FS-UAE for OS X PPC (Not officially supported)

[tlosm]

Hi mike,
i will check it on X5000 and report

[SnakeCoils]

Okay, since I am on vacation now I have a bit more spare time to experiment.
I have found which part of FS-UAE is to blame because of empty menu text and missing drive sounds: it is the fs-uae.dat file in FS-UAE/Contents/MacOS in fact if I replace the one found in 2.7.14 with the fs-uae.dat from 2.5.20 (the last I know to work) everything works as expected again.
If I remember well this issue has already occoured in the past and then was fixed but now has surfaced again.

[SnakeCoils]

Latest FS-UAE 2.7.15dev for OSX PPC is up in The Zone! and ready for download.
This version is still missing the OSD menu and messages because the executable seems to dislike the fs-uae.dat file inside the App bundle. If replaced with the same from 2.5.20 everything start to work as expected :-) Maybe some later modifications to the fs-uae.dat packaging has broken the loading on OSX PPC side?

[FrodeSolheim]

Originally Posted by SnakeCoils

This version is still missing the OSD menu and messages because the executable seems to dislike the fs-uae.dat file inside the App bundle. If replaced with the same from 2.5.20 everything start to work as expected :-) Maybe some later modifications to the fs-uae.dat packaging has broken the loading on OSX PPC side?

Maybe; the more recent FS-UAE versions have support for compressed files within fs-uae.dat. Earlier versions only supported uncompressed files. Maybe the decompression does not work correctly with PPC/big-endian.

You can add try adding

-Z store

to the zip commands in Makefile.am to add the data files uncompressed to fs-uae.dat.

EDIT: Yes, this is a very likely explanation. I can see in the source code that the check for compression method is not endian safe. The compression method field is 16 bit (2 bytes), and must be checked with byte swapping on PPC. There are other endian issues with the decompression as well. Easy to fix, though.

[SnakeCoils]

Originally Posted by FrodeSolheim

You can add try adding

-Z store

to the zip commands in Makefile.am to add the data files uncompressed to fs-uae.dat.

Thank you for the hint Frode :-) however for the OSX "zip" the correct option for storing only without compression is simply "-0" and after changed two lines in the makefile.am as suggested the resulting App bundle now has floppy sounds and text menus again! :-D

I have simply modified the fs-uae-dat packaging lines in:

cd $(s) && $(ZIP) -0 -r $(abs_builddir)/fs-uae.dat share
cd $(b) && $(ZIP) -0 -r $(abs_builddir)/fs-uae.dat share

and now FS-UAE works perfectly with all sounds and menus you expect :-) I am going to upload this fixed version to the Zone!

[SnakeCoils]

Latest FS-UAE 2.7.16dev for OSX PPC is up in The Zone! and ready for download.
This version has been fixed with the same method described for the 2.7.15 release to bring back drive sounds, OSD messages and options menu.

[SnakeCoils]

Latest FS-UAE 2.9.0dev for OSX PPC is up in The Zone! and ready for download.
This version has been fixed with the same method described for the 2.7.15 release to bring back drive sounds, OSD messages and options menu.
Additionally I had to remove all these lines from src/cfgfile.cpp :

p->cachesize = MAX_JIT_CACHE;

because otherwise the compiler stopped with the following error:

src/cfgfile.cpp:6677:17: error: 'MAX_JIT_CACHE' was not declared in this scope

This is strange because the option "--disable-jit" was correctly declared in the ./configure command.

[FrodeSolheim]

Not that strange.. Toni does not test building WinUAE with JIT (and neither does I for FS-UAE, really). Anyway, fixed in commit https://github.com/FrodeSolheim/fs-u...00ee2f97790d8d

[FrodeSolheim]

@SnakeCoils Do you want to check if this patch fixes support for compression in fs-uae.dat?

diff --git a/fs-uae/libfsemu/src/data.c b/fs-uae/libfsemu/src/data.c
index 2d597ad..0da1fc1 100644
--- a/fs-uae/libfsemu/src/data.c
+++ b/fs-uae/libfsemu/src/data.c
@@ -175,8 +175,9 @@ int fs_data_file_content(const char *name, char **data, int *size)
         /* no signature found, not a local file header */
         return 8;
     }
-    if (lfh.compression_method != 0 && lfh.compression_method != 8) {
-        /* file is stored compressed, not supported right now */
+    if (le16toh(lfh.compression_method) != 0 &&
+            le16toh(lfh.compression_method) != 8) {
+        /* Unsupported compression type. */
         return 9;
     }
     if (data == NULL) {
@@ -200,10 +201,10 @@ int fs_data_file_content(const char *name, char **data, int *size)
         return 13;
     }
 
-    if (lfh.compression_method == 0) {
+    if (le16toh(lfh.compression_method) == 0) {
         // printf("STORE %s\n", name);
         /* Compression method "store" */
-        if (lfh.compressed_size != lfh.uncompressed_size) {
+        if (le32toh(lfh.compressed_size) != le32toh(lfh.uncompressed_size)) {
             return 9;
         }
         if (fread(*data, *size, 1, g_dat_file) != 1) {
@@ -211,7 +212,7 @@ int fs_data_file_content(const char *name, char **data, int *size)
             return 14;
         }
 #ifdef USE_ZLIB
-    } else if (lfh.compression_method == 8) {
+    } else if (le16toh(lfh.compression_method) == 8) {
         // printf("DEFLATE %s\n", name);
         /* Compression method "deflate". Borrowed some public domain code from
          * http://codeandlife.com/2014/01/01/unzip-library-for-c/ */
@@ -232,8 +233,8 @@ int fs_data_file_content(const char *name, char **data, int *size)
         }
 
         /* Inflate compressed data */
-        int compressed_left = lfh.compressed_size;
-        int uncompressed_left = lfh.uncompressed_size;
+        int compressed_left = le32toh(lfh.compressed_size);
+        int uncompressed_left = le32toh(lfh.uncompressed_size);
         while (compressed_left && uncompressed_left && ret != Z_STREAM_END) {
             /* Read next chunk */
             strm.avail_in = fread(buffer, 1,

[SnakeCoils]

Both patches applied, now let's wait the compiler to finish it's job...

EDIT1 : No more JIT CACHE errors, this part seems OK now

EDIT2 : Drive sounds, menu texts and OSD warning messages everything works great! :-)

Thank you very very much Frode, I will place the new binary in the Zone!

[SnakeCoils]

Latest FS-UAE 2.9.1dev for OSX PPC Leopard is up in The Zone! and ready for download.

[SnakeCoils]

Latest FS-UAE 2.9.2dev for OSX PPC Leopard is up in The Zone! and ready for download.

[eastone]

Thanks @SnakeCoils for this release. I'm always testing every Amiga emulator on PPC with this demo: http://www.pouet.net/prod.php?which=12336
and on FS-UAE looks perfect compared to E-UAE even with JIT but sounds unfortunately same bad as on other emulators
Could you test it on your configuration?
One more thing, demo requires standard A1200 configuration to run.

Hey SnakeCoils, I haven't actually tested this or really read through the thread entirely yet, but does this port just do 68k stuff or can I also use PPC?

[SnakeCoils]

@eastone
On my configuration the performance of the A1200 demo you mentioned is not that bad , of course there are some slowdowns when the emulated AGA chipset is squeezed but not the worse case I've seen, let's try the "Nexus 7" demo for example and you will see the real limit of available CPU resources for a smooth emulation.
Anyway this is the FS-UAE configuration (2.9.2dev binary) I used for running the demo:

amiga_model = A1200
accuracy = -1
floppy_drive_volume = 0
video_sync = off
fullscreen = 1
fullscreen_mode = fullscreen
keep_aspect = 1
viewport = * * * * => 0 0 752 574
floppy_drive_0 = Digital-A1200.adf

My PPC machine is an MDD dual G4 at 1.5 GHZ (overclocked from 1.42) with a Radeon X800XT (flashed from a FireGL X3) and 2 Gb of CL2 DRAM (4 identical Corsair DDR400 modules). Recently I added an SSD unit mounted on a Acard ARS-2133 and now system loading time is almost half the previous I had with a Raptor 15k.


@esc
At present my FS-UAE builds supports 68k code only, for Amiga PPC support on OSX PPC hosts some modifications to source are needed but the result, even on faster G5 machine is disappointing. I've heard that on Linux builds the performances are not so bad but I know very little about that world.

Ah, bummer. Interesting, though, is there any reason we can't pass a lot of the emulated ppc calls to the actual ppc hardware? I suspect that's probably very complicated but at first glance it doesn't seem totally necessary to emulate CPU instructions which already exist in the hardware.

[SnakeCoils]

I think this question has been already answered in some part of this long thread (or elsewhere in the forum), If I remember well the host CPU and memory must be accessed as separate virtual machine (QEMU) and results are then shared between FS-UAE and the VM plugin, in this way is possible to allow direct calls to the host CPU from the emulated Amiga without interferences to other host resources since the two environment operates as two separate VMs. In the past some experiments have been done but again the CPU resources of PowerMacs are not enough for smoothly running a basic FS-UAE 68k environment so imagine a 68k+PPC emulation. Honestly I have to admit to be not very interested in the AmigaOS 4.x world because the heart of Amiga is in the 68k code apps and games, however I appriciate new developments of this platform but I remain convinced that if you want to do "modern" things you need "newer" machines.

Cool, thanks for the reply


Sent from my iPad using Tapatalk Pro

[SnakeCoils]

Latest FS-UAE 2.9.3dev for OSX PPC Leopard is up in The Zone! and ready for download.

[eastone]

Newest version does not work on my G5:

Dyld Error Message:
Library not loaded: @executable_path/../Frameworks/SDL2.framework/Versions/A/SDL2
Referenced from: /Users/eastone/Downloads/FS-UAE293.app/Contents/MacOS/fs-uae
Reason: image not found

Previous version works well.

OK, I put SDL2 manually into Framework folder and it works now.