llvm dev - Apr 2018 - LLVM Alias Analysis (Load and store from same address is not showed up in same set)

Hi
I have this simple c code for which I would like to use for alias analysis.


#include <stdio.h>

#include <stdlib.h>


static int (*fp) (void);

void ind_call (int (*compr)(void)){

    fp = compr;

    fp();

}

int hello(void){

    return printf("hello world\n");

}

int main(){

    ind_call(hello);

    return 0;

}

So, I do the following:

bin/opt -basicaa -cfl-anders-aa -print-alias-sets bin/test/test.bc

I get the following result:

Alias sets for function 'ind_call':

Alias Set Tracker: 2 alias sets for 2 pointer values.

  AliasSet[0x91a5820, 1] must alias, Mod/Ref   Pointers: (i32 ()** %compr.addr,
8)

  AliasSet[0x91a58c0, 2] may alias, Mod/Ref   Pointers: (i32 ()** @fp, 8)

    1 Unknown instructions: i32 %call


Alias sets for function 'hello':

Alias Set Tracker: 1 alias sets for 0 pointer values.

  AliasSet[0x91a5910, 1] may alias, Mod/Ref

    1 Unknown instructions: i32 %call


Alias sets for function 'main':

Alias Set Tracker: 2 alias sets for 1 pointer values.

  AliasSet[0x91a5a50, 1] must alias, Mod       Pointers: (i32* %retval, 4)

  AliasSet[0x91a5aa0, 1] may alias, Mod/Ref

    1 Unknown instructions: void <badref>

My question is from c code when we can notice that fp and compr target to same
address (because of fp = compr), why they are not in the same alias set. As you
can see, for function ind_call, the alias set that contain fp is along, while
from my understanding there should be another alias compr.

For better understanding, I also include the LLVM IR here (generated by
LLVM-7.0):

; ModuleID = 'test.c'

source_filename = "test.c"

target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"

target triple = "x86_64-unknown-linux-gnu"


@fp = internal global i32 ()* null, align 8

@.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00",
align 1


; Function Attrs: noinline nounwind optnone uwtable

define dso_local void @ind_call(i32 ()* %compr) #0 {

entry:

  %compr.addr = alloca i32 ()*, align 8

  store i32 ()* %compr, i32 ()** %compr.addr, align 8

  %0 = load i32 ()*, i32 ()** %compr.addr, align 8

  store i32 ()* %0, i32 ()** @fp, align 8

  %1 = load i32 ()*, i32 ()** @fp, align 8

  %call = call i32 %1()

  ret void

}


; Function Attrs: noinline nounwind optnone uwtable

define dso_local i32 @hello() #0 {

entry:

  %call = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([13 x i8], [13
x i8]* @.str, i32 0, i32 0))

  ret i32 %call

}


declare dso_local i32 @printf(i8*, ...) #1


; Function Attrs: noinline nounwind optnone uwtable

define dso_local i32 @main() #0 {

entry:

  %retval = alloca i32, align 4

  store i32 0, i32* %retval, align 4

  call void @ind_call(i32 ()* @hello)

  ret i32 0

}


Best Regards
Mustakimur Rahman Khandaker
Graduate Student and Research Assistant
Department of Computer Science
Florida State University
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180415/f395ab0a/attachment.html>

On 4/15/2018 11:44 AM, Mustakimur Khandaker via llvm-dev
wrote:
> Hi
> I have this simple c code for which I would like to use for alias 
> analysis.
>
>     #include <stdio.h>
>     #include <stdlib.h>
>
>     static int (*fp) (void);
>     void ind_call (int (*compr)(void)){
>         fp = compr;
>         fp();
>     }
>     int hello(void){
>         return printf("hello world\n");
>     }
>     int main(){
>         ind_call(hello);
>         return 0;
>     }
>
>
> So, I do the following:
>
>     bin/opt -basicaa -cfl-anders-aa -print-alias-sets bin/test/test.bc
>
>
> I get the following result:
>
>     Alias sets for function 'ind_call':
>     Alias Set Tracker: 2 alias sets for 2 pointer values.
>       AliasSet[0x91a5820, 1] must alias, Mod/Ref   Pointers: (i32 ()**
>     %compr.addr, 8)
>       AliasSet[0x91a58c0, 2] may alias, Mod/Ref   Pointers: (i32 ()**
>     @fp, 8)
>         1 Unknown instructions: i32 %call
>
LLVM's alias analysis is not very smart; it depends on other 
optimization passes to produce reasonable results.  You should at least 
run mem2reg before trying to do any alias analysis queries.

-Eli


-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux
Foundation Collaborative Project

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180416/f243179c/attachment.html>

Okay, I have tried the following:

 bin/opt -mem2reg -basicaa -cfl-anders-aa -print-alias-sets bin/test/test.bc

But, the result is same as last time. Do you mean something like this to try?
Can you suggest what else could be an important factor for this issue?


Best Regards
Mustakimur Rahman Khandaker
Graduate Student and Teaching Assistant
Department of Computer Science
Florida State University
________________________________
From: Friedman, Eli <efriedma at codeaurora.org>
Sent: Monday, April 16, 2018 1:51:14 PM
To: Mustakimur Khandaker; llvm-dev at lists.llvm.org
Subject: Re: [llvm-dev] LLVM Alias Analysis (Load and store from same address is
not showed up in same set)

On 4/15/2018 11:44 AM, Mustakimur Khandaker via llvm-dev wrote:
Hi
I have this simple c code for which I would like to use for alias analysis.

#include <stdio.h>
#include <stdlib.h>

static int (*fp) (void);
void ind_call (int (*compr)(void)){
    fp = compr;
    fp();
}
int hello(void){
    return printf("hello world\n");
}
int main(){
    ind_call(hello);
    return 0;
}

So, I do the following:
bin/opt -basicaa -cfl-anders-aa -print-alias-sets bin/test/test.bc

I get the following result:
Alias sets for function 'ind_call':
Alias Set Tracker: 2 alias sets for 2 pointer values.
  AliasSet[0x91a5820, 1] must alias, Mod/Ref   Pointers: (i32 ()** %compr.addr,
8)
  AliasSet[0x91a58c0, 2] may alias, Mod/Ref   Pointers: (i32 ()** @fp, 8)
    1 Unknown instructions: i32 %call

LLVM's alias analysis is not very smart; it depends on other optimization
passes to produce reasonable results.  You should at least run mem2reg before
trying to do any alias analysis queries.

-Eli


--
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux
Foundation Collaborative Project
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180416/3b149eef/attachment.html>