From 5181255696e182a3b545328574f7e56dc0e20077 Mon Sep 17 00:00:00 2001
From: Filipe Cabecinhas <me@filcab.net>
Date: Fri, 6 Jun 2014 18:07:06 +0000
Subject: [PATCH] Fixed a bug in lowering shuffle_vectors to insertps

Summary:
We were being too strict and not accounting for undefs.
Added a test case and fixed another one where we improved codegen.

Reviewers: grosbach, nadav, delena

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D4039

llvm-svn: 210361
---
 llvm/lib/Target/X86/X86ISelLowering.cpp | 29 ++++++++++++++++++++---------
 llvm/test/CodeGen/X86/avx-shuffle.ll    |  6 ++++--
 llvm/test/CodeGen/X86/sse41.ll          | 11 +++++++++++
 3 files changed, 35 insertions(+), 11 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 610be3f..9e728c7 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -3964,14 +3964,22 @@ static bool isINSERTPSMask(ArrayRef<int> Mask, MVT VT) {
 
   unsigned CorrectPosV1 = 0;
   unsigned CorrectPosV2 = 0;
-  for (int i = 0, e = (int)VT.getVectorNumElements(); i != e; ++i)
+  for (int i = 0, e = (int)VT.getVectorNumElements(); i != e; ++i) {
+    if (Mask[i] == -1) {
+      ++CorrectPosV1;
+      ++CorrectPosV2;
+      continue;
+    }
+
     if (Mask[i] == i)
       ++CorrectPosV1;
     else if (Mask[i] == i + 4)
       ++CorrectPosV2;
+  }
 
   if (CorrectPosV1 == 3 || CorrectPosV2 == 3)
-    // We have 3 elements from one vector, and one from another.
+    // We have 3 elements (undefs count as elements from any vector) from one
+    // vector, and one from another.
     return true;
 
   return false;
@@ -7462,8 +7470,9 @@ static SDValue getINSERTPS(ShuffleVectorSDNode *SVOp, SDLoc &dl,
   assert((VT == MVT::v4f32 || VT == MVT::v4i32) &&
          "unsupported vector type for insertps/pinsrd");
 
-  int FromV1 = std::count_if(Mask.begin(), Mask.end(),
-                             [](const int &i) { return i < 4; });
+  auto FromV1Predicate = [](const int &i) { return i < 4 && i > -1; };
+  auto FromV2Predicate = [](const int &i) { return i >= 4; };
+  int FromV1 = std::count_if(Mask.begin(), Mask.end(), FromV1Predicate);
 
   SDValue From;
   SDValue To;
@@ -7471,15 +7480,17 @@ static SDValue getINSERTPS(ShuffleVectorSDNode *SVOp, SDLoc &dl,
   if (FromV1 == 1) {
     From = V1;
     To = V2;
-    DestIndex = std::find_if(Mask.begin(), Mask.end(),
-                             [](const int &i) { return i < 4; }) -
+    DestIndex = std::find_if(Mask.begin(), Mask.end(), FromV1Predicate) -
                 Mask.begin();
   } else {
+    assert(std::count_if(Mask.begin(), Mask.end(), FromV2Predicate) == 1 &&
+           "More than one element from V1 and from V2, or no elements from one "
+           "of the vectors. This case should not have returned true from "
+           "isINSERTPSMask");
     From = V2;
     To = V1;
-    DestIndex = std::find_if(Mask.begin(), Mask.end(),
-                             [](const int &i) { return i >= 4; }) -
-                Mask.begin();
+    DestIndex =
+        std::find_if(Mask.begin(), Mask.end(), FromV2Predicate) - Mask.begin();
   }
 
   if (MayFoldLoad(From)) {
diff --git a/llvm/test/CodeGen/X86/avx-shuffle.ll b/llvm/test/CodeGen/X86/avx-shuffle.ll
index f407ba4..f3f7e55 100644
--- a/llvm/test/CodeGen/X86/avx-shuffle.ll
+++ b/llvm/test/CodeGen/X86/avx-shuffle.ll
@@ -5,8 +5,10 @@ define <4 x float> @test1(<4 x float> %a) nounwind {
   %b = shufflevector <4 x float> zeroinitializer, <4 x float> %a, <4 x i32> <i32 2, i32 5, i32 undef, i32 undef>
   ret <4 x float> %b
 ; CHECK-LABEL: test1:
-; CHECK: vshufps
-; CHECK: vpshufd
+;; TODO: This test could be improved by removing the xor instruction and
+;; having vinsertps zero out the needed elements.
+; CHECK: vxorps
+; CHECK: vinsertps
 }
 
 ; rdar://10538417
diff --git a/llvm/test/CodeGen/X86/sse41.ll b/llvm/test/CodeGen/X86/sse41.ll
index a3c6201..a77ede2 100644
--- a/llvm/test/CodeGen/X86/sse41.ll
+++ b/llvm/test/CodeGen/X86/sse41.ll
@@ -692,3 +692,14 @@ define <4 x float> @insertps_from_broadcast_multiple_use(<4 x float> %a, <4 x fl
   %13 = fadd <4 x float> %11, %12
   ret <4 x float> %13
 }
+
+define <4 x float> @insertps_with_undefs(<4 x float> %a, float* %b) {
+; CHECK-LABEL: insertps_with_undefs:
+; CHECK-NOT: shufps
+; CHECK: insertps    $32, %xmm0
+; CHECK: ret
+  %1 = load float* %b, align 4
+  %2 = insertelement <4 x float> undef, float %1, i32 0
+  %result = shufflevector <4 x float> %a, <4 x float> %2, <4 x i32> <i32 4, i32 undef, i32 0, i32 7>
+  ret <4 x float> %result
+}
-- 
2.7.4