164 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=slp-vectorizer,dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s3 4target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"5target triple = "i386-apple-macosx10.8.0"6 7define double @foo(ptr nocapture %D) {8; CHECK-LABEL: @foo(9; CHECK-NEXT: br label [[TMP1:%.*]]10; CHECK: 1:11; CHECK-NEXT: [[I_02:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP11:%.*]], [[TMP1]] ]12; CHECK-NEXT: [[SUM_01:%.*]] = phi double [ 0.000000e+00, [[TMP0]] ], [ [[TMP10:%.*]], [[TMP1]] ]13; CHECK-NEXT: [[TMP2:%.*]] = shl nsw i32 [[I_02]], 114; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[D:%.*]], i32 [[TMP2]]15; CHECK-NEXT: [[TMP4:%.*]] = load <2 x double>, ptr [[TMP3]], align 416; CHECK-NEXT: [[TMP5:%.*]] = fmul <2 x double> [[TMP4]], [[TMP4]]17; CHECK-NEXT: [[TMP6:%.*]] = fmul <2 x double> [[TMP5]], [[TMP5]]18; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x double> [[TMP6]], i32 019; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[TMP6]], i32 120; CHECK-NEXT: [[TMP9:%.*]] = fadd double [[TMP7]], [[TMP8]]21; CHECK-NEXT: [[TMP10]] = fadd double [[SUM_01]], [[TMP9]]22; CHECK-NEXT: [[TMP11]] = add nsw i32 [[I_02]], 123; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP11]], 10024; CHECK-NEXT: br i1 [[EXITCOND]], label [[TMP12:%.*]], label [[TMP1]]25; CHECK: 12:26; CHECK-NEXT: ret double [[TMP10]]27;28 br label %129 30; <label>:1 ; preds = %1, %031 %i.02 = phi i32 [ 0, %0 ], [ %10, %1 ]32 %sum.01 = phi double [ 0.000000e+00, %0 ], [ %9, %1 ]33 %2 = shl nsw i32 %i.02, 134 %3 = getelementptr inbounds double, ptr %D, i32 %235 %4 = load double, ptr %3, align 436 %A4 = fmul double %4, %437 %A42 = fmul double %A4, %A438 %5 = or disjoint i32 %2, 139 %6 = getelementptr inbounds double, ptr %D, i32 %540 %7 = load double, ptr %6, align 441 %A7 = fmul double %7, %742 %A72 = fmul double %A7, %A743 %8 = fadd double %A42, %A7244 %9 = fadd double %sum.01, %845 %10 = add nsw i32 %i.02, 146 %exitcond = icmp eq i32 %10, 10047 br i1 %exitcond, label %11, label %148 49; <label>:11 ; preds = %150 ret double %951}52 53define i1 @two_wide_fcmp_reduction(<2 x double> %a0) {54; CHECK-LABEL: @two_wide_fcmp_reduction(55; CHECK-NEXT: [[A:%.*]] = fcmp ogt <2 x double> [[A0:%.*]], splat (double 1.000000e+00)56; CHECK-NEXT: [[B:%.*]] = extractelement <2 x i1> [[A]], i32 057; CHECK-NEXT: [[C:%.*]] = extractelement <2 x i1> [[A]], i32 158; CHECK-NEXT: [[D:%.*]] = and i1 [[B]], [[C]]59; CHECK-NEXT: ret i1 [[D]]60;61 %a = fcmp ogt <2 x double> %a0, <double 1.0, double 1.0>62 %b = extractelement <2 x i1> %a, i32 063 %c = extractelement <2 x i1> %a, i32 164 %d = and i1 %b, %c65 ret i1 %d66}67 68define double @fadd_reduction(<2 x double> %a0) {69; CHECK-LABEL: @fadd_reduction(70; CHECK-NEXT: [[A:%.*]] = fadd fast <2 x double> [[A0:%.*]], splat (double 1.000000e+00)71; CHECK-NEXT: [[B:%.*]] = extractelement <2 x double> [[A]], i32 072; CHECK-NEXT: [[C:%.*]] = extractelement <2 x double> [[A]], i32 173; CHECK-NEXT: [[D:%.*]] = fadd fast double [[B]], [[C]]74; CHECK-NEXT: ret double [[D]]75;76 %a = fadd fast <2 x double> %a0, <double 1.000000e+00, double 1.000000e+00>77 %b = extractelement <2 x double> %a, i32 078 %c = extractelement <2 x double> %a, i32 179 %d = fadd fast double %b, %c80 ret double %d81}82 83; PR43745 https://bugs.llvm.org/show_bug.cgi?id=4374584 85define i1 @fcmp_lt_gt(double %a, double %b, double %c) {86; CHECK-LABEL: @fcmp_lt_gt(87; CHECK-NEXT: entry:88; CHECK-NEXT: [[FNEG:%.*]] = fneg double [[B:%.*]]89; CHECK-NEXT: [[MUL:%.*]] = fmul double [[A:%.*]], 2.000000e+0090; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x double> poison, double [[C:%.*]], i32 191; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x double> [[TMP0]], double [[FNEG]], i32 092; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> poison, <2 x i32> <i32 1, i32 poison>93; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x double> [[TMP2]], double [[B]], i32 194; CHECK-NEXT: [[TMP4:%.*]] = fsub <2 x double> [[TMP1]], [[TMP3]]95; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x double> poison, double [[MUL]], i32 096; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x double> [[TMP5]], <2 x double> poison, <2 x i32> zeroinitializer97; CHECK-NEXT: [[TMP7:%.*]] = fdiv <2 x double> [[TMP4]], [[TMP6]]98; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[TMP7]], i32 199; CHECK-NEXT: [[CMP:%.*]] = fcmp olt double [[TMP8]], 0x3EB0C6F7A0B5ED8D100; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x double> [[TMP7]], i32 0101; CHECK-NEXT: [[CMP4:%.*]] = fcmp olt double [[TMP9]], 0x3EB0C6F7A0B5ED8D102; CHECK-NEXT: [[OR_COND:%.*]] = and i1 [[CMP]], [[CMP4]]103; CHECK-NEXT: br i1 [[OR_COND]], label [[CLEANUP:%.*]], label [[LOR_LHS_FALSE:%.*]]104; CHECK: lor.lhs.false:105; CHECK-NEXT: [[TMP10:%.*]] = fcmp ule <2 x double> [[TMP7]], splat (double 1.000000e+00)106; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP10]], i32 0107; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[TMP10]], i32 1108; CHECK-NEXT: [[NOT_OR_COND9:%.*]] = or i1 [[TMP11]], [[TMP12]]109; CHECK-NEXT: ret i1 [[NOT_OR_COND9]]110; CHECK: cleanup:111; CHECK-NEXT: ret i1 false112;113entry:114 %fneg = fneg double %b115 %add = fsub double %c, %b116 %mul = fmul double %a, 2.000000e+00117 %div = fdiv double %add, %mul118 %sub = fsub double %fneg, %c119 %div3 = fdiv double %sub, %mul120 %cmp = fcmp olt double %div, 0x3EB0C6F7A0B5ED8D121 %cmp4 = fcmp olt double %div3, 0x3EB0C6F7A0B5ED8D122 %or.cond = and i1 %cmp, %cmp4123 br i1 %or.cond, label %cleanup, label %lor.lhs.false124 125lor.lhs.false:126 %cmp5 = fcmp ule double %div, 1.000000e+00127 %cmp7 = fcmp ule double %div3, 1.000000e+00128 %not.or.cond9 = or i1 %cmp7, %cmp5129 ret i1 %not.or.cond9130 131cleanup:132 ret i1 false133}134 135define i1 @fcmp_lt(double %a, double %b, double %c) {136; CHECK-LABEL: @fcmp_lt(137; CHECK-NEXT: [[FNEG:%.*]] = fneg double [[B:%.*]]138; CHECK-NEXT: [[MUL:%.*]] = fmul double [[A:%.*]], 2.000000e+00139; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x double> poison, double [[C:%.*]], i32 1140; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double [[FNEG]], i32 0141; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <2 x double> [[TMP2]], <2 x double> poison, <2 x i32> <i32 1, i32 poison>142; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x double> [[TMP3]], double [[B]], i32 1143; CHECK-NEXT: [[TMP5:%.*]] = fsub <2 x double> [[TMP2]], [[TMP4]]144; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x double> poison, double [[MUL]], i32 0145; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x double> [[TMP6]], <2 x double> poison, <2 x i32> zeroinitializer146; CHECK-NEXT: [[TMP8:%.*]] = fdiv <2 x double> [[TMP5]], [[TMP7]]147; CHECK-NEXT: [[TMP9:%.*]] = fcmp uge <2 x double> [[TMP8]], splat (double 0x3EB0C6F7A0B5ED8D)148; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP9]], i32 0149; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP9]], i32 1150; CHECK-NEXT: [[NOT_OR_COND:%.*]] = or i1 [[TMP10]], [[TMP11]]151; CHECK-NEXT: ret i1 [[NOT_OR_COND]]152;153 %fneg = fneg double %b154 %add = fsub double %c, %b155 %mul = fmul double %a, 2.000000e+00156 %div = fdiv double %add, %mul157 %sub = fsub double %fneg, %c158 %div3 = fdiv double %sub, %mul159 %cmp = fcmp uge double %div, 0x3EB0C6F7A0B5ED8D160 %cmp4 = fcmp uge double %div3, 0x3EB0C6F7A0B5ED8D161 %not.or.cond = or i1 %cmp4, %cmp162 ret i1 %not.or.cond163}164