cd2e7ffcc6
Eliminate uneeded lookups by passing a Virt2PhysMap::iterator instead
...
of the virtual register to certain functions.
llvm-svn: 11143
2004-02-06 03:15:40 +00:00
2dd1c8d8ce
Fix another dominator update bug. These bugs keep getting exposed because GCSE
...
keeps finding more code motion opportunities now that the dominators are correct!
llvm-svn: 11142
2004-02-05 23:20:59 +00:00
a97abeb342
Change live interval representation. Machine instructions now have two
...
slots each. As a concequence they get numbered as 0, 2, 4 and so
on. The first slot is used for operand uses and the second for
defs. Here's an example:
0: A = ...
2: B = ...
4: C = A + B ;; last use of A
The live intervals should look like:
A = [1, 5)
B = [3, x)
C = [5, y)
llvm-svn: 11141
2004-02-05 22:55:25 +00:00
c0c953f0bc
Fix bug updating dominators
...
llvm-svn: 11140
2004-02-05 22:33:26 +00:00
f978c421e5
Add debug output
...
llvm-svn: 11139
2004-02-05 22:33:19 +00:00
aaaaec43e7
Bug fixed
...
llvm-svn: 11138
2004-02-05 21:16:42 +00:00
14ab84a483
Fix PR223: Loopsimplify incorrectly updates dominator information
...
The problem is that the dominator update code didn't "realize" that it's
possible for the newly inserted basic block to dominate anything. Because
it IS possible, stuff was getting updated wrong.
llvm-svn: 11137
2004-02-05 21:12:24 +00:00
eba3c36f99
New testcase for PR223: Loopsimplify incorrectly updates dominator information
...
llvm-svn: 11136
2004-02-05 21:11:38 +00:00
1165c1081e
We don't need to scan the blocks that we are live-in on every
...
access. Rather we only have to do it on the creation of the interval.
llvm-svn: 11135
2004-02-05 20:45:40 +00:00
66c16d144e
s/gcse/scalarrepl
...
llvm-svn: 11133
2004-02-05 17:31:37 +00:00
e3f247bcac
In a "seeing the forest through the trees" kinda situation, I realized that a
...
complete rewrite of load-vn will make it a bit faster. This changes speeds up
the gcse pass (which uses load-vn) from 25.45s to 0.42s on the testcase in
PR209.
I've also verified that this gives the exact same results as the old one.
llvm-svn: 11132
2004-02-05 17:20:00 +00:00
d3e4d81ad7
This is a big diff with no functionality change. We just reorder some code,
...
which causes big reindentation. While I'm at it, I fix the fixme by removing
some dead code.
llvm-svn: 11131
2004-02-05 05:56:23 +00:00
90b9043ea5
finegrainify namespacification
...
llvm-svn: 11130
2004-02-05 05:51:40 +00:00
dd0c255d14
Added missing include.
...
llvm-svn: 11129
2004-02-05 05:04:39 +00:00
cb435541c4
Fixed Chris' typo.
...
llvm-svn: 11128
2004-02-05 04:45:21 +00:00
f127d9cf33
qoi bug fixed
...
llvm-svn: 11127
2004-02-05 00:48:41 +00:00
5bab6301ac
Implement optimizations for handling large basic blocks.
...
llvm-svn: 11126
2004-02-05 00:36:43 +00:00
327426411e
Modify the two address instruction pass to remove the duplicate
...
operand of the instruction and thus simplify the register allocation.
llvm-svn: 11124
2004-02-04 22:17:40 +00:00
39ad6f2772
Minor speedup, don't query ValueMap each time through the loop
...
llvm-svn: 11123
2004-02-04 21:44:26 +00:00
fe66239860
Always replace instead of appending when creating archive files. It may be
...
slightly slower, but I think we can handle it, especially if it means
BytecodeLibs are correctly regenerated.
llvm-svn: 11122
2004-02-04 21:41:23 +00:00
9ce98a7e9f
Include <iosfwd> and <string> instead of <iostream>.
...
Take away the default iostream argument of createMachineFunctionPrinterPass(),
at Chris's request.
llvm-svn: 11121
2004-02-04 21:41:10 +00:00
94b5c3cb91
Take away the default iostream argument of createMachineFunctionPrinterPass(),
...
at Chris's request.
llvm-svn: 11120
2004-02-04 21:41:01 +00:00
5e442bb3f1
Clean out DESTLIBBYTECODE when making clean in runtime.
...
llvm-svn: 11119
2004-02-04 21:40:49 +00:00
6f8865bf9f
Two changes:
...
1. Don't scan to the end of alloca instructions in the caller function to
insert inlined allocas, just insert at the top. This saves a lot of
time inlining into functions with a lot of allocas.
2. Use splice to move the alloca instructions over, instead of remove/insert.
This allows us to transfer a block at a time, and eliminates a bunch of
silly symbol table manipulations.
This speeds up the inliner on the testcase in PR209 from 1.73s -> 1.04s (67%)
llvm-svn: 11118
2004-02-04 21:33:42 +00:00
dbf4b42fde
IMULri* instructions do not require their first two registers operands
...
to be the same (IOW they are not two address instructions).
llvm-svn: 11117
2004-02-04 17:21:04 +00:00
0fa8c7c321
Optimize the case where we are inlining a function that contains only one basic block,
...
and that basic block ends with a return instruction. In this case, we can just splice
the cloned "body" of the function directly into the source basic block, avoiding a lot
of rearrangement and splitBasicBlock's linear scan over the split block. This speeds up
the inliner on the testcase in PR209 from 2.3s to 1.7s, a 35% reduction.
llvm-svn: 11116
2004-02-04 04:17:06 +00:00
4babe055d3
Check in header file I forgot before.
...
llvm-svn: 11115
2004-02-04 03:59:08 +00:00
8d414ad035
Adjust to the new BasicBlock ctor, which requires a function parameter
...
llvm-svn: 11114
2004-02-04 03:58:28 +00:00
bb5f0db8d6
Adjust to the new BB ctor
...
llvm-svn: 11113
2004-02-04 03:57:50 +00:00
040ad3afa8
Delete the BasicBlock ctor that only takes a BasicBlock to insert before. This
...
fails when the basic block points to the function->end. Instead, require that
the client pass in the function AND the basicblock to insert into.
llvm-svn: 11112
2004-02-04 03:57:34 +00:00
0ff9da5fed
Remove unneeded code now that splitBasicBlock does the "right thing"
...
llvm-svn: 11111
2004-02-04 03:21:51 +00:00
9cda4bf8e5
When splitting a basic block, insert the new half immediately after the first
...
half.
llvm-svn: 11110
2004-02-04 03:21:31 +00:00
18ef3fda57
More refactoring. Move alloca instructions and handle invoke instructions
...
before we delete the original call site, allowing slight simplifications of
code, but nothing exciting.
llvm-svn: 11109
2004-02-04 02:51:48 +00:00
9fc977eac4
Move the cloning of the function body much earlier in the inlinefunction
...
process. The only optimization we did so far is to avoid creating a
PHI node, then immediately destroying it in the common case where the
callee has one return statement. Instead, we just don't create the return
value. This has no noticable performance impact, but paves the way for
future improvements.
llvm-svn: 11108
2004-02-04 01:41:09 +00:00
a6578ef318
Give CloneBasicBlock an optional function argument to specify which function
...
to add the cloned block to. This allows the block to be added to the function
immediately, and all of the instructions to be immediately added to the function
symbol table, which speeds up the inliner from 3.7 -> 3.38s on the PR209.
llvm-svn: 11107
2004-02-04 01:19:43 +00:00
d8a232b716
If changing a parent, don't add then remove the object from the leak detector
...
llvm-svn: 11106
2004-02-04 01:06:38 +00:00
c4c7ea5288
In BasicBlock::splitBasicBlock, just use islist::splice to move the instructions,
...
instead of a loop that is really inefficient with large basic blocks.
This speeds up the inliner pass on the testcase in PR209 from 13.8s to 2.24s
which still isn't exactly speedy, but is a lot better. :)
llvm-svn: 11105
2004-02-03 23:11:21 +00:00
135dcc024b
Revised version of a patch for PR222 that works with my version of gmake.
...
llvm-svn: 11104
2004-02-03 23:05:24 +00:00
3d96eb6fa6
Bug fixed
...
llvm-svn: 11103
2004-02-03 22:59:56 +00:00
6bd75a6223
Apply Michael Kahl's patch to fix the 'make tools-only' target when OBJDIR != SRCDIR.
...
See PR222.
llvm-svn: 11102
2004-02-03 22:56:40 +00:00
ae51cae111
Bunch up all locally used allocas by the block they are allocated in, and
...
process them all as a group. This speeds up SRoA/mem2reg from 28.46s to
0.62s on the testcase from PR209.
llvm-svn: 11100
2004-02-03 22:34:12 +00:00
3784188620
Handle extremely trivial cases extremely efficiently. This speeds up
...
SRoA/mem2reg from 41.2s to 27.5s on the testcase in PR209.
llvm-svn: 11099
2004-02-03 22:00:33 +00:00
d1b1992495
Generate ftst instructions for comparison against zero
...
llvm-svn: 11098
2004-02-03 18:54:04 +00:00
8a63001939
bug fixed
...
llvm-svn: 11097
2004-02-03 08:03:41 +00:00
3c8c72c54f
Add the ftst instruction
...
llvm-svn: 11095
2004-02-03 07:27:50 +00:00
8161306139
Add support for one argument OneArgFP instructions
...
llvm-svn: 11094
2004-02-03 07:27:34 +00:00
3d100ef0ca
When an instruction like: A += B had both A and B virtual registers
...
spilled, A was loaded from its stack location twice. This fixes the bug.
llvm-svn: 11093
2004-02-03 01:13:07 +00:00
74b26f77a3
Revert changes. Will implement this using a different set of primitives
...
llvm-svn: 11091
2004-02-02 23:08:58 +00:00
6e34205653
Fix debugging output.
...
llvm-svn: 11088
2004-02-02 22:00:32 +00:00
6471e66fbb
Correctly update def/use information for modified machine operands.
...
llvm-svn: 11087
2004-02-02 21:56:40 +00:00
5f03f684ed
Add MachineOperand::setDef() and MachineOperand::setUse() so that the
...
TwoAddressInstructionPass can correctly update use/def information.
llvm-svn: 11086
2004-02-02 21:55:18 +00:00
86ff30deab
Should be more careful. The previously applied change made all counts
...
outside of loops = 0.
llvm-svn: 11085
2004-02-02 20:29:57 +00:00
bcade4b8e6
Floating point negates are -0.0 - X, not 0.0 - X
...
llvm-svn: 11084
2004-02-02 20:21:29 +00:00
c2f0aa58df
Disable (x - (y - z)) => (x + (z - y)) optimization for floating point.
...
llvm-svn: 11083
2004-02-02 20:09:56 +00:00
cacd30b957
Update comment
...
llvm-svn: 11082
2004-02-02 20:09:22 +00:00
63b61e8739
No need to declare implicit uses/defs of ST0
...
llvm-svn: 11081
2004-02-02 19:57:45 +00:00
d932cbf613
Bug fixed
...
llvm-svn: 11080
2004-02-02 19:40:06 +00:00
6204e75c4a
Make deadarghaX0r warning louder.
...
(I just love typing haX0r. haX0r haX0r haX0r.)
llvm-svn: 11079
2004-02-02 19:32:27 +00:00
30d26ac561
Generate the fchs instruction to negate a floating point number
...
llvm-svn: 11078
2004-02-02 19:31:38 +00:00
7af8ad6444
Add support for OneArgFPRW instructions, fix a couple of typeos
...
llvm-svn: 11077
2004-02-02 19:23:15 +00:00
ac2ced8554
Include PowerPC.h. Flesh out the stub versions of addPassesToEmitAssembly()
...
and addPassesToJITCompile() slightly.
llvm-svn: 11076
2004-02-02 19:06:36 +00:00
57b8ef2693
Add a -march=powerpc option. Automatically select it if this looks like a
...
big-endian, 32-bit module, or if __ppc__, __POWERPC__, or __APPLE__ are
defined.
llvm-svn: 11075
2004-02-02 19:06:12 +00:00
53f4c48393
Link in the PowerPC back-end.
...
llvm-svn: 11074
2004-02-02 19:05:52 +00:00
2458bd6507
Add comments describing how you would add prototypes for factory methods for
...
PowerPC-specific passes here.
llvm-svn: 11073
2004-02-02 19:05:28 +00:00
fea8eb8938
Add prototype for llvm::allocatePowerPCTargetMachine().
...
llvm-svn: 11072
2004-02-02 19:05:08 +00:00
298fdd7eb1
Codegen -0.0 correctly. Do not use fldz! This is another -0.0 == +0.0 problem, arg.
...
llvm-svn: 11070
2004-02-02 18:56:30 +00:00
6fe75845f9
Correct the method I just added to actually return false sometimes
...
llvm-svn: 11069
2004-02-02 18:53:04 +00:00
29734bccc5
Add a new method to ConstantFP
...
llvm-svn: 11068
2004-02-02 18:40:29 +00:00
7bae4ce64f
Bug fixed
...
llvm-svn: 11067
2004-02-02 17:48:56 +00:00
2cecace9b2
Create an object for tracking physical register usage. This will look
...
much better when I get rid of the reserved registers.
llvm-svn: 11066
2004-02-02 07:30:36 +00:00
e2d6189513
Bug fixed
...
llvm-svn: 11065
2004-02-01 22:52:12 +00:00
ac80ea4970
Second half of a fix for PR218 & test/Regression/Assembler/2004-02-01-NegativeZero.llx.
...
Basically we store floating point values as their integral components, instead of relying
on the semantics of floating point < to differentiate between values. This is likely to
make the map search be faster anyway.
llvm-svn: 11064
2004-02-01 22:49:04 +00:00
5e2feb4a0f
The first half of a fix for PR218 & test/Regression/Assembler/2004-02-01-NegativeZero.llx
...
llvm-svn: 11063
2004-02-01 22:48:09 +00:00
e55e30917a
Add another test
...
llvm-svn: 11062
2004-02-01 22:47:16 +00:00
9b9c0485fd
Change weight array into a vector and make it as big as the number of
...
registers (not as the max number of registers).
Change toSpill from a std::set into a std::vector<bool>.
Use the reverse iterator adapter to do a reverse scan of allocatable
registers.
llvm-svn: 11061
2004-02-01 20:13:26 +00:00
5e05a59bf8
Use std::map::count() instead of std::map::find() != std::map::end()
...
where appropriate.
llvm-svn: 11060
2004-02-01 18:39:53 +00:00
e1146140fc
New testcase for PR218
...
llvm-svn: 11059
2004-02-01 18:16:06 +00:00
0df32f636e
Add MRegisterInfo::getNumRegs().
...
llvm-svn: 11058
2004-02-01 17:14:20 +00:00
a3a56fa041
Change string for joined intervals.
...
llvm-svn: 11057
2004-02-01 16:13:05 +00:00
aeb8a80c71
FpMOV is also a move instruction.
...
llvm-svn: 11055
2004-02-01 08:22:16 +00:00
84a84c38b7
Change xor to ^.
...
llvm-svn: 11054
2004-02-01 08:19:25 +00:00
eddfe437c7
Now that tailduplication does not bork SSA form, run mem2reg earlier in gccas.
...
This tremendously improves the code generated by the LLVM optimizer, primarily
by making the inliner more aggressive. For example, it improves the stepanov
benchmark from 55.56 mega-additions/sec to 98.04 Ma/s. It also improves the
oopack/iterator benchmark from 338.3MFLOPS/s to 1103.4MFLOPS/s. Less noteworthy,
it improves oopack/matrix from 573 -> 641 MFLOPS/s.
llvm-svn: 11053
2004-02-01 07:24:53 +00:00
ed9b12c31a
Disable tail duplication in any "hard" cases, where it might break SSA form.
...
llvm-svn: 11052
2004-02-01 06:32:28 +00:00
cae054f59f
Fix a bug in a recent checkin
...
llvm-svn: 11050
2004-02-01 05:25:07 +00:00
7c91a6176c
Fix the count of the number of instructions removed
...
llvm-svn: 11049
2004-02-01 05:15:07 +00:00
8d4f06ed84
Missed one silly assert :-)
...
llvm-svn: 11048
2004-02-01 02:21:31 +00:00
69deb4d236
Simplify joinIntervals() code.
...
llvm-svn: 11047
2004-02-01 02:18:31 +00:00
8c285b20b7
Remove all of the annoying statistics now that I'm finished (for the near
...
term) working on bytecode size stuff.
llvm-svn: 11046
2004-02-01 01:50:31 +00:00
91dcc03c39
Use MRegisterInfo::isPhysicalRegister and
...
MRegisterInfo::isVirtualRegister.
llvm-svn: 11045
2004-02-01 01:27:01 +00:00
125ed54fdc
Fix a crasher bug in my constant folding rewrite
...
llvm-svn: 11044
2004-02-01 01:23:19 +00:00
93f7c40879
Print an error message if we can't materialize the bytecode file
...
llvm-svn: 11043
2004-02-01 01:07:25 +00:00
87d16ffdb0
Add comments
...
llvm-svn: 11042
2004-02-01 00:32:48 +00:00
8225d83e7b
Print an error message if there is an error materialize the bc file.
...
llvm-svn: 11041
2004-02-01 00:32:35 +00:00
662bb3f495
Apply final part of Chris' patch.
...
llvm-svn: 11040
2004-01-31 23:48:47 +00:00
459a67c309
Be a little smarter on the way we handle physical register defs.
...
llvm-svn: 11038
2004-01-31 23:13:30 +00:00
c330b98c1c
Finegrainify namespacification, use new MRegisterInfo::isVirtualRegister
...
method
llvm-svn: 11037
2004-01-31 21:27:19 +00:00
a2ef296994
Fix, correctly this time, the computation of the return value
...
Fix a spello
Tighten up the assertion checking
No functionality changes.
llvm-svn: 11036
2004-01-31 21:21:43 +00:00
626f431165
* Fix incorrect computation of the runOnMachineFunction return value
...
* Turn a bunch of instance variables into automatic variables
llvm-svn: 11035
2004-01-31 21:14:04 +00:00
d835aa6711
Remove unneeded #includes
...
Move Passes.h (which defines the interface to this file) to the top.
Move statistics to the top of the file.
Add a comment
llvm-svn: 11034
2004-01-31 21:07:15 +00:00
2f58ffc571
Avoid referencing deleted DSgraphs when merging an SCC into a larger SCC. This
...
fixes the crash in 176.gcc.
llvm-svn: 11033
2004-01-31 21:02:18 +00:00
Alkis Evlogimenos