80cbed4f61
Another nice speedup for the register allocator. This time, we replace
...
the Virt2PhysRegMap std::map with an std::vector. This speeds up the
register allocator another (almost) 40%, from .72->.45s in a release build
of LLC on 253.perlbmk.
llvm-svn: 11219
2004-02-09 02:12:04 +00:00
56f9b190e1
Add a new (hidden) option that is useful for profiling.
...
llvm-svn: 11218
2004-02-09 01:47:10 +00:00
bbd9a43d20
Ugh, perform an optimization that GCC should be able to do itself. This
...
speeds up livevar from .48/.32s -> .45/.31s in LLC on perlbmk
llvm-svn: 11217
2004-02-09 01:43:23 +00:00
26407384ec
Only do stuff for the REAL number of physical registers we have, not 1024.
...
This speeds up live variables a lot, from .60/.39s -> .47/.26s in LLC, for
the first/second pass respectively.
llvm-svn: 11216
2004-02-09 01:35:21 +00:00
490627a404
Change the PhysRegsUsed map into a dense array. Seeing that this is a mapping
...
from physical registers, and they are always dense, it makes sense to not have
a ton of RBtree overhead. This change speeds up regalloclocal about ~30% on
253.perlbmk, from .35s -> .27s in the JIT (in LLC, it goes from .74 -> .55).
Now live variable analysis is the slowest codegen pass. Of course it doesn't
help that we have to run it twice, because regalloclocal doesn't update it,
but even if it did it would be the slowest pass (now it's just the 2x slowest
pass :(
llvm-svn: 11215
2004-02-09 01:26:13 +00:00
2eaf58775e
Two problems with these lines of code:
...
1. The "work" was not in the assert, so it was punishing the optimized release
2. getNamedFunction is _very_ expensive in large programs. It is not designed
to be used like this, and was taking 7% of the execution time of the code
generator on perlbmk.
Since the assert "can never fail", I'm just killing it.
llvm-svn: 11214
2004-02-09 00:59:07 +00:00
57ea2e3294
The ConstantExpr::getCast call can cause a CPR to be generated. If so,
...
strip it off.
llvm-svn: 11213
2004-02-09 00:20:55 +00:00
5b6892e37c
Fix PR215: [bcwriter] Problem compactifying ConstantPointerRefs
...
Have I ever mentioned how much I _hate_ constantpointerrefs?
llvm-svn: 11212
2004-02-09 00:15:41 +00:00
3480e935d0
Fix grammar-o.
...
llvm-svn: 11210
2004-02-08 22:27:33 +00:00
c5b7b24949
All of spec, including 253.perlbmk should now work. There is still the 254.gap
...
regression, but that will definitely be fixed by 1.2
llvm-svn: 11209
2004-02-08 22:23:33 +00:00
3b7f6b2217
Improve compatibility with programs that already have a prototype for 'write',
...
even if it is wierd in some way.
llvm-svn: 11207
2004-02-08 22:14:44 +00:00
85195147e1
vi failed me again. :)
...
llvm-svn: 11206
2004-02-08 21:52:30 +00:00
144f0efd4c
Rename the invoke 'except' destination to the 'unwind' destination
...
llvm-svn: 11205
2004-02-08 21:52:04 +00:00
17235712fb
Change the 'exception' destination to the 'unwind' destination. We will always
...
allow 'except' instead of 'unwind' here though.
llvm-svn: 11203
2004-02-08 21:48:25 +00:00
fae8ab3088
rename the "exceptional" destination of an invoke instruction to the 'unwind' dest
...
llvm-svn: 11202
2004-02-08 21:44:31 +00:00
35630adf54
Add new bug, add note about llvm-db
...
llvm-svn: 11201
2004-02-08 21:20:42 +00:00
56997dd283
Fix PR225: [pruneeh] -pruneeh pass removes invoke instructions it shouldn't
...
llvm-svn: 11200
2004-02-08 21:15:59 +00:00
071bc60450
splitBasicBlock "does the right thing" now, no reason to reposition it.
...
llvm-svn: 11199
2004-02-08 20:49:07 +00:00
50ceee1ead
Bug fixed
...
llvm-svn: 11198
2004-02-08 19:59:05 +00:00
108cadc274
Implement proper invoke/unwind lowering.
...
This fixed PR16 "[lowerinvoke] The -lowerinvoke pass does not insert calls to setjmp/longjmp"
llvm-svn: 11195
2004-02-08 19:53:56 +00:00
8a9da302a8
New testcase, code generators should support invoke/unwind
...
llvm-svn: 11192
2004-02-08 19:40:58 +00:00
dc631735e6
Print out all globals as they are emitted, not just those emitted from
...
emitGlobals
llvm-svn: 11191
2004-02-08 19:33:23 +00:00
6604615a16
There is no reason to #define fd
...
llvm-svn: 11190
2004-02-08 19:33:07 +00:00
d7b895a0c5
add pr105
...
llvm-svn: 11189
2004-02-08 18:49:13 +00:00
3e666c1897
Fix unterminated-quote typo in echo command
...
llvm-svn: 11188
2004-02-08 08:01:00 +00:00
8613f13cc3
Document automake as a build dependency because we use aclocal
...
llvm-svn: 11187
2004-02-08 07:49:04 +00:00
b921039ee9
Move rules for regenerating autoconf files into this shell script.
...
llvm-svn: 11186
2004-02-08 07:44:48 +00:00
4f99c4f75f
Remove rules for autoconf'ing files from top-level Makefile
...
llvm-svn: 11185
2004-02-08 07:44:30 +00:00
476488e669
Add a call to 'write' right before the call to abort() in the unwind path.
...
This causes the JIT, or LLC'd program to print out a nice message, explaining
WHY the program aborted.
llvm-svn: 11184
2004-02-08 07:30:29 +00:00
d97d8faa17
Fix indentation of selectiondag.
...
I don't know why its indentation has been bugging me, but it has.
llvm-svn: 11180
2004-02-08 05:49:29 +00:00
fa616e14fc
Add one that I missed
...
llvm-svn: 11179
2004-02-08 01:53:10 +00:00
9ab85275ef
Instead of callign removeTriviallyDeadNodes on the global graph every time
...
removeDeadNodes is called, only call it at the end of the pass being run.
This saves 1.3 seconds running DSA on 177.mesa (5.3->4.0s), which is
pretty big. This is only possible because of the automatic garbage
collection done on forwarding nodes.
llvm-svn: 11178
2004-02-08 01:51:48 +00:00
e1531f3eb1
Remove another unneeded call.
...
llvm-svn: 11177
2004-02-08 01:40:40 +00:00
985282c78e
This call is no longer needed now that merging does not produce garbage
...
llvm-svn: 11176
2004-02-08 01:38:34 +00:00
b0f32183e4
Substantially improve the DSA code by removing 'forwarding' nodes from
...
DSGraphs while they are forwarding. When the last reference to the forwarding
node is dropped, the forwarding node is autodeleted. This should simplify
removeTriviallyDead nodes, and is only (efficiently) possible because we are
using an ilist of dsnodes now.
llvm-svn: 11175
2004-02-08 01:27:18 +00:00
5592af7927
Bugfix for ilist conversion. The ilist wants to make an 'end' node which has
...
G == 0
llvm-svn: 11174
2004-02-08 01:05:37 +00:00
f3861ce4f6
Switch the Nodes list from being an std::vector<DSNode*> to an ilist<DSNode>
...
llvm-svn: 11173
2004-02-08 00:53:26 +00:00
74a99554d1
One of the 'annoying' things about ilists is that the iterators don't behave
...
quite the same as for non-intrusive lists of pointers to nodes. To support
transitioning code bases, add a new 'compatibility' iterator.
llvm-svn: 11172
2004-02-08 00:51:31 +00:00
bd16efb88f
Change to use node_iterators instead of direct access to Nodes
...
llvm-svn: 11171
2004-02-08 00:23:16 +00:00
d9a063c9a2
Change to use iterators instead of direct access
...
llvm-svn: 11170
2004-02-08 00:22:41 +00:00
8c146638b6
Minor change to breakpoint (lack of) support
...
llvm-svn: 11169
2004-02-08 00:06:20 +00:00
ce74c2f50b
getNodes() is gone, use node_begin/end instead
...
Rename stats from dsnode -> dsa
Add a new stat
llvm-svn: 11167
2004-02-07 23:58:05 +00:00
06834e6d7b
getNodes() is gone
...
llvm-svn: 11166
2004-02-07 23:57:26 +00:00
7b3d404e28
Abstract out the Nodes collection. Instead of providing a getNodes() method,
...
provide node_begin/end iterators, which are only guaranteed to be
bidirectional, not random access.
llvm-svn: 11165
2004-02-07 23:57:09 +00:00
6e69a5aa5d
Actually USE isForwarding method
...
llvm-svn: 11160
2004-02-07 23:02:32 +00:00
e37a472ef0
As Alkis pointed out to me, I forgot to commit this... :(
...
llvm-svn: 11159
2004-02-07 22:54:19 +00:00
40862c7578
There is no need to clone over nodes that are going to be dead anyway
...
llvm-svn: 11157
2004-02-07 22:00:03 +00:00
92affee3e7
Use autoconf answers from config.h (FIXME, should autoconf this file
...
directly instead).
Fix LockHolder/MutexLocker typo.
llvm-svn: 11156
2004-02-06 22:33:17 +00:00
a1d7fcb15f
Add a missing link
...
llvm-svn: 11154
2004-02-06 20:27:41 +00:00
b596679737
Fix grammar.
...
llvm-svn: 11153
2004-02-06 18:40:35 +00:00
Chris Lattner