Inline ASM and multithreading

[Blitz Amateur]

Well, I liked MrP's way of using ASM to plot a pixel, so I adapted it. I also decided I wanted to try multithreading.

The multithreading is disgustingly faster here, however, the program doesn't wair for old threads to close before opeinig new ones. So that is a potential leak there. No promises your system won't lock up =/ ( Just for reassurance though, it works beautifully here )

To use the multithreading, change the "mode" variable to 1

Code Select Expand

#define PTC_WIN
#include "tinyptc.bi"

type threadstack
    thread as integer
    nextthread as threadstack pointer
end type


declare sub putpixel(byval params as integer pointer)
declare sub DrawScreen(byval off as integer)
declare sub DrawScreen2()

dim shared as integer pointer buffer
dim shared as integer threadlock
dim shared as threadstack pointer currentthread

dim shared as integer params(3)

dim as byte mode = 0 ' Setting mode to 1 uses multithreading

threadlock = MutexCreate()


buffer = callocate(640*480,Len(Integer))

ptc_open("test",640,480)
dim as integer t
dim as double ttime
ttime = timer
while inkey$=""
ptc_update(buffer)
if mode = 1 then
t = threadcreate (@DrawScreen,0)
t2 = threadcreate (@DrawScreen,1)
else
DrawScreen2
endif
print 1.0/(timer-ttime)
ttime = timer

wend

sub DrawScreen(byval off as integer)
    mutexlock threadlock
    dim as integer pointer params
    params = callocate(3,Len(integer))
    for x=off to 639 step 2
        for y=0 to 479
            params[0] = x
            params[1] = y
            params[2] = &hFF0000
            putpixel(params)
        next
    next
    mutexunlock threadlock
end sub

sub DrawScreen2()
    dim as integer pointer params
    params = callocate(3,Len(integer))
    for x=off to 639
        for y=0 to 479
            params[0] = x
            params[1] = y
            params[2] = &hFF0000
            putpixel(params)
        next
    next
end sub

sub PutPixel(byval params as integer pointer)
    dim as integer x,y,colour
    x=params[0]
    y=params[1]
    colour=params[2]

    asm
        mov eax, [x]
        mov ebx, [y]
        mov ecx, [colour]
        mov edx, [buffer]
        imul ebx, 640
        add eax, ebx
        lea edx, [edx+eax*4]
        mov [edx], ecx
    end asm
end sub

[Jim]

That's not quite right.Ã,  I think you're trying to write the even pixels in one thread and the odd pixels in the other, and then flip the screen, but that's not what you're doing.Ã,  Because you have the mutex in DrawScreen, then only one thread at a time is running (only 1 thread can lock the mutex at a time).Ã,  You should take the mutex lock out.Ã,  Also, since you don't wait for either thread to finish, then the frame rate is only counting how long it takes to create two threads, not how long it takes for each of the threads to finish the job.Ã,  Try changing pixel colours to see what's going on.

To fix it you should add

Code Select Expand


threadwait(t)
threadwait(t2)

to wait for the threads to finish, and remove the two mutexlock/mutexunlock lines.

As long as you have at least Hyperthreading on your CPU you should see some speed-up, maybe up to 50%.Ã,  I did some tests with a mandelbrot generator on my HT and it's amazing the speed difference, but with that it's pure computation and no memory accesses, which is why you won't see that much of a speedup.

You're also leaking the data you callocate for each call.Ã,  Make sure you use the -mt option when you compile else the two callocates going off at the same time might interfere with each other -Ã,  you need a thread-safe runtime library for it to work reliably.

Jim

[Blitz Amateur]

Thanks for the help, I changed a couple things and relized that with all of the memory access, it's actually quite a bit slower on my system with the muilt-threading. This being a first attempt and all, It was destined to not work quite right

[Jim]

Maybe if you change one thread to do the top half of the screen and the other half to do the bottom it will be better?
Anyway, because of all the computation, there's definitely an opportnuity to use multithreading to speed up your water sim.  I'd like to see that.

Jim