[MemoryDependence] Fix invariant group store

[wsmoses]

Summary

Fix bug in MemoryDependence [and thus GVN] for invariant group

[jdoerfert]

The test needs to be simplified, remove malloc and print, all geps and bitcasts, metadata, flags, ... everything except the store %p, %p_ptr and the load %p.

Explain in the commit message what the patch does, sure it fixes an error, but what error, e.g., a store is only invariant to a load if the pointer operands are the same.

Other than that, the logic is sound and fixes the problem.

[wsmoses]

Minimize test

[jdoerfert]

LG, with the updated commit message.

llvm/lib/Analysis/MemoryDependenceAnalysis.cpp

for (const Use &Us : Ptr->uses()) {
         if (GEP->hasAllZeroIndices()) {
           LoadOperandsQueue.push_back(U);
           continue;
         }
 
       // If we hit load/store with the same invariant.group metadata (and the
       // same pointer operand) we can assume that value pointed by pointer
       // operand didn't change.
-      if ((isa<LoadInst>(U) || isa<StoreInst>(U)) &&
+      if ((isa<LoadInst>(U) ||
+           (isa<StoreInst>(U) &&
+            cast<StoreInst>(U)->getPointerOperand() == Ptr)) &&
           U->hasMetadata(LLVMContext::MD_invariant_group))
         ClosestDependency = GetClosestDependency(ClosestDependency, U);
     }
   }
 
   if (!ClosestDependency)
     return MemDepResult::getUnknown();
   if (ClosestDependency->getParent() == BB)

llvm/test/Transforms/GVN/storeinvgroup.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -gvn -S -o - < %s | FileCheck %s
+
+; ModuleID = 'tc.ll'
+source_filename = "Main.cpp"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.c = type { double, double }
+
+@.str = private unnamed_addr constant [4 x i8] c"%f\0A\00", align 1
+
+; Function Attrs: norecurse uwtable mustprogress
+define i32 @main() local_unnamed_addr {
+; CHECK-LABEL: @main(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CALL:%.*]] = tail call noalias i8* @malloc(i64 32)
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[CALL]] to %struct.c*
+; CHECK-NEXT:    [[A1:%.*]] = getelementptr inbounds [[STRUCT_C:%.*]], %struct.c* [[TMP0]], i64 1, i32 0
+; CHECK-NEXT:    store double 1.234500e+00, double* [[A1]], align 8
+; CHECK-NEXT:    [[ARRAYIDX7_I_I:%.*]] = getelementptr inbounds i8, i8* [[CALL]], i64 16
+; CHECK-NEXT:    [[MALLOCCALL19_I:%.*]] = call noalias i8* @malloc(i64 8)
+; CHECK-NEXT:    [[_MALLOCCACHE_I:%.*]] = bitcast i8* [[MALLOCCALL19_I]] to %struct.c**
+; CHECK-NEXT:    [[C2:%.*]] = bitcast i8* [[ARRAYIDX7_I_I]] to %struct.c*
+; CHECK-NEXT:    store %struct.c* [[C2]], %struct.c** [[_MALLOCCACHE_I]], align 8, !invariant.group !2
+; CHECK-NEXT:    [[A10_I_I:%.*]] = getelementptr inbounds [[STRUCT_C]], %struct.c* [[C2]], i64 0, i32 0
+; CHECK-NEXT:    [[CALL4:%.*]] = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), double 1.234500e+00)
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %call = tail call noalias i8* @malloc(i64 32)
+  %0 = bitcast i8* %call to %struct.c*
+  %a1 = getelementptr inbounds %struct.c, %struct.c* %0, i64 1, i32 0
+  store double 1.23450000e+00, double* %a1, align 8
+  %arrayidx7.i.i = getelementptr inbounds i8, i8* %call, i64 16
+  %malloccall19.i = call noalias i8* @malloc(i64 8)
+  %_malloccache.i = bitcast i8* %malloccall19.i to %struct.c**
+  %c2 = bitcast i8* %arrayidx7.i.i to %struct.c*
+  %g3 = getelementptr inbounds %struct.c*, %struct.c** %_malloccache.i, i64 0
+  store %struct.c* %c2, %struct.c** %g3, align 8, !invariant.group !7
+  %a10.i.i = getelementptr inbounds %struct.c, %struct.c* %c2, i64 0, i32 0
+  %iload = load double, double* %a10.i.i, align 8, !invariant.group !8
+  %call4 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), double %iload)
+  ret i32 0
+}
+
+; Function Attrs: inaccessiblememonly nofree nounwind willreturn
+declare dso_local noalias noundef i8* @malloc(i64 noundef)
+
+; Function Attrs: nofree nounwind
+declare dso_local noundef i32 @printf(i8* nocapture noundef readonly, ...)
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 13.0.0 (git%github.com:llvm/llvm-project d163e75c81c1609855b98cbdffa0141c70d8578d)"}
+!2 = distinct !{!2, !3}
+!3 = !{!"llvm.loop.mustprogress"}
+!4 = !{!5}
+!5 = distinct !{!5, !6, !"diffe_Z1pP1cPd: %q"}
+!6 = distinct !{!6, !"diffe_Z1pP1cPd"}
+!7 = distinct !{}
+!8 = distinct !{}
+!9 = distinct !{!9, !3}