RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-28 14:32:51 +01:00
Code Issues Projects Releases Wiki Activity
40,858 Commits 4 Branches 5 Tags 704 MiB
cf2cb4da82
Commit Graph

5542 Commits

Author SHA1 Message Date
Owen Anderson
667d9c95c2 Remove the -disable-correct-folding option, which was ugly and is no longer needed. 
llvm-svn: 54361
 2008-08-05 18:27:54 +00:00
Dan Gohman
de163bb39d Fix several const-correctness issues, resolving some -Wcast-qual warnings. 
llvm-svn: 54349
 2008-08-05 14:45:15 +00:00
Evan Cheng
1b731b8504 Remove #if 0. 
llvm-svn: 54347
 2008-08-05 07:20:57 +00:00
Evan Cheng
754148a2ec Fix PR2568: Fix bug that cause redudant kill marker after its live interval has been extended due to coalescing. 
llvm-svn: 54346
 2008-08-05 07:10:38 +00:00
Owen Anderson
7b5fe60440 Remove unneeded iteration. Thanks to Dan for the feedback. 
llvm-svn: 54337
 2008-08-05 00:30:10 +00:00
Owen Anderson
32ae9380f1 This option doesn't need to be a target option. It can be in SDISel instead. 
llvm-svn: 54336
 2008-08-05 00:27:28 +00:00
Owen Anderson
84fbc312d4 - Fix SelectionDAG to generate correct CFGs. 
- Add a basic machine-level dead block eliminator.

These two have to go together, since many other parts of the code generator are unable to handle the unreachable blocks otherwise created.

llvm-svn: 54333
 2008-08-04 23:54:43 +00:00
Dan Gohman
60ea311ec8 Fix SDISel lowering of PHI nodes to use ComputeValueVTs. 
This allows it to work correctly on aggregate values.
This fixes PR2623.

llvm-svn: 54331
 2008-08-04 23:42:46 +00:00
Dan Gohman
af429b3e52 Fix SDISel lowering of zeroinitializer and undef to use ComputeValueVTs. 
This allows it to work correctly on nested aggregate values.
This fixes PR2625.

llvm-svn: 54330
 2008-08-04 23:30:41 +00:00
Dale Johannesen
f669e7c14f Add a flag to disable jump table generation (all 
switches use the binary search algorithm) for
environments that don't support it.  PPC64 JIT
is such an environment; turn the flag on for that.

llvm-svn: 54248
 2008-07-31 18:13:12 +00:00
Dan Gohman
f691fc703d Improve dagcombining for sext-loads and sext-in-reg nodes. 
llvm-svn: 54239
 2008-07-31 00:50:31 +00:00
Dan Gohman
926a91d664 Move SelectionDAG::viewGraph() out of line; as an inline function 
it isn't always visible to gdb.

llvm-svn: 54228
 2008-07-30 18:48:53 +00:00
Dan Gohman
4ca56a8993 Don't look for leaf values to store when lowering stores of 
empty structs. This fixes PR2612.

llvm-svn: 54226
 2008-07-30 18:36:51 +00:00
Owen Anderson
4e5db8d80f Use existing LiveInterval methods to simplify live interval merging. Thanks to Evan for pointing these out. 
llvm-svn: 54225
 2008-07-30 18:27:35 +00:00
Owen Anderson
e6dca3efcb Value numbers whose def index is a special sentinel value should not be remapped. 
llvm-svn: 54218
 2008-07-30 17:42:47 +00:00
Owen Anderson
377ad6cb31 More fixes for corner cases when remapping live range indices. 
llvm-svn: 54186
 2008-07-30 00:22:56 +00:00
Owen Anderson
6147af0f8c When merging live intervals, we also need to merge in any live ranges that are inputs to two-address instructions 
that themselves define a range we already care about.

llvm-svn: 54185
 2008-07-30 00:21:16 +00:00
Owen Anderson
ab575176fc When merging a PHI operand's live interval into the PHI's live interval, we need to merge over all liveranges in 
the operand's interval that share the relevant value number, not just the range that immediately precedes the PHI.

llvm-svn: 54174
 2008-07-29 21:17:08 +00:00
Owen Anderson
bebb8e5d20 Don't decrement the BB remap when we don't need to. 
llvm-svn: 54173
 2008-07-29 21:15:44 +00:00
Duncan Sands
c3d73fbfc0 Fix PR2609. If a label is deleted, then it needs 
to be marked invalid regardless of whether it is
a debug, an exception handling or (hopefully) a
GC label.

llvm-svn: 54172
 2008-07-29 20:56:02 +00:00
Nate Begeman
f76a814673 Fix broken CellSPU lowering, re-instate braces in Legalize 
llvm-svn: 54168
 2008-07-29 19:07:27 +00:00
Nate Begeman
20b060cd54 Disable a fix in the previous patch, since it breaks CellSPU. 
The CellSPU codegen is broken, but needs to be fixed before we can
put this back in.

llvm-svn: 54164
 2008-07-29 18:28:31 +00:00
Nate Begeman
9a71580e21 Add vector shifts to the IR, patch by Eli Friedman. 
CodeGen & Clang work coming next.

llvm-svn: 54161
 2008-07-29 15:49:41 +00:00
Dan Gohman
9653b21dc2 Fold the useful features of alist and alist_node into ilist, and 
a new ilist_node class, and remove them. Unlike alist_node,
ilist_node doesn't attempt to manage storage itself, so it avoids
the associated problems, including being opaque in gdb.

Adjust the Recycler class so that it doesn't depend on alist_node.
Also, change it to use explicit Size and Align parameters, allowing
it to work when the largest-sized node doesn't have the greatest
alignment requirement.

Change MachineInstr's MachineMemOperand list from a pool-backed
alist to a std::list for now.

llvm-svn: 54146
 2008-07-28 21:51:04 +00:00
Dan Gohman
f6391108c6 Fix a typo in a comment. 
llvm-svn: 54136
 2008-07-28 18:43:51 +00:00
Dan Gohman
6bfdeb2ed1 Make the ScheduleDAG's GraphRoot edge be blue and dashed too, like 
the SelectionDAG's.

llvm-svn: 54129
 2008-07-27 22:46:49 +00:00
Dan Gohman
9742f7772d Rename SDOperand to SDValue. 
llvm-svn: 54128
 2008-07-27 21:46:04 +00:00
Dan Gohman
47c5cdbc34 Tidy SDNode::use_iterator, and complete the transition to have it 
parallel its analogue, Value::value_use_iterator. The operator* method
now returns the user, rather than the use.

llvm-svn: 54127
 2008-07-27 20:43:25 +00:00
Dan Gohman
2a0bc8f6e1 Rename isOnlyUseOf to isOnlyUserOf. 
llvm-svn: 54124
 2008-07-27 18:06:42 +00:00
Duncan Sands
545c21768f Some binary operations were being treated as 
unary operations!  Add support for softening
some additional unary operations like fp_to_sint.

llvm-svn: 54122
 2008-07-27 12:28:43 +00:00
Owen Anderson
25a3bf3f0e Fix the issues originally addressed in r54070. After thinking about it some more, I realized that the right thing to do 
is to have StrongPHIElimination use its knowledge of the PHIs before they're erased to update the intervals appropriate.  This is
both simpler and more accurate than the alternative, which was having LIA figure it out when it renumbered things, plus it's just
the right thing to do!

llvm-svn: 54077
 2008-07-25 23:38:08 +00:00
Owen Anderson
8f9ab550a2 Revert my previous patch. In retrospect, this is completely the wrong way to fix this problem. 
llvm-svn: 54072
 2008-07-25 23:06:59 +00:00
Owen Anderson
9408721750 Special cases are needed in renumbering when dealing with renumbering after a PHI has been removed. The interval previously defined 
by the PHI needs to be extended to the beginning of its basic block, and the intervals that were inputs need to be trimmed to the end 
of their basic blocks.

llvm-svn: 54070
 2008-07-25 22:32:01 +00:00
Owen Anderson
9e3ea5a60c In order to avoid reprocessing a register more than once, we need to add it 
to the handled set so it will get filtered out in future iterations.

llvm-svn: 54065
 2008-07-25 21:35:43 +00:00
Owen Anderson
987bd1dc5b Remove live interval entries for an interval if we're eliminating its only VN. 
llvm-svn: 54062
 2008-07-25 21:08:41 +00:00
Owen Anderson
fe97082ddd Properly remap live ranges whose end indices are the end of the function. 
llvm-svn: 54061
 2008-07-25 21:07:13 +00:00
Owen Anderson
cefa4df4c1 Make the remapping of interval indices (particularly ending indices) more robust. 
This is tricky business, and will probably take a few more iterations to get
the last kinks out of it.

llvm-svn: 54043
 2008-07-25 19:50:48 +00:00
Dan Gohman
25c825ede1 Disable the new aggressive remat logic introduced in 54000; it causes some 
regressions, such as PR2595. Also, there is a significant code-quality
issue in SPEC 464.h264ref and a few others.

llvm-svn: 54014
 2008-07-25 15:08:37 +00:00
Mon P Wang
0fba42aefd When splitting a vector shuffle, fixed which type we used for the hi part 
llvm-svn: 54007
 2008-07-25 01:30:26 +00:00
Dan Gohman
68487f0859 Use AliasAnalysis::pointsToConstantMemory in SDISel to avoid unnecessary 
dependencies with constant load nodes. This allows them to be scheduled
freely.

llvm-svn: 54001
 2008-07-25 00:04:14 +00:00
Dan Gohman
680e1bd958 Enable rematerialization of constants using AliasAnalysis::pointsToConstantMemory, 
and knowledge of PseudoSourceValues. This unfortunately isn't sufficient to allow
constants to be rematerialized in PIC mode -- the extra indirection is a
complication.

llvm-svn: 54000
 2008-07-25 00:02:30 +00:00
Owen Anderson
cb7fb7aa4a Store the predecessor MBB in the PHIUnion, rather than an index, since the indices will change after renumbering. 
llvm-svn: 53985
 2008-07-24 17:12:16 +00:00
Owen Anderson
08961cbffe Enable the insertion of empty indices into LiveInterals, thereby making renumbering possible. 
llvm-svn: 53961
 2008-07-23 21:37:49 +00:00
Owen Anderson
58d9213c3a Fix a compile-time regression introduced by my heuristic-changing patch. I forgot 
to multiply the instruction count by a constant factor in a few places, which
caused the register allocator to require many more iterations.

llvm-svn: 53959
 2008-07-23 19:47:27 +00:00
Dan Gohman
6564581be0 Enable first-class aggregates support. 
Remove the GetResultInst instruction. It is still accepted in LLVM assembly
and bitcode, where it is now auto-upgraded to ExtractValueInst. Also, remove
support for return instructions with multiple values. These are auto-upgraded
to use InsertValueInst instructions.

The IRBuilder still accepts multiple-value returns, and auto-upgrades them
to InsertValueInst instructions.

llvm-svn: 53941
 2008-07-23 00:34:11 +00:00
Duncan Sands
550e0de239 LegalizeTypes support for VSETCC. Fixes PR2575. 
llvm-svn: 53938
 2008-07-22 23:54:03 +00:00
Owen Anderson
7b8947cc31 Change the heuristics used in the coalescer, register allocator, and within 
live intervals itself to use an instruction count approximation that is 
not affected by inserting empty indices.

llvm-svn: 53937
 2008-07-22 22:46:49 +00:00
Evan Cheng
1aa928a8e6 Fix pr2566: incorrect assumption about bit_convert. It doesn't not have to output a vector value. Patch by Nicolas Capens! 
llvm-svn: 53932
 2008-07-22 20:42:56 +00:00
Dan Gohman
727b49bc26 Make the GraphRoot edge look like a chain edge, which is more accurate, 
and use the right result number, in the off chance that the graph root
has multiple result values.

llvm-svn: 53923
 2008-07-22 17:52:59 +00:00
Bill Wendling
87b1a0950c Another buildbot test commit. 
llvm-svn: 53896
 2008-07-22 00:53:37 +00:00
Bill Wendling
e0e8ecd5e9 Trivial check-in to test buildbot. No functionality change. 
llvm-svn: 53889
 2008-07-22 00:28:47 +00:00
Dan Gohman
e0ed9af577 Fix grammaros in comments. 
llvm-svn: 53884
 2008-07-21 22:38:59 +00:00
Dan Gohman
b024b0eedd Enhance the GraphWriter support for edge destinations, and teach the 
SelectionDAG graph writer to make use of them. Now, nodes with multiple
values are displayed as such, with incoming edges pointing to the
specific value they use.

llvm-svn: 53875
 2008-07-21 21:06:55 +00:00
Dan Gohman
dd84251894 After early-lowering the FORMAL_ARGUMENTS node, delete it. 
llvm-svn: 53874
 2008-07-21 21:04:07 +00:00
Dan Gohman
b91bef08a7 Add titles to the various SelectionDAG viewGraph calls 
that include useful information like the name of the
block being viewed and the current phase of compilation.

llvm-svn: 53872
 2008-07-21 20:00:07 +00:00
Dan Gohman
8a07776d39 Fix uses of underscore-capital names. 
llvm-svn: 53870
 2008-07-21 19:48:15 +00:00
Dan Gohman
8f718c75f0 Now that the MachineInstr leaks are fixed, enable leak checking 
in the MachineInstr clone code.

llvm-svn: 53868
 2008-07-21 18:47:29 +00:00
Duncan Sands
6e31474e71 Add VerifyNode, a place to put sanity checks on 
generic SDNode's (nodes with their own constructors
should do sanity checking in the constructor).  Add
sanity checks for BUILD_VECTOR and fix all the places
that were producing bogus BUILD_VECTORs, as found by
"make check".  My favorite is the BUILD_VECTOR with
only two operands that was being used to build a
vector with four elements!

llvm-svn: 53850
 2008-07-21 10:20:31 +00:00
Bill Wendling
73f5ecdbea Pull r53795 from Gaz into mainline: 
If .loc and .file aren't used, always emit the "debug_line" section. This
requires at least one entry in the line matrix. So if there's nothing to emit
into the matrix, emit an end of matrix value anyway.

llvm-svn: 53803
 2008-07-20 00:11:19 +00:00
Evan Cheng
5ed6b9f0ae Fix a memory leak in LiveIntervalAnalysis. 
llvm-svn: 53779
 2008-07-19 00:37:25 +00:00
Duncan Sands
ef45c602b6 Softfloat support for FDIV. Patch by 
Richard Pennington.

llvm-svn: 53773
 2008-07-18 21:18:48 +00:00
Duncan Sands
39da9f4a43 Eliminate unused variable. 
llvm-svn: 53772
 2008-07-18 21:07:41 +00:00
Duncan Sands
3d68e2ff9c Revert 53729, after waking up in the middle of 
the night realising that it was wrong :)  I
think the reason the same type was being used
for the shufflevec of indices as for the actual
indices is so that if one of them needs splitting
then so does the other.  After my patch it might
be that the indices need splitting but not the
rest, yet there is no good way of handling that.
I think the right solution is to not have the
shufflevec be an operand at all: just have it
be the list of numbers it actually is, stored
as extra info in the node.

llvm-svn: 53768
 2008-07-18 20:12:05 +00:00
Dan Gohman
e119f3db46 Fix a LocalSpiller leak. This fixes tramp3d-v4. 
llvm-svn: 53766
 2008-07-18 18:28:56 +00:00
Dan Gohman
f38f99ccc5 Re-introduce LeakDetector support for MachineInstrs and MachineBasicBlocks. 
Fix a leak that this turned up in LowerSubregs.cpp.
And, comment a leak in LiveIntervalAnalysis.cpp.

llvm-svn: 53746
 2008-07-17 23:49:46 +00:00
Dan Gohman
c666c01e65 When printing MemOperand nodes, only use print() for 
PseudoSourceValue values, which never have names. Use getName()
for all other values, because we want to print just a short summary
of the value, not the entire instruction.

llvm-svn: 53738
 2008-07-17 21:12:16 +00:00
Evan Cheng
d26080487b Subreg live interval valno may not have a corresponding def machineinstr since it's less precise. 
llvm-svn: 53734
 2008-07-17 19:48:53 +00:00
Duncan Sands
08ea7c0351 Use a legal type for elements of the vector_shuffle 
mask.  These are just indices into the shuffled vector
so their type is unrelated to the type of the
shuffled elements (which is what was being used before).
This fixes vec_shuffle-11.ll when using LegalizeTypes.
What seems to have happened is that Dan's recent change
r53687, which corrected the result type of the shuffle,
somehow caused LegalizeTypes to notice that the mask
operand was a BUILD_VECTOR with a legal type but elements
of an illegal type (i64).  LegalizeTypes legalized this
by introducing a new BUILD_VECTOR of i32 and bitcasting
it to the old type.  But the mask operand is not supposed
to be a bitcast but a straight BUILD_VECTOR of constants,
causing a crash.

llvm-svn: 53729
 2008-07-17 19:28:41 +00:00
Dan Gohman
8981962672 Add a new function, ReplaceAllUsesOfValuesWith, which handles bulk 
replacement of multiple values. This is slightly more efficient
than doing multiple ReplaceAllUsesOfValueWith calls, and theoretically
could be optimized even further. However, an important property of this
new function is that it handles the case where the source value set and
destination value set overlap. This makes it feasible for isel to use
SelectNodeTo in many very common cases, which is advantageous because
SelectNodeTo avoids a temporary node and it doesn't require CSEMap
updates for users of values that don't change position.

Revamp MorphNodeTo, which is what does all the work of SelectNodeTo, to
handle operand lists more efficiently, and to correctly handle a number
of corner cases to which its new wider use exposes it.

This commit also includes a change to the encoding of post-isel opcodes
in SDNodes; now instead of being sandwiched between the target-independent
pre-isel opcodes and the target-dependent pre-isel opcodes, post-isel
opcodes are now represented as negative values. This makes it possible
to test if an opcode is pre-isel or post-isel without having to know
the size of the current target's post-isel instruction set.

These changes speed up llc overall by 3% and reduce memory usage by 10%
on the InstructionCombining.cpp testcase with -fast and -regalloc=local.

llvm-svn: 53728
 2008-07-17 19:10:17 +00:00
Duncan Sands
c3331602f9 LegalizeTypes support for what seems to be the 
only missing ppc long double operations: FNEG
and FP_EXTEND.

llvm-svn: 53723
 2008-07-17 17:35:14 +00:00
Duncan Sands
778e45e748 Turn LegalizeTypes back off again for the moment: 
it is breaking Darwin bootstrap due to missing
functionality.

llvm-svn: 53721
 2008-07-17 17:06:03 +00:00
Duncan Sands
41aeb22c1d Factorize some code for determining which libcall to use. 
llvm-svn: 53713
 2008-07-17 02:36:29 +00:00
Dan Gohman
0025513482 Fix the result type of a VECTOR_SHUFFLE+BIT_CONVERT dagcombine. This 
was turned up by some new SelectionDAG assertion checks that I'm
working on.

llvm-svn: 53687
 2008-07-16 16:13:58 +00:00
Duncan Sands
3448d4087f Add support for promoting and expanding AssertZext 
and AssertSext.  Needed when passing huge integer
parameters with the zeroext or signext attributes.

llvm-svn: 53684
 2008-07-16 16:03:07 +00:00
Dan Gohman
2638ec5b4c Fix a comment to say nonnegative instead of positive. 
llvm-svn: 53681
 2008-07-16 15:57:10 +00:00
Dan Gohman
bc29205ccc Add an assert to check for empty flags for MachineMemOperand. 
llvm-svn: 53680
 2008-07-16 15:56:42 +00:00
Duncan Sands
b753d27377 Reorder methods alphabetically. No functionality change. 
While this is not a wonderful organizing principle, it
does make it easy to find routines, and clear where to
insert new ones.

llvm-svn: 53672
 2008-07-16 11:41:33 +00:00
Duncan Sands
b2e1ddbd0b Turn on LegalizeTypes by default. 
llvm-svn: 53671
 2008-07-16 11:36:51 +00:00
Dan Gohman
c4400b8618 SelectionDAG::AssignNodeIds is unused. 
llvm-svn: 53636
 2008-07-15 18:29:32 +00:00
Dan Gohman
c9b72806d7 Don't sort SDNodes by their addresses in SelectionDAG::dump. Instead, 
just use the AllNodes order, which is at least relatively stable
across runs.

llvm-svn: 53632
 2008-07-15 18:18:54 +00:00
Duncan Sands
7ca2df2319 LegalizeTypes support for fabs on ppc long double. 
llvm-svn: 53613
 2008-07-15 15:02:44 +00:00
Duncan Sands
58eb5e35da LegalizeTypes support for promotion of bswap. 
In LegalizeDAG the value is zero-extended to
the new type before byte swapping.  It doesn't
matter how the extension is done since the new
bits are shifted off anyway after the swap, so
extend by any old rubbish bits.  This results
in the final assembler for the testcase being
one line shorter.

llvm-svn: 53604
 2008-07-15 10:18:22 +00:00
Duncan Sands
710be60c23 LegalizeTypes support for promotion of SIGN_EXTEND_INREG. 
llvm-svn: 53603
 2008-07-15 10:14:24 +00:00
Duncan Sands
d2097e5b92 Reorder the integer promotion methods alphabetically. 
No change in functionality.

llvm-svn: 53602
 2008-07-15 10:12:34 +00:00
Mon P Wang
834c054b1d Fixed potential bug if the source and target of a bit convert have different alignment 
llvm-svn: 53590
 2008-07-15 05:28:34 +00:00
Dan Gohman
e76fd33ee5 Reapply 53476 and 53480, with a fix so that it properly updates 
the BB member to the current basic block after emitting
instructions.

llvm-svn: 53567
 2008-07-14 18:19:29 +00:00
Dan Gohman
7b61b66334 Improve debug output for MemOperandSDNode. PseudoSourceValue nodes 
don't have value names, so use print instead of getName() to get a
useful string.

llvm-svn: 53563
 2008-07-14 17:51:24 +00:00
Dan Gohman
bf764c6a4c Fix edito in the PseudoSourceValue name list. 
llvm-svn: 53562
 2008-07-14 17:45:47 +00:00
Duncan Sands
ef690bbffe I don't think BUILD_PAIR can have a vector result. 
Remove support for this.

llvm-svn: 53559
 2008-07-14 17:34:19 +00:00
Duncan Sands
bb3251e1f4 Tighten up some checks. Fix FPOWI splitting for 
non-power-of-two vectors.

llvm-svn: 53558
 2008-07-14 17:33:37 +00:00
Duncan Sands
134155a263 An INSERT_VECTOR_ELT can insert a larger value 
than the vector element type.  Don't forget to
handle this when the insertion index is not a
constant.

llvm-svn: 53556
 2008-07-14 17:32:02 +00:00
Duncan Sands
5a4bab9849 According to the docs, it is possible to have an 
extending load of a vector.  Handle this case when
splitting vector loads.  I'm not completely sure
what is supposed to happen, but I think it means
hi should be set to undef.  LegalizeDAG does not
consider this case.

llvm-svn: 53555
 2008-07-14 17:27:46 +00:00
Duncan Sands
fd63c72205 There should be no extending loads or truncating 
stores of one-element vectors.  Also, neaten the
handling of INSERT_VECTOR_ELT when the inserted
type is larger than the vector element type.

llvm-svn: 53554
 2008-07-14 17:22:31 +00:00
Duncan Sands
a9d2c2042e Ignore TargetConstant with an illegal type. These 
are used for passing huge immediates in inline ASM
from the front-end straight down to the ASM writer.
Of course this is a hack, but it is simple, limited
in scope, works in practice, and is what LegalizeDAG
does.

llvm-svn: 53553
 2008-07-14 17:15:45 +00:00
Evan Cheng
7457283357 Typos. 
llvm-svn: 53504
 2008-07-12 02:22:07 +00:00
Evan Cheng
05e5317cab Fix PR2536: a nasty spiller bug. If a two-address instruction uses a register but the use portion of its live range is not part of its liveinterval, it must be defined by an implicit_def. In that case, do not spill the use. e.g. 
8   %reg1024<def> = IMPLICIT_DEF
12  %reg1024<def> = INSERT_SUBREG %reg1024<kill>, %reg1025, 2

The live range [12, 14) are not part of the r1024 live interval since it's defined by an implicit def. It will not conflicts with live interval of r1025. Now suppose both registers are spilled, you can easily see a situation where both registers are reloaded before the INSERT_SUBREG and both target registers that would overlap.

llvm-svn: 53503
 2008-07-12 01:56:02 +00:00
Evan Cheng
ebf7ff4531 Back out 53476 and 53480 for now. Somehow they cause llc to miscompile 179.art. 
llvm-svn: 53502
 2008-07-12 01:38:51 +00:00
Dan Gohman
4c18394001 Include a frame index in the "fixed stack" pseudo source value 
instead of using the frame index for the SVOffset, which was
inconsistent.

llvm-svn: 53486
 2008-07-11 22:44:52 +00:00
Dan Gohman
b21183f645 Fix an obsolete top-level comment. 
llvm-svn: 53481
 2008-07-11 22:39:58 +00:00
Dan Gohman
3b91ef27b5 Factor out debugging code into the common base class. 
llvm-svn: 53480
 2008-07-11 22:36:22 +00:00
Dan Gohman
6ee2195ab1 Add support for putting NamedRegionTimers in TimerGroups, and 
use a timer group for the timers in SelectionDAGISel. Also,
Split scheduling out from emitting, to give each their own
timer.

llvm-svn: 53476
 2008-07-11 21:54:34 +00:00
Dan Gohman
d7a32ecde2 Trim unnecessary #includes. 
llvm-svn: 53471
 2008-07-11 20:38:31 +00:00
Duncan Sands
eb04d356e3 Remove an apparently useless routine: there should 
be no need to split the result of a vector RET node,
since they are always already legal.

llvm-svn: 53462
 2008-07-11 17:02:09 +00:00
Duncan Sands
1e3fe05638 It is pointless to turn a UINT_TO_FP into an 
SINT_TO_FP libcall plus additional operations:
it might as well be a direct UINT_TO_FP libcall.
So only turn it into an SINT_TO_FP if the target
has special handling for SINT_TO_FP.

llvm-svn: 53461
 2008-07-11 17:00:14 +00:00
Duncan Sands
58130a26f4 Add two missing SINT_TO_FP libcalls. 
llvm-svn: 53460
 2008-07-11 16:57:02 +00:00
Duncan Sands
52f1dbf139 Port a shift-by-1 optimization from LegalizeDAG: it 
was presumably added after the rest of the code was
copied to LegalizeTypes.

llvm-svn: 53459
 2008-07-11 16:54:57 +00:00
Duncan Sands
669d4d6909 Add support for 128 bit shifts and 32 bit shifts 
on 16 bit machines.

llvm-svn: 53458
 2008-07-11 16:52:29 +00:00
Chris Lattner
78d60a220b Fix a bug in the soft-float handling of FCOPYSIGN that Duncan noticed 
when working on legalizetypes.  Both legalizetypes and legalizeops now
produce hte same code for CodeGen/ARM/fcopysign.ll.

llvm-svn: 53435
 2008-07-10 23:46:13 +00:00
Chris Lattner
5e784a8ebd make legalize types be a command line option: -enable-legalize-types. 
llvm-svn: 53434
 2008-07-10 23:37:50 +00:00
Dan Gohman
1f42b96c7e Make stack slot coloring's debug output more consistent with 
other passes.

llvm-svn: 53415
 2008-07-10 19:49:32 +00:00
Evan Cheng
d7023b66ea Change StackSlotForVirtReg (which maps vregs to frame indices) from std::map to IndexedMap. 
llvm-svn: 53414
 2008-07-10 18:23:23 +00:00
Duncan Sands
a6c09579f0 Add support for 128 bit multiplicative operations. 
Lack of these caused a bootstrap failure with Fortran
on x86-64 with LegalizeTypes turned on.  While there,
be nice to 16 bit machines and support expansion of
i32 too.

llvm-svn: 53408
 2008-07-10 15:35:05 +00:00
Duncan Sands
90a916539d Add a mysteriously missing libcall, FPTOSINT_F80_I32. 
Be nice to 16 bit machines by supporting FP_TO_XINT
expansion for these.

llvm-svn: 53407
 2008-07-10 15:33:02 +00:00
Duncan Sands
91912e046a Fix a FIXME: use an apint in CTTZ legalization. 
llvm-svn: 53406
 2008-07-10 15:30:54 +00:00
Duncan Sands
f2734eabf2 Remove PromoteIntRes_FP_ROUND - not sure what it 
was doing there: FP_ROUND returns a float, not an
integer.

llvm-svn: 53405
 2008-07-10 15:29:55 +00:00
Duncan Sands
16f5ab06b0 Make sure the alignment of the temporary created 
in CreateStackStoreLoad is good enough for both
the source and destination types.

llvm-svn: 53404
 2008-07-10 15:26:17 +00:00
Duncan Sands
a219bb03e1 Make the LegalizeType method naming scheme more regular. 
llvm-svn: 53403
 2008-07-10 15:25:04 +00:00
Duncan Sands
8afe39dc56 Don't barf when dumping a constant that contains 
a ginormous value (eg: i128 -1).

llvm-svn: 53402
 2008-07-10 11:23:14 +00:00
Evan Cheng
3ec3113431 - Change the horrible N^2 isRegReDefinedByTwoAddr. Now callers must supply the operand index of def machineoperand and at most one full scan of non-implicit operands is needed. 
- Change local register allocator to use the new isRegReDefinedByTwoAddr instead of reinventing the wheel.

llvm-svn: 53394
 2008-07-10 07:35:43 +00:00
Owen Anderson
65e43ae1e8 Use DenseMap instead of std::map in local register allocation. This improves the time on instcombine from .31s to .22s 
llvm-svn: 53390
 2008-07-10 01:56:35 +00:00
Owen Anderson
189609bb36 Fix 403.gcc. Finally got the check for two-address-ness correct. 
llvm-svn: 53389
 2008-07-10 01:53:01 +00:00
Owen Anderson
35080ed650 Revert r53367, which was breaking things. 
llvm-svn: 53378
 2008-07-09 23:09:10 +00:00
Dan Gohman
e10fc737ca Simplify hasNUsesOfValue and hasAnyUsesOfValue even more. This 
makes their special-case checks of use_size() less beneficial,
so remove them. This eliminates all but one use of use_size(),
which is in AssignTopologicalOrder, which uses it only once for
each node, and so can reasonably afford to recompute it, as
this allows the UsesSize field of SDNode to be removed
altogether.

llvm-svn: 53377
 2008-07-09 23:03:14 +00:00
Dan Gohman
6ba50d7fc1 hasAnyUseOfValue can check SDUse nodes of its users directly instead 
of examining every operand of every user.

llvm-svn: 53374
 2008-07-09 22:39:01 +00:00
Dan Gohman
679872b05f Move MemoryVT out of LSBaseNode into MemSDNode, allowing the 
getMemOperand function to be moved into the base class as well
and made non-virtual.

llvm-svn: 53372
 2008-07-09 22:08:04 +00:00
Evan Cheng
082f7f802e Avoid creating expensive comment string if it's not going to be printed. 
llvm-svn: 53369
 2008-07-09 21:53:02 +00:00
Owen Anderson
8174cf28a3 Loosen our check here. Local regalloc only cares that the reg is used and def'd by the same instruction, but about the details of 
the relationship.

llvm-svn: 53367
 2008-07-09 21:34:36 +00:00
Dan Gohman
f6e651fcf2 Move the IsVolatile and SVOffset fields into the MemSDNode base 
class, and store IsVolatile and Alignment in a more compact form.
This makes AtomicSDNode slightly larger, but it shrinks LoadSDNode
and StoreSDNode, which are much more common and are the largest of
the SDNode subclasses. Also, this lets the isVolatile() and
getAlignment() accessors be non-virtual.

llvm-svn: 53361
 2008-07-09 21:23:02 +00:00
Owen Anderson
50a64de51e Don't use an expensive check for two-address-ness when we have the information sitting around to determine it much more quickly, 
This speeds up the local register allocator from 0.37s to 0.31s on instcombine.

llvm-svn: 53359
 2008-07-09 21:15:10 +00:00
Owen Anderson
58912683fd Factor local liveness computation out into its own function. 
llvm-svn: 53352
 2008-07-09 20:14:53 +00:00
Dan Gohman
0975e33ffc Reuse the MO variable instead of recomputing it in RegAllocLocal. 
Keep RegAllocSimple in sync.

llvm-svn: 53351
 2008-07-09 20:12:26 +00:00
Dan Gohman
c27e4e9651 Give RegAllocSimple a TargetInstrInfo member to keep it consistent 
with RegAllocLocal.

llvm-svn: 53347
 2008-07-09 19:56:01 +00:00
Dan Gohman
ad2c9877ce RegAllocLocal has a TargetInstrInfo data member. Use it instead 
of having local variables duplicate it.

llvm-svn: 53346
 2008-07-09 19:55:19 +00:00
Dan Gohman
4d221ae533 Use find with std::map, when that's what's needed, instead of lower_bound 
with extra checks.

llvm-svn: 53344
 2008-07-09 19:51:00 +00:00
Anton Korobeynikov
f16db15839 Switch to new section name handling facility 
llvm-svn: 53316
 2008-07-09 13:27:16 +00:00
Duncan Sands
e66e7b80f0 Remove some unneeded includes. 
llvm-svn: 53289
 2008-07-09 12:08:25 +00:00
Duncan Sands
828e080f3b Redo LegalizeTypes soft float support for 
SINT_TO_FP and UINT_TO_FP.  This now produces
the same code as LegalizeDAG (the previous
code was based on a mistaken idea of what
LegalizeDAG did in this case).

llvm-svn: 53288
 2008-07-09 12:07:22 +00:00
Duncan Sands
bc816b10f7 Forgot to update the chain result when softening 
loads.

llvm-svn: 53287
 2008-07-09 11:15:31 +00:00
Duncan Sands
cb4999041b LegalizeTypes soft float support for FP_TO_SINT and 
FP_TO_UINT.

llvm-svn: 53286
 2008-07-09 11:13:46 +00:00
Duncan Sands
c8d39ae369 LegalizeTypes support for powi soft float. 
llvm-svn: 53285
 2008-07-09 11:11:47 +00:00
Duncan Sands
d5c6d06980 Make the role of MVT::i32 clearer here, and add a 
note since it is not clear whether it is correct.

llvm-svn: 53284
 2008-07-09 08:07:41 +00:00
Evan Cheng
098905497f Missed alignment argument on stores lowered from memcpy. 
llvm-svn: 53281
 2008-07-09 06:38:06 +00:00
Bill Wendling
3119083ee6 Make the DICountVisitor not a visitor. This keeps us from calling virtual 
functions and junk.

llvm-svn: 53279
 2008-07-09 06:02:33 +00:00
Dan Gohman
2e23cd5a18 const-ify SelectionDAG::getNodeValueTypes. 
llvm-svn: 53264
 2008-07-09 00:00:42 +00:00
Dan Gohman
9f1547684c It's no longer necessary to test if a MachineBasicBlock's 
parent is non-null. It now always is.

llvm-svn: 53263
 2008-07-08 23:59:09 +00:00
Dan Gohman
286e8f7295 Verify that MachineMemOperand alignment is a non-zero power of 2. 
llvm-svn: 53262
 2008-07-08 23:47:04 +00:00
Dan Gohman
5b24b3baef Factor out the code for computing an alignment value, and make it 
available to getAtomic in addition to just getLoad and getStore,
to prevent MachineMemOperands with 0 alignment.

llvm-svn: 53261
 2008-07-08 23:46:32 +00:00
Owen Anderson
d90ae5127e Fix the build. Apparently MachineInstr& is no longer implicitly convertable to MachineBasicBlock::iterator. 
llvm-svn: 53260
 2008-07-08 23:36:37 +00:00
Owen Anderson
caf8cb40e8 Make the local register allocator compute (purely local) liveness information for itself 
rather than depending on LiveVariables.  This decreases compile time from:
0.5909s (LV + Regalloc) to 0.421s (just regalloc).

llvm-svn: 53256
 2008-07-08 22:24:50 +00:00
Dale Johannesen
31b41acdd4 Remove some dead code. 
llvm-svn: 53253
 2008-07-08 21:53:43 +00:00
Evan Cheng
a5621a1c54 Do not CSE DEBUG_LOC, DBG_LABEL, DBG_STOPPOINT, DECLARE, and EH_LABEL SDNode's. This improves compile time slightly at -O0 -g. 
llvm-svn: 53246
 2008-07-08 20:06:39 +00:00
Duncan Sands
9255845ac7 Remove custom expansion from LegalizeTypes when doing 
soft float: experiments show that targets aren't
expecting this for results or for operands.  Add
support select/select_cc result soft float and
correct operand soft float for these.

llvm-svn: 53245
 2008-07-08 20:03:24 +00:00
Duncan Sands
a41982af51 Add missing select_cc libcall line, somehow omitted 
in LegalizeTypes.

llvm-svn: 53244
 2008-07-08 20:00:05 +00:00
Evan Cheng
6af015292e Unbreak C++ tests on x86 Darwin. 
llvm-svn: 53237
 2008-07-08 16:40:43 +00:00
Duncan Sands
f9f4389fb7 LegalizeTypes support for FP_ROUND and FP_EXTEND 
soft float.

llvm-svn: 53231
 2008-07-08 10:50:55 +00:00
Evan Cheng
5be1103646 Avoid unnecessary string construction during asm printing. 
llvm-svn: 53215
 2008-07-08 00:55:58 +00:00
Dan Gohman
cd25487258 Pool-allocation for MachineInstrs, MachineBasicBlocks, and 
MachineMemOperands. The pools are owned by MachineFunctions.

This drastically reduces the number of calls to malloc/free made
during the "Emit" phase of scheduling, as well as later phases
in CodeGen. Combined with other changes, this speeds up the
"instruction selection" phase of CodeGen by 10% in some cases.

llvm-svn: 53212
 2008-07-07 23:14:23 +00:00
Dan Gohman
93e0208361 Pool-allocation for SDNodes. The pool is allocated once for each function, 
and reused across SelectionDAGs.

This drastically reduces the number of calls to malloc/free made during
instruction selection, and improves memory locality.

llvm-svn: 53211
 2008-07-07 23:02:41 +00:00
Bill Wendling
b30a16ca2c Use the canonical way to get an empty structure. 
llvm-svn: 53206
 2008-07-07 21:41:57 +00:00
Bill Wendling
d7da49634d Use StringMap for greater justice! 
llvm-svn: 53202
 2008-07-07 20:59:31 +00:00
Dan Gohman
d787dcfa3d Fix SDNode::MorphNodeTo (a function used by by SelectNodeTo) to 
properly track dead nodes that are on the original SDNode's operand
list but not the new one, and have no other uses.

llvm-svn: 53201
 2008-07-07 20:57:48 +00:00
Dan Gohman
255eb1d296 Move MachineMemOperand's constructor out of line, to avoid a 
#include dependency on Support/MathExtras.h in the header file.

llvm-svn: 53200
 2008-07-07 20:32:02 +00:00
Dan Gohman
ed34aec9a8 Use of operator* is redundant and confusing here. 
llvm-svn: 53197
 2008-07-07 20:08:05 +00:00
Dan Gohman
7fb6670108 Minor const-correctness fixes. 
llvm-svn: 53196
 2008-07-07 20:06:06 +00:00
Dan Gohman
713d153ea3 Assert that all MachineInstrs update PhysRegUseDefLists in 
their cleanup code. 

llvm-svn: 53194
 2008-07-07 19:55:35 +00:00
Dan Gohman
d263602463 Remove most of the uses of SDOperandPtr, usually replacing it with a 
simple const SDOperand*, which is what's usually needed.

For AddNodeIDOperands, which is small, just duplicate the function to
accept an SDUse*.

For SelectionDAG::getNode - Add an overload that accepts SDUse* that
copies the operands into a temporary SDOperand array, but also has
special-case checks for 0 through 3 operands to avoid the copy in
the common cases.

llvm-svn: 53183
 2008-07-07 18:26:29 +00:00
Dan Gohman
955fdc7a4c Add explicit keywords. 
llvm-svn: 53179
 2008-07-07 18:00:37 +00:00
Dan Gohman
c97817aac3 Make DenseMap's insert return a pair, to more closely resemble std::map. 
llvm-svn: 53177
 2008-07-07 17:46:23 +00:00
Evan Cheng
38ce408540 LegalizeSetCCOperands should legalize the result of ExpandLibCall. Patch by Richard Osborne. 
llvm-svn: 53169
 2008-07-07 07:18:09 +00:00
Bill Wendling
d6bd5ea0b7 Prevent option name conflict. 
llvm-svn: 53166
 2008-07-07 05:42:27 +00:00
Duncan Sands
0caf597378 LegalizeTypes soft-float support for stores of a 
float value.

llvm-svn: 53165
 2008-07-07 00:08:12 +00:00
Mon P Wang
a43d6758ac Fixed generating incorrect aligned stores that I backout of r53031 
that fixed problems in EmitStackConvert where the source and target type
have different alignment by creating a stack slot with the max
alignment of source and target type.

llvm-svn: 53150
 2008-07-05 20:40:31 +00:00
Duncan Sands
3ea6f15708 Rather than having a different custom legalization 
hook for each way in which a result type can be
legalized (promotion, expansion, softening etc),
just use one: ReplaceNodeResults, which returns
a node with exactly the same result types as the
node passed to it, but presumably with a bunch of
custom code behind the scenes.  No change if the
new LegalizeTypes infrastructure is not turned on.

llvm-svn: 53137
 2008-07-04 11:47:58 +00:00
Duncan Sands
aac5c915ed Linux also does not require exception handling 
moves in order to get correct debug info.  Since
I can't imagine how any target could possibly
be any different, I've just stripped out the
option: now all the world's like Darwin!

llvm-svn: 53134
 2008-07-04 09:55:48 +00:00
Bill Wendling
e3ad6f2249 Don't return std::vector by value, but pass it in by reference to be filled. 
llvm-svn: 53123
 2008-07-03 23:13:02 +00:00
Bill Wendling
2c9a84d788 Revert my previous check-in that split up MachineModuleInfo. It turns out to 
slow the compiler down at -O0 some 30% or more. Ooops.

llvm-svn: 53120
 2008-07-03 22:53:42 +00:00
Evan Cheng
01e53a3406 Backed out 53031. 
llvm-svn: 53110
 2008-07-03 18:20:14 +00:00
Evan Cheng
1f6148a84c - Remove calls to copyKillDeadInfo which is an N^2 function. Instead, propagate kill / dead markers as new instructions are constructed in foldMemoryOperand, convertToThressAddress, etc. 
- Also remove LiveVariables::instructionChanged, etc. Replace all calls with cheaper calls which update VarInfo kill list.

llvm-svn: 53097
 2008-07-03 09:09:37 +00:00
Dan Gohman
ca3abd7409 Reapply r52988, "Simplify addRegisterKilled and addRegisterDead." The 
254.gap failure was not due to this mod.

llvm-svn: 53068
 2008-07-03 01:18:51 +00:00
Dan Gohman
3a3ea83ef7 Avoid unnecessarily copying APInt objects. 
llvm-svn: 53065
 2008-07-03 00:52:03 +00:00
Evan Cheng
67ab36cb9d Use std::replace instead of std::find and push_back. 
llvm-svn: 53063
 2008-07-03 00:28:27 +00:00
Evan Cheng
7c3920e692 - Add LiveVariables::replaceKillInstruction. This does a subset of instructionChanged. That is, it only update the VarInfo.kills if the new instruction is known to have the correct dead and kill markers. 
- CommuteInstruction copies kill / dead markers over to new instruction. So use replaceKillInstruction instead.

llvm-svn: 53061
 2008-07-03 00:07:19 +00:00
Owen Anderson
604f9f722d Make LiveVariables even more optional, by making it optional in the call to TargetInstrInfo::convertToThreeAddressInstruction 
Also, if LV isn't around, then TwoAddr doesn't need to be updating flags, since they won't have been set in the first place.

llvm-svn: 53058
 2008-07-02 23:41:07 +00:00
Dan Gohman
741cda0ac8 Replace a few uses of SelectionDAG::getTargetNode with 
SelectionDAG::SelectNodeTo in the instruction selector. This
updates existing nodes in place instead of creating new ones.

Go back to selecting ISD::DBG_LABEL nodes into
TargetInstrInfo::DBG_LABEL nodes instead of leaving them
unselected, now that SelectNodeTo allows us to update them
in place.

llvm-svn: 53057
 2008-07-02 23:23:19 +00:00
Dan Gohman
d9458e2d39 Revert r52988. It broke 254.gap on x86-64. 
llvm-svn: 53050
 2008-07-02 22:12:55 +00:00
Owen Anderson
a420723feb TwoAddressInstructionPass doesn't really require LiveVariables, it just needs to update it if it's already around. 
llvm-svn: 53049
 2008-07-02 21:28:58 +00:00
Duncan Sands
21e2a711e3 Add a new getMergeValues method that does not need 
to be passed the list of value types, and use this
where appropriate.  Inappropriate places are where
the value type list is already known and may be
long, in which case the existing method is more
efficient.

llvm-svn: 53035
 2008-07-02 17:40:58 +00:00
Mon P Wang
55f4371807 Fixed problem in EmitStackConvert where the source and target type 
have different alignment by creating a stack slot with the max
alignment of source and target type.

llvm-svn: 53031
 2008-07-02 17:07:12 +00:00
Chris Lattner
8bcad39c9a instead of aborting on shifts of i1, just implicitly fold them. 
The dag combiner can produce a shift of i1 when folding icmp i1's.

llvm-svn: 53030
 2008-07-02 17:01:57 +00:00
Duncan Sands
b7d8348f0d Fix typo compounded by a cut-and-pasto. 
llvm-svn: 53012
 2008-07-02 10:03:53 +00:00
Duncan Sands
5d5f1e800d Let AnalyzeNewNode take care of calling ExpungeNode. 
This makes sure that all new nodes are expunged, not
just those the top node of a new subtree.

llvm-svn: 53011
 2008-07-02 09:56:41 +00:00
Evan Cheng
3921085102 - Use a faster priority comparison function if -fast. 
- Code clean up.

llvm-svn: 53010
 2008-07-02 09:23:51 +00:00
Chris Lattner
06545d1c4a Add a new (simple) StringMap::clear method, patch by Pratik 
Solanki!

llvm-svn: 53008
 2008-07-02 05:26:32 +00:00
Bill Wendling
83caf5a9d5 Use the canonical form for getting an empty structure. 
llvm-svn: 53003
 2008-07-02 00:50:02 +00:00
Bill Wendling
aa45e38a17 Sorry. I couldn't sleep at night knowing I put these ugly casts into the source tree. 
llvm-svn: 53001
 2008-07-02 00:35:47 +00:00
Bill Wendling
27c38cee90 Darwin doesn't need exception handling information for the "move" info when 
debug information is being output, because it's leet!

llvm-svn: 52994
 2008-07-01 23:34:48 +00:00
Evan Cheng
5389352c98 Avoid creating expensive comment string if it's not going to be printed. 
llvm-svn: 52992
 2008-07-01 23:18:29 +00:00
Owen Anderson
b258a71a37 No need to use std::distance. We can just count the number of operands 
much more cheaply.

llvm-svn: 52990
 2008-07-01 22:34:11 +00:00
Evan Cheng
508e82fcfc Simplify addRegisterKilled and addRegisterDead. 
llvm-svn: 52988
 2008-07-01 22:21:21 +00:00
Bill Wendling
c695823456 - Update comments. 
- Don't use GlobalVariable::LinkageTypes when unsigned works.

llvm-svn: 52987
 2008-07-01 22:08:01 +00:00