[llvm] [AMDGPU] Recognize uniform expressions on tidx (PR #178533)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -1628,5 +1628,26 @@ GCNTTIImpl::getInstructionUniformity(const Value *V) const {
   if (isSourceOfDivergence(V))
     return InstructionUniformity::NeverUniform;
 
+  if (!ST->isWaveSizeKnown())
+    return InstructionUniformity::Default;
+
+  using namespace PatternMatch;
+  uint64_t C;
+  if (match(V, m_LShr(m_Intrinsic<Intrinsic::amdgcn_workitem_id_x>(),
+                      m_ConstantInt(C))) &&
+      C >= ST->getWavefrontSizeLog2())
+    return InstructionUniformity::AlwaysUniform;
+
+  uint64_t LaneMask = ST->getWavefrontSize() - 1;
+  if (match(V, m_And(m_Intrinsic<Intrinsic::amdgcn_workitem_id_x>(),
----------------
rampitec wrote:

By convention constant is always RHS. LLVM takes care of it.

https://github.com/llvm/llvm-project/pull/178533