InstCombine: Improve the result bitvect type when folding (cmp pred (load (gep GV, i)) C) to a bit test.

The original code used i32, and i64 if legal. This introduced unneeded
casts when they aren't legal, or when the index variable i has another
type. In order of preference: try to use i's type; use the smallest
fitting legal type (using an added DataLayout method); default to i32.
A testcase checks that this works when the index gep operand is i16.

Patch by : Ahmed Bougacha <ahmed.bougacha@gmail.com>
Reviewed by : Duncan

llvm-svn: 177712

4 files changed, 43 insertions, 11 deletions

diff --git a/llvm/include/llvm/IR/DataLayout.h b/llvm/include/llvm/IR/DataLayout.h
index 72b4eedabdc..547d857b7b7 100644
--- a/llvm/include/llvm/IR/DataLayout.h
+++ b/llvm/include/llvm/IR/DataLayout.h
@@ -352,6 +352,10 @@ public:
   /// type.
   Type *getIntPtrType(Type *) const;
 
+  /// getSmallestLegalIntType - Return the smallest integer type with size at
+  /// least as big as Width bits.
+  Type *getSmallestLegalIntType(LLVMContext &C, unsigned Width = 0) const;
+
   /// getIndexedOffset - return the offset from the beginning of the type for
   /// the specified indices.  This is used to implement getelementptr.
   uint64_t getIndexedOffset(Type *Ty, ArrayRef<Value *> Indices) const;
diff --git a/llvm/lib/IR/DataLayout.cpp b/llvm/lib/IR/DataLayout.cpp
index 95966749127..ecd5216f20a 100644
--- a/llvm/lib/IR/DataLayout.cpp
+++ b/llvm/lib/IR/DataLayout.cpp
@@ -621,6 +621,13 @@ Type *DataLayout::getIntPtrType(Type *Ty) const {
   return IntTy;
 }
 
+Type *DataLayout::getSmallestLegalIntType(LLVMContext &C, unsigned Width) const {
+  for (unsigned i = 0, e = (unsigned)LegalIntWidths.size(); i != e; ++i)
+    if (Width <= LegalIntWidths[i])
+      return Type::getIntNTy(C, LegalIntWidths[i]);
+  return 0;
+}
+
 uint64_t DataLayout::getIndexedOffset(Type *ptrTy,
                                       ArrayRef<Value *> Indices) const {
   Type *Ty = ptrTy;
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 32fdb9b708b..a4e117e4c4b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -443,20 +443,29 @@ FoldCmpLoadFromIndexedGlobal(GetElementPtrInst *GEP, GlobalVariable *GV,
   }
 
 
-  // If a 32-bit or 64-bit magic bitvector captures the entire comparison state
+  // If a magic bitvector captures the entire comparison state
   // of this load, replace it with computation that does:
   //   ((magic_cst >> i) & 1) != 0
-  if (ArrayElementCount <= 32 ||
-      (TD && ArrayElementCount <= 64 && TD->isLegalInteger(64))) {
-    Type *Ty;
-    if (ArrayElementCount <= 32)
+  {
+    Type *Ty = 0;
+
+    // Look for an appropriate type:
+    // - The type of Idx if the magic fits
+    // - The smallest fitting legal type if we have a DataLayout
+    // - Default to i32
+    if (ArrayElementCount <= Idx->getType()->getIntegerBitWidth())
+      Ty = Idx->getType();
+    else if (TD)
+      Ty = TD->getSmallestLegalIntType(Init->getContext(), ArrayElementCount);
+    else if (ArrayElementCount <= 32)
       Ty = Type::getInt32Ty(Init->getContext());
-    else
-      Ty = Type::getInt64Ty(Init->getContext());
-    Value *V = Builder->CreateIntCast(Idx, Ty, false);
-    V = Builder->CreateLShr(ConstantInt::get(Ty, MagicBitvector), V);
-    V = Builder->CreateAnd(ConstantInt::get(Ty, 1), V);
-    return new ICmpInst(ICmpInst::ICMP_NE, V, ConstantInt::get(Ty, 0));
+
+    if (Ty != 0) {
+      Value *V = Builder->CreateIntCast(Idx, Ty, false);
+      V = Builder->CreateLShr(ConstantInt::get(Ty, MagicBitvector), V);
+      V = Builder->CreateAnd(ConstantInt::get(Ty, 1), V);
+      return new ICmpInst(ICmpInst::ICMP_NE, V, ConstantInt::get(Ty, 0));
+    }
   }
 
   return 0;
diff --git a/llvm/test/Transforms/InstCombine/load-cmp.ll b/llvm/test/Transforms/InstCombine/load-cmp.ll
index 5cafb7787e3..d88188e4109 100644
--- a/llvm/test/Transforms/InstCombine/load-cmp.ll
+++ b/llvm/test/Transforms/InstCombine/load-cmp.ll
@@ -47,6 +47,18 @@ define i1 @test4(i32 %X) {
 ; CHECK-NEXT: ret i1 %R
 }
 
+define i1 @test4_i16(i16 %X) {
+  %P = getelementptr inbounds [10 x i16]* @G16, i32 0, i16 %X
+  %Q = load i16* %P
+  %R = icmp sle i16 %Q, 73
+  ret i1 %R
+; CHECK: @test4_i16
+; CHECK-NEXT: lshr i16 933, %X
+; CHECK-NEXT: and i16 {{.*}}, 1
+; CHECK-NEXT: %R = icmp ne i16 {{.*}}, 0
+; CHECK-NEXT: ret i1 %R
+}
+
 define i1 @test5(i32 %X) {
   %P = getelementptr inbounds [10 x i16]* @G16, i32 0, i32 %X
   %Q = load i16* %P