94 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -verify-machineinstrs -mcpu=pwr8 -mtriple=powerpc64-unknown-linux-gnu -mattr=+power8-vector < %s | FileCheck %s3; RUN: llc -verify-machineinstrs -mcpu=pwr8 -mtriple=powerpc64-ibm-aix-xcoff -vec-extabi -mattr=+power8-vector < %s | FileCheck %s4; RUN: llc -verify-machineinstrs -mcpu=pwr7 -mtriple=powerpc64-unknown-linux-gnu < %s | FileCheck -check-prefix=CHECK-PWR7 %s5; RUN: llc -verify-machineinstrs -mcpu=pwr7 -mtriple=powerpc64-ibm-aix-xcoff -vec-extabi < %s | FileCheck -check-prefix=CHECK-PWR7-AIX %s6 7define void @VPKUDUM_unary(ptr %A) {8; CHECK-LABEL: VPKUDUM_unary:9; CHECK: # %bb.0: # %entry10; CHECK-NEXT: lxvw4x 34, 0, 311; CHECK-NEXT: vpkudum 2, 2, 212; CHECK-NEXT: stxvw4x 34, 0, 313; CHECK-NEXT: blr14;15; CHECK-PWR7-LABEL: VPKUDUM_unary:16; CHECK-PWR7: # %bb.0: # %entry17; CHECK-PWR7-NEXT: addis 4, 2, .LCPI0_0@toc@ha18; CHECK-PWR7-NEXT: lxvw4x 34, 0, 319; CHECK-PWR7-NEXT: addi 4, 4, .LCPI0_0@toc@l20; CHECK-PWR7-NEXT: lxvw4x 35, 0, 421; CHECK-PWR7-NEXT: vperm 2, 2, 2, 322; CHECK-PWR7-NEXT: stxvw4x 34, 0, 323; CHECK-PWR7-NEXT: blr24;25; CHECK-PWR7-AIX-LABEL: VPKUDUM_unary:26; CHECK-PWR7-AIX: # %bb.0: # %entry27; CHECK-PWR7-AIX-NEXT: ld 4, L..C0(2) # %const.028; CHECK-PWR7-AIX-NEXT: lxvw4x 34, 0, 329; CHECK-PWR7-AIX-NEXT: lxvw4x 35, 0, 430; CHECK-PWR7-AIX-NEXT: vperm 2, 2, 2, 331; CHECK-PWR7-AIX-NEXT: stxvw4x 34, 0, 332; CHECK-PWR7-AIX-NEXT: blr33entry:34 %tmp = load <2 x i64>, ptr %A35 %tmp2 = bitcast <2 x i64> %tmp to <4 x i32>36 %tmp3 = extractelement <4 x i32> %tmp2, i32 137 %tmp4 = extractelement <4 x i32> %tmp2, i32 338 %tmp5 = insertelement <4 x i32> undef, i32 %tmp3, i32 039 %tmp6 = insertelement <4 x i32> %tmp5, i32 %tmp4, i32 140 %tmp7 = insertelement <4 x i32> %tmp6, i32 %tmp3, i32 241 %tmp8 = insertelement <4 x i32> %tmp7, i32 %tmp4, i32 342 %tmp9 = bitcast <4 x i32> %tmp8 to <2 x i64>43 store <2 x i64> %tmp9, ptr %A44 ret void45}46 47define void @VPKUDUM(ptr %A, ptr %B) {48; CHECK-LABEL: VPKUDUM:49; CHECK: # %bb.0: # %entry50; CHECK-NEXT: lxvw4x 34, 0, 351; CHECK-NEXT: lxvw4x 35, 0, 452; CHECK-NEXT: vpkudum 2, 2, 353; CHECK-NEXT: stxvw4x 34, 0, 354; CHECK-NEXT: blr55;56; CHECK-PWR7-LABEL: VPKUDUM:57; CHECK-PWR7: # %bb.0: # %entry58; CHECK-PWR7-NEXT: lxvw4x 35, 0, 459; CHECK-PWR7-NEXT: addis 4, 2, .LCPI1_0@toc@ha60; CHECK-PWR7-NEXT: lxvw4x 34, 0, 361; CHECK-PWR7-NEXT: addi 4, 4, .LCPI1_0@toc@l62; CHECK-PWR7-NEXT: lxvw4x 36, 0, 463; CHECK-PWR7-NEXT: vperm 2, 2, 3, 464; CHECK-PWR7-NEXT: stxvw4x 34, 0, 365; CHECK-PWR7-NEXT: blr66;67; CHECK-PWR7-AIX-LABEL: VPKUDUM:68; CHECK-PWR7-AIX: # %bb.0: # %entry69; CHECK-PWR7-AIX-NEXT: lxvw4x 35, 0, 470; CHECK-PWR7-AIX-NEXT: ld 4, L..C1(2) # %const.071; CHECK-PWR7-AIX-NEXT: lxvw4x 34, 0, 372; CHECK-PWR7-AIX-NEXT: lxvw4x 36, 0, 473; CHECK-PWR7-AIX-NEXT: vperm 2, 2, 3, 474; CHECK-PWR7-AIX-NEXT: stxvw4x 34, 0, 375; CHECK-PWR7-AIX-NEXT: blr76entry:77 %tmp = load <2 x i64>, ptr %A78 %tmp2 = bitcast <2 x i64> %tmp to <4 x i32>79 %tmp3 = load <2 x i64>, ptr %B80 %tmp4 = bitcast <2 x i64> %tmp3 to <4 x i32>81 %tmp5 = extractelement <4 x i32> %tmp2, i32 182 %tmp6 = extractelement <4 x i32> %tmp2, i32 383 %tmp7 = extractelement <4 x i32> %tmp4, i32 184 %tmp8 = extractelement <4 x i32> %tmp4, i32 385 %tmp9 = insertelement <4 x i32> undef, i32 %tmp5, i32 086 %tmp10 = insertelement <4 x i32> %tmp9, i32 %tmp6, i32 187 %tmp11 = insertelement <4 x i32> %tmp10, i32 %tmp7, i32 288 %tmp12 = insertelement <4 x i32> %tmp11, i32 %tmp8, i32 389 %tmp13 = bitcast <4 x i32> %tmp12 to <2 x i64>90 store <2 x i64> %tmp13, ptr %A91 ret void92}93 94