It's on the tip of my tongue.. until then, try searching in the DirectX mailing list archive: http://discuss.microsoft.com/archives/DIRECTXDEV.html

Some cards can accelerate that function, but not all.

It's always best to use as much DirectX in your DirectX programs as possible as new cards add new features all the time. By doing so, you get at least a limited degree of support without recompiling.

You should be able to check if your card supports sys to vram acceleration by using the DirectX properties program (for DX5.0 anyway. I'm sure DX6.0 has some form of this as well).

Also, you may want to check to see if your memcpy() copies DWORDs, WORDs, or BYTEs at a time. Most compilers use DWORDs for 32 bit apps, but I think some don't. The only other thing I can think of is if you're copying unaligned data, DirectX MAY do some dynamic ordering to at least TRY to setup aligned copying, which if it did would beat out your unaligned memcpy. Always try to keep mem movement aligned if possible.

Rock

if you want to use memcpy() like fn to do the blt, use mmx. it'll allow you to move qwords and *might* be a bit faster than dd's bltfast.

I say just stick with the directx bltfast or blt statment. Alot of hardware ive seen accelerates it and your not likly to beat that. Plus its genraly going to work on any system with a directx compatable driver unlike some of the other routines i have seen.

Thanks for all the replies guys. I wasn't considering not using DirectX for the blit, since it's so much faster, I was just wondering why this was happening. Also I am considering allowing the user to select either DirectX blitting or regular memcpy (I hear memcpy is noticeably faster than the HEL implementation of BltFast sys->vid). Anyway I will have to give it a go on an older video card and see what difference it makes.

As Rock suggested I checked out the DX Device Viewer tool, which tells me that Sys->Vid memory Blt's are accelerated. I was also wondering (technically) how this would be accelerated, since the bus is limited to a certain speed anyway. If anyone has any thoughts there I'd be interested to hear about it as I don't know a lot about video hardware acceleration/capabilities.

Also thanks for the suggestion on using MMX Queasy, I'm going to look into that.

Phoenix

I'm just pondering some unusual results I got when testing a project. When blitting from a 480x480 physical memory buffer into the back buffer (in video memory), I get better results using BltFast than memcpy'ing it line by line, contrary to what I have been told. Can the video card accelerate this operation (I didn't think it could, but this makes me think it must be). Using memcpy the speed is around half as slow... any ideas? Thanks!

Phoenix

Few references from DirectX SDK Help:

"The IDirectDraw4::GetCaps method fills in the capabilities of the device driver for the hardware and the hardware-emulation layer (HEL)."

From the filled DDCAPS structure check if dwCaps has DDCAPS_CANBLTSYSMEM 'flag checked' or something

"DDCAPS_CANBLTSYSMEM
Indicates that display hardware is capable of blitting to or from system memory."

You could check that and then use preferred blitting method, am I right?

------------------
"Tank, I need an exit. Fast!"