1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-02 00:42:52 +01:00
Commit Graph

49703 Commits

Author SHA1 Message Date
Bruno Cardoso Lopes
0fa8b71a55 Enable SSEDomainFix pass for AVX mode.
llvm-svn: 139816
2011-09-15 18:27:32 +00:00
Jakob Stoklund Olesen
74cb8c037a Count correctly when a COPY turns into a spill or reload.
The number of spills could go negative since a folded COPY is just a
spill, and it may be eliminated.

llvm-svn: 139815
2011-09-15 18:22:52 +00:00
Jim Grosbach
9d7aa9bcbc Thumb2 assembly parsing and encoding for REV.
llvm-svn: 139813
2011-09-15 18:13:30 +00:00
Benjamin Kramer
1adbd30351 DWARF: Print line tables per compile unit, so they get the right address size.
llvm-svn: 139808
2011-09-15 18:02:20 +00:00
Jim Grosbach
e341895736 Assmebler symbol attribute directives don't work on temporary symbols.
Assembler private local symbols aren't legal targets of symbol attributes,
so issue a diagnostic for them.

Based on patch by Stepan Dyatkovskiy.

llvm-svn: 139807
2011-09-15 17:56:49 +00:00
Jakob Stoklund Olesen
444ae3ddc7 Count inserted spills and reloads more accurately.
Adjust counters when removing spill and reload instructions.

We still don't account for reloads being removed by eliminateDeadDefs().

llvm-svn: 139806
2011-09-15 17:54:28 +00:00
Benjamin Kramer
900209189c DWARF: wire up .debug_str dumping.
llvm-svn: 139799
2011-09-15 16:57:13 +00:00
Jim Grosbach
00e28bee78 Handle missing newline at EOF more gracefully in MC AsmLexer.
If we see an EOF w/o a preceding end-of-line, return an EndOfStatement
token before returning the Eof token.

Based on patch by Stepan Dyatkovskiy.

llvm-svn: 139798
2011-09-15 16:52:06 +00:00
Jakob Stoklund Olesen
f608389173 Trace through sibling PHIs in bulk.
When traceSiblingValue() encounters a PHI-def value created by live
range splitting, don't look at all the predecessor blocks.  That can be
very expensive in a complicated CFG.

Instead, consider that all the non-PHI defs jointly dominate all the
PHI-defs.  Tracing directly to all the non-PHI defs is much faster that
zipping around in the CFG when there are many PHIs with many
predecessors.

This significantly improves compile time for indirectbr interpreters.

llvm-svn: 139797
2011-09-15 16:41:12 +00:00
Jim Grosbach
69ddec5ff7 ARM support the pre-UAL mnemonic 'qsubaddx' for 'qsax.'
llvm-svn: 139796
2011-09-15 16:16:50 +00:00
Jim Grosbach
d428c970e3 Thumb2 push/pop mnemonic recognition.
llvm-svn: 139794
2011-09-15 15:55:04 +00:00
Jakob Stoklund Olesen
1c42989edf Speed up LiveIntervals::shrinkToUse with some caching.
Blocks with multiple PHI successors only need to go on the worklist
once.  Use a SmallPtrSet to track the live-out blocks that have already
been handled.  This is a lot faster than the two live range check we
would otherwise do.

Also stop recomputing hasPHIKill flags.  Like RenumberValues(), it is
conservatively correct to leave them in, and they are not used for
anything important.

llvm-svn: 139792
2011-09-15 15:24:16 +00:00
Jakob Stoklund Olesen
08e2c3836e Revert r139782, "RemoveCopyByCommutingDef doesn't need hasPHIKill()."
It does, after all.

RemoveCopyByCommutingDef rewrites the uses of one particular value
number in A. It doesn't know how to rewrite phi uses, so there can't be
any.

llvm-svn: 139787
2011-09-15 06:27:32 +00:00
Benjamin Kramer
346abfde05 DWARF: Make DIE printing more bulletproof.
llvm-svn: 139786
2011-09-15 05:43:00 +00:00
Nick Lewycky
d984768671 Add some more DWARF extensions from:
1. http://gcc.gnu.org/wiki/TemplateParmsDwarf
2. ftp://ftp.software.ibm.com/software/os390/czos/dwarf/mips_extensions.pdf

llvm-svn: 139784
2011-09-15 05:21:03 +00:00
Jakob Stoklund Olesen
83f23f23cd Stop verifying hasPHIKill() flags.
There is only one legitimate use remaining, in addIntervalsForSpills().
All other calls to hasPHIKill() are only used to update PHIKill flags.

The addIntervalsForSpills() function is part of the old spilling
framework, only used by linearscan.

llvm-svn: 139783
2011-09-15 05:16:30 +00:00
Jakob Stoklund Olesen
f6c29ab2d4 RemoveCopyByCommutingDef doesn't need hasPHIKill().
Instead, let HasOtherReachingDefs() test for defs in B that overlap any
phi-defs in A as well.  This test is slightly different, but almost
identical.

A perfectly precise test would only check those phi-defs in A that are
reachable from AValNo.

llvm-svn: 139782
2011-09-15 05:03:50 +00:00
Jakob Stoklund Olesen
2c232210b1 It is safe to remat a value killed by phis.
The source live range is recomputed using shrinkToUses() which does
handle phis correctly.  The hasPHIKill() condition was relevant in the
old days when ReMaterializeTrivialDef() tried to recompute the live
range itself.

The shrinkToUses() function will mark the original def as dead when no
more uses and phi kills remain.  It is then removed by
runOnMachineFunction().

llvm-svn: 139781
2011-09-15 04:52:06 +00:00
Jakob Stoklund Olesen
a7631a56b4 Leave hasPHIKill flags alone in LiveInterval::RenumberValues.
It is conservatively correct to keep the hasPHIKill flags, even after
deleting PHI-defs.

The calculation can be very expensive after taildup has created a
quadratic number of indirectbr edges in the CFG, and the hasPHIKill flag
isn't used for anything after RenumberValues().

llvm-svn: 139780
2011-09-15 04:37:18 +00:00
Nick Lewycky
c76a2c21bc Update Dwarf enums list for DWARF 4.
Note that DW_TAG_rvalue_reference_type is officially 0x42, not 0x41.

llvm-svn: 139779
2011-09-15 04:23:44 +00:00
Benjamin Kramer
2b93dac69c DWARF: Print the number for unknown abbrev fields.
Thanks Nick!

llvm-svn: 139778
2011-09-15 04:15:59 +00:00
Benjamin Kramer
2dd05a53d3 DWARF: Fail gracefully when encountering unknown values in an abbrev.
llvm-svn: 139777
2011-09-15 04:00:58 +00:00
Nick Lewycky
4b85977895 Give structs with virtual methods a virtual destructor.
llvm-svn: 139776
2011-09-15 03:41:51 +00:00
Benjamin Kramer
8f332f8767 DWARF: Silence GCC -Wsign-compare warning.
llvm-svn: 139775
2011-09-15 03:20:04 +00:00
Benjamin Kramer
551244d899 DWARF: Fix indentation.
llvm-svn: 139774
2011-09-15 03:11:09 +00:00
Benjamin Kramer
8013b8f7e0 DWARF: Include <algorithm> explicitly.
llvm-svn: 139773
2011-09-15 02:19:33 +00:00
Benjamin Kramer
ceca872d69 DWARF: Add basic support for line tables.
The llvm-dwarfdump output isn't very verbose yet.

llvm-svn: 139771
2011-09-15 02:12:05 +00:00
Eli Friedman
2109f34467 Make demanded-elt simplification for shufflevector slightly stronger. Spotted by inspection.
llvm-svn: 139768
2011-09-15 01:14:29 +00:00
Andrew Trick
e5bb7267ff [regcoalescing] bug fix for RegistersDefinedFromSameValue.
An improper SlotIndex->VNInfo lookup was leading to unsafe copy removal.
Fixes PR10920 401.bzip2 miscompile with no IV rewrite.

llvm-svn: 139765
2011-09-15 01:09:33 +00:00
Eli Friedman
7cb90dcbce Fix the code creating VZEXT_LOAD so that it creates the right memoperand. Issue spotted in -debug output. I can't think of any practical effects at the moment, but it might matter if we start doing more aggressive alias analysis in CodeGen.
llvm-svn: 139758
2011-09-14 23:42:45 +00:00
Douglas Gregor
66d5bb28aa Update the comment for system_temp_directory() to indicate when it
will ignore the erasedOnReboot option, and properly escape the
backslash in "C:\TEMP". Thanks to Aaron and Francois.

llvm-svn: 139755
2011-09-14 23:21:47 +00:00
Jim Grosbach
669e269758 Thumb2 assembly parsing and encoding for PKH.
llvm-svn: 139754
2011-09-14 23:16:41 +00:00
Jim Grosbach
c1475b0f3f ARMv7a has the PKH instructions.
llvm-svn: 139753
2011-09-14 23:16:34 +00:00
Devang Patel
e5dc3a1f1e Add support to emit debug info for C++0x nullptr type.
llvm-svn: 139751
2011-09-14 23:13:28 +00:00
Jim Grosbach
1a4f264f52 ARM tighten up the register classes for the PKH instructions.
llvm-svn: 139748
2011-09-14 22:52:14 +00:00
Owen Anderson
86f1fb2955 Fix a crasher in Thumb2 MOV-immediate encoding for certain inputs.
llvm-svn: 139747
2011-09-14 22:46:14 +00:00
Bill Wendling
7694b6d655 Include limits.h to make sure PATH_MAX is known on Solaris 10.
Patch by Joakim Johansson!

llvm-svn: 139743
2011-09-14 21:49:42 +00:00
Jim Grosbach
e841adae12 Thumb2 assembly parsing and encoding for MVN.
llvm-svn: 139739
2011-09-14 21:24:41 +00:00
Owen Anderson
04d8803035 Nested IT blocks are UNPREDICTABLE. Mark them as such when disassembling them.
llvm-svn: 139736
2011-09-14 21:06:21 +00:00
Jim Grosbach
585e3c779f Thumb2 assembly parsing and encoding for MUL.
llvm-svn: 139735
2011-09-14 21:00:40 +00:00
Benjamin Kramer
00648a5c53 DWARF: Generate the address lookup table from the DIE tree if .debug_aranges is not available.
Ported from LLDB.

llvm-svn: 139732
2011-09-14 20:52:27 +00:00
Douglas Gregor
96b7e66374 Add a simple routine to determine the typical system directory for
temporary data. 

llvm-svn: 139725
2011-09-14 20:27:01 +00:00
Jim Grosbach
b1c70aab3e Thumb2 assembly parsing and encoding for MSR/MRS.
Fix a bug in handling default flags for both ARM and Thumb encodings.

llvm-svn: 139721
2011-09-14 20:03:46 +00:00
Jim Grosbach
932d409524 Thumb2 assembly parsing for MOV in IT block.
Select the right 16 vs. 32 bit encoding in an IT block.

llvm-svn: 139714
2011-09-14 19:12:11 +00:00
Benjamin Kramer
3f753c2372 DWARF: Reorder fields to reduce padding.
llvm-svn: 139712
2011-09-14 18:34:47 +00:00
Dan Gohman
f4d52b917e Don't mark objc_retainBlock as nounwind. It calls user copy constructors
which could theoretically throw.

llvm-svn: 139710
2011-09-14 18:33:34 +00:00
Dan Gohman
223fdfc56f objc_retainBlock is not NoModRef because it can update forwarding pointers
in memory relevant to the optimizer. rdar://10050579.

llvm-svn: 139708
2011-09-14 18:13:00 +00:00
Jim Grosbach
41c8bdfdd9 ARM fix assembly parser handling of ranges in register lists.
Clean up register list handling in general a bit to explicitly check things
like all the registers being from the same register class.

rdar://8883573

llvm-svn: 139707
2011-09-14 18:08:35 +00:00
Benjamin Kramer
fd97743d22 DWARF: Improve indentation of DIE dumping so it's easier to see the structure.
llvm-svn: 139705
2011-09-14 17:54:56 +00:00
Jakob Stoklund Olesen
44b1ea8bae Ignore the cloning of unknown registers.
THe LRE_DidCloneVirtReg callback may be called with vitual registers
that RAGreedy doesn't even know about yet.  In that case, there are no
data structures to update.

llvm-svn: 139702
2011-09-14 17:34:37 +00:00