With alchemy, we can compile AS3 source files assembly, which means we can use some AVM2 inline assembly language in our AS3 source file. Here is the code snippet for using inline asm and accessing memory in AS3:
/* [Adobe Alchemy Hacks] AlchemyAD_hack.as {Simple example for using inline asm and accessing memory in AS3} By Bruce Jawn (January/23/2011) [http://bruce-lab.blogspot.com] To compile this source file with Alchemy: cd /cygdrive/f/alchemy/ java -Xms16M -Xmx196M -jar F:/alchemy/bin/asc.jar -AS3 -strict -import F:/alchemy/flashlibs/global.abc -import F:/alchemy/flashlibs/playerglobal.abc -config Alchemy::Debugger=false -config Alchemy::NoDebugger=true -config Alchemy::Shell=false -config Alchemy::NoShell=true -config Alchemy::LogLevel=0 -config Alchemy::Vector=true -config Alchemy::NoVector=false -config Alchemy::SetjmpAbuse=false -swf AlchemyAD_hack,800,600,60 AlchemyAD_hack.as */ package { import flash.display.Sprite; import flash.text.TextField; import flash.utils.ByteArray; import flash.utils.Endian; import flash.system.ApplicationDomain; public class AlchemyAD_hack extends Sprite{ public function AlchemyAD_hack () { /*Create the print shell*/ var MyShell:TextField=new TextField(); MyShell.height=600; addChild(MyShell); function print(output:*):void { MyShell.appendText(String(output)); MyShell.appendText("\n"); } /*Test Memory Write*/ //ByteArray for the test var testData:ByteArray = new ByteArray(); testData.endian = Endian.LITTLE_ENDIAN; testData.length=0xffff*4; //select testdata in memory ApplicationDomain.currentDomain.domainMemory=testData; var AdrInt:int=0; //the test value we will write into testData via memory var testValue:int=123; //write the testValue into testData ApplicationDomain.currentDomain.domainMemory[0] = testValue; //Check if testValue has been written into testData print(testData[0]);//should print 123 /*Test Memory Read*/ var readedValue:int=ApplicationDomain.currentDomain.domainMemory[0]; print(readedValue);//should print 123 /*Test Inline ASM*/ //label and jump __asm(jump, target('myLable')); print("not jumped!");//this line will be skipped __asm(label, lbl('myLable')); print("jumped!"); //switch jump var myState:int=1; __asm(push(myState), switchjump('state0','state1','state2')); __asm(lbl('state0')); print("This is state0."); __asm(lbl('state1')); print("This is state1."); __asm(lbl('state2')); print("This is state2."); //iftrue jump var temp:int=1; __asm(push(temp!=0), iftrue, target('turejump')); print("iftrue not jumped!");//this line will be skipped __asm(label, lbl('turejump')); print("iftrue jumped!"); /*Test Alchemy Memory Instructions*/ //All memory opcodes listed here: /* Get a 32 bit value at the location addr and return as an int: _mr32(addr:int):int{ return __xasm======(push(addr), op(0x37)); } Get a 16 bit unsigned value at the location addr and return as an int:. _mru16(addr:int):int { return __xasm (push(addr), op(0x36)); } Get a 16 bit signed value at the location addr and return as an int: _mrs16(addr:int):int { return __xasm (push(addr), op(0x36)); } // li16 Get a 8 bit value at the location addr and return as an int: _mru8(addr:int):int { return __xasm (push(addr), op(0x35)); } Get a 8 bit value at the location addr and return as an int: _mrs8(addr:int):int { return __xasm (push(addr), op(0x35)); } Get a float value at the location addr and return as an Number: _mrf(addr:int):Number { return __xasm (push(addr), op(0x38)); } Get a double value at the location addr and return as an Number: _mrd(addr:int):Number { return __xasm (push(addr), op(0x39)); } Write an int as a 32 bit value at the location addr: _mw32(addr:int, val:int):void { __asm(push(val), push(addr), op(0x3c)); } Write an int as a 16 bit value at the location addr: _mw16(addr:int, val:int):void { __asm(push(val), push(addr), op(0x3b)); } Write an int as a 8 bit value at the location addr: _mw8(addr:int, val:int):void { __asm(push(val), push(addr), op(0x3a)); } Write a Number as a float at the location addr: _mwf(addr:int, val:Number):void { __asm(push(val), push(addr), op(0x3d)); } Write a Number as a double at the location addr: _mwd(addr:int, val:Number):void { __asm(push(val), push(addr), op(0x3e)); } */ //Write an int 654321 as a 32 bit value at the location 1000 __asm(push(654321),push(1000),op(0x3c)); //Trace the memory ApplicationDomain.currentDomain.domainMemory.position=1000; print(ApplicationDomain.currentDomain.domainMemory.readInt());//should print 654321 //Get a 32 bit value at the location 1000 and return as an int var temp:int=__xasm (push(1000), op(0x37)); print(temp);//should print 654321 /*Test some AVM2 Instructions*/ //More AVM2 Instructions can be found at: //http://www.anotherbigidea.com/javaswf/avm2/AVM2Instructions.html //test add: 0xA0 var var1:int=123; var var2:int=321; //write (var1+var2)=444 to testData[0] via memory //ApplicationDomain.currentDomain.domainMemory.position=AdrInt; //ApplicationDomain.currentDomain.domainMemory.writeInt(var1+var2); __asm(push(var1), push(var2), op(0xA0), push(AdrInt), op(0x3c)); testData.position=0; print(testData.readInt());//should print 444 //test subtract: 0xA1 var result:int=__xasm(push(var1), push(var2), op(0xA1));//var result=var1-var2; print(result);//should print -198 //write (var1-var2)=-198 to testData[1] via memory in a different way __asm(push(result),push(AdrInt+4),op(0x3c)); testData.position=4; print(testData.readInt());//should print -198 }//end of function AlchemyAD_hack }//end of class }//end of pacakge /* References: http://labs.adobe.com/wiki/index.php/Alchemy:Documentation:Developing_with_Alchemy:AS3_API http://unitzeroone.com/blog/2009/05/22/another-scream-on-flash-alchemy-memory-and-compilers/ http://blog.frankula.com/?p=211 http://forums.adobe.com/message/2616985 http://forums.adobe.com/message/3001861 Special thanks to Bernd Paradies (http://forums.adobe.com/people/Bernd%20Paradies) */
Update: 20, Feb., 2011
All AVM2 Opcode names for Alchemy can be found here:(ASC source code)
http://opensource.adobe.com/svn/opensource/flex/sdk/trunk/modules/asc/src/java/macromedia/abc/Opcodes.java
so for example, the following code
var result:int=__xasm(push(var1), push(var2), op(0xA1));
can be also writen as
var result:int=__xasm(push(var1), push(var2), subtract);
======
The output:
123
123
jumped!
This is state2.
iftrue jumped!
654321
654321
444
-198
-198
And when we compile C/C++ code to swf or swc, we can manuly modify the *.as file generated during the compilation process, for optimization and then use Alchemy asc to compile the modified *.as to swf or swc. By default the temp *.as file will be deleted, to get that file, we can simply copy and paste that file before it deleted during the compilation, or modify the "gcc" file in "alchemy\achacks" folder, remove/commentize the last two lines:
# remove junk TODO failure leaves stuff around!
if(!$ENV{ACHACKS_TMPS})
{ sys("rm", "-f", <$$.achacks.*>) }
Now we have the generated *.as files from the compiler, something like "19048.achacks.as".
It's easy to compile the *.as to a swf, use the command in the code snippet above.
To compile the *.as to swc, there are several ways.
First way, you can compile the *.as to swf, unzip the swc compiled before and replace the "library.swf" with the new swf. I've tried this but there are some problems I haven't solved.
Second way, modify the gcc file, you can follow this post: http://blog.frankula.com/?p=211.
Third way, use the makefile from this project:
http://alchemy-hacks.googlecode.com/svn/trunk/tricks/
Fourth way, follow this post
http://unitzeroone.com/blog/2009/05/22/another-scream-on-flash-alchemy-memory-and-compilers/
Fifth way, mentioned by Bernd Paradies, can be found here: http://forums.adobe.com/message/3001861
Final way, this is what I recommend, use the wrapper by Ed McManus.
Ed McManus posted the script here: http://forums.adobe.com/message/2616985,
but there are some errors caused by the forum formatting. I fixed the errors,
one obvious error is extra spaces in "nbsp;", another big problem is when you try to use the compiled swc, flashdevelop will throw the error "Target Matching “[xX][mM][lL]” is Not Allowed", thanks to this post http://www.anujgakhar.com/2009/02/17/the-processing-instruction-target-matching-xxmmll-is-not-allowed/, I figure out this problem is caused by the spaces before
catalog.xml's header. I made changes to the script and now it works properly.
You can download the fixed version(alc-asc) here: http://flaswf.googlecode.com/svn/trunk/QuickAlchemy/Hack/SWC/
To use it, put it into your "alchemy/achacks" folder, go to cygwin use command
such as "alc-asc modifiedAlchemy.as outputLib.swc".
Links:
http://forums.adobe.com/message/2616985
http://forums.adobe.com/message/3001861
http://blog.frankula.com/?p=211
http://www.anujgakhar.com/2009/02/17/the-processing-instruction-target-matching-xxmmll-is-not-allowed/
http://unitzeroone.com/blog/2009/05/22/another-scream-on-flash-alchemy-memory-and-compilers/
And Ed McManus wrote some very good documents on Alchemy -
General Porting Tips:
https://github.com/emcmanus/flashsnes/blob/master/docs/General_Porting_Tips
Alchemy VM Architecture:
https://github.com/emcmanus/flashsnes/blob/master/docs/Alchemy_VM_Architecture.txt
Performance:
https://github.com/emcmanus/flashsnes/blob/master/docs/Performance