Hi all,
Using IDL_IDLBridge has been very helpful in some of our work lately, helping us
to get more processing done while one IDL process is waiting on a DLL, and other
fun. I thought I'd share some findings that show how results can come back 4-10
times faster using shared memory (SHMMAP, etc.) rather than GetVar(). Any
comments on this are very welcome, and I'd be interested to see how this plays
out on other hardware.
Notes:
I split out the SHM Cleanup timing since some applications may not require
cleanup between successive calls.
I thought it a bit odd to have to do this when assigning the value on the bridge
process:
oBridge -> Execute, "shmA[0,0,0,0] = a"
... but if you use [0] then only the first row of values is filled in. Sure,
it's faster, but... :-) (and if you use [0,0] only the first 2D plane is filled
in, etc.) Here's my !Version:
{ x86 Win32 Windows Microsoft Windows 6.3 Mar 23 2006 32 64}
I'll send another posting with the .pro file as an attachment, in case that's
more convenient for some to use.
Cheers,
-Dick
--
Dick Jackson Software Consulting http://www.d-jackson.com
Victoria, BC, Canada +1-250-220-6117 dick@d-jackson.com
=====
PRO SHMvsGetVar
;; Using an IDL_IDLBridge process, test timings of getting results back
;; using shared memory (SHM) vs. oBridge -> GetVar()
;;
;; SHM appears to be four to ten times faster in my testing.
;; Dick Jackson - dick@d-jackson.com
oBridge = Obj_New('IDL_IDLBridge')
shmName = 'SHMvsGetVar'
;; Describe sizes of arrays to use for testing (as strings)
sizes = ['10', '1E7', '[10,1E6]', '[1E6,10]', '[1E4,1E3]', '[1E2,1E2,1E3]']
;; '1E8','[10,1E7]','[1E7,10]','[1E4,1E4]','[1E3,1E2,1E3]','[1E 2,1E2,1E2,1E2]']
nTests = N_Elements(sizes)
FOR testI=0, nTests-1 DO BEGIN
sizeStr = sizes[testI]
ok = Execute('sizeNum = Long('+sizeStr+')')
IF ~ok THEN Message, 'Execute() failed: check sizeStr "'+sizeStr+'"'
Print, 'Testing with byte array of size: '+sizeStr
oBridge -> Execute, 'a=BIndGen('+sizeStr+')'
localA = BIndGen(sizeNum)
;; Test SHM method
t0 = SysTime(/Seconds)
oBridge -> Execute, "SHMMap,'"+shmName+"',/Byte,Dimension="+sizeStr
oBridge -> Execute, "shmA = SHMVar('"+shmName+"')"
oBridge -> Execute, "shmA[0,0,0,0] = a" ; Must have N_Dims(a) or more (bug?)
SHMMap, shmName,/Byte,Dimension=sizeNum
shmA = SHMVar(shmName)
shmTime = SysTime(/Seconds)-t0
Print, Format="(' SHM: ',F0.3,' s')",shmTime
;; Check result
IF ~(Array_Equal(Size(shmA), Size(localA)) && $
Array_Equal(shmA, localA)) THEN Print, ' *** Result check failed!'
;; Remove references to mapped variables and unmap memory segments
t0 = SysTime(/Seconds)
shmA = 0B
SHMUnmap, shmName
oBridge -> Execute, "shmA = 0B"
oBridge -> Execute, "SHMUnmap,'"+shmName+"'"
Print, Format="(' SHM+Cleanup:',F0.3,' s')",shmTime+(SysTime(/Seconds)-t0)
;; Test GetVar method
t0 = SysTime(/Seconds)
getVarA = oBridge -> GetVar('a')
Print, Format="(' GetVar: ',F0.3,' s')",SysTime(/Seconds)-t0
;; Check result
IF ~(Array_Equal(Size(getVarA), Size(localA)) && $
Array_Equal(getVarA, localA)) THEN Print, ' *** Result check failed!'
Wait, 0.001 ; To allow Print statements to flush
ENDFOR
Obj_Destroy, oBridge
END
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
