Jump to content
  • Advertisement

ziomau

Member
  • Content Count

    19
  • Joined

  • Last visited

Everything posted by ziomau

  1. Hi all, I know that this matter has been covered several times, but in all previous posts, there are clear indication of interest about this matter. So, having done my own performance comparison I submit it to the community. I understand that the scope of this comparison is very specific, but that comes from a particular need I have. I currently build automation machines with a real time control done with a PC. Machine personalization is done with a series of programs currently written in Small. These programs are normally compiled into byte code, and the byte code is then injected into the real time environment where a virtual machine does the real execution. The virtual machine exposes to the script several system function call to operate on the physical machine hardware (reading/setting input/outputs, moving axes, setting variables, ecc ...)   Due to the limitations of the Small language (lack of structures, no typing, no doubles, ...) I am investigating the possibility to switch to AngelScript.   Having solved most of the interface problems, now I have an AngelScript compiler and a real time virtual machine for execution running, so I am able to make some performance comparisons.   I started with a very simple script. Here is the AS version: int TestNum; void main(void) { int count = 0; for(int col = 0; col < 10000; col++) { count++; ExtVar = count; // Line A: Makes one call to external environment TestNum = count; // Line B: Only set a script variable } } Here is the Small version: new TestNum; main() { new count = 0; for(new col = 0; col < 10000; col++) { count++; Set_ExtVar(count); // Line A: Makes one call to external environment TestNum = count; // Line B: Only set a script variable } } As you can see they are very similar, and basically they are made of single loop making one local variable increment and than an assignment. Line A an B that you see in the code are alternatives. Only one of them were present during the evaluation. This are the results in mS: AS Small Line A 4.37 1.85 Line B 3.05 1.07 Looks like AS is two to three times slower than Small. Few final notes: The test was done on the same machine with similar load and same real time environment I didn't investigate the JIT, because I am not sure it can even work in the real time environment The external call is the same in both tests executing the same code, but it causes an additional 0.8mS in Small and 1.3mS is AS. Thus it is possible to conclude that the external function call is at least 0.5mS slower The jitter between test runs is in the order of than 0.05 in all cases The slight syntax difference between the two sources in line A is due to the fact that Small does not allow the definition of an external opAssing method, so en explicit function call is made. I believe that this result in a very similar byte code anyway. I am under the impression that AS virtual machine makes some allocation/deallocation during the run of the test, even if everything looks static (or at least allocated at the beginning). I tell this because if I let the test run repeatedly for one hour,, when I shout down the real time system the call to engine->Release takes as much as 5 minutes to complete. I am not able at the moment to tell exactly why. Should you need further information or have me make further tests, just ask. Regards. Mau.   PS: Unfortunately these results forces us to stay with current solution, but I will keep a eye of further AS development, especially in the performance area.
  2.   That's what I imagined. The problem for me is that the real time environment is not equivalent to the OS. Somewhat similar, not identical and far from complete. I will investigate if the page request call is available in my environment, or if some sort of workaround is possible.   Anyway, thanks for the hint. Mau.
  3. Hi All, Thanks for all the information and suggestions.   GarbageCollect I was under the impression that disabling the GC, would cause objects to be released immediately when needed. Ok, my fault, I will add an explicit call and see. I will report here my results. One additional question here: I can call the GarbageCollect on a thread with no strict real time requirements, but I need to know how the execution of the GC will interact with the execution of the script interpreter which must continue to run in real time on alive objects. I mean: does the GC just rise a semaphore at the beginning and drop it at the end (600mS later) or will protect single critical access to shared areas with finer granularity? Or (better) since the GC is disabled in script execution, it just runs on its" to be released" memory being sure that nothing will be added there?   JIT Regarding the Jit probably this is not the best place to ask, but I need some general information to understand if it can be applicable to my case. Just a few questions if someone can give me a simple answer or redirect me to the available information ...   1) Does the JIT generate processor (x86 in my case) instructions, or it is some additional byte code optimization but still interpreted by a virtual machine?   2) In case real processor instructions are generated, I expect that they are first generated as data. So what is the method used to "jump" to the data? Nowadays processor normally disable data execution and only privileged instructions are allowed to change this.   3) Instead, in case only some internal optimization is done and not real processor instructions, what is the expected execution speed gain against the average script code?   Real Time As previous posters have indicated there are several kinds of "real time". Sometimes they have to be certified and/or undergo particular scrutiny. Fortunately, in my case, my code is not involved in deep space or in life savings, so no need for certification (that's also why I can use code like AS or Small without the need to certify them). Nevertheless I need to control machine automation and compute space trajectory of several motors, and this need strict timing requirements. This requirement is what it is normally called "hard real time" which means the deadline should never be missed. Opposite to the "soft real time" where this requirement have to be fulfilled on average, but sparse deadline miss is acceptable (games, or audio processors fall into this second category where this problem is normally solved by a sufficiently deep buffer). The "hard real time" requirement means that I must be sure there aren't bottlenecks or unnecessary critical sections that may cause priority inversion or other nasty (for real time) effects.   Thanks. Mau.
  4. I nailed down a little bit my problem, that happened only if I reload my script several times. I explain a little what I am doing: In my environment I have a single asIScriptEngine class instantiated. This engine is initialized once with the following code: engine->SetDefaultAccessMask(0xFFFFFFFF); RegisterScriptMath(engine); // Register math and trig functions add on RegisterEngineProperties(engine); // Register engine properties RegisterStdString(engine); // Register string class add on RegisterScriptArray(engine, true); // Register generic array add on RegisterSystemVarsDecl(engine); // Register 3 system class types RegisterSystemEnums(engine); // Register about 60 system enums with a total of about 1000 values RegisterSystemFunctions(engine); // Register about 100 system functions RegisterSystemVariables(engine); // Register about 400 system variables of the declared types RegisterUserVariables(engine); // Register about 350 user variables of the declared types RegisterUserIo(engine); // Register about 100 IO definitions (one of the declared types) If needed I can show the classes I am using, but I can say they are very simple, with only few methods and operators declared. Actually they simply represent a system integer, float and boolean types respectively. Variables of these types live outside the script and are accessed by the script when needed through the instance declared during the engine initialization.   Then I create a new module, load the byte code (compiled and saved separately), create a context, set the line callback. Apart from error checking, and some simplification the code is: mod = engine->GetModule(modname, asGM_ALWAYS_CREATE); mod->LoadByteCode(&bs); ctx = engine->CreateContext(); ctx->SetLineCallback(asFUNCTION(LineCallback), &lplc, asCALL_CDECL); mod->ResetGlobalVars(ctx); mainEntry = mod->GetFunctionByDecl("void main(void)"); ctx->Prepare(mainEntry); After this, the real time loop executes the code repeatedly every real time tick as in: while(WaitForNextInterrupt()) { st = ctx->Execute(); if(st == asEXECUTION_FINISHED) ctx->Prepare(plc->mainEntry); } If I leave the above code running for long time this does not show the problem. If I shutdown the program immediately, or even after one hour the results are the same: I have, in both cases, a GC statistics printout like: GetGCStatistics: currentSize:732, totalDestroyed:0, totalDetected:0, newObjects:732, totalNewDestroyed:0 and the full shutdown takes about 600mS. This time is surprisingly long, for what I expect should be done, but at least it is not minutes! And the fact that it does not show an increase with time, makes me feel I am on the safe side here. Now, regarding the problem ... Every now and then, while the real time environment is still running, we need to change/fix, our code and, as fast as possible, replace the running version of the code with the fixed code. Apart from some synchronization between load phase and real time execution, and the obvious error checking, basically the code does the same sequence as above from the GetModule up to the Prepare included. But, this time the code creates a new module, and uses a different set of mod and ctx, leaving the current ones alone (they may be actually running by the real time tick). Then, I wait until the engine is not executing any code (between a real time tick and the next), and atomically swap the two sets of ctx and mod. Thus the next real time tick, the engine that was sitting in the while above executes the new code. After we are sure everything went ok, the old mod and ctx are then destroyed/released. For completeness I must say that the preparation code and the execution are done on different threads. Setting or not the AS_NO_THREADS does not change the behavior as far as this problem is concerned.   When I reload the code again and again, the GC objects start increasing at about 700 object each time, and proportionally increases the time it takes to shutdown. The sequence I tested demonstrates that each time the preparation code is executed, about 700 objects hang out in GC. Things get worse when the reload is repeated, but even the first time we have those 700 objects around.   So the point is: Why I have about 700 objects in the garbage collector when I explicitly disabled it? Additionally, Is the above sequence correct, or should I call some other cleanup function during the load/reload process?   Sorry for these long posts, but I am trying to be as clear and detailed as possible. Thanks for any help. Mau.  
  5. Hi some more comments, before doing some more tests.     "I've not really designed the LineCallback for high performance" Actually, most of the time my LineCallback does nothing. But sometimes, during machine setup, we need to show what the real time code is doing. Since I cannot stop the execution, what I do is to collect the line that are executed at each code scan, to show them highlighted in the editor. That's why I need to collect the line number, and the function executing, nothing more. Sometimes I need to inspect the value of some global variable, but I believe this can be done asking the engine, or execution contest and I can do that in relaxed time. What look still strange to me, as I said, is that the information I need should be at hand in the execution context, so why it takes so long?     "The garbage collection that happens when shutting down ..." I will try to make a printout with GetGCStatistics(), and submit it here. I will try to run for longer time too.     "Perhaps you can share the source code ..." The source will not even compile without the real time extension. I will try to send to you some fragment to show how I initialize and manage the virtual machine, in order to have you check if I made something very wrong.     "I had to increase the number of iterations to 10 million ..." Probably you missed the fact that my timings were in mS. You increased the loop 1000 times to get into a seconds range. That means that my PC is running at the same speed as yours, or at least in the same magnitude order. As I described I measure time using the processor clock. In case the processor clock does not change, the timing is very precise. Some green technology allows the processor to reduce its own speed when under low load, but this normally may be disabled with a bios switch. In our case having to run in real time, we need fastest response even in case the processor has nothing else to do, so we normally disable all "green" setting in the bios. The TSC is an internal 64bit counter that counts up from zero when the processor is powered up by the processor clock. It can be read in C/C++ with this simple code fragment (MSVC): inline __declspec(naked) __int64 RdPTime(void) { _asm { rdtsc ret } }; In Linux (GCC) I have made some test some time ago. I cannot bet it is still working, but here is the code: inline int64_t RdPTime(void) { int64_t ret; rdtscll(ret); return(ret); }; My measurements are done getting the counter just before and just after the code I need to time and taking the difference. The minimum granularity is the processor clock. With nowadays 2Ghz machines or more we are talking about less than a nS (10^-9 Sec). Normally far more resolution than necessary. The only problem is that usually the processor clock is not a know measure. A mainboard sold as 2.8Ghz nominal, may actually generate a clock of 2753Mhz, so if you need to precisely convert ticks to seconds you need to know the exact clock running. Normally I compare the TSC counter difference with a OS known timing (Like Sleep, or other suspend with timeout) for one second,and that should be precise enough for our needs. You get the idea ...     "... at what moment do you need to shutdown the script engine?" Actually you are right. I need to shut down the engine only when the application is closed. Actually, on real machines this will never happen. Machines are just turned off ... no shutdown procedure. So this may be a non problem for me. Nevertheless I prefer is everything works cleanly, before committing to this solution. I don't want to hide maybe some memory leak or a memory fragmentation problem hiding in the code, that may byte me back in some other situation. Automatic machines may run for days or even weeks without being shutdown, so any problem that tend to increase in time is a no go.   Thanks all. I will be back with further investigations in the shutdown area. Regards. Mau.
  6. I made a quick test placing some time reference markers inside the engine destructor in order to track down the reason it take so long to complete in real time. I placed a time marker between every major loop or instruction group inside the destructor code. Here are the results (in mS) together with a small description to identify the code section. Follow the ~asCScriptEngine source code for further details. 1 Calling asDELETE(const_cast<asCScriptEngine*>(this), asCScriptEngine); 0 Calling ~asCScriptEngine 0 Call to SetContextCallbacks 1178 Loop on scriptModules[n]->Discard() 11997 Call to GarbageCollect 0 if and call to defaultArrayObjectType->Release() 13 Loop on templateInstanceTypes 332 Call to GarbageCollect, FreeUnusedGlobalProperties, ClearUnusedTypes 0 Loop on scriptTypes 0 Call to GarbageCollect, FreeUnusedGlobalProperties, ClearUnusedTypes 0 Loop on scriptFunctions 0 Call to GarbageCollect, ClearUnusedTypes 0 Test on refCount 0 Test on ReportAndReleaseUndestroyedObjects 4 Loop on mapTypeIdToDataType 305 Call to RemoveConfiguration and loop on configGroups 117 Iteration on registeredGlobalProps 0 Loop on templateInstanceTypes 0 Loop on allRegisteredTypes and on templateSubTypes 3 Loop on registeredGlobalFuncs and 4 calls to ReleaseAllFunctions 0 Loop on funcDefs 2 Loop on stringConstants 1 Loop on scriptSectionNames 0 Loop on userData 0 Loop on nameSpaces 81 Call to Unprepare and return to main The largest time it takes (12 seconds!) is the first Garbage Collect call, even if I should have disabled it and I have no object that exposes some GC behavior.   I believe I have not enough knowledge to understand this matter or to delve deeper inside the code, but if someone needs further investigation or some specific test, I am available. Remember only that I cannot place breakpoints or examine variables in real time. I can only put time markers, printouts, or counters to be dumped at the end or every now and then during operation.   Regards. Mau.
  7. ziomau

    Limited support for stringstreams

    It is a real time extension to the windows environment (RTX). The real time runs like a windows driver, and normally drivers are written in pure C. This environment offer some support to C++, but it is largely incomplete. Generally it is risky to use the full power of C++ in rel time, since an apparently simple instruction like an assignment or a comparison may hide very expensive library or objects calls. So the most advanced features of std and generally the containers are missing there. Unfortunately stringstream falls into this category.   Hope this helps. Mau.
  8. Hi All, I am running AngelScript in an environment where there is limited or missing support for std::stringstream I have changed the two files of the add-on scriptstdstring.cpp and scriptstdstring.h I submit the here for inclusion in future releases. I understand that this may be a too specific problem to be considered, but anyway, I submit the (partial) solution here for anyone interested. I cannot attach the cpp file, but I can sent it to anyone interested.The changes in the include file are just the addition of the following lines: // Some environment have a limited support of strstreams. // Setting this define to 0 reverts all formatting options to general functions // // 0 = off // 1 = on #ifndef AS_USE_STRSTREAMS #define AS_USE_STRSTREAMS 0 #endif I must admit that the change is not complete. I am not using the generic interface, so in that case I just placed a TODO comment. If someone else is interested I can develop further on this matter, otherwise, I will keep these changes in my AngelScript version only.   Regards. Mau.  
  9. Hi all, thanks for all suggestions. I want just add a few clarifications.   I wasn't aware of the fine tuning page, but having a look at it now, I can tell that most of the suggestions were already included: My functions does not use anything inside the virtual machine, so, no need for caching. I have already the threads and the GC disabled, but I did notice the following issue.   I do need to have the line callback, for debugging purposes. The minimum I need is a call to get the line number and module executing, i.e. a call to ctx->GetLineNumber. But I noticed that while doing my evaluations I included also a call to ctx->GetCallStackSize and ctx->GetFunction that were not really needed, So I did again the same test with an empty callback first and with a single call to GetLineNumber next. Here are the results: Empty GetLineNumber Line A 2.5 4.3 Line B 0.9 3.0 With the GetLineNumber call alone I got similar results as previous tests (meaning that probably GetCallStackSize and GetFunction are not particularly expensive) but there is a significant change removing the GetLineNumber call too. I ignore what this call does internally but it looks strange to me that is is so expensive. Current line and module, should be something at hand in the virtual machine. Probably the fact that it returns a string too causes linear search and string copy/allocate. Probably this is not needed here. Maybe it is worth to investigate here. Just to make some small suggestion, I don't know how applicable here, but I can tell how Small manages this issue. Small, after loading a script, requires a special scan of the script itself where the virtual machine patches the script with Ids instead of strings, function calls and other things. During this scan the virtual machine calls also the line callback function passing the corresponding id for each thing substituted, so the application can build a dictionary or a table. During real execution only Ids have to be manipulated. The advantage of this approach is that the real time environment can work with integers only, and only when there is a need to show something to the user (not in real time, obviously) a time consuming dictionary access and string manipulation is performed. Hope this helps.   Continuing with your comments (somewhat in order):   "Release call" I do not know what causes the 5 minutes delay in the Release call. I cannot debug it since it is running in the real time portion of the code. I can do some further investigation placing some printout. I will be back with details, but I noticed that this behavior looks incremental. I mean, if I run the test for few seconds, that the Release call returns immediately. If I run the test for 10 minutes or more, than the Release call will take more and more time. I suspect memory fragmentation or worse memory leak. In the real time environment the memory management is done differently in order to grant predictable time allocation. This means that some allocation/deallocation made during execution that may go unnoticed in standard environment, may cause problems in real time. I will make some more test in this area.   "Compiler settings" I am using the same compiler, with the same setting and the whole application has only the minimum difference in the code base to be able to run the two different script. I am not trying to denigrate AS here. I am really trying to make my best efforts to be unbiased, and I am really willing to switch to AS, so I am really willing to see good numbers there ... I may have made mistakes, of course (like the line callback above) but I want to state here that I want to be as fair as I can.   "Performance measurement" I am measuring the performance using the CPU clock (rdtsc processor instruction), so the result are precise to the nanosecond, and the accuracy is that of a typical crystal (~100ppm). And yes the speed step or other clock optimization/variations are switched off. Moreover the point of measure is the same in both cases. I am quite confident of the measuring method, also because it is the same timing method we are using since years to time internal happenings in the real time environment. Additionally this method is available in all operating systems since actually it is a processor instruction. A small fragment of inline asm allows it to be used in C/C++   "Support" I had some support in the past from the Small community, but now they turned to a different way more oriented to application scripting and making the language less suitable for real time execution, So I froze my sources at some point (before the switch to the new Pawn naming) and evolved from there with my own code. On the opposite I appreciate very much the support offered by Andreas at first and by the whole community here. Good job!   Thanks all, Mau.
  10. ziomau

    opConv and opImplConv with bool

    That would be great! Thanks. Mau.
  11. ziomau

    Declaring System objects array

    It seems I have still problems managing arrays ... What I am trying to do now is to register a global property which is an array I have seen many examples where an array is registered with RegisterGlobalProperty as: "array<Type> @TypeArray" In my case I the registration returns successfully, but I have a crash in engine->allRegisteredTypes.Erase(cursor), down in an  asASSERT(value < 1000000);   I am trying to set the array dimension too, without success. I tried the following with the indicated outcome: "array<Type> @TypeArray(10)" error -10 "array<Type> TypeArray(10)" error -10 "Type[] @TypeArray(10)" error -10 "Type[] TypeArray(10)" error -10 "array<Type> @TypeArray" crash in engine->allRegisteredTypes.Erase(cursor) "array<Type> TypeArray" crash in engine->allRegisteredTypes.Erase(cursor) "Type[] @TypeArray" crash in engine->allRegisteredTypes.Erase(cursor) "Type[] TypeArray" crash in engine->allRegisteredTypes.Erase(cursor) Does the "Type" object need to have some particular method or declaration to work? What else am I missing?   Thanks for helping. Mau.
  12. ziomau

    Declaring System objects array

    "Type" was an obvious generalization. My real object "SIVar". The relevant code to register SIVar type is: r = engine->RegisterObjectType("SIVar", sizeof(int), asOBJ_VALUE | asOBJ_APP_CLASS_CD); assert(r >= 0); r = engine->RegisterObjectBehaviour("SIVar", asBEHAVE_CONSTRUCT, "void f()", asFUNCTION(ConstructSIVarDef), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectBehaviour("SIVar", asBEHAVE_DESTRUCT, "void f()", asFUNCTION(DestructSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "const uint32 GetSelector()", asFUNCTION(GetSelectorSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "int opImplConv() const", asFUNCTION(IntConvertSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "uint32 opImplConv() const", asFUNCTION(UIntConvertSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); //  r = engine->RegisterObjectMethod("SIVar", "bool opImplConv() const", asFUNCTION(BoolConvertSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "bool b() const", asFUNCTION(BoolConvertSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "int32 opAssign(int32)", asFUNCTION(IntAssignSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); r = engine->RegisterObjectMethod("SIVar", "uint32 opAssign(uint32)", asFUNCTION(UIntAssignSIVar), asCALL_CDECL_OBJLAST); assert(r >= 0); I have a lot of objects of this type and arrays of them living in an external hardware, and the script is not allowed to create/destroy them but only access reading the value (through the implicit conversions to int and uint and possibly bool, when will be allowed) or writing (through the two opAssign).   After declaring this type, I tried to declare those external object withing the script using the RegisterGlobalProperty. The actual code that does this, actually read the definitions of the objects from an xml file and builds on the fly the RegisterGlobalProperty parameters, so the real code fragment is useless here. But at the end RegisterGlobalProperty receives a string and a this pointer. The code is: r = engine->RegisterGlobalProperty(vDef.c_str(), (void*)VarSelector); Where the VarSelector is actually a handle to each of the existing external object, and the vDef string contains something like: "SIVar CurrentWs" "SIVar UsedWs" "SIVar SystemStart" .... Finally I tried to register to the script arrays of such objects, and I tried the strings in the original post (just replace Type with SIVar), getting the described errors.   Thanks for any help. Mau.
  13. ziomau

    opConv and opImplConv with bool

    Following the above discussion, I am too in a situation where a bool conversion would be very usefull. I have a lot of system objects registered as variables into Angel Script, that represent physical digital inputs from hardware. What I am currently trying to do is to allow operation like: out1 = In3 || (In4 && in7); or if(In3 && In8) ...   Is this even possible without an implicit bool conversion? Could be possible to allow this type of operations maybe subject to a configuration variable? Thanks. Mau.
  14. Dear all, I have problems to pass script arrays to registered functions. I searched for examples without success.   In the script I have: int[] Arr = { 123, 456, 789 }; void main(void) { PassArray(Arr); } Then I registered a function: engine->RegisterGlobalFunction("void PassArray(const array<uint32> &in Arr)", asFUNCTION(PassArray), asCALL_CDECL); But compiling the script I have the error: No matching signatures to 'PassArray(int[])' I also tryed all possible variations adding and removing & and @ in both registration and script, without finding any correct way. I coudn't check the real execution of this function but I thought I could declare it in C++ code as: void PassArray(CScriptArray &Arr); Any Idea on how to pass arrays to native functions?   Thanks. Mau.  
  15. Dear all, I need to run the scripts in a real time environment where there is a limited implementation of the standard C++ library, and particularly a very poor stream class. This causes a lot of errors when trying to include the add-on implementation of the string factory and class.   I have two choices to get out of this problem: 1) Implement a different string factory not using streams 2) Implement a different string native object   As for point 1) I know what to do, and for sure this is the quickest and easiest solution, but let me investigate further the second option, that for me is more interesting because it may solve a different category of problems too (see later).   The basic question is: how much is complex to add to the language a simple static "C" type array? In order to have something like: int foo[20]; char buf[100]; SomeDefinedType[3]; ... and possibly ... double Matrix[3][3];   This addition whould automatically solve my string problem too, by using char arrays, and all standard "C" routine strcat, strcpy, strstr, ... by adding them as external calls or even implementing in a script module.   Having the "C" array, can solve my other problem too. Sometimes I have some data structures, that I want to map and exchange between the script enviromìnment and the host interpreter. These structure come from far ago and I cannot easily change them without breaking a lot of code. some of those structures are defined like: struct msg {   int Id;   int len;   char data[24]; };   As it is, I cannot exchange this data type, other than copying over one type to the other. I cannot even include a common definition file to have it defined into the two environment, because the above structure whould not compile in as.   Enough for the description, now the real question: I am willing to help adding this feature, but I need some help to start. Where I need to start looking first to add this thing to the compiler? Any suggestion? Disclamers?   Thanks a lot. Mau.  
  16. ziomau

    Implementation of simple (static) array/strings

    Thanks a lot for your answer, and thanks for pointing me to the array access function too.   The problem with access functions is that expressions using array elements becomes soon unreadable. Even simple expressions like: msg m; int aWord = m.data[0] | (m.data[1] << 8) becomes: msg m; int aWord = m.get_array(0) | (m.get_array(1) << 8); But anyway, ... we can live with it.   Regards. Mau.
  17. ziomau

    Problem loading bytecode

    Dear all, I need to compile the scripts into bytecode, and load and execute them in a completely different environment. I compile the scripts with a slightly modified version of the asbuild. The the other environment I implemente the corresponding slightly modified version of asrun. The problem is when the script references a cdecl external function declared in the environment. I boiled down the issue to this script: void main(void) { SystemFunc(0, 123); } I succesfully register this function in the engine both when I succesfully compile the script and when I try to execute it, with: engine->RegisterGlobalFunction("void SystemFunc(uint, int)", asFUNCTION(SystemFunc), asCALL_CDECL); But then the LoadByteCode function fails apparently because it does not find the function. Further analysis led me to the call sequence into the asCReader object: Read -> ReadInner -> ReadUsedFunctions -> IsSignatureEqual. After few nested calls it end up into asCScriptFunction::IsSignatureExceptNameAndReturnTypeEqual where there is the following code:   if(this->isReadOnly        != readOnly) return false;   if(this->inOutFlags        != paramInOut) return false;   if(this->parameterTypes    != paramTypes) return false;   if((this->objectType != 0) != (objType != 0)) return false;   return true; Where everything matches but the pointer to parameterType. I am quite sure that the contents of the parameters type are pretty the same, but simply they are two different copies of the same thing. And they do not match. I need some suggestion about the way out of this issue. Maybe I have made something wrong, and any help is appreciated. Note that if I compile the script in the same environment everything works as aexpected. The best way, I think to solve is to mach the parameters when they are created and make only one copy, but I donìt know where to look. The other way is to match the contents of the parameters, rather than pointers ... Any suggestion? Thanks. Mau.    
  18. ziomau

    Problem loading bytecode

    Thanks Andreas for your answer. For Your information the environment is a real time kernel, but the platform is a normal win32 with VS2010 compiler, so everythink applies to x86 environment should work into this kernel too. Parameter conventions are the same as the more conventional win32 programming. This is also the reason I want to compile in plain win32 and execute only the code in real time ... it looks like a waste of time to "compile" in real time ... I am working with the latest svn version. My fault was an innocent missing "const" in the signature used running the bytecode, with respect to that used by the compiler, and that was why it didn't match the function. I discovered this tonight, well after I posted here. So sorry for the rised flag ... it was my fault.   Regards. Mau.
  19. ziomau

    Problem loading bytecode

    Sorry for the above noise ... Looking further I've noticed that indeed the parameterTypes array contents are checked so probably there is some other fault in my code ...   Continuing to try ...   Regards. Mau.  
  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

GameDev.net is your game development community. Create an account for your GameDev Portfolio and participate in the largest developer community in the games industry.

Sign me up!