| 1 |
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| 2 |
; RUN: opt -S -passes='default<O2>' -mtriple=powerpc64le-- -o - %s | \ |
| 3 |
; RUN: FileCheck %s |
| 4 |
|
| 5 |
%0 = type <{ double }> |
| 6 |
|
| 7 |
define dso_local void @test(ptr %arg) #0 { |
| 8 |
; CHECK-LABEL: @test( |
| 9 |
; CHECK-NEXT: bb: |
| 10 |
; CHECK-NEXT: [[I19:%.*]] = getelementptr i8, ptr [[ARG:%.*]], i64 -32 |
| 11 |
; CHECK-NEXT: [[I21:%.*]] = getelementptr inbounds nuw i8, ptr [[ARG]], i64 48 |
| 12 |
; CHECK-NEXT: [[I21_REPACK1:%.*]] = getelementptr inbounds nuw i8, ptr [[ARG]], i64 64 |
| 13 |
; CHECK-NEXT: br label [[BB16:%.*]] |
| 14 |
; CHECK: bb16: |
| 15 |
; CHECK-NEXT: [[I20:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 16 |
; CHECK-NEXT: [[I24:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20]]) |
| 17 |
; CHECK-NEXT: [[I24_ELT:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24]], 0 |
| 18 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT]], ptr [[I21]], align 16 |
| 19 |
; CHECK-NEXT: [[I24_ELT1:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24]], 1 |
| 20 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1]], ptr [[I21_REPACK1]], align 16 |
| 21 |
; CHECK-NEXT: [[I20_1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 22 |
; CHECK-NEXT: [[I24_1:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_1]]) |
| 23 |
; CHECK-NEXT: [[I24_ELT_1:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_1]], 0 |
| 24 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_1]], ptr [[I21]], align 16 |
| 25 |
; CHECK-NEXT: [[I24_ELT1_1:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_1]], 1 |
| 26 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_1]], ptr [[I21_REPACK1]], align 16 |
| 27 |
; CHECK-NEXT: [[I20_2:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 28 |
; CHECK-NEXT: [[I24_2:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_2]]) |
| 29 |
; CHECK-NEXT: [[I24_ELT_2:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_2]], 0 |
| 30 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_2]], ptr [[I21]], align 16 |
| 31 |
; CHECK-NEXT: [[I24_ELT1_2:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_2]], 1 |
| 32 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_2]], ptr [[I21_REPACK1]], align 16 |
| 33 |
; CHECK-NEXT: [[I20_3:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 34 |
; CHECK-NEXT: [[I24_3:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_3]]) |
| 35 |
; CHECK-NEXT: [[I24_ELT_3:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_3]], 0 |
| 36 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_3]], ptr [[I21]], align 16 |
| 37 |
; CHECK-NEXT: [[I24_ELT1_3:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_3]], 1 |
| 38 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_3]], ptr [[I21_REPACK1]], align 16 |
| 39 |
; CHECK-NEXT: [[I20_4:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 40 |
; CHECK-NEXT: [[I24_4:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_4]]) |
| 41 |
; CHECK-NEXT: [[I24_ELT_4:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_4]], 0 |
| 42 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_4]], ptr [[I21]], align 16 |
| 43 |
; CHECK-NEXT: [[I24_ELT1_4:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_4]], 1 |
| 44 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_4]], ptr [[I21_REPACK1]], align 16 |
| 45 |
; CHECK-NEXT: [[I20_5:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 46 |
; CHECK-NEXT: [[I24_5:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_5]]) |
| 47 |
; CHECK-NEXT: [[I24_ELT_5:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_5]], 0 |
| 48 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_5]], ptr [[I21]], align 16 |
| 49 |
; CHECK-NEXT: [[I24_ELT1_5:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_5]], 1 |
| 50 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_5]], ptr [[I21_REPACK1]], align 16 |
| 51 |
; CHECK-NEXT: [[I20_6:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 52 |
; CHECK-NEXT: [[I24_6:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_6]]) |
| 53 |
; CHECK-NEXT: [[I24_ELT_6:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_6]], 0 |
| 54 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_6]], ptr [[I21]], align 16 |
| 55 |
; CHECK-NEXT: [[I24_ELT1_6:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_6]], 1 |
| 56 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_6]], ptr [[I21_REPACK1]], align 16 |
| 57 |
; CHECK-NEXT: [[I20_7:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 58 |
; CHECK-NEXT: [[I24_7:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_7]]) |
| 59 |
; CHECK-NEXT: [[I24_ELT_7:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_7]], 0 |
| 60 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_7]], ptr [[I21]], align 16 |
| 61 |
; CHECK-NEXT: [[I24_ELT1_7:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_7]], 1 |
| 62 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_7]], ptr [[I21_REPACK1]], align 16 |
| 63 |
; CHECK-NEXT: [[I20_8:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 64 |
; CHECK-NEXT: [[I24_8:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_8]]) |
| 65 |
; CHECK-NEXT: [[I24_ELT_8:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_8]], 0 |
| 66 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_8]], ptr [[I21]], align 16 |
| 67 |
; CHECK-NEXT: [[I24_ELT1_8:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_8]], 1 |
| 68 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_8]], ptr [[I21_REPACK1]], align 16 |
| 69 |
; CHECK-NEXT: [[I20_9:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 70 |
; CHECK-NEXT: [[I24_9:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_9]]) |
| 71 |
; CHECK-NEXT: [[I24_ELT_9:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_9]], 0 |
| 72 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_9]], ptr [[I21]], align 16 |
| 73 |
; CHECK-NEXT: [[I24_ELT1_9:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_9]], 1 |
| 74 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_9]], ptr [[I21_REPACK1]], align 16 |
| 75 |
; CHECK-NEXT: [[I20_10:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 76 |
; CHECK-NEXT: [[I24_10:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_10]]) |
| 77 |
; CHECK-NEXT: [[I24_ELT_10:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_10]], 0 |
| 78 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_10]], ptr [[I21]], align 16 |
| 79 |
; CHECK-NEXT: [[I24_ELT1_10:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_10]], 1 |
| 80 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_10]], ptr [[I21_REPACK1]], align 16 |
| 81 |
; CHECK-NEXT: [[I20_11:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 82 |
; CHECK-NEXT: [[I24_11:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_11]]) |
| 83 |
; CHECK-NEXT: [[I24_ELT_11:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_11]], 0 |
| 84 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_11]], ptr [[I21]], align 16 |
| 85 |
; CHECK-NEXT: [[I24_ELT1_11:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_11]], 1 |
| 86 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_11]], ptr [[I21_REPACK1]], align 16 |
| 87 |
; CHECK-NEXT: [[I20_12:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 88 |
; CHECK-NEXT: [[I24_12:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_12]]) |
| 89 |
; CHECK-NEXT: [[I24_ELT_12:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_12]], 0 |
| 90 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_12]], ptr [[I21]], align 16 |
| 91 |
; CHECK-NEXT: [[I24_ELT1_12:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_12]], 1 |
| 92 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_12]], ptr [[I21_REPACK1]], align 16 |
| 93 |
; CHECK-NEXT: [[I20_13:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 94 |
; CHECK-NEXT: [[I24_13:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_13]]) |
| 95 |
; CHECK-NEXT: [[I24_ELT_13:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_13]], 0 |
| 96 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_13]], ptr [[I21]], align 16 |
| 97 |
; CHECK-NEXT: [[I24_ELT1_13:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_13]], 1 |
| 98 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_13]], ptr [[I21_REPACK1]], align 16 |
| 99 |
; CHECK-NEXT: [[I20_14:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 100 |
; CHECK-NEXT: [[I24_14:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_14]]) |
| 101 |
; CHECK-NEXT: [[I24_ELT_14:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_14]], 0 |
| 102 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_14]], ptr [[I21]], align 16 |
| 103 |
; CHECK-NEXT: [[I24_ELT1_14:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_14]], 1 |
| 104 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_14]], ptr [[I21_REPACK1]], align 16 |
| 105 |
; CHECK-NEXT: [[I20_15:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[I19]]) |
| 106 |
; CHECK-NEXT: [[I24_15:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[I20_15]]) |
| 107 |
; CHECK-NEXT: [[I24_ELT_15:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_15]], 0 |
| 108 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT_15]], ptr [[I21]], align 16 |
| 109 |
; CHECK-NEXT: [[I24_ELT1_15:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[I24_15]], 1 |
| 110 |
; CHECK-NEXT: store <16 x i8> [[I24_ELT1_15]], ptr [[I21_REPACK1]], align 16 |
| 111 |
; CHECK-NEXT: br label [[BB16]], !llvm.loop [[LOOP0:![0-9]+]] |
| 112 |
; |
| 113 |
bb: |
| 114 |
%i = alloca ptr, align 8 |
| 115 |
store ptr %arg, ptr %i, align 8 |
| 116 |
%i1 = alloca ptr, align 8 |
| 117 |
%i2 = alloca ptr, align 8 |
| 118 |
%i3 = alloca i32, align 4 |
| 119 |
%i4 = alloca i32, align 4 |
| 120 |
%i5 = alloca i64, align 8 |
| 121 |
%i6 = alloca i64, align 8 |
| 122 |
%i7 = alloca <256 x i1>, align 32 |
| 123 |
%i8 = load ptr, ptr %i, align 8 |
| 124 |
%i9 = load i32, ptr %i8, align 4 |
| 125 |
%i10 = sub nsw i32 %i9, 0 |
| 126 |
store i32 %i10, ptr %i4, align 4 |
| 127 |
%i11 = load i32, ptr %i4, align 4 |
| 128 |
%i12 = ashr i32 %i11, 5 |
| 129 |
%i13 = sext i32 %i12 to i64 |
| 130 |
%i14 = load i64, ptr %i6, align 8 |
| 131 |
%i15 = sub nsw i64 %i14, 1 |
| 132 |
br label %bb16 |
| 133 |
|
| 134 |
bb16: ; preds = %bb16, %bb |
| 135 |
%i17 = load i64, ptr %i5, align 8 |
| 136 |
%i18 = icmp sge i64 %i17, 1 |
| 137 |
%i19 = getelementptr i8, ptr %arg, i64 -32 |
| 138 |
%i20 = call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %i19) |
| 139 |
store <256 x i1> %i20, ptr %i7, align 32 |
| 140 |
%i21 = getelementptr inbounds i8, ptr %arg, i64 48 |
| 141 |
%i23 = load <256 x i1>, ptr %i7, align 32 |
| 142 |
%i24 = call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> %i23) |
| 143 |
store { <16 x i8>, <16 x i8> } %i24, ptr %i21, align 16 |
| 144 |
br label %bb16, !llvm.loop !1 |
| 145 |
} |
| 146 |
|
| 147 |
; Function Attrs: argmemonly nounwind readonly |
| 148 |
declare <256 x i1> @llvm.ppc.vsx.lxvp(ptr) #1 |
| 149 |
|
| 150 |
; Function Attrs: nounwind readnone |
| 151 |
declare { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1>) #2 |
| 152 |
|
| 153 |
attributes #0 = { "no-trapping-math"="true" } |
| 154 |
attributes #1 = { argmemonly nounwind readonly } |
| 155 |
attributes #2 = { nounwind readnone } |
| 156 |
|
| 157 |
!1 = distinct !{!1, !2, !3, !4} |
| 158 |
!2 = !{!"llvm.loop.vectorize.width", i32 1} |
| 159 |
!3 = !{!"llvm.loop.interleave.count", i32 1} |
| 160 |
!4 = !{!"llvm.loop.unroll.count", i32 16} |