1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -mcpu=penryn | FileCheck %s --check-prefix=SSE 2*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -mcpu=sandybridge | FileCheck %s --check-prefix=AVX 3*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -mcpu=haswell | FileCheck %s --check-prefix=AVX2 4*9880d681SAndroid Build Coastguard Worker; This checks that lowering for creation of constant vectors is sane and 5*9880d681SAndroid Build Coastguard Worker; doesn't use redundant shuffles. (fixes PR22276) 6*9880d681SAndroid Build Coastguard Workertarget triple = "x86_64-unknown-unknown" 7*9880d681SAndroid Build Coastguard Worker 8*9880d681SAndroid Build Coastguard Workerdefine <4 x i32> @zero_vector() { 9*9880d681SAndroid Build Coastguard Worker; SSE-LABEL: zero_vector: 10*9880d681SAndroid Build Coastguard Worker; SSE: xorps %xmm0, %xmm0 11*9880d681SAndroid Build Coastguard Worker; SSE-NEXT: retq 12*9880d681SAndroid Build Coastguard Worker; AVX-LABEL: zero_vector: 13*9880d681SAndroid Build Coastguard Worker; AVX: vxorps %xmm0, %xmm0, %xmm0 14*9880d681SAndroid Build Coastguard Worker; AVX-NEXT: retq 15*9880d681SAndroid Build Coastguard Worker; AVX2-LABEL: zero_vector: 16*9880d681SAndroid Build Coastguard Worker; AVX2: vxorps %xmm0, %xmm0, %xmm0 17*9880d681SAndroid Build Coastguard Worker; AVX2-NEXT: retq 18*9880d681SAndroid Build Coastguard Worker %zero = insertelement <4 x i32> undef, i32 0, i32 0 19*9880d681SAndroid Build Coastguard Worker %splat = shufflevector <4 x i32> %zero, <4 x i32> undef, <4 x i32> zeroinitializer 20*9880d681SAndroid Build Coastguard Worker ret <4 x i32> %splat 21*9880d681SAndroid Build Coastguard Worker} 22*9880d681SAndroid Build Coastguard Worker 23*9880d681SAndroid Build Coastguard Worker; Note that for the "const_vector" versions, lowering that uses a shuffle 24*9880d681SAndroid Build Coastguard Worker; instead of a load would be legitimate, if it's a single broadcast shuffle. 25*9880d681SAndroid Build Coastguard Worker; (as opposed to the previous mess) 26*9880d681SAndroid Build Coastguard Worker; However, this is not the current preferred lowering. 27*9880d681SAndroid Build Coastguard Workerdefine <4 x i32> @const_vector() { 28*9880d681SAndroid Build Coastguard Worker; SSE-LABEL: const_vector: 29*9880d681SAndroid Build Coastguard Worker; SSE: movaps {{.*}}, %xmm0 # xmm0 = [42,42,42,42] 30*9880d681SAndroid Build Coastguard Worker; SSE-NEXT: retq 31*9880d681SAndroid Build Coastguard Worker; AVX-LABEL: const_vector: 32*9880d681SAndroid Build Coastguard Worker; AVX: vmovaps {{.*}}, %xmm0 # xmm0 = [42,42,42,42] 33*9880d681SAndroid Build Coastguard Worker; AVX-NEXT: retq 34*9880d681SAndroid Build Coastguard Worker; AVX2-LABEL: const_vector: 35*9880d681SAndroid Build Coastguard Worker; AVX2: vbroadcastss {{[^%].*}}, %xmm0 36*9880d681SAndroid Build Coastguard Worker; AVX2-NEXT: retq 37*9880d681SAndroid Build Coastguard Worker %const = insertelement <4 x i32> undef, i32 42, i32 0 38*9880d681SAndroid Build Coastguard Worker %splat = shufflevector <4 x i32> %const, <4 x i32> undef, <4 x i32> zeroinitializer 39*9880d681SAndroid Build Coastguard Worker ret <4 x i32> %splat 40*9880d681SAndroid Build Coastguard Worker} 41