SDL2

OpenGL context seemed to work, well at least for one colourful triangle. Currently it's 16-bit similar to SDL1 but we can change that.

Great!

@AmigaBlitter

I have fixed issue in framebuffer that caused stripes similar to picture in post #58. Please update your codebase and retest.

There are also a couple of open tickets on project page.

https://github.com/segrax/openfodder Open Fodder, it uses SDL2. Perhaps portable some day :)

@AmigaBlitter

Regarding "test" directory. I think sdl2-config has problem in native env - or I just can't use it properly. Anyway, after setting prefix as /sdk/local/newlib, as with sdl(1)-config, it was possible to configure tests. This assumes that you have SDK:local/newlib/include/SDL2 and libs in SDK:local/newlib/lib.

Then, I had to hack sdl2-config to add -use-dynld -ldl, to make tests link in my own environment.

Please note that you need to copy also build/SDL2_test.a into SDK:local/newlib/lib/.

I believe there has to be some configure script modifications to fix sdl2-config for natives.

Oh, and BMP loading is broken, image tests didn't work. I will open a ticket.

Comments allowed...

@Capehill

thank you for this report.

Downloading now new version to do some tests.

@AmigaBlitter

BMP loading should work now after two workarounds: 1) define byte order in SDL_endian.h 2) use fopen64 instead of fopen, to make fseeko64 work.

Please try r49.

@Capehill

ok

thank you


reuild the r49 :)

@AmigaBlitter

Fixed a few lines in the configure files.

Now the test's suite compile properly and the library is installed ok.

@AmigaBlitter

With r51 audio should work.

@Capehill

R51 seems ok.

Strange ?bug? in configure? I can't disable the altivec by passing the parameter --disable-altivec, so a little hack is needed.

@Capehill

Fixed the configure issue by forcing the --disable-altivec with a var.

Capehill pointed out that testsprite demo doesn't properly quit the SDL. This causes non freed signal that lead to crash.

Verified that many demo, actually, doesn't use the SDL_CommonQuit() and SDL_CommonState. This will be fixed asap.


host immagini


image hosting



image share

@AmigaBlitter

After configuring with --disable-altivec, "testrendercopyex" works with Sam440. If there is some AltiVec expert lurking, perhaps (s)he could help to understand why GCC generates some vector instructions for RenderCopyEx function, causing illegal instruction.

@AmigaBlitter

Fixed "testspriteminimal" and "testdrawchessboard". Neither did any attempt of cleaning up after test. There are probably more :(

@Capehill

Thank you.

Will test new version tomorrow

@Capehill

Sound correct to me, you need AltiVec to use Altivec, CPU's without AltiVec like Sam440, will get a illegal instruction, the same thing if you use instructions that are only available on AMCC440, like ISEL this won't work on CPU's without ISEL opcode.

This way there is two versions of Mplayer, one with AltiVec and one without. It’s a hazel but it might worth it.

be sure that no -maltivec switch is passed to GCC otherwise the compiler will generate optimized altivec function even if no altivec inline instructions are present into the code

@AmigaBlitter

Quote:

This is quite common in SDL programming as far as I have experienced. makes me think that other SDL implmentations will clean up for the lazy prgrammer on exit().

@thread

Noticed that using the -mcpu=440 flag produces strange results.

@afxgroup

Has anybody tested AltiVec blitters? Are they worth dual-building?

Or shall we just forget about it and use OpenGL (and compositing) where available?