wth -- Thread time .vs Subroutine time

General FreeBASIC programming questions.
Post Reply
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

'ptrBaseBuffer1' and 'ptrBaseBuffer1plus' are fixed until a buffer becomes exhausted when we go back to Sub SwitchBuffer.
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

@fxm
Yours truly wrote:In your post Feb 13, 2021 where you use " the current version of 'ThreadPooling' " and " the new version of 'ThreadPooling' " I put the new version to PractRand and got a catastrophic failure at 256MB; three times.
I went back to the post Feb 13.

I saved the code " the current version of 'ThreadPooling' " into a file called A.txt. I saved the code " the new version of 'ThreadPooling' " into a file called B.txt.

My text editor, TextPad, has a facility to compare txt files.

Yes, B.txt had references to _mutex3 and used memmove that A.txt did not have. However, there were some statements in A.txt which I would have thought should have carried over into B.txt.

Now I am no way near to fully understanding your code, but I am now wondering whether one or more statements have been inadvertently removed from A that are still required in B.

If you do a txt comparison you may spot something quickly.
fxm
Moderator
Posts: 12082
Joined: Apr 22, 2009 12:46
Location: Paris suburbs, FRANCE

Re: wth -- Thread time .vs Subroutine time

Post by fxm »

I have demonstrated by test a case of blocking for the version that you call "pre mutex3".
This is due to the line:
If pThis->_end = 1 And Ubound(pThis->_pThread) = 0 Then Exit Sub
where the shared variable 'pThis->_pThread' is accessed without mutex exclusion with 'PoolingSubmit'.
Similar code for the version "new".
I corrected this bug (in both '_Thread').

- Version "pre _mutex3" + correction "blocking":

Code: Select all

Type ThreadPooling  '' Version "pre _mutex3" + correction "blocking"
    Public:
        Declare Constructor()
        Declare Sub PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
        Declare Sub PoolingWait()
        Declare Sub PoolingWait(values() As String)
        
        Declare Property PoolingState() As Ubyte
        
        Declare Destructor()
    Private:
        Dim As Function(Byval p As Any Ptr) As String _pThread0
        Dim As Any Ptr _p0
        Dim As Function(Byval p As Any Ptr) As String _pThread(Any)
        Dim As Any Ptr _p(Any)
        Dim As Any Ptr _mutex1
        Dim As Any Ptr _mutex2
        Dim As Any Ptr _pt
        Dim As Byte _end
        Dim As String _returnF(Any)
        Dim As Ubyte _state
        Declare Static Sub _Thread(Byval p As Any Ptr)
End Type

Constructor ThreadPooling()
    Redim This._pThread(0)
    Redim This._p(0)
    Redim This._returnF(0)
    This._mutex1 = Mutexcreate()
    This._mutex2 = Mutexcreate()
    Mutexlock(This._mutex2)
    This._pt= Threadcreate(@ThreadPooling._Thread, @This)
End Constructor

Sub ThreadPooling.PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
    Mutexlock(This._mutex1)
    If Ubound(This._pThread) = 0 Then
        Mutexunlock(This._mutex2)
    End If
    Redim Preserve This._pThread(Ubound(This._pThread) + 1)
    This._pThread(Ubound(This._pThread)) = pThread
    Redim Preserve This._p(Ubound(This._p) + 1)
    This._p(Ubound(This._p)) = p
    Mutexunlock(This._mutex1)
    This._state = 1
End Sub

Sub ThreadPooling.PoolingWait()
    This._end = 1
    Mutexunlock(This._mutex2)
    .ThreadWait(This._pt)
    This._end = 0
    Redim This._returnF(0)
    This._state = 0
    This._pt= Threadcreate(@ThreadPooling._Thread, @This)
End Sub

Sub ThreadPooling.PoolingWait(values() As String)
    This._end = 1
    Mutexunlock(This._mutex2)
    .ThreadWait(This._pt)
    This._end = 0
    If Ubound(This._returnF) > 0 Then
        Redim values(1 To Ubound(This._returnF))
        For I As Integer = 1 To Ubound(This._returnF)
            values(I) = This._returnF(I)
        Next I
        Redim This._returnF(0)
    Else
        Erase values
    End If
    This._state = 0
    This._pt= Threadcreate(@ThreadPooling._Thread, @This)
End Sub

Property ThreadPooling.PoolingState() As Ubyte
    Return This._state
End Property

Sub ThreadPooling._Thread(Byval p As Any Ptr)
    Dim As ThreadPooling Ptr pThis = p
    Dim As Integer ub
    Do
        Mutexlock(pThis->_mutex1)
        While ub = 0
            Mutexunlock(pThis->_mutex1)
            pThis->_state = 4
            Mutexlock(pThis->_mutex2)
            If pThis->_end = 1 Then Exit Sub
            Mutexlock(pThis->_mutex1)
            ub = Ubound(pThis->_pThread)
        Wend
        pThis->_pThread0 = pThis->_pThread(1)
        pThis->_p0 = pThis->_p(1)
        For I As Integer = 2 To Ubound(pThis->_pThread)
            pThis->_pThread(I - 1) = pThis->_pThread(I)
            pThis->_p(I - 1) = pThis->_p(I)
        Next I
        Redim Preserve pThis->_pThread(Ubound(pThis->_pThread) - 1)
        Redim Preserve pThis->_p(Ubound(pThis->_p) - 1)
        ub = Ubound(pThis->_pThread)
        Mutexunlock(pThis->_mutex1)
        Redim Preserve pThis->_ReturnF(Ubound(pThis->_returnF) + 1)
        pThis->_state = 2
        pThis->_returnF(Ubound(pThis->_returnF)) = pThis->_pThread0(pThis->_p0)
    Loop
End Sub

Destructor ThreadPooling()
    This._end = 1
    Mutexunlock(This._mutex2)
    .ThreadWait(This._pt)
    Mutexdestroy(This._mutex1)
    Mutexdestroy(This._mutex2)
End Destructor
- Version "pre _mutex3" + correction "blocking" + change #2 (but not change #1):

Code: Select all

Type ThreadPooling  '' Version "pre _mutex3" + correction "blocking" + change #2 (but not change #1)
    Public:
        Declare Constructor()
        Declare Sub PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
        Declare Sub PoolingWait()
        Declare Sub PoolingWait(values() As String)
        
        Declare Property PoolingState() As Ubyte
        
        Declare Destructor()
    Private:
        Dim As Function(Byval p As Any Ptr) As String _pThread0
        Dim As Any Ptr _p0
        Dim As Function(Byval p As Any Ptr) As String _pThread(Any)
        Dim As Any Ptr _p(Any)
        Dim As Any Ptr _mutex1
        Dim As Any Ptr _mutex2
        Dim As Any Ptr _mutex3
        Dim As Any Ptr _pt
        Dim As Byte _end
        Dim As String _returnF(Any)
        Dim As Ubyte _state
        Declare Static Sub _Thread(Byval p As Any Ptr)
End Type

Constructor ThreadPooling()
    Redim This._pThread(0)
    Redim This._p(0)
    Redim This._returnF(0)
    This._mutex1 = Mutexcreate()
    This._mutex2 = Mutexcreate()
    Mutexlock(This._mutex2)
    This._mutex3 = Mutexcreate()
    Mutexlock(This._mutex3)
    This._pt= Threadcreate(@ThreadPooling._Thread, @This)
End Constructor

Sub ThreadPooling.PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
    Mutexlock(This._mutex1)
    If Ubound(This._pThread) = 0 Then
        Mutexunlock(This._mutex2)
    End If
    Redim Preserve This._pThread(Ubound(This._pThread) + 1)
    This._pThread(Ubound(This._pThread)) = pThread
    Redim Preserve This._p(Ubound(This._p) + 1)
    This._p(Ubound(This._p)) = p
    Mutexunlock(This._mutex1)
    This._state = 1
End Sub

Sub ThreadPooling.PoolingWait()
    Mutexlock(This._mutex3)
    Redim This._returnF(0)
    This._state = 0
    Mutexunlock(This._mutex3)
End Sub

Sub ThreadPooling.PoolingWait(values() As String)
    Mutexlock(This._mutex3)
    If Ubound(This._returnF) > 0 Then
        Redim values(1 To Ubound(This._returnF))
        For I As Integer = 1 To Ubound(This._returnF)
            values(I) = This._returnF(I)
        Next I
        Redim This._returnF(0)
    Else
        Erase values
    End If
    This._state = 0
    Mutexunlock(This._mutex3)
End Sub

Property ThreadPooling.PoolingState() As Ubyte
    Return This._state
End Property

Sub ThreadPooling._Thread(Byval p As Any Ptr)
    Dim As ThreadPooling Ptr pThis = p
    Dim As Integer ub
    Do
        Mutexlock(pThis->_mutex1)
        While ub = 0
            Mutexunlock(pThis->_mutex3)
            Mutexunlock(pThis->_mutex1)
            pThis->_state = 4
            Mutexlock(pThis->_mutex2)
            If pThis->_end = 1 Then Exit Sub
            Mutexlock(pThis->_mutex1)
            ub = Ubound(pThis->_pThread)
            Mutexlock(pThis->_mutex3)
        Wend
        pThis->_pThread0 = pThis->_pThread(1)
        pThis->_p0 = pThis->_p(1)
        For I As Integer = 2 To Ubound(pThis->_pThread)
            pThis->_pThread(I - 1) = pThis->_pThread(I)
            pThis->_p(I - 1) = pThis->_p(I)
        Next I
        Redim Preserve pThis->_pThread(Ubound(pThis->_pThread) - 1)
        Redim Preserve pThis->_p(Ubound(pThis->_p) - 1)
        ub = Ubound(pThis->_pThread)
        Mutexunlock(pThis->_mutex1)
        Redim Preserve pThis->_ReturnF(Ubound(pThis->_returnF) + 1)
        pThis->_state = 2
        pThis->_returnF(Ubound(pThis->_returnF)) = pThis->_pThread0(pThis->_p0)
    Loop
End Sub

Destructor ThreadPooling()
    This._end = 1
    Mutexunlock(This._mutex2)
    .ThreadWait(This._pt)
    Mutexdestroy(This._mutex1)
    Mutexdestroy(This._mutex2)
    Mutexdestroy(This._mutex3)
End Destructor
For the moment, I do not include the change #1 (use of 'memmove') because I have still a doubt on 'memmove' when multi-threading.

If you had the time and the inclination to try both, I would be grateful.
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

- Version "pre _mutex3" + correction "blocking": Failed 4MB
- Version "pre _mutex3" + correction "blocking" + change #2 (but not change #1): Failed 32MB
fxm
Moderator
Posts: 12082
Joined: Apr 22, 2009 12:46
Location: Paris suburbs, FRANCE

Re: wth -- Thread time .vs Subroutine time

Post by fxm »

All of this is incomprehensible !

Do you use a gcc optimization level ('-Ox') when compiling ?
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

fxm wrote:Do you use a gcc optimization level ('-Ox') when compiling ?
Yes, -O2, for all tests including the version which works.

I have just used gas and the last two versions failed with that.
fxm
Moderator
Posts: 12082
Joined: Apr 22, 2009 12:46
Location: Paris suburbs, FRANCE

Re: wth -- Thread time .vs Subroutine time

Post by fxm »

Well, all of this is not satisfactory.
Neither 'ThreadPooling' version satisfies us both.
So, for the moment I do not valid any 'ThreadPooling' version.
I only valid the 'ThreadInitThenMultiStart' version.
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

@fxm

Well, you certainly gave it a good try.

'ThreadInitThenMultiStart' is working with CryptoRndII to at least 1TB is readable and easy to use.

If it is any consolation if anyone else other than FreeBASIC users want to use thread pooling as CryptoRndII does then they will have to use Microsoft's method.

As I have indicated previously that is, in my opinion, a monumental achievement and should be part of the FreeBASIC package. Some PowerBASIC users would give their right arm to use a MultiStart method with their threading applications without the need to use Mirosoft's method; which is what I will do in future.
Lost Zergling
Posts: 534
Joined: Dec 02, 2011 22:51
Location: France

Re: wth -- Thread time .vs Subroutine time

Post by Lost Zergling »

When I posted that one viewtopic.php?f=17&t=29102#p279074 I could not imagine someone could be able to take me at my word. 'However, try to come up with a design(not technique, just 'coup de crayon') that is same time original, intuitive, and relevant'. I was wrong.
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

... and we should thank our lucky stars that we have someone with fxm's calibre maintaining the Documentation Forum and Programmer's Guide.
fxm
Moderator
Posts: 12082
Joined: Apr 22, 2009 12:46
Location: Paris suburbs, FRANCE

Re: wth -- Thread time .vs Subroutine time

Post by fxm »

After thinking calmly, I believe I solved the 'ThreadPooling' Type problem.

It is indeed necessary to be able to do (for the 'PoolingSubmit', 'PoolingWait' and 'Destructeur' methods, all in competition with '_Thread') atomic mutex unlockings, which is not possible with simple mutexlocks / mutexunlocks.
This therefore requires the use of conditional variables (condwait / condsignal).

New version 'ThreadPooling' "super_new":

Code: Select all

#include once "crt/string.bi"
Type ThreadPooling  '' version super-new
    Public:
        Declare Constructor()
        Declare Sub PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
        Declare Sub PoolingWait()
        Declare Sub PoolingWait(values() As String)
        
        Declare Property PoolingState() As Ubyte
        
        Declare Destructor()
    Private:
        Dim As Function(Byval p As Any Ptr) As String _pThread0
        Dim As Any Ptr _p0
        Dim As Function(Byval p As Any Ptr) As String _pThread(Any)
        Dim As Any Ptr _p(Any)
        Dim As Any Ptr _mutex
        Dim As Any Ptr _cond1
        Dim As Any Ptr _cond2
        Dim As Any Ptr _pt
        Dim As Byte _end
        Dim As String _returnF(Any)
        Dim As Ubyte _state
        Declare Static Sub _Thread(Byval p As Any Ptr)
End Type

Constructor ThreadPooling()
    Redim This._pThread(0)
    Redim This._p(0)
    Redim This._returnF(0)
    This._mutex = Mutexcreate()
    This._cond1 = Condcreate()
    This._cond2 = Condcreate()
    This._pt= Threadcreate(@ThreadPooling._Thread, @This)
End Constructor

Sub ThreadPooling.PoolingSubmit(Byval pThread As Function(Byval As Any Ptr) As String, Byval p As Any Ptr = 0)
    Mutexlock(This._mutex)
    Redim Preserve This._pThread(Ubound(This._pThread) + 1)
    This._pThread(Ubound(This._pThread)) = pThread
    Redim Preserve This._p(Ubound(This._p) + 1)
    This._p(Ubound(This._p)) = p
    CondSignal(This._cond2)
    This._state = 1
    Mutexunlock(This._mutex)
End Sub

Sub ThreadPooling.PoolingWait()
    Mutexlock(This._mutex)
    While (This._state And 11) > 0
        Condwait(This._Cond1, This._mutex)
    Wend
    Redim This._returnF(0)
    This._state = 0
    Mutexunlock(This._mutex)
End Sub

Sub ThreadPooling.PoolingWait(values() As String)
    Mutexlock(This._mutex)
    While (This._state And 11) > 0
        Condwait(This._Cond1, This._mutex)
    Wend
    If Ubound(This._returnF) > 0 Then
        Redim values(1 To Ubound(This._returnF))
        For I As Integer = 1 To Ubound(This._returnF)
            values(I) = This._returnF(I)
        Next I
        Redim This._returnF(0)
    Else
        Erase values
    End If
    This._state = 0
    Mutexunlock(This._mutex)
End Sub

Property ThreadPooling.PoolingState() As UByte
    If UBound(This._p) > 0 Then
        Return 8 + This._state
    Else
        Return This._state
    End If
End Property

Sub ThreadPooling._Thread(Byval p As Any Ptr)
    Dim As ThreadPooling Ptr pThis = p
    Do
        Mutexlock(pThis->_mutex)
        If Ubound(pThis->_pThread) = 0 Then
            pThis->_state = 4
            CondSignal(pThis->_cond1)
            While Ubound(pThis->_pThread) = 0
                Condwait(pThis->_cond2, pThis->_mutex)
                If pThis->_end = 1 Then Exit Sub
            Wend
        End If
        pThis->_pThread0 = pThis->_pThread(1)
        pThis->_p0 = pThis->_p(1)
        If Ubound(pThis->_pThread) > 1 Then
            memmove(@pThis->_pThread(1), @pThis->_pThread(2), (Ubound(pThis->_pThread) - 1) * Sizeof(pThis->_pThread))
            memmove(@pThis->_p(1), @pThis->_p(2), (Ubound(pThis->_p) - 1) * Sizeof(pThis->_p))
        End If
        Redim Preserve pThis->_pThread(Ubound(pThis->_pThread) - 1)
        Redim Preserve pThis->_p(Ubound(pThis->_p) - 1)
        Mutexunlock(pThis->_mutex)
        Redim Preserve pThis->_ReturnF(Ubound(pThis->_returnF) + 1)
        pThis->_state = 2
        pThis->_returnF(Ubound(pThis->_returnF)) = pThis->_pThread0(pThis->_p0)
    Loop
End Sub

Destructor ThreadPooling()
    Mutexlock(This._mutex)
    This._end = 1
    CondSignal(This._cond2)
    Mutexunlock(This._mutex)
    .ThreadWait(This._pt)
    Mutexdestroy(This._mutex)
    Conddestroy(This._cond1)
    Conddestroy(This._cond2)
End Destructor
Last edited by fxm on Mar 04, 2023 9:35, edited 5 times in total.
Reason: Added state flag for 'ThreadPooling' and corrected case of blocking for 'ThreadPooling' and therefore also for 'ThreadDispatching' + optimization.
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

I have just gone past 64GB.

It is looking good, Houston. Image
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

256GB! Image
deltarho[1859]
Posts: 4292
Joined: Jan 02, 2017 0:34
Location: UK
Contact:

Re: wth -- Thread time .vs Subroutine time

Post by deltarho[1859] »

Break out the champagne - 1TB and only a couple of small anomalies.

A superb result.

Image

Code: Select all

Microsoft Windows [Version 10.0.19042.804]
(c) 2019 Microsoft Corporation. All rights reserved.
 
C:\Users\deltarho>E:
 
E:\>cd pr64
 
E:\PR64>My_RNG | rng_test stdin32 -tlmin 1KB -multithreaded
RNG_test using PractRand version 0.94
RNG = RNG_stdin32, seed = unknown
test set = core, folding = standard (32 bit)
 
rng=RNG_stdin32, seed=unknown
length= 1 kilobyte (2^10 bytes), time= 0.1 seconds
  no anomalies in 6 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 2 kilobytes (2^11 bytes), time= 0.2 seconds
  no anomalies in 8 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 4 kilobytes (2^12 bytes), time= 0.2 seconds
  no anomalies in 18 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 8 kilobytes (2^13 bytes), time= 0.4 seconds
  no anomalies in 21 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 16 kilobytes (2^14 bytes), time= 0.6 seconds
  no anomalies in 26 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 32 kilobytes (2^15 bytes), time= 0.9 seconds
  no anomalies in 34 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 64 kilobytes (2^16 bytes), time= 1.2 seconds
  no anomalies in 42 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 128 kilobytes (2^17 bytes), time= 1.5 seconds
  no anomalies in 50 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 256 kilobytes (2^18 bytes), time= 1.8 seconds
  no anomalies in 59 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 512 kilobytes (2^19 bytes), time= 2.2 seconds
  no anomalies in 67 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 1 megabyte (2^20 bytes), time= 2.5 seconds
  no anomalies in 74 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 2 megabytes (2^21 bytes), time= 2.9 seconds
  no anomalies in 87 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 4 megabytes (2^22 bytes), time= 3.2 seconds
  no anomalies in 97 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 8 megabytes (2^23 bytes), time= 3.6 seconds
  no anomalies in 106 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 16 megabytes (2^24 bytes), time= 4.0 seconds
  no anomalies in 118 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 32 megabytes (2^25 bytes), time= 4.5 seconds
  Test Name                         Raw       Processed     Evaluation
  [Low8/32]DC6-9x1Bytes-1           R=  +6.0  p =  4.5e-3   unusual
  ...and 127 test result(s) without anomalies
 
rng=RNG_stdin32, seed=unknown
length= 64 megabytes (2^26 bytes), time= 5.0 seconds
  no anomalies in 140 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 128 megabytes (2^27 bytes), time= 5.7 seconds
  no anomalies in 154 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 256 megabytes (2^28 bytes), time= 6.7 seconds
  no anomalies in 165 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 512 megabytes (2^29 bytes), time= 8.1 seconds
  no anomalies in 178 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 1 gigabyte (2^30 bytes), time= 10.6 seconds
  Test Name                         Raw       Processed     Evaluation
  [Low8/32]BCFN(2+1,13-2,T)         R=  +9.1  p =  3.0e-4   unusual
  ...and 191 test result(s) without anomalies
 
rng=RNG_stdin32, seed=unknown
length= 2 gigabytes (2^31 bytes), time= 15.2 seconds
  no anomalies in 204 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 4 gigabytes (2^32 bytes), time= 23.9 seconds
  no anomalies in 216 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 8 gigabytes (2^33 bytes), time= 42.1 seconds
  no anomalies in 229 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 16 gigabytes (2^34 bytes), time= 78.0 seconds
  no anomalies in 240 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 32 gigabytes (2^35 bytes), time= 147 seconds
  no anomalies in 251 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 64 gigabytes (2^36 bytes), time= 294 seconds
  no anomalies in 263 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 128 gigabytes (2^37 bytes), time= 577 seconds
  no anomalies in 273 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 256 gigabytes (2^38 bytes), time= 1105 seconds
  no anomalies in 284 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 512 gigabytes (2^39 bytes), time= 2262 seconds
  no anomalies in 295 test result(s)
 
rng=RNG_stdin32, seed=unknown
length= 1 terabyte (2^40 bytes), time= 4522 seconds
  no anomalies in 304 test result(s)
Lost Zergling
Posts: 534
Joined: Dec 02, 2011 22:51
Location: France

Re: wth -- Thread time .vs Subroutine time

Post by Lost Zergling »

When several threads, possibly distributed over separate cores or processors, share the same mutexes, the additional management of an atomic transactional problem on the mutex is imposed at the low level. However, the structure of the multistart syntax could avoid this constraint, and it has already been observed that CondWait and CondSignal induce a delay compared to mutexes (moreover, it is possible to suppose that the transactional problem was not absolutely evacuated, it will more likely have changed the order of magnitude of occurrence). If the goal of ThreadPooling is to simplify the syntax for managing thread pools, then it might seem relevant to wrap around multiStart rather than start from the paradigm of a code inside, because it's not clear that this would be technically less relevant. If the objective of ThreadPooling is mainly technical (reducing the number of mutexes), then ThreadPooling is relevant versus MultiStart. Over-implementing a ThreadPooling syntax around MultiStart could have two advantages: offloading the low level of a 'free' atomic transactional problem (even if the speed suffers little, the load (processor) may not be the same) , and would also allow a naturally compatible syntax (hierarchichal). I speak a lot, but I help little, thank you for being indulgent.
Post Reply