1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-24 13:33:37 +02:00
llvm-mirror/test/CodeGen/Thumb2
Wei Mi 00d0d9c981 [SCEV] Try to reuse existing value during SCEV expansion
Current SCEV expansion will expand SCEV as a sequence of operations
and doesn't utilize the value already existed. This will introduce
redundent computation which may not be cleaned up throughly by
following optimizations.

This patch introduces an ExprValueMap which is a map from SCEV to the
set of equal values with the same SCEV. When a SCEV is expanded, the
set of values is checked and reused whenever possible before generating
a sequence of operations.

The original commit triggered regressions in Polly tests. The regressions
exposed two problems which have been fixed in current version.

1. Polly will generate a new function based on the old one. To generate an
instruction for the new function, it builds SCEV for the old instruction,
applies some tranformation on the SCEV generated, then expands the transformed
SCEV and insert the expanded value into new function. Because SCEV expansion
may reuse value cached in ExprValueMap, the value in old function may be
inserted into new function, which is wrong.
   In SCEVExpander::expand, there is a logic to check the cached value to
be used should dominate the insertion point. However, for the above
case, the check always passes. That is because the insertion point is
in a new function, which is unreachable from the old function. However
for unreachable node, DominatorTreeBase::dominates thinks it will be
dominated by any other node.
   The fix is to simply add a check that the cached value to be used in
expansion should be in the same function as the insertion point instruction.

2. When the SCEV is of scConstant type, expanding it directly is cheaper than
reusing a normal value cached. Although in the cached value set in ExprValueMap,
there is a Constant type value, but it is not easy to find it out -- the cached
Value set is not sorted according to the potential cost. Existing reuse logic
in SCEVExpander::expand simply chooses the first legal element from the cached
value set.
   The fix is that when the SCEV is of scConstant type, don't try the reuse
logic. simply expand it.

Differential Revision: http://reviews.llvm.org/D12090

llvm-svn: 259736
2016-02-04 01:27:38 +00:00
..
2009-07-17-CrossRegClassCopy.ll
2009-07-21-ISelBug.ll
2009-07-23-CPIslandBug.ll
2009-07-30-PEICrash.ll
2009-08-01-WrongLDRBOpc.ll
2009-08-02-CoalescerBug.ll
2009-08-04-CoalescerAssert.ll
2009-08-04-CoalescerBug.ll
2009-08-04-ScavengerAssert.ll
2009-08-04-SubregLoweringBug2.ll
2009-08-04-SubregLoweringBug3.ll
2009-08-04-SubregLoweringBug.ll Revert "Change memcpy/memset/memmove to have dest and source alignments." 2015-11-19 05:56:52 +00:00
2009-08-06-SpDecBug.ll
2009-08-07-CoalescerBug.ll
2009-08-07-NeonFPBug.ll
2009-08-08-ScavengerAssert.ll
2009-08-10-ISelBug.ll
2009-08-21-PostRAKill4.ll
2009-09-01-PostRAProlog.ll
2009-09-28-ITBlockBug.ll
2009-10-15-ITBlockBranch.ll
2009-11-01-CopyReg2RegBug.ll
2009-11-11-ScavengerAssert.ll
2009-11-13-STRDBug.ll
2009-12-01-LoopIVUsers.ll [SCEV] Try to reuse existing value during SCEV expansion 2016-02-04 01:27:38 +00:00
2010-01-06-TailDuplicateLabels.ll
2010-01-19-RemovePredicates.ll
2010-02-11-phi-cycle.ll
2010-02-24-BigStack.ll
2010-03-08-addi12-ccout.ll
2010-03-15-AsmCCClobber.ll
2010-04-15-DynAllocBug.ll
2010-04-26-CopyRegCrash.ll
2010-05-24-rsbs.ll
2010-06-14-NEONCoalescer.ll
2010-06-19-ITBlockCrash.ll
2010-06-21-TailMergeBug.ll
2010-08-10-VarSizedAllocaBug.ll
2010-11-22-EpilogueBug.ll
2010-12-03-AddSPNarrowing.ll
2011-04-21-FILoweringBug.ll
2011-06-07-TwoAddrEarlyClobber.ll
2011-12-16-T2SizeReduceAssert.ll
2012-01-13-CBNZBug.ll Revert "Change memcpy/memset/memmove to have dest and source alignments." 2015-11-19 05:56:52 +00:00
2013-02-19-tail-call-register-hint.ll
2013-03-02-vduplane-nonconstant-source-index.ll
2013-03-06-vector-sext-operand-scalarize.ll
aapcs.ll ARMLoadStoreOptimizer: Create LDRD/STRD on thumb2 2015-07-21 00:18:59 +00:00
aligned-constants.ll [MC] Use .p2align instead of .align 2016-01-26 00:03:25 +00:00
aligned-spill.ll
bfi.ll
bfx.ll
buildvector-crash.ll
carry.ll Address buildbot fallout from r259065 2016-01-28 18:59:04 +00:00
cbnz.ll
constant-islands-jump-table.ll
constant-islands-new-island-padding.ll
constant-islands-new-island.ll
constant-islands.ll
cortex-fp.ll
crash.ll [ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions 2015-09-30 10:56:37 +00:00
cross-rc-coalescing-1.ll
cross-rc-coalescing-2.ll
div.ll
emit-unwinding.ll [ARM] Handle t2ADDri in ARMAsmPrinter::EmitUnwindingInstruction. 2015-11-10 00:10:41 +00:00
float-cmp.ll [SDAG] Optimize unordered comparison in soft-float mode (patch by Anton Nadolskiy) 2015-07-15 08:39:35 +00:00
float-intrinsics-double.ll LegalizeDAG: Fix and improve FCOPYSIGN/FABS legalization 2015-11-12 01:02:47 +00:00
float-intrinsics-float.ll [ARM] Use correct half-precision functions in EABI mode 2015-10-07 16:58:49 +00:00
float-ops.ll
frameless2.ll
frameless.ll
ifcvt-compare.ll [ARM] Enable shrink-wrapping by default. 2015-11-18 00:40:54 +00:00
ifcvt-neon.ll
inflate-regs.ll
inlineasm.ll
large-call.ll
large-stack.ll
ldr-str-imm12.ll
lit.local.cfg
longMACt.ll
lsr-deficiency.ll
machine-licm.ll [ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions 2015-09-30 10:56:37 +00:00
mul_const.ll
pic-load.ll ARM: Add scheduling information for LDRLIT instructions to swift scheduling model 2015-07-17 23:18:26 +00:00
segmented-stacks.ll
setjmp_longjmp.ll Arm: Don't define a label twice with two setjmps in a function. 2015-07-16 22:34:20 +00:00
stack_guard_remat.ll
tail-call-r9.ll
thumb2-adc.ll
thumb2-add2.ll
thumb2-add3.ll
thumb2-add4.ll
thumb2-add5.ll
thumb2-add6.ll
thumb2-add.ll
thumb2-and2.ll
thumb2-and.ll
thumb2-asr2.ll
thumb2-asr.ll
thumb2-bcc.ll
thumb2-bfc.ll
thumb2-bic.ll
thumb2-branch.ll
thumb2-call-tc.ll
thumb2-call.ll
thumb2-cbnz.ll
thumb2-clz.ll
thumb2-cmn2.ll
thumb2-cmn.ll
thumb2-cmp2.ll
thumb2-cmp.ll
thumb2-eor2.ll
thumb2-eor.ll
thumb2-ifcvt1-tc.ll
thumb2-ifcvt1.ll [ARM] Enable shrink-wrapping by default. 2015-11-18 00:40:54 +00:00
thumb2-ifcvt2.ll [ARM] Enable shrink-wrapping by default. 2015-11-18 00:40:54 +00:00
thumb2-ifcvt3.ll
thumb2-jtb.ll
thumb2-ldm.ll
thumb2-ldr_ext.ll
thumb2-ldr_post.ll
thumb2-ldr_pre.ll
thumb2-ldr.ll
thumb2-ldrb.ll
thumb2-ldrd.ll
thumb2-ldrh.ll
thumb2-lsl2.ll
thumb2-lsl.ll
thumb2-lsr2.ll
thumb2-lsr3.ll
thumb2-lsr.ll
thumb2-mla.ll
thumb2-mls.ll
thumb2-mov.ll
thumb2-mul.ll
thumb2-mulhi.ll [ARM] Renaming +t2dsp feature into +dsp, as discussed on llvm-dev 2015-10-23 17:19:19 +00:00
thumb2-mvn2.ll
thumb2-mvn.ll
thumb2-neg.ll
thumb2-orn2.ll
thumb2-orn.ll
thumb2-orr2.ll
thumb2-orr.ll
thumb2-pack.ll
thumb2-rev16.ll
thumb2-rev.ll
thumb2-ror.ll
thumb2-rsb2.ll
thumb2-rsb.ll
thumb2-sbc.ll
thumb2-select_xform.ll
thumb2-select.ll
thumb2-shifter.ll
thumb2-smla.ll [ARM] Renaming +t2dsp feature into +dsp, as discussed on llvm-dev 2015-10-23 17:19:19 +00:00
thumb2-smul.ll [ARM] Renaming +t2dsp feature into +dsp, as discussed on llvm-dev 2015-10-23 17:19:19 +00:00
thumb2-spill-q.ll [ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions 2015-09-30 10:56:37 +00:00
thumb2-str_post.ll
thumb2-str_pre.ll
thumb2-str.ll
thumb2-strb.ll
thumb2-strh.ll
thumb2-sub2.ll
thumb2-sub3.ll
thumb2-sub4.ll
thumb2-sub5.ll
thumb2-sub.ll
thumb2-sxt_rot.ll
thumb2-sxt-uxt.ll
thumb2-tbb.ll [MC] Use .p2align instead of .align 2016-01-26 00:03:25 +00:00
thumb2-tbh.ll
thumb2-teq2.ll
thumb2-teq.ll
thumb2-tst2.ll
thumb2-tst.ll
thumb2-uxt_rot.ll Fix test case label check 2015-11-20 20:24:49 +00:00
thumb2-uxtb.ll
tls1.ll
tls2.ll
tpsoft.ll
v8_IT_1.ll [ARM][NEON] Use address space in vld([1234]|[234]lane) and vst([1234]|[234]lane) instructions 2015-09-30 10:56:37 +00:00
v8_IT_2.ll
v8_IT_3.ll Distribute the weight on the edge from switch to default statement to edges generated in lowering switch. 2015-09-01 01:42:16 +00:00
v8_IT_4.ll
v8_IT_5.ll [ARM] Enable shrink-wrapping by default. 2015-11-18 00:40:54 +00:00
v8_IT_6.ll