InstCombine now optimizes vector udiv by power of 2 to shifts

Fixes r8429

llvm-svn: 144036

2 files changed, 24 insertions, 5 deletions

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index 7f48125a97a..2f82b7b4a91 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -441,19 +441,23 @@ Instruction *InstCombiner::visitUDiv(BinaryOperator &I) {
   // Handle the integer div common cases
   if (Instruction *Common = commonIDivTransforms(I))
     return Common;
-
-  if (ConstantInt *C = dyn_cast<ConstantInt>(Op1)) {
+  
+  { 
     // X udiv 2^C -> X >> C
     // Check to see if this is an unsigned division with an exact power of 2,
     // if so, convert to a right shift.
-    if (C->getValue().isPowerOf2()) { // 0 not included in isPowerOf2
+    const APInt *C;
+    if (match(Op1, m_Power2(C))) {
       BinaryOperator *LShr =
-        BinaryOperator::CreateLShr(Op0, 
-            ConstantInt::get(Op0->getType(), C->getValue().logBase2()));
+      BinaryOperator::CreateLShr(Op0, 
+                                 ConstantInt::get(Op0->getType(), 
+                                                  C->logBase2()));
       if (I.isExact()) LShr->setIsExact();
       return LShr;
     }
+  }
 
+  if (ConstantInt *C = dyn_cast<ConstantInt>(Op1)) {
     // X udiv C, where C >= signbit
     if (C->getValue().isNegative()) {
       Value *IC = Builder->CreateICmpULT(Op0, C);
diff --git a/llvm/test/CodeGen/X86/vec_udiv_to_shift.ll b/llvm/test/CodeGen/X86/vec_udiv_to_shift.ll
new file mode 100644
index 00000000000..e325f61c9f3
--- /dev/null
+++ b/llvm/test/CodeGen/X86/vec_udiv_to_shift.ll
@@ -0,0 +1,15 @@
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+define <8 x i16> @udiv_vec8x16(<8 x i16> %var) {
+entry:
+; CHECK: lshr <8 x i16> %var, <i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5>
+%0 = udiv <8 x i16> %var, <i16 32, i16 32, i16 32, i16 32, i16 32, i16 32, i16 32, i16 32>
+ret <8 x i16> %0
+}
+
+define <4 x i32> @udiv_vec4x32(<4 x i32> %var) {
+entry:
+; CHECK: lshr <4 x i32> %var, <i32 4, i32 4, i32 4, i32 4>
+%0 = udiv <4 x i32> %var, <i32 16, i32 16, i32 16, i32 16>
+ret <4 x i32> %0
+}
\ No newline at end of file