Matthias Braun [Wed, 1 Oct 2008 11:38:47 +0000 (11:38 +0000)]
create an index, as this heavily speeds up some queries
[r22391]
Sebastian Buchwald [Wed, 1 Oct 2008 11:23:05 +0000 (11:23 +0000)]
PBQP now doesn't find a solution, if we modified the costs a little bit.
- removed unused global variables
- added documentation :-)
now the PBQP solver need some new features...
[r22390]
Michael Beck [Tue, 30 Sep 2008 23:43:40 +0000 (23:43 +0000)]
- BugFix: cleanup cannot work with a precalculated step, use antic_in as initial clean nodes
- use avail_out as the closure for phi_translate()
- add missing lookup() in insert_nodes()
[r22387]
Sebastian Buchwald [Tue, 30 Sep 2008 23:04:21 +0000 (23:04 +0000)]
Added users for symconst.
[r22386]
Sebastian Buchwald [Tue, 30 Sep 2008 21:22:55 +0000 (21:22 +0000)]
Don't panic, the train is waiting for you...
[r22385]
Sebastian Buchwald [Tue, 30 Sep 2008 19:31:40 +0000 (19:31 +0000)]
Add users for symconst.
[r22384]
Moritz Kroll [Tue, 30 Sep 2008 19:01:17 +0000 (19:01 +0000)]
Optimize store after stores with partial overwrites with same values
[r22382]
Moritz Kroll [Tue, 30 Sep 2008 18:55:17 +0000 (18:55 +0000)]
Push ConvP through AddI
[r22381]
Moritz Kroll [Tue, 30 Sep 2008 18:54:08 +0000 (18:54 +0000)]
Mallon: Fixed smaller_mode()
[r22380]
Matthias Braun [Tue, 30 Sep 2008 16:37:42 +0000 (16:37 +0000)]
fix warning in optimized mode
[r22379]
Matthias Braun [Tue, 30 Sep 2008 16:34:19 +0000 (16:34 +0000)]
cleanup statistics stuff
[r22378]
Matthias Braun [Tue, 30 Sep 2008 16:32:27 +0000 (16:32 +0000)]
fix some warnings in optimized mode
[r22377]
Michael Beck [Tue, 30 Sep 2008 14:49:45 +0000 (14:49 +0000)]
place FIRM_DBG_REGISTER into the init function
[r22375]
Matthias Braun [Tue, 30 Sep 2008 14:05:58 +0000 (14:05 +0000)]
put memory operand folding to spill_apply time
[r22373]
Matthias Braun [Tue, 30 Sep 2008 13:53:57 +0000 (13:53 +0000)]
old version was correct...
[r22371]
Sebastian Buchwald [Tue, 30 Sep 2008 13:12:40 +0000 (13:12 +0000)]
Removed dependency that prevents full address mode.
[r22370]
Sebastian Buchwald [Tue, 30 Sep 2008 12:53:21 +0000 (12:53 +0000)]
Right nodes, right order but still not crashing...
[r22368]
Sebastian Buchwald [Tue, 30 Sep 2008 12:10:39 +0000 (12:10 +0000)]
Another variant without index, but shifts with different modes.
[r22367]
Sebastian Buchwald [Tue, 30 Sep 2008 11:27:05 +0000 (11:27 +0000)]
All heuristic applications at right place, but there's a conv between :-(
[r22363]
Sebastian Buchwald [Tue, 30 Sep 2008 10:48:57 +0000 (10:48 +0000)]
Use PBQP heuristic at shift const and afterwards at computed value.
[r22362]
Sebastian Buchwald [Tue, 30 Sep 2008 09:24:48 +0000 (09:24 +0000)]
Provoke heuristical PBQP decision for computed value.
[r22360]
Matthias Braun [Tue, 30 Sep 2008 08:19:50 +0000 (08:19 +0000)]
push compilation unit in context
[r22357]
Matthias Braun [Tue, 30 Sep 2008 08:08:47 +0000 (08:08 +0000)]
when we can do it for tuples, we can do it for Ids too
[r22356]
Matthias Braun [Tue, 30 Sep 2008 08:07:39 +0000 (08:07 +0000)]
we have code duplication in determine_entity_usage and is_address_taken (from scalar_replacement). Port the recent skip_tupe/id fixes
[r22355]
Michael Beck [Mon, 29 Sep 2008 23:36:49 +0000 (23:36 +0000)]
- add a combo flag again, so inlining can use combo or df/cf depending on command line option
[r22352]
Sebastian Buchwald [Mon, 29 Sep 2008 23:00:06 +0000 (23:00 +0000)]
Added more users to calculated value, but there are no real skip pattern for this nodes which is a little bit strange.
Maybe there's a bug which prevents consuming direct users of "memop"-nodes.
[r22351]
Michael Beck [Mon, 29 Sep 2008 22:54:04 +0000 (22:54 +0000)]
- Look through Tuple when checking for address taken (helps inlining)
[r22350]
Michael Beck [Mon, 29 Sep 2008 22:29:16 +0000 (22:29 +0000)]
- BugFix: removed missing debugging artefact, causing partitions with
constant type place on the workqueue again (fix ack/test2.c)
- Add debug output when a partition is placed on the workqueue
[r22349]
Sebastian Buchwald [Mon, 29 Sep 2008 19:57:32 +0000 (19:57 +0000)]
Don't try to use address multiple times, use calculated value instead.
[r22348]
Sebastian Buchwald [Mon, 29 Sep 2008 18:23:04 +0000 (18:23 +0000)]
Tried to create a PBQP graph with can't be solved heuristically (not finished yet)
[r22347]
Matthias Braun [Mon, 29 Sep 2008 13:43:09 +0000 (13:43 +0000)]
be robust against event names containing special chars
[r22346]
Matthias Braun [Mon, 29 Sep 2008 13:16:57 +0000 (13:16 +0000)]
we WANT to push forced graphs on the prioqueue
[r22345]
Matthias Braun [Mon, 29 Sep 2008 09:14:05 +0000 (09:14 +0000)]
here it is
[r22341]
Michael Beck [Sun, 28 Sep 2008 23:38:03 +0000 (23:38 +0000)]
- removed C99 features
[r22340]
Sebastian Buchwald [Sun, 28 Sep 2008 22:23:09 +0000 (22:23 +0000)]
- setted store value to symconst + const which can replaced by an immediate
[r22339]
Sebastian Buchwald [Sun, 28 Sep 2008 22:03:11 +0000 (22:03 +0000)]
- added missing base in two test cases
- added test cases for destination address mode
[r22338]
Sebastian Buchwald [Sun, 28 Sep 2008 21:37:44 +0000 (21:37 +0000)]
Added test with all (?) address mode variants.
[r22337]
Matthias Braun [Sun, 28 Sep 2008 21:16:06 +0000 (21:16 +0000)]
output register pressure in register class context
[r22336]
Matthias Braun [Sun, 28 Sep 2008 21:11:52 +0000 (21:11 +0000)]
also count remats
[r22335]
Matthias Braun [Sun, 28 Sep 2008 21:08:51 +0000 (21:08 +0000)]
improve fix some statistics in bechordal
[r22334]
Matthias Braun [Sun, 28 Sep 2008 18:25:59 +0000 (18:25 +0000)]
- split array.h into array.h/array_t.h and make array.h independent of
compiletime options like NDEBUG (as it is public API!)
- renamed some statevs (append a prefix for their module)
[r22329]
Matthias Braun [Sun, 28 Sep 2008 18:24:25 +0000 (18:24 +0000)]
fix bug in statev script for empty contexts
[r22328]
Matthias Braun [Sun, 28 Sep 2008 18:12:08 +0000 (18:12 +0000)]
disable some expensive statevs
[r22327]
Matthias Braun [Sun, 28 Sep 2008 17:37:43 +0000 (17:37 +0000)]
fix warning
[r22326]
Matthias Braun [Sun, 28 Sep 2008 17:37:15 +0000 (17:37 +0000)]
fix warning
[r22325]
Matthias Braun [Sun, 28 Sep 2008 17:36:57 +0000 (17:36 +0000)]
fix warning in optimized build
[r22324]
Moritz Kroll [Sun, 28 Sep 2008 00:41:24 +0000 (00:41 +0000)]
Updated Visual Studio 2005/2008 project
[r22322]
Michael Beck [Sun, 28 Sep 2008 00:28:58 +0000 (00:28 +0000)]
- add and fix some comments
[r22321]
Christoph Mallon [Sat, 27 Sep 2008 16:05:36 +0000 (16:05 +0000)]
Teach Mr. Gauß, Mr. Seidel and Mr. Grund to ignore the start loop.
[r22318]
Christoph Mallon [Sat, 27 Sep 2008 15:14:51 +0000 (15:14 +0000)]
Remove wrong constlike flags from PPC32 Addi, Ori and Andi_dot.
[r22317]
Michael Beck [Sat, 27 Sep 2008 14:57:43 +0000 (14:57 +0000)]
add missing exception attribute
[r22316]
Christoph Mallon [Sat, 27 Sep 2008 14:21:42 +0000 (14:21 +0000)]
Do not choke on Bad nodes while transforming for PPC32.
[r22315]
Christoph Mallon [Sat, 27 Sep 2008 14:17:18 +0000 (14:17 +0000)]
Use set_transformed_and_mark() instead of mark_irn_visited() + be_set_transformed_node().
[r22314]
Michael Beck [Sat, 27 Sep 2008 13:31:10 +0000 (13:31 +0000)]
- revert big benefice for irg_inline_forced methods,
seems to be less usefull
[r22313]
Michael Beck [Sat, 27 Sep 2008 10:32:04 +0000 (10:32 +0000)]
- BugFix: irg_inline_forced is not a bit
- revert to always inline forced functions for now
[r22311]
Michael Beck [Sat, 27 Sep 2008 03:13:41 +0000 (03:13 +0000)]
- fixed some typos
[r22310]
Michael Beck [Sat, 27 Sep 2008 02:50:50 +0000 (02:50 +0000)]
- add missing initializations
[r22309]
Michael Beck [Sat, 27 Sep 2008 02:45:49 +0000 (02:45 +0000)]
Various Fixes:
- put call entries into a double linked list, so we can remove them in any order
- correctly handle recursive calls: use the new (copied) callee_env
- correctly handle weights for self-recursive calls: due to inlining recursive calls might get promoted to self-recursive calls
- maxsize is a hard limit now: even forced functions are not inlined anymore (otherwise it would be possible to "force" a recursive one)
- forced inlines are handled like function with BIG benefice and inlined first (sorted by loop depth)
- fixed some comments, simplified code
[r22308]
Sebastian Buchwald [Fri, 26 Sep 2008 18:56:49 +0000 (18:56 +0000)]
typo
[r22303]
Christoph Mallon [Fri, 26 Sep 2008 17:21:14 +0000 (17:21 +0000)]
Indentation.
[r22302]
Christoph Mallon [Fri, 26 Sep 2008 17:17:40 +0000 (17:17 +0000)]
Fix comments.
[r22301]
Christoph Mallon [Fri, 26 Sep 2008 17:11:10 +0000 (17:11 +0000)]
Fix comments.
[r22299]
Christoph Mallon [Fri, 26 Sep 2008 17:05:49 +0000 (17:05 +0000)]
SymConsts represent constant entities irrespective of variability.
[r22297]
Michael Beck [Fri, 26 Sep 2008 15:05:34 +0000 (15:05 +0000)]
removed C99 code
[r22296]
Michael Beck [Fri, 26 Sep 2008 15:05:13 +0000 (15:05 +0000)]
simplified code, fixed comments
[r22294]
Sebastian Buchwald [Fri, 26 Sep 2008 13:38:12 +0000 (13:38 +0000)]
Removed senseless assert.
[r22291]
Christoph Mallon [Fri, 26 Sep 2008 11:02:20 +0000 (11:02 +0000)]
Fix crash in debug output.
[r22287]
Christoph Mallon [Fri, 26 Sep 2008 10:48:27 +0000 (10:48 +0000)]
Respect stickyness, when calculating entity usage.
[r22285]
Christoph Mallon [Thu, 25 Sep 2008 15:05:32 +0000 (15:05 +0000)]
Use be_put_ignore_regs() instead of duplicating its logic.
[r22271]
Christoph Mallon [Thu, 25 Sep 2008 15:04:52 +0000 (15:04 +0000)]
Resolve warning.
[r22270]
Christoph Mallon [Thu, 25 Sep 2008 14:50:06 +0000 (14:50 +0000)]
Remove unnecessary assure_dom_front().
[r22269]
Christoph Mallon [Thu, 25 Sep 2008 14:46:16 +0000 (14:46 +0000)]
Kick beraextern.c.
[r22268]
Christoph Mallon [Thu, 25 Sep 2008 14:29:42 +0000 (14:29 +0000)]
panic() instead of assert(0).
[r22264]
Sebastian Buchwald [Thu, 25 Sep 2008 14:04:06 +0000 (14:04 +0000)]
Fixed compile error and moved cse en/disabling into pbqp transformer.
[r22263]
Sebastian Buchwald [Thu, 25 Sep 2008 13:33:41 +0000 (13:33 +0000)]
Added getter for non address mode heuristic.
[r22262]
Sebastian Buchwald [Thu, 25 Sep 2008 13:14:44 +0000 (13:14 +0000)]
Set cse_opt to old value instead of enabling it.
[r22260]
Matthias Braun [Thu, 25 Sep 2008 11:19:38 +0000 (11:19 +0000)]
just a few random mips fixes
[r22256]
Matthias Braun [Wed, 24 Sep 2008 14:38:41 +0000 (14:38 +0000)]
push functions into a priority queue and inline in benefice order
[r22242]
Matthias Braun [Wed, 24 Sep 2008 14:38:04 +0000 (14:38 +0000)]
rename key to priority in pqueue
[r22241]
Matthias Braun [Wed, 24 Sep 2008 12:10:04 +0000 (12:10 +0000)]
cleanup pqueue a little bit
[r22235]
Matthias Braun [Wed, 24 Sep 2008 11:24:02 +0000 (11:24 +0000)]
more debuginfo
[r22234]
Matthias Braun [Wed, 24 Sep 2008 11:23:40 +0000 (11:23 +0000)]
hack to make it possible to mark spill,reload and remat nodes
[r22233]
Christoph Mallon [Wed, 24 Sep 2008 09:28:01 +0000 (09:28 +0000)]
Do not forcefully rebuild outedges in compute_execfreq(), but handle this at the caller site if necessary.
[r22231]
Matthias Braun [Wed, 24 Sep 2008 09:19:03 +0000 (09:19 +0000)]
fix error introduced by r22215
[r22230]
Matthias Braun [Wed, 24 Sep 2008 08:27:27 +0000 (08:27 +0000)]
fix warning
[r22229]
Matthias Braun [Wed, 24 Sep 2008 08:26:41 +0000 (08:26 +0000)]
improved? inlining heuristic
[r22228]
Matthias Braun [Wed, 24 Sep 2008 07:52:00 +0000 (07:52 +0000)]
fix the assert(0)->panic commit
[r22227]
Matthias Braun [Wed, 24 Sep 2008 07:00:52 +0000 (07:00 +0000)]
removed outdated comment
[r22226]
Michael Beck [Tue, 23 Sep 2008 23:30:44 +0000 (23:30 +0000)]
- replace a lot of assert(0) by panic()
[r22225]
Christoph Mallon [Tue, 23 Sep 2008 20:49:57 +0000 (20:49 +0000)]
panic() instead of assert(0)ing.
[r22221]
Michael Beck [Tue, 23 Sep 2008 19:39:22 +0000 (19:39 +0000)]
- detected const methods with possible endless loops cannot float and must have a memory ...
[r22219]
Michael Beck [Tue, 23 Sep 2008 18:09:52 +0000 (18:09 +0000)]
- implemented mtp_property_weak
[r22215]
Matthias Braun [Tue, 23 Sep 2008 14:52:59 +0000 (14:52 +0000)]
move declaration of inline functions around
[r22209]
Matthias Braun [Tue, 23 Sep 2008 14:52:24 +0000 (14:52 +0000)]
reenable floatin of const functions (don't care about obscure endless loop cases for now)
[r22208]
Matthias Braun [Tue, 23 Sep 2008 14:51:59 +0000 (14:51 +0000)]
always remove critical edges before doing code placement
[r22207]
Christoph Mallon [Tue, 23 Sep 2008 14:32:23 +0000 (14:32 +0000)]
Fix comment.
[r22205]
Sebastian Buchwald [Tue, 23 Sep 2008 14:08:26 +0000 (14:08 +0000)]
Fixed test case.
[r22203]
Sebastian Buchwald [Tue, 23 Sep 2008 12:58:03 +0000 (12:58 +0000)]
Added test case for pbqp construction: There seems to be missing skip patterns, so that the pbqp transformer produce unnecessary ia32 nodes.
[r22199]
Matthias Braun [Tue, 23 Sep 2008 12:42:42 +0000 (12:42 +0000)]
revert 22196
[r22198]
Matthias Braun [Tue, 23 Sep 2008 12:32:08 +0000 (12:32 +0000)]
free_methods was a bad base for callgraph walks, use methods which have no callers (in the current compilation unit)
[r22197]