[llvm] [AMDGPU][Disassembler] Use target feature for `.amdhsa_reserve_xnack_mask` instead of hard code zero (PR #161771)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -2303,7 +2303,12 @@ Expected<bool> AMDGPUDisassembler::decodeCOMPUTE_PGM_RSRC1(
   KdStream << Indent << ".amdhsa_reserve_vcc " << 0 << '\n';
   if (!hasArchitectedFlatScratch())
     KdStream << Indent << ".amdhsa_reserve_flat_scratch " << 0 << '\n';
-  KdStream << Indent << ".amdhsa_reserve_xnack_mask " << 0 << '\n';
+  bool ReservedXnackMask = STI.hasFeature(AMDGPU::FeatureXNACK);
+#ifndef NDEBUG
+  assert(!ReservedXnackMask || STI.hasFeature(AMDGPU::FeatureSupportsXNACK));
----------------
rampitec wrote:

assert is already !NDEBUG.

https://github.com/llvm/llvm-project/pull/161771