145 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -mtriple=aarch64 -mcpu=neoverse-v2 -p loop-vectorize %s -S | FileCheck %s3 4define double @fp128_fmuladd_reduction(ptr %start0, ptr %start1, ptr %end0, ptr %end1, double %x, i64 %n) {5; CHECK-LABEL: define double @fp128_fmuladd_reduction(6; CHECK-SAME: ptr [[START0:%.*]], ptr [[START1:%.*]], ptr [[END0:%.*]], ptr [[END1:%.*]], double [[X:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {7; CHECK-NEXT: [[ITER_CHECK:.*]]:8; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 29; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]10; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:11; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], 812; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]13; CHECK: [[VECTOR_PH]]:14; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 815; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]16; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]17; CHECK: [[VECTOR_BODY]]:18; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]19; CHECK-NEXT: [[VEC_PHI:%.*]] = phi double [ [[X]], %[[VECTOR_PH]] ], [ [[TMP21:%.*]], %[[VECTOR_BODY]] ]20; CHECK-NEXT: [[TMP0:%.*]] = mul i64 [[INDEX]], 1621; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[START0]], i64 [[TMP0]]22; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[INDEX]], 823; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[START1]], i64 [[TMP2]]24; CHECK-NEXT: [[TMP24:%.*]] = getelementptr fp128, ptr [[TMP1]], i64 225; CHECK-NEXT: [[TMP4:%.*]] = getelementptr fp128, ptr [[TMP1]], i64 426; CHECK-NEXT: [[TMP5:%.*]] = getelementptr fp128, ptr [[TMP1]], i64 627; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x fp128>, ptr [[TMP1]], align 1628; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <2 x fp128>, ptr [[TMP24]], align 1629; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x fp128>, ptr [[TMP4]], align 1630; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <2 x fp128>, ptr [[TMP5]], align 1631; CHECK-NEXT: [[TMP28:%.*]] = getelementptr double, ptr [[TMP3]], i64 232; CHECK-NEXT: [[TMP35:%.*]] = getelementptr double, ptr [[TMP3]], i64 433; CHECK-NEXT: [[TMP36:%.*]] = getelementptr double, ptr [[TMP3]], i64 634; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <2 x double>, ptr [[TMP3]], align 1635; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x double>, ptr [[TMP28]], align 1636; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x double>, ptr [[TMP35]], align 1637; CHECK-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x double>, ptr [[TMP36]], align 1638; CHECK-NEXT: [[TMP10:%.*]] = fptrunc <2 x fp128> [[WIDE_LOAD]] to <2 x double>39; CHECK-NEXT: [[TMP11:%.*]] = fptrunc <2 x fp128> [[WIDE_LOAD3]] to <2 x double>40; CHECK-NEXT: [[TMP12:%.*]] = fptrunc <2 x fp128> [[WIDE_LOAD4]] to <2 x double>41; CHECK-NEXT: [[TMP13:%.*]] = fptrunc <2 x fp128> [[WIDE_LOAD5]] to <2 x double>42; CHECK-NEXT: [[TMP14:%.*]] = fmul <2 x double> [[TMP10]], [[WIDE_LOAD6]]43; CHECK-NEXT: [[TMP15:%.*]] = fmul <2 x double> [[TMP11]], [[WIDE_LOAD7]]44; CHECK-NEXT: [[TMP16:%.*]] = fmul <2 x double> [[TMP12]], [[WIDE_LOAD8]]45; CHECK-NEXT: [[TMP17:%.*]] = fmul <2 x double> [[TMP13]], [[WIDE_LOAD9]]46; CHECK-NEXT: [[TMP18:%.*]] = call double @llvm.vector.reduce.fadd.v2f64(double [[VEC_PHI]], <2 x double> [[TMP14]])47; CHECK-NEXT: [[TMP19:%.*]] = call double @llvm.vector.reduce.fadd.v2f64(double [[TMP18]], <2 x double> [[TMP15]])48; CHECK-NEXT: [[TMP20:%.*]] = call double @llvm.vector.reduce.fadd.v2f64(double [[TMP19]], <2 x double> [[TMP16]])49; CHECK-NEXT: [[TMP21]] = call double @llvm.vector.reduce.fadd.v2f64(double [[TMP20]], <2 x double> [[TMP17]])50; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 851; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]52; CHECK-NEXT: br i1 [[TMP22]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]53; CHECK: [[MIDDLE_BLOCK]]:54; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]55; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]56; CHECK: [[VEC_EPILOG_ITER_CHECK]]:57; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[N_VEC]], 1658; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[START0]], i64 [[TMP6]]59; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[N_VEC]], 860; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[START1]], i64 [[TMP7]]61; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 262; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]63; CHECK: [[VEC_EPILOG_PH]]:64; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]65; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi double [ [[TMP21]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[X]], %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]66; CHECK-NEXT: [[N_MOD_VF10:%.*]] = urem i64 [[N]], 267; CHECK-NEXT: [[N_VEC11:%.*]] = sub i64 [[N]], [[N_MOD_VF10]]68; CHECK-NEXT: [[TMP25:%.*]] = mul i64 [[N_VEC11]], 1669; CHECK-NEXT: [[TMP26:%.*]] = getelementptr i8, ptr [[START0]], i64 [[TMP25]]70; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[N_VEC11]], 871; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[START1]], i64 [[TMP8]]72; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]73; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:74; CHECK-NEXT: [[INDEX12:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT19:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]75; CHECK-NEXT: [[VEC_PHI13:%.*]] = phi double [ [[BC_MERGE_RDX]], %[[VEC_EPILOG_PH]] ], [ [[TMP33:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]76; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX12]], 1677; CHECK-NEXT: [[NEXT_GEP14:%.*]] = getelementptr i8, ptr [[START0]], i64 [[OFFSET_IDX]]78; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[INDEX12]], 879; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[START1]], i64 [[TMP9]]80; CHECK-NEXT: [[WIDE_LOAD17:%.*]] = load <2 x fp128>, ptr [[NEXT_GEP14]], align 1681; CHECK-NEXT: [[WIDE_LOAD18:%.*]] = load <2 x double>, ptr [[NEXT_GEP8]], align 1682; CHECK-NEXT: [[TMP31:%.*]] = fptrunc <2 x fp128> [[WIDE_LOAD17]] to <2 x double>83; CHECK-NEXT: [[TMP32:%.*]] = fmul <2 x double> [[TMP31]], [[WIDE_LOAD18]]84; CHECK-NEXT: [[TMP33]] = call double @llvm.vector.reduce.fadd.v2f64(double [[VEC_PHI13]], <2 x double> [[TMP32]])85; CHECK-NEXT: [[INDEX_NEXT19]] = add nuw i64 [[INDEX12]], 286; CHECK-NEXT: [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT19]], [[N_VEC11]]87; CHECK-NEXT: br i1 [[TMP34]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]88; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:89; CHECK-NEXT: [[CMP_N20:%.*]] = icmp eq i64 [[N]], [[N_VEC11]]90; CHECK-NEXT: br i1 [[CMP_N20]], label %[[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]]91; CHECK: [[VEC_EPILOG_SCALAR_PH]]:92; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[TMP26]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[NEXT_GEP3]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[START0]], %[[ITER_CHECK]] ]93; CHECK-NEXT: [[BC_RESUME_VAL21:%.*]] = phi ptr [ [[NEXT_GEP7]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[NEXT_GEP6]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[START1]], %[[ITER_CHECK]] ]94; CHECK-NEXT: [[BC_RESUME_VAL22:%.*]] = phi i64 [ [[N_VEC11]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]95; CHECK-NEXT: [[BC_MERGE_RDX23:%.*]] = phi double [ [[TMP33]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[TMP21]], %[[VEC_EPILOG_ITER_CHECK]] ], [ [[X]], %[[ITER_CHECK]] ]96; CHECK-NEXT: br label %[[LOOP:.*]]97; CHECK: [[LOOP]]:98; CHECK-NEXT: [[PTR0:%.*]] = phi ptr [ [[PTR0_NEXT:%.*]], %[[LOOP]] ], [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ]99; CHECK-NEXT: [[PTR1:%.*]] = phi ptr [ [[PTR1_NEXT:%.*]], %[[LOOP]] ], [ [[BC_RESUME_VAL21]], %[[VEC_EPILOG_SCALAR_PH]] ]100; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ [[BC_RESUME_VAL22]], %[[VEC_EPILOG_SCALAR_PH]] ]101; CHECK-NEXT: [[RED:%.*]] = phi double [ [[RED_NEXT:%.*]], %[[LOOP]] ], [ [[BC_MERGE_RDX23]], %[[VEC_EPILOG_SCALAR_PH]] ]102; CHECK-NEXT: [[PTR0_NEXT]] = getelementptr i8, ptr [[PTR0]], i64 16103; CHECK-NEXT: [[PTR1_NEXT]] = getelementptr i8, ptr [[PTR1]], i64 8104; CHECK-NEXT: [[LOAD0:%.*]] = load fp128, ptr [[PTR0]], align 16105; CHECK-NEXT: [[LOAD1:%.*]] = load double, ptr [[PTR1]], align 16106; CHECK-NEXT: [[TRUNC:%.*]] = fptrunc fp128 [[LOAD0]] to double107; CHECK-NEXT: [[RED_NEXT]] = tail call double @llvm.fmuladd.f64(double [[TRUNC]], double [[LOAD1]], double [[RED]])108; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1109; CHECK-NEXT: [[CMP1_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]110; CHECK-NEXT: br i1 [[CMP1_NOT]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]111; CHECK: [[EXIT]]:112; CHECK-NEXT: [[LCSSA:%.*]] = phi double [ [[RED_NEXT]], %[[LOOP]] ], [ [[TMP21]], %[[MIDDLE_BLOCK]] ], [ [[TMP33]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ]113; CHECK-NEXT: ret double [[LCSSA]]114;115entry:116 br label %loop117 118loop:119 %ptr0 = phi ptr [ %ptr0.next, %loop ], [ %start0, %entry ]120 %ptr1 = phi ptr [ %ptr1.next, %loop ], [ %start1, %entry ]121 %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]122 %red = phi double [ %red.next, %loop ], [ %x, %entry ]123 %ptr0.next = getelementptr i8, ptr %ptr0, i64 16124 %ptr1.next = getelementptr i8, ptr %ptr1, i64 8125 %load0 = load fp128, ptr %ptr0, align 16126 %load1 = load double, ptr %ptr1, align 16127 %trunc = fptrunc fp128 %load0 to double128 %red.next = tail call double @llvm.fmuladd.f64(double %trunc, double %load1, double %red)129 %iv.next = add i64 %iv, 1130 %cmp1.not = icmp eq i64 %iv.next, %n131 br i1 %cmp1.not, label %exit, label %loop132 133exit:134 %lcssa = phi double [ %red.next, %loop ]135 ret double %lcssa136}137;.138; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}139; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}140; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}141; CHECK: [[PROF3]] = !{!"branch_weights", i32 2, i32 6}142; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}143; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}144;.145