diff --git a/llvm/include/llvm/IR/KnowledgeRetention.h b/llvm/include/llvm/IR/KnowledgeRetention.h
--- a/llvm/include/llvm/IR/KnowledgeRetention.h
+++ b/llvm/include/llvm/IR/KnowledgeRetention.h
@@ -31,6 +31,11 @@
 /// The returned instruction is not inserted anywhere.
 IntrinsicInst *buildAssumeFromInst(Instruction *I);
 
+/// Calls BuildAssumeFromInst and if the resulting llvm.assume is valid insert
+/// if before I. This is usually what need to be done to salvage the knowledge
+/// contained in the instruction I.
+void salvageKnowledge(Instruction* I);
+
 /// It is possible to have multiple Value for the argument of an attribute in
 /// the same llvm.assume on the same llvm::Value. This is rare but need to be
 /// dealt with.
diff --git a/llvm/lib/IR/KnowledgeRetention.cpp b/llvm/lib/IR/KnowledgeRetention.cpp
--- a/llvm/lib/IR/KnowledgeRetention.cpp
+++ b/llvm/lib/IR/KnowledgeRetention.cpp
@@ -232,6 +232,11 @@
   return Builder.build();
 }
 
+void llvm::salvageKnowledge(Instruction* I) {
+  if (Instruction* Intr = buildAssumeFromInst(I))
+    Intr->insertBefore(I);
+}
+
 static bool bundleHasArgument(const CallBase::BundleOpInfo &BOI,
                                unsigned Idx) {
   return BOI.End - BOI.Begin > Idx;
diff --git a/llvm/lib/Transforms/Scalar/EarlyCSE.cpp b/llvm/lib/Transforms/Scalar/EarlyCSE.cpp
--- a/llvm/lib/Transforms/Scalar/EarlyCSE.cpp
+++ b/llvm/lib/Transforms/Scalar/EarlyCSE.cpp
@@ -38,6 +38,7 @@
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/KnowledgeRetention.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/PatternMatch.h"
@@ -947,6 +948,7 @@
         continue;
       }
 
+      salvageKnowledge(&Inst);
       salvageDebugInfoOrMarkUndef(Inst);
       removeMSSA(Inst);
       Inst.eraseFromParent();
@@ -1013,6 +1015,7 @@
                 cast<ConstantInt>(KnownCond)->isOne()) {
               LLVM_DEBUG(dbgs()
                          << "EarlyCSE removing guard: " << Inst << '\n');
+              salvageKnowledge(&Inst);
               removeMSSA(Inst);
               Inst.eraseFromParent();
               Changed = true;
@@ -1048,6 +1051,7 @@
           Changed = true;
         }
         if (isInstructionTriviallyDead(&Inst, &TLI)) {
+          salvageKnowledge(&Inst);
           removeMSSA(Inst);
           Inst.eraseFromParent();
           Changed = true;
@@ -1073,6 +1077,7 @@
         if (auto *I = dyn_cast<Instruction>(V))
           I->andIRFlags(&Inst);
         Inst.replaceAllUsesWith(V);
+        salvageKnowledge(&Inst);
         removeMSSA(Inst);
         Inst.eraseFromParent();
         Changed = true;
@@ -1133,6 +1138,7 @@
           }
           if (!Inst.use_empty())
             Inst.replaceAllUsesWith(Op);
+          salvageKnowledge(&Inst);
           removeMSSA(Inst);
           Inst.eraseFromParent();
           Changed = true;
@@ -1176,6 +1182,7 @@
         }
         if (!Inst.use_empty())
           Inst.replaceAllUsesWith(InVal.first);
+        salvageKnowledge(&Inst);
         removeMSSA(Inst);
         Inst.eraseFromParent();
         Changed = true;
@@ -1228,6 +1235,7 @@
           LLVM_DEBUG(dbgs() << "Skipping due to debug counter\n");
           continue;
         }
+        salvageKnowledge(&Inst);
         removeMSSA(Inst);
         Inst.eraseFromParent();
         Changed = true;
@@ -1263,6 +1271,7 @@
             if (!DebugCounter::shouldExecute(CSECounter)) {
               LLVM_DEBUG(dbgs() << "Skipping due to debug counter\n");
             } else {
+              salvageKnowledge(&Inst);
               removeMSSA(*LastStore);
               LastStore->eraseFromParent();
               Changed = true;
diff --git a/llvm/test/Transforms/EarlyCSE/guards.ll b/llvm/test/Transforms/EarlyCSE/guards.ll
--- a/llvm/test/Transforms/EarlyCSE/guards.ll
+++ b/llvm/test/Transforms/EarlyCSE/guards.ll
@@ -10,10 +10,16 @@
 define i32 @test0(i32* %ptr, i1 %cond) {
 ; We can do store to load forwarding over a guard, since it does not
 ; clobber memory
-; CHECK-LABEL: @test0(
-; CHECK-NEXT:    store i32 40, i32* [[PTR:%.*]]
-; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
-; CHECK-NEXT:    ret i32 40
+; NO_ASSUME-LABEL: @test0(
+; NO_ASSUME-NEXT:    store i32 40, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
+; NO_ASSUME-NEXT:    ret i32 40
+;
+; USE_ASSUME-LABEL: @test0(
+; USE_ASSUME-NEXT:    store i32 40, i32* [[PTR:%.*]]
+; USE_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    ret i32 40
 ;
 
   store i32 40, i32* %ptr
@@ -24,10 +30,16 @@
 
 define i32 @test1(i32* %val, i1 %cond) {
 ; We can CSE loads over a guard, since it does not clobber memory
-; CHECK-LABEL: @test1(
-; CHECK-NEXT:    [[VAL0:%.*]] = load i32, i32* [[VAL:%.*]]
-; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: @test1(
+; NO_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[VAL:%.*]]
+; NO_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: @test1(
+; USE_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[VAL:%.*]]
+; USE_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[COND:%.*]]) [ "deopt"() ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[VAL]], i64 4), "nonnull"(i32* [[VAL]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
 
   %val0 = load i32, i32* %val
@@ -205,12 +217,21 @@
 define void @test08(i32 %a, i32 %b, i32* %ptr) {
 ; Check that we deal correctly with stores when removing guards in the same
 ; block in case when the condition is not recalculated.
-; CHECK-LABEL: @test08(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    store i32 100, i32* [[PTR:%.*]]
-; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
-; CHECK-NEXT:    store i32 400, i32* [[PTR]]
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test08(
+; NO_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; NO_ASSUME-NEXT:    store i32 100, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
+; NO_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test08(
+; USE_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; USE_ASSUME-NEXT:    store i32 100, i32* [[PTR:%.*]]
+; USE_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; USE_ASSUME-NEXT:    ret void
 ;
 
   %cmp = icmp eq i32 %a, %b
@@ -228,21 +249,39 @@
 ; Similar to test08, but with more control flow.
 ; TODO: Can we get rid of the store in the end of entry given that it is
 ; post-dominated by other stores?
-; CHECK-LABEL: @test09(
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    store i32 100, i32* [[PTR:%.*]]
-; CHECK-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
-; CHECK-NEXT:    store i32 400, i32* [[PTR]]
-; CHECK-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
-; CHECK:       if.true:
-; CHECK-NEXT:    store i32 500, i32* [[PTR]]
-; CHECK-NEXT:    br label [[MERGE:%.*]]
-; CHECK:       if.false:
-; CHECK-NEXT:    store i32 600, i32* [[PTR]]
-; CHECK-NEXT:    br label [[MERGE]]
-; CHECK:       merge:
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test09(
+; NO_ASSUME-NEXT:  entry:
+; NO_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; NO_ASSUME-NEXT:    store i32 100, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
+; NO_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; NO_ASSUME-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
+; NO_ASSUME:       if.true:
+; NO_ASSUME-NEXT:    store i32 500, i32* [[PTR]]
+; NO_ASSUME-NEXT:    br label [[MERGE:%.*]]
+; NO_ASSUME:       if.false:
+; NO_ASSUME-NEXT:    store i32 600, i32* [[PTR]]
+; NO_ASSUME-NEXT:    br label [[MERGE]]
+; NO_ASSUME:       merge:
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test09(
+; USE_ASSUME-NEXT:  entry:
+; USE_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; USE_ASSUME-NEXT:    store i32 100, i32* [[PTR:%.*]]
+; USE_ASSUME-NEXT:    call void (i1, ...) @llvm.experimental.guard(i1 [[CMP]]) [ "deopt"() ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
+; USE_ASSUME:       if.true:
+; USE_ASSUME-NEXT:    store i32 500, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br label [[MERGE:%.*]]
+; USE_ASSUME:       if.false:
+; USE_ASSUME-NEXT:    store i32 600, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br label [[MERGE]]
+; USE_ASSUME:       merge:
+; USE_ASSUME-NEXT:    ret void
 ;
 
 entry:
@@ -361,11 +400,20 @@
 
 define void @test13(i32 %a, i32 %b, i32* %ptr) {
 ; Check that we deal correctly with stores when removing guards due to assume.
-; CHECK-LABEL: @test13(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP]])
-; CHECK-NEXT:    store i32 400, i32* [[PTR:%.*]]
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test13(
+; NO_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; NO_ASSUME-NEXT:    call void @llvm.assume(i1 [[CMP]])
+; NO_ASSUME-NEXT:    store i32 400, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test13(
+; USE_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 [[CMP]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR:%.*]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; USE_ASSUME-NEXT:    ret void
 ;
 
   %cmp = icmp eq i32 %a, %b
@@ -384,20 +432,38 @@
 ; Similar to test13, but with more control flow.
 ; TODO: Can we get rid of the store in the end of entry given that it is
 ; post-dominated by other stores?
-; CHECK-LABEL: @test14(
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP]])
-; CHECK-NEXT:    store i32 400, i32* [[PTR:%.*]]
-; CHECK-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
-; CHECK:       if.true:
-; CHECK-NEXT:    store i32 500, i32* [[PTR]]
-; CHECK-NEXT:    br label [[MERGE:%.*]]
-; CHECK:       if.false:
-; CHECK-NEXT:    store i32 600, i32* [[PTR]]
-; CHECK-NEXT:    br label [[MERGE]]
-; CHECK:       merge:
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test14(
+; NO_ASSUME-NEXT:  entry:
+; NO_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; NO_ASSUME-NEXT:    call void @llvm.assume(i1 [[CMP]])
+; NO_ASSUME-NEXT:    store i32 400, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
+; NO_ASSUME:       if.true:
+; NO_ASSUME-NEXT:    store i32 500, i32* [[PTR]]
+; NO_ASSUME-NEXT:    br label [[MERGE:%.*]]
+; NO_ASSUME:       if.false:
+; NO_ASSUME-NEXT:    store i32 600, i32* [[PTR]]
+; NO_ASSUME-NEXT:    br label [[MERGE]]
+; NO_ASSUME:       merge:
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test14(
+; USE_ASSUME-NEXT:  entry:
+; USE_ASSUME-NEXT:    [[CMP:%.*]] = icmp eq i32 [[A:%.*]], [[B:%.*]]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 [[CMP]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR:%.*]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    store i32 400, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br i1 [[C:%.*]], label [[IF_TRUE:%.*]], label [[IF_FALSE:%.*]]
+; USE_ASSUME:       if.true:
+; USE_ASSUME-NEXT:    store i32 500, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br label [[MERGE:%.*]]
+; USE_ASSUME:       if.false:
+; USE_ASSUME-NEXT:    store i32 600, i32* [[PTR]]
+; USE_ASSUME-NEXT:    br label [[MERGE]]
+; USE_ASSUME:       merge:
+; USE_ASSUME-NEXT:    ret void
 ;
 
 entry:
diff --git a/llvm/test/Transforms/EarlyCSE/invariant-loads.ll b/llvm/test/Transforms/EarlyCSE/invariant-loads.ll
--- a/llvm/test/Transforms/EarlyCSE/invariant-loads.ll
+++ b/llvm/test/Transforms/EarlyCSE/invariant-loads.ll
@@ -6,12 +6,21 @@
 declare void @clobber_and_use(i32)
 
 define void @f_0(i32* %ptr) {
-; CHECK-LABEL: @f_0(
-; CHECK-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @f_0(
+; NO_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @f_0(
+; USE_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    ret void
 ;
 
   %val0 = load i32, i32* %ptr, !invariant.load !{}
@@ -25,11 +34,18 @@
 
 define void @f_1(i32* %ptr) {
 ; We can forward invariant loads to non-invariant loads.
-; CHECK-LABEL: @f_1(
-; CHECK-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @f_1(
+; NO_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @f_1(
+; USE_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    ret void
 ;
 
   %val0 = load i32, i32* %ptr, !invariant.load !{}
@@ -41,11 +57,18 @@
 
 define void @f_2(i32* %ptr) {
 ; We can forward a non-invariant load into an invariant load.
-; CHECK-LABEL: @f_2(
-; CHECK-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]]
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @f_2(
+; NO_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]]
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @f_2(
+; USE_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    ret void
 ;
 
   %val0 = load i32, i32* %ptr
@@ -56,15 +79,26 @@
 }
 
 define void @f_3(i1 %cond, i32* %ptr) {
-; CHECK-LABEL: @f_3(
-; CHECK-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[LEFT:%.*]], label [[RIGHT:%.*]]
-; CHECK:       left:
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
-; CHECK-NEXT:    ret void
-; CHECK:       right:
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @f_3(
+; NO_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    br i1 [[COND:%.*]], label [[LEFT:%.*]], label [[RIGHT:%.*]]
+; NO_ASSUME:       left:
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; NO_ASSUME-NEXT:    ret void
+; NO_ASSUME:       right:
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @f_3(
+; USE_ASSUME-NEXT:    [[VAL0:%.*]] = load i32, i32* [[PTR:%.*]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    br i1 [[COND:%.*]], label [[LEFT:%.*]], label [[RIGHT:%.*]]
+; USE_ASSUME:       left:
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[PTR]], i64 4), "nonnull"(i32* [[PTR]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[VAL0]])
+; USE_ASSUME-NEXT:    ret void
+; USE_ASSUME:       right:
+; USE_ASSUME-NEXT:    ret void
 ;
   %val0 = load i32, i32* %ptr, !invariant.load !{}
   call void @clobber_and_use(i32 %val0)
@@ -114,10 +148,16 @@
 ; for the moment we chose to be conservative and just assume it's valid
 ; to restore the same unchanging value.
 define void @test_dse1(i32* %p) {
-; CHECK-LABEL: @test_dse1(
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[V1]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test_dse1(
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test_dse1(
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret void
 ;
   %v1 = load i32, i32* %p, !invariant.load !{}
   call void @clobber_and_use(i32 %v1)
@@ -142,12 +182,21 @@
 ; If we remove the load, we still start an invariant scope since
 ; it lets us remove later loads not explicitly marked invariant
 define void @test_scope_start_without_load(i32* %p) {
-; CHECK-LABEL: @test_scope_start_without_load(
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]]
-; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[ADD]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[V1]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test_scope_start_without_load(
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]]
+; NO_ASSUME-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[ADD]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test_scope_start_without_load(
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]]
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[ADD]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; USE_ASSUME-NEXT:    ret void
 ;
   %v1 = load i32, i32* %p
   %v2 = load i32, i32* %p, !invariant.load !{}
@@ -162,13 +211,23 @@
 ; with a potentially greater generation.  This hides the earlier invariant
 ; load
 define void @test_scope_restart(i32* %p) {
-; CHECK-LABEL: @test_scope_restart(
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[V1]])
-; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[ADD]])
-; CHECK-NEXT:    call void @clobber_and_use(i32 [[V1]])
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: @test_scope_restart(
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; NO_ASSUME-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[ADD]])
+; NO_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: @test_scope_restart(
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P:%.*]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    [[ADD:%.*]] = add i32 [[V1]], [[V1]]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[ADD]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    call void @clobber_and_use(i32 [[V1]])
+; USE_ASSUME-NEXT:    ret void
 ;
   %v1 = load i32, i32* %p, !invariant.load !{}
   call void @clobber_and_use(i32 %v1)
diff --git a/llvm/test/Transforms/EarlyCSE/invariant.start.ll b/llvm/test/Transforms/EarlyCSE/invariant.start.ll
--- a/llvm/test/Transforms/EarlyCSE/invariant.start.ll
+++ b/llvm/test/Transforms/EarlyCSE/invariant.start.ll
@@ -9,11 +9,18 @@
 ; Check that we do load-load forwarding over invariant.start, since it does not
 ; clobber memory
 define i8 @test_bypass1(i8 *%P) {
-; CHECK-LABEL: define {{[^@]+}}@test_bypass1
-; CHECK-SAME: (i8* [[P:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i8, i8* [[P]]
-; CHECK-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
-; CHECK-NEXT:    ret i8 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_bypass1
+; NO_ASSUME-SAME: (i8* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i8, i8* [[P]]
+; NO_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; NO_ASSUME-NEXT:    ret i8 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_bypass1
+; USE_ASSUME-SAME: (i8* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i8, i8* [[P]]
+; USE_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i8* [[P]], i64 1), "nonnull"(i8* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i8 0
 ;
 
   %V1 = load i8, i8* %P
@@ -26,11 +33,18 @@
 
 ; Trivial Store->load forwarding over invariant.start
 define i8 @test_bypass2(i8 *%P) {
-; CHECK-LABEL: define {{[^@]+}}@test_bypass2
-; CHECK-SAME: (i8* [[P:%.*]])
-; CHECK-NEXT:    store i8 42, i8* [[P]]
-; CHECK-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
-; CHECK-NEXT:    ret i8 42
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_bypass2
+; NO_ASSUME-SAME: (i8* [[P:%.*]])
+; NO_ASSUME-NEXT:    store i8 42, i8* [[P]]
+; NO_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; NO_ASSUME-NEXT:    ret i8 42
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_bypass2
+; USE_ASSUME-SAME: (i8* [[P:%.*]])
+; USE_ASSUME-NEXT:    store i8 42, i8* [[P]]
+; USE_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i8* [[P]], i64 1), "nonnull"(i8* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i8 42
 ;
 
   store i8 42, i8* %P
@@ -43,11 +57,18 @@
 ; %P is valid, and the second store is actually unreachable based on semantics
 ; of invariant.start.
 define void @test_bypass3(i8* %P) {
-; CHECK-LABEL: define {{[^@]+}}@test_bypass3
-; CHECK-SAME: (i8* [[P:%.*]])
-; CHECK-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
-; CHECK-NEXT:    store i8 60, i8* [[P]]
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_bypass3
+; NO_ASSUME-SAME: (i8* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; NO_ASSUME-NEXT:    store i8 60, i8* [[P]]
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_bypass3
+; USE_ASSUME-SAME: (i8* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 1, i8* [[P]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i8* [[P]], i64 1), "nonnull"(i8* [[P]]) ]
+; USE_ASSUME-NEXT:    store i8 60, i8* [[P]]
+; USE_ASSUME-NEXT:    ret void
 ;
 
   store i8 50, i8* %P
@@ -83,12 +104,20 @@
 declare void @llvm.invariant.end.p0i32({}*, i64, i32* nocapture) nounwind
 
 define i32 @test_before_load(i32* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test_before_load
-; CHECK-SAME: (i32* [[P:%.*]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_before_load
+; NO_ASSUME-SAME: (i32* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_before_load
+; USE_ASSUME-SAME: (i32* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
   %v1 = load i32, i32* %p
@@ -99,12 +128,20 @@
 }
 
 define i32 @test_before_clobber(i32* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test_before_clobber
-; CHECK-SAME: (i32* [[P:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_before_clobber
+; NO_ASSUME-SAME: (i32* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_before_clobber
+; USE_ASSUME-SAME: (i32* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   %v1 = load i32, i32* %p
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
@@ -115,13 +152,22 @@
 }
 
 define i32 @test_duplicate_scope(i32* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test_duplicate_scope
-; CHECK-SAME: (i32* [[P:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    [[TMP2:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_duplicate_scope
+; NO_ASSUME-SAME: (i32* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    [[TMP2:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_duplicate_scope
+; USE_ASSUME-SAME: (i32* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    [[TMP2:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   %v1 = load i32, i32* %p
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
@@ -133,13 +179,22 @@
 }
 
 define i32 @test_unanalzyable_load(i32* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test_unanalzyable_load
-; CHECK-SAME: (i32* [[P:%.*]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_unanalzyable_load
+; NO_ASSUME-SAME: (i32* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_unanalzyable_load
+; USE_ASSUME-SAME: (i32* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
   call void @clobber()
@@ -169,16 +224,28 @@
 }
 
 define i32 @test_merge(i32* %p, i1 %cnd) {
-; CHECK-LABEL: define {{[^@]+}}@test_merge
-; CHECK-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
-; CHECK:       taken:
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    br label [[MERGE]]
-; CHECK:       merge:
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_merge
+; NO_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
+; NO_ASSUME:       taken:
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    br label [[MERGE]]
+; NO_ASSUME:       merge:
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_merge
+; USE_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
+; USE_ASSUME:       taken:
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    br label [[MERGE]]
+; USE_ASSUME:       merge:
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   %v1 = load i32, i32* %p
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
@@ -250,17 +317,30 @@
 }
 
 define i32 @test_merge_unanalyzable_load(i32* %p, i1 %cnd) {
-; CHECK-LABEL: define {{[^@]+}}@test_merge_unanalyzable_load
-; CHECK-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
-; CHECK:       taken:
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    br label [[MERGE]]
-; CHECK:       merge:
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_merge_unanalyzable_load
+; NO_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
+; NO_ASSUME:       taken:
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    br label [[MERGE]]
+; NO_ASSUME:       merge:
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_merge_unanalyzable_load
+; USE_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    br i1 [[CND]], label [[MERGE:%.*]], label [[TAKEN:%.*]]
+; USE_ASSUME:       taken:
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    br label [[MERGE]]
+; USE_ASSUME:       merge:
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
   call void @clobber()
@@ -277,12 +357,20 @@
 }
 
 define void @test_dse_before_load(i32* %p, i1 %cnd) {
-; CHECK-LABEL: define {{[^@]+}}@test_dse_before_load
-; CHECK-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_dse_before_load
+; NO_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_dse_before_load
+; USE_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret void
 ;
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
   %v1 = load i32, i32* %p
@@ -292,12 +380,20 @@
 }
 
 define void @test_dse_after_load(i32* %p, i1 %cnd) {
-; CHECK-LABEL: define {{[^@]+}}@test_dse_after_load
-; CHECK-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
-; CHECK-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret void
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_dse_after_load
+; NO_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; NO_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret void
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_dse_after_load
+; USE_ASSUME-SAME: (i32* [[P:%.*]], i1 [[CND:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]]
+; USE_ASSUME-NEXT:    [[TMP1:%.*]] = call {}* @llvm.invariant.start.p0i32(i64 4, i32* [[P]])
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret void
 ;
   %v1 = load i32, i32* %p
   call {}* @llvm.invariant.start.p0i32(i64 4, i32* %p)
@@ -410,11 +506,18 @@
 
 ; Invariant load defact starts an invariant.start scope of the appropriate size
 define i32 @test_invariant_load_scope(i32* %p) {
-; CHECK-LABEL: define {{[^@]+}}@test_invariant_load_scope
-; CHECK-SAME: (i32* [[P:%.*]])
-; CHECK-NEXT:    [[V1:%.*]] = load i32, i32* [[P]], !invariant.load !0
-; CHECK-NEXT:    call void @clobber()
-; CHECK-NEXT:    ret i32 0
+; NO_ASSUME-LABEL: define {{[^@]+}}@test_invariant_load_scope
+; NO_ASSUME-SAME: (i32* [[P:%.*]])
+; NO_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]], !invariant.load !0
+; NO_ASSUME-NEXT:    call void @clobber()
+; NO_ASSUME-NEXT:    ret i32 0
+;
+; USE_ASSUME-LABEL: define {{[^@]+}}@test_invariant_load_scope
+; USE_ASSUME-SAME: (i32* [[P:%.*]])
+; USE_ASSUME-NEXT:    [[V1:%.*]] = load i32, i32* [[P]], !invariant.load !0
+; USE_ASSUME-NEXT:    call void @clobber()
+; USE_ASSUME-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(i32* [[P]], i64 4), "nonnull"(i32* [[P]]) ]
+; USE_ASSUME-NEXT:    ret i32 0
 ;
   %v1 = load i32, i32* %p, !invariant.load !{}
   call void @clobber()