From 30d519d92487a35136bac458d48694879b4a2060 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Tue, 29 May 2018 20:04:10 +0000
Subject: [PATCH] [X86] Fix a potential crash that occur after r333419.

The code could issue a truncate from a small type to larger type. We need to extend in that case instead.

llvm-svn: 333460
---
 lib/Target/X86/X86ISelLowering.cpp |  3 +--
 test/CodeGen/X86/combine-select.ll | 20 ++++++++++++++++++++
 2 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index ad4c9528776..fb1d30b28e1 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -32142,8 +32142,7 @@ static SDValue combineSelect(SDNode *N, SelectionDAG &DAG,
         cast<ConstantSDNode>(AndNode.getOperand(1))->getAPIntValue() == 1) {
       // LHS and RHS swapped due to
       // setcc outputting 1 when AND resulted in 0 and vice versa.
-      if (AndNode.getValueType() != MVT::i8)
-        AndNode = DAG.getNode(ISD::TRUNCATE, DL, MVT::i8, AndNode);
+      AndNode = DAG.getZExtOrTrunc(AndNode, DL, MVT::i8);
       return DAG.getNode(ISD::SELECT, DL, VT, AndNode, RHS, LHS);
     }
   }
diff --git a/test/CodeGen/X86/combine-select.ll b/test/CodeGen/X86/combine-select.ll
index 58be308f505..7cbb9831ba9 100644
--- a/test/CodeGen/X86/combine-select.ll
+++ b/test/CodeGen/X86/combine-select.ll
@@ -105,3 +105,23 @@ entry:
   %6 = insertelement <4 x float> %a, float %5, i32 0
   ret <4 x float> %6
 }
+
+; Make sure we don't crash trying to truncate the and instruction i4->i8. We need to extend instead.
+define <4 x float> @select_mask_add_ss_small_mask_type(<4 x float> %w, i4 %u, <4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: select_mask_add_ss_small_mask_type:
+; CHECK:       ## %bb.0: ## %entry
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vaddss %xmm2, %xmm1, %xmm0 {%k1}
+; CHECK-NEXT:    retq
+entry:
+  %0 = extractelement <4 x float> %b, i32 0
+  %1 = extractelement <4 x float> %a, i32 0
+  %2 = fadd float %1, %0
+  %3 = and i4 %u, 1
+  %4 = icmp eq i4 %3, 0
+  %5 = extractelement <4 x float> %w, i32 0
+  %6 = select i1 %4, float %5, float %2
+  %7 = insertelement <4 x float> %a, float %6, i32 0
+  ret <4 x float> %7
+}
+