[llvm] [AMDGPU][IGLP] SingleWaveOpt: Cache DSW Counters from PreRA (PR #67759)

[via llvm-commits]

================
@@ -1105,14 +1112,20 @@ class MFMASmallGemmSingleWaveOpt final : public IGLPStrategy {
   }
 };
 
+static unsigned DSWCount = 0;
+static unsigned DSWWithPermCount = 0;
+static unsigned DSWWithSharedVMEMCount = 0;
+
 void MFMASmallGemmSingleWaveOpt::applyIGLPStrategy(
     DenseMap<int, SUnitsToCandidateSGsMap> &SyncedInstrs,
-    DenseMap<int, SmallVector<SchedGroup, 4>> &SyncedSchedGroups) {
+    DenseMap<int, SmallVector<SchedGroup, 4>> &SyncedSchedGroups,
+    bool IsPostRA) {
   unsigned MFMACount = 0;
-  unsigned DSWCount = 0;
-  unsigned DSWWithPermCount = 0;
-  unsigned DSWWithSharedVMEMCount = 0;
   unsigned DSRCount = 0;
+
+  assert((IsPostRA ||
+          DSWCount == DSWWithPermCount == DSWWithSharedVMEMCount == 0) &&
----------------
mikaelholmen wrote:

Hi @jrbyrnes 
I don't know this code at all, but gcc warned on this assert and suggested some paratheses, so it caught my eye.
Does it really check what the assert string says?
I think e.g. the assert passes if the counts are 1, 1, 0 but the string says all should be 0?


https://github.com/llvm/llvm-project/pull/67759