GFX_WRITE_VRAM_ADDR is expensive? (was: code benchmarking)

Chilly Willy · Post by **Chilly Willy** » Sat Mar 17, 2012 12:16 am

I'll try those two using 4.6.2 and see what I get. I might also try making it use short addressing on the ram to see how that affects the speed.

djcouchycouch · Post by **djcouchycouch** » Mon Mar 19, 2012 12:39 am

Hi,

Just wanted to let you guys know I tried out your suggestions and it's definitely helped. The benchmark numbers for that routine has dropped to 13045, a pretty dramatic improvement!.

Thanks again for your help!
DJCC

Stef · Post by **Stef** » Mon Mar 19, 2012 8:50 am

Chilly Willy wrote:I'll try those two using 4.6.2 and see what I get. I might also try making it use short addressing on the ram to see how that affects the speed.

Nice, i think it could be interesting

How can you force the compiler for short RAM addressing ? define RAM area at FFFF0000-FFFFFFFF is enough ?

Stef · Post by **Stef** » Mon Mar 19, 2012 8:51 am

djcouchycouch wrote:Hi,

Just wanted to let you guys know I tried out your suggestions and it's definitely helped. The benchmark numbers for that routine has dropped to 13045, a pretty dramatic improvement!.

Thanks again for your help!
DJCC

Glad you get it, it's a nice improvement indeed

Chilly Willy · Post by **Chilly Willy** » Mon Mar 19, 2012 5:07 pm

Stef wrote:
Chilly Willy wrote:I'll try those two using 4.6.2 and see what I get. I might also try making it use short addressing on the ram to see how that affects the speed.
Nice, i think it could be interesting How can you force the compiler for short RAM addressing ? define RAM area at FFFF0000-FFFFFFFF is enough ?

Yeah, just change the ram address in the linker script to 0xFFFF8000 and the length to 0x8000. I also change the stack to 0xFFFF8000 at the same time so the stack doesn't "waste" any of the short addressing space.

I did this once before, but I've forgotten if I used any switches with that. It's not something I normally do for MD homebrew.

If you use the built-in allocator, you could also change the linker script to set _end/__end to 0xFF0000 so that the heap gets put at the start of ram; then you could malloc up to the stack.