[llvm-commits] [llvm] r157127 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/avx2-vbroadcast.ll
Nadav Rotem
nadav.rotem at intel.com
Sat May 19 12:57:37 PDT 2012
Author: nadav
Date: Sat May 19 14:57:37 2012
New Revision: 157127
URL: http://llvm.org/viewvc/llvm-project?rev=157127&view=rev
Log:
Add support for additional in-reg vbroadcast patterns
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/avx2-vbroadcast.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=157127&r1=157126&r2=157127&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat May 19 14:57:37 2012
@@ -5026,12 +5026,18 @@
}
}
+ bool IsLoad = ISD::isNormalLoad(Ld.getNode());
+ unsigned ScalarSize = Ld.getValueType().getSizeInBits();
+
+ // Handle AVX2 in-register broadcasts.
+ if (!IsLoad && Subtarget->hasAVX2() &&
+ (ScalarSize == 32 || (Is256 && ScalarSize == 64)))
+ return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
+
// The scalar source must be a normal load.
- if (!ISD::isNormalLoad(Ld.getNode()))
+ if (!IsLoad)
return SDValue();
- unsigned ScalarSize = Ld.getValueType().getSizeInBits();
-
if (ScalarSize == 32 || (Is256 && ScalarSize == 64))
return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
Modified: llvm/trunk/test/CodeGen/X86/avx2-vbroadcast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx2-vbroadcast.ll?rev=157127&r1=157126&r2=157127&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx2-vbroadcast.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx2-vbroadcast.ll Sat May 19 14:57:37 2012
@@ -222,3 +222,40 @@
ret:
ret void
}
+
+; CHECK: _inreg0
+; CHECK: broadcastss
+; CHECK: ret
+define <8 x i32> @_inreg0(i32 %scalar) nounwind uwtable readnone ssp {
+ %in = insertelement <8 x i32> undef, i32 %scalar, i32 0
+ %wide = shufflevector <8 x i32> %in, <8 x i32> undef, <8 x i32> zeroinitializer
+ ret <8 x i32> %wide
+}
+
+; CHECK: _inreg1
+; CHECK: broadcastss
+; CHECK: ret
+define <8 x float> @_inreg1(float %scalar) nounwind uwtable readnone ssp {
+ %in = insertelement <8 x float> undef, float %scalar, i32 0
+ %wide = shufflevector <8 x float> %in, <8 x float> undef, <8 x i32> zeroinitializer
+ ret <8 x float> %wide
+}
+
+; CHECK: _inreg2
+; CHECK: broadcastss
+; CHECK: ret
+define <4 x float> @_inreg2(float %scalar) nounwind uwtable readnone ssp {
+ %in = insertelement <4 x float> undef, float %scalar, i32 0
+ %wide = shufflevector <4 x float> %in, <4 x float> undef, <4 x i32> zeroinitializer
+ ret <4 x float> %wide
+}
+
+; CHECK: _inreg3
+; CHECK: broadcastsd
+; CHECK: ret
+define <4 x double> @_inreg3(double %scalar) nounwind uwtable readnone ssp {
+ %in = insertelement <4 x double> undef, double %scalar, i32 0
+ %wide = shufflevector <4 x double> %in, <4 x double> undef, <4 x i32> zeroinitializer
+ ret <4 x double> %wide
+}
+
More information about the llvm-commits
mailing list