1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; MemCpy optimizations should take place even in presence of invariant.start 3; RUN: opt < %s -basic-aa -memcpyopt -S -enable-memcpyopt-memoryssa=0 | FileCheck %s --check-prefixes=CHECK,NO_MSSA 4; RUN: opt < %s -basic-aa -memcpyopt -S -enable-memcpyopt-memoryssa=1 -verify-memoryssa | FileCheck %s --check-prefixes=CHECK,MSSA 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 8target triple = "i686-apple-darwin9" 9 10%0 = type { x86_fp80, x86_fp80 } 11declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i1) nounwind 12declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i1) 13declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) 14 15declare {}* @llvm.invariant.start.p0i8(i64, i8* nocapture) nounwind readonly 16 17; FIXME: The invariant.start does not modify %P. 18; The intermediate alloca and one of the memcpy's should be eliminated, the 19; other should be transformed to a memmove. 20define void @test1(i8* %P, i8* %Q) nounwind { 21; NO_MSSA-LABEL: @test1( 22; NO_MSSA-NEXT: [[MEMTMP:%.*]] = alloca [[TMP0:%.*]], align 16 23; NO_MSSA-NEXT: [[R:%.*]] = bitcast %0* [[MEMTMP]] to i8* 24; NO_MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[R]], i8* align 16 [[P:%.*]], i32 32, i1 false) 25; NO_MSSA-NEXT: [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 32, i8* [[P]]) 26; NO_MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[R]], i32 32, i1 false) 27; NO_MSSA-NEXT: ret void 28; 29; MSSA-LABEL: @test1( 30; MSSA-NEXT: [[MEMTMP:%.*]] = alloca [[TMP0:%.*]], align 16 31; MSSA-NEXT: [[R:%.*]] = bitcast %0* [[MEMTMP]] to i8* 32; MSSA-NEXT: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 [[R]], i8* align 16 [[P:%.*]], i32 32, i1 false) 33; MSSA-NEXT: [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 32, i8* [[P]]) 34; MSSA-NEXT: call void @llvm.memmove.p0i8.p0i8.i32(i8* align 16 [[Q:%.*]], i8* align 16 [[P]], i32 32, i1 false) 35; MSSA-NEXT: ret void 36; 37 %memtmp = alloca %0, align 16 38 %R = bitcast %0* %memtmp to i8* 39 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %R, i8* align 16 %P, i32 32, i1 false) 40 %i = call {}* @llvm.invariant.start.p0i8(i64 32, i8* %P) 41 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %Q, i8* align 16 %R, i32 32, i1 false) 42 ret void 43} 44 45 46; The invariant.start intrinsic does not inhibit tranforming the memcpy to a 47; memset. 48define void @test2(i8* %dst1, i8* %dst2, i8 %c) { 49; CHECK-LABEL: @test2( 50; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* [[DST1:%.*]], i8 [[C:%.*]], i64 128, i1 false) 51; CHECK-NEXT: [[I:%.*]] = call {}* @llvm.invariant.start.p0i8(i64 32, i8* [[DST1]]) 52; CHECK-NEXT: call void @llvm.memset.p0i8.i64(i8* align 8 [[DST2:%.*]], i8 [[C]], i64 128, i1 false) 53; CHECK-NEXT: ret void 54; 55 call void @llvm.memset.p0i8.i64(i8* %dst1, i8 %c, i64 128, i1 false) 56 %i = call {}* @llvm.invariant.start.p0i8(i64 32, i8* %dst1) 57 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %dst2, i8* align 8 %dst1, i64 128, i1 false) 58 ret void 59} 60