Which Endian routine is better?
in [Visual Basic]
|
Prev: VB 6 & VS?
Next: C:\WINDOWS\system32\ieframe.dll\1
From: Larry Serflaten on 13 Mar 2010 11:23 "MM" <kylix_is(a)yahoo.co.uk> wrote > No, it's not. It's even faster than SwapEndian08. Here's my latest > attempt: > > Function Endian32_UsingShift(ByVal n As Long) As Long > > Endian32_UsingShift = ((ShiftLL(n And &HFF&, 24)) _ > Or (ShiftLL(n And &HFF00&, 8)) _ > Or (ShiftLR(n And &HFF0000, 8)) _ > Or (ShiftLR(n And &HFF000000, 24))) > > End Function > Something else to try.... Function EndianSwap(Value as long) As Long Dim tmp as long tmp = ShiftLL(Value And &HFF00FF&, 8) Or ShiftLR(Value And &HFF00FF00, 8) EndianSwap = ShiftLL(tmp, 16) Or ShiftLR(tmp, 16) End Function LFS
From: Mike Williams on 13 Mar 2010 11:53 "MM" <kylix_is(a)yahoo.co.uk> wrote in message news:4m9np5tsba87an9ns7volhm0ke112gl46h(a)4ax.com... > Here's my latest attempt: > Function Endian32_UsingShift(ByVal n As Long) As Long > Endian32_UsingShift = ((ShiftLL(n And &HFF&, 24)) _ > Or (ShiftLL(n And &HFF00&, 8)) _ > Or (ShiftLR(n And &HFF0000, 8)) _ > Or (ShiftLR(n And &HFF000000, 24))) > End Function > returns 18 secs for 1 million iterations on my Sempron > 2.8 on Windows 98 SE. ?? 18 secs for 1 million ?? Presumably you mean 18 milliseconds? > However, SwapEndian08 returns 19 secs. > The shifts are from Crescent's QuickPak Professional. I think to get the ultimate speed you're going to need to perform the whole job (the shifts and the ORs / ANDs) in the same machine code routine so that you call the function just once rather than calling a shift function four times and performing the ANDs and ORs in VB. Maybe someone with up to date ASM experience could write such a routine? Mike
From: Jim Mack on 13 Mar 2010 12:41 Mike Williams wrote: > > ?? 18 secs for 1 million ?? Presumably you mean 18 milliseconds? Probably not. It's guaranteed to be slower, but not by that much. There's simply no way that making 4 external ASM calls, especially to a Declared function, is going to beat optimized, compiled VB6. I don't believe it. The reason I don't believe it is that I'm an ASM expert myself, and sell a library that competed with QuickPak Pro (Ethan and I are old friends). In ASM (386+), you can perform this operation with two instructions that literally take nanoseconds to execute: Endian4 PROC PUBLIC mov eax, [esp+4] bswap eax ret 4 Endian4 ENDP From VB, then: Swapped& = Endian4(Unswapped&) When we had this exact discussion here a couple of years ago, I placed that function into a DLL, using a Typelib, and others here timed it against the vbspeed code. It lost. The reason is that VB jumps through a lot of hoops to make an external call. For a Declared function, it must also call SetLastDllError internally on each call. So even though the code itself is perfectly optimal, it can't compete with inline code optimized by the VB compiler. If there's interest I'll do it again, making the DLL available for anyone to DL and test. And if anyone here has our Stamina library, try a call to DxEndian(unswapped&, 4) and tell us how it fares. -- Jim Mack MicroDexterity Inc www.microdexterity.com
From: MM on 13 Mar 2010 13:17 On Sat, 13 Mar 2010 12:41:10 -0500, "Jim Mack" <jmack(a)mdxi.nospam.com> wrote: >Mike Williams wrote: >> >> ?? 18 secs for 1 million ?? Presumably you mean 18 milliseconds? > >Probably not. It's guaranteed to be slower, but not by that much. > >There's simply no way that making 4 external ASM calls, especially to >a Declared function, is going to beat optimized, compiled VB6. I don't >believe it. I did it. It does. MM
From: Jim Mack on 13 Mar 2010 14:15
MM wrote: > On Sat, 13 Mar 2010 12:41:10 -0500, "Jim Mack" > <jmack(a)mdxi.nospam.com> wrote: > >> Mike Williams wrote: >>> >>> ?? 18 secs for 1 million ?? Presumably you mean 18 milliseconds? >> >> Probably not. It's guaranteed to be slower, but not by that much. >> >> There's simply no way that making 4 external ASM calls, especially >> to a Declared function, is going to beat optimized, compiled VB6. >> I don't believe it. > > I did it. It does. Are you doing this compiled to native code with all optimizations? Just for comparison, try this: http://www.microdexterity.com/other/bendian.dll Read the text file first. I don't know what's different this time, but I get a 40% speed advantage using the ASM code over the Swap08 code. The last time we tried this people reported a 10% disadvantage IIRC. -- Jim |