Commit Graph

2545 Commits

Author SHA1 Message Date
Evan Cheng
1cc3984148 Allow targets more controls on what nodes are scheduled by reg pressure, what for latency in hybrid mode.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104293 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-20 23:26:43 +00:00
Bob Wilson
63b8845e78 Handle Neon v2f64 and v2i64 vector shuffles as register copies.
This fixes the remaining issue with pr7167.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104257 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-20 18:39:53 +00:00
Evan Cheng
211ffa1351 Code refactoring: pull SchedPreference enum from TargetLowering.h to TargetMachine.h and put it in its own namespace.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104147 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 20:19:50 +00:00
Evan Cheng
9085f98b32 t2LEApcrel and tLEApcrel are re-materializable. This makes it possible to hoist more loads during machine LICM.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104115 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 07:28:01 +00:00
Evan Cheng
27fa722311 Use 'adr' for LEApcrel and LEApcrel. Mark LEApcrel re-materializable.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104114 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 07:26:50 +00:00
Evan Cheng
5fd1c9be2d Mark pattern-less mayLoad / mayStore instructions neverHasSideEffects. These do not have other un-modeled side effects.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104111 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 06:07:03 +00:00
Evan Cheng
3c3195cbf1 Target instruction selection should copy memoperands.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104110 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 06:06:09 +00:00
Evan Cheng
ea420b20d4 Mark a few more pattern-less instructions with neverHasSideEffects. This is especially important on instructions like t2LEApcreal which are prime candidate for machine LICM.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104102 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 01:52:25 +00:00
Evan Cheng
28dad2a5ca Sink dag combine's post index load / store code that swap base ptr and index into the target hook. Only the target knows whether the swap is safe. In Thumb2 mode, the offset must be an immediate. rdar://7998649
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104060 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-18 21:31:17 +00:00
Jakob Stoklund Olesen
535af4a320 ARMBaseRegisterInfo::estimateRSStackSizeLimit() could return prematurely with a
too large limit.

The function would return immediately when finding an addrmode 3/5 instruction.
It needs to keep scanning in case there is an addrmode 6 instruction which drops
the limit to 0.

A test case is very difficult to produce because it will only fail when the
scavenger is used.

rdar://problem/7894847

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103995 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-17 23:29:23 +00:00
Evan Cheng
47006be498 vmov of immediates are trivially re-materializable.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103982 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-17 21:54:50 +00:00
Bob Wilson
7f43fd84db Fix a regression in 464.h264 for thumb1 and thumb2 nightly tests.
Obvious in retrospect but not fun to debug.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103969 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-17 20:31:13 +00:00
Evan Cheng
6206124250 Turn on -neon-reg-sequence by default.
Using NEON load / store multiple instructions will no longer create gobs of vmov of D registers!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-17 19:51:20 +00:00
Evan Cheng
9c207ac0dc No reason not to run the NEON domain croassing fix up pass in thumb2 mode.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103917 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-17 01:11:46 +00:00
Anton Korobeynikov
bd91ea53f8 Chris said that the comment char should be escaped. Fix all the occurences of "@" in *.td
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103903 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-16 09:15:36 +00:00
Anton Korobeynikov
4878b8415f Generalize the ARM DAG combiner of mul with constants to all power-of-two cases.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103901 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-16 08:54:20 +00:00
Evan Cheng
8f6de385d6 Model vst lane instructions with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103898 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-16 03:27:48 +00:00
Anton Korobeynikov
a9790d739a Some cheap DAG combine goodness for multiplication with a particular constant.
This can be extended later on to handle more "complex" constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103881 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 18:16:59 +00:00
Anton Korobeynikov
418d1d954d "trap" pseudo-op turned out to be apple-local.
Temporary emit it as raw bytes until it will be added to binutils as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103878 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 17:19:20 +00:00
Evan Cheng
7189fd03fa Model 128-bit vld lane with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103868 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 07:53:37 +00:00
Evan Cheng
4782b1e2ca v4i64 and v8i64 are only synthesizable when NEON is available.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103855 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 02:20:21 +00:00
Evan Cheng
06b666c705 Allow TargetLowering::getRegClassFor() to be called on illegal types. Also
allow target to override it in order to map register classes to illegal
but synthesizable types. e.g. v4i64, v8i64 for ARM / NEON.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103854 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 02:18:07 +00:00
Evan Cheng
7092c2bfcb Model 64-bit lane vld with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103851 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-15 01:36:29 +00:00
Evan Cheng
b990a2f249 Teach two-address pass to do some coalescing while eliminating REG_SEQUENCE
instructions.

e.g.
%reg1026<def> = VLDMQ %reg1025<kill>, 260, pred:14, pred:%reg0
%reg1027<def> = EXTRACT_SUBREG %reg1026, 6
%reg1028<def> = EXTRACT_SUBREG %reg1026<kill>, 5
...
%reg1029<def> = REG_SEQUENCE %reg1028<kill>, 5, %reg1027<kill>, 6, %reg1028, 7, %reg1027, 8, %reg1028, 9, %reg1027, 10, %reg1030<kill>, 11, %reg1032<kill>, 12

After REG_SEQUENCE is eliminated, we are left with:

%reg1026<def> = VLDMQ %reg1025<kill>, 260, pred:14, pred:%reg0
%reg1029:6<def> = EXTRACT_SUBREG %reg1026, 6
%reg1029:5<def> = EXTRACT_SUBREG %reg1026<kill>, 5

The regular coalescer will not be able to coalesce reg1026 and reg1029 because it doesn't
know how to combine sub-register indices 5 and 6. Now 2-address pass will consult the
target whether sub-registers 5 and 6 of reg1026 can be combined to into a larger
sub-register (or combined to be reg1026 itself as is the case here). If it is possible, 
it will be able to replace references of reg1026 with reg1029 + the larger sub-register
index.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103835 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 23:21:14 +00:00
Evan Cheng
12c24690c7 Model VST*_UPD and VST*oddUPD pair with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103833 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 22:54:52 +00:00
Bill Wendling
b92187a410 Rename "HasCalls" in MachineFrameInfo to "AdjustsStack" to better describe what
the variable actually tracks.

N.B., several back-ends are using "HasCalls" as being synonymous for something
that adjusts the stack. This isn't 100% correct and should be looked into.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103802 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 21:14:32 +00:00
Evan Cheng
5c6aba2e3a Model VLD*_UPD and VLD*odd_UPD pair with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103790 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 18:54:59 +00:00
Evan Cheng
22c687b642 Added a QQQQ register file to model 4-consecutive Q registers.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103760 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 02:13:41 +00:00
Evan Cheng
7f68719517 Fix comments.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103749 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-14 00:21:45 +00:00
Evan Cheng
c4ca40eb5e Add comment about the pseudo registers QQ, each of which is a pair of Q registers.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103731 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 20:02:08 +00:00
Bob Wilson
1190c14b54 Fix pr7110: For non-Darwin targets UnspilledCS1GPRs may include high registers.
Do not use those for Thumb1 functions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103730 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 19:58:24 +00:00
Daniel Dunbar
1860e7dcfd Fix -Asserts warning.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103694 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 03:19:36 +00:00
Evan Cheng
69b9f9883e Bring back VLD1q and VST1q and use them for reloading / spilling Q registers. This allows folding loads and stores into VMOVQ.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103692 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 01:12:06 +00:00
Evan Cheng
d929f77738 Expand VMOVQQ into a pair of VMOVQ.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103684 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 00:17:02 +00:00
Evan Cheng
020cc1b4d0 Mark some pattern-less instructions as neverHasSideEffects.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103683 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 00:16:46 +00:00
Evan Cheng
431300797b Fix some potential issues in the pseudo instruction expansion phase: copy implicit operands and memoperands. Also, expand instructions even if their defs are "dead" since they may have implicit kill operands.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103667 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-12 23:13:12 +00:00
Evan Cheng
a4d73d01c4 Remove a dead fixme.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103642 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-12 20:20:22 +00:00
Rafael Espindola
18c1021ec1 Add support for movi32 of global values to the new (MC) asm printer.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103576 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-12 05:16:34 +00:00
Evan Cheng
5bdc2aa264 vst instructions are modeled as this:
v1024 = REG_SEQUENCE ...
v1025 = EXTRACT_SUBREG v1024, 5
v1026 = EXTRACR_SUBREG v1024, 6
      = VSTxx <addr>, v1025, v1026

The REG_SEQUENCE ensures the sources that feed into the VST instruction
are getting the right register allocation so they form a large super-
register. The extract_subreg will be coalesced away all would just work:
v1024 = REG_SEQUENCE ...
      = VSTxx <addr>, v1024:5, v1024:6

The problem is if the coalescer isn't run, the extract_subreg instructions
would stick around and there is no assurance v1025 and v1026 will get the
right registers.

As a short term workaround, teach the NEON pre-allocation pass to transfer
the sub-register indices over. An alternative would be do it 2addr pass
when reg_sequence's are eliminated. But that *seems* wrong and require
updating liveness information.

Another alternative is to do this in the scheduler when the instructions are
created. But that would mean somehow the scheduler this has to be done for
correctness reason. That's yucky as well. So for now, we are leaving this
in the target specific pass.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103540 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-12 01:42:50 +00:00
Evan Cheng
9647f3d981 Avoid breaking vstd when reg_sequence is not used.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103513 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 21:07:36 +00:00
Duncan Sands
16d8f8bd91 I got tired of VISIBILITY_HIDDEN colliding with the gcc enum. Rename it
to LLVM_LIBRARY_VISIBILITY and introduce LLVM_GLOBAL_VISIBILITY, which is
the opposite, for future use by dragonegg.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103495 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 20:16:09 +00:00
Dan Gohman
ff7a562751 Implement a bunch more TargetSelectionDAGInfo infrastructure.
Move EmitTargetCodeForMemcpy, EmitTargetCodeForMemset, and
EmitTargetCodeForMemmove out of TargetLowering and into
SelectionDAGInfo to exercise this.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103481 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 17:31:57 +00:00
Dan Gohman
419e4f9263 Remove the TargetLowering::getSubtarget() virtual function, which
was unused. TargetMachine::getSubtarget() is used instead.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103474 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 16:21:03 +00:00
Evan Cheng
fb3611daad Select @llvm.trap to the special B with 1111 condition (i.e. trap) instruction.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103459 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 07:26:32 +00:00
Evan Cheng
0ce537a9db Model some vst3 and vst4 with reg_sequence.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103453 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-11 01:19:40 +00:00
Evan Cheng
e9e2ba05de Model some vld3 instructions with REG_SEQUENCE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103437 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-10 21:26:24 +00:00
Evan Cheng
603afbfe2a Model vld2 / vst2 with reg_sequence.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103411 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-10 17:34:18 +00:00
Jim Grosbach
4b77f6a85a Clean up the conditional for handling of sign_extend_inreg based on
whether the extract instructions are available.

rdar://7956878



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103277 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-07 18:34:55 +00:00
Evan Cheng
435d499177 Use VLD2q32 / VST2q32 to reload / spill QQ (pair of Q) registers when stack slot is sufficiently aligned. Use VLDMD / VSTMD otherwise.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103235 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-07 02:04:02 +00:00
Evan Cheng
07a6d9391c Use VSTMD / VLDMD for spills and reloads of Q registers instead of VSTMQ / VLDQ. The later are aliases which ought to be eliminated but we can't because they are used for storing and loading v2f64 values.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103234 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-07 01:54:08 +00:00