gistfile1.diff

diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 0bc3ac7..e0bd6b5 100644
--- a/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -554,6 +554,51 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     break;
   }
 
+  // Constant fold <A x Bi> << Ci.
+  case Intrinsic::x86_sse2_psll_d:
+  case Intrinsic::x86_sse2_psll_dq:
+  case Intrinsic::x86_sse2_psll_dq_bs:
+  case Intrinsic::x86_sse2_psll_q:
+  case Intrinsic::x86_sse2_psll_w:
+  case Intrinsic::x86_sse2_pslli_d:
+  case Intrinsic::x86_sse2_pslli_q:
+  case Intrinsic::x86_sse2_pslli_w:
+  case Intrinsic::x86_avx2_psll_d:
+  case Intrinsic::x86_avx2_psll_dq:
+  case Intrinsic::x86_avx2_psll_dq_bs:
+  case Intrinsic::x86_avx2_psll_q:
+  case Intrinsic::x86_avx2_psll_w:
+  case Intrinsic::x86_avx2_pslli_d:
+  case Intrinsic::x86_avx2_pslli_q:
+  case Intrinsic::x86_avx2_pslli_w:
+  case Intrinsic::x86_avx512_psll_dq:
+  case Intrinsic::x86_avx512_psll_dq_bs: {
+    // Simplify if count is constant. To 0 if > BitWidth, otherwise to shl.
+    auto CDV = dyn_cast<ConstantDataVector>(II->getArgOperand(1));
+    auto CInt = dyn_cast<ConstantInt>(II->getArgOperand(1));
+    if (!CDV && !CInt)
+      break;
+    ConstantInt *Count;
+    if (CDV)
+      Count = cast<ConstantInt>(CDV->getElementAsConstant(0));
+    else
+      Count = CInt;
+
+    auto Vec = II->getArgOperand(0);
+    auto VT = cast<VectorType>(Vec->getType());
+    if (Count->getZExtValue() >
+        VT->getElementType()->getPrimitiveSizeInBits() - 1)
+      return ReplaceInstUsesWith(
+          CI, ConstantAggregateZero::get(Vec->getType()));
+    else {
+      unsigned VWidth = VT->getNumElements();
+      // Get a constant vector of the same type as the first operand.
+      auto VTCI = ConstantInt::get(VT->getElementType(), Count->getZExtValue());
+      return BinaryOperator::CreateShl(
+          Vec, Builder->CreateVectorSplat(VWidth, VTCI));
+    }
+    break;
+  }
 
   case Intrinsic::x86_sse41_pmovsxbw:
   case Intrinsic::x86_sse41_pmovsxwd:
diff --git a/test/Transforms/InstCombine/vec_demanded_elts.ll b/test/Transforms/InstCombine/vec_demanded_elts.ll
index d12412a..cb62fc6 100644
--- a/test/Transforms/InstCombine/vec_demanded_elts.ll
+++ b/test/Transforms/InstCombine/vec_demanded_elts.ll
@@ -209,4 +209,22 @@ define <4 x float> @test_select(float %f, float %g) {
   ret <4 x float> %ret
 }
 
+define <4 x i32> @test_pslli_s() nounwind {
+entry:
+; Constant fold.
+; CHECK: test_pslli_s
+; CHECK: <i32 0, i32 2, i32 4, i32 6>
+  %0 = tail call <4 x i32> @llvm.x86.sse2.pslli.d(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, i32 1)
+  ret <4 x i32> %0
+}
+
+define <4 x i32> @test_pslli_l() nounwind {
+entry:
+; Constant 0
+; CHECK: test_pslli_l
+; CHECK: zeroinitializer
+  %0 = tail call <4 x i32> @llvm.x86.sse2.pslli.d(<4 x i32> <i32 0, i32 1, i32 2, i32 3>, i32 32)
+  ret <4 x i32> %0
+}
 
+declare <4 x i32> @llvm.x86.sse2.pslli.d(<4 x i32>, i32)