Michael Beck [Fri, 1 Aug 2008 10:03:27 +0000 (10:03 +0000)]
fixed const
[r20887]
Michael Beck [Fri, 1 Aug 2008 09:56:22 +0000 (09:56 +0000)]
ptr add/sub test
[r20886]
Matthias Braun [Fri, 1 Aug 2008 09:24:36 +0000 (09:24 +0000)]
firm already had kill_node, no need to be_kill_node
[r20885]
Christoph Mallon [Fri, 1 Aug 2008 08:51:22 +0000 (08:51 +0000)]
Add two break; missing in r20857.
[r20881]
Christoph Mallon [Fri, 1 Aug 2008 08:14:28 +0000 (08:14 +0000)]
Skip a single downconv even if it has multiple users.
[r20878]
Christoph Mallon [Fri, 1 Aug 2008 07:47:33 +0000 (07:47 +0000)]
Include limits.h for LONG_{MAX,MIN}.
[r20877]
Moritz Kroll [Fri, 1 Aug 2008 02:16:23 +0000 (02:16 +0000)]
Testing helps... *yawn*
[r20875]
Moritz Kroll [Fri, 1 Aug 2008 01:53:20 +0000 (01:53 +0000)]
Implemented simple switch lowering
[r20872]
Michael Beck [Thu, 31 Jul 2008 22:19:53 +0000 (22:19 +0000)]
- removed a now superfluous function
[r20871]
Sebastian Buchwald [Thu, 31 Jul 2008 21:33:59 +0000 (21:33 +0000)]
Deactivate edges before freeing an irg.
[r20869]
Sebastian Buchwald [Thu, 31 Jul 2008 21:26:36 +0000 (21:26 +0000)]
Fixed more memory leaks.
[r20868]
Sebastian Buchwald [Thu, 31 Jul 2008 21:13:33 +0000 (21:13 +0000)]
typos
[r20867]
Sebastian Buchwald [Thu, 31 Jul 2008 21:02:34 +0000 (21:02 +0000)]
Fixed revision 20864.
[r20866]
Michael Beck [Thu, 31 Jul 2008 20:58:58 +0000 (20:58 +0000)]
combo test case
[r20865]
Sebastian Buchwald [Thu, 31 Jul 2008 20:28:49 +0000 (20:28 +0000)]
Fixed one more memory leak.
[r20864]
Moritz Kroll [Thu, 31 Jul 2008 20:26:19 +0000 (20:26 +0000)]
Panic on switch size bigger than 256000... WTF?!
[r20863]
Sebastian Buchwald [Thu, 31 Jul 2008 20:15:39 +0000 (20:15 +0000)]
Fixed one more memory leak.
[r20862]
Michael Beck [Thu, 31 Jul 2008 19:26:42 +0000 (19:26 +0000)]
- allow add and sub with reference values
[r20857]
Sebastian Buchwald [Thu, 31 Jul 2008 19:18:09 +0000 (19:18 +0000)]
Call firm_be_finish, but who calls me?
[r20856]
Sebastian Buchwald [Thu, 31 Jul 2008 19:16:53 +0000 (19:16 +0000)]
Added function firm_be_finish.
[r20855]
Sebastian Buchwald [Thu, 31 Jul 2008 19:15:44 +0000 (19:15 +0000)]
Call free functions.
[r20854]
Michael Beck [Thu, 31 Jul 2008 18:47:00 +0000 (18:47 +0000)]
- use double-linked lists instead of flexible arrays for the
split_by_what() function
- BugFix: do not replace the Bad node
- BugFix: wrong var used
[r20853]
Sebastian Buchwald [Thu, 31 Jul 2008 18:19:40 +0000 (18:19 +0000)]
Fixed one more memory leak.
[r20851]
Sebastian Buchwald [Thu, 31 Jul 2008 18:18:06 +0000 (18:18 +0000)]
typo
[r20850]
Sebastian Buchwald [Thu, 31 Jul 2008 18:10:09 +0000 (18:10 +0000)]
Fixed one more memory leak.
[r20849]
Moritz Kroll [Thu, 31 Jul 2008 18:03:49 +0000 (18:03 +0000)]
Added Beck's VC obstack version. Updated VC project
[r20847]
Michael Beck [Thu, 31 Jul 2008 14:41:53 +0000 (14:41 +0000)]
- BugFix: when splitting by input, ensure than Z is split by ALL inputs
- BugFix: to detect that a Cmp must be reavaluated, we give the Cmp node the type true IF there is a chance that it could be evaluated
[r20841]
Christoph Mallon [Thu, 31 Jul 2008 14:13:14 +0000 (14:13 +0000)]
GC attributes left over by r20834.
[r20838]
Matthias Braun [Thu, 31 Jul 2008 12:52:31 +0000 (12:52 +0000)]
- reverted the "small, very small" optimisation, because:
* The underlying idea is simply wrong
* the is_inside function was wrong too, so the wrong optimisation nearly never triggered
[r20834]
Matthias Braun [Thu, 31 Jul 2008 09:17:21 +0000 (09:17 +0000)]
fix warning
[r20828]
Matthias Braun [Thu, 31 Jul 2008 09:17:10 +0000 (09:17 +0000)]
generalize support for segments
[r20827]
Matthias Braun [Wed, 30 Jul 2008 18:34:01 +0000 (18:34 +0000)]
fix/improve peephole
[r20820]
Matthias Braun [Wed, 30 Jul 2008 18:32:14 +0000 (18:32 +0000)]
use peephoole_exchange, IncSP->Push should use ia32_Unknown and not ProduceVal
[r20819]
Michael Beck [Wed, 30 Jul 2008 13:16:46 +0000 (13:16 +0000)]
fehler119 is C99
[r20806]
Matthias Braun [Wed, 30 Jul 2008 12:51:51 +0000 (12:51 +0000)]
simply deactivate edges instead of asserting it
[r20803]
Michael Beck [Wed, 30 Jul 2008 11:58:40 +0000 (11:58 +0000)]
- inline_method moves copied constants to start block yet
[r20799]
Matthias Braun [Wed, 30 Jul 2008 11:22:53 +0000 (11:22 +0000)]
improved fak.c
[r20798]
Matthias Braun [Wed, 30 Jul 2008 11:21:31 +0000 (11:21 +0000)]
cosmetics
[r20797]
Matthias Braun [Wed, 30 Jul 2008 11:20:21 +0000 (11:20 +0000)]
don't prefer 1 cpu architecture over another (and wonder why we don't construct inc/dec)
[r20796]
Michael Beck [Wed, 30 Jul 2008 09:59:02 +0000 (09:59 +0000)]
fix new fallthrough handling
[r20793]
Christoph Mallon [Wed, 30 Jul 2008 09:56:18 +0000 (09:56 +0000)]
Sort the inputs of a Sync before it is being created to aid CSE.
[r20792]
Christoph Mallon [Wed, 30 Jul 2008 09:55:44 +0000 (09:55 +0000)]
Reduce worst time complexity of mem parallelisation from O(2**n) to O(n**2).
[r20791]
Christoph Mallon [Wed, 30 Jul 2008 09:52:08 +0000 (09:52 +0000)]
panic() on internal error instead of assert(0)ing.
[r20790]
Christoph Mallon [Wed, 30 Jul 2008 09:51:20 +0000 (09:51 +0000)]
Remove handling for 32bit as smaller mode in emit_ia32_Conv_I2I(), because it is not sensible.
[r20789]
Christoph Mallon [Wed, 30 Jul 2008 09:43:58 +0000 (09:43 +0000)]
Replace be_peephole_before_exchange()+sched_remove()+exchange()+be_peephole_new_node() by the new function be_peephole_exchange().
[r20788]
Matthias Braun [Wed, 30 Jul 2008 09:23:39 +0000 (09:23 +0000)]
irprofile fix
[r20787]
Matthias Braun [Wed, 30 Jul 2008 09:21:43 +0000 (09:21 +0000)]
- refactor emitter next/prev block handling a bit
- only emit ret $0 on athlon if block has a label (= there is at least 1 jump
to the block)
[r20786]
Christoph Mallon [Wed, 30 Jul 2008 08:46:02 +0000 (08:46 +0000)]
Remove last remaining SCHEDULE_PROJS.
[r20785]
Christoph Mallon [Wed, 30 Jul 2008 07:15:32 +0000 (07:15 +0000)]
Fix fehler120: Do only skip int->int, not float->int, Convs before the shift amount.
[r20782]
Michael Beck [Wed, 30 Jul 2008 01:12:52 +0000 (01:12 +0000)]
fixed the parameter
[r20781]
Michael Beck [Wed, 30 Jul 2008 00:59:31 +0000 (00:59 +0000)]
- do not kill the start block when folding control flow
[r20778]
Michael Beck [Wed, 30 Jul 2008 00:58:43 +0000 (00:58 +0000)]
- opt_tail_recursion() must run before optimize_funccalls():
Add this not-unimportant info to the doxygen docu
- typo fixed
- add a set_using_irn_link()/clear_using_irn_link() pair
[r20777]
Michael Beck [Wed, 30 Jul 2008 00:38:46 +0000 (00:38 +0000)]
the famous Factorial example
[r20776]
Michael Beck [Wed, 30 Jul 2008 00:32:10 +0000 (00:32 +0000)]
- commented out the wrong line
[r20775]
Michael Beck [Wed, 30 Jul 2008 00:28:05 +0000 (00:28 +0000)]
- add compute_Confirm(), compute_Bad() and compute_Unknown()
- fold the control flow by removing unreachable inputs using the analysis info
[r20774]
Christoph Mallon [Tue, 29 Jul 2008 21:36:29 +0000 (21:36 +0000)]
fehler120: Backend discards float->int Conv for shift amount.
[r20772]
Christoph Mallon [Tue, 29 Jul 2008 21:35:22 +0000 (21:35 +0000)]
Add main().
[r20771]
Michael Beck [Tue, 29 Jul 2008 19:55:51 +0000 (19:55 +0000)]
- BugFix: of by one error
[r20769]
Moritz Kroll [Tue, 29 Jul 2008 19:45:41 +0000 (19:45 +0000)]
fehler119: crash in edg fe for C99 sizeof.
[r20768]
Michael Beck [Tue, 29 Jul 2008 19:27:17 +0000 (19:27 +0000)]
- using tarval_top as an alias for tarval_unreachable simplifies code
- work-around: when the inputs of Sub/Cmp are congruent BUT its type is already a constant, set the type to BOTTOM: I can not understand how it converges WITHOUT this fix
[r20767]
Matthias Braun [Tue, 29 Jul 2008 15:05:31 +0000 (15:05 +0000)]
don't do the store-float-conv optimisation with double->unsigned
[r20766]
Matthias Braun [Tue, 29 Jul 2008 15:05:24 +0000 (15:05 +0000)]
don't do the store-float-conv optimisation with double->unsigned
[r20765]
Michael Beck [Tue, 29 Jul 2008 14:37:39 +0000 (14:37 +0000)]
Add NaN condition to Sub AND Cmp
[r20763]
Michael Beck [Tue, 29 Jul 2008 14:37:00 +0000 (14:37 +0000)]
combo error
[r20762]
Sebastian Buchwald [Tue, 29 Jul 2008 14:13:27 +0000 (14:13 +0000)]
Added simplified test case for conversion double2int.
[r20760]
Michael Beck [Tue, 29 Jul 2008 13:14:05 +0000 (13:14 +0000)]
- BugFix: constants must be evaluated BEFORE congruence checks
[r20758]
Michael Beck [Tue, 29 Jul 2008 11:55:18 +0000 (11:55 +0000)]
fixed new type_or_ent type handling
[r20756]
Michael Beck [Tue, 29 Jul 2008 11:49:04 +0000 (11:49 +0000)]
- BugFix: fixed the "type is neither Top nor constant" condition causing U,R nodes node now put on the touched list
- add test for monotonic behavior
- fixed typo
[r20755]
Matthias Braun [Tue, 29 Jul 2008 10:25:46 +0000 (10:25 +0000)]
make testcase return 0 in main
[r20752]
Matthias Braun [Tue, 29 Jul 2008 09:30:49 +0000 (09:30 +0000)]
fix/improve proj generation
[r20751]
Matthias Braun [Tue, 29 Jul 2008 09:10:20 +0000 (09:10 +0000)]
cleanups/fixes for ASM handling
[r20749]
Michael Beck [Mon, 28 Jul 2008 14:37:49 +0000 (14:37 +0000)]
- BugFix: Proj nodes must compute TOP if they are in a unreachable Block
[r20739]
Matthias Braun [Mon, 28 Jul 2008 12:41:20 +0000 (12:41 +0000)]
add an asm test with an inout parameter
[r20734]
Michael Beck [Mon, 28 Jul 2008 11:48:39 +0000 (11:48 +0000)]
- improved doxygen comments about edges_activate() and the dep API
[r20732]
Michael Beck [Mon, 28 Jul 2008 11:43:21 +0000 (11:43 +0000)]
- add an assert tat checks for backend phase when added dep edges
- edges_activate()/edges_deactivate() handles DEP edges only in BACKEND phase
[r20731]
Andreas Zwinkau [Mon, 28 Jul 2008 11:01:55 +0000 (11:01 +0000)]
Adding some test cases used for building the PBQP code selection
[r20729]
Sebastian Buchwald [Mon, 28 Jul 2008 10:45:41 +0000 (10:45 +0000)]
added support for ASM nodes
[r20728]
Matthias Braun [Mon, 28 Jul 2008 09:15:52 +0000 (09:15 +0000)]
forgot to adjust mips backend for asm stuff
[r20727]
Matthias Braun [Mon, 28 Jul 2008 09:13:52 +0000 (09:13 +0000)]
public API for asm constraints
[r20726]
Matthias Braun [Mon, 28 Jul 2008 09:13:28 +0000 (09:13 +0000)]
- handle parsing of assembler constraints in backends. Provide functions for
frontend to determine what is input/output/memory operand
- prefix some globally visible ia32 functions with ia32_
[r20725]
Matthias Braun [Mon, 28 Jul 2008 09:10:36 +0000 (09:10 +0000)]
accept directories or single tests as parameter
[r20724]
Matthias Braun [Mon, 28 Jul 2008 09:10:16 +0000 (09:10 +0000)]
fix for latest makereport
[r20723]
Michael Beck [Sat, 26 Jul 2008 22:52:36 +0000 (22:52 +0000)]
BugFix:
- due to the way we implement compute_Phi)(, we must place all Phi's
of a block on the cprop list if the block is placed
- implemented local cprop list as double-linked list: this ensures the
fifo character of this list (is it really needed? It might generate
lesser rounds)
[r20714]
Michael Beck [Sat, 26 Jul 2008 21:47:51 +0000 (21:47 +0000)]
- BugFix: must check the input of the user node y, not def node x
- Opcode splits Sel nodes by its entity
- do not replace Bad nodes
[r20713]
Michael Beck [Sat, 26 Jul 2008 20:08:58 +0000 (20:08 +0000)]
fixed warnings
[r20711]
Michael Beck [Sat, 26 Jul 2008 11:19:13 +0000 (11:19 +0000)]
- BugFix: a node might fall 2 times, so we need a on_fallen flag
- BugFix: the "Opcode" of a Proj node must contains its projection number, else Projs with different numbers might be congruent ...
[r20709]
Michael Beck [Sat, 26 Jul 2008 11:16:28 +0000 (11:16 +0000)]
- BugFix: Dep-Edges where calculated WRONG if the node on which another depends is dead ...
- removed useless if
[r20708]
Michael Beck [Sat, 26 Jul 2008 11:14:04 +0000 (11:14 +0000)]
typo fixed
[r20707]
Michael Beck [Sat, 26 Jul 2008 11:12:54 +0000 (11:12 +0000)]
include MS debug heap by default, so debug macros can be called from everywhere
[r20706]
Michael Beck [Sat, 26 Jul 2008 08:16:55 +0000 (08:16 +0000)]
BugFixes:
- nodes must be initialized with Top not Bottom
- right order for mode_X and mode_BB nodes
- default_compute() assigns now U, R to mode_X nodes
- cannot handle symconst_addr_name yet
- compute_Phi() must check the predecessor block for U, because Firm constants live in the start block and are always != Top ...
- removed the late on_* removement: its assumptions were wrong
- set the node for newly created IR-nodes
Improvements:
- ensure reachable End and EndBlock in apply_result()
- can now replace ProjX(Cond) nodes by Jmp or Bad
Finally runs on Clicks example 3.1 :-)
[r20705]
Michael Beck [Sat, 26 Jul 2008 08:06:04 +0000 (08:06 +0000)]
- BugFix: although tarval_b_false and tarval_b_true are reserved tarvals, they are valid constants
[r20704]
Michael Beck [Sat, 26 Jul 2008 05:36:46 +0000 (05:36 +0000)]
- first working combo version (at least one example :-), no identities yet
[r20703]
Michael Beck [Sat, 26 Jul 2008 05:35:52 +0000 (05:35 +0000)]
- add tarval_is_constant() to check whether a tarval represents a constant or an reserved value
- add explicit mode_X tarval_unreachable and tarval_reachable tarvals needed for combo
- fixed a doxygen comment
[r20702]
Sebastian Buchwald [Fri, 25 Jul 2008 14:49:08 +0000 (14:49 +0000)]
handle Unknown nodes
[r20695]
Michael Beck [Fri, 25 Jul 2008 13:51:29 +0000 (13:51 +0000)]
add sorting of out edges, some small fixes
[r20693]
Michael Beck [Fri, 25 Jul 2008 13:08:19 +0000 (13:08 +0000)]
- out edges now preserve the input position, needed for combo.c()
[r20692]
Andreas Zwinkau [Fri, 25 Jul 2008 09:22:55 +0000 (09:22 +0000)]
Moved assertion into if clause
It is not necessary to assert anything about lsmode, if
it isn't used anyways.
[r20683]
Michael Beck [Fri, 25 Jul 2008 00:07:59 +0000 (00:07 +0000)]
- fixed type_or_ent type: get rod of casts
[r20678]
Michael Beck [Thu, 24 Jul 2008 14:22:25 +0000 (14:22 +0000)]
- implemented compute_Proj_Cond(), compute_Add()
- fixed initial state
- simplyfied def_compute()
[r20671]
Michael Beck [Thu, 24 Jul 2008 12:41:11 +0000 (12:41 +0000)]
- add prototype for combo()
- fixed gvn-pre comment (not buggy anymore)
[r20665]