08.Mar.2002
- uae_jit: minor optimizations in the mid-layer (and_l_ri)
02.Mar.2002
- uae_jit: add support for FreeBSD (Michael Alyn Mille)
- uae_fpu: fix asm constraint in fpu_x86.cpp/etos for FPU_DEBUG
- uae_fpu: rename FP types from fp_* to fpu_* to avoid name clash
15.Jan.2002
- uae_jit: add "jitdebug" prefs item so that runtime disassemblers
(m68k, x86 et al.) are enabled and extra checkpoints are generated.
This requires cxmon builtin.
07.Jan.2002
- uae_jit: fix again fflags_into_flags_internal() when the native
register clobbers condition codes. This is for the generation of
alternate code for fucomi instructions that don't exist on
non-P6+ cores.
06.Jan.2002
- CVS: synchronized with CVS sources from 06/01/2002
- uae_jit: merge with AmigaXL JIT sources (06/12/2001)
- uae_jit: enable separate BlockInfos allocation. i.e. move block
info data out of the translation cache. It needas more physical
memory but this option can save up to 30% compilation time when
many lazy flushes occur (e.g. applications in MacOS 8.x)
- uae_fpu: fix typos (replace NAN with NaN)
20.Dec.2001
- CVS: synchronized with CVS sources from 20/12/2001
- uae_jit: fix codegen for shift/rotate instructions
- uae_cpu: fix condition codes handling for BFINS instructions
20.Aug.2001
- CVS: synchronized with CVS sources from 18/08/2001
- uae_jit: fix a little bug in the register allocator
- uae_fpu: fix compilation problems in non-JIT mode
15.Jul.2001
- CVS: synchronized with CVS sources from 14/07/2001
- unix: runtime depth and resolution switching
- unix: AppleTalk-over-UDP tunnelling, activated by seeing "udptunnel"
to "true" in the prefs file
- unix: added "JIT Compiler" panel to the preferences editor
- table68k: NEGX may use X & Z flags
- uae_cpu: DIVS/DIVU: clear V flag when dividing by zero
- uae_cpu: completely removed spcflags hacks
- uae_cpu: better optimized flags handling for x86 (uses SAHF/SETO)
- uae_fpu: fixed 68040-specific FPU operations
- uae_fpu: rewrote functions from math lib for long double support (win32)
- uae_fpu: fix some rounding code for the IEEE-based fpu core
- uae_jit: added some peepholes for 8-bit immediates
- uae_jit: added a flag and a prefs item ("jit") to enable or
not the JIT compiler
- uae_jit: it is now possible to choose the translation cache
flush mechanism from a new prefs item ("jitlazyflush")
- jit_fpu: fucomi does not exist on non P6+ cores
- jit_fpu: fixed FMOVEM code translator once more. This solves Photoshop
hangs or crashes at startup.
- jit_fpu: removed FREM code translator because too many code would
have to be generated for correct handling of the quotient byte. This
fixes the "free rotation" feature of ClarisWorks in Drawing mode.
15.Feb.2001 ("NJ" release)
- windows: fixed 24 bpp blitter (Windows seems to use packed BGR888)
- uae_cpu: fixed stupid bug in CINV/CPUSH that prevented from flushing
the instruction cache as requested. This solves BBEdit Lite 4.1 crashes
and probably others...
- uae_jit: added CMOV "emulation" for processors that don't support them
- uae_jit: the "jitwantcmov" prefs item is now deprecated
- uae_jit: added code to determine processors suffering from partial register stalls
- jit_fpu: tried to improve long double support
- jit_fpu: fixed scrollbar bug caused by an incorrect FMOVEM code generator
14.Feb.2001
- CVS: synchronized with CVS sources from 11/02/2001
- Patches for fake Hardware Bases
- Integrated the Windows port
- VOSF code: improved blitters selection and performance
13.Jan.2001
- CVS: synchronized with CVS sources from 12/01/2001
- table68k: fixed CINV and CPUSH bitmasks to match the Cache Field
- table68k: EMULOP is now considered as a JUMP instruction and therefore
terminates a basic block
- uae_jit: imported "Ipswitch Town" Service Pack 1
- uae_jit: added or_l_rm() and raw_or_l_rm() used in compemu_fpp.cpp
- uae_jit: disabled spcflags hacks when compiling with the JIT compiler
since the gain (if any!) would be much too anecdotal to be worth
supporting them
- uae_jit: dropped support of banked memory addressing
- uae_cpu: removed all references to USE_COMPILER
- uae_cpu: removed compiler.h and compiler.cpp
- uae_fpu: fixed the FTAN bug in the assembly optimized core
- uae_fpu: IEEE-based core now features lazy evaluation of condition codes
- uae_mem: added get_virtual_address() for completeness
- uae_mem: removed declaration for the (unused) "filesysory" table
- uae_mem: removed the unused "check" slot in mem_banks
- cpuopti: if GCC version is older than 2.95, the optimization used to
delete useless register saves/restores is disabled
- configure: enabling the JIT compiler now enables JIT compilation for
FPU instructions and the Lazy Translation Cache Flush mechanism
- configure: added --enable-profile=gdb option to let GCC compile with '-g'
- configure: added -fno-strength-reduce to GCC/i386 options
- configure: added UNALIGNED_PROFITABLE to i386 macro definitions
- video_vosf.h: do_handle_screen_fault() now ends up with the default
SIGSEGV handler if we could not handle the faultive address ourselves
12.Nov.2000
- Now use the ASM_VOLATILE macro instead of __asm__ __volatile__
- New framework for FPU emulation cores and moved all FPU-related files
to the uae_cpu/fpu directory
- Enabled JIT compilation for floating-point instructions
(Configure option: --enable-jit-fpu --enable-fpe=uae)
- Fixed my lazy translation cache flusher
(Configure option: --enable-jit-lazy-flush)
- Optimized spcflags handling in m68k_run_compile() after compiled code
is run. Indeed, compiled code now returns a value indicating whether or
not, spcflags have to be checked or not.
(Default behavior)
- Imported on-demand spcflags checking. i.e. a new table filled up with
fake instructions handlers pointing to a dummy function checking spcflags
when necessary.
(Configure option: --enable-spcflags-hack)
- Attempted sort of atomic handling of spcflags through use of lock
instructions on X86 architectures and pthread mutexes where available.
(Configure option: --enable-spcflags-excl)
15.Oct.2000
- Better platform-independance
- Now follows unconditional branches if the lazy flusher is not used
- Added jitfpu prefs item
- Added jitwantcmov prefs item
- Enabled conditional moves
- Fixed a compilation bug when GCC tried to unroll loop in m68k_execute()
From Bernie Meyer's UAE-JIT (Everton release) :
- Optimized block lookup code, giving only up to 10% speedup
- Native registers can now hold more than one virtual register.
Avoids some pointless register moves, and thus gives a very slight
speedup
- Low level instruction optimizer. Currently broken :-(
- Compiler functions are now allowed to fail at runtime (see compemu_fpp)
14.Oct.2000
- Initial public release
