OS4 compatible UAE expansion development status

[Toni Wilen]

Originally Posted by Aegis

Ok, I'll speak to Chris (NetSurf dev) 'cause I'm sure he'll be curious about NetSurf bugging out whilst OWB/Odyssey are working - maybe it's easier for him to fix it his end.

I don't think it helps because there is TCP/IP stack (bsdsocket.library) between the driver (uaenet.device) and application (netsurf) and it probably is more or less a black box.

Totally different network layers. (ethernet frames vs socket api calls) Can't really know if some TCP stream packet was lost and stack kept waiting for it and finally returned error or something else.

[Toni Wilen]

OS4 question: How to call TrackSegmentList() from m68k code? It would hopefully help OS4 to show better crash information if something goes wrong in UAE boot ROM code.

[psykicko]

I got this from a person:

If you can't call them via the nromal DOSBase ptr they are probably PPC only.
Some function are PPC only, I don't know if that's one of them or not. but being V50 and above it's possible.

But he told me also i should ask on Hyperion forum. Do you want me to ?

[thellier]

It is possible as there is way to call ppc .libray from 68k
code
I will give an example monday as i dont have my pc here

I think aminet/wazp3d got those glue functions in his sources
It is called something 68k_to_ppc

Alain Thellier - Wazp3D

[Toni Wilen]

I was told by OS4 developer that there is no other way than to LoadSeg() the file, all other routines will not mark the segment list as JIT enabled.

So.. Executable it is (mostly empty dummy exe, all the code is in boot ROM).

winuae.7z updated.

Attached is exe that does the boot rom "remapping". Most commonly used routines are relocated (trap system, directory filesystem handler, uaegfx).

If executable returns (without any messages): something failed. If it does not return (without any messages): it is active. winuaelog.txt also shows RAM base address and some other information. Must be run from CLI. Does not (yet) affect UAE expansions that are initialized after uaebootmaprom has been run = do not run from startup-sequence.

Question is: Can anyone notice any performance differences? (with vs without running uaebootmaprom)

[radzik]

Someone has tested emulated SB128 sound card with AOS 4.1 (FE)? I have installed drivers, but in AHI preferences, when I test a sound I get only noise. On Paula I have normal test sound, but on the SB128 (drivers from OS4Depot) no. There are problems with PCI DMA? I'm emulating A4000PPC with Mediator 4000 MkII. All is under stable WinUAE 3.2 version.

[MickJT]

I've tested SB128 using Prometheus and also GREX. Works fine here.

[MickJT]

Started up WinUAE. Ran uaebootmaprom from Work: in the Shell, and then WinUAE crashed (no dump file).

[Snake79]

Originally Posted by radzik

Someone has tested emulated SB128 sound card with AOS 4.1 (FE)?

Cześć Radzik I'm using similar config (attached) and it works fine with AOS4.1FE (on latest betas and 3.2.2). Pozdrawiam.
PS. You are doing great job with podcast of yours, keep it up

[Toni Wilen]

Originally Posted by MickJT

Started up WinUAE. Ran uaebootmaprom from Work: in the Shell, and then WinUAE crashed (no dump file).

Is it repeatable? )(try at least 3 times).

Originally Posted by radzik

Someone has tested emulated SB128 sound card with AOS 4.1 (FE)? I have installed drivers, but in AHI preferences, when I test a sound I get only noise. On Paula I have normal test sound, but on the SB128 (drivers from OS4Depot) no. There are problems with PCI DMA? I'm emulating A4000PPC with Mediator 4000 MkII. All is under stable WinUAE 3.2 version.

Switch on "Full PCI DMA" in PCI bridge options.

Real hardware does not support PCI DMA: emulated PCI DMA is not enabled by default.

(btw, this does not have anything to do with UAE expansions!)

[MickJT]

Originally Posted by Toni Wilen

Is it repeatable? )(try at least 3 times)

Unfortunately, yes.

I've also changed Accelerator SCSI to UAE now.

It also randomly freezes. I have a dump file but you said earlier that they're useless on these betas.

[Toni Wilen]

Originally Posted by MickJT

Unfortunately, yes.

I've also changed Accelerator SCSI to UAE now.

It also randomly freezes. I have a dump file but you said earlier that they're useless on these betas.

Does it work if you switch off JIT? (Which is useless in PPC-only CPU emulation) JIT usually only adds more noise to logs..

[MickJT]

Seeing as it's useless in PPC mode, I didn't think having JIT on would cause any problems, but nevertheless now uaebootmaprom runs without crashing WinUAE. Thanks for the tip.

Nothing immediately jumps out at me in terms of any speed difference, but I'll let others decide on that.

Edit: Still have random freezes.
Edit2: And graphical glitches on uaegfx (32bit / "All" enabled in every cycle gadget that has that option). Today is the first time I've tried one of these expansion related betas, so I haven't been running it long enough to know if these glitches occur without uaebootmaprom running, yet.

[Snake79]

Originally Posted by Toni Wilen

Question is: Can anyone notice any performance differences? (with vs without running uaebootmaprom)

Don't know if i'm getting the idea right, but have some test results (with updated winuae.7z)...
PS. sorry for upper offtopic...

[Toni Wilen]

Originally Posted by MickJT

Seeing as it's useless in PPC mode, I didn't think having JIT on would cause any problems, but nevertheless now uaebootmaprom runs without crashing WinUAE. Thanks for the tip.

Fixed.

JIT adds extra stuff here and there, including hardware level memory write protection.

Edit: Still have random freezes.

Can't do anything until someone narrows it down to something specific.

Edit2: And graphical glitches on uaegfx (32bit / "All" enabled in every cycle gadget that has that option). Today is the first time I've tried one of these expansion related betas, so I haven't been running it long enough to know if these glitches occur without uaebootmaprom running, yet.

Will be ignored because there is no real hardware Zorro board that has 32-bit mode to confirm it. It could be OS4 classic bug/feature.

[Aegis]

Zero difference in Quake - same fps without and with uaebootmaprom + latest beta.

[Toni Wilen]

Originally Posted by Aegis

Zero difference in Quake - same fps without and with uaebootmaprom + latest beta.

It probably does not affect Quake too much because it is mainly direct writes to VRAM. (Previous speed increase most likely came from removal of interrupt/task context switches in trap system)

Some graphics graphics micro benchmark probably should some differences, like something that writes lots of text strings or moves windows around etc..

EDIT: uaehf.device and directory filesystem disk speed test should also show some differences.

[Snake79]

Have in mind that i am still not sure if i am doing this right (using uaebootmaprom)... CgxBenchmark results in 2 tries (don't know if that help in something)...
Left (results in a screenshot) are before and right (in the same screenshot) are after using uaebootmaprom.
PS. No JIT (today updated winuae.7z).

[Aegis]

Ragemem benchmark:

[thellier]

>How to call TrackSegmentList() from m68k code

There are some .library with both a ppc jump table and a 68k jump table

You can generate such a 68k jump table with fdtrans from an .sfd file

fdtrans dos.sfd --all


An sfd contain all the functions description with 68k register usage

I have made a dos.sfd with only your function :
--------------------------------------------------------------------------------
==id $Id: Dos_lib.sfd,v 1.0 2012/10/12 13:30:00 noname Exp $
==base _DosBase
==basetype struct Library *
==libname Dos.library
==bias 30
==public
==include <exec/types.h>

LONG TrackSegmentList( CONST_STRPTR name, BPTR segment, CONST APTR extra_info, LONG extra_info_size ) (a0,d0,a1,d1)

==end
--------------------------------------------------------------------------------
It give that
--------------------------------------------------------------------------------
/*
** This file was automatically generated by fdtrans 52.1.
** Do not edit it by hand. Instead, edit the sfd file
** that was used to generate this file
*/

#ifdef __USE_INLINE__
#undef __USE_INLINE__
#endif
#ifndef __NOGLOBALIFACE__
#define __NOGLOBALIFACE__
#endif

#include <exec/interfaces.h>
#include <exec/libraries.h>
#include <exec/emulation.h>
#include <interfaces/exec.h>
#include <interfaces/dos.h>
#include <proto/dos.h>


static inline int8 convert_int8 (uint32 x) { return x; }
static inline int16 convert_int16(uint32 x) { return x; }


STATIC struct Library * stub_OpenPPC(ULONG *regarray)
{
struct Library *Base = (struct Library *) regarray[REG68K_A6/4];
struct ExtendedLibrary *ExtLib = (struct ExtendedLibrary *) ((ULONG)Base + Base->lib_PosSize);
struct LibraryManagerInterface *Self = (struct LibraryManagerInterface *) ExtLib->ILibrary;

return Self->Open(0);
}
STATIC CONST struct EmuTrap stub_Open = { TRAPINST, TRAPTYPE, (ULONG (*)(ULONG *))stub_OpenPPC };

STATIC APTR stub_ClosePPC(ULONG *regarray)
{
struct Library *Base = (struct Library *) regarray[REG68K_A6/4];
struct ExtendedLibrary *ExtLib = (struct ExtendedLibrary *) ((ULONG)Base + Base->lib_PosSize);
struct LibraryManagerInterface *Self = (struct LibraryManagerInterface *) ExtLib->ILibrary;

return Self->Close();
}
STATIC CONST struct EmuTrap stub_Close = { TRAPINST, TRAPTYPE, (ULONG (*)(ULONG *))stub_ClosePPC };

STATIC APTR stub_ExpungePPC(ULONG *regarray __attribute__((unused)))
{
return NULL;
}
STATIC CONST struct EmuTrap stub_Expunge = { TRAPINST, TRAPTYPE, (ULONG (*)(ULONG *))stub_ExpungePPC };

STATIC ULONG stub_ReservedPPC(ULONG *regarray __attribute__((unused)))
{
return 0UL;
}
STATIC CONST struct EmuTrap stub_Reserved = { TRAPINST, TRAPTYPE, stub_ReservedPPC };

static LONG stub_TrackSegmentListPPC(uint32 *regarray)
{
struct Library *Base = (struct Library *) regarray[REG68K_A6/4];
struct ExtendedLibrary *ExtLib = (struct ExtendedLibrary *) ((uint32)Base + Base->lib_PosSize);
struct DosIFace *Self = (struct DosIFace *) ExtLib->MainIFace;

return Self->TrackSegmentList(
(CONST_STRPTR)regarray[8],
(BPTR)regarray[0],
(CONST APTR)regarray[9],
(LONG)regarray[1]
);
}
STATIC CONST struct EmuTrap stub_TrackSegmentList = { TRAPINST, TRAPTYPE, (uint32 (*)(uint32 *))stub_TrackSegmentListPPC };

CONST CONST_APTR VecTable68K[] =
{
&stub_Open,
&stub_Close,
&stub_Expunge,
&stub_Reserved,
&stub_TrackSegmentList,
(CONST_APTR)-1
};

--------------------------------------------------------------------------------

Dont know if this helps in this case .....

Alain Thellier

BTW perhaps existing os4 dos.library can be called from 68k code if it (already) have such a 68k jump table