Path: ...!news.nobody.at!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!raubtier-asyl.eternal-september.org!.POSTED!not-for-mail
From: Bonita Montero <Bonita.Montero@gmail.com>
Newsgroups: comp.lang.c
Subject: Re: Top 10 most common hard skills listed on resumes...
Date: Fri, 30 Aug 2024 09:37:47 +0200
Organization: A noiseless patient Spider
Lines: 79
Message-ID: <varss8$de8j$1@raubtier-asyl.eternal-september.org>
References: <vab101$3er$1@reader1.panix.com> <vad8lr$1fv5u$1@dont-email.me>
 <vaf7f0$k51$2@reader1.panix.com> <vafgb2$1to4v$2@dont-email.me>
 <vafkdk$1ut4h$2@raubtier-asyl.eternal-september.org>
 <20240825192810.0000672c@yahoo.com>
 <vafs6u$21ofd$1@raubtier-asyl.eternal-september.org>
 <vafsst$20j4p$3@dont-email.me> <vaj3c4$2lb2c$1@dont-email.me>
 <vaj46o$2kusd$2@dont-email.me> <vajvoh$2t849$1@dont-email.me>
 <vak35f$2tphj$1@raubtier-asyl.eternal-september.org>
 <vak6f8$2tsqj$2@dont-email.me>
 <vak7c0$2ufit$1@raubtier-asyl.eternal-september.org>
 <vaki4u$303sg$1@dont-email.me>
 <vakjff$30c4f$1@raubtier-asyl.eternal-september.org>
 <val7d6$33e83$1@dont-email.me>
 <vamb0t$3btll$2@raubtier-asyl.eternal-september.org>
 <vamqfc$3e42u$1@dont-email.me>
 <vamrra$3e9lv$1@raubtier-asyl.eternal-september.org>
 <20240828134323.00003ce1@yahoo.com>
 <van039$3es48$1@raubtier-asyl.eternal-september.org>
 <20240828150632.000032aa@yahoo.com>
 <van5sl$3fnkh$1@raubtier-asyl.eternal-september.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Fri, 30 Aug 2024 09:37:44 +0200 (CEST)
Injection-Info: raubtier-asyl.eternal-september.org; posting-host="22fa54b4466af970c3e8d7b644a2159b";
	logging-data="440595"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/8Hr+iHQvFOnax750n81ZaMe88T/7zFeA="
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:E9DvTO0wGr6Jr/MrYMSKgGnqYAk=
Content-Language: de-DE
In-Reply-To: <van5sl$3fnkh$1@raubtier-asyl.eternal-september.org>
Bytes: 4522

I've run the numbers on different kinds of virtual indirection:

#include <iostream>
#include <functional>
#include <chrono>
#include <atomic>
#include <memory>

using namespace std;
using namespace chrono;

static atomic_int x( 0 );
auto adder = []( int a, int b ) { return a + b + ::x.load( 
memory_order_relaxed ); };

struct V
{
	virtual int operator ()( int a, int b )
	{
		return adder( a, b );
	}
};

function<int ( int, int )> fn( adder );
function<int ( int, int )> fnRef( ref( adder ) );
function<int ( int, int )> fnC( +adder );
atomic<int (*)( int, int )> cFn( +adder );
unique_ptr<V> v( make_unique<V>() );

int main()
{
	int sum = 0;
	auto bench = [&]<typename Fn>( char const *what, Fn fn )
	{
		using hr_t = high_resolution_clock;
		using dur_t = hr_t::duration;
		dur_t tMin = dur_t::max();
		for( int turn = 100; turn--; )
		{
			auto start = hr_t::now();
			for( int a = 100; a--; )
				for( int b = 100; b--; )
					sum += fn( a, b );
			dur_t dur = hr_t::now() - start;
			tMin = tMin > dur ? dur : tMin;
		}
		cout << what << duration_cast<nanoseconds>( tMin ).count() / 1.0e4 << 
endl;
	};
	bench( "function<>: ", []( int a, int b ) { return ::fn( a, b ); } );
	bench( "function<> ref'd: ", []( int a, int b ) { return ::fnRef( a, b 
); } );
	bench( "function<> C'd: ", []( int a, int b ) { return ::fnC( a, b ); } );
	bench( "virtual method: ", []( int a, int b ) { return (*::v)( a, b ); } );
	bench( "c-function ptr: ", []( int a, int b ) { return ::cFn.load( 
memory_order_relaxed )( a, b ); } );
	bench( "compile-time polymorphism: ", ::adder );
	return sum;
}

This are the results for my Zen4-CPU:

	function<>: 0.91
	function<> ref'd: 1.09
	function<> C'd: 1.1
	virtual method: 0.92
	c-function ptr: 0.95
	compile-time polymorphism: 0.41

So there's only half a nanosecond more overhead when using a function<>
-object with external storage (similar to a virtual function, but only
with one indirection since there's no virtual function table) over di-
rect inlining. No matter if you have a function<>-object, whether it is
initialized with a reference_wrapper (no external storage) or a C func-
tion pointer or if you use a C function pointler directly: it's all the
same performance. Although there are different complexities there are
enough calculation units in the Zen4-CPU to do it all in about the same
time.