[PATCH] D27491: [LVI] Meet rules for non-null pointers

[Philip Reames via Phabricator via llvm-commits]

reames created this revision.
reames added reviewers: nicholas, apilipenko, igmyrj, sanjoy, jmolloy.
reames added a subscriber: llvm-commits.
Herald added a subscriber: mcrosier.

The patch itself is fairly straight-forward, it simply extends LVI conservative meet operator to remember non-null facts when merging non-null constants.

However, the fact I had to write this is really making me question the value of pointer types in LVI as a whole.  The lack of any decent meet rule basically means that pointer handling in LVI reduces to equality propagation and a dereferenceability based non-null analysis.  Given GVN implements a version of the former (minus the meet of two constants case), I'm wondering whether pointers should just be removed from LVI and handled separately.  I want to stew on that a bit further before actually landing this patch.


https://reviews.llvm.org/D27491

Files:
  lib/Analysis/LazyValueInfo.cpp
  test/Transforms/CorrelatedValuePropagation/select.ll


Index: test/Transforms/CorrelatedValuePropagation/select.ll
===================================================================
--- test/Transforms/CorrelatedValuePropagation/select.ll
+++ test/Transforms/CorrelatedValuePropagation/select.ll
@@ -216,3 +216,53 @@
 ; CHECK: ret i1 true
   ret i1 true
 }
+
+define i1 @test7(i1 %unknown, i32* nonnull %p1, i32* nonnull %p2 ) {
+; CHECK-LABEL: @test7
+  %sel = select i1 %unknown, i32* %p1, i32* %p2
+  ;; TODO: This pointless branch shouldn't be neccessary
+  br label %next2
+next2:
+; CHECK-LABEL: next2:
+; CHECK: ret i1 true
+  %res = icmp ne i32* %sel, null
+  ret i1 %res
+}
+ at g1 = external global i32
+ at g2 = external global i32
+
+define i1 @test8a(i1 %unknown, i32* nonnull %p1) {
+; CHECK-LABEL: @test8
+  %sel = select i1 %unknown, i32* %p1, i32* @g1
+  ;; TODO: This pointless branch shouldn't be neccessary
+  br label %next2
+next2:
+; CHECK-LABEL: next2:
+; CHECK: ret i1 true
+  %res = icmp ne i32* %sel, null
+  ret i1 %res
+}
+
+define i1 @test8b(i1 %unknown, i32* nonnull %p1) {
+; CHECK-LABEL: @test8
+  %sel = select i1 %unknown, i32* %p1, i32* @g1
+  ;; TODO: This pointless branch shouldn't be neccessary
+  br label %next2
+next2:
+; CHECK-LABEL: next2:
+; CHECK: ret i1 true
+  %res = icmp ne i32* %sel, null
+  ret i1 %res
+}
+
+define i1 @test9(i1 %unknown) {
+; CHECK-LABEL: @test9
+  %sel = select i1 %unknown, i32* @g1, i32* @g2
+  ;; TODO: This pointless branch shouldn't be neccessary
+  br label %next2
+next2:
+; CHECK-LABEL: next2:
+; CHECK: ret i1 true
+  %res = icmp ne i32* %sel, null
+  ret i1 %res
+}
Index: lib/Analysis/LazyValueInfo.cpp
===================================================================
--- lib/Analysis/LazyValueInfo.cpp
+++ lib/Analysis/LazyValueInfo.cpp
@@ -222,12 +222,22 @@
       if (RHS.isConstant()) {
         if (Val == RHS.Val)
           return false;
+
+        PointerType *PT = dyn_cast<PointerType>(Val->getType());
+        if (PT && isKnownNonNull(Val) && isKnownNonNull(RHS.Val))
+          return markNotConstant(ConstantPointerNull::get(PT));
+        
         return markOverdefined();
       }
 
       if (RHS.isNotConstant()) {
         if (Val == RHS.Val)
           return markOverdefined();
+
+        PointerType *PT = dyn_cast<PointerType>(Val->getType());
+        if (PT && isa<ConstantPointerNull>(RHS.Val) && isKnownNonNull(Val))
+          return markNotConstant(ConstantPointerNull::get(PT));
+        
         return markOverdefined();
       }
 
@@ -238,6 +248,11 @@
       if (RHS.isConstant()) {
         if (Val == RHS.Val)
           return markOverdefined();
+
+        PointerType *PT = dyn_cast<PointerType>(Val->getType());
+        if (PT && isa<ConstantPointerNull>(Val) && isKnownNonNull(RHS.Val))
+          return false;
+        
         return markOverdefined();
       }
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D27491.80501.patch
Type: text/x-patch
Size: 2837 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161207/2d740344/attachment.bin>