1; RUN: llc -mcpu=pwr8 -mtriple=powerpc64-unknown-linux-gnu -mattr=+power8-vector < %s | FileCheck %s 2; RUN: llc -mcpu=pwr7 -mtriple=powerpc64-unknown-linux-gnu < %s | FileCheck -check-prefix=CHECK-PWR7 %s 3 4define void @VPKUDUM_unary(<2 x i64>* %A) { 5entry: 6 %tmp = load <2 x i64>, <2 x i64>* %A 7 %tmp2 = bitcast <2 x i64> %tmp to <4 x i32> 8 %tmp3 = extractelement <4 x i32> %tmp2, i32 1 9 %tmp4 = extractelement <4 x i32> %tmp2, i32 3 10 %tmp5 = insertelement <4 x i32> undef, i32 %tmp3, i32 0 11 %tmp6 = insertelement <4 x i32> %tmp5, i32 %tmp4, i32 1 12 %tmp7 = insertelement <4 x i32> %tmp6, i32 %tmp3, i32 2 13 %tmp8 = insertelement <4 x i32> %tmp7, i32 %tmp4, i32 3 14 %tmp9 = bitcast <4 x i32> %tmp8 to <2 x i64> 15 store <2 x i64> %tmp9, <2 x i64>* %A 16 ret void 17} 18 19; CHECK-LABEL: @VPKUDUM_unary 20; CHECK-NOT: vperm 21; CHECK-NOT: vmrglw 22; CHECK-NOT: vmrghw 23; CHECK: vpkudum 24; CHECK-PWR7: vmrglw 25; CHECK-PWR7: vmrghw 26; CHECK-PWR7: vmrglw 27 28define void @VPKUDUM(<2 x i64>* %A, <2 x i64>* %B) { 29entry: 30 %tmp = load <2 x i64>, <2 x i64>* %A 31 %tmp2 = bitcast <2 x i64> %tmp to <4 x i32> 32 %tmp3 = load <2 x i64>, <2 x i64>* %B 33 %tmp4 = bitcast <2 x i64> %tmp3 to <4 x i32> 34 %tmp5 = extractelement <4 x i32> %tmp2, i32 1 35 %tmp6 = extractelement <4 x i32> %tmp2, i32 3 36 %tmp7 = extractelement <4 x i32> %tmp4, i32 1 37 %tmp8 = extractelement <4 x i32> %tmp4, i32 3 38 %tmp9 = insertelement <4 x i32> undef, i32 %tmp5, i32 0 39 %tmp10 = insertelement <4 x i32> %tmp9, i32 %tmp6, i32 1 40 %tmp11 = insertelement <4 x i32> %tmp10, i32 %tmp7, i32 2 41 %tmp12 = insertelement <4 x i32> %tmp11, i32 %tmp8, i32 3 42 %tmp13 = bitcast <4 x i32> %tmp12 to <2 x i64> 43 store <2 x i64> %tmp13, <2 x i64>* %A 44 ret void 45} 46 47; CHECK-LABEL: @VPKUDUM 48; CHECK-NOT: vperm 49; CHECK-NOT: vmrglw 50; CHECK-NOT: vmrghw 51; CHECK: vpkudum 52; CHECK-PWR7: vmrglw 53; CHECK-PWR7: vmrghw 54; CHECK-PWR7: vmrglw 55