[llvm-commits] [llvm] r61555 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/vec_extract.ll test/CodeGen/X86/vec_insert-2.ll test/CodeGen/X86/vec_insert-3.ll test/CodeGen/X86/vec_insert.ll test/CodeGen/X86/vec_shuffle-12.ll test/CodeGen/X86/vec_shuffle-28.ll test/CodeGen/X86/widen_arith-1.ll
Evan Cheng
evan.cheng at apple.com
Thu Jan 1 21:29:09 PST 2009
Author: evancheng
Date: Thu Jan 1 23:29:08 2009
New Revision: 61555
URL: http://llvm.org/viewvc/llvm-project?rev=61555&view=rev
Log:
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.
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/vec_extract.ll
llvm/trunk/test/CodeGen/X86/vec_insert-2.ll
llvm/trunk/test/CodeGen/X86/vec_insert-3.ll
llvm/trunk/test/CodeGen/X86/vec_insert.ll
llvm/trunk/test/CodeGen/X86/vec_shuffle-12.ll
llvm/trunk/test/CodeGen/X86/vec_shuffle-28.ll
llvm/trunk/test/CodeGen/X86/widen_arith-1.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu Jan 1 23:29:08 2009
@@ -4215,6 +4215,14 @@
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,
Modified: llvm/trunk/test/CodeGen/X86/vec_extract.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_extract.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_extract.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_extract.ll Thu Jan 1 23:29:08 2009
@@ -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
Modified: llvm/trunk/test/CodeGen/X86/vec_insert-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_insert-2.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_insert-2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_insert-2.ll Thu Jan 1 23:29:08 2009
@@ -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
Modified: llvm/trunk/test/CodeGen/X86/vec_insert-3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_insert-3.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_insert-3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_insert-3.ll Thu Jan 1 23:29:08 2009
@@ -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
Modified: llvm/trunk/test/CodeGen/X86/vec_insert.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_insert.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_insert.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_insert.ll Thu Jan 1 23:29:08 2009
@@ -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]
Modified: llvm/trunk/test/CodeGen/X86/vec_shuffle-12.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_shuffle-12.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_shuffle-12.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_shuffle-12.ll Thu Jan 1 23:29:08 2009
@@ -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
}
Modified: llvm/trunk/test/CodeGen/X86/vec_shuffle-28.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_shuffle-28.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_shuffle-28.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_shuffle-28.ll Thu Jan 1 23:29:08 2009
@@ -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
Modified: llvm/trunk/test/CodeGen/X86/widen_arith-1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/widen_arith-1.ll?rev=61555&r1=61554&r2=61555&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/widen_arith-1.ll (original)
+++ llvm/trunk/test/CodeGen/X86/widen_arith-1.ll Thu Jan 1 23:29:08 2009
@@ -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
More information about the llvm-commits
mailing list