[RS4GC] Handle uses of extractelement for conversion from vector to scalar base

[anna]

Summary

As mentioned in the comments, extractelement is special
since we actually want a scalar base for that element we extracted from
the vector (i.e. not a vector base).
This same logic should apply to uses of the extractelement such as phis
and selects which have the same BDV as the extractelement.

Added testcase showcases the problem where the BDV identification chokes
on the incorrect cast from vector to scalar for the phi use of
extractelement.

Tests-Run: make check

[anna]

llvm/lib/Transforms/Scalar/RewriteStatepointsForGC.cpp

949

Just realized this won't work in all cases since the base ptr identified may not dominate the derived ptr. Needs fixing.

[anna]

fixed dominator error.

[dantrushin]

LGTM, but what about test failure? Please fix it

[anna]

In D75704#1909467, @dantrushin wrote:

LGTM, but what about test failure? Please fix it

will do, thanks! forgot to update the test once I tested the patch internally.

[anna]

changed to early-return, updated comments and test.

[anna]

removed extra line. cleaned up whiitespace. NFC w.r.t. prev diff.

llvm/lib/Transforms/Scalar/RewriteStatepointsForGC.cpp

 
 #ifndef NDEBUG
   LLVM_DEBUG(dbgs() << "States after meet iteration:\n");
   for (auto Pair : States) {
     LLVM_DEBUG(dbgs() << " " << Pair.second << " for " << *Pair.first << "\n");
   }
 #endif
 
-  // Insert Phis for all conflicts
-  // TODO: adjust naming patterns to avoid this order of iteration dependency
+  // Handle extractelement instructions and their uses.
   for (auto Pair : States) {
     Instruction *I = cast<Instruction>(Pair.first);
     BDVState State = Pair.second;
     assert(!isKnownBaseResult(I) && "why did it get added?");
     assert(!State.isUnknown() && "Optimistic algorithm didn't complete!");
 
     // extractelement instructions are a bit special in that we may need to
     // insert an extract even when we know an exact base for the instruction.
     // The problem is that we need to convert from a vector base to a scalar
     // base for the particular indice we're interested in.
-    if (State.isBase() && isa<ExtractElementInst>(I) &&
-        isa<VectorType>(State.getBaseValue()->getType())) {
-      auto *EE = cast<ExtractElementInst>(I);
-      // TODO: In many cases, the new instruction is just EE itself.  We should
-      // exploit this, but can't do it here since it would break the invariant
-      // about the BDV not being known to be a base.
-      auto *BaseInst = ExtractElementInst::Create(
-          State.getBaseValue(), EE->getIndexOperand(), "base_ee", EE);
-      BaseInst->setMetadata("is_base_value", MDNode::get(I->getContext(), {}));
-      States[I] = BDVState(BDVState::Base, BaseInst);
-    }
+    if (!State.isBase() || !isa<ExtractElementInst>(I) ||
+        !isa<VectorType>(State.getBaseValue()->getType()))
+      continue;
+
+    // We need to handle uses of the extractelement that have the same vector
+    // base as well but the use is a scalar type. Since we cannot reuse the
+    // same BaseInst above (may not satisfy property that base pointer should
+    // always dominate derived pointer), we conservatively set this as conflict.
+    // Setting the base value for these conflicts is handled in the next loop
+    // which traverses States.
+    for (User *U : I->users()) {
+      auto *UseI = dyn_cast<Instruction>(U);
+      if (!UseI || !States.count(UseI))
+        continue;
+      if (!isa<VectorType>(UseI->getType()) && States[UseI] == State)
+        States[UseI] = BDVState(BDVState::Conflict);
+    }
 
+  // Insert Phis for all conflicts
+  // TODO: adjust naming patterns to avoid this order of iteration dependency
+  for (auto Pair : States) {
+    Instruction *I = cast<Instruction>(Pair.first);
+    BDVState State = Pair.second;
+    assert(!isKnownBaseResult(I) && "why did it get added?");
+    assert(!State.isUnknown() && "Optimistic algorithm didn't complete!");
+
     // Since we're joining a vector and scalar base, they can never be the
     // same.  As a result, we should always see insert element having reached
     // the conflict state.
     assert(!isa<InsertElementInst>(I) || State.isConflict());
 
     if (!State.isConflict())
       continue;
 

llvm/test/Transforms/RewriteStatepointsForGC/scalar-base-vector.ll

 ;
 entry:
   %base.i32 = bitcast <2 x i8 addrspace(1)*> %base to <2 x i32 addrspace(1)*>
   %vec = getelementptr i32, <2 x i32 addrspace(1)*> %base.i32, <2 x i64> %offsets
   call void @do_safepoint()
   %ptr = extractelement <2 x i32 addrspace(1)*> %vec, i32 0
   ret i32 addrspace(1)* %ptr
 }
+
+define void @test6() gc "statepoint-example" {
+; CHECK-LABEL: @test6(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    br label [[HEADER:%.*]]
+; CHECK:       header:
+; CHECK-NEXT:    [[TMP_BASE:%.*]] = phi i8 addrspace(1)* [ [[TMP6_BASE:%.*]], [[LATCH:%.*]] ], [ null, [[BB:%.*]] ], !is_base_value !0
+; CHECK-NEXT:    [[TMP:%.*]] = phi i8 addrspace(1)* [ [[TMP6:%.*]], [[LATCH]] ], [ undef, [[BB]] ]
+; CHECK-NEXT:    br label [[BB10:%.*]]
+; CHECK:       bb10:
+; CHECK-NEXT:    [[STATEPOINT_TOKEN:%.*]] = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 2882400000, i32 0, void ()* @spam, i32 0, i32 0, i32 0, i32 1, i8 addrspace(1)* [[TMP]], i8 addrspace(1)* [[TMP]], i8 addrspace(1)* [[TMP_BASE]])
+; CHECK-NEXT:    [[TMP_RELOCATED:%.*]] = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token [[STATEPOINT_TOKEN]], i32 9, i32 8)
+; CHECK-NEXT:    [[TMP_BASE_RELOCATED:%.*]] = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token [[STATEPOINT_TOKEN]], i32 9, i32 9)
+; CHECK-NEXT:    br label [[BB25:%.*]]
+; CHECK:       bb25:
+; CHECK-NEXT:    [[STATEPOINT_TOKEN1:%.*]] = call token (i64, i32, <2 x i8 addrspace(1)*> ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_v2p1i8f(i64 2882400000, i32 0, <2 x i8 addrspace(1)*> ()* @baz, i32 0, i32 0, i32 0, i32 0)
+; CHECK-NEXT:    [[TMP262:%.*]] = call <2 x i8 addrspace(1)*> @llvm.experimental.gc.result.v2p1i8(token [[STATEPOINT_TOKEN1]])
+; CHECK-NEXT:    [[BASE_EE:%.*]] = extractelement <2 x i8 addrspace(1)*> [[TMP262]], i32 0, !is_base_value !0
+; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <2 x i8 addrspace(1)*> [[TMP262]], i32 0
+; CHECK-NEXT:    br i1 undef, label [[BB7:%.*]], label [[LATCH]]
+; CHECK:       bb7:
+; CHECK-NEXT:    br label [[LATCH]]
+; CHECK:       latch:
+; CHECK-NEXT:    [[TMP6_BASE]] = phi i8 addrspace(1)* [ [[BASE_EE]], [[BB25]] ], [ [[BASE_EE]], [[BB7]] ], !is_base_value !0
+; CHECK-NEXT:    [[TMP6]] = phi i8 addrspace(1)* [ [[TMP27]], [[BB25]] ], [ [[TMP27]], [[BB7]] ]
+; CHECK-NEXT:    br label [[HEADER]]
+;
+bb:
+  br label %header
+
+header:                                              ; preds = %latch, %bb
+  %tmp = phi i8 addrspace(1)* [ %tmp6, %latch ], [ undef, %bb ]
+  br label %bb10
+
+bb10:                                             ; preds = %bb2
+  call void @spam() [ "deopt"(i8 addrspace(1)* %tmp) ]
+  br label %bb25
+
+bb25:                                             ; preds = %bb24
+  %tmp26 = call <2 x i8 addrspace(1)*> @baz()
+  %tmp27 = extractelement <2 x i8 addrspace(1)*> %tmp26, i32 0
+  br i1 undef, label %bb7, label %latch
+
+bb7:                                              ; preds = %bb25
+  br label %latch
+
+latch:                                              ; preds = %bb25, %bb7
+  %tmp6 = phi i8 addrspace(1)* [ %tmp27, %bb25 ], [ %tmp27, %bb7 ]
+  br label %header
+}
+
+declare void @spam()
+declare <2 x i8 addrspace(1)*> @baz()