[llvm] [AMDGPU] Allow negative offsets in scratch instructions (PR #166979)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,232 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -verify-machineinstrs -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 | FileCheck %s --check-prefix=GFX950
+
+; Ensure we don't crash with: "Cannot scavenge register in FI elimination!"
----------------
arsenm wrote:

Allowing negative offsets isn't a fix for this 

https://github.com/llvm/llvm-project/pull/166979