Michael Beck [Fri, 12 Sep 2008 21:35:33 +0000 (21:35 +0000)]
- C99 features removed
[r21913]
Matthias Braun [Fri, 12 Sep 2008 11:13:23 +0000 (11:13 +0000)]
adjustements
[r21893]
Michael Beck [Fri, 12 Sep 2008 00:41:03 +0000 (00:41 +0000)]
- BugFix: fc_get_plusinf()/fc_get_minusinf() must set INF class
- BugFix: fixed printing of Inf values
[r21869]
Michael Beck [Fri, 12 Sep 2008 00:33:22 +0000 (00:33 +0000)]
Improved handling of fc_exact: NaN and Inf results are NOT exact
[r21868]
Michael Beck [Fri, 12 Sep 2008 00:14:14 +0000 (00:14 +0000)]
reverted r21775: too restrictive because ieee 754 uses binary representation for the mantissa
[r21867]
Michael Beck [Thu, 11 Sep 2008 23:20:54 +0000 (23:20 +0000)]
- reset the default stack alignment to 2^2
[r21864]
Michael Beck [Thu, 11 Sep 2008 23:06:30 +0000 (23:06 +0000)]
- fix for r21852: do not remove ALL Conv optimizations,
yust those that causing a problem (Conv to mode_b leading to Phib)
[r21863]
Christoph Mallon [Thu, 11 Sep 2008 13:42:59 +0000 (13:42 +0000)]
Fix return value.
[r21853]
Andreas Zwinkau [Thu, 11 Sep 2008 13:00:56 +0000 (13:00 +0000)]
Removing conv-on-Phi local opt
It leads to remaining Phib nodes later, because
the conv_b_lower operation is instantly reverted, when
it tries to insert a Convb.
[r21852]
Christoph Mallon [Thu, 11 Sep 2008 12:38:18 +0000 (12:38 +0000)]
Resolve race condition: First set handler, then call alarm().
[r21851]
Christoph Mallon [Thu, 11 Sep 2008 12:13:58 +0000 (12:13 +0000)]
In be_get_Proj_for_pn() before getting the proj number make sure that the node really is a Proj (could be End due to keep alive).
[r21847]
Christoph Mallon [Thu, 11 Sep 2008 12:11:05 +0000 (12:11 +0000)]
Indentation, braces.
[r21846]
Christoph Mallon [Thu, 11 Sep 2008 12:09:02 +0000 (12:09 +0000)]
Typo in comment.
[r21844]
Christoph Mallon [Thu, 11 Sep 2008 10:20:05 +0000 (10:20 +0000)]
Do not mark functions with kept basic blocks (endless loops, dead ends) as "const" or "pure".
[r21838]
Christoph Mallon [Thu, 11 Sep 2008 10:18:22 +0000 (10:18 +0000)]
Prevent beuses from choking on basic blocks without control flow successors.
[r21837]
Andreas Zwinkau [Thu, 11 Sep 2008 09:27:44 +0000 (09:27 +0000)]
Adding mini testcase for incmem
[r21836]
Christoph Mallon [Thu, 11 Sep 2008 08:24:32 +0000 (08:24 +0000)]
Remove unused variable.
[r21833]
Christoph Mallon [Thu, 11 Sep 2008 08:21:51 +0000 (08:21 +0000)]
s/get_irn_op(x) {==,!=} op_FOO/{,!}is_FOO(x)/.
[r21832]
Christoph Mallon [Thu, 11 Sep 2008 07:30:06 +0000 (07:30 +0000)]
s/Block_not_block_visited(x)/!Block_block_visited(x)/.
[r21831]
Christoph Mallon [Thu, 11 Sep 2008 07:18:53 +0000 (07:18 +0000)]
s/get_irn_op(x) == op_Block/is_Block(x)/.
[r21830]
Christoph Mallon [Thu, 11 Sep 2008 06:22:30 +0000 (06:22 +0000)]
Typo in comment.
[r21828]
Michael Beck [Wed, 10 Sep 2008 16:21:07 +0000 (16:21 +0000)]
remove assert in keepalive()
[r21824]
Matthias Braun [Wed, 10 Sep 2008 15:06:03 +0000 (15:06 +0000)]
fix testcases, yet another endless loop test
[r21823]
Matthias Braun [Wed, 10 Sep 2008 15:00:22 +0000 (15:00 +0000)]
more tricky endless loop stuff
[r21822]
Michael Beck [Wed, 10 Sep 2008 14:06:24 +0000 (14:06 +0000)]
- Changed keep logic: all nodes can be keeped yet, thi is necessary to fix fehler152
[r21820]
Michael Beck [Wed, 10 Sep 2008 11:05:37 +0000 (11:05 +0000)]
another strange endless loop failure
[r21814]
Michael Beck [Wed, 10 Sep 2008 10:58:08 +0000 (10:58 +0000)]
wrong endless loop handling
[r21813]
Christoph Mallon [Wed, 10 Sep 2008 09:04:45 +0000 (09:04 +0000)]
Use Block-specific functions instead of generic get_irn_*().
[r21812]
Christoph Mallon [Wed, 10 Sep 2008 05:53:39 +0000 (05:53 +0000)]
Fix typo in comment.
[r21811]
Michael Beck [Wed, 10 Sep 2008 01:31:56 +0000 (01:31 +0000)]
fixed warning
[r21810]
Michael Beck [Wed, 10 Sep 2008 01:23:54 +0000 (01:23 +0000)]
- use unsigned instead of int where appropriate
- use int instead of double for affinity edge weight
[r21809]
Michael Beck [Wed, 10 Sep 2008 01:09:23 +0000 (01:09 +0000)]
simplified more
[r21808]
Michael Beck [Wed, 10 Sep 2008 01:08:07 +0000 (01:08 +0000)]
typo fixed
[r21807]
Michael Beck [Wed, 10 Sep 2008 01:03:04 +0000 (01:03 +0000)]
- fixed comment: bs cannot be NULL anymore (and was never NULL previously)
[r21806]
Michael Beck [Wed, 10 Sep 2008 01:01:21 +0000 (01:01 +0000)]
arch_register_class_put() is only used locally
[r21805]
Michael Beck [Wed, 10 Sep 2008 00:54:33 +0000 (00:54 +0000)]
- if 0'ed unused function
[r21804]
Michael Beck [Wed, 10 Sep 2008 00:44:04 +0000 (00:44 +0000)]
use panic instead of assert(0 & "...")
[r21803]
Michael Beck [Wed, 10 Sep 2008 00:09:45 +0000 (00:09 +0000)]
- fixed typo, indentation
[r21802]
Michael Beck [Wed, 10 Sep 2008 00:00:22 +0000 (00:00 +0000)]
- made the be_stack_layout part of be_abi_irg_t to reduce indirection
- if(..) -> if (...)
- removed unused dont_save set
- moved initializations of variables to first usage reducing life time
[r21801]
Matthias Braun [Tue, 9 Sep 2008 15:01:16 +0000 (15:01 +0000)]
fix
[r21788]
Matthias Braun [Tue, 9 Sep 2008 14:56:56 +0000 (14:56 +0000)]
match immediate for SubSP and AddSP
[r21787]
Matthias Braun [Tue, 9 Sep 2008 14:51:49 +0000 (14:51 +0000)]
test without alignment
[r21786]
Matthias Braun [Tue, 9 Sep 2008 14:09:43 +0000 (14:09 +0000)]
intialize initial_bias
[r21785]
Michael Beck [Tue, 9 Sep 2008 14:02:28 +0000 (14:02 +0000)]
- Allowing an initial stack_bias to be transmitted from the prolog generating function
- reverted 21738: old_bp is again part of the between type for ia32
[r21783]
Michael Beck [Tue, 9 Sep 2008 13:04:58 +0000 (13:04 +0000)]
checked in "too much"
[r21782]
Michael Beck [Tue, 9 Sep 2008 13:00:05 +0000 (13:00 +0000)]
fixed r21771
[r21781]
Christoph Mallon [Tue, 9 Sep 2008 12:06:26 +0000 (12:06 +0000)]
Typo in comment.
[r21780]
Christoph Mallon [Tue, 9 Sep 2008 11:13:37 +0000 (11:13 +0000)]
Add should_be_same constraint on second input to second output of ia32_Minus64Bit.
[r21778]
Michael Beck [Tue, 9 Sep 2008 10:43:30 +0000 (10:43 +0000)]
- finally kicked size_t, both calls return unsigned
[r21776]
Michael Beck [Tue, 9 Sep 2008 10:41:43 +0000 (10:41 +0000)]
- even more restrictive: we only know that twos-complement can be placed into a IEEE 754 mantissa
[r21775]
Michael Beck [Tue, 9 Sep 2008 10:10:09 +0000 (10:10 +0000)]
- currently we support IEEE 754 only, so change the condition
[r21774]
Michael Beck [Tue, 9 Sep 2008 10:08:01 +0000 (10:08 +0000)]
- give enums tags so MSVC debugger is happy
[r21773]
Michael Beck [Tue, 9 Sep 2008 10:06:17 +0000 (10:06 +0000)]
tarval_ieee754_get_mantissa_size() implemented
[r21772]
Michael Beck [Tue, 9 Sep 2008 09:54:42 +0000 (09:54 +0000)]
- C99 features removed
- renamed second var i
[r21771]
Christoph Mallon [Tue, 9 Sep 2008 09:20:30 +0000 (09:20 +0000)]
80bit floats have no implicit 1. Remove unnecessary condition.
[r21770]
Christoph Mallon [Tue, 9 Sep 2008 09:19:06 +0000 (09:19 +0000)]
Slightly reduce code duplication.
[r21769]
Matthias Braun [Tue, 9 Sep 2008 07:43:44 +0000 (07:43 +0000)]
push unused livethroughs through loops in belady3
[r21766]
Matthias Braun [Tue, 9 Sep 2008 07:43:19 +0000 (07:43 +0000)]
improved PIC support
[r21765]
Christoph Mallon [Tue, 9 Sep 2008 06:38:17 +0000 (06:38 +0000)]
Do not warn about wrong register for a Phi input if that register is a joker (i.e. the infamous unknown register).
[r21764]
Christoph Mallon [Tue, 9 Sep 2008 06:36:19 +0000 (06:36 +0000)]
Make warning about registers assigned to Phi inputs more clear.
[r21763]
Christoph Mallon [Mon, 8 Sep 2008 18:20:50 +0000 (18:20 +0000)]
Correctly transform memory for dest AM when Syncs are present.
[r21752]
Michael Beck [Mon, 8 Sep 2008 12:35:59 +0000 (12:35 +0000)]
testcase for failing gvn-pre
[r21749]
Matthias Braun [Sun, 7 Sep 2008 16:35:22 +0000 (16:35 +0000)]
fix warnings
[r21739]
Matthias Braun [Sun, 7 Sep 2008 16:34:40 +0000 (16:34 +0000)]
old_bp should not be part of the between type, this fixes stack alignment for functions with base pointer
[r21738]
Sebastian Buchwald [Fri, 5 Sep 2008 21:43:00 +0000 (21:43 +0000)]
typo
[r21732]
Matthias Braun [Fri, 5 Sep 2008 18:50:32 +0000 (18:50 +0000)]
don't compare attributes but their addresses
[r21730]
Matthias Braun [Fri, 5 Sep 2008 16:55:03 +0000 (16:55 +0000)]
vf regs are not virtual (well not for our definition)
[r21728]
Matthias Braun [Fri, 5 Sep 2008 16:54:30 +0000 (16:54 +0000)]
new and improved regalloc verifier
[r21727]
Matthias Braun [Fri, 5 Sep 2008 16:54:15 +0000 (16:54 +0000)]
reset chunk ids for new funtions
[r21726]
Andreas Zwinkau [Fri, 5 Sep 2008 11:44:33 +0000 (11:44 +0000)]
Removed -after_pbqp_transform, because it is the same as -transformed
[r21719]
Christoph Mallon [Thu, 4 Sep 2008 19:53:05 +0000 (19:53 +0000)]
Fix typo in BE_REGISTER_MODULE_CONSTRUCTOR() user.
[r21711]
Matthias Braun [Thu, 4 Sep 2008 12:15:39 +0000 (12:15 +0000)]
simplified and improved
[r21699]
Michael Beck [Thu, 4 Sep 2008 12:11:27 +0000 (12:11 +0000)]
different cparser / eccp code
[r21698]
Matthias Braun [Thu, 4 Sep 2008 10:43:34 +0000 (10:43 +0000)]
remove a wrong and some unnecessary comments
[r21693]
Michael Beck [Thu, 4 Sep 2008 10:32:17 +0000 (10:32 +0000)]
strange difference between eccp and cparser
[r21692]
Christoph Mallon [Thu, 4 Sep 2008 09:32:17 +0000 (09:32 +0000)]
Rename -fluffig to -fsync, perform opt_sync() (formerly opt_ldst2()) last to not interfere with load store optimisation.
[r21689]
Christoph Mallon [Thu, 4 Sep 2008 08:57:47 +0000 (08:57 +0000)]
Indentation.
[r21686]
Matthias Braun [Thu, 4 Sep 2008 08:42:11 +0000 (08:42 +0000)]
fix the reload folding mode stuff in a more correct way
[r21685]
Matthias Braun [Thu, 4 Sep 2008 08:31:06 +0000 (08:31 +0000)]
revert last change
[r21684]
Matthias Braun [Thu, 4 Sep 2008 08:26:38 +0000 (08:26 +0000)]
don't extend load_mode when folding reloads
[r21683]
Andreas Zwinkau [Wed, 3 Sep 2008 14:18:29 +0000 (14:18 +0000)]
Use march=pentium3 to use if-conv
[r21671]
Christoph Mallon [Wed, 3 Sep 2008 13:50:49 +0000 (13:50 +0000)]
Fix conv skip in gen_lowered_64bit_shifts() (still necessary at all?).
[r21669]
Christoph Mallon [Wed, 3 Sep 2008 13:48:00 +0000 (13:48 +0000)]
Only skip Convs within the same register class when trying to create dest AM (is this still necessary at all?).
[r21668]
Christoph Mallon [Wed, 3 Sep 2008 13:02:06 +0000 (13:02 +0000)]
Fix fehler147: Do not deconv floats.
[r21666]
Matthias Braun [Wed, 3 Sep 2008 12:43:25 +0000 (12:43 +0000)]
simplify
[r21665]
Matthias Braun [Wed, 3 Sep 2008 12:42:46 +0000 (12:42 +0000)]
fehler147 (produces invisible lightning gun in quake3)
[r21664]
Sebastian Buchwald [Tue, 2 Sep 2008 21:17:41 +0000 (21:17 +0000)]
Added test case for Load-Add-Load conflict, that also seems to have missing skip patterns.
[r21657]
Christoph Mallon [Tue, 2 Sep 2008 14:06:54 +0000 (14:06 +0000)]
Use backtracking in find_original_value() for the flags emitter.
[r21655]
Michael Beck [Tue, 2 Sep 2008 13:45:35 +0000 (13:45 +0000)]
removed C99 feature
[r21653]
Michael Beck [Tue, 2 Sep 2008 13:42:39 +0000 (13:42 +0000)]
extending r21638: check sign when classifying float constants
[r21652]
Christoph Mallon [Tue, 2 Sep 2008 13:23:40 +0000 (13:23 +0000)]
Fix printing tarvals into long double.
[r21650]
Christoph Mallon [Tue, 2 Sep 2008 13:11:12 +0000 (13:11 +0000)]
Style fix.
[r21648]
Andreas Zwinkau [Tue, 2 Sep 2008 13:01:54 +0000 (13:01 +0000)]
Dump ins_permuted and cmp_unsigned attributes to vcg
[r21647]
Michael Beck [Tue, 2 Sep 2008 12:47:08 +0000 (12:47 +0000)]
removed C99 features
[r21646]
Michael Beck [Tue, 2 Sep 2008 12:35:32 +0000 (12:35 +0000)]
demonstartes a bug in the tarval module
[r21645]
Matthias Braun [Tue, 2 Sep 2008 11:54:51 +0000 (11:54 +0000)]
overspilling testcase by mallon
[r21644]
Matthias Braun [Tue, 2 Sep 2008 11:35:55 +0000 (11:35 +0000)]
fix too conservative delay taking step
[r21643]
Andreas Zwinkau [Tue, 2 Sep 2008 10:41:14 +0000 (10:41 +0000)]
New test cases for Mux and ia32_Test
[r21641]
Christoph Mallon [Tue, 2 Sep 2008 10:00:52 +0000 (10:00 +0000)]
Be nice: Use the async-signal safe _exit() instead of exit().
[r21640]
Christoph Mallon [Tue, 2 Sep 2008 09:47:13 +0000 (09:47 +0000)]
fehler143: eccp makes the __assert_fail() call in an assert() macro disappear.
[r21639]