peter
db81b4333a
* fatal error when procedure is too complex
2004-11-14 21:08:27 +00:00
florian
8ffe3c6a9e
* debug writeln removed
2004-11-06 18:58:18 +00:00
florian
42e14b6825
+ additional extdebug check for wrong add_reg_instructions added
...
* too long manglednames are cut off at 200 chars using a crc
2004-11-06 17:44:47 +00:00
peter
1cf3aef0f0
* regalloc bind to instructions need to get real ait_instruction
2004-11-01 10:34:08 +00:00
Jonas Maebe
dc71fc8290
* make sure live_start/end is never a tai_regalloc, as those can be
...
removed by the register allocator and thus become invalid. This fixed
make cycle with -Or for ppc, but I'm not sure what the warning on
symsym.pas:1663 means. Since the tlocation change, even regular make
cycle doesn't work anymore though...
2004-10-31 23:18:29 +00:00
peter
c95a859f0a
* generic tlocation
...
* move tlocation to cgutils
2004-10-31 21:45:02 +00:00
florian
36ca15cf07
* fixed compilation of system unit on arm
2004-10-31 16:04:30 +00:00
florian
e04b172854
* fixed generic optimizer
...
* enabled generic optimizer for sparc
2004-10-30 15:21:37 +00:00
peter
30be9e8e5d
* during translation only process regalloc for the current regtype
2004-10-24 17:04:01 +00:00
mazen
581b52422c
- remove $IFDEF DELPHI and related code
...
- remove $IFDEF FPCPROCVAR and related code
2004-10-15 09:14:16 +00:00
peter
60c73cc0e5
* -Or fixes for open array
2004-10-13 21:12:51 +00:00
peter
b0ebc06ce6
* removed warning about register used only once
2004-10-11 15:47:03 +00:00
peter
eaf46c4fb1
* live_registers must be restored after the spilling store code
...
is generate to add correct conflicts for extra temporary registers
2004-10-06 20:14:08 +00:00
peter
f5471aef6e
* more spilling rewrites
2004-10-05 20:41:01 +00:00
peter
ee3585f56f
* spilling code rewritten for x86. It now used the generic
...
spilling routines. Special x86 optimization still needs
to be added.
* Spilling fixed when both operands needed to be spilled
* Cleanup of spilling routine, do_spill_readwritten removed
2004-10-04 20:46:22 +00:00
peter
8fb3536f6e
* simple regvar support, not yet finished
2004-09-26 17:45:29 +00:00
peter
65c3ba277c
* ungetregister is now only used for cpuregisters, renamed to
...
ungetcpuregister
* renamed (get|unget)explicitregister(s) to ..cpuregister
* removed location-release/reference_release
2004-09-25 14:23:54 +00:00
peter
33a834821f
* paraloc branch merged
2004-09-21 17:25:12 +00:00
florian
1eb575fa22
* fixed longbool(<int64>) on sparc
2004-08-24 21:02:32 +00:00
daniel
c4e909e316
* Add check <= 255 when adding to adj_colours
2004-07-09 21:38:30 +00:00
daniel
649bd81edb
* Use a normal pascal set in assign_colours, since it only will contain
...
real registers
2004-07-08 09:57:55 +00:00
daniel
d2a2aeac9f
* supregset_reset clears 8kb of memory. However, it is being called in
...
inner loops, see for example colour_registers. According to profile data
this causes fillchar to be the most time consuming procedure.
Some modifications done to make it clear less than 8kb of memory each
call. Divides time spent in fillchar by two, but it still is the no.1
procedure.
2004-07-07 17:35:26 +00:00
florian
34d634c64d
* fixed arm compilation
2004-06-22 18:24:18 +00:00
florian
8a9758c5e2
* logs truncated
2004-06-20 08:55:28 +00:00
florian
e9bb27df36
* spilling of doubles on sparc fixed
2004-06-20 08:47:33 +00:00
florian
588e2c38bf
* dwarf branch merged
2004-06-16 20:07:06 +00:00
peter
5358278ca1
tai_regalloc.allocation changed to ratype to notify rgobj of register size changes
2004-05-22 23:34:27 +00:00
Jonas Maebe
b20d72aa36
* do not remove "allocation,deallocation" pairs, as those are important
...
for the optimizer
2004-04-26 19:57:50 +00:00
peter
3a5717c1b7
* rewrote add_to_movelist, it now uses a field to store the number
...
of allocated entries. Also made it using less hardcoded values
2004-03-14 22:50:04 +00:00
peter
56c727153a
* check if movelist is valid
2004-03-14 20:06:17 +00:00
peter
ee569d7817
* make extcycle is working again
2004-02-12 15:54:03 +00:00
olle
193e41b455
+ check that register allocation is not made at the wrong moment
2004-02-09 20:12:23 +00:00
Jonas Maebe
71202a141d
* taicpu.is_same_reg_move() now gets a regtype parameter so it only
...
removes moves of that particular register type. This is necessary so
we don't remove the live_start instruction of a register before it
has been processed
2004-02-08 23:10:21 +00:00
daniel
9868155d53
* Register allocator speed boost
2004-02-08 14:26:28 +00:00
daniel
46d241a666
* Take advantage of our new with statement optimization
2004-02-07 23:28:34 +00:00
daniel
be3eabc26f
* Some changes to better accomodate very large movelists
...
* movelist resizing now exponential (avoids heap fragmentation, saves
300 kb memory in make cycle)
* Trgobj.combine hand-optimized (still too slow)
2004-02-06 13:34:46 +00:00
peter
8caaa7740d
* more record alignment fixes
2004-01-28 22:16:31 +00:00
florian
8ef2f96215
* made DoSpill* overrideable
...
+ add_cpu_interferences added
2004-01-26 17:40:11 +00:00
daniel
97e87aaebc
* reginfo now also only allocated during register allocation
...
* third round of gdb cleanups: kick out most of concatstabto
2004-01-26 16:12:27 +00:00
daniel
ecf7accc38
* Keep interference bitmap only allocated during register allocation.
...
Saves 2 mb of memory.
2004-01-25 23:21:02 +00:00
peter
c34c9af5ce
* moved spilling code from taicpu to rg
2004-01-12 16:37:59 +00:00
daniel
61d29c4e4e
* Register allocation bug fixed
2004-01-11 13:21:35 +00:00
daniel
cdbb601a4e
* Degree=0 problem fixed
...
* Degree to high problem fixed
2004-01-09 22:02:29 +00:00
peter
1c7e6fc380
* sparc updates
...
* use registertype in spill_register
2003-12-26 14:02:30 +00:00
peter
0734904796
* only report unreleased imaginary registers
2003-12-22 23:09:34 +00:00
peter
cc8d2226ae
* made decrease_degree working, but not really fixed
2003-12-22 22:13:46 +00:00
florian
78aeec22ca
* arm compiler compilation fixed
2003-12-18 17:06:21 +00:00
peter
bc4f29e15c
* don't insert dealloc before alloc of the same register
2003-12-17 21:59:05 +00:00
daniel
23aa8e1cda
* Automatic conversion from integer constants to pointer constants is no
...
longer done except in Delphi mode
2003-12-16 09:41:44 +00:00
peter
1367e342db
* reg allocations for imaginary register are now inserted just
...
before reg allocation
* tregister changed to enum to allow compile time check
* fixed several tregister-tsuperregister errors
2003-12-15 21:25:48 +00:00
