How is array-access compiled?
BlitzMax Forums/BlitzMax Beginners Area/How is array-access compiled?
| ||
I am just wondering, if you set up an array for example...Local myarray:Int[50]and then you want to access it with... For Local i:int=0 to 49 Print myarray[i] Nextdoes BlitzMax compile this down to being `hardcoded` to access that array in 4-byte increments? Or does it have to check the type of the array on every access and multiple that size by the offset (plus the base) to get to the data? Or is it more hardcoded - ie it multiplies i by 4 and adds it to the base address then reads? Also, would it be faster then to use an array of Bytes rather than an array of Ints, so that the multiplication (or left shift) to jump to the appropriate location can be eliminated? |
| ||
| When you compile this code, BMax generates an assembler-code file. As far as I can tell (and as far as my knowledge goes for assembler), it seems that the type of the array is analysed during runtime. I've compiled this code: FrameWork BRL.Basic Local myarray:Int[50] This gives the smallest code. Inside this assembler-code file (untitled1.bmx.d.s), there are only 2 lines that change, when you change the type of the array from Int to Byte: _17 db "[]i",0 _11 db "i",0 Those change to: _17 db "[]b",0 _11 db "b",0 When the For Next loop is included, this block is changed when it's an Int-array: _21: mov eax,dword [ebp-4] movzx eax,byte [eax+ebx+24] mov eax,eax push eax call _bbStringFromInt add esp,4 push eax call _brl_standardio_Print add esp,4 To _21: mov eax,dword [ebp-4] push dword [eax+ebx*4+24] call _bbStringFromInt add esp,4 push eax call _brl_standardio_Print add esp,4 For printing the contents of the array, it seems that it jumps 4 bytes (hardcoded) for an Int and only 1 byte for a Byte array. |
| ||
| Hmm ok, so maybe the bytes are a bit faster then. Having said that, on a PowerPC most instructions run in 1 cycle so even if it is a more complex instruction like including the *4, maybe it makes no difference. |
| ||
| I think you won't notice the difference in access-time. Maybe when you're about to read from that array for about 1.000.000 times per frame and you want to have your game running at 50fps, maybe then you will see the difference. My nickname may suggest that I own a PowerPC processor, but I don't. In my Amiga days, I always dreamed of buying the BlizzardPPC 1260 turboboard (with 68060 AND PPC 603e), that's where my nickname comes from. But that's another story... :-) |
| ||
| Well, conservative estimate, I may need to call these function pointers, say, at least several thousand times per frame at a high frame rate. Re Amiga, yah I never got further than a 25Mhz 68040, and a PPC even at a few hundred Mhz seemed super fast. Now I'm enjoying a 300Mhz G3 and a 1GHz G4, which even though it's by no means top of the line, sure lets you do a lot more stuff. |
| ||
| If you're calling several thousand function pointers per frame, then some extra instructions in the calling sequence are not going to matter. They'll likely fall into holes in the pipeline anyway. They'll be dwarfed by cache issues and branch mispredictions...and what you're doing in each of the functions. |
| ||
| Internally there is no type checking. It just acceses the 4 bytes, gives you the value. In the case of an array of strings its just pointers to the strings. Same for objects, and shock and horrer, ARRAYS OF ARRAYS! :P |
| ||
| Re Amiga, yah I never got further than a 25Mhz 68040, and a PPC even at a few hundred Mhz seemed super fast. Now I'm enjoying a 300Mhz G3 and a 1GHz G4, which even though it's by no means top of the line, sure lets you do a lot more stuff. I started from an A500, then bought a A1200 (14MHz 68020) and installed a Blizzard 1260 (68060 50MHz) with 32Mb RAM + SCSI-II controller. I could open DirectoryOpus (kinda like Windows Explorer) about 9 times, before running out of graphical memory (2Mb) on a res of 1024x256 (couldn't use 1024x512, because of the flickering). Try that on a pc. Back then I had also a pc (Pentium MMX 180MHz) and it seemed that with some programs, my Amiga was faster. Now I've got a P4 2.4GHz and my Amiga is only used for scanning pictures with my flatbad scanner (+/- once a year). But back to the topic: Those lines, that hold the "[]b" and "[]i", for what purpose are they used then? Is it just to indicate the type of pointer (to indicate that the arrays points to bytes or integer objects)? |
| ||
| Oh. Started with an A500, got a 1/2 meg upgrade, then a 2meg upgrade, then an A1200, then an Apollo A1240 (i think), then put it all in a tower with a graphics card. Yes Amiga's were cool, but in terms of gfx power or resolution most modern systems are faster now. |
| ||
| 1024x256.. lol squashed?:P |
| ||
| Yah, that sounds like `superhires`... and it wasnt 1024 it was 1280x256. or 1280x512 with interlacing. Actually, A1200 could do 640x512 non-interlaced I think |
| ||
| Remember you scroll screens sideways in AmyOS so he could have set his screensize to any res irrespective of the actual resolution. I constantly used a multiscan screenmode that was 2 or 3 screens in width... |
   |