132 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=loop-vectorize --force-vector-width=4 --force-vector-interleave=0 -S -o - < %s | FileCheck %s3 4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"5target triple = "x86_64-unknown-linux-gnu"6 7%0 = type { i32 }8%1 = type { i64 }9 10define ptr @foo(ptr %p, ptr %p.last) unnamed_addr #0 {11; CHECK-LABEL: @foo(12; CHECK-NEXT: entry:13; CHECK-NEXT: [[P3:%.*]] = ptrtoint ptr [[P:%.*]] to i6414; CHECK-NEXT: [[P_LAST1:%.*]] = ptrtoint ptr [[P_LAST:%.*]] to i6415; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[P_LAST1]], -102416; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[P3]]17; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 1018; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 119; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 1620; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]21; CHECK: vector.ph:22; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 1623; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]24; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[N_VEC]], 102425; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP4]]26; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]27; CHECK: vector.body:28; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]29; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]30; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1024, i64 2048, i64 3072>31; CHECK-NEXT: [[STEP_ADD:%.*]] = getelementptr i8, <4 x ptr> [[TMP5]], <4 x i64> splat (i64 4096)32; CHECK-NEXT: [[STEP_ADD_2:%.*]] = getelementptr i8, <4 x ptr> [[STEP_ADD]], <4 x i64> splat (i64 4096)33; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, <4 x ptr> [[STEP_ADD_2]], <4 x i64> splat (i64 4096)34; CHECK-NEXT: [[WIDE_MASKED_GATHER6:%.*]] = call <4 x ptr> @llvm.masked.gather.v4p0.v4p0(<4 x ptr> align 8 [[TMP8]], <4 x i1> splat (i1 true), <4 x ptr> poison)35; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1636; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 1638437; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]38; CHECK-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]39; CHECK: middle.block:40; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x ptr> [[WIDE_MASKED_GATHER6]], i32 341; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]42; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]43; CHECK: scalar.ph:44; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ]45; CHECK-NEXT: br label [[LOOP:%.*]]46; CHECK: loop:47; CHECK-NEXT: [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P_INC:%.*]], [[LOOP]] ]48; CHECK-NEXT: [[P_INC]] = getelementptr inbounds i64, ptr [[P2]], i64 12849; CHECK-NEXT: [[V:%.*]] = load ptr, ptr [[P2]], align 850; CHECK-NEXT: [[B:%.*]] = icmp eq ptr [[P_INC]], [[P_LAST]]51; CHECK-NEXT: br i1 [[B]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]52; CHECK: exit:53; CHECK-NEXT: [[V_LCSSA:%.*]] = phi ptr [ [[V]], [[LOOP]] ], [ [[TMP10]], [[MIDDLE_BLOCK]] ]54; CHECK-NEXT: ret ptr [[V_LCSSA]]55;56entry:57 br label %loop58 59loop:60 %p2 = phi ptr [ %p, %entry ], [ %p.inc, %loop ]61 %p.inc = getelementptr inbounds i64, ptr %p2, i64 12862 %v = load ptr, ptr %p2, align 863 %b = icmp eq ptr %p.inc, %p.last64 br i1 %b, label %exit, label %loop65 66exit:67 ret ptr %v68}69 70define ptr @bar(ptr %p, ptr %p.last) unnamed_addr #0 {71; CHECK-LABEL: @bar(72; CHECK-NEXT: entry:73; CHECK-NEXT: [[P3:%.*]] = ptrtoint ptr [[P:%.*]] to i6474; CHECK-NEXT: [[P_LAST1:%.*]] = ptrtoint ptr [[P_LAST:%.*]] to i6475; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[P_LAST1]], -102476; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[P3]]77; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 1078; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 179; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 1680; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]81; CHECK: vector.ph:82; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 1683; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]84; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[N_VEC]], 102485; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP4]]86; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]87; CHECK: vector.body:88; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]89; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]90; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1024, i64 2048, i64 3072>91; CHECK-NEXT: [[STEP_ADD:%.*]] = getelementptr i8, <4 x ptr> [[TMP5]], <4 x i64> splat (i64 4096)92; CHECK-NEXT: [[STEP_ADD_2:%.*]] = getelementptr i8, <4 x ptr> [[STEP_ADD]], <4 x i64> splat (i64 4096)93; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, <4 x ptr> [[STEP_ADD_2]], <4 x i64> splat (i64 4096)94; CHECK-NEXT: [[WIDE_MASKED_GATHER6:%.*]] = call <4 x ptr> @llvm.masked.gather.v4p0.v4p0(<4 x ptr> align 8 [[TMP8]], <4 x i1> splat (i1 true), <4 x ptr> poison)95; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1696; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 1638497; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]98; CHECK-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]99; CHECK: middle.block:100; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x ptr> [[WIDE_MASKED_GATHER6]], i32 3101; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]102; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]103; CHECK: scalar.ph:104; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ]105; CHECK-NEXT: br label [[LOOP:%.*]]106; CHECK: loop:107; CHECK-NEXT: [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P_INC:%.*]], [[LOOP]] ]108; CHECK-NEXT: [[P_INC]] = getelementptr inbounds i64, ptr [[P2]], i64 128109; CHECK-NEXT: [[V:%.*]] = load ptr, ptr [[P2]], align 8110; CHECK-NEXT: [[B:%.*]] = icmp eq ptr [[P_INC]], [[P_LAST]]111; CHECK-NEXT: br i1 [[B]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]112; CHECK: exit:113; CHECK-NEXT: [[V_LCSSA:%.*]] = phi ptr [ [[V]], [[LOOP]] ], [ [[TMP10]], [[MIDDLE_BLOCK]] ]114; CHECK-NEXT: ret ptr [[V_LCSSA]]115;116entry:117 br label %loop118 119loop:120 %p2 = phi ptr [ %p, %entry ], [ %p.inc, %loop ]121 %p.inc = getelementptr inbounds i64, ptr %p2, i64 128122 %v = load ptr, ptr %p2, align 8123 %b = icmp eq ptr %p.inc, %p.last124 br i1 %b, label %exit, label %loop125 126exit:127 ret ptr %v128}129 130attributes #0 = { "target-cpu"="skylake" }131 132