if (!Subtarget->hasAVX())
return false;
- // Match any permutation of 128-bit vector with 64-bit types
- if (NumLanes == 1 && NumElts != 2)
- return false;
-
- // Only match 256-bit with 32 types
- if (VT.getSizeInBits() == 256 && NumElts != 4)
+ // Only match 256-bit with 64-bit types
+ if (VT.getSizeInBits() != 256 || NumElts != 4)
return false;
// The mask on the high lane is independent of the low. Both can match
if (!Subtarget->hasAVX())
return false;
- // Match any permutation of 128-bit vector with 32-bit types
- if (NumLanes == 1 && NumElts != 4)
- return false;
-
- // Only match 256-bit with 32 types
- if (VT.getSizeInBits() == 256 && NumElts != 8)
+ // Only match 256-bit with 32-bit types
+ if (VT.getSizeInBits() != 256 || NumElts != 8)
return false;
// The mask on the high lane should be the same as the low. Actually,
--- /dev/null
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s
+
+; PR11102
+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: test1:
+; CHECK: vshufps
+; CHECK: vpshufd
+}