Smart Pascal FastCode inspiration
Remember the Fastcode project? The alternative memory management and memory operation routines which almost doubled the speed of Delphi 5-7 applications? And they did it by replacing the bog standard code from Borland with hand-optimized assembly language alternatives.
Well, today I wrote the Smart Pascal equivalent of the Fastcode project for Delphi. Or at least that’s what it reminded me of. So here is the Smart Pascal variations of the “Fastcode” like replacement modules 🙂
And yes, I did test these against 2-3 variations of the same procedures and these are as fast as I can make them without introducing lookup tables.
How fast are these routines? Well, the following code snippet was used for benchmarking:
for x:=1 to 1000000 do begin mMemory:=AllocMem(1024); try WriteMem(mMemory,14,TDatatype.BytesToTypedArray ([10,20,30,40,50,60,70,80,90,100])); mMemory:=ReAllocMem(mMemory,512); Readmem(mMemory,14,10); finally Freemem(mMemory); end; end;
The original code executes with time-code 551855 (which is approx 2.3 seconds) while this new variation executes with time-code 131816 which is a quite significant boost! But this is naturally because I have completely bypassed marshaled pointers which we are about to introduce in the next update of Smart Mobile Studio.
As explained earlier, a marshaled pointer is ultimately an object which is managed, or takes care of something for you. In this case it assumes the role of a pointer, taking care of the reference to the memory segment (TMemoryHandle, as used above) and the offset into that segment.
But for the rest of us that like to squeeze every drop of performance out of both clients and server, this will give you something to play with. And it’s good practice for learning to optimize code which targets the most widely used virtual-machine in the world, namely the browser. A technology which eventually will take over the world and replace most of our native systems. We are a few years ahead of our time with SMS, but hey- world domination is hard work!
function AllocMem(const Size:Integer):TMemoryHandle; begin if Size>0 then Result:=new JUInt8ClampedArray(Size); end; procedure Freemem(var Memory:TMemoryHandle); begin if (memory) then begin // decouple buffer from type // this does not release memory, but "hints" to the GC // to mark the segment for level 1 release classification JUInt8ClampedArray(Memory).buffer := NIL; Memory := null; end; end; function ReAllocMem(Memory:TMemoryHandle; Size:Integer):TMemoryHandle; begin if (Memory) then begin if Size>0 then begin if Memory.length > Size then Memory:=JUInt8ClampedArray(memory).SubArray(0,Size-1); result:=new JUInt8ClampedArray(Size); JUInt8ClampedArray(result).Set(JUInt8ClampedArray(Memory),0); end; end else result:=AllocMem(Size); end; function WriteMem(const Memory:TMemoryHandle; const Offset:Integer; const Data:TMemoryHandle):Integer; begin if (Memory) then begin if (Data) then begin if offset + data.length-1 > memory.length then result:=(offset + data.length-1) - memory.length else result:=data.length; if offset + Data.length > memory.length then JUInt8ClampedArray(Memory).Set( new JUInt8ClampedArray (JTypedArray( JUInt8ClampedArray(data).buffer.Slice(0,result-1))), offset) else JUInt8ClampedArray(Memory).Set(JTypedArray(data),offset); end; end; end; function ReadMem(const Memory:TMemoryHandle; const Offset:Integer; Size:Integer):TMemoryHandle; begin if (Memory) then begin if Offset>=0 then begin if offset + Size-1 > Memory.length then dec(Size,(offset + Size-1)-Memory.length); result:=new JUInt8ClampedArray(JTypedArray( JUInt8ClampedArray(Memory).buffer.Slice(Offset,Offset + Size))); end; end; end;