X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
X-Trace-PostClient-IP: 68.147.232.190
From: Brian Inglis <Brian DOT Inglis AT SystematicSW DOT Invalid>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: new gcc 4.10 behaviour
Organization: Systematic Software
Message-ID: <6q6v225evfpih3nbj2ts20qre7br1i8an7@4ax.com>
References: <1143542847 DOT 613686 DOT 87820 AT v46g2000cwv DOT googlegroups DOT com> <48sjvaFlir4cU1 AT news DOT dfncis DOT de> <48v4k4Fm65abU1 AT news DOT dfncis DOT de> <48vd7eFm0rukU5 AT news DOT dfncis DOT de>
X-Newsreader: Forte Agent 1.93/32.576 English (American)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Lines: 47
Date: Sun, 02 Apr 2006 10:01:14 GMT
NNTP-Posting-Host: 64.59.135.176
X-Complaints-To: abuse AT shaw DOT ca
X-Trace: pd7tw2no 1143972074 64.59.135.176 (Sun, 02 Apr 2006 04:01:14 MDT)
NNTP-Posting-Date: Sun, 02 Apr 2006 04:01:14 MDT
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

On 29 Mar 2006 12:34:54 GMT in comp.os.msdos.djgpp, Hans-Bernhard
Broeker <broeker AT physik DOT rwth-aachen DOT de> wrote:

>Martin Steuer <martin AT x DOT invalid> wrote:
>
>> And if not, why? Would the optimizer have to assume that the first call 
>> to clock has changed the internal state of that function?
>
>Generally it would --- but not if the two calles are initializers of
>variables.

The initializer guarantees a sequence point: see below. 

>> And in general: how can one guarantee a certain order of execution for 
>> code like that from the OP? 
>
>By making them statements, rather than series of variable definitions.
>That changes introduces what the C standard calls "sequence points",
>which the optimizer must respect, and thus maintains order of things.

ISO C 9899-1999 Annex C Sequence points
"1 The following are the sequence points described in 5.1.2.3:
...
- The end of a full declarator: declarators (6.7.5);
- The end of a full expression: an initializer (6.7.8);
..."

>This would appear to be a side effect of the C99 feature allowing
>variable definitions to occur in the middle of a block, rather than at
>the start of it only.

Then it's a bug, perhaps over zealous optimization, perhaps in (an
internal) definition of the clock function assuming attributes const
or pure apply (when they do not); in such cases, an optimizer could
hoist or rearrange a function call, but it should not in this case. 

It's also a suspicious coincidence that the poster reports a speed
increase of 5 times, and that ticks are 18.2/sec but multiplied by 5
so that CLK_TCK and CLOCKS_PER_SEC can be kept as an integral 91;
perhaps the ticks are being multiplied by 5 elsewhere, as well as in
clock.c? 

-- 
Thanks. Take care, Brian Inglis 	Calgary, Alberta, Canada

Brian DOT Inglis AT CSi DOT com 	(Brian[dot]Inglis{at}SystematicSW[dot]ab[dot]ca)
    fake address		use address above to reply