[llvm] [AMDGPU] Use larger immediate values in S_NOP (PR #158990)

[Matt Arsenault via llvm-commits]

================
@@ -688,6 +688,14 @@ unsigned GCNSubtarget::getNSAThreshold(const MachineFunction &MF) const {
   return NSAThreshold;
 }
 
+unsigned GCNSubtarget::getSNopBits() const {
+  if (getGeneration() >= AMDGPUSubtarget::GFX12)
+    return 7;
+  if (getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
+    return 4;
+  return 3;
+}
----------------
arsenm wrote:

Pull definition into header? 

https://github.com/llvm/llvm-project/pull/158990