[GVN] Small tweaks to comments, style, and missed vector handling

Noticed these while doing a final sweep of the code to make sure I hadn't missed anything in my last couple of patches.  The (minor) missed optimization was noticed because of the stylistic fix to avoid an overly specific cast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@354412 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Utils/VNCoercion.cpp b/lib/Transforms/Utils/VNCoercion.cpp
index 181e53e7605c1dd6b5f9c65f8de06acf3b82c6e0..19593054c23a2881be5778748c67caa6f48eeee7 100644 (file)
--- a/lib/Transforms/Utils/VNCoercion.cpp
+++ b/lib/Transforms/Utils/VNCoercion.cpp
@@ -36,8 +36,8 @@ bool canCoerceMustAliasedValueToLoad(Value *StoredVal, Type *LoadTy,
     // As a special case, allow coercion of memset used to initialize
     // an array w/null.  Despite non-integral pointers not generally having a
     // specific bit pattern, we do assume null is zero.
-    if (auto *CI = dyn_cast<ConstantInt>(StoredVal))
-      return CI->isZero();
+    if (auto *CI = dyn_cast<Constant>(StoredVal))
+      return CI->isNullValue();
     return false;
   }
   
@@ -287,9 +287,8 @@ int analyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
   // If this is memset, we just need to see if the offset is valid in the size
   // of the memset..
   if (MI->getIntrinsicID() == Intrinsic::memset) {
-    Value *StoredVal = cast<MemSetInst>(MI)->getValue();
     if (DL.isNonIntegralPointerType(LoadTy->getScalarType())) {
-      auto *CI = dyn_cast<ConstantInt>(StoredVal);
+      auto *CI = dyn_cast<ConstantInt>(cast<MemSetInst>(MI)->getValue());
       if (!CI || !CI->isZero())
         return -1;
     }
@@ -316,7 +315,8 @@ int analyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
   if (Offset == -1)
     return Offset;
 
-  // Don't coerce non-integral pointers to integers or vice versa.
+  // Don't coerce non-integral pointers to integers or vice versa, and the
+  // memtransfer is implicitly a raw byte code
   if (DL.isNonIntegralPointerType(LoadTy->getScalarType()))
     // TODO: Can allow nullptrs from constant zeros
     return -1;

diff --git a/test/Transforms/GVN/non-integral-pointers.ll b/test/Transforms/GVN/non-integral-pointers.ll
index 7c26063918544721a9cc31f5cce1e27e111d8a86..254defe0690eaa5b4a10e557c5d21698854dc3f7 100644 (file)
--- a/test/Transforms/GVN/non-integral-pointers.ll
+++ b/test/Transforms/GVN/non-integral-pointers.ll
@@ -139,18 +139,32 @@ define <1 x i8 addrspace(4)*> @neg_forward_store_vload(<1 x i8 addrspace(4)*> ad
   ret <1 x i8 addrspace(4)*> %ref
 }
 
-; TODO: missed optimization, we can forward the null.
+; Nulls have known bit patterns, so we can forward
 define i8 addrspace(4)* @forward_store_zero(i8 addrspace(4)* addrspace(4)* %loc) {
 ; CHECK-LABEL: @forward_store_zero(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[LOC_BC:%.*]] = bitcast i8 addrspace(4)* addrspace(4)* [[LOC:%.*]] to i64 addrspace(4)*
-; CHECK-NEXT:    store i64 5, i64 addrspace(4)* [[LOC_BC]]
-; CHECK-NEXT:    [[REF:%.*]] = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* [[LOC]]
-; CHECK-NEXT:    ret i8 addrspace(4)* [[REF]]
+; CHECK-NEXT:    store i64 0, i64 addrspace(4)* [[LOC_BC]]
+; CHECK-NEXT:    ret i8 addrspace(4)* null
 ;
   entry:
   %loc.bc = bitcast i8 addrspace(4)* addrspace(4)* %loc to i64 addrspace(4)*
-  store i64 5, i64 addrspace(4)* %loc.bc
+  store i64 0, i64 addrspace(4)* %loc.bc
+  %ref = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* %loc
+  ret i8 addrspace(4)* %ref
+}
+
+; Nulls have known bit patterns, so we can forward
+define i8 addrspace(4)* @forward_store_zero2(i8 addrspace(4)* addrspace(4)* %loc) {
+; CHECK-LABEL: @forward_store_zero2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[LOC_BC:%.*]] = bitcast i8 addrspace(4)* addrspace(4)* [[LOC:%.*]] to <2 x i32> addrspace(4)*
+; CHECK-NEXT:    store <2 x i32> zeroinitializer, <2 x i32> addrspace(4)* [[LOC_BC]]
+; CHECK-NEXT:    ret i8 addrspace(4)* null
+;
+  entry:
+  %loc.bc = bitcast i8 addrspace(4)* addrspace(4)* %loc to <2 x i32> addrspace(4)*
+  store <2 x i32> zeroinitializer, <2 x i32> addrspace(4)* %loc.bc
   %ref = load i8 addrspace(4)*, i8 addrspace(4)* addrspace(4)* %loc
   ret i8 addrspace(4)* %ref
 }