Use movaps / movd to extract vector element 0 even with sse4.1. It's still cheaper than pextrw especially if the value is in memory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61555 91177308-0d34-0410-b5e6-96231b3b80d8

8 files changed, 23 insertions, 15 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 43a7bb5bb5..5ed6342bb1 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -4215,6 +4215,14 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT_SSE4(SDValue Op,
                                     DAG.getValueType(VT));
     return DAG.getNode(ISD::TRUNCATE, VT, Assert);
   } else if (VT.getSizeInBits() == 16) {
+    unsigned Idx = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+    // If Idx is 0, it's cheaper to do a move instead of a pextrw.
+    if (Idx == 0)
+      return DAG.getNode(ISD::TRUNCATE, MVT::i16,
+                         DAG.getNode(ISD::EXTRACT_VECTOR_ELT, MVT::i32,
+                                     DAG.getNode(ISD::BIT_CONVERT, MVT::v4i32,
+                                                 Op.getOperand(0)),
+                                     Op.getOperand(1)));
     SDValue Extract = DAG.getNode(X86ISD::PEXTRW, MVT::i32,
                                     Op.getOperand(0), Op.getOperand(1));
     SDValue Assert  = DAG.getNode(ISD::AssertZext, MVT::i32, Extract,
diff --git a/test/CodeGen/X86/vec_extract.ll b/test/CodeGen/X86/vec_extract.ll
index b45f9398e4..f1f009ec81 100644
--- a/test/CodeGen/X86/vec_extract.ll
+++ b/test/CodeGen/X86/vec_extract.ll
@@ -1,4 +1,4 @@
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 -o %t -f
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 -o %t -f
 ; RUN: grep movss    %t | count 3
 ; RUN: grep movhlps  %t | count 1
 ; RUN: grep pshufd   %t | count 1
diff --git a/test/CodeGen/X86/vec_insert-2.ll b/test/CodeGen/X86/vec_insert-2.ll
index a6d4f014d2..8d0bcc4fbf 100644
--- a/test/CodeGen/X86/vec_insert-2.ll
+++ b/test/CodeGen/X86/vec_insert-2.ll
@@ -1,8 +1,8 @@
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep {\$36,} | count 2
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep shufps | count 2
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep pinsrw | count 1
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep movhpd | count 1
-; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+sse2 | grep unpcklpd | count 1
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | grep {\$36,} | count 2
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | grep shufps | count 2
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | grep pinsrw | count 1
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | grep movhpd | count 1
+; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+sse2,-sse41 | grep unpcklpd | count 1
 
 define <4 x float> @t1(float %s, <4 x float> %tmp) nounwind {
         %tmp1 = insertelement <4 x float> %tmp, float %s, i32 3
diff --git a/test/CodeGen/X86/vec_insert-3.ll b/test/CodeGen/X86/vec_insert-3.ll
index e42a368489..e43eca4b87 100644
--- a/test/CodeGen/X86/vec_insert-3.ll
+++ b/test/CodeGen/X86/vec_insert-3.ll
@@ -1,4 +1,4 @@
-; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+sse2 | grep punpcklqdq | count 1
+; RUN: llvm-as < %s | llc -march=x86-64 -mattr=+sse2,-sse41 | grep punpcklqdq | count 1
 
 define <2 x i64> @t1(i64 %s, <2 x i64> %tmp) nounwind {
         %tmp1 = insertelement <2 x i64> %tmp, i64 %s, i32 1
diff --git a/test/CodeGen/X86/vec_insert.ll b/test/CodeGen/X86/vec_insert.ll
index c8c9f141ec..e032c5b854 100644
--- a/test/CodeGen/X86/vec_insert.ll
+++ b/test/CodeGen/X86/vec_insert.ll
@@ -1,5 +1,5 @@
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep movss | count 1
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | not grep pinsrw
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | grep movss | count 1
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2,-sse41 | not grep pinsrw
 
 define void @test(<4 x float>* %F, i32 %I) {
 	%tmp = load <4 x float>* %F		; <<4 x float>> [#uses=1]
diff --git a/test/CodeGen/X86/vec_shuffle-12.ll b/test/CodeGen/X86/vec_shuffle-12.ll
index 3f49b02543..aad27ea2f8 100644
--- a/test/CodeGen/X86/vec_shuffle-12.ll
+++ b/test/CodeGen/X86/vec_shuffle-12.ll
@@ -5,24 +5,24 @@
 ; RUN: grep pshuflw %t | count 3
 ; RUN: grep pshufhw %t | count 2
 
-define <8 x i16> @t1(<8 x i16>* %A, <8 x i16>* %B) {
+define <8 x i16> @t1(<8 x i16>* %A, <8 x i16>* %B) nounwind {
 	%tmp1 = load <8 x i16>* %A
 	%tmp2 = load <8 x i16>* %B
 	%tmp3 = shufflevector <8 x i16> %tmp1, <8 x i16> %tmp2, <8 x i32> < i32 8, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 >
 	ret <8 x i16> %tmp3
 }
 
-define <8 x i16> @t2(<8 x i16> %A, <8 x i16> %B) {
+define <8 x i16> @t2(<8 x i16> %A, <8 x i16> %B) nounwind {
 	%tmp = shufflevector <8 x i16> %A, <8 x i16> %B, <8 x i32> < i32 9, i32 1, i32 2, i32 9, i32 4, i32 5, i32 6, i32 7 >
 	ret <8 x i16> %tmp
 }
 
-define <8 x i16> @t3(<8 x i16> %A, <8 x i16> %B) {
+define <8 x i16> @t3(<8 x i16> %A, <8 x i16> %B) nounwind {
 	%tmp = shufflevector <8 x i16> %A, <8 x i16> %A, <8 x i32> < i32 8, i32 3, i32 2, i32 13, i32 7, i32 6, i32 5, i32 4 >
 	ret <8 x i16> %tmp
 }
 
-define <8 x i16> @t4(<8 x i16> %A, <8 x i16> %B) {
+define <8 x i16> @t4(<8 x i16> %A, <8 x i16> %B) nounwind {
 	%tmp = shufflevector <8 x i16> %A, <8 x i16> %B, <8 x i32> < i32 0, i32 7, i32 2, i32 3, i32 1, i32 5, i32 6, i32 5 >
 	ret <8 x i16> %tmp
 }
diff --git a/test/CodeGen/X86/vec_shuffle-28.ll b/test/CodeGen/X86/vec_shuffle-28.ll
index e73b824d02..0c81e77b95 100644
--- a/test/CodeGen/X86/vec_shuffle-28.ll
+++ b/test/CodeGen/X86/vec_shuffle-28.ll
@@ -1,6 +1,6 @@
 ; RUN: llvm-as < %s | llc -march=x86 -mattr=sse41 -o %t -f
 ; RUN: grep punpcklwd %t | count 1
-; RUN: grep pextrw %t | count 8
+; RUN: grep pextrw %t | count 6
 ; RUN: grep pinsrw %t | count 8
 
 
diff --git a/test/CodeGen/X86/widen_arith-1.ll b/test/CodeGen/X86/widen_arith-1.ll
index 13683b159f..419078174d 100644
--- a/test/CodeGen/X86/widen_arith-1.ll
+++ b/test/CodeGen/X86/widen_arith-1.ll
@@ -1,7 +1,7 @@
 ; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse42 -disable-mmx -o %t -f
 ; RUN: grep paddb  %t | count 1
 ; RUN: grep pextrb %t | count 1
-; RUN: grep pextrw %t | count 1
+; RUN: not grep pextrw %t
 
 ; Widen a v3i8 to v16i8 to use a vector add