- std::vector<>::capacity()-growth - 7 Updates
- Now what? - 5 Updates
- Is there a way to use literal-suffixes without selecting a namespace ? - 7 Updates
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 08:18AM +0100 So I just had the idea if the growth-overhead isn't really constant. To check that I wrote this tiny test-program: #include <iostream> #include <vector> #include <chrono> using namespace std; using namespace chrono; int main() { using hrc_tp = time_point<high_resolution_clock>; for( size_t n = (size_t)1 << 8; n <= (size_t)1 << 24; n *= 2 ) { vector<size_t> vi; hrc_tp start = high_resolution_clock::now(); for( size_t i = n; i; --i ) vi.emplace_back( i ); double ns = (double)(int64_t)duration_cast<nanoseconds>( high_resolution_clock::now() - start ).count() / n; cout << n << "\t" << ns << endl; } } The results on my ThreadRipper 3990X are following: 256 31.56250 512 20.15620 1024 8.28125 2048 7.65625 4096 6.81641 8192 11.10350 16384 8.15430 32768 6.22803 65536 5.69458 131072 6.26465 262144 4.69452 524288 5.28549 1048576 3.97713 2097152 4.02081 4194304 4.27639 8388608 4.72761 16777216 3.93250 So the overhead is about to be constant with 32768 elements and further. I'm aware that it is hard to ensure because the memory-allocator won't have a linear behaviour. Please post your results and tell the compiler- / standard-libary -version ! |
| MrSpook_Jo962q_c@3k5_1.biz: Mar 19 10:40AM On Fri, 19 Mar 2021 08:18:06 +0100 > double ns = (double)(int64_t)duration_cast<nanoseconds>( >high_resolution_clock::now() - start ).count() / n; A triple cast? Seriously? |
| Paavo Helde <myfirstname@osa.pri.ee>: Mar 19 01:00PM +0200 19.03.2021 09:18 Bonita Montero kirjutas: > have a linear behaviour. > Please post your results and tell the compiler- / standard-libary > -version ! MSVC2019 + tbbmalloc 256 383.594 512 17.3828 1024 13.0859 2048 7.08008 4096 6.56738 8192 6.26221 16384 4.03442 32768 4.76074 65536 4.80347 131072 3.35541 262144 4.07028 524288 5.70507 1048576 4.62179 2097152 4.47669 4194304 6.01001 8388608 7.32294 16777216 5.90116 |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 01:00PM +0100 >> double ns = (double)(int64_t)duration_cast<nanoseconds>( >> high_resolution_clock::now() - start ).count() / n; > A triple cast? Seriously? I do this because the x86-FPUs don't have load-instructions for unsigned values. And further the compiler would complain that converting from an int64_t to a double might drop information; having an explicit cast does not cause such warnings. |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 01:07PM +0100 Look at this code ... #include <cstdint> using namespace std; double f( uint64_t x ) { return x; } It results in this ... ?f@@YAN_K@Z PROC ; f ; File C:\Users\Olli\x.cpp ; Line 7 xorps xmm0, xmm0 test rcx, rcx js SHORT $LN3@f cvtsi2sd xmm0, rcx ; Line 8 ret 0 $LN3@f: ; Line 7 mov rax, rcx and ecx, 1 shr rax, 1 or rax, rcx cvtsi2sd xmm0, rax addsd xmm0, xmm0 ; Line 8 ret 0 ?f@@YAN_K@Z ENDP ; f While this ... #include <cstdint> using namespace std; double f( uint64_t x ) { return (int64_t)x; } ... results in this ... ?f@@YAN_K@Z PROC ; f ; File C:\Users\Olli\x.cpp ; Line 6 xorps xmm0, xmm0 ; Line 7 cvtsi2sd xmm0, rcx ; Line 8 ret 0 ?f@@YAN_K@Z ENDP ; f |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 01:16PM +0100 >> double ns = (double)(int64_t)duration_cast<nanoseconds>( >> high_resolution_clock::now() - start ).count() / n; > A triple cast? Seriously? And if you want to really improve the above code to be perfect, it would look like the following: double ns = (double)(int64_t)duration_cast<nanoseconds>( high_resolution_clock::now() - start ).count() / (ptrdiff_t)n; I cast to ptrdiff_t because today it's safe to assume that it has the same size like size_t, and ptrdiff_t is implicitly casted to double in this statement, so I don't need an additioncal cast. |
| Ian Collins <ian-news@hotmail.com>: Mar 20 11:23AM +1300 On 20/03/2021 00:00, Paavo Helde wrote: >> -version ! > MSVC2019 + tbbmalloc > 256 383.594 Blimey! > 4194304 6.01001 > 8388608 7.32294 > 16777216 5.90116 AMD Ryzen 7 2700X, Ubuntu clang version 13.0.0 256 13.1914 512 5.32227 1024 4.36426 2048 2.48486 4096 2.66382 8192 2.16602 16384 2.32263 32768 2.64426 65536 3.06087 131072 3.24371 262144 3.36548 524288 3.67879 1048576 4.68726 2097152 5.22924 4194304 4.80952 8388608 5.18689 16777216 5.08674 -- Ian. |
| Brian Wood <woodbrian77@gmail.com>: Mar 18 05:20PM -0700 On Saturday, May 11, 2019 at 12:28:33 AM UTC-5, Brian Wood wrote: > Archimedes > Jerusalem > c-plus-plus The C++ Middleware Writer merges two important areas: code generation and services. By the grace of G-d and with your help, I hope to improve on the areas above. What do you think needs work? Thanks in advance. Brian Ebenezer Enteprises - Enjoying programming again. https://webEbenezer.net https://github.com/Ebenezer-group/onwards |
| Mr Flibble <flibble@i42.REMOVETHISBIT.co.uk>: Mar 19 01:05AM On 19/03/2021 00:20, Brian Wood wrote: > Ebenezer Enteprises - Enjoying programming again. > https://webEbenezer.net > https://github.com/Ebenezer-group/onwards Give it a fucking rest, mate. Best thing you can do right now is start a brand new unrelated project as this project isn't going anywhere: the fact this isn't obvious to you after all this time is depressing. YOU ARE WASTING YOUR TIME BY DOING FUCK ALL EXCEPT RELIGIOUS PROSELYTISM ORIENTED PROCRASTINATION, OBVIOUS TO ALL AND SUNDRY. /Flibble -- π |
| gazelle@shell.xmission.com (Kenny McCormack): Mar 19 01:56AM In article <vNS4I.55821$ut29.21508@fx45.ams4>, Mr Flibble <flibble@i42.REMOVETHISBIT.co.uk> wrote: ... >you after all this time is depressing. YOU ARE WASTING YOUR TIME BY >DOING FUCK ALL EXCEPT RELIGIOUS PROSELYTISM ORIENTED PROCRASTINATION, >OBVIOUS TO ALL AND SUNDRY. Could we all chip in to fund Brian and Rick's honeymoon? If ever two people should just marry and ride off into the sunset, it is these two. -- You are a dreadful man, Kenny, for all your ways are the ways of death. - Rick C Hodgin - (P.S. -> https://www.youtube.com/watch?v=sMmTkKz60W8) |
| Brian Wood <woodbrian77@gmail.com>: Mar 18 06:57PM -0700 On Thursday, March 18, 2021 at 8:05:53 PM UTC-5, Mr Flibble wrote: > > Ebenezer Enteprises - Enjoying programming again. > > https://webEbenezer.net > > https://github.com/Ebenezer-group/onwards Don't swear here, please. Sorry to disappoint you, but this is the way things have to be now. One man's trash is another man's treasure .... |
| red floyd <no.spam.here@its.invalid>: Mar 19 12:28PM -0700 On 3/18/2021 6:57 PM, Brian Wood wrote: > Don't swear here, please. Fuck off. Who died and made you the ruler of c.l.c++? |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 08:21AM +0100 Is there a way to use literal-suffixes without selecting a namespace ? F.e. "1std::s" doesn't work when I didn't say using chrono. Is there something similar that would work ? |
| "Alf P. Steinbach" <alf.p.steinbach+usenet@gmail.com>: Mar 19 12:30PM +0100 On 19.03.2021 08:21, Bonita Montero wrote: > Is there a way to use literal-suffixes without selecting a namespace ? > F.e. "1std::s" doesn't work when I didn't say using chrono. Is there > something similar that would work ? Just using namespace std::literals; and be done with it. ;-) - Alf |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 02:06PM +0100 > Just > using namespace std::literals; > and be done with it. ;-) As you can imagine I'm already aware of that. But that's not what I asked for. |
| "Alf P. Steinbach" <alf.p.steinbach+usenet@gmail.com>: Mar 19 05:19PM +0100 On 19.03.2021 14:06, Bonita Montero wrote: >> and be done with it. ;-) > As you can imagine I'm already aware of that. > But that's not what I asked for. You could call the operator like a function, e.g. std::literals::operator""s( "Blah" ) You could even create a local synonym: auto s( const char* const s ) -> string { return std::literals::operator""s( s ); } Using string literal example because I'm not so familiar with chrono. Disclaimer: off the cuff code, not touched by compilers' dirty hands. - Alf |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 06:49PM +0100 > { return std::literals::operator""s( s ); } > Using string literal example because I'm not so familiar with chrono. > Disclaimer: off the cuff code, not touched by compilers' dirty hands. As my question suggests I was asking for a nearly as convenient solution than the example I gave, i.e. 1std::s f.e.. |
| Juha Nieminen <nospam@thanks.invalid>: Mar 19 07:02PM > Is there a way to use literal-suffixes without selecting a namespace ? > F.e. "1std::s" doesn't work when I didn't say using chrono. Is there > something similar that would work ? While the suffixes are cool and all, I myself just prefer to write eg. std::chrono::seconds(123) Sure, it's longer, but a lot easier to read and understand at a quick glance. |
| Bonita Montero <Bonita.Montero@gmail.com>: Mar 19 08:14PM +0100 > std::chrono::seconds(123) > Sure, it's longer, but a lot easier to read and understand at a > quick glance. I always select the std::/chrono::-namespaces in my .cpp-file globally. |
| You received this digest because you're subscribed to updates for this group. You can change your settings on the group membership page. To unsubscribe from this group and stop receiving emails from it send an email to comp.lang.c+++unsubscribe@googlegroups.com. |
No comments:
Post a Comment