mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-09-10 06:54:54 +00:00
aligned. Teach memcpyopt to not give up all hope when confonted with an underaligned memcpy feeding an overaligned byval. If the *source* of the memcpy can be determined to be adequeately aligned, or if it can be forced to be, we can eliminate the memcpy. This addresses PR9794. We now compile the example into: define i32 @f(%struct.p* nocapture byval align 8 %q) nounwind ssp { entry: %call = call i32 @g(%struct.p* byval align 8 %q) nounwind ret i32 %call } in both x86-64 and x86-32 mode. We still don't get a tailcall though, because tailcalls apparently can't handle byval. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131884 91177308-0d34-0410-b5e6-96231b3b80d8
132 lines
4.5 KiB
LLVM
132 lines
4.5 KiB
LLVM
; RUN: opt < %s -basicaa -memcpyopt -dse -S | FileCheck %s
|
|
|
|
target 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"
|
|
target triple = "i686-apple-darwin9"
|
|
|
|
define void @test1({ x86_fp80, x86_fp80 }* sret %agg.result, x86_fp80 %z.0, x86_fp80 %z.1) nounwind {
|
|
entry:
|
|
%tmp2 = alloca { x86_fp80, x86_fp80 } ; <{ x86_fp80, x86_fp80 }*> [#uses=1]
|
|
%memtmp = alloca { x86_fp80, x86_fp80 }, align 16 ; <{ x86_fp80, x86_fp80 }*> [#uses=2]
|
|
%tmp5 = fsub x86_fp80 0xK80000000000000000000, %z.1 ; <x86_fp80> [#uses=1]
|
|
call void @ccoshl( { x86_fp80, x86_fp80 }* sret %memtmp, x86_fp80 %tmp5, x86_fp80 %z.0 ) nounwind
|
|
%tmp219 = bitcast { x86_fp80, x86_fp80 }* %tmp2 to i8* ; <i8*> [#uses=2]
|
|
%memtmp20 = bitcast { x86_fp80, x86_fp80 }* %memtmp to i8* ; <i8*> [#uses=1]
|
|
call void @llvm.memcpy.i32( i8* %tmp219, i8* %memtmp20, i32 32, i32 16 )
|
|
%agg.result21 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8* ; <i8*> [#uses=1]
|
|
call void @llvm.memcpy.i32( i8* %agg.result21, i8* %tmp219, i32 32, i32 16 )
|
|
|
|
; Check that one of the memcpy's are removed.
|
|
;; FIXME: PR 8643 We should be able to eliminate the last memcpy here.
|
|
|
|
; CHECK: @test1
|
|
; CHECK: call void @ccoshl
|
|
; CHECK: call void @llvm.memcpy
|
|
; CHECK-NOT: llvm.memcpy
|
|
; CHECK: ret void
|
|
ret void
|
|
}
|
|
|
|
declare void @ccoshl({ x86_fp80, x86_fp80 }* sret , x86_fp80, x86_fp80) nounwind
|
|
|
|
declare void @llvm.memcpy.i32(i8*, i8*, i32, i32) nounwind
|
|
|
|
|
|
; The intermediate alloca and one of the memcpy's should be eliminated, the
|
|
; other should be related with a memmove.
|
|
define void @test2(i8* %P, i8* %Q) nounwind {
|
|
%memtmp = alloca { x86_fp80, x86_fp80 }, align 16
|
|
%R = bitcast { x86_fp80, x86_fp80 }* %memtmp to i8*
|
|
call void @llvm.memcpy.i32( i8* %R, i8* %P, i32 32, i32 16 )
|
|
call void @llvm.memcpy.i32( i8* %Q, i8* %R, i32 32, i32 16 )
|
|
ret void
|
|
|
|
; CHECK: @test2
|
|
; CHECK-NEXT: call void @llvm.memmove{{.*}}(i8* %Q, i8* %P
|
|
; CHECK-NEXT: ret void
|
|
}
|
|
|
|
|
|
|
|
|
|
@x = external global { x86_fp80, x86_fp80 }
|
|
|
|
define void @test3({ x86_fp80, x86_fp80 }* noalias sret %agg.result) nounwind {
|
|
%x.0 = alloca { x86_fp80, x86_fp80 }
|
|
%x.01 = bitcast { x86_fp80, x86_fp80 }* %x.0 to i8*
|
|
call void @llvm.memcpy.i32( i8* %x.01, i8* bitcast ({ x86_fp80, x86_fp80 }* @x to i8*), i32 32, i32 16 )
|
|
%agg.result2 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8*
|
|
call void @llvm.memcpy.i32( i8* %agg.result2, i8* %x.01, i32 32, i32 16 )
|
|
ret void
|
|
; CHECK: @test3
|
|
; CHECK-NEXT: %agg.result2 = bitcast
|
|
; CHECK-NEXT: call void @llvm.memcpy
|
|
; CHECK-NEXT: ret void
|
|
}
|
|
|
|
|
|
; PR8644
|
|
define void @test4(i8 *%P) {
|
|
%A = alloca {i32, i32}
|
|
%a = bitcast {i32, i32}* %A to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %P, i64 8, i32 4, i1 false)
|
|
call void @test4a(i8* byval align 1 %a)
|
|
ret void
|
|
; CHECK: @test4
|
|
; CHECK-NEXT: call void @test4a(
|
|
}
|
|
|
|
declare void @test4a(i8* byval align 1)
|
|
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
|
|
|
|
%struct.S = type { i128, [4 x i8]}
|
|
|
|
@sS = external global %struct.S, align 16
|
|
|
|
declare void @test5a(%struct.S* byval align 16) nounwind ssp
|
|
|
|
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
|
|
|
|
; rdar://8713376 - This memcpy can't be eliminated.
|
|
define i32 @test5(i32 %x) nounwind ssp {
|
|
entry:
|
|
%y = alloca %struct.S, align 16
|
|
%tmp = bitcast %struct.S* %y to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* bitcast (%struct.S* @sS to i8*), i64 32, i32 16, i1 false)
|
|
%a = getelementptr %struct.S* %y, i64 0, i32 1, i64 0
|
|
store i8 4, i8* %a
|
|
call void @test5a(%struct.S* byval align 16 %y)
|
|
ret i32 0
|
|
; CHECK: @test5(
|
|
; CHECK: store i8 4
|
|
; CHECK: call void @test5a(%struct.S* byval align 16 %y)
|
|
}
|
|
|
|
;; Noop memcpy should be zapped.
|
|
define void @test6(i8 *%P) {
|
|
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %P, i64 8, i32 4, i1 false)
|
|
ret void
|
|
; CHECK: @test6
|
|
; CHECK-NEXT: ret void
|
|
}
|
|
|
|
|
|
; PR9794 - Should forward memcpy into byval argument even though the memcpy
|
|
; isn't itself 8 byte aligned.
|
|
%struct.p = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 }
|
|
|
|
define i32 @test7(%struct.p* nocapture byval align 8 %q) nounwind ssp {
|
|
entry:
|
|
%agg.tmp = alloca %struct.p, align 4
|
|
%tmp = bitcast %struct.p* %agg.tmp to i8*
|
|
%tmp1 = bitcast %struct.p* %q to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* %tmp1, i64 48, i32 4, i1 false)
|
|
%call = call i32 @g(%struct.p* byval align 8 %agg.tmp) nounwind
|
|
ret i32 %call
|
|
; CHECK: @test7
|
|
; CHECK: call i32 @g(%struct.p* byval align 8 %q) nounwind
|
|
}
|
|
|
|
declare i32 @g(%struct.p* byval align 8)
|
|
|
|
|