[llvm] [AMDGPU] Skip register uses in AMDGPUResourceUsageAnalysis (PR #133242)

[Diana Picus via llvm-commits]

================
@@ -12,7 +12,7 @@ declare float @llvm.fma.f32(float, float, float)
 ; CHECK:  v_mov_b32_e32 v{{[0-9]+}}, 0
 ; CHECK:  v_mov_b32_e32 v{{[0-9]+}}, 0
 ; It's probably OK if this is slightly higher:
-; CHECK: ; NumVgprs: 8
+; CHECK: ; NumVgprs: 5
----------------
rovka wrote:

I could use another pair of eyes on this test. From what I can tell, it uses v[4:7] (`buffer_store_dwordx4 v[4:7], off, s[0:3], 0`) but only defines v4. Am I missing smth, or is this just broken? 

https://github.com/llvm/llvm-project/pull/133242