmless.

https://github.com/llvm/llvm-project/pull/79136