1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; Test that the memset library call simplifier works correctly. 3; 4; RUN: opt < %s -passes=instcombine -S | FileCheck %s 5 6target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" 7 8declare ptr @memset(ptr, i32, i32) 9declare void @llvm.memset.p0.i32(ptr nocapture writeonly, i8, i32, i32, i1) 10declare noalias ptr @malloc(i32) #1 11 12; Check memset(mem1, val, size) -> llvm.memset(mem1, val, size, 1). 13 14define ptr @test_simplify1(ptr %mem, i32 %val, i32 %size) { 15; CHECK-LABEL: @test_simplify1( 16; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[VAL:%.*]] to i8 17; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 1 [[MEM:%.*]], i8 [[TMP1]], i32 [[SIZE:%.*]], i1 false) 18; CHECK-NEXT: ret ptr [[MEM]] 19; 20 %ret = call ptr @memset(ptr %mem, i32 %val, i32 %size) 21 ret ptr %ret 22} 23 24define ptr @test_simplify1_tail(ptr %mem, i32 %val, i32 %size) { 25; CHECK-LABEL: @test_simplify1_tail( 26; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[VAL:%.*]] to i8 27; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr align 1 [[MEM:%.*]], i8 [[TMP1]], i32 [[SIZE:%.*]], i1 false) 28; CHECK-NEXT: ret ptr [[MEM]] 29; 30 %ret = tail call ptr @memset(ptr %mem, i32 %val, i32 %size) 31 ret ptr %ret 32} 33 34define ptr @test_simplify1_musttail(ptr %mem, i32 %val, i32 %size) { 35; CHECK-LABEL: @test_simplify1_musttail( 36; CHECK-NEXT: [[RET:%.*]] = musttail call ptr @memset(ptr [[MEM:%.*]], i32 [[VAL:%.*]], i32 [[SIZE:%.*]]) 37; CHECK-NEXT: ret ptr [[RET]] 38; 39 %ret = musttail call ptr @memset(ptr %mem, i32 %val, i32 %size) 40 ret ptr %ret 41} 42 43; Malloc + memset pattern is now handled by DSE in a more general way. 44 45define ptr @pr25892_lite(i32 %size) #0 { 46; CHECK-LABEL: @pr25892_lite( 47; CHECK-NEXT: [[CALL1:%.*]] = call ptr @malloc(i32 [[SIZE:%.*]]) #[[ATTR0:[0-9]+]] 48; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 1 [[CALL1]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 49; CHECK-NEXT: ret ptr [[CALL1]] 50; 51 %call1 = call ptr @malloc(i32 %size) #1 52 %call2 = call ptr @memset(ptr %call1, i32 0, i32 %size) #1 53 ret ptr %call2 54} 55 56; A memset intrinsic should be handled similarly to a memset() libcall. 57; Notice that malloc + memset pattern is now handled by DSE in a more general way. 58 59define ptr @malloc_and_memset_intrinsic(i32 %n) #0 { 60; CHECK-LABEL: @malloc_and_memset_intrinsic( 61; CHECK-NEXT: [[CALL:%.*]] = call ptr @malloc(i32 [[N:%.*]]) 62; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 1 [[CALL]], i8 0, i32 [[N]], i1 false) 63; CHECK-NEXT: ret ptr [[CALL]] 64; 65 %call = call ptr @malloc(i32 %n) 66 call void @llvm.memset.p0.i32(ptr %call, i8 0, i32 %n, i32 1, i1 false) 67 ret ptr %call 68} 69 70; This should not create a calloc and should not crash the compiler. 71; Notice that malloc + memset pattern is now handled by DSE in a more general way. 72 73define ptr @notmalloc_memset(i32 %size, ptr %notmalloc) { 74; CHECK-LABEL: @notmalloc_memset( 75; CHECK-NEXT: [[CALL1:%.*]] = call ptr [[NOTMALLOC:%.*]](i32 [[SIZE:%.*]]) #[[ATTR0]] 76; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 1 [[CALL1]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 77; CHECK-NEXT: ret ptr [[CALL1]] 78; 79 %call1 = call ptr %notmalloc(i32 %size) #1 80 %call2 = call ptr @memset(ptr %call1, i32 0, i32 %size) #1 81 ret ptr %call2 82} 83 84; This doesn't fire currently because the malloc has more than one use. 85; Notice that malloc + memset pattern is now handled by DSE in a more general way. 86 87define ptr @pr25892(i32 %size) #0 { 88; CHECK-LABEL: @pr25892( 89; CHECK-NEXT: entry: 90; CHECK-NEXT: [[CALL:%.*]] = tail call ptr @malloc(i32 [[SIZE:%.*]]) #[[ATTR0]] 91; CHECK-NEXT: [[CMP:%.*]] = icmp eq ptr [[CALL]], null 92; CHECK-NEXT: br i1 [[CMP]], label [[CLEANUP:%.*]], label [[IF_END:%.*]] 93; CHECK: if.end: 94; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr nonnull align 1 [[CALL]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 95; CHECK-NEXT: br label [[CLEANUP]] 96; CHECK: cleanup: 97; CHECK-NEXT: [[RETVAL_0:%.*]] = phi ptr [ [[CALL]], [[IF_END]] ], [ null, [[ENTRY:%.*]] ] 98; CHECK-NEXT: ret ptr [[RETVAL_0]] 99; 100entry: 101 %call = tail call ptr @malloc(i32 %size) #1 102 %cmp = icmp eq ptr %call, null 103 br i1 %cmp, label %cleanup, label %if.end 104if.end: 105 %call2 = tail call ptr @memset(ptr nonnull %call, i32 0, i32 %size) #1 106 br label %cleanup 107cleanup: 108 %retval.0 = phi ptr [ %call, %if.end ], [ null, %entry ] 109 ret ptr %retval.0 110} 111 112; If there's a calloc transform, the store must also be eliminated. 113 114define ptr @buffer_is_modified_then_memset(i32 %size) { 115; CHECK-LABEL: @buffer_is_modified_then_memset( 116; CHECK-NEXT: [[PTR:%.*]] = tail call ptr @malloc(i32 [[SIZE:%.*]]) #[[ATTR0]] 117; CHECK-NEXT: store i8 1, ptr [[PTR]], align 1 118; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr nonnull align 1 [[PTR]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 119; CHECK-NEXT: ret ptr [[PTR]] 120; 121 %ptr = tail call ptr @malloc(i32 %size) #1 122 store i8 1, ptr %ptr ;; fdata[0] = 1; 123 %memset = tail call ptr @memset(ptr nonnull %ptr, i32 0, i32 %size) #1 124 ret ptr %memset 125} 126 127define ptr @memset_size_select(i1 %b, ptr %ptr) { 128; CHECK-LABEL: @memset_size_select( 129; CHECK-NEXT: [[SIZE:%.*]] = select i1 [[B:%.*]], i32 10, i32 50 130; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(10) [[PTR:%.*]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 131; CHECK-NEXT: ret ptr [[PTR]] 132; 133 %size = select i1 %b, i32 10, i32 50 134 %memset = tail call ptr @memset(ptr nonnull %ptr, i32 0, i32 %size) #1 135 ret ptr %memset 136} 137 138 139define ptr @memset_size_select2(i1 %b, ptr %ptr) { 140; CHECK-LABEL: @memset_size_select2( 141; CHECK-NEXT: [[SIZE:%.*]] = select i1 [[B:%.*]], i32 10, i32 50 142; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(80) [[PTR:%.*]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 143; CHECK-NEXT: ret ptr [[PTR]] 144; 145 %size = select i1 %b, i32 10, i32 50 146 %memset = tail call ptr @memset(ptr nonnull dereferenceable(80) %ptr, i32 0, i32 %size) #1 147 ret ptr %memset 148} 149 150define ptr @memset_size_select3(i1 %b, ptr %ptr) { 151; CHECK-LABEL: @memset_size_select3( 152; CHECK-NEXT: [[SIZE:%.*]] = select i1 [[B:%.*]], i32 10, i32 50 153; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(40) [[PTR:%.*]], i8 0, i32 [[SIZE]], i1 false) 154; CHECK-NEXT: ret ptr [[PTR]] 155; 156 %size = select i1 %b, i32 10, i32 50 157 %memset = tail call ptr @memset(ptr dereferenceable_or_null(40) %ptr, i32 0, i32 %size) 158 ret ptr %memset 159} 160 161define ptr @memset_size_select4(i1 %b, ptr %ptr) { 162; CHECK-LABEL: @memset_size_select4( 163; CHECK-NEXT: [[SIZE:%.*]] = select i1 [[B:%.*]], i32 10, i32 50 164; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(40) [[PTR:%.*]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 165; CHECK-NEXT: ret ptr [[PTR]] 166; 167 %size = select i1 %b, i32 10, i32 50 168 %memset = tail call ptr @memset(ptr nonnull dereferenceable_or_null(40) %ptr, i32 0, i32 %size) #1 169 ret ptr %memset 170} 171 172define ptr @memset_size_ashr(i1 %b, ptr %ptr, i32 %v) { 173; CHECK-LABEL: @memset_size_ashr( 174; CHECK-NEXT: [[SIZE:%.*]] = ashr i32 -2, [[V:%.*]] 175; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr noundef nonnull align 1 dereferenceable(1) [[PTR:%.*]], i8 0, i32 [[SIZE]], i1 false) #[[ATTR0]] 176; CHECK-NEXT: ret ptr [[PTR]] 177; 178 %size = ashr i32 -2, %v 179 %memset = tail call ptr @memset(ptr nonnull %ptr, i32 0, i32 %size) #1 180 ret ptr %memset 181} 182 183define ptr @memset_attrs1(i1 %b, ptr %ptr, i32 %size) { 184; CHECK-LABEL: @memset_attrs1( 185; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr align 1 dereferenceable_or_null(40) [[PTR:%.*]], i8 0, i32 [[SIZE:%.*]], i1 false) #[[ATTR0]] 186; CHECK-NEXT: ret ptr [[PTR]] 187; 188 %memset = tail call ptr @memset(ptr dereferenceable_or_null(40) %ptr, i32 0, i32 %size) #1 189 ret ptr %memset 190} 191 192; be sure to drop nonnull since size is unknown and can be 0 193; do not change dereferenceable attribute 194define ptr @memset_attrs2(i1 %b, ptr %ptr, i32 %size) { 195; CHECK-LABEL: @memset_attrs2( 196; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr nonnull align 1 dereferenceable(40) [[PTR:%.*]], i8 0, i32 [[SIZE:%.*]], i1 false) #[[ATTR0]] 197; CHECK-NEXT: ret ptr [[PTR]] 198; 199 %memset = tail call ptr @memset(ptr nonnull dereferenceable(40) %ptr, i32 0, i32 %size) #1 200 ret ptr %memset 201} 202 203; size is unknown, just copy attrs, no changes in attrs 204define ptr @memset_attrs3(i1 %b, ptr %ptr, i32 %size) { 205; CHECK-LABEL: @memset_attrs3( 206; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr nonnull align 1 dereferenceable_or_null(40) [[PTR:%.*]], i8 0, i32 [[SIZE:%.*]], i1 false) #[[ATTR0]] 207; CHECK-NEXT: ret ptr [[PTR]] 208; 209 %memset = tail call ptr @memset(ptr nonnull dereferenceable_or_null(40) %ptr, i32 0, i32 %size) #1 210 ret ptr %memset 211} 212 213; be sure to drop nonnull since size is unknown and can be 0 214define ptr @memset_attrs4(i1 %b, ptr %ptr, i32 %size) { 215; CHECK-LABEL: @memset_attrs4( 216; CHECK-NEXT: tail call void @llvm.memset.p0.i32(ptr nonnull align 1 [[PTR:%.*]], i8 0, i32 [[SIZE:%.*]], i1 false) #[[ATTR0]] 217; CHECK-NEXT: ret ptr [[PTR]] 218; 219 %memset = tail call ptr @memset(ptr nonnull %ptr, i32 0, i32 %size) #1 220 ret ptr %memset 221} 222 223define ptr @test_no_incompatible_attr(ptr %mem, i32 %val, i32 %size) { 224; CHECK-LABEL: @test_no_incompatible_attr( 225; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[VAL:%.*]] to i8 226; CHECK-NEXT: call void @llvm.memset.p0.i32(ptr align 1 [[MEM:%.*]], i8 [[TMP1]], i32 [[SIZE:%.*]], i1 false) 227; CHECK-NEXT: ret ptr [[MEM]] 228; 229 %ret = call dereferenceable(1) ptr @memset(ptr %mem, i32 %val, i32 %size) 230 ret ptr %ret 231} 232 233attributes #0 = { nounwind ssp uwtable } 234attributes #1 = { nounwind } 235attributes #2 = { nounwind readnone } 236 237