| Dan Gohman | c1be92f | 2010-10-18 18:04:47 +0000 | [diff] [blame] | 1 | ; RUN: opt < %s -basicaa -memcpyopt -S | FileCheck %s |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 2 | ; These memmoves should get optimized to memcpys. |
| 3 | |
| 4 | target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128" |
| 5 | target triple = "x86_64-apple-darwin9.0" |
| 6 | |
| Chris Lattner | b85e4eb | 2011-06-18 06:05:24 +0000 | [diff] [blame] | 7 | declare void @llvm.memmove.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 8 | |
| 9 | define i8* @test1(i8* nocapture %src) nounwind { |
| 10 | entry: |
| 11 | ; CHECK: @test1 |
| 12 | ; CHECK: call void @llvm.memcpy |
| 13 | |
| Chris Lattner | 26b0000 | 2011-06-17 03:14:27 +0000 | [diff] [blame] | 14 | %malloccall = tail call i8* @malloc(i32 trunc (i64 mul nuw (i64 ptrtoint (i8* getelementptr (i8* null, i32 1) to i64), i64 13) to i32)) |
| 15 | %call3 = bitcast i8* %malloccall to [13 x i8]* |
| Chris Lattner | b85e4eb | 2011-06-18 06:05:24 +0000 | [diff] [blame] | 16 | %call3.sub = getelementptr inbounds [13 x i8]* %call3, i64 0, i64 0 |
| 17 | tail call void @llvm.memmove.p0i8.p0i8.i64(i8* %call3.sub, i8* %src, i64 13, i32 1, i1 false) |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 18 | ret i8* %call3.sub |
| 19 | } |
| Chris Lattner | 26b0000 | 2011-06-17 03:14:27 +0000 | [diff] [blame] | 20 | declare noalias i8* @malloc(i32) |
| 21 | |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 22 | |
| 23 | define void @test2(i8* %P) nounwind { |
| 24 | entry: |
| 25 | ; CHECK: @test2 |
| 26 | ; CHECK: call void @llvm.memcpy |
| Chris Lattner | b85e4eb | 2011-06-18 06:05:24 +0000 | [diff] [blame] | 27 | %add.ptr = getelementptr i8* %P, i64 16 |
| 28 | tail call void @llvm.memmove.p0i8.p0i8.i64(i8* %P, i8* %add.ptr, i64 16, i32 1, i1 false) |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 29 | ret void |
| 30 | } |
| 31 | |
| 32 | ; This cannot be optimize because the src/dst really do overlap. |
| 33 | define void @test3(i8* %P) nounwind { |
| 34 | entry: |
| 35 | ; CHECK: @test3 |
| 36 | ; CHECK: call void @llvm.memmove |
| Chris Lattner | b85e4eb | 2011-06-18 06:05:24 +0000 | [diff] [blame] | 37 | %add.ptr = getelementptr i8* %P, i64 16 |
| 38 | tail call void @llvm.memmove.p0i8.p0i8.i64(i8* %P, i8* %add.ptr, i64 17, i32 1, i1 false) |
| Chris Lattner | f41eaac | 2009-09-01 17:56:32 +0000 | [diff] [blame] | 39 | ret void |
| 40 | } |