Commit Graph

302 Commits

Author SHA1 Message Date
daniel
e60da116b0 * More microoptimizations 2003-12-15 16:37:47 +00:00
peter
2057dbd01a * fix statedebug compile 2003-12-15 15:58:58 +00:00
daniel
7061f04355 * Register allocator speed optimizations
- Worklist no longer a ringbuffer
    - No find operations are left
    - Simplify now done in constant time
    - unusedregs is now a Tsuperregisterworklist
    - Microoptimizations
2003-12-14 20:24:28 +00:00
peter
bce5a1e252 * rg[tregistertype] added in tcg 2003-12-12 17:16:17 +00:00
peter
1407a03ff1 * remove redundant calls to add_edge_used 2003-12-04 23:27:32 +00:00
peter
825d5b4ed7 * check for add_move_instruction 2003-11-29 17:36:41 +00:00
florian
4584775cbe * changed some types to prevend range check errors 2003-11-24 15:17:37 +00:00
peter
f44380e4a9 * fixed alias/colouring > 255 2003-11-10 19:05:50 +00:00
florian
bbf7300a0c * Florian's culmutative nr. 1; contains:
- invalid calling conventions for a certain cpu are rejected
    - arm softfloat calling conventions
    - -Sp for cpu dependend code generation
    - several arm fixes
    - remaining code for value open array paras on heap
2003-11-07 15:58:32 +00:00
peter
9e66b09843 * call firstpass before allocation and codegeneration is started
* move leftover code from pass_2.generatecode() to psub
2003-10-30 16:22:40 +00:00
Jonas Maebe
cf8b181ff2 * some ALLOWDUPREG improvements 2003-10-29 21:29:14 +00:00
peter
96f9973b46 * taicpu_abstract.oper[] changed to pointers 2003-10-21 15:15:35 +00:00
florian
13a4e41b0c * improved speed; reduced memory usage of the interference bitmap 2003-10-19 12:36:36 +00:00
florian
05fcae577a * some ppc stuff fixed
* memory leak fixed
2003-10-19 01:34:30 +00:00
peter
9f54e1b7c5 * made worklists dynamic in size 2003-10-18 15:41:26 +00:00
peter
ec7fb064d0 * fixed last commit 2003-10-17 16:16:08 +00:00
florian
5b33ab3643 * fixed more ppc stuff 2003-10-17 15:25:18 +00:00
peter
d0de3b3ea8 * 64k registers supported
* fixed some memory leaks
2003-10-17 14:38:32 +00:00
florian
1df13c788c * fixed some MMX<->SSE
* started to fix ppc, needs an overhaul
  + stabs info improve for spilling, not sure if it works correctly/completly
  - MMX_SUPPORT removed from Makefile.fpc
2003-10-11 16:06:42 +00:00
peter
fb81b7ebbb * old trgobj moved to x86/rgcpu and renamed to trgx86fpu
* tregisteralloctor renamed to trgobj
  * removed rgobj from a lot of units
  * moved location_* and reference_* to cgobj
  * first things for mmx register allocation
2003-10-10 17:48:13 +00:00
daniel
b2343f4b16 * Register allocator splitted, ans abstract now 2003-10-09 21:31:37 +00:00
peter
70fe77ca7c * procinfo unit contains tprocinfo
* cginfo renamed to cgbase
  * moved cgmessage to verbose
  * fixed ppc and sparc compiles
2003-10-01 20:34:48 +00:00
peter
2771f49e22 * reuse registers with the least conflicts 2003-09-30 19:54:42 +00:00
peter
540691bf02 * optimized releasing of registers 2003-09-29 20:58:55 +00:00
peter
dea558b5c7 * return reg 255 when allowdupreg is defined 2003-09-28 13:41:12 +00:00
peter
bf66a6e4ab * fix filepositions
* insert spill temp allocations at the start of the proc
2003-09-25 16:19:32 +00:00
peter
76a53a375d * varspez in calls to push_addr_param 2003-09-16 16:17:01 +00:00
daniel
90e1ca64f3 * Fixed fast spilling functionality by re-adding the code that initializes
precoloured nodes to degree 255. I would like to play hangman on the one
    who removed that code.
2003-09-12 19:07:42 +00:00
florian
8f66389b46 * improved arm code generation
* move some protected and private field around
  * the temp. register for register parameters/arguments are now released
    before the move to the parameter register is done. This improves
    the code in a lot of cases.
2003-09-11 11:54:59 +00:00
daniel
49e539ac8d * Adding register allocation order 2003-09-09 20:59:27 +00:00
peter
05a15df3c2 * use register with least interferences in spillregister 2003-09-09 15:55:44 +00:00
peter
b9d09a4e5c * preparations for different default calling conventions
* various RA fixes
2003-09-07 22:09:34 +00:00
peter
610cbec252 * fixes for FPU register allocation 2003-09-03 21:06:45 +00:00
peter
6a8d5eb25d * NEWRA branch merged 2003-09-03 15:55:00 +00:00
florian
220e05dd5e * fixed arm concatcopy
+ arm support in the common compiler sources added
  * moved some generic cg code around
  + tfputype added
  * ...
2003-09-03 11:18:36 +00:00
daniel
889bd81bdb * Register allocator bugfix for h2pas 2003-08-23 10:46:21 +00:00
Jonas Maebe
4f8c390198 * fixed regvars so they work with newra (at least for ppc)
* fixed some volatile register bugs
  + -dnotranslation option for -dnewra, which causes the registers not to
    be translated from virtual to normal registers. Requires support in
    the assembler writer as well, which is only implemented in aggas/
    agppcgas currently
2003-08-17 16:59:20 +00:00
daniel
7d3bfaaed4 * Precoloured nodes now have an infinite degree approached with 255,
like they should.
2003-08-17 14:32:48 +00:00
daniel
45f9851e87 * Another register allocator bug fixed.
* cpu_registers set to 6 for i386
2003-08-17 08:48:02 +00:00
daniel
42c320cb29 * cs_regalloc renamed to cs_regvars to avoid confusion with register
allocator
  * Some preventive changes to i386 spillinh code
2003-08-09 18:56:54 +00:00
daniel
557ac8b02e * Fixed a register allocator bug
* Figured out why -dnewra generates superfluous "mov reg1,reg2"
    statements: changes in location_force. These moves are now no longer
    constrained so they are optimized away.
2003-08-03 14:09:50 +00:00
Jonas Maebe
7bc1557f5c * add_edges_used() is now also called for registers allocated with
getexplicitregisterint()
  * writing the intereference graph is now only done with -dradebug2 and
    the created files are now called "igraph.<module_name>"
2003-07-21 13:32:39 +00:00
daniel
268bfcf784 * Fixed register allocator. *Lots* of fixes. 2003-07-06 15:31:20 +00:00
Jonas Maebe
80d6905b7b * fixed my previous completely broken commit. It's not perfect though,
registers > last_supreg and < max_intreg may still be "translated"
2003-07-06 15:00:47 +00:00
Jonas Maebe
b1917a614b * support integer registers that are not managed by newra (ie. don't
translate register numbers that fall outside the range
    first_supreg..last_supreg)
2003-07-06 14:45:05 +00:00
peter
b7d99ec934 * paraloc splitted in callerparaloc,calleeparaloc
* sparc calling convention updates
2003-07-02 22:18:04 +00:00
Jonas Maebe
404d344d92 * lots of newra fixes (need getfuncretparaloc implementation for i386)!
* renamed all_intregisters to volatile_intregisters and made it
    processor dependent
2003-06-17 16:34:44 +00:00
Jonas Maebe
d29c96896f * fixed newra cycle for x86
* added constants for indicating source and destination operands of the
    "move reg,reg" instruction to aasmcpu (and use those in rgobj)
2003-06-14 14:53:50 +00:00
peter
0dceec9a64 * current_procdef removed, use current_procinfo.procdef instead 2003-06-13 21:19:30 +00:00
peter
d179aea20d * ungetregisterfpu gets size parameter 2003-06-12 21:11:10 +00:00
peter
a0b9306652 * newra compiles for sparc 2003-06-12 16:43:07 +00:00
Jonas Maebe
55509b199b * (de)allocation of registers for parameters is now performed properly
(and checked on the ppc)
  - removed obsolete allocation of all parameter registers at the start
    of a procedure (and deallocation at the end)
2003-06-09 14:54:26 +00:00
peter
d91c880148 * cg.a_load_* get a from and to size specifier
* makeregsize only accepts newregister
  * i386 uses generic tcgnotnode,tcgunaryminus
2003-06-03 21:11:09 +00:00
daniel
41e0bc4cec * Register allocator finished 2003-06-03 13:01:59 +00:00
peter
230a14ff68 * getregisterfpu size parameter added
* op_const_reg size parameter added
  * sparc updates
2003-06-01 21:38:06 +00:00
Jonas Maebe
f67e370ce1 * set inital costs of assigning a variable to a register to 120 for
non-i386, because the used register must be store to memory at the
    start and loaded again at the end
2003-05-31 20:31:11 +00:00
Jonas Maebe
0357f71e14 * fixed several regvar related bugs for non-i386. make cycle with -Or now
works for ppc
2003-05-30 18:55:21 +00:00
Jonas Maebe
f433c21ead * use as little different registers on the ppc until newra is released,
since every used register must be saved
2003-05-30 12:36:13 +00:00
Jonas Maebe
cb279b2029 * changed tt_persistant to tt_persistent :)
* tempcreatenode now doesn't accept a boolean anymore for persistent
    temps, but a ttemptype, so you can also create ansistring temps etc
2003-05-17 13:30:08 +00:00
peter
abca40f041 * regvar fixes 2003-05-16 14:33:31 +00:00
daniel
b899ea5469 * Bug fix in simplify 2003-04-26 20:03:49 +00:00
peter
f4b818fc1d * removed funcretn,funcretsym, function result is now in varsym
and aliases for result and function name are added using absolutesym
  * vs_hidden parameter for funcret passed in parameter
  * vs_hidden fixes
  * writenode changed to printnode and released from extdebug
  * -vp option added to generate a tree.log with the nodetree
  * nicer printnode for statements, callnode
2003-04-25 20:59:33 +00:00
daniel
6aa01a99a9 * Ifdefs around a lot of calls to cleartempgen
* Fixed registers that are allocated but not freed in several nodes
  * Tweak to register allocator to cause less spills
  * 8-bit registers now interfere with esi,edi and ebp
    Compiler can now compile rtl successfully when using new register
    allocator
2003-04-25 08:25:26 +00:00
peter
c351d2a524 * compile fix for no-newra 2003-04-23 20:23:06 +00:00
daniel
9e2084b8d5 * Further register allocator work. Compiler now smaller with new
allocator than without.
  * Somebody forgot to adjust ppu version number
2003-04-23 14:42:07 +00:00
peter
47489f2376 * firstpass uses expectloc
* checks if there are differences between the expectloc and
    location.loc from secondpass in EXTDEBUG
2003-04-22 23:50:22 +00:00
daniel
25059e21b6 + Implemented the actual register allocator
+ Scratch registers unavailable when new register allocator used
  + maybe_save/maybe_restore unavailable when new register allocator used
2003-04-22 10:09:34 +00:00
peter
e21dab9739 * count address regs separate 2003-04-21 19:16:49 +00:00
daniel
0c93e261e7 * Added some code to keep track of move instructions in register
allocator
2003-04-17 16:48:21 +00:00
daniel
a32926e2fc * Some work on interference graph construction 2003-04-17 07:50:24 +00:00
peter
c4ddcc78a8 * generic constructor working for i386
* remove fixed self register
  * esi added as address register for i386
2003-03-28 19:16:56 +00:00
Jonas Maebe
d168ae15df * lots of new regallocator fixes, both in generic and ppc-specific code
(ppc compiler still can't compile the linux system unit though)
2003-03-11 21:46:24 +00:00
olle
03f1fef7aa + added cutils to the uses clause 2003-03-09 21:18:59 +00:00
daniel
eedc447400 + Added newra version of Ti386shlshrnode
+ Added interference graph construction code
2003-03-08 20:36:41 +00:00
daniel
fb55235acc * Work to handle new register notation in ag386nsm
+ Added newra version of Ti386moddivnode
2003-03-08 13:59:16 +00:00
daniel
2801609d3f * Created newra version of secondmul in n386add.pas 2003-03-08 10:53:48 +00:00
daniel
e8f649f5c2 + $define newra will enable new register allocator
+ getregisterint will return imaginary registers with $newra
  + -sr switch added, will skip register allocation so you can see
    the direct output of the code generator before register allocation
2003-03-08 08:59:07 +00:00
daniel
5912bc972c * Fixed ungetreference 2003-02-26 20:50:45 +00:00
daniel
4c450c3bb0 * Fixed a few issues 2003-02-19 22:39:56 +00:00
daniel
86bcea6ef5 * Code generator converted to new register notation
- Horribily outdated todo.txt removed
2003-02-19 22:00:14 +00:00
carl
59d3edeec7 * Several bugfixes for m68k target (register alloc., opcode emission)
+ VIS target
  + Generic add more complete (still not verified)
2003-02-02 19:25:54 +00:00
daniel
55a161bbbf * Tregister changed into a record 2003-01-08 18:43:56 +00:00
carl
67486c96c3 * fixes for Delphi 6 compilation
(warning : Some features do not work under Delphi)
2002-10-05 12:43:23 +00:00
peter
4b81e16fe2 * tempgen cleanup
* tt_noreuse temp type added that will be used in genentrycode
2002-08-23 16:14:48 +00:00
florian
4144773f01 * result type handling in tcgcal.pass_2 overhauled
* better tnode.dowrite
  * some ppc stuff fixed
2002-08-17 22:09:43 +00:00
florian
e313bab4ff * first part of procinfo rewrite 2002-08-17 09:23:33 +00:00
florian
b136cecacd * first part of ppc calling conventions fix 2002-08-06 20:55:20 +00:00
carl
eff31e8524 + more more more documentation
+ first version include/exclude (can't test though, not enough scratch for i386 :()...
2002-08-05 18:27:48 +00:00
carl
32f3f65a26 + added generic exception support (still does not work!)
+ more documentation
2002-08-04 19:06:41 +00:00
florian
07783d2a0d * powerpc target fixed, very simple units can be compiled
* some basic stuff for better callparanode handling, far from being finished
2002-07-07 09:52:32 +00:00
peter
4285e99853 * internal linker
* reorganized aasm layer
2002-07-01 18:46:20 +00:00
peter
06ebac4e27 * readded missing revisions 2002-05-18 13:34:04 +00:00
carl
21b3a10f02 + defines.inc -> fpcdefs.inc to avoid conflicts if compiling by hand
+ try to fix temp allocation (still in ifdef)
+ generic constructor calls
+ start of tassembler / tmodulebase class cleanup
2002-05-16 19:46:34 +00:00
peter
2992e1819c * removed old logs and updated copyright year 2002-05-14 19:34:38 +00:00
carl
c3dfda6e7c + makeregsize
+ changeregsize is now a local routine
2002-04-21 15:23:03 +00:00
carl
87aa88e9b7 + generic FPC_CHECKPOINTER
+ first parameter offset in stack now portable
* rename some constants
+ move some cpu stuff to other units
- remove unused constents
* fix stacksize for some targets
* fix generic size problems which depend now on EXTEND_SIZE constant
2002-04-20 21:32:23 +00:00
carl
17f229006a + reg2str -> std_reg2str() 2002-04-15 19:03:31 +00:00
Jonas Maebe
71a52a4aeb * several powerpc-related additions and fixes 2002-04-06 18:10:42 +00:00
peter
5e7a1e3729 * removed unused units
* use tlocation.size in cg.a_*loc*() routines
2002-04-04 19:05:54 +00:00
peter
5b242f7741 * tlocation,treference update
* LOC_CONSTANT added for better constant handling
  * secondadd splitted in multiple routines
  * location_force_reg added for loading a location to a register
    of a specified size
  * secondassignment parses now first the right and then the left node
    (this is compatible with Kylix). This saves a lot of push/pop especially
    with string operations
  * adapted some routines to use the new cg methods
2002-04-02 17:11:27 +00:00
Jonas Maebe
43368392e2 * fixed different parameter name in interface and implementation
declaration of a method (only 1.0.x detected this)
2002-04-01 19:24:25 +00:00
Jonas Maebe
7fb55bf4e4 + a_loadfpu_* and a_loadmm_* methods in tcg
* register allocation is now handled by a class and is mostly processor
    independent (+rgobj.pas and i386/rgcpu.pas)
  * temp allocation is now handled by a class (+tgobj.pas, -i386\tgcpu.pas)
  * some small improvements and fixes to the optimizer
  * some register allocation fixes
  * some fpuvaroffset fixes in the unary minus node
  * push/popusedregisters is now called rg.save/restoreusedregisters and
    (for i386) uses temps instead of push/pop's when using -Op3 (that code is
    also better optimizable)
  * fixed and optimized register saving/restoring for new/dispose nodes
  * LOC_FPU locations now also require their "register" field to be set to
    R_ST, not R_ST0 (the latter is used for LOC_CFPUREGISTER locations only)
  - list field removed of the tnode class because it's not used currently
    and can cause hard-to-find bugs
2002-03-31 20:26:33 +00:00