[llvm] [AMDGPU] Convert PrologEpilogSGPRSpills from DenseMap to sorted vector (PR #90957)

[Jay Foad via llvm-commits]

================
@@ -18,9 +18,9 @@ define void @vector_reg_liverange_split() #0 {
 ; GFX90A-NEXT:    buffer_store_dword v40, off, s[0:3], s33 ; 4-byte Folded Spill
 ; GFX90A-NEXT:    buffer_store_dword a32, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
 ; GFX90A-NEXT:    s_mov_b64 exec, s[18:19]
+; GFX90A-NEXT:    v_writelane_b32 v40, s16, 4
 ; GFX90A-NEXT:    v_writelane_b32 v40, s28, 2
 ; GFX90A-NEXT:    v_writelane_b32 v40, s29, 3
----------------
jayfoad wrote:

Note these spills are now in ascending order of SGPR number. Same in other tests.

https://github.com/llvm/llvm-project/pull/90957