[llvm] r301036 - ARM: make sure we use all entries in a vector before forming a vpaddl.
Tim Northover via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 21 13:35:52 PDT 2017
Author: tnorthover
Date: Fri Apr 21 15:35:52 2017
New Revision: 301036
URL: http://llvm.org/viewvc/llvm-project?rev=301036&view=rev
Log:
ARM: make sure we use all entries in a vector before forming a vpaddl.
Otherwise there's some mismatch, and we'll either form an illegal type or an
illegal node.
Thanks to Eli Friedman for pointing out the problem with my original solution.
Modified:
llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
llvm/trunk/test/CodeGen/ARM/vpadd.ll
Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=301036&r1=301035&r2=301036&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Fri Apr 21 15:35:52 2017
@@ -9480,11 +9480,11 @@ AddCombineBUILD_VECTORToVPADDL(SDNode *N
return SDValue();
}
- // Don't generate vpaddl+vmovn; we'll match it to vpadd later. Also don't try
- // to handle an i8 -> i32 situation (or similar). vpaddl can only double the
- // size.
- if (2 * Vec.getValueType().getVectorElementType().getSizeInBits() !=
- VT.getVectorElementType().getSizeInBits())
+ // Don't generate vpaddl+vmovn; we'll match it to vpadd later. Also make sure
+ // we're using the entire input vector, otherwise there's a size/legality
+ // mismatch somewhere.
+ if (nextIndex != Vec.getValueType().getVectorNumElements() ||
+ Vec.getValueType().getVectorElementType() == VT.getVectorElementType())
return SDValue();
// Create VPADDL node.
Modified: llvm/trunk/test/CodeGen/ARM/vpadd.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/vpadd.ll?rev=301036&r1=301035&r2=301036&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/vpadd.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/vpadd.ll Fri Apr 21 15:35:52 2017
@@ -495,6 +495,15 @@ define <2 x i8> @fromExtendingExtractVec
ret <2 x i8> %x
}
+define <2 x i16> @fromExtendingExtractVectorElt_2i16(<8 x i16> %in) {
+; CHECK-LABEL: fromExtendingExtractVectorElt_2i16:
+; CHECK: vadd.i32
+ %tmp1 = shufflevector <8 x i16> %in, <8 x i16> undef, <2 x i32> <i32 0, i32 2>
+ %tmp2 = shufflevector <8 x i16> %in, <8 x i16> undef, <2 x i32> <i32 1, i32 3>
+ %x = add <2 x i16> %tmp2, %tmp1
+ ret <2 x i16> %x
+}
+
declare <4 x i16> @llvm.arm.neon.vpaddls.v4i16.v8i8(<8 x i8>) nounwind readnone
declare <2 x i32> @llvm.arm.neon.vpaddls.v2i32.v4i16(<4 x i16>) nounwind readnone
More information about the llvm-commits
mailing list