[llvm] [TableGen] Introduce a less aggressive suppression for HwMode Decoder… (PR #86060)

[Craig Topper via llvm-commits]

================
@@ -2417,21 +2436,67 @@ static bool Check(DecodeStatus &Out, DecodeStatus In) {
 
 // Collect all HwModes referenced by the target for encoding purposes,
 // returning a vector of corresponding names.
-static void
-collectHwModesReferencedForEncodings(const CodeGenHwModes &HWM,
-                                     std::vector<StringRef> &Names) {
+static void collectHwModesReferencedForEncodings(
+    const CodeGenHwModes &HWM, std::vector<StringRef> &Names,
+    NamespacesHwModesMap &NamespacesWithHwModes) {
   SmallBitVector BV(HWM.getNumModeIds());
   for (const auto &MS : HWM.getHwModeSelects()) {
     for (const HwModeSelect::PairType &P : MS.second.Items) {
-      if (P.second->isSubClassOf("InstructionEncoding"))
+      if (P.second->isSubClassOf("InstructionEncoding")) {
+        std::string DecoderNamespace =
+            std::string(P.second->getValueAsString("DecoderNamespace"));
+        if (P.first == DefaultMode) {
+          NamespacesWithHwModes[DecoderNamespace][""] = 1;
----------------
topperc wrote:

Thanks, I thought a std::set would probably work.

https://github.com/llvm/llvm-project/pull/86060