[llvm] AMDGPU: Implement tensor_save and tensor_stop for gfx1250 (PR #146590)

[Changpeng Fang via llvm-commits]

================
@@ -1052,6 +1069,12 @@ let SubtargetPredicate = isGFX12Plus in {
   def GLOBAL_WBINV  : FLAT_Global_Invalidate_Writeback<"global_wbinv">;
 } // End SubtargetPredicate = isGFX12Plus
 
+let SubtargetPredicate = isGFX1250Plus in {
+
----------------
changpeng wrote:

> Is this blank line intended?

Yes. There is asymmetric empty lines for "let SubtargetPredicate = isGFX1250Plu", just want to avoid merge conflict.
Or we can fix the downstream.

https://github.com/llvm/llvm-project/pull/146590