Commit Graph

28 Commits

Author SHA1 Message Date
Evan Cheng
48575f6ea7 Making use of VFP / NEON floating point multiply-accumulate / subtraction is
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
   of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
   additional pipeline stall. So it's frequently better to single codegen
   vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
   stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
   vmla + vmla is very bad. But this isn't ideal either:
     vmul
     vadd
     vmla
   Instead, we want to expand the second vmla:
     vmla
     vmul
     vadd
   Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
   faster.

Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.

A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
   compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
   fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
   vmla / vmls will trigger one of the special hazards.

Work in progress, only A+B are enabled.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
Evan Cheng
8239daf7c8 Two sets of changes. Sorry they are intermingled.
1. Fix pre-ra scheduler so it doesn't try to push instructions above calls to
   "optimize for latency". Call instructions don't have the right latency and
   this is more likely to use introduce spills.
2. Fix if-converter cost function. For ARM, it should use instruction latencies,
   not # of micro-ops since multi-latency instructions is completely executed
   even when the predicate is false. Also, some instruction will be "slower"
   when they are predicated due to the register def becoming implicit input.
   rdar://8598427


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118135 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-03 00:45:17 +00:00
Owen Anderson
e3cc84a43d Thread the determination of branch prediction hit rates back through the if-conversion heuristic APIs. For now,
stick with a constant estimate of 90% (branch predictors are good!), but we might find that we want to provide
more nuanced estimates in the future.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115364 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 22:45:50 +00:00
Owen Anderson
aa9f0a57d0 Provide an option to restore old-style if-conversion heuristics for Thumb2.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115339 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 20:28:06 +00:00
Owen Anderson
b20b85168c Part one of switching to using a more sane heuristic for determining if-conversion profitability.
Rather than having arbitrary cutoffs, actually try to cost model the conversion.

For now, the constants are tuned to more or less match our existing behavior, but these will be
changed to reflect realistic values as this work proceeds.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114973 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-28 18:32:13 +00:00
Evan Cheng
3ef1c8759a Teach if-converter to be more careful with predicating instructions that would
take multiple cycles to decode.
For the current if-converter clients (actually only ARM), the instructions that
are predicated on false are not nops. They would still take machine cycles to
decode. Micro-coded instructions such as LDM / STM can potentially take multiple
cycles to decode. If-converter should take treat them as non-micro-coded
simple instructions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113570 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-10 01:29:16 +00:00
Jakob Stoklund Olesen
ac27366700 Replace copyRegToReg with copyPhysReg for ARM.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108078 91177308-0d34-0410-b5e6-96231b3b80d8
2010-07-11 06:33:54 +00:00
Evan Cheng
13151432ed Change if-conversion block size limit checks to add some flexibility.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106901 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-25 22:42:03 +00:00
Evan Cheng
4d54e5b2dd Tail merging pass shall not break up IT blocks. rdar://8115404
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106517 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-22 01:18:16 +00:00
Evan Cheng
86050dc8cc Allow ARM if-converter to be run after post allocation scheduling.
- This fixed a number of bugs in if-converter, tail merging, and post-allocation
  scheduler. If-converter now runs branch folding / tail merging first to
  maximize if-conversion opportunities.
- Also changed the t2IT instruction slightly. It now defines the ITSTATE
  register which is read by instructions in the IT block.
- Added Thumb2 specific hazard recognizer to ensure the scheduler doesn't
  change the instruction ordering in the IT block (since IT mask has been
  finalized). It also ensures no other instructions can be scheduled between
  instructions in the IT block.

This is not yet enabled.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106344 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-18 23:09:54 +00:00
Evan Cheng
68fc2daf8f Allow target to place 2-address pass inserted copies in better spots. Thumb2 will use this to try to avoid breaking up IT blocks.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@105745 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-09 19:26:01 +00:00
Dan Gohman
34dcc6fadc Add a DebugLoc argument to TargetInstrInfo::copyRegToReg, so that it
doesn't have to guess.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103194 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-06 20:33:48 +00:00
Evan Cheng
746ad69e08 Add argument TargetRegisterInfo to loadRegFromStackSlot and storeRegToStackSlot.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103193 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-06 19:06:44 +00:00
Dan Gohman
864e2efce2 Remove the target hook TargetInstrInfo::BlockHasNoFallThrough in favor of
MachineBasicBlock::canFallThrough(), which is target-independent and more
thorough.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90634 91177308-0d34-0410-b5e6-96231b3b80d8
2009-12-05 00:44:40 +00:00
Evan Cheng
fdc834046e Refactor code.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86423 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-08 00:15:23 +00:00
Jim Grosbach
31c24bf5b3 80-column cleanup of file header comments
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86408 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-07 22:00:39 +00:00
Evan Cheng
d457e6e9a5 Refactor code. Fix a potential missing check. Teach isIdentical() about tLDRpci_pic.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86330 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-07 04:04:34 +00:00
Evan Cheng
78e5c1140a - Add TargetInstrInfo::isIdentical(). It's similar to MachineInstr::isIdentical
except it doesn't care if the definitions' virtual registers differ. This is
  used by machine LICM and other MI passes to perform CSE.
- Teach Thumb2InstrInfo::isIdentical() to check two t2LDRpci_pic are identical.
  Since pc relative constantpool entries are always different, this requires it
  it check if the values can actually the same.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86328 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-07 03:52:02 +00:00
Jim Grosbach
bd79fc8ef2 80-columns
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86310 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-07 00:13:30 +00:00
Evan Cheng
b9803a8fa6 - Add pseudo instructions tLDRpci_pic and t2LDRpci_pic which does a pc-relative
load of a GV from constantpool and then add pc. It allows the code sequence to
  be rematerializable so it would be hoisted by machine licm.
- Add a late pass to break these pseudo instructions into a number of real
  instructions. Also move the code in Thumb2 IT pass that breaks up t2MOVi32imm
  to this pass. This is done before post regalloc scheduling to allow the
  scheduler to proper schedule these instructions. It also allow them to be
  if-converted and shrunk by later passes.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@86304 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-06 23:52:48 +00:00
Evan Cheng
6495f63945 - More refactoring. This gets rid of all of the getOpcode calls.
- This change also makes it possible to switch between ARM / Thumb on a
  per-function basis.
- Fixed thumb2 routine which expand reg + arbitrary immediate. It was using
  using ARM so_imm logic.
- Use movw and movt to do reg + imm when profitable.
- Other code clean ups and minor optimizations.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77300 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-28 05:48:47 +00:00
Evan Cheng
5732ca084a Use t2LDRi12 and t2STRi12 to load / store to / from stack frames. Eliminate more getOpcode calls.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@77181 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-27 03:14:20 +00:00
David Goodwin
5ff58b5c3a Correctly handle the Thumb-2 imm8 addrmode. Specialize frame index elimination more exactly for Thumb-2 to get better code gen.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@76919 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-24 00:16:18 +00:00
David Goodwin
b53cc014d0 Fix frame index elimination to correctly handle thumb-2 addressing modes that don't allow negative offsets. During frame elimination convert *i12 opcode to a *i8 when necessary due to a negative offset.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@76883 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-23 17:06:46 +00:00
Anton Korobeynikov
b8e9ac834a Emit cross regclass register moves for thumb2.
Minor code duplication cleanup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@76124 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-16 23:26:06 +00:00
David Goodwin
f1daf7d8ab Use common code for both ARM and Thumb-2 instruction and register info.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@75067 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-08 23:10:31 +00:00
David Goodwin
334c26473b Checkpoint Thumb2 Instr info work. Generalized base code so that it can be shared between ARM and Thumb2. Not yet activated because register information must be generalized first.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@75010 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-08 16:09:28 +00:00
David Goodwin
b50ea5c48f Checkpoint refactoring of ThumbInstrInfo and ThumbRegisterInfo into Thumb1InstrInfo, Thumb2InstrInfo, Thumb1RegisterInfo and Thumb2RegisterInfo. Move methods from ARMInstrInfo to ARMBaseInstrInfo to prepare for sharing with Thumb2.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@74731 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-02 22:18:33 +00:00