Il ne s’agit pas vraiment d’un \u00e9mulateur mais plutot d’une machine virtuelle pour d\u00e9velopper et faire tourner des applications Atari ST, mais aussi TT & Falcon tournant sous Windows et en Open Source. Maintenant ARAnyM est fourni avec EmuTOS, un TOS gratuit. Deux versions sont disponible : Une avec et l’autre sans l’\u00e9mulation du DSP.<\/p>\n
<\/p>\n
Les am\u00e9liorations sont les suivantes:<\/p>\n
– input.cpp: improved the help message in window title<\/em> <\/p>\n
\n– various files: version++ and other minor updates for the new release<\/em>
\n– MacAranym.xcodeproj, configure.ac: Fixing build on Mac OS 10.9 \/ Xcode 5.1<\/em>
\n– New config setting [LILO] LoadToFastRAM to load kernel in FastRAM<\/em>
\n– Delete some FreeMiNT’s specific code, like usb_device struct, leave<\/em>
\n\u00a0 the updates to the members of this struct to FreeMiNT’s driver.<\/em>
\n– Remove some endianess swapping where not necessary, and add them<\/em>
\n\u00a0 where was needed. Use SDL function instead our owns.<\/em>
\n– Update API version, because some function have now less parameters.<\/em>
\n– Fix some return values.<\/em>
\n– ikbd.cpp – force transition for GPIP pin<\/em>
\n– fpu_mpfr.cpp – Be more careful about SNaN<\/em>
\n– newcpu.cpp – Fix printing of sign-extended operands in disassembler<\/em>
\n– fpu_mpfr.cpp – Preserve the bits of NaN<\/em>
\n– fpu_mpfr.c – Improve accuracy of denormal results, fix decoding of FBcc.L<\/em>
\n– Unix\/vm_alloc.cpp – Use MAP_32BIT only when requested<\/em>
\n– m68k.h – Added ARM flags optimization<\/em>
\n– configure.ac – Added ARM flags optimization<\/em>
\n– sysdeps.h – Improved clobber list<\/em>
\n– configure.ac : Added ARM Assembly compiler flags (build armv6 optimized<\/em>
\n\u00a0 assembly on armv6 and above) <\/em>
\n– src\/Unix\/sysdeps.h : Improved __asm__ memory access (supports armv6, armv7<\/em>
\n\u00a0 and common arm)<\/em>
\n\u00a0 (ARMV6_ASSEMBLY for armv6 -> armv8; or ARM_ASSEMBLY for armv1 -> armv5)<\/em>
\n\u00a0 because extended optimation is for armv6 and above<\/em>
\n\u00a0 i.e. –build=armv5l-unknown-linux-gnueabi will build an<\/em>
\n\u00a0 generic arm version with common ARM instructions<\/em>
\n– sigsegv_linux_arm.cpp\u00a0 : Bug fixed with word\/int byte order<\/em>
\n– configure.ac : Added sigsegv_linux.arm.cpp for linux ARM Architecture<\/em>
\n– sigsegv_linux_arm.cpp\u00a0 : sigsegv support for ARM Linux<\/em>
\n– Support for –enable-addr-check=sseg on ARM Linux<\/em>
\n– configure.ac : Reverted to previous FIXED_ADDRESS implementation for 64 Bit<\/em>
\n\u00a0 previous change wasn’t necessary<\/em>
\n– sigsegv_darwin_x86.cpp : 64 Bit code<\/em>
\n– sigsegv_darwin_x86.cpp : xor added statement<\/em>
\n– New files: MachServer.c, MachUser.c and gen_machfiles.sh<\/em>
\n– MacAranym.xcodeproj: New build phase gen_machfiles<\/em>
\n– sigsegv_linux_x86_64.cpp : New 64 Bit Jit for Linux<\/em>
\n– MacAranym.xcodeproj: Fixed Profile settings<\/em>
\n– configure.ac : Fixed typo<\/em>
\n– bpf_helper.c: Set executable flag in fix permissions and wait for child<\/em>
\n\u00a0 processes to finish.<\/em>
\n– configure.ac prepared for 64Bit Fixed addressing vm_allocate on darwin targets<\/em>
\n– MacAranym.xcodeproj: Fixed Packaging scheme for 32\/64 Bit Targets<\/em>
\n– sigsegv_darwin_x86.cpp: added break in set_eflags. (Not sure if the eflags<\/em>
\n\u00a0 are really relevant)<\/em>
\n– sigsegv_darwin_x86.cpp: Fixed X86_REG_EDX<\/em>
\n– sigsegv_darwin_x86.cpp: Limit to BAD_ACCESS and forward_exception to enable<\/em>
\n\u00a0 XCode Debugging<\/em>
\n– MacAranym.xcodeproj: Removed specific compilation flags of bpf_helper target.<\/em>
\n\u00a0 This should enable error-free compilation with Xcode 3.1.<\/em>
\n– newcpu.cpp\/compemu_fpp.cpp\/fpu_ieee.cpu: fixed bugs which were reported as<\/em>
\n\u00a0 warnings by LLVM.<\/em>
\n– ethernet_macosx.cpp: BPF packet filtering improvements: added a new filter for<\/em>
\n\u00a0 multi-cast packets (MAC based) and another filter solely based on the IP<\/em>
\n\u00a0 address (ARP and IP protocol only).<\/em>
\n\u00a0 \u00ab\u00a0nofilter\u00a0\u00bb should only be used for debugging, as it causes memory leaks on the<\/em>
\n\u00a0 Atari\/MiNT side and therefore system crashes.<\/em>
\n\u00a0 Implemented multi-packet reads from BPF device (one read operation can deliver<\/em>
\n\u00a0 multiple ethernet frames).<\/em>
\n– uae_cpu\/compiler\/compemu.h: Avoid conflict with definitions from<\/em>
\n\u00a0 <byteswap.h>, imported by SDL.<\/em>
\n– ethernet_macosx.cpp: fixed memory leak on Atari side by implementing a <\/em>
\n\u00a0 working packet filter which ensures that only valid ethernet frames are <\/em>
\n\u00a0 forwarded to the Atari side. Removed the \u00ab\u00a0nofilter\u00a0\u00bb keyword.<\/em>
\n\u00a0 Thanks to Zorro for pointing me to this severe bug.<\/em>
\n– natfeat\/ethernet.h: Increase MAX_PACKET_SIZE to 9000 to allow enabling<\/em>
\n\u00a0 Ethernet on systems with MTU bigger than 1514.<\/em>
\n– ethernet_macosx.cpp: Fix memory overflow in BPF driver.<\/em>
\n– input.cpp: fixed click to grab input, added proper host<->Atari mouse sync<\/em>
\n– ikbd.cpp: fixed SendMouseMotion() and MergeMousePacket()<\/em>
\n– include\/aradata.h: added setAtariMousePosition()<\/em>
\n– natfeat\/nfvdi.cpp: fixes a bug (sort of a race condition) in NFVDI.<\/em>
\n\u00a0 When fVDI switches to 32-bit display in VdiDriver::setResolution(),<\/em>
\n\u00a0 a new HostSurface is created immediately for the VDI driver.<\/em>
\n\u00a0 Then the video mode change is applied to the host window only on next VBL,<\/em>
\n\u00a0 in HostScreen::refreshSurface().<\/em>
\n\u00a0 But in VdiDriver::expandArea(), there was a bogus test:<\/em>
\n\u00a0\u00a0\u00a0 if (host->video->getBpp() == 8) {<\/em>
\n\u00a0 This was wrong, because the VDI made a test on the host window surface,<\/em>
\n\u00a0 which would only be correct on next VBL. Found and fixed by Vincent Riviere.<\/em>
\n– bootos_emutos.cpp: fix for now correct BCD date format in EmuTOS by Vincent.<\/em>
\n– include\/natfeats.h: added proper check for end of memory block that<\/em>
\n\u00a0 is to be copied between host and guest using the LIBC memcpy().<\/em>
\n\u00a0 Re-enabled NATFEAT_LIBC_MEMCPY fast copying in NatFeats because<\/em>
\n\u00a0 I believe it’s relatively safe now when there are checks for<\/em>
\n\u00a0 begin and end of the Atari memory block. Atari memory is contiguous<\/em>
\n\u00a0 so when both begin and end of the block is readable\/writeable<\/em>
\n\u00a0 the whole block should be OK to copy by LIBC memcpy.<\/em>
\n– input.cpp: \u00ab\u00a0Reseting weird mouse packet\u00a0\u00bb error message changed to<\/em>
\n\u00a0 \u00ab\u00a0Resetting suspicious mouse packet\u00a0\u00bb debug message thus it’ll no longer<\/em>
\n\u00a0 annoy you. Thanks to Helmut Karlowski for pointing it out.<\/em>
\n– include\/natfeats.h: added switch for previously default and dangerous<\/em>
\n\u00a0 direct LIBC memcpy() to\/from guest provided memory pointer. Now it is<\/em>
\n\u00a0 disabled by default and byte by byte copying routine is used instead<\/em>
\n\u00a0 (guest programs should no longer be able to kill the whole ARAnyM).<\/em>
\n\u00a0 It might affect the NatFeats performance negatively so if required<\/em>
\n\u00a0 set the newly introduced NATFEAT_LIBC_MEMCPY define to 1.<\/em>
\n\u00a0 In addition the direct LIBC memcpy() is now guarded by proper pointer check<\/em>
\n\u00a0 so even if you enable it for speed reasons you should still be relatively<\/em>
\n\u00a0 safe. Thanks to Thorsten Glaser for providing simple linux user program<\/em>
\n\u00a0 that was able to kill ARAnyM by issuing NF_NAME NatFeat.<\/em>
\n– natfeat\/nf_basicset.cpp, natfeat\/ethernet.cpp: removed incorrect ValidAddr<\/em>
\n\u00a0 checks.<\/em>
\n– Xcode 4.5 project\/build system: fixes for building with OS X 10.8, changed <\/em>
\n\u00a0 packaging DMG to include README_ethernet.rtf instead of TunTap description,<\/em>
\n\u00a0 adjusted detection of aclocal and makedepend to support usage of MacPorts<\/em>
\n– Disabled loading of WMIcon on MacAranym as the correct icon is already bundled<\/em>
\n\u00a0 with the application.<\/em>
\n– README, NEWS: updated for a new release<\/em>
\n– aranym.spec, version.h, export: increased version to 0.9.14<\/em>
\n– input.cpp: use host middle mouse button for keyboard\/mouse ungrab. Works only<\/em>
\n\u00a0 if Eiffel mouse middle button\/wheel mode is not enabled.<\/em>
\n– Added new option –enable-bpf-ethernet for Mac OS X. This allows using bridged<\/em>
\n\u00a0 ethernet without installing TunTap kernel extensions. See README_ethernet.rtf.<\/em>
\n– ethernet_macosx.cpp: Berkley Packet Filter (BPF) based ethernet driver<\/em>
\n\u00a0 for Mac OS X. This driver supports only \u00ab\u00a0bridge\u00a0\u00bb mode, i.e. Aranym <\/em>
\n\u00a0 is part of your LAN, with its own IP address.<\/em>
\n– bpf_helper.c,fd_trans.c: a small utility used to do privileged operations<\/em>
\n\u00a0 like opening a BPF device. When run for the first time it tries to<\/em>
\n\u00a0 fix the permissions (chown root, chmod ug+s).<\/em>
\n\u00a0 The file descriptor to the device is then passed using Unix sockets.<\/em>
\n– Added Xcode 4 scheme files: added shared schemes for building the targets<\/em>
\n– Source cleanup\/remove warnings for compilation with latest LLVM compiler of<\/em>
\n\u00a0 Xcode 4.5<\/em>
\n– Building on Mac OS 10.7: Do not set \u00ab\u00a0-isysroot\u00a0\u00bb during compilation. Targetting<\/em>
\n\u00a0 Mac OS 10.5 is still possible without explicitely specifying the SDK to use<\/em>
\n\u00a0 for compilation. This was only necessary during Universal Binary builds where<\/em>
\n\u00a0 explicitely the MacOSX10.4u.sdk had to be used to build multi architecture<\/em>
\n\u00a0 builds.<\/em>
\n– hostscreen.cpp: \u00ab\u00a0enforce VIDEL refresh\u00a0\u00bb was missing => no screen updates <\/em>
\n\u00a0 during boot up when no nfVDI was active.<\/em>
\n– gui-osx\/SDLMain.m: Fix \u00ab\u00a0deprecated\u00a0\u00bb warning for \u00ab\u00a0poseAsClass\u00a0\u00bb<\/em>
\n– Xcode project: enable X86_64 compilation beside i386 architecture and build<\/em>
\n\u00a0 fat binaries. 64 bit support in SDL is only available as of Mac OS 10.6.<\/em>
\n\u00a0 Therefore on 10.5 only the 32 bit version will run.<\/em>
\n– hostscreen.cpp: fix missing check for \u00ab\u00a0SingleBlitRefresh\u00a0\u00bb configuration in<\/em>
\n\u00a0 refreshScreen().<\/em>
\n– nfvdi_soft.cpp: Performance improvements by skipping memory translation<\/em>
\n\u00a0 if MMU emulation is disabled.<\/em>
\n– dirty_rect.cpp: maintain a surrounding rectangle containing the whole<\/em>
\n\u00a0 dirtied area.<\/em>
\n– hostscreen.cpp: Two possible modes for updating the atari screen surface <\/em>
\n\u00a0 and the host screen surface: \u00ab\u00a0chunky\u00a0\u00bb (as in previous versions, uses <\/em>
\n\u00a0 multiple, small rectangular updates) and \u00ab\u00a0single\u00a0\u00bb (which uses one single<\/em>
\n\u00a0 rectangular update over the whole \u00ab\u00a0dirty\u00a0\u00bb surface)<\/em>
\n– parameters.cpp\/dlgVideo.cpp: new boolean \u00ab\u00a0SingleBlitRefresh\u00a0\u00bb <\/em>
\n\u00a0 and \u00ab\u00a0SingleBlitComposing\u00a0\u00bb for the video configuration segment. <\/em>
\n\u00a0 If set to \u00ab\u00a0Yes\u00a0\u00bb the screen to be composed\/screen to be refreshed will be sent<\/em>
\n\u00a0 in a single BitBlit to the screen\/graphics card. If set to \u00ab\u00a0No\u00a0\u00bb, only the <\/em>
\n\u00a0 dirty screen areas are sent by using multiple BitBlits. <\/em>
\n\u00a0 Default is \u00ab\u00a0No\u00a0\u00bb.<\/em>
\n\u00a0 => Checkout what combination gives you the best performance.<\/em>
\n– hostscreen: Merge consecutive dirty blocks to reduce number of update to<\/em>
\n\u00a0 screen operations.<\/em>
\n– MacAranym.xcodeproj: Support compilation with SDL 1.2.15.<\/em>
\n\u00a0 The standard Mac OS X deployment target is now 10.6, Intel only.<\/em>
\n\u00a0 It’s still possible to compile binaries with PowerPC and OS 10.5<\/em>
\n\u00a0 but no more Universal Binaries are going to be built.<\/em>
\n\u00a0 Build configuration \u00ab\u00a0Release G5\u00a0\u00bb has been removed.<\/em>
\n– MacAranym.xcodeproj: Added new target \u00ab\u00a0MacAranym MMU MPFR\u00a0\u00bb with MPFR FPU <\/em>
\n\u00a0 engine enabled. This target needs libgmp and libmpfr from MacPorts.<\/em>
\n\u00a0 Please checkout the README_Xcode.rtf for more details on how to build.<\/em>
\n– MacAranym.xcodeproj: Renamed \u00ab\u00a0MacAranym JIT IEEE\u00a0\u00bb to \u00ab\u00a0MacAranym JIT\u00a0\u00bb as JIT <\/em>
\n\u00a0 already implies IEEE FPU.<\/em>
\n\u00a0 Further this target has been modified to include always \u00ab\u00a0–enable-jit-fpu\u00a0\u00bb.<\/em>
\n– MacAranym.xcodeproj: Clean-up of project global and target specific configure<\/em>
\n\u00a0 flags (CONFIGURE_OPTIONS_BASE and CONFIGURE_OPTIONS_TARGET).<\/em>
\n\u00a0 Automatic detection when COMPILE_DEFS do not match what is generated in the<\/em>
\n\u00a0 Makefile during configure. It is now easier to change configure options on a<\/em>
\n\u00a0 target and really getting the right stuff compiled.<\/em>
\n– sigsegv_darwin_x86.cpp: Fix compilation of JIT if Mac OS X Deployment target >= 10.5<\/em>
\n– fpu\/mathlib.h: fix compilation problem with in fp_do_expm1 inline assembly<\/em>
\n\u00a0 and LLVM GCC compiler.<\/em>
\n– dirty_rects.cpp\/h: implemented setDirtyLine which calculates the surrounding<\/em>
\n\u00a0 rectangle and adds it to the dirty rectangle list.<\/em>
\n– nfvdi_soft.cpp: use setDirtyLine instead of setDirtyRect whenever a line is<\/em>
\n\u00a0 drawn. This fixes redraw glitches whenever a line is not drawn from top left<\/em>
\n\u00a0 to bottom right.<\/em>
\n– configure.ac: don’t stop if libusb not found (apparently Debian on kfreebsd<\/em>
\n\u00a0 has the libusb as a fake package or what so the build was failing there)<\/em>
\n– fpu_mpfr.cpp: Fix range check when converting to long word<\/em>
\n– Fix MacAranym JIT: set\/enforce deployment target to OS 10.4<\/em>
\n– fpu_mpfr.cpp: Fix missing fpu exception update<\/em><\/p>\n