Sega Genesis Dev Kit (SGDK)

Chilly Willy · Post by **Chilly Willy** » Thu Sep 20, 2012 4:35 pm

KanedaFr wrote:little bug on the joy support :
readJoypad return joy state and type
unfortunatly type is also used when it calls joyEventCB
so change is not 0 when released but type << 12 ...
I would prefer to keep the initial mode : 0 for release, button for press

The "backwards compatibility" has its limits... for controllers, there is now a mask you should use for just the buttons - use (val & BUTTON_ALL) instead of plain old (val). If you force the support for pads to three button pad mode (for example, JOY_setSupport(PORT_1, JOY_SUPPORT_3BTN); ), then you'll have complete backwards compatibility since the TYPE for 3 button pads is 0.

Stef · Post by **Stef** » Thu Sep 20, 2012 8:16 pm

Well that is not a big deal anyway, i separated type and state because i think it is less confusing this way too

Now we have :

Code: Select all

u8 JOY_getJoypadType(u16 joy);

which return type for the specified joypad and the readJoypadState(..) does not return anymore joypad type in high nibble.

Chilly Willy · Post by **Chilly Willy** » Thu Sep 20, 2012 9:58 pm

I assume you changed the example, too, right?

Stef · Post by **Stef** » Thu Sep 20, 2012 10:01 pm

djcouchycouch · Post by **djcouchycouch** » Sat Sep 22, 2012 4:28 pm

I'm seeing what looks like to be a bug with fastMemsetU16.

Code: Select all


#define MAX_COINS_IN_LEVEL 256
u16 coinMap[MAX_COINS_IN_LEVEL];

fastMemsetU16((u16*)coinMap, 1, MAX_COINS_IN_LEVEL);

PrintNumber(coinMap[0]); // helper printing function 
PrintNumber(coinMap[1]);
PrintNumber(coinMap[2]);

Output (values are in hexidecimal):

Code: Select all

Message : 00000001 
Message : 00000101
Message : 00000101

switching to memsetU16 gives the correct output:

Code: Select all

Message : 00000001
Message : 00000001
Message : 00000001

I kept wondering why my coins were all screwed up

DJCC

Chilly Willy · Post by **Chilly Willy** » Sat Sep 22, 2012 5:00 pm

Change line 347 of memory.c from

Code: Select all

    value32 = cnv_8to32_tab[value];

to

Code: Select all

    asm("move.w %0,%1\n\t"
        "swap %1\n\t"
        "move.w %0,%1\n"
        : "=d" (value),  "=d" (value32) : : "cc" );

djcouchycouch · Post by **djcouchycouch** » Sat Sep 22, 2012 5:06 pm

Tried it out and now it's giving me:

Code: Select all

Message : 00000001 
Message : 00000000 
Message : 00000000

Chilly Willy · Post by **Chilly Willy** » Sat Sep 22, 2012 7:18 pm

A first word can be written by itself for odd word addresses, then it tries writing longs. So it seems like the asm isn't working... okay, try this:

Code: Select all

    asm("move.w %1,%0\n\t"
        "swap %0\n\t"
        "move.w %1,%0\n"
        : "=d" (value32) : "=d" (value) : "cc" );

Had value as an output register, which was probably the problem.

djcouchycouch · Post by **djcouchycouch** » Sat Sep 22, 2012 7:22 pm

That doesn't compile.

/sgdk092/src/memory.c:356: error: input operand constraint contains `='

Chilly Willy · Post by **Chilly Willy** » Sat Sep 22, 2012 10:05 pm

djcouchycouch wrote:That doesn't compile.

/sgdk092/src/memory.c:356: error: input operand constraint contains `='

Get rid of the second =... just use "d" (value). The = is for indicating write access, if I remember correctly.

djcouchycouch · Post by **djcouchycouch** » Sat Sep 22, 2012 10:17 pm

That seems to work. Thanks!

Chilly Willy · Post by **Chilly Willy** » Sun Sep 23, 2012 12:20 am

djcouchycouch wrote:That seems to work. Thanks!

No problem. Glad it finally worked out. Pulling inline assembly off the top of my head is fun, but takes a little debugging. I always forget which field is inputs and which are the outputs... things like that.

Stef · Post by **Stef** » Sun Sep 23, 2012 3:08 pm

Very stupid bug, to keep all C you just need to change the

Code: Select all

value32 = cnv_8to32_tab[value];

by

Code: Select all

value32 = (value << 16) | value;

but asm part is faster as you can use the swap instruction.

Chilly Willy · Post by **Chilly Willy** » Sun Sep 23, 2012 6:53 pm

Stef wrote:Very stupid bug, to keep all C you just need to change the
Code: Select all
value32 = cnv_8to32_tab[value];
by
Code: Select all
value32 = (value << 16) | value;
but asm part is faster as you can use the swap instruction.

I was fairly certain << 16 wasn't going to use swap, which is why I recommended the asm. If it didn't use swap, it wouldn't be "fast" at that point, and it was supposed to be the fast version of the routine.

Stef · Post by **Stef** » Sun Sep 23, 2012 7:22 pm

Chilly Willy wrote: I was fairly certain << 16 wasn't going to use swap, which is why I recommended the asm. If it didn't use swap, it wouldn't be "fast" at that point, and it was supposed to be the fast version of the routine.

Hehe is was sure that it wouldn't too but in fact it does use the swap instruction ! But the rest of the code is really not the best we can do so i finally rewrote all these methods in pure asm.
I made fastxxx deprecated as now the normal memset and memcpy methods use fast asm code now