242 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes="default<O3>" -S %s | FileCheck %s3 4target triple = "arm64-apple-darwin"5 6; Make sure we can vectorize a loop that uses a function to clamp a double to7; be between a given minimum and maximum value.8 9define internal double @clamp(double %v) {10entry:11 %retval = alloca double, align 812 %v.addr = alloca double, align 813 store double %v, ptr %v.addr, align 814 %0 = load double, ptr %v.addr, align 815 %cmp = fcmp olt double %0, 0.000000e+0016 br i1 %cmp, label %if.then, label %if.end17 18if.then: ; preds = %entry19 store double 0.000000e+00, ptr %retval, align 820 br label %return21 22if.end: ; preds = %entry23 %1 = load double, ptr %v.addr, align 824 %cmp1 = fcmp ogt double %1, 6.000000e+0025 br i1 %cmp1, label %if.then2, label %if.end326 27if.then2: ; preds = %if.end28 store double 6.000000e+00, ptr %retval, align 829 br label %return30 31if.end3: ; preds = %if.end32 %2 = load double, ptr %v.addr, align 833 store double %2, ptr %retval, align 834 br label %return35 36return: ; preds = %if.end3, %if.then2, %if.then37 %3 = load double, ptr %retval, align 838 ret double %339}40 41define void @loop(ptr %X, ptr %Y) {42; CHECK-LABEL: @loop(43; CHECK-NEXT: entry:44; CHECK-NEXT: [[X6:%.*]] = ptrtoint ptr [[X:%.*]] to i6445; CHECK-NEXT: [[Y7:%.*]] = ptrtoint ptr [[Y:%.*]] to i6446; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[X6]], [[Y7]]47; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 3248; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_BODY:%.*]]49; CHECK: vector.body:50; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]51; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw double, ptr [[Y]], i64 [[INDEX]]52; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP1]], i64 1653; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP1]], align 854; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x double>, ptr [[TMP2]], align 855; CHECK-NEXT: [[TMP3:%.*]] = fcmp olt <2 x double> [[WIDE_LOAD]], zeroinitializer56; CHECK-NEXT: [[TMP4:%.*]] = fcmp olt <2 x double> [[WIDE_LOAD8]], zeroinitializer57; CHECK-NEXT: [[TMP5:%.*]] = fcmp ogt <2 x double> [[WIDE_LOAD]], splat (double 6.000000e+00)58; CHECK-NEXT: [[TMP6:%.*]] = fcmp ogt <2 x double> [[WIDE_LOAD8]], splat (double 6.000000e+00)59; CHECK-NEXT: [[TMP7:%.*]] = select <2 x i1> [[TMP5]], <2 x double> splat (double 6.000000e+00), <2 x double> [[WIDE_LOAD]]60; CHECK-NEXT: [[TMP8:%.*]] = select <2 x i1> [[TMP6]], <2 x double> splat (double 6.000000e+00), <2 x double> [[WIDE_LOAD8]]61; CHECK-NEXT: [[TMP9:%.*]] = select <2 x i1> [[TMP3]], <2 x double> zeroinitializer, <2 x double> [[TMP7]]62; CHECK-NEXT: [[TMP10:%.*]] = select <2 x i1> [[TMP4]], <2 x double> zeroinitializer, <2 x double> [[TMP8]]63; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw double, ptr [[X]], i64 [[INDEX]]64; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP11]], i64 1665; CHECK-NEXT: store <2 x double> [[TMP9]], ptr [[TMP11]], align 866; CHECK-NEXT: store <2 x double> [[TMP10]], ptr [[TMP12]], align 867; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 468; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 2000069; CHECK-NEXT: br i1 [[TMP13]], label [[FOR_COND_CLEANUP:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]70; CHECK: for.cond.cleanup:71; CHECK-NEXT: ret void72; CHECK: for.body:73; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY]] ]74; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw double, ptr [[Y]], i64 [[INDVARS_IV]]75; CHECK-NEXT: [[TMP14:%.*]] = load double, ptr [[ARRAYIDX]], align 876; CHECK-NEXT: [[CMP_I:%.*]] = fcmp olt double [[TMP14]], 0.000000e+0077; CHECK-NEXT: [[CMP1_I:%.*]] = fcmp ogt double [[TMP14]], 6.000000e+0078; CHECK-NEXT: [[DOTV_I:%.*]] = select i1 [[CMP1_I]], double 6.000000e+00, double [[TMP14]]79; CHECK-NEXT: [[RETVAL_0_I:%.*]] = select i1 [[CMP_I]], double 0.000000e+00, double [[DOTV_I]]80; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw double, ptr [[X]], i64 [[INDVARS_IV]]81; CHECK-NEXT: store double [[RETVAL_0_I]], ptr [[ARRAYIDX2]], align 882; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 183; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 2000084; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]85;86entry:87 %X.addr = alloca ptr, align 888 %Y.addr = alloca ptr, align 889 %i = alloca i32, align 490 store ptr %X, ptr %X.addr, align 891 store ptr %Y, ptr %Y.addr, align 892 call void @llvm.lifetime.start.p0(ptr %i) #293 store i32 0, ptr %i, align 494 br label %for.cond95 96for.cond: ; preds = %for.inc, %entry97 %0 = load i32, ptr %i, align 498 %cmp = icmp ult i32 %0, 2000099 br i1 %cmp, label %for.body, label %for.cond.cleanup100 101for.cond.cleanup: ; preds = %for.cond102 call void @llvm.lifetime.end.p0(ptr %i) #2103 br label %for.end104 105for.body: ; preds = %for.cond106 %1 = load ptr, ptr %Y.addr, align 8107 %2 = load i32, ptr %i, align 4108 %idxprom = zext i32 %2 to i64109 %arrayidx = getelementptr inbounds double, ptr %1, i64 %idxprom110 %3 = load double, ptr %arrayidx, align 8111 %call = call double @clamp(double %3)112 %4 = load ptr, ptr %X.addr, align 8113 %5 = load i32, ptr %i, align 4114 %idxprom1 = zext i32 %5 to i64115 %arrayidx2 = getelementptr inbounds double, ptr %4, i64 %idxprom1116 store double %call, ptr %arrayidx2, align 8117 br label %for.inc118 119for.inc: ; preds = %for.body120 %6 = load i32, ptr %i, align 4121 %inc = add i32 %6, 1122 store i32 %inc, ptr %i, align 4123 br label %for.cond124 125for.end: ; preds = %for.cond.cleanup126 ret void127}128 129; Test that requires sinking/hoisting of instructions for vectorization.130 131define void @loop2(ptr %A, ptr %B, ptr %C, float %x) {132; CHECK-LABEL: @loop2(133; CHECK-NEXT: entry:134; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 40000135; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[C:%.*]], i64 40000136; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 40000137; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[B]], [[SCEVGEP2]]138; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[C]], [[SCEVGEP]]139; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]140; CHECK-NEXT: [[BOUND04:%.*]] = icmp ult ptr [[B]], [[SCEVGEP3]]141; CHECK-NEXT: [[BOUND15:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]]142; CHECK-NEXT: [[FOUND_CONFLICT6:%.*]] = and i1 [[BOUND04]], [[BOUND15]]143; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT6]]144; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label [[LOOP_BODY:%.*]], label [[VECTOR_PH:%.*]]145; CHECK: vector.ph:146; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x float> poison, float [[X:%.*]], i64 0147; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x float> [[BROADCAST_SPLATINSERT]], <4 x float> poison, <4 x i32> zeroinitializer148; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]149; CHECK: vector.body:150; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]151; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds nuw i32, ptr [[C]], i64 [[INDEX]]152; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 16153; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4, !alias.scope [[META4:![0-9]+]]154; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4, !alias.scope [[META4]]155; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD]], splat (i32 20)156; CHECK-NEXT: [[TMP3:%.*]] = icmp eq <4 x i32> [[WIDE_LOAD7]], splat (i32 20)157; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDEX]]158; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP4]], i64 16159; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <4 x float>, ptr [[TMP4]], align 4, !alias.scope [[META7:![0-9]+]]160; CHECK-NEXT: [[WIDE_LOAD9:%.*]] = load <4 x float>, ptr [[TMP5]], align 4, !alias.scope [[META7]]161; CHECK-NEXT: [[TMP6:%.*]] = fmul <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD8]]162; CHECK-NEXT: [[TMP7:%.*]] = fmul <4 x float> [[BROADCAST_SPLAT]], [[WIDE_LOAD9]]163; CHECK-NEXT: [[TMP8:%.*]] = getelementptr float, ptr [[B]], i64 [[INDEX]]164; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[TMP8]], i64 16165; CHECK-NEXT: [[WIDE_LOAD10:%.*]] = load <4 x float>, ptr [[TMP8]], align 4, !alias.scope [[META9:![0-9]+]], !noalias [[META11:![0-9]+]]166; CHECK-NEXT: [[WIDE_LOAD11:%.*]] = load <4 x float>, ptr [[TMP9]], align 4, !alias.scope [[META9]], !noalias [[META11]]167; CHECK-NEXT: [[TMP10:%.*]] = fadd <4 x float> [[TMP6]], [[WIDE_LOAD10]]168; CHECK-NEXT: [[TMP11:%.*]] = fadd <4 x float> [[TMP7]], [[WIDE_LOAD11]]169; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP2]], <4 x float> [[TMP6]], <4 x float> [[TMP10]]170; CHECK-NEXT: [[PREDPHI12:%.*]] = select <4 x i1> [[TMP3]], <4 x float> [[TMP7]], <4 x float> [[TMP11]]171; CHECK-NEXT: store <4 x float> [[PREDPHI]], ptr [[TMP8]], align 4, !alias.scope [[META9]], !noalias [[META11]]172; CHECK-NEXT: store <4 x float> [[PREDPHI12]], ptr [[TMP9]], align 4, !alias.scope [[META9]], !noalias [[META11]]173; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8174; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000175; CHECK-NEXT: br i1 [[TMP13]], label [[EXIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]176; CHECK: loop.body:177; CHECK-NEXT: [[IV1:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ], [ 0, [[ENTRY:%.*]] ]178; CHECK-NEXT: [[C_GEP:%.*]] = getelementptr inbounds nuw i32, ptr [[C]], i64 [[IV1]]179; CHECK-NEXT: [[C_LV:%.*]] = load i32, ptr [[C_GEP]], align 4180; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[C_LV]], 20181; CHECK-NEXT: [[A_GEP_0:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[IV1]]182; CHECK-NEXT: [[A_LV_0:%.*]] = load float, ptr [[A_GEP_0]], align 4183; CHECK-NEXT: [[MUL2_I81_I:%.*]] = fmul float [[X]], [[A_LV_0]]184; CHECK-NEXT: [[B_GEP_0:%.*]] = getelementptr inbounds nuw float, ptr [[B]], i64 [[IV1]]185; CHECK-NEXT: br i1 [[CMP]], label [[LOOP_LATCH]], label [[ELSE:%.*]]186; CHECK: else:187; CHECK-NEXT: [[B_LV:%.*]] = load float, ptr [[B_GEP_0]], align 4188; CHECK-NEXT: [[ADD:%.*]] = fadd float [[MUL2_I81_I]], [[B_LV]]189; CHECK-NEXT: br label [[LOOP_LATCH]]190; CHECK: loop.latch:191; CHECK-NEXT: [[ADD_SINK:%.*]] = phi float [ [[ADD]], [[ELSE]] ], [ [[MUL2_I81_I]], [[LOOP_BODY]] ]192; CHECK-NEXT: store float [[ADD_SINK]], ptr [[B_GEP_0]], align 4193; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV1]], 1194; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 10000195; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[LOOP_BODY]], !llvm.loop [[LOOP13:![0-9]+]]196; CHECK: exit:197; CHECK-NEXT: ret void198;199entry:200 br label %loop.header201 202loop.header:203 %iv = phi i64 [ %iv.next, %loop.latch ], [ 0, %entry ]204 %cmp.0 = icmp ult i64 %iv, 10000205 br i1 %cmp.0, label %loop.body, label %exit206 207loop.body:208 %C.gep = getelementptr inbounds i32, ptr %C, i64 %iv209 %C.lv = load i32, ptr %C.gep210 %cmp = icmp eq i32 %C.lv, 20211 br i1 %cmp, label %then, label %else212 213then:214 %A.gep.0 = getelementptr inbounds float, ptr %A, i64 %iv215 %A.lv.0 = load float, ptr %A.gep.0, align 4216 %mul2.i81.i = fmul float %A.lv.0, %x217 %B.gep.0 = getelementptr inbounds float, ptr %B, i64 %iv218 store float %mul2.i81.i, ptr %B.gep.0, align 4219 br label %loop.latch220 221else:222 %A.gep.1 = getelementptr inbounds float, ptr %A, i64 %iv223 %A.lv.1 = load float, ptr %A.gep.1, align 4224 %mul2 = fmul float %A.lv.1, %x225 %B.gep.1 = getelementptr inbounds float, ptr %B, i64 %iv226 %B.lv = load float, ptr %B.gep.1, align 4227 %add = fadd float %mul2, %B.lv228 store float %add, ptr %B.gep.1, align 4229 br label %loop.latch230 231loop.latch:232 %iv.next = add nuw nsw i64 %iv, 1233 br label %loop.header234 235exit:236 ret void237}238 239declare void @llvm.lifetime.start.p0(ptr nocapture)240 241declare void @llvm.lifetime.end.p0(ptr nocapture)242