How does Delphi XE7 handle Multiple Threading in Parallel Tasks - multithreading

I am new to parallel programming and am trying to find out why I occasionally get an EmonitorLockException:Object lock not owned when I increase the number of parrallel tasks to run. Is the case that the Threads become tangled the more tasks I run. Or is my code code not correct?
{$APPTYPE CONSOLE}
uses
System.SysUtils, System.Threading, System.Classes, System.SyncObjs, System.StrUtils;
const
WorkerCount = 10000; // this is the number of tasks to run in parallel note:when this number is increased by repeated factors of 10
// it takes longer to produce the result and sometimes the program crashes
// with an EmonitorLockException:Object lock not owned . Is it that my program is not written correctly or efficiently to run
// a large number of parallel taks and the Threads become entagled.
// Eventually I would like to optimeize the program to find the optimal number of tasks to run in parallel so as to find the result in the shortest time.
// This becomes important when the word sequence is increased to six or more letters.
sequencetofind='help'; // letter sequence to find randomly
sequencelengthplus1=5; // the ength of the letter sequence plus 1 extra letter for a check to see if it is working
var
Ticks: Cardinal;
i,k,m: Integer;
sequencenum: Integer;
alphabetarray:array[1..26] of string;
v:char;
copyarray,letters:array[1..sequencelengthplus1] of string;
tasks: array of ITask;
LTask: ITask;
Event1:TEvent;
sequencesection:TCriticalSection;
function findsequencex(index: Integer): TProc;
begin
Result := procedure
var
counter,m:integer;
r:integer;
z:string;
lettersx:array[1..sequencelengthplus1] of string;
begin
for m:=1 to sequencelengthplus1-1 do
lettersx[m]:=letters[m];
randomize;
counter:=1;
repeat
r:=random(26)+1;
z:=alphabetarray[r]; //randomly find letters until matched with the sequence
if z=letters[counter] then
begin
copyarray[counter]:=z;
counter:=counter+1; // increase counter when successfully found a match
end
else
counter:=1; // if match fails start again and look for the first letter
if (counter=sequencelengthplus1) then
begin // if all letters found in correct order find one more letter as a check
sequencesection.Acquire; //critical section start
r:=random(26)+1;
z:=alphabetarray[r];
TInterlocked.CompareExchange(sequencenum,r,0);
copyarray[sequencelengthplus1]:=z;
Event1.SetEvent; // set in motion the process to stop all other tasks
sequencesection.release; // critical section end
end;
until (Event1.WaitFor(0)=wrSignaled); // check to see if all letters of the sequence has been found
end;
end;
procedure Parallel2;
var
i,sequencevalue,j: Integer;
begin
Event1:=TEvent.Create(nil,true,false,'noname'); // sequence checker
Event1.resetevent;
sequencenum := 0;
Ticks := TThread.GetTickCount;
SetLength(Tasks, WorkerCount); // number of parallel tasks to undertake
for i := 0 to WorkerCount-1 do
Tasks[i]:=TTask.Run(findsequencex(i));
TTask.WaitForAny(Tasks); // wait for the first one to successfully finish
TThread.Synchronize(nil,
procedure
begin
for LTask in Tasks do
LTask.Cancel; // kill the remaining tasks
TInterlocked.Add (sequencevalue, sequencenum); // note the random letter check
end);
Ticks := TThread.GetTickCount - Ticks;
writeln('Parallel time ' + Ticks.ToString + ' ms, last random alphabet sequence number: ' + sequencenum.ToString+' random letter is = '+alphabetarray[sequencevalue]);
end;
begin
sequencesection:=TCriticalSection.Create;
for m:=1 to (sequencelengthplus1-1) do
begin
letters[m]:=copy(sequencetofind,m,1);
writeln(letters[m]);
end;
i:=0;
for v:='a' to 'z' do
begin
i:=i+1;
alphabetarray[i]:=v;
end;
try
begin
Parallel2; // call the parrallel procedure
writeln('finished');
for m:=1 to sequencelengthplus1 do
writeln(copyarray[m]);
if (Event1.WaitFor(0)=wrSignaled) then
begin
writeln('event signaled');
if (sequencenum=0) then writeln('sequence is null');
end;
Event1.Free;
sequencesection.free;
end;
except
on E: Exception do
Writeln(E.ClassName, ': ', E.Message);
end;
Readln;
end.

You've got masses of shared global variables that you access without any synchronization. For instance:
if z = letters[counter] then
begin
copyarray[counter] := z;
counter := counter + 1;
// increase counter when successfully found a match
end
Here copyarray is a global that is shared between all the tasks. This data race alone could result in the error you see. There are other similar races. And I am sure that there are many many more problems. The entire code cannot be salvaged. You need to throw it away and start again.
Here are some tips:
Pick a simpler task to begin learning about parallel programming.
Find a good book or tutorial on the subject. Start small and work up to your actual problem.
Stop using global variables. Sharing data with globals can only lead to pain. Sharing is your enemy. Keep it to a minimum.
If you need to share data do so in an explicit manner rather than using globals. And make sure the access to shared data is synchronized.
Don't call Randomize from inside your thread. It's not threadsafe. Call it once at startup.
Random is not threadsafe. Find a thread safe PRNG, or synchronise calls to Random. The former option is to be preferred.
Don't call TThread.Synchronize from the main thread.
Manage lifetime the standard way. When you create an object, use try and finally to protect its lifetime. Don't create objects in one function and destroy them in some other function.
Format your code so that it is readable. If you cannot read your code, how will you ever understand it?
With the greatest of respect, it's clear that you've not mastered serial programming yet. You should aim to be proficient at serial programming before moving to parallel programming. Parallel programming is at least an order of magnitude harder.
With that in mind, try to write a good, clean version of your program in serial form. Then think about how to transform it to a parallel version.

Related

Paint Polygons Multithreading delphi

For quite some time I tried to improve the speed of my painting program. Unfortunately I just achieved some small improvements by using the OmnithreadLibrary and also by parallelizing the painting process and the loading process of Objects.
My Task in detail:
I stored >1.000.000 Objects in my Database ( Polygons, Rectangles and also Circles). The user should be able to select and paint elements by Type/Position ... .
The selected number of Elements from the user varies from 1 to the max Number of Elements stored in the Database.
Painting a big number of Polygons (>100000) is time consuming. Currently I achieved with my code an improvement of 25%.
How would you speed up the painting process? Where is the mistake?
I would be very grateful for any advice. :)
My Code in detail
Start loading Objects from SQL DB into an ElementArray. This is done by several loading threads. After loading the first Object, the painting thread begins to transform the Data into an Array of TPoints. Converting Data and also Painting Data is done in several Threads. All processes with one exception (Merging Bitmaps) runs parallel.
procedure TbmpthreadForm.StartPaintingPolygons(Sender: TObject);
var
elementsPerThread: Integer;
begin
// 1. Load Data from Database by multithreaded sql queries
// EVery single thread loads the same number of elements
For begin CreateTask(loadTask, IntToStr(i)).MonitorWith(otlMonitor1)
.SetParameter('SQL', sqlStr[i]).Run;
end;
// Save all Array indices in queue
dynamicQueue := TOmniBaseQueue.Create(655365, 4);
// CREATE QUERIES WITH SAME INSTANCE COUNT And Start load DB Objects
for
begin
CreateTask(loadTask, IntToStr(i)).MonitorWith(otlMonitor1)
.SetParameter('SQL', sqlStr).Run;
end;
// START MULTITHREADED PAINT PROCESS
// Single Thread -> Single BMP -> Merge BMPs
Parallel.ParallelTask.NumTasks(4).OnStop(
procedure
begin
masterBitmap.SaveToFile('c:\temp\myimage.bmp');
end).Execute(
procedure
var
value: TOmniValue;
k: Integer;
threadBitMap: TBITMAP;
begin
threadNum.value := threadNum.value + 1;
threadBitMap := TBITMAP.Create;
repeat
// ELEMENT IN QUEUE???? YES-> Paint ELEMENT
if dynamicQueue.TryDequeue(value) then
begin
k := value.AsInteger;
PaintSingleObject(elementList[k], threadBitMap);
end;
until (flag and dynamicQueue.IsEmpty);
// Merge all Bitmaps, after painting all objects
canvas.lock;
BitBlt(masterBitmap.canvas.Handle, 0, 0, masterBitmap.Width,
masterBitmap.Height, threadBitMap.canvas.Handle, 0, 0, SRCAND);
canvas.unlock;
threadBitMap.Free;
end);
end;
LOADING THE Database is done in a few seconds. Painting Process is the bottle neck!
procedure TbmpthreadForm.PaintSingleObject(DS: TObjectTableRecord;
threadBMP: TBITMAP);
var
i, j: Integer;
MyPoly: TPolygon;
aTFPolygon: TFPolygon;
OldPen, NewPen: HPen;
begin
SetPenParameters(threadBMP.canvas, DS, line_pixel, NewPen, OldPen);
...
// Convert a Polygon from string
StringToPolygon(AnsiString(DS.ObjectOutLineString), aTFPolygon);
// Convert Real Position Value to Pixel Value
... MyPoly[j] := TransformLengthToPixel(P2RWMatrix, aTFPolygon[i])
// now Select BrushSetting ...
threadBMP.Canvas.Polygon(aPoly);
end;
Paint_ObjectLabels(threadBMP.canvas, DS, aUnit);
end;
Best,
Michael

Parallel processing strings Delphi full available CPU usage

The goal is to achieve full usage of the available cores, in converting floats to strings in a single Delphi application. I think this problem applies to the general processing of string. Yet in my example I am specifically using the FloatToStr method.
What I am doing (I've kept this very simple so there is little ambiguity around the implementation):
Using Delphi XE6
Create thread objects which inherit from TThread, and start them.
In the thread execute procedure it will convert a large amount of
doubles into strings via the FloatToStr method.
To simplify, these doubles are just the same constant, so there is no
shared or global memory resource required by the threads.
Although multiple cores are used, the CPU usage % always will max out on the amount of a single core. I understand this is an established issue. So I have some specific questions.
In a simple way the same operation could be done by multiple app instances, and thereby achieve more full usage of the available CPU. Is it possible to do this effectively within the same executable ?
I.e. assign threads different process ids on the OS level or some equivalent division recognised by the OS ? Or is this simply not possible in out of the box Delphi ?
On scope :
I know there are different memory managers available & other groups have tried changing some of the lower level asm lock usage http://synopse.info/forum/viewtopic.php?id=57
But, I am asking this question in the scope of not doing things at such a low level.
Thanks
Hi J. My code is deliberately very simple :
TTaskThread = class(TThread)
public
procedure Execute; override;
end;
procedure TTaskThread.Execute;
var
i: integer;
begin
Self.FreeOnTerminate := True;
for i := 0 to 1000000000 do
FloatToStr(i*1.31234);
end;
procedure TfrmMain.Button1Click(Sender: TObject);
var
t1, t2, t3: TTaskThread;
begin
t1 := TTaskThread.Create(True);
t2 := TTaskThread.Create(True);
t3 := TTaskThread.Create(True);
t1.Start;
t2.Start;
t3.Start;
end;
This is a 'test code', where the CPU (via performance monitor) maxes out at 25% (I have 4 cores). If the FloatToStr line is swapped for a non string operation, e.g. Power(i, 2), then the performance monitor shows the expected 75% usage.
(Yes there are better ways to measure this, but I think this is sufficient for the scope of this question)
I have explored this issue fairly thoroughly. The purpose of the question was to put forth the crux of the issue in a very simple form.
I am asking about limitations when using the FloatToStr method. And asking is there an implementation incarnation which will permit better usage of available cores.
Thanks.
I second what everyone else has said in the comments. It is one of the dirty little secrets of Delphi that the FastMM memory manager is not scalable.
Since memory managers can be replaced you can simply replace FastMM with a scalable memory manager. This is a rapidly changing field. New scalable memory managers pop up every few months. The problem is that it is hard to write a correct scalable memory manager. What are you prepared to trust? One thing that can be said in FastMM's favour is that it is robust.
Rather than replacing the memory manager, it is better to replace the need to replace the memory manager. Simply avoid heap allocation. Find a way to do your work with need for repeated calls to allocate dynamic memory. Even if you had a scalable heap manager, heap allocation would still cost.
Once you decide to avoid heap allocation the next decision is what to use instead of FloatToStr. In my experience the Delphi runtime library does not offer much support. For example, I recently discovered that there is no good way to convert an integer to text using a caller supplied buffer. So, you may need to roll your own conversion functions. As a simple first step to prove the point, try calling sprintf from msvcrt.dll. This will provide a proof of concept.
If you can't change the memory manager (MM) the only thing to do is to avoid using it where MM could be a bottleneck.
As for float to string conversion (Disclamer: I tested the code below with Delphi XE) instead of
procedure Test1;
var
i: integer;
S: string;
begin
for i := 0 to 10 do begin
S:= FloatToStr(i*1.31234);
Writeln(S);
end;
end;
you can use
procedure Test2;
var
i: integer;
S: string;
Value: Extended;
begin
SetLength(S, 64);
for i := 0 to 10 do begin
Value:= i*1.31234;
FillChar(PChar(S)^, 64, 0);
FloatToText(PChar(S), Value, fvExtended, ffGeneral, 15, 0);
Writeln(S);
end;
end;
which produce the same result but does not allocate memory inside the loop.
And take attention
function FloatToStr(Value: Extended): string; overload;
function FloatToStr(Value: Extended; const FormatSettings: TFormatSettings): string; overload;
The first form of FloatToStr is not thread-safe, because it uses localization information contained in global variables. The second form of FloatToStr, which is thread-safe, refers to localization information contained in the FormatSettings parameter. Before calling the thread-safe form of FloatToStr, you must populate FormatSettings with localization information. To populate FormatSettings with a set of default locale values, call GetLocaleFormatSettings.
Much thanks for your knowledge and help so far. As per your suggestions I've attempted to write an equivalent FloatToStr method in a way which avoids heap allocation. To some success. This is by no means a solid fool proof implementation, just nice and simple proof of concept which could be extended upon to achieve a more satisfying solution.
(Should also note using XE6 64-bit)
Experiment result/observations:
the CPU usage % was proportional to the number of threads started
(i.e. each thread = 1 core maxed out via performance monitor).
as expected, with more threads started, performance degraded somewhat for each individual one (i.e. time measured to perform task - see code).
times are just rough averages
8 cores 3.3GHz - 1 thread took 4200ms. 6 threads took 5200ms each.
8 cores 2.5GHz - 1 thread took 4800ms. 2=>4800ms, 4=>5000ms, 6=>6300ms.
I did not calculate the overall time for a total multi thread run. Just observed CPU usage % and measured individual thread times.
Personally I find it a little hilarious that this actually works :) Or perhaps I have done something horribly wrong ?
Surely there are library units out there which resolve these things ?
The code:
unit Main;
interface
uses
Winapi.Windows, Winapi.Messages, System.SysUtils, System.Variants, System.Classes, Vcl.Graphics,
Vcl.Controls, Vcl.Forms, Vcl.Dialogs, Vcl.StdCtrls,
Generics.Collections,
DateUtils;
type
TfrmParallel = class(TForm)
Button1: TButton;
Memo1: TMemo;
procedure Button1Click(Sender: TObject);
private
{ Private declarations }
public
{ Public declarations }
end;
TTaskThread = class(TThread)
private
Fl: TList<double>;
public
procedure Add(l: TList<double>);
procedure Execute; override;
end;
var
frmParallel: TfrmParallel;
implementation
{$R *.dfm}
{ TTaskThread }
procedure TTaskThread.Add(l: TList<double>);
begin
Fl := l;
end;
procedure TTaskThread.Execute;
var
i, j: integer;
s, xs: shortstring;
FR: TFloatRec;
V: double;
Precision, D: integer;
ZeroCount: integer;
Start, Finish: TDateTime;
procedure AppendByteToString(var Result: shortstring; const B: Byte);
const
A1 = '1';
A2 = '2';
A3 = '3';
A4 = '4';
A5 = '5';
A6 = '6';
A7 = '7';
A8 = '8';
A9 = '9';
A0 = '0';
begin
if B = 49 then
Result := Result + A1
else if B = 50 then
Result := Result + A2
else if B = 51 then
Result := Result + A3
else if B = 52 then
Result := Result + A4
else if B = 53 then
Result := Result + A5
else if B = 54 then
Result := Result + A6
else if B = 55 then
Result := Result + A7
else if B = 56 then
Result := Result + A8
else if B = 57 then
Result := Result + A9
else
Result := Result + A0;
end;
procedure AppendDP(var Result: shortstring);
begin
Result := Result + '.';
end;
begin
Precision := 9;
D := 1000;
Self.FreeOnTerminate := True;
//
Start := Now;
for i := 0 to Fl.Count - 1 do
begin
V := Fl[i];
// //orignal way - just for testing
// xs := shortstring(FloatToStrF(V, TFloatFormat.ffGeneral, Precision, D));
//1. get float rec
FloatToDecimal(FR, V, TFloatValue.fvExtended, Precision, D);
//2. check sign
if FR.Negative then
s := '-'
else
s := '';
//2. handle negative exponent
if FR.Exponent < 1 then
begin
AppendByteToString(s, 0);
AppendDP(s);
for j := 1 to Abs(FR.Exponent) do
AppendByteToString(s, 0);
end;
//3. count consecutive zeroes
ZeroCount := 0;
for j := Precision - 1 downto 0 do
begin
if (FR.Digits[j] > 48) and (FR.Digits[j] < 58) then
Break;
Inc(ZeroCount);
end;
//4. build string
for j := 0 to Length(FR.Digits) - 1 do
begin
if j = Precision then
Break;
//cut off where there are only zeroes left up to precision
if (j + ZeroCount) = Precision then
Break;
//insert decimal point - for positive exponent
if (FR.Exponent > 0) and (j = FR.Exponent) then
AppendDP(s);
//append next digit
AppendByteToString(s, FR.Digits[j]);
end;
// //use just to test agreement with FloatToStrF
// if s <> xs then
// frmParallel.Memo1.Lines.Add(string(s + '|' + xs));
end;
Fl.Free;
Finish := Now;
//
frmParallel.Memo1.Lines.Add(IntToStr(MillisecondsBetween(Start, Finish)));
//!YES LINE IS NOT THREAD SAFE!
end;
procedure TfrmParallel.Button1Click(Sender: TObject);
var
i: integer;
t: TTaskThread;
l: TList<double>;
begin
//pre generating the doubles is not required, is just a more useful test for me
l := TList<double>.Create;
for i := 0 to 10000000 do
l.Add(Now/(-i-1)); //some double generation
//
t := TTaskThread.Create(True);
t.Add(l);
t.Start;
end;
end.
FastMM4, by default, on thread contention, when one thread cannot acquire access to data, locked by another thread, calls Windows API function Sleep(0), and then, if the lock is still not available enters a loop by calling Sleep(1) after each check of the lock.
Each call to Sleep(0) experiences the expensive cost of a context switch, which can be 10000+ cycles; it also suffers the cost of ring 3 to ring 0 transitions, which can be 1000+ cycles. As about Sleep(1) – besides the costs associated with Sleep(0) – it also delays execution by at least 1 millisecond, ceding control to other threads, and, if there are no threads waiting to be executed by a physical CPU core, puts the core into sleep, effectively reducing CPU usage and power consumption.
That’s why, in your case, CPU use never reached 100% - because of the Sleep(1) issued by FastMM4.
This way of acquiring locks is not optimal.
A better way would have been a spin-lock of about 5000 pause instructions, and, if the lock was still busy, calling SwitchToThread() API call. If pause is not available (on very old processors with no SSE2 support) or SwitchToThread() API call was not available (on very old Windows versions, prior to Windows 2000), the best solution would be to utilize EnterCriticalSection / LeaveCriticalSection, that don’t have latency associated by Sleep(1), and which also very effectively cedes control of the CPU core to other threads.
I have modified FastMM4 to use a new approach to waiting for a lock: CriticalSections instead of Sleep(). With these options, the Sleep() will never be used but EnterCriticalSection / LeaveCriticalSection will be used instead. Testing has shown that the approach of using CriticalSections instead of Sleep (which was used by default before in FastMM4) provides significant gain in situations when the number of threads working with the memory manager is the same or higher than the number of physical cores. The gain is even more evident on computers with multiple physical CPUs and Non-Uniform Memory Access (NUMA). I have implemented compile-time options to take away the original FastMM4 approach of using Sleep(InitialSleepTime) and then Sleep(AdditionalSleepTime) (or Sleep(0) and Sleep(1)) and replace them with EnterCriticalSection / LeaveCriticalSection to save valuable CPU cycles wasted by Sleep(0) and to improve speed (reduce latency) that was affected each time by at least 1 millisecond by Sleep(1), because the Critical Sections are much more CPU-friendly and have definitely lower latency than Sleep(1).
When these options are enabled, FastMM4-AVX it checks:
whether the CPU supports SSE2 and thus the "pause" instruction, and
whether the operating system has the SwitchToThread() API call, and,
and in this case uses "pause" spin-loop for 5000 iterations and then SwitchToThread() instead of critical sections; If a CPU doesn't have the "pause" instrcution or Windows doesn't have the SwitchToThread() API function, it will use EnterCriticalSection / LeaveCriticalSection.
I have made available the fork called FastMM4-AVX at https://github.com/maximmasiutin/FastMM4
Here are the comparison of the Original FastMM4 version 4.992, with default options compiled for Win64 by Delphi 10.2 Tokyo (Release with Optimization), and the current FastMM4-AVX branch. Under some scenarios, the FastMM4-AVX branch is more than twice as fast comparing to the Original FastMM4. The tests have been run on two different computers: one under Xeon E6-2543v2 with 2 CPU sockets, each has 6 physical cores (12 logical threads) - with only 5 physical core per socket enabled for the test application. Another test was done under a i7-7700K CPU.
Used the "Multi-threaded allocate, use and free" and "NexusDB" test cases from the FastCode Challenge Memory Manager test suite, modified to run under 64-bit.
Xeon E6-2543v2 2*CPU i7-7700K CPU
(allocated 20 logical (allocated 8 logical
threads, 10 physical threads, 4 physical
cores, NUMA) cores)
Orig. AVX-br. Ratio Orig. AVX-br. Ratio
------ ----- ------ ----- ----- ------
02-threads realloc 96552 59951 62.09% 65213 49471 75.86%
04-threads realloc 97998 39494 40.30% 64402 47714 74.09%
08-threads realloc 98325 33743 34.32% 64796 58754 90.68%
16-threads realloc 116708 45855 39.29% 71457 60173 84.21%
16-threads realloc 116273 45161 38.84% 70722 60293 85.25%
31-threads realloc 122528 53616 43.76% 70939 62962 88.76%
64-threads realloc 137661 54330 39.47% 73696 64824 87.96%
NexusDB 02 threads 122846 90380 73.72% 79479 66153 83.23%
NexusDB 04 threads 122131 53103 43.77% 69183 43001 62.16%
NexusDB 08 threads 124419 40914 32.88% 64977 33609 51.72%
NexusDB 12 threads 181239 55818 30.80% 83983 44658 53.18%
NexusDB 16 threads 135211 62044 43.61% 59917 32463 54.18%
NexusDB 31 threads 134815 48132 33.46% 54686 31184 57.02%
NexusDB 64 threads 187094 57672 30.25% 63089 41955 66.50%
Your code that calls FloatToStr is OK, since it allocates a result string using the memory manager, then reallocates it, etc. Even better idea would have been to explicitly deallocate it, for example:
procedure TTaskThread.Execute;
var
i: integer;
s: string;
begin
for i := 0 to 1000000000 do
begin
s := FloatToStr(i*1.31234);
Finalize(s);
end;
end;
You can find better tests of the memory manager in the FastCode challenge test suite at https://github.com/maximmasiutin/FastCodeBenchmark
Also, please note that reference counters in Delphi strings use locking operations, which are inherently slow. For example, on an Intel 2400MHz processor with Tiger Lake microarchitecture (released in October 2020), LOCK ADD is about 18 CPU cycles (7.5ns), while non-locked simple ADD is about 0.75 CPU cycles (0.3ns). If your code ensures that the strings are not assigned and modified from different threads, then you may not need this locking. One of the approaches to ensure that a string with multiple references is not manipulated from different threads is to call UniquesString() before such use. Therefore, to improve speed, you may modify the System.pas and to remove the LOCK prefix from the assembly instructions that operate the string reference counters. For example, instead of
LOCK INC [EDX-skew].StrRec.refCnt
there will be
INC [EDX-skew].StrRec.refCnt
However, compiling and using your own, custom version of System.pas may not be an easy task. You can find more information about reference counter locking in Delphi strings in a separate answer.

How to make multiple threads do work on a single TStringList

I want to be able to make multiple threads do work on a single TStringList. This is my current Thread code for using a single thread to do the work;
type
TFilterThread = class(TThread)
protected
procedure Execute; override;
public
lCombos : TStringList;
//Public Vars
end;
procedure TFilterThread.Execute;
var
I: integer;
HTML: String;
frm1 : TForm1;
splitLogin: TStringList;
validCount: Integer;
begin
validCount := 0;
for I := 0 to lCombos.Count - 1 do
begin
if Terminated then Exit();
Unit1.Form1.ListBox1.ItemIndex := i;
try
HTML := Unit1.Form1.IdHTTP1.Get(EmailCheckURL + lCombos[i]);
if AnsiPos('You indicated you', HTML) > 0 then
begin
//Do stuff
end;
except
Continue;
end;
end;
Showmessage('Finished!');
Unit1.Form1.Button1.Caption := 'Start';
end;
To Start the thread I use;
lComboLsit := TStringList.Create;
for I := 0 to listBox1.Items.Count -1 do
lComboLsit.Add(listBox1.Items[i]);`
iTFilterThread := TFilterThread.Create(True);
iTFilterThread.FreeOnTerminate := True;
iTFilterThread.lCombos := lComboLsit;
iTFilterThread.Start;
How would I introduce another thread to also do work on the lCombos list so that the operation would complete quicker?
The answer is that it depends. If the string list is not modified, then there's nothing to do. If the string list is modified then what is best depends critically on your specific usage.
Looking at your code, your program may not be CPU bound and so adding more threads may not help. Your program's bottleneck will the HTTP communication. However, despite not being CPU bound it is plausible that running multiple threads will reduce the impact HTTP latency. You can benchmark to find out. You don't appear to be modifying the string list so there's no race problems to be concerned with there.
However, here's a problem:
Unit1.Form1.IdHTTP1.Get(EmailCheckURL + lCombos[i]);
That will work, and is thread safe so long as TIdHTTP is thread safe. But it's pretty ugly to allow a thread to access a component on a form like that. And I don't see any real sense or need to share the TIdHTTP instance between threads. It would be far cleaner to let each thread instantiate and use their own TIdHTTP component.
Of course, you will need to decide on a policy for dividing the work between all your threads. You could have a shared index that keeps track of the next item to process. Have each thread increment it atomically each time they take an item. A parallel for loop would be a good fit here. That's available in the latest version of Delphi, or in any decent third party parallel library.
You do have some problems with your code. In the thread procedure you do this:
Unit1.Form1.ListBox1.ItemIndex := i;
....
Unit1.Form1.Button1.Caption := 'Start';
You cannot access VCL components from a thread.
And then you do this
ShowMessage('Finished!');
Don't show UI from a thread.
A point of idiom. Instead of looping over the items in your list box you can simply do this:
lComboLsit.Assign(listBox1.Items);

Using the Delphi XE7 Parallel Library

I have a time consuming routine which I'd like to process in parallel using Delphi XE7's new parallel library.
Here is the single threaded version:
procedure TTerritoryList.SetUpdating(const Value: boolean);
var
i, n: Integer;
begin
if (fUpdating <> Value) or not Value then
begin
fUpdating := Value;
for i := 0 to Count - 1 do
begin
Territory[i].Updating := Value; // <<<<<< Time consuming routine
if assigned(fOnCreateShapesProgress) then
fOnCreateShapesProgress(Self, 'Reconfiguring ' + Territory[i].Name, i / (Count - 1));
end;
end;
end;
There is really nothing complex going on. If the territory list variable is changed or set to false then the routine loops around all the sales territories and recreates the territory border (which is the time consuming task).
So here is my attempt to make it parallel:
procedure TTerritoryList.SetUpdating(const Value: boolean);
var
i, n: Integer;
begin
if (fUpdating <> Value) or not Value then
begin
fUpdating := Value;
n := Count;
i := 0;
TParallel.For(0, Count - 1,
procedure(Index: integer)
begin
Territory[Index].Updating := fUpdating; // <<<<<< Time consuming routine
TInterlocked.Increment(i);
TThread.Queue(TThread.CurrentThread,
procedure
begin
if assigned(fOnCreateShapesProgress) then
fOnCreateShapesProgress(nil, 'Reconfiguring ', i / n);
end);
end
);
end;
end;
I've replaced the for-loop with a parallel for-loop. The counter, 'i' is locked as it is incremented to show progress. I then wrap the OnCreateShapeProgress event in a TThread.Queue, which will be handled by the main thread. The OnCreateShapeProgress event is handled by a routine which updates the progress bar and label describing the task.
The routine works if I exclude the call to the OnCreateShapeProgress event. It crashes with an EAurgumentOutOfRange error.
So my question is simple:
Am I doing anything anything stupid?
How to do you call an event handler from within a TParallel.For loop or TTask?
The most obvious problem that I can see is that you queue to the worker thread.
Your call to TThread.Queue passes TThread.CurrentThread. That is the very thread on which you are calling TThread.Queue. I think it is safe to say that you should never pass TThread.CurrentThread to TThread.Queue.
Instead, remove that parameter. Use the one parameter overload that just accepts a thread procedure.
Otherwise I'd note that the incrementing of the progress counter i is not really handled correctly. Well, the incrementing is fine, but you then read it later and that's a race. You can report progress out of order if thread 1 increments before thread 2 but thread 2 queues progress before thread 1. Solve that by moving the counter increment code to the main thread. Simply increment it inside the queued anonymous method. Added bonus to that is you no longer need to use an atomic increment since all modifications are on the main thread.
Beyond that, this QC report seems rather similar to what you report: http://qc.embarcadero.com/wc/qcmain.aspx?d=128392
Finally, AtomicIncrement is the idiomatic way to perform lock free incrementing in the latest versions of Delphi.

Updating each thread's timer in a TListView without CPU overload

Delphi used: 2007
Hello everyone,
I have a TListView with ViewStyle set to vsReport. When I click on a button, I launch about 50 threads. Each thread has a TListItem component. Each TListItem has a SubItem that is a timer. It starts at 250 and goes all the way down to 0. The user is able to see each timer decreasing in the TListView.
I have written the following code:
procedure TThreadWorker.DeleteTickets;
begin
ListItem.Delete;
end;
procedure TThreadWorker.UpdateTimer;
begin
ListItem.SubItems[1] := IntToStr(Timer);
end;
procedure TThreadWorker.TimerCounter;
begin
Timer := 300;
repeat
Sleep(1000);
Dec(Timer);
Synchronize(UpdateTimer);
until (Timer = 0);
Synchronize(DeleteTickets);
end;
And... it works! But here's the thing: all these synchronizations seem to unnecessarily overload the CPU. Obviously, it's a bigger problem when I launch more threads (100, 200 or 300) or when I use a weaker computer. At first, I wasn't sure it was the synchronizations; but if I deactivate them, the CPU is no more overloaded.
To be frank, it's not that much of an issue. However, I have the feeling that decrementing timers shouldn't cause any sort of CPU overload: my code is probably not right. I tried calling UpdateTimer less often, and while it softens the CPU overload, it doesn't fix it in the end. Furthermore, I'd like the user to see the timer updated each second. The timer also needs to be as precise as possible.
Thank you.
I think you have placed the cart ahead of the horse here. Having all your threads synchronize into the main thread, each with their own timer and message queue, will place a heavy burden on the system. What's more, often times you don't want to burden your threads with running a message loop.
A better approach in my view is to place a single timer in the main thread. When it ticks, have it retrieve progress from each thread or task that it needs to report on. You'll need to serialize access to that progress, but that's not expensive.
I think threads are more expensive to the CPU than you might think. From what I remember, the CPU has overhead to swap each thread in and out of the cache. With the CPU swapping out 50 different threads, I'm not surprised its overloading
One solution might be to extend the TTimer component then dynamically create 50 of those rather than 50 threads. TTimer uses the windows api instead of threads. (The code below is untested, but should at least give you the idead)
TMyTimer = class(TTimer)
begin
public
Timer: integer;
ListItem: TListItem;
end;
...
procedure ButtonClick(Sender: TObject)
begin
for i := 0 to 50 do
begin
ltimer = TMyTimer.Create;
ltimer.Timer := 300;
ltimer.ListItem := TListItem.Create;
//initialize list item here
ltimer.OnTimer := DecTimer;
end;
end;
procedure DecTimer(Sender: TObject)
begin
dec(TMytimer(Sender).Timer);
TMyTimer(Sender).ListItem.SubItem[1] := StrToInt(TMytimer(Sender).Timer)
end;
If the threads are all starting at the same time, try doing something like having one thread control up to 25 timer. i.e. For 50 timers you only have two threads. The timer event would then just loop through its 25 counters and decrement them. You'd still need to use synchronize for this.
The answer to this question might be of interest:
How expensive are threads?
Here is example using TThread.Queue and a TSimpleEvent for timing the counter instead of a Sleep().
Type
TThreadWorker = Class(TThread)
private
FTimer : Integer;
FListItem : TListItem;
procedure Execute; override;
procedure UpdateTimer;
procedure DeleteTicket;
public
constructor Create( aListItem : TListItem);
End;
constructor TThreadWorker.Create(aListItem : TListItem);
begin
Inherited Create(false);
FListItem := aListItem;
Self.FreeOnTerminate := true;
end;
procedure TThreadWorker.Execute;
var
anEvent : TSimpleEvent;
begin
anEvent := TSimpleEvent.Create(nil,true,false,'');
try
FTimer := 300;
repeat
anEvent.WaitFor(1000);
Queue(UpdateTimer);
Dec(FTimer);
until (FTimer = 0);
Self.Synchronize( DeleteTicket); // <-- Do not Queue this !
finally
anEvent.Free;
end;
end;
procedure TThreadWorker.UpdateTimer;
begin
FListItem.SubItems[1] := IntToStr(FTimer);
end;
procedure TThreadWorker.DeleteTicket;
begin
FListItem.Delete;
end;
Just a note, the DeleteTicket must be synchronized. The thread is terminated when execute is done, and anything on the queue will be left dangling.

Resources