brintos

brintos / llvm-project-archived public Read only

0
0
Text · 14.9 KiB · e13dfce Raw
251 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -mtriple=riscv64 -mcpu=sifive-x280 -passes=slp-vectorizer -S -slp-revec -slp-max-reg-size=1024 -slp-threshold=-100 %s | FileCheck --check-prefixes=CHECK,POWEROF2 %s3; RUN: opt -mtriple=riscv64 -mcpu=sifive-x280 -passes=slp-vectorizer -S -slp-revec -slp-max-reg-size=1024 -slp-threshold=-100 -slp-vectorize-non-power-of-2 %s | FileCheck --check-prefixes=CHECK,NONPOWEROF2 %s4 5define i32 @test() {6; CHECK-LABEL: @test(7; CHECK-NEXT:  entry:8; CHECK-NEXT:    br label [[IF_END_I87:%.*]]9; CHECK:       if.end.i87:10; CHECK-NEXT:    [[TMP0:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> align 4 getelementptr (i32, <4 x ptr> <ptr inttoptr (i64 64036 to ptr), ptr inttoptr (i64 64036 to ptr), ptr inttoptr (i64 64064 to ptr), ptr inttoptr (i64 64064 to ptr)>, <4 x i64> <i64 0, i64 1, i64 0, i64 1>), <4 x i1> splat (i1 true), <4 x i32> poison)11; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> <i32 undef, i32 undef, i32 0, i32 0>, <4 x i32> <i32 0, i32 1, i32 6, i32 7>12; CHECK-NEXT:    switch i32 0, label [[SW_BB509_I:%.*]] [13; CHECK-NEXT:      i32 1, label [[SW_BB509_I]]14; CHECK-NEXT:      i32 0, label [[IF_THEN458_I:%.*]]15; CHECK-NEXT:    ]16; CHECK:       if.then458.i:17; CHECK-NEXT:    br label [[SW_BB509_I]]18; CHECK:       sw.bb509.i:19; CHECK-NEXT:    [[TMP4:%.*]] = phi <4 x i32> [ [[TMP0]], [[IF_THEN458_I]] ], [ [[TMP3]], [[IF_END_I87]] ], [ [[TMP3]], [[IF_END_I87]] ]20; CHECK-NEXT:    ret i32 021;22entry:23  %getelementptr0 = getelementptr i8, ptr null, i64 6403624  %getelementptr1 = getelementptr i8, ptr null, i64 6406425  br label %if.end.i8726 27if.end.i87:                                       ; preds = %entry28  %0 = load <2 x i32>, ptr %getelementptr0, align 429  %1 = load <2 x i32>, ptr %getelementptr1, align 830  switch i32 0, label %sw.bb509.i [31  i32 1, label %sw.bb509.i32  i32 0, label %if.then458.i33  ]34 35if.then458.i:                                     ; preds = %if.end.i8736  br label %sw.bb509.i37 38sw.bb509.i:                                       ; preds = %if.then458.i, %if.end.i87, %if.end.i8739  %4 = phi <2 x i32> [ %0, %if.then458.i ], [ %0, %if.end.i87 ], [ %0, %if.end.i87 ]40  %5 = phi <2 x i32> [ %1, %if.then458.i ], [ zeroinitializer, %if.end.i87 ], [ zeroinitializer, %if.end.i87 ]41  ret i32 042}43 44define void @test2() {45; CHECK-LABEL: @test2(46; CHECK-NEXT:  entry:47; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr null, i64 13248; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr null, i64 20049; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr null, i64 30050; CHECK-NEXT:    [[TMP3:%.*]] = load <8 x float>, ptr [[TMP1]], align 451; CHECK-NEXT:    [[TMP4:%.*]] = load <8 x float>, ptr [[TMP2]], align 452; CHECK-NEXT:    [[TMP5:%.*]] = load <16 x float>, ptr [[TMP0]], align 453; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <8 x float> [[TMP4]], <8 x float> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>54; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <8 x float> [[TMP3]], <8 x float> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>55; CHECK-NEXT:    [[TMP10:%.*]] = shufflevector <32 x float> [[TMP6]], <32 x float> [[TMP7]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>56; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <16 x float> [[TMP5]], <16 x float> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>57; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <32 x float> [[TMP10]], <32 x float> [[TMP11]], <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47>58; CHECK-NEXT:    [[TMP9:%.*]] = fpext <32 x float> [[TMP8]] to <32 x double>59; CHECK-NEXT:    [[TMP14:%.*]] = fadd <32 x double> zeroinitializer, [[TMP9]]60; CHECK-NEXT:    [[TMP15:%.*]] = fptrunc <32 x double> [[TMP14]] to <32 x float>61; CHECK-NEXT:    [[TMP16:%.*]] = fcmp ogt <32 x float> zeroinitializer, [[TMP15]]62; CHECK-NEXT:    ret void63;64entry:65  %0 = getelementptr i8, ptr null, i64 13266  %1 = getelementptr i8, ptr null, i64 16467  %2 = getelementptr i8, ptr null, i64 20068  %3 = getelementptr i8, ptr null, i64 30069  %4 = load <8 x float>, ptr %0, align 470  %5 = load <8 x float>, ptr %1, align 471  %6 = load <8 x float>, ptr %2, align 472  %7 = load <8 x float>, ptr %3, align 473  %8 = fpext <8 x float> %4 to <8 x double>74  %9 = fpext <8 x float> %5 to <8 x double>75  %10 = fpext <8 x float> %6 to <8 x double>76  %11 = fpext <8 x float> %7 to <8 x double>77  %12 = fadd <8 x double> zeroinitializer, %878  %13 = fadd <8 x double> zeroinitializer, %979  %14 = fadd <8 x double> zeroinitializer, %1080  %15 = fadd <8 x double> zeroinitializer, %1181  %16 = fptrunc <8 x double> %12 to <8 x float>82  %17 = fptrunc <8 x double> %13 to <8 x float>83  %18 = fptrunc <8 x double> %14 to <8 x float>84  %19 = fptrunc <8 x double> %15 to <8 x float>85  %20 = fcmp ogt <8 x float> zeroinitializer, %1686  %21 = fcmp ogt <8 x float> zeroinitializer, %1787  %22 = fcmp ogt <8 x float> zeroinitializer, %1888  %23 = fcmp ogt <8 x float> zeroinitializer, %1989  ret void90}91 92define void @test3(float %0) {93; CHECK-LABEL: @test3(94; CHECK-NEXT:  entry:95; CHECK-NEXT:    br label [[FOR_BODY_LR_PH:%.*]]96; CHECK:       for.body.lr.ph:97; CHECK-NEXT:    br i1 false, label [[FOR_COND_CLEANUP:%.*]], label [[FOR_BODY:%.*]]98; CHECK:       for.cond.cleanup:99; CHECK-NEXT:    [[TMP1:%.*]] = phi <4 x float> [ zeroinitializer, [[FOR_BODY_LR_PH]] ], [ [[TMP7:%.*]], [[FOR_BODY]] ]100; CHECK-NEXT:    ret void101; CHECK:       for.body:102; CHECK-NEXT:    [[TMP4:%.*]] = load <2 x float>, ptr null, align 4103; CHECK-NEXT:    [[TMP5:%.*]] = fcmp olt <2 x float> zeroinitializer, [[TMP4]]104; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <2 x i1> [[TMP5]], <2 x i1> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>105; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <4 x i1> <i1 true, i1 true, i1 undef, i1 undef>, <4 x i1> [[TMP8]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>106; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>107; CHECK-NEXT:    [[TMP7]] = select <4 x i1> [[TMP9]], <4 x float> [[TMP6]], <4 x float> zeroinitializer108; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]109;110entry:111  br label %for.body.lr.ph112 113for.body.lr.ph:114  br i1 false, label %for.cond.cleanup, label %for.body115 116for.cond.cleanup:                                 ; preds = %for.body, %for.body.lr.ph117  %1 = phi <2 x float> [ zeroinitializer, %for.body.lr.ph ], [ %5, %for.body ]118  %2 = phi <2 x float> [ zeroinitializer, %for.body.lr.ph ], [ %6, %for.body ]119  ret void120 121for.body:122  %3 = load <2 x float>, ptr null, align 4123  %4 = fcmp olt <2 x float> zeroinitializer, %3124  %5 = select <2 x i1> <i1 true, i1 true>, <2 x float> %3, <2 x float> zeroinitializer125  %6 = select <2 x i1> %4, <2 x float> %3, <2 x float> zeroinitializer126  br label %for.cond.cleanup127}128 129define ptr @test4() {130; POWEROF2-LABEL: @test4(131; POWEROF2-NEXT:    [[TMP1:%.*]] = fadd <8 x float> zeroinitializer, zeroinitializer132; POWEROF2-NEXT:    [[TMP2:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 1, i32 2>133; POWEROF2-NEXT:    [[TMP3:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 5, i32 6>134; POWEROF2-NEXT:    [[TMP4:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <2 x i32> <i32 0, i32 4>135; POWEROF2-NEXT:    [[TMP5:%.*]] = shufflevector <2 x float> [[TMP2]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>136; POWEROF2-NEXT:    [[TMP16:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>137; POWEROF2-NEXT:    [[TMP6:%.*]] = shufflevector <4 x float> [[TMP5]], <4 x float> [[TMP16]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>138; POWEROF2-NEXT:    br label [[TMP8:%.*]]139; POWEROF2:       8:140; POWEROF2-NEXT:    br label [[TMP8]]141; POWEROF2:       9:142; POWEROF2-NEXT:    [[TMP9:%.*]] = phi <2 x float> [ poison, [[TMP7:%.*]] ], [ [[TMP4]], [[TMP0:%.*]] ]143; POWEROF2-NEXT:    [[TMP10:%.*]] = phi <4 x float> [ poison, [[TMP7]] ], [ [[TMP6]], [[TMP0]] ]144; POWEROF2-NEXT:    br label [[TMP11:%.*]]145; POWEROF2:       12:146; POWEROF2-NEXT:    [[TMP12:%.*]] = shufflevector <4 x float> [[TMP10]], <4 x float> poison, <2 x i32> <i32 0, i32 1>147; POWEROF2-NEXT:    [[TMP13:%.*]] = fmul <2 x float> [[TMP12]], zeroinitializer148; POWEROF2-NEXT:    [[TMP14:%.*]] = shufflevector <4 x float> [[TMP10]], <4 x float> poison, <2 x i32> <i32 2, i32 3>149; POWEROF2-NEXT:    [[TMP15:%.*]] = fmul <2 x float> zeroinitializer, [[TMP14]]150; POWEROF2-NEXT:    [[TMP30:%.*]] = extractelement <2 x float> [[TMP9]], i32 0151; POWEROF2-NEXT:    [[TMP17:%.*]] = fmul float 0.000000e+00, [[TMP30]]152; POWEROF2-NEXT:    [[TMP18:%.*]] = extractelement <2 x float> [[TMP9]], i32 1153; POWEROF2-NEXT:    [[TMP19:%.*]] = fmul float [[TMP18]], 0.000000e+00154; POWEROF2-NEXT:    [[TMP20:%.*]] = extractelement <2 x float> [[TMP13]], i32 0155; POWEROF2-NEXT:    [[TMP21:%.*]] = fadd reassoc nsz float [[TMP20]], [[TMP17]]156; POWEROF2-NEXT:    [[TMP22:%.*]] = extractelement <2 x float> [[TMP15]], i32 0157; POWEROF2-NEXT:    [[TMP23:%.*]] = fadd reassoc nsz float [[TMP22]], [[TMP19]]158; POWEROF2-NEXT:    [[TMP24:%.*]] = extractelement <2 x float> [[TMP13]], i32 1159; POWEROF2-NEXT:    [[TMP25:%.*]] = fadd reassoc nsz float [[TMP21]], [[TMP24]]160; POWEROF2-NEXT:    [[TMP26:%.*]] = extractelement <2 x float> [[TMP15]], i32 1161; POWEROF2-NEXT:    [[TMP27:%.*]] = fadd reassoc nsz float [[TMP23]], [[TMP26]]162; POWEROF2-NEXT:    [[TMP28:%.*]] = tail call float @llvm.sqrt.f32(float [[TMP25]])163; POWEROF2-NEXT:    [[TMP29:%.*]] = tail call float @llvm.sqrt.f32(float [[TMP27]])164; POWEROF2-NEXT:    ret ptr null165;166; NONPOWEROF2-LABEL: @test4(167; NONPOWEROF2-NEXT:    [[TMP1:%.*]] = fadd <8 x float> zeroinitializer, zeroinitializer168; NONPOWEROF2-NEXT:    [[TMP2:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>169; NONPOWEROF2-NEXT:    [[TMP3:%.*]] = shufflevector <8 x float> [[TMP1]], <8 x float> poison, <3 x i32> <i32 4, i32 5, i32 6>170; NONPOWEROF2-NEXT:    [[TMP4:%.*]] = shufflevector <3 x float> [[TMP2]], <3 x float> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>171; NONPOWEROF2-NEXT:    [[TMP18:%.*]] = shufflevector <3 x float> [[TMP3]], <3 x float> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>172; NONPOWEROF2-NEXT:    [[TMP5:%.*]] = shufflevector <6 x float> [[TMP4]], <6 x float> [[TMP18]], <6 x i32> <i32 0, i32 1, i32 2, i32 6, i32 7, i32 8>173; NONPOWEROF2-NEXT:    br label [[TMP7:%.*]]174; NONPOWEROF2:       7:175; NONPOWEROF2-NEXT:    br label [[TMP7]]176; NONPOWEROF2:       8:177; NONPOWEROF2-NEXT:    [[TMP8:%.*]] = phi <6 x float> [ poison, [[TMP6:%.*]] ], [ [[TMP5]], [[TMP0:%.*]] ]178; NONPOWEROF2-NEXT:    br label [[TMP9:%.*]]179; NONPOWEROF2:       10:180; NONPOWEROF2-NEXT:    [[TMP10:%.*]] = shufflevector <6 x float> [[TMP8]], <6 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>181; NONPOWEROF2-NEXT:    [[TMP11:%.*]] = fmul <3 x float> zeroinitializer, [[TMP10]]182; NONPOWEROF2-NEXT:    [[TMP12:%.*]] = shufflevector <6 x float> [[TMP8]], <6 x float> poison, <3 x i32> <i32 3, i32 4, i32 5>183; NONPOWEROF2-NEXT:    [[TMP13:%.*]] = fmul <3 x float> zeroinitializer, [[TMP12]]184; NONPOWEROF2-NEXT:    [[TMP14:%.*]] = call reassoc nsz float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP11]])185; NONPOWEROF2-NEXT:    [[TMP15:%.*]] = call reassoc nsz float @llvm.vector.reduce.fadd.v3f32(float 0.000000e+00, <3 x float> [[TMP13]])186; NONPOWEROF2-NEXT:    [[TMP16:%.*]] = tail call float @llvm.sqrt.f32(float [[TMP14]])187; NONPOWEROF2-NEXT:    [[TMP17:%.*]] = tail call float @llvm.sqrt.f32(float [[TMP15]])188; NONPOWEROF2-NEXT:    ret ptr null189;190  %1 = fadd <8 x float> zeroinitializer, zeroinitializer191  %2 = extractelement <8 x float> %1, i64 0192  %3 = extractelement <8 x float> %1, i64 1193  %4 = extractelement <8 x float> %1, i64 2194  %5 = extractelement <8 x float> %1, i64 4195  %6 = extractelement <8 x float> %1, i64 5196  %7 = extractelement <8 x float> %1, i64 6197  br label %9198 1998:200  br label %9201 2029:203  %10 = phi float [ 0.000000e+00, %8 ], [ %7, %0 ]204  %11 = phi float [ 0.000000e+00, %8 ], [ %6, %0 ]205  %12 = phi float [ 0.000000e+00, %8 ], [ %5, %0 ]206  %13 = phi float [ 0.000000e+00, %8 ], [ %4, %0 ]207  %14 = phi float [ 0.000000e+00, %8 ], [ %3, %0 ]208  %15 = phi float [ 0.000000e+00, %8 ], [ %2, %0 ]209  br label %16210 21116:212  %17 = fmul float %14, 0.000000e+00213  %18 = fmul float 0.000000e+00, %11214  %19 = fmul float 0.000000e+00, %15215  %20 = fmul float %12, 0.000000e+00216  %21 = fadd reassoc nsz float %17, %19217  %22 = fadd reassoc nsz float %18, %20218  %23 = fmul float %13, 0.000000e+00219  %24 = fmul float %10, 0.000000e+00220  %25 = fadd reassoc nsz float %21, %23221  %26 = fadd reassoc nsz float %22, %24222  %27 = tail call float @llvm.sqrt.f32(float %25)223  %28 = tail call float @llvm.sqrt.f32(float %26)224  ret ptr null225}226 227define i32 @test5() {228; CHECK-LABEL: @test5(229; CHECK-NEXT:  entry:230; CHECK-NEXT:    br label [[FOR_END47:%.*]]231; CHECK:       for.end47:232; CHECK-NEXT:    [[TMP0:%.*]] = phi <8 x double> [ <double 0x7FF8000000000000, double 0x7FF8000000000000, double 0.000000e+00, double 0.000000e+00, double 0x7FF8000000000000, double 0x7FF8000000000000, double 0.000000e+00, double 0.000000e+00>, [[ENTRY:%.*]] ]233; CHECK-NEXT:    ret i32 0234;235entry:236  %div0 = fdiv <2 x double> zeroinitializer, zeroinitializer237  %div1 = fdiv <2 x double> zeroinitializer, zeroinitializer238  %add0 = fadd <2 x double> zeroinitializer, %div0239  %add1 = fadd <2 x double> zeroinitializer, zeroinitializer240  %add2 = fadd <2 x double> %div1, zeroinitializer241  %add3 = fadd <2 x double> zeroinitializer, zeroinitializer242  br label %for.end47243 244for.end47:                                        ; preds = %entry245  %add0.lcssa = phi <2 x double> [ %add0, %entry ]246  %add1.lcssa = phi <2 x double> [ %add1, %entry ]247  %add2.lcssa = phi <2 x double> [ %add2, %entry ]248  %add3.lcssa = phi <2 x double> [ %add3, %entry ]249  ret i32 0250}251