[Mono-dev] [PATCH] Redirect certain string constructors to CreateString
Kornél Pál
kornelpal at gmail.com
Sun Jul 16 09:17:59 EDT 2006
Hi,
In thread "[Mono-dev] COM Interop Patch" there is managed-to-managed wrapper
as well so I think it makes sense to add a MONO_WRAPPER_MANAGED_TO_MANAGED
wrapper type.
I understand that using mono_runtime_invoke () makes the things easier but
it will result in a managed-to-native wrapper that wraps a native-to-managed
wrapper so there always will be an overhead.
I would prefer if there were no native code involved because these
constructors don't need icalls at all they just need some wrapper because
constructors cannot return a value.
How difficult would it to introduce a MONO_WRAPPER_MANAGED_TO_MANAGED
wrapper type?
Kornél
----- Original Message -----
From: "Zoltan Varga" <vargaz at gmail.com>
To: "Kornél Pál" <kornelpal at gmail.com>
Cc: <mono-devel-list at lists.ximian.com>
Sent: Sunday, July 16, 2006 1:55 PM
Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors to
CreateString
Hi,
Thinking about it, I think the runtime_invoke () based approach is
better. It might not have
the best performance, but it doesn't have the problems with the
wrapper based approach.
Also, some of the perf overhead is by mono_runtime_invoke () which
needs to be find
the compiled version of the runtime invoke wrapper etc. Later, we
could add an API to
look this up and cache it, making this much faster. If you go with the
wrapper based
approach, you should add a new wrapper type, since MANAGED_TO_NATIVE is for
calling native code, and a lot of runtime parts depend on this.
Zoltan
On 7/16/06, Kornél Pál <kornelpal at gmail.com> wrote:
> Hi,
>
> The previous test results were incorrect. And I added this time I cached
> the
> result of mono_class_get_method_from_name ().
>
> I used the same test with the following parameters:
> string length: 8; iterations: 10000000
> string length: 128; iterations: 10000000
> string length: 4096; iterations: 1000000
>
> And I got the following results for call:
> 6562
> 13922
> 35031
>
> And the following for invoke:
> 14750
> 23141
> 37047
>
> Altough the measures in my previous message were incorrect, a
> managed-to-managed wrapper is much faster.
>
> Kornél
>
> ----- Original Message -----
> From: "Kornél Pál" <kornelpal at gmail.com>
> To: "Zoltan Varga" <vargaz at gmail.com>
> Cc: <mono-devel-list at lists.ximian.com>
> Sent: Sunday, July 16, 2006 5:40 AM
> Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors to
> CreateString
>
>
> > Hi,
> >
> > Using mono_runtime_invoke () is simpler indeed but it's much slower than
> > using managed only implementation. Using the attached test I got 19265
> > for
> > CEE_CALL while 34672 for mono_runtime_invoke (). So I think it's worth
> > to
> > generate managed wrappers unless they cause some serious problems.
> >
> > Is it OK to use MONO_WRAPPER_MANAGED_TO_NATIVE for the generated
> > managed-to-managed wrappers to be consistent with other icalls?
> >
> > Kornél
> >
> > ----- Original Message -----
> > From: "Zoltan Varga" <vargaz at gmail.com>
> > To: "Kornél Pál" <kornelpal at gmail.com>
> > Cc: <mono-devel-list at lists.ximian.com>
> > Sent: Sunday, July 16, 2006 1:40 AM
> > Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors to
> > CreateString
> >
> >
> > Hi,
> >
> > I would prefer the wrapper approach. Another (simpler) approach
> > would be to simply
> > call the CreateString methods from the unmanaged ctors using
> > mono_runtime_invoke ().
> >
> > Zoltan
> >
> > On 7/15/06, Kornél Pál <kornelpal at gmail.com> wrote:
> >> Hi,
> >>
> >> Could you please let me know which solution do you prefer? (redirecting
> >> .ctor to CreateString or using a wrapper that calls CreateString)
> >>
> >> According to my test using a wrapper makes the code slower. I tried to
> >> use
> >> CEE_JMP as well but in that case .ctor is lost from the stack trace as
> >> well
> >> so I think there is no use to create a wrapper using CEE_JMP.
> >>
> >> Thanks for your help.
> >>
> >> Kornél
> >>
> >> ----- Original Message -----
> >> From: "Kornél Pál" <kornelpal at gmail.com>
> >> To: "Zoltan Varga" <vargaz at gmail.com>
> >> Cc: <mono-devel-list at lists.ximian.com>
> >> Sent: Thursday, July 13, 2006 2:48 PM
> >> Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors to
> >> CreateString
> >>
> >>
> >> > Hi,
> >> >
> >> > I reallized that simply redirecting the method may not be a good
> >> > solution
> >> > because the .ctor method is lost from the stack trace this way. And I
> >> > think
> >> > other problems may occur altough I never experienced such problems.
> >> >
> >> > I attached an extended version of the previous patch that creates a
> >> > wrapper.
> >> >
> >> > In addition now CreateString () methods are properly implemented and
> >> > I've
> >> > patched ASCIIEncoding and Latin1Encoding classes so that their
> >> > GetString
> >> > ()
> >> > methods no longer recurse to CreateString () methods.
> >> >
> >> > If these modifications are approved I'll commit this patch and later
> >> > it
> >> > will
> >> > be easier to extend CreateString () to use faster method
> >> > implementations
> >> > for
> >> > internal encoding classes. (I changed my mind.:)
> >> >
> >> > Kornél
> >> >
> >> > ----- Original Message -----
> >> > From: "Kornél Pál" <kornelpal at gmail.com>
> >> > To: "Zoltan Varga" <vargaz at gmail.com>
> >> > Cc: <mono-devel-list at lists.ximian.com>
> >> > Sent: Wednesday, July 12, 2006 5:27 PM
> >> > Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors
> >> > to
> >> > CreateString
> >> >
> >> >
> >> >> Hi,
> >> >>
> >> >> CreateString methods are not yet complete. The attached CreateString
> >> >> methods were forged for testing only.
> >> >>
> >> >> Encoding.GetString() methods are currently using new String (char
> >> >> [])
> >> >> so
> >> >> the performance gain were probably insignificant and
> >> >> Latin1Encoding.GetString () methods currently use new String
> >> >> (sbyte*,
> >> >> int,
> >> >> int) that would result in infinite recursion.
> >> >>
> >> >> I'm going to check-in the modifications made to the runtime after I
> >> >> finish
> >> >> the work on Encoding classes. (And when those modifications will be
> >> >> approved as well.)
> >> >>
> >> >> Thanks for the approval and your help in implementing the code.
> >> >>
> >> >> Kornél
> >> >>
> >> >> ----- Original Message -----
> >> >> From: "Zoltan Varga" <vargaz at gmail.com>
> >> >> To: "Kornél Pál" <kornelpal at gmail.com>
> >> >> Cc: <mono-devel-list at lists.ximian.com>
> >> >> Sent: Wednesday, July 12, 2006 4:30 PM
> >> >> Subject: Re: [Mono-dev] [PATCH] Redirect certain string constructors
> >> >> to
> >> >> CreateString
> >> >>
> >> >>
> >> >> Hi,
> >> >>
> >> >> This looks ok to check-in, except the Console.WriteLine in
> >> >> CreateString
> >> >> ().
> >> >>
> >> >> Zoltan
> >> >>
> >> >> On 7/12/06, Kornél Pál <kornelpal at gmail.com> wrote:
> >> >>> Attached a modified version. Using a function instead of (-1) is a
> >> >>> good
> >> >>> idea
> >> >>> not because g_assert_not_reached () can be used but because (-1)
> >> >>> will
> >> >>> not
> >> >>> be
> >> >>> monopolized to string constructors (altought it's very unlikely to
> >> >>> need
> >> >>> any
> >> >>> other kind of redirection).
> >> >>>
> >> >>> And I added two more g_asserts to ensure that CreateString methods
> >> >>> are
> >> >>> implemented using managed code.
> >> >>>
> >> >>> If you have any other comments please let me know.
> >> >>>
> >> >>> Kornél
> >> >>>
> >> >>> ----- Original Message -----
> >> >>> From: "Zoltan Varga" <vargaz at gmail.com>
> >> >>> To: "Kornél Pál" <kornelpal at gmail.com>
> >> >>> Sent: Wednesday, July 12, 2006 12:43 PM
> >> >>> Subject: Re: [Mono-dev] [PATCH] Redirect certain string
> >> >>> constructors
> >> >>> to
> >> >>> CreateString
> >> >>>
> >> >>>
> >> >>> Hi,
> >> >>>
> >> >>> This approach seems workable. It would be better to have a dummy
> >> >>> string
> >> >>> ctor (eg ves_icall_System_String_ctor_CreateString with a body of
> >> >>> g_assert_not_reached ()) instead of the ICALL_... constant.
> >> >>> Probably
> >> >>> only
> >> >>> mono_marshal_lock () is needed.
> >> >>>
> >> >>> On 7/11/06, Kornél Pál <kornelpal at gmail.com> wrote:
> >> >>> > The previous one did not cache the results. Now this is fixed.
> >> >>> >
> >> >>> > BTW is mono_loader_lock () and mono_marshal_lock () necessary as
> >> >>> > well?
> >> >>> >
> >> >>> > Kornél
> >> >>> >
> >> >>> > ----- Original Message -----
> >> >>> > From: "Kornél Pál" <kornelpal at gmail.com>
> >> >>> > To: <mono-devel-list at lists.ximian.com>
> >> >>> > Sent: Tuesday, July 11, 2006 9:20 AM
> >> >>> > Subject: [PATCH] Redirect certain string constructors to
> >> >>> > CreateString
> >> >>> >
> >> >>> >
> >> >>> > > Hi,
> >> >>> > >
> >> >>> > > The attached patch implements what I was planning to do.
> >> >>> > >
> >> >>> > > Note that if the patch is approved I'm not going to commit is
> >> >>> > > yet
> >> >>> > > because
> >> >>> > > the encoding classes should be updated first.
> >> >>> > >
> >> >>> > > This patch contains String.cs modifications as well that should
> >> >>> > > compile
> >> >>> > > with
> >> >>> > > the current encoding classes as well.
> >> >>> > >
> >> >>> > > I think this is a good solution and works for me but comments
> >> >>> > > are
> >> >>> > > welcome.
> >> >>> > >
> >> >>> > > Please review and approve the patch.
> >> >>> > >
> >> >>> > > Kornél
> >> >>> > >
> >> >>> >
> >> >>> >
> >> >>> > _______________________________________________
> >> >>> > Mono-devel-list mailing list
> >> >>> > Mono-devel-list at lists.ximian.com
> >> >>> > http://lists.ximian.com/mailman/listinfo/mono-devel-list
> >> >>> >
> >> >>> >
> >> >>> >
> >> >>> >
> >> >>>
> >> >>>
> >> >>>
> >> >>
> >> >
> >>
> >>
> >
>
>
More information about the Mono-devel-list
mailing list