public inbox for cygwin-patches@cygwin.com
 help / color / mirror / Atom feed
From: Mark Geisert <mark@maxrnd.com>
To: cygwin-patches@cygwin.com
Subject: Re: [PATCH] Cygwin: Make gcc-specific code in <sys/cpuset.h> compiler-agnostic
Date: Fri, 7 Jul 2023 03:13:40 -0700	[thread overview]
Message-ID: <465f8863-6559-e061-684a-a2a812e9c4c6@maxrnd.com> (raw)
In-Reply-To: <ZKfeaMftPy8HmXyy@calimero.vinschen.de>

Hi Corinna,

Corinna Vinschen wrote:
> On Jul  7 00:41, Mark Geisert wrote:
>> The current version of <sys/cpuset.h> cannot be compiled by Clang due to
>> the use of __builtin* functions.  Their presence here was a dubious
>> optimization anyway, so their usage has been converted to standard
>> library functions.  A popcnt (population count of 1 bits in a word)
>> function is provided here because there isn't one in the standard library
>> or elsewhere in the Cygwin DLL.
> 
> And clang really doesn't provide it?  That's unfortunate.
> 
> Do you really think it's not worth to use it if it's available?

I don't know for sure.  I'd guess the popcnt op should be optimized if available; 
the others probably don't need it.

> You could workaround it like this:
> 
>> +/* Modern CPUs have popcnt* instructions but the need here is not worth
>> + * worrying about builtins or inline assembler for different compilers. */
>> +static inline int
>> +__maskpopcnt (__cpu_mask mask)
>> +{
> #if (__GNUC__ >= 4)
>       return __builtin_popcountl (mask);
> #else
>> +  int res = 0;
>> +  unsigned long ulmask = (unsigned long) mask;
>> +
>> +  while (ulmask != 0)
>> +    {
>> +      if (ulmask & 1)
>> +        ++res;
>> +      ulmask >>= 1;
>> +    }
>> +  return res;
> #endif
>> +}
>> +

The first version of the patch (unsubmitted) worked something like that, though it 
was a chore figuring out how to tell the difference between gcc and clang.  clang 
#defines __GNUC__ (?!) for example.  I ended up using __GNUC_PREREQ__ with the 
hope clang version numbers stay lower than gcc version numbers.  Has to be a 
better way than that.

On the other hand, one compilation with clang or clang++, I forget which, and with 
some optimization flag, recognized the 'while' loop in that function and turned it 
into the Hackers Delight algorithm for popcnt in ~20 instructions and no loop.

TL;DR let me ponder this over the weekend.
Thanks for listening,

..mark


  reply	other threads:[~2023-07-07 10:13 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-07-07  7:41 Mark Geisert
2023-07-07  9:44 ` Corinna Vinschen
2023-07-07 10:13   ` Mark Geisert [this message]
2023-07-07 10:45     ` Corinna Vinschen
2023-07-07 15:46   ` Jon Turney
2023-07-07 15:49     ` Corinna Vinschen
2023-07-07 18:54   ` Brian Inglis
2023-07-08 19:22     ` Brian Inglis
2023-07-08 20:59       ` Mark Geisert
2023-07-08 21:53         ` Mark Geisert
2023-07-09  4:27           ` Brian Inglis
2023-07-09  7:58             ` Mark Geisert

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=465f8863-6559-e061-684a-a2a812e9c4c6@maxrnd.com \
    --to=mark@maxrnd.com \
    --cc=cygwin-patches@cygwin.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).