1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -passes='loop-mssa(loop-predication),gvn,simplifycfg' -loop-predication-insert-assumes-of-predicated-guards-conditions=true < %s 2>&1 | FileCheck %s 3 4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2" 5target triple = "x86_64-unknown-linux-gnu" 6 7define i32 @test0(ptr %p1, ptr %p2, ptr %p3, ptr %p4, ptr %p5, i1 %c, i32 %x) { 8; CHECK-LABEL: @test0( 9; CHECK-NEXT: entry: 10; CHECK-NEXT: [[P1_1:%.*]] = getelementptr i32, ptr [[P1:%.*]], i64 1 11; CHECK-NEXT: [[P1_2:%.*]] = getelementptr i32, ptr [[P1]], i64 2 12; CHECK-NEXT: [[P1_3:%.*]] = getelementptr i32, ptr [[P1]], i64 3 13; CHECK-NEXT: [[IV_1_START:%.*]] = load i32, ptr [[P1_1]], align 4, !range [[RNG0:![0-9]+]] 14; CHECK-NEXT: [[IV_1_END:%.*]] = load i32, ptr [[P1_2]], align 4, !range [[RNG0]] 15; CHECK-NEXT: [[IV_2_END:%.*]] = load i32, ptr [[P1_3]], align 4, !range [[RNG0]] 16; CHECK-NEXT: [[LOOP_COND:%.*]] = icmp ult i32 [[IV_2_END]], [[IV_1_END]] 17; CHECK-NEXT: br i1 [[LOOP_COND]], label [[LOOP_PREHEADER:%.*]], label [[EXIT:%.*]] 18; CHECK: loop.preheader: 19; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[IV_1_END]], -1 20; CHECK-NEXT: [[TMP1:%.*]] = sub i32 [[TMP0]], [[IV_1_START]] 21; CHECK-NEXT: [[TMP2:%.*]] = icmp ule i32 [[IV_2_END]], [[TMP1]] 22; CHECK-NEXT: [[TMP3:%.*]] = icmp ult i32 [[IV_1_START]], [[IV_1_END]] 23; CHECK-NEXT: [[TMP4:%.*]] = and i1 [[TMP3]], [[TMP2]] 24; CHECK-NEXT: [[TMP5:%.*]] = freeze i1 [[TMP4]] 25; CHECK-NEXT: br label [[LOOP:%.*]] 26; CHECK: loop: 27; CHECK-NEXT: [[IV_1:%.*]] = phi i32 [ [[IV_1_NEXT:%.*]], [[LOOP_NEXT:%.*]] ], [ [[IV_1_START]], [[LOOP_PREHEADER]] ] 28; CHECK-NEXT: [[IV_2:%.*]] = phi i32 [ [[IV_2_NEXT:%.*]], [[LOOP_NEXT]] ], [ 0, [[LOOP_PREHEADER]] ] 29; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr i8, ptr [[P2:%.*]], i32 [[IV_1]] 30; CHECK-NEXT: [[VALUE:%.*]] = load i8, ptr [[GEP_1]], align 1 31; CHECK-NEXT: [[COND_1:%.*]] = icmp ult i32 [[IV_1]], [[IV_1_END]] 32; CHECK-NEXT: [[WC:%.*]] = call i1 @llvm.experimental.widenable.condition() 33; CHECK-NEXT: [[TMP6:%.*]] = and i1 [[TMP5]], [[WC]] 34; CHECK-NEXT: br i1 [[TMP6]], label [[LOOP_NEXT]], label [[DEOPT:%.*]] 35; CHECK: loop.next: 36; CHECK-NEXT: call void @llvm.assume(i1 [[COND_1]]) 37; CHECK-NEXT: [[GEP_3:%.*]] = getelementptr i8, ptr [[P4:%.*]], i32 [[IV_1]] 38; CHECK-NEXT: store i8 [[VALUE]], ptr [[GEP_3]], align 1 39; CHECK-NEXT: [[IV_1_NEXT]] = add nuw nsw i32 [[IV_1]], 1 40; CHECK-NEXT: [[IV_2_NEXT]] = add nuw nsw i32 [[IV_2]], 1 41; CHECK-NEXT: [[LATCH_COND:%.*]] = icmp ult i32 [[IV_2]], [[IV_2_END]] 42; CHECK-NEXT: br i1 [[LATCH_COND]], label [[LOOP]], label [[EXIT]] 43; CHECK: deopt: 44; CHECK-NEXT: [[DEOPT_RES:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ] 45; CHECK-NEXT: ret i32 [[DEOPT_RES]] 46; CHECK: exit: 47; CHECK-NEXT: [[RES:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[IV_2]], [[LOOP_NEXT]] ] 48; CHECK-NEXT: ret i32 [[RES]] 49; 50entry: 51 %p1.1 = getelementptr i32, ptr %p1, i64 1 52 %p1.2 = getelementptr i32, ptr %p1, i64 2 53 %p1.3 = getelementptr i32, ptr %p1, i64 3 54 %iv.1.start = load i32, ptr %p1.1, !range !0 55 %iv.1.end = load i32, ptr %p1.2, !range !0 56 %iv.2.end = load i32, ptr %p1.3, !range !0 57 %loop.cond = icmp ult i32 %iv.2.end, %iv.1.end 58 br i1 %loop.cond, label %loop, label %exit 59 60loop: 61 %iv.1 = phi i32 [ %iv.1.start, %entry ], [ %iv.1.next, %latch ] 62 %iv.2 = phi i32 [ 0, %entry ], [ %iv.2.next, %latch ] 63 %gep.1 = getelementptr i8, ptr %p2, i32 %iv.1 64 %value = load i8, ptr %gep.1 65 %cond.1 = icmp ult i32 %iv.1, %iv.1.end 66 %wc = call i1 @llvm.experimental.widenable.condition() 67 %explicit_guard_cond = and i1 %cond.1, %wc 68 br i1 %explicit_guard_cond, label %loop.next, label %deopt 69 70loop.next: 71 br i1 %cond.1, label %if.true, label %if.false 72 73if.true: 74 %gep.3 = getelementptr i8, ptr %p4, i32 %iv.1 75 store i8 %value, ptr %gep.3 76 br label %latch 77 78if.false: 79 %gep.4 = getelementptr i8, ptr %p4, i32 %iv.2 80 store i8 %value, ptr %gep.4 81 br label %latch 82 83latch: 84 %iv.1.next = add nuw nsw i32 %iv.1, 1 85 %iv.2.next = add nuw nsw i32 %iv.2, 1 86 %latch.cond = icmp ult i32 %iv.2, %iv.2.end 87 br i1 %latch.cond, label %loop, label %exit 88 89deopt: 90 %deopt_res = call i32 (...) @llvm.experimental.deoptimize.i32() [ "deopt"() ] 91 ret i32 %deopt_res 92 93exit: 94 %res = phi i32 [ 0, %entry ], [ %iv.2, %latch ] 95 ret i32 %res 96} 97 98; Function Attrs: inaccessiblememonly nocallback nofree nosync nounwind willreturn 99declare void @llvm.assume(i1) #0 100 101; Function Attrs: inaccessiblememonly nocallback nofree nosync nounwind speculatable willreturn 102declare i1 @llvm.experimental.widenable.condition() #1 103 104declare i32 @llvm.experimental.deoptimize.i32(...) 105 106attributes #0 = { inaccessiblememonly nocallback nofree nosync nounwind willreturn } 107attributes #1 = { inaccessiblememonly nocallback nofree nosync nounwind speculatable willreturn } 108 109!0 = !{i32 0, i32 2147483646} 110