1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -S -basic-aa -memcpyopt -enable-memcpyopt-memoryssa=0 | FileCheck %s 3; RUN: opt < %s -S -basic-aa -memcpyopt -enable-memcpyopt-memoryssa=1 -verify-memoryssa | FileCheck %s 4target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64" 5 6declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1) nounwind 7declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) nounwind 8 9; The resulting memset is only 4-byte aligned, despite containing 10; a 16-byte aligned store in the middle. 11 12define void @foo(i32* %p) { 13; CHECK-LABEL: @foo( 14; CHECK-NEXT: [[A0:%.*]] = getelementptr i32, i32* [[P:%.*]], i64 0 15; CHECK-NEXT: [[A1:%.*]] = getelementptr i32, i32* [[P]], i64 1 16; CHECK-NEXT: [[A2:%.*]] = getelementptr i32, i32* [[P]], i64 2 17; CHECK-NEXT: [[A3:%.*]] = getelementptr i32, i32* [[P]], i64 3 18; CHECK-NEXT: [[TMP1:%.*]] = bitcast i32* [[A0]] to i8* 19; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 4 [[TMP1]], i8 0, i64 16, i1 false) 20; CHECK-NEXT: ret void 21; 22 %a0 = getelementptr i32, i32* %p, i64 0 23 store i32 0, i32* %a0, align 4 24 %a1 = getelementptr i32, i32* %p, i64 1 25 store i32 0, i32* %a1, align 16 26 %a2 = getelementptr i32, i32* %p, i64 2 27 store i32 0, i32* %a2, align 4 28 %a3 = getelementptr i32, i32* %p, i64 3 29 store i32 0, i32* %a3, align 4 30 ret void 31} 32 33; Replacing %a8 with %a4 in the memset requires boosting the alignment of %a4. 34 35define void @bar() { 36; CHECK-LABEL: @bar( 37; CHECK-NEXT: [[A4:%.*]] = alloca i32, align 8 38; CHECK-NEXT: [[A8:%.*]] = alloca i32, align 8 39; CHECK-NEXT: [[A8_CAST:%.*]] = bitcast i32* [[A8]] to i8* 40; CHECK-NEXT: [[A4_CAST:%.*]] = bitcast i32* [[A4]] to i8* 41; CHECK-NEXT: [[A41:%.*]] = bitcast i32* [[A4]] to i8* 42; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 8 [[A41]], i8 0, i64 4, i1 false) 43; CHECK-NEXT: ret void 44; 45 %a4 = alloca i32, align 4 46 %a8 = alloca i32, align 8 47 %a8.cast = bitcast i32* %a8 to i8* 48 %a4.cast = bitcast i32* %a4 to i8* 49 call void @llvm.memset.p0i8.i64(i8* align 8 %a8.cast, i8 0, i64 4, i1 false) 50 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %a4.cast, i8* align 4 %a8.cast, i64 4, i1 false) 51 ret void 52} 53