llvm dev - Aug 2018 - A question about the widely used "xxx | true" inside llvm

This is better suited to llvm-dev, so I'm directing it there.

From: llvm-commits <llvm-commits-bounces at lists.llvm.org> on behalf of
Qing Shan Zhang via llvm-commits <llvm-commits at lists.llvm.org>
Reply-To: Qing Shan Zhang <qshanz at cn.ibm.com>
Date: Sunday, August 19, 2018 at 10:10 PM
To: "llvm-commits at lists.llvm.org" <llvm-commits at
lists.llvm.org>
Subject: A question about the widely used "xxx | true" inside llvm

Hi, guys,

I notice that, we are widely using the statement "xxx | true" in llvm,
such as "return SimplifyCFG(BB) | true;".  I have no idea if there is
any downside to use the logic operator "||" here. Instead, I indeed
see the potential problem to use bitwise operator "|". Because the
bitwise operator do not short-circuit and the language std didn't specify
the evaluation order for its lhs and rhs expression. That means, compiler is
free to optimize this statement to "return true".

In shorts, IMO, it is legal to do the following optimization:
bool bar() {
  return foo() | true;
}
-->
bool bar() {
  return true;  // the call to foo() is missing.
}

Best regards

steven.zhang(张青山)
XLC++ Compiler Frontend Developer

IBM China Development Lab, Shanghai
Tel: (8621)609-28454 Mobile: +8615900986116
E-mail: qshanz at cn.ibm.com<mailto:qshanz at cn.ibm.com>
关注IBM中国编译器开发团队 - 新浪微博:
http://www.weibo.com/ibmcompiler<https://urldefense.proofpoint.com/v2/url?u=http-3A__www.weibo.com_ibmcompiler&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=A_Y7Fzz24WHgMcnZjkQNCxix8GiaTORg0B7J-WtGp9c&e=>
| developerWorks:
http://t.cn/SPHWF4<https://urldefense.proofpoint.com/v2/url?u=http-3A__t.cn_SPHWF4&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=EJ9ZnKu_kAhWtR9SgL-C_ooRh7TCL-EGc0wAotg4wb4&e=>

"All things are difficult before they are easy."


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180820/4e3c2453/attachment.html>

The language requires both sides of the | need to be evaluated. The
compiler can't remove the function call if it has side effects.


~Craig


On Sun, Aug 19, 2018 at 10:15 PM Shoaib Meenai via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> This is better suited to llvm-dev, so I'm directing it there.
>
>
>
> *From: *llvm-commits <llvm-commits-bounces at lists.llvm.org> on
behalf of
> Qing Shan Zhang via llvm-commits <llvm-commits at lists.llvm.org>
> *Reply-To: *Qing Shan Zhang <qshanz at cn.ibm.com>
> *Date: *Sunday, August 19, 2018 at 10:10 PM
> *To: *"llvm-commits at lists.llvm.org" <llvm-commits at
lists.llvm.org>
> *Subject: *A question about the widely used "xxx | true" inside
llvm
>
>
>
> Hi, guys,
>
>
>
> I notice that, we are widely using the statement "xxx | true" in
llvm,
> such as "return SimplifyCFG(BB) | true;".  I have no idea if
there is any
> downside to use the logic operator "||" here. Instead, I indeed
see the
> potential problem to use bitwise operator "|". Because the
bitwise operator
> do not short-circuit and the language std didn't specify the evaluation
> order for its lhs and rhs expression. That means, compiler is free to
> optimize this statement to "return true".
>
>
>
> In shorts, IMO, it is legal to do the following optimization:
> bool bar() {
>   return foo() | true;
> }
> -->
> bool bar() {
>   return true;  // the call to foo() is missing.
>
> }
>
>
> Best regards
>
> steven.zhang(张青山)
> XLC++ Compiler Frontend Developer
>
> IBM China Development Lab, Shanghai
> Tel: (8621)609-28454 Mobile: +8615900986116
> E-mail: qshanz at cn.ibm.com
> 关注IBM中国编译器开发团队 - 新浪微博: http://www.weibo.com/ibmcompiler
>
<https://urldefense.proofpoint.com/v2/url?u=http-3A__www.weibo.com_ibmcompiler&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=A_Y7Fzz24WHgMcnZjkQNCxix8GiaTORg0B7J-WtGp9c&e=>
> | developerWorks: http://t.cn/SPHWF4
>
<https://urldefense.proofpoint.com/v2/url?u=http-3A__t.cn_SPHWF4&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=EJ9ZnKu_kAhWtR9SgL-C_ooRh7TCL-EGc0wAotg4wb4&e=>
>
> "All things are difficult before they are easy."
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180819/2f94c8c5/attachment.html>

Shoaib Meenai via llvm-dev <llvm-dev at lists.llvm.org>
writes:
> This is better suited to llvm-dev, so I'm directing it there.
>
> From: llvm-commits <llvm-commits-bounces at lists.llvm.org> on behalf
of
> Qing Shan Zhang via llvm-commits <llvm-commits at lists.llvm.org>
> Reply-To: Qing Shan Zhang <qshanz at cn.ibm.com>
> Date: Sunday, August 19, 2018 at 10:10 PM
> To: "llvm-commits at lists.llvm.org" <llvm-commits at
lists.llvm.org>
> Subject: A question about the widely used "xxx | true" inside
llvm
>
> Hi, guys,
>
> I notice that, we are widely using the statement "xxx | true" in
llvm,
> such as "return SimplifyCFG(BB) | true;".  I have no idea if
there is
> any downside to use the logic operator "||" here. Instead, I
indeed
> see the potential problem to use bitwise operator "|". Because
the
> bitwise operator do not short-circuit and the language std didn't
> specify the evaluation order for its lhs and rhs expression. That
> means, compiler is free to optimize this statement to "return
true".
I think "widely using" is overstating the issue. This seems to be
happening in a lot of places in SimplifyCFG though, and from looking at
the code and the history it's pretty clear to me that the uses of the
bitwise or are either stylistic or unintentional. I've gone ahead and
changed all of those to use || for clarity in r340153.
> In shorts, IMO, it is legal to do the following optimization:
> bool bar() {
>   return foo() | true;
> }
> -->
> bool bar() {
>   return true;  // the call to foo() is missing.
> }
This is only legal if the compiler can prove that foo() has no side
effects, which isn't true in this case. Using bitwise or here is
certainly confusing, but there isn't a correctness issue.
> Best regards
>
> steven.zhang(张青山)
> XLC++ Compiler Frontend Developer
>
> IBM China Development Lab, Shanghai
> Tel: (8621)609-28454 Mobile: +8615900986116
> E-mail: qshanz at cn.ibm.com<mailto:qshanz at cn.ibm.com>
> 关注IBM中国编译器开发团队 - 新浪微博:
>
http://www.weibo.com/ibmcompiler<https://urldefense.proofpoint.com/v2/url?u=http-3A__www.weibo.com_ibmcompiler&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=A_Y7Fzz24WHgMcnZjkQNCxix8GiaTORg0B7J-WtGp9c&e=>
> | developerWorks:
>
http://t.cn/SPHWF4<https://urldefense.proofpoint.com/v2/url?u=http-3A__t.cn_SPHWF4&d=DwMFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=Pq-hk0wEUZUsW1TCR5WSeKCTC9p-kFZHFQXEwhAc7-k&s=EJ9ZnKu_kAhWtR9SgL-C_ooRh7TCL-EGc0wAotg4wb4&e=>
>
> "All things are difficult before they are easy."
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

On Aug 19, 2018, at 10:33 PM, Craig Topper via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> The language requires both sides of the | need to be evaluated. The
compiler can't remove the function call if it has side effects.
Correct.  If you are somehow really bothered by this, we could also use:

bool bar() {
  return foo(), true;
}

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20180822/30b1b148/attachment.html>