Chandler Carruth 908e62868c [SDAG] Make the DAGCombine worklist not grow endlessly due to duplicate ... insertions. The old behavior could cause arbitrarily bad memory usage in the DAG combiner if there was heavy traffic of adding nodes already on the worklist to it. This commit switches the DAG combine worklist to work the same way as the instcombine worklist where we null-out removed entries and only add new entries to the worklist. My measurements of codegen time shows slight improvement. The memory utilization is unsurprisingly dominated by other factors (the IR and DAG itself I suspect). This change results in subtle, frustrating churn in the particular order in which DAG combines are applied which causes a number of minor regressions where we fail to match a pattern previously matched by accident. AFAICT, all of these should be using AddToWorklist to directly or should be written in a less brittle way. None of the changes seem drastically bad, and a few of the changes seem distinctly better. A major change required to make this work is to significantly harden the way in which the DAG combiner handle nodes which become dead (zero-uses). Previously, we relied on the ability to "priority-bump" them on the combine worklist to achieve recursive deletion of these nodes and ensure that the frontier of remaining live nodes all were added to the worklist. Instead, I've introduced a routine to just implement that precise logic with no indirection. It is a significantly simpler operation than that of the combiner worklist proper. I suspect this will also fix some other problems with the combiner. I think the x86 changes are really minor and uninteresting, but the avx512 change at least is hiding a "regression" (despite the test case being just noise, not testing some performance invariant) that might be looked into. Not sure if any of the others impact specific "important" code paths, but they didn't look terribly interesting to me, or the changes were really minor. The consensus in review is to fix any regressions that show up after the fact here. Thanks to the other reviewers for checking the output on other architectures. There is a specific regression on ARM that Tim already has a fix prepped to commit. Differential Revision: http://reviews.llvm.org/D4616 llvm-svn: 213727		2014-07-23 07:08:53 +00:00
..
2004-11-29-ShrCrash.ll
2004-11-30-shift-crash.ll
2004-11-30-shr-var-crash.ll
2004-12-12-ZeroSizeCommon.ll
2005-01-14-SetSelectCrash.ll
2005-01-14-UndefLong.ll
2005-08-12-rlwimi-crash.ll
2005-09-02-LegalizeDuplicatesCalls.ll
2005-10-08-ArithmeticRotate.ll
2005-11-30-vastart-crash.ll
2006-01-11-darwin-fp-argument.ll
2006-01-20-ShiftPartsCrash.ll
2006-04-01-FloatDoubleExtend.ll
2006-04-05-splat-ish.ll
2006-04-19-vmaddfp-crash.ll
2006-05-12-rlwimi-crash.ll
2006-07-07-ComputeMaskedBits.ll
2006-07-19-stwbrx-crash.ll
2006-08-11-RetVector.ll
2006-08-15-SelectionCrash.ll
2006-09-28-shift_64.ll
2006-10-13-Miscompile.ll
2006-10-17-brcc-miscompile.ll
2006-10-17-ppc64-alloca.ll
2006-11-10-DAGCombineMiscompile.ll
2006-11-29-AltivecFPSplat.ll
2006-12-07-LargeAlloca.ll
2006-12-07-SelectCrash.ll
2007-01-04-ArgExtension.ll
2007-01-15-AsmDialect.ll
2007-01-29-lbrx-asm.ll
2007-01-31-InlineAsmAddrMode.ll
2007-02-16-AlignPacked.ll
2007-02-16-InlineAsmNConstraint.ll
2007-02-23-lr-saved-twice.ll
2007-03-24-cntlzd.ll
2007-03-30-SpillerCrash.ll
2007-04-24-InlineAsm-I-Modifier.ll
2007-04-30-InlineAsmEarlyClobber.ll
2007-05-03-InlineAsm-S-Constraint.ll
2007-05-14-InlineAsmSelectCrash.ll
2007-05-22-tailmerge-3.ll
2007-05-30-dagcombine-miscomp.ll
2007-06-28-BCCISelBug.ll
2007-08-04-CoalescerAssert.ll
2007-09-04-AltivecDST.ll
2007-09-07-LoadStoreIdxForms.ll
2007-09-08-unaligned.ll
2007-09-11-RegCoalescerAssert.ll
2007-09-12-LiveIntervalsAssert.ll
2007-10-16-InlineAsmFrameOffset.ll
2007-10-18-PtrArithmetic.ll
2007-10-21-LocalRegAllocAssert2.ll
2007-10-21-LocalRegAllocAssert.ll
2007-11-04-CoalescerCrash.ll
2007-11-16-landingpad-split.ll	Add the ability to use GEPs for address sinking in CGP	2014-04-12 00:59:48 +00:00
2007-11-19-VectorSplitting.ll
2008-02-05-LiveIntervalsAssert.ll
2008-02-09-LocalRegAllocAssert.ll
2008-03-05-RegScavengerAssert.ll
2008-03-17-RegScavengerCrash.ll
2008-03-18-RegScavengerAssert.ll
2008-03-24-AddressRegImm.ll
2008-03-24-CoalescerBug.ll
2008-03-26-CoalescerBug.ll
2008-04-10-LiveIntervalCrash.ll
2008-04-16-CoalescerBug.ll
2008-04-23-CoalescerCrash.ll
2008-05-01-ppc_fp128.ll
2008-06-19-LegalizerCrash.ll
2008-06-21-F128LoadStore.ll
2008-06-23-LiveVariablesCrash.ll
2008-07-10-SplatMiscompile.ll	[PATCH] Correct type used for VADD_SPLAT optimization on PowerPC	2014-05-27 15:57:51 +00:00
2008-07-15-Bswap.ll
2008-07-15-Fabs.ll
2008-07-15-SignExtendInreg.ll
2008-07-17-Fneg.ll
2008-07-24-PPC64-CCBug.ll
2008-09-12-CoalescerBug.ll
2008-10-17-AsmMatchingOperands.ll
2008-10-28-f128-i32.ll
2008-10-28-UnprocessedNode.ll
2008-10-31-PPCF128Libcalls.ll
2008-12-02-LegalizeTypeAssert.ll
2009-01-16-DeclareISelBug.ll
2009-03-17-LSRBug.ll
2009-05-28-LegalizeBRCC.ll
2009-07-16-InlineAsm-M-Operand.ll
2009-08-17-inline-asm-addr-mode-breakage.ll
2009-09-18-carrybit.ll
2009-11-15-ProcImpDefsBug.ll
2009-11-25-ImpDefBug.ll
2010-02-04-EmptyGlobal.ll
2010-02-12-saveCR.ll
2010-03-09-indirect-call.ll
2010-04-01-MachineCSEBug.ll
2010-05-03-retaddr1.ll
2010-10-11-Fast-Varargs.ll
2010-12-18-PPCStackRefs.ll
2011-12-05-NoSpillDupCR.ll
2011-12-06-SpillAndRestoreCR.ll
2011-12-08-DemandedBitsMiscompile.ll
2012-09-16-TOC-entry-check.ll
2012-10-11-dynalloc.ll
2012-10-12-bitcast.ll
2012-11-16-mischedcall.ll
2013-05-15-preinc-fold.ll
2013-07-01-PHIElimBug.ll
a2-fp-basic.ll
a2q-stackalign.ll
a2q.ll
aa-tbaa.ll	Reenable use of TBAA during CodeGen	2014-04-12 01:26:00 +00:00
addc.ll
addi-reassoc.ll
addrfuncstr.ll
alias.ll	[PPC] Use alias symbols in address computation.	2014-05-29 15:41:38 +00:00
align.ll
allocate-r0.ll
altivec-ord.ll
and_add.ll
and_sext.ll
and_sra.ll
and-branch.ll
and-elim.ll
and-imm.ll
anon_aggr.ll	[PowerPC] Fix FrameIndex handling in SelectAddressRegImm	2014-07-20 22:26:40 +00:00
ashr-neg1.ll
asm-dialect.ll
asm-Zy.ll
asym-regclass-copy.ll
atomic-1.ll	IR: add "cmpxchg weak" variant to support permitted failure.	2014-06-13 14:24:07 +00:00
atomic-2.ll	IR: add "cmpxchg weak" variant to support permitted failure.	2014-06-13 14:24:07 +00:00
Atomics-32.ll	IR: add "cmpxchg weak" variant to support permitted failure.	2014-06-13 14:24:07 +00:00
Atomics-64.ll
available-externally.ll	[PowerPC] 32-bit ELF PIC support	2014-07-18 23:29:49 +00:00
bdzlr.ll
big-endian-actual-args.ll
big-endian-call-result.ll
big-endian-formal-args.ll
branch-opt.ll
bswap-load-store.ll
buildvec_canonicalize.ll
bv-pres-v8i1.ll
bv-widen-undef.ll
byval-agg-info.ll
calls.ll
can-lower-ret.ll
cc.ll	[PowerPC] Add a full condition code register to make the "cc" clobber work	2014-04-04 15:15:57 +00:00
cmp-cmp.ll
coalesce-ext.ll
compare-duplicate.ll
compare-simm.ll
complex-return.ll	[SDAG] Make the DAGCombine worklist not grow endlessly due to duplicate	2014-07-23 07:08:53 +00:00
constants.ll
copysignl.ll
cr1eq-no-extra-moves.ll
cr1eq.ll
cr_spilling.ll
cr-spills.ll
crash.ll
crbit-asm.ll
crbits.ll
crsave.ll
ctr-cleanup.ll
ctrloop-asm.ll
ctrloop-cpsgn.ll
ctrloop-fp64.ll
ctrloop-i64.ll
ctrloop-large-ec.ll
ctrloop-le.ll	[PowerPC] Implement some additional TLI callbacks	2014-04-12 21:52:38 +00:00
ctrloop-lt.ll	[PowerPC] Implement some additional TLI callbacks	2014-04-12 21:52:38 +00:00
ctrloop-ne.ll
ctrloop-reg.ll
ctrloop-s000.ll
ctrloop-sh.ll	[PowerPC] On PPC32, 128-bit shifts might be runtime calls	2014-05-11 16:23:29 +00:00
ctrloop-sums.ll
ctrloop-udivti3.ll
ctrloops.ll
cttz.ll
darwin-labels.ll
dbg.ll	DebugInfo: Sure up subprogram variable list handling with more assertions and fewer conditionals.	2014-05-14 21:52:46 +00:00
DbgValueOtherTargets.test
dcbt-sched.ll
delete-node.ll
div-2.ll
dyn-alloca-aligned.ll
early-ret2.ll	Rename loop unrolling and loop vectorizer metadata to have a common prefix.	2014-06-25 15:41:00 +00:00
early-ret.ll
empty-functions.ll
emptystruct.ll
eqv-andc-orc-nor.ll
extsh.ll
fabs.ll
fast-isel-binary.ll
fast-isel-br-const.ll
fast-isel-call.ll
fast-isel-cmp-imm.ll
fast-isel-conversion-p5.ll	[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction)	2014-06-24 20:05:18 +00:00
fast-isel-conversion.ll	[PPC64] Fix PR20071 (fctiduz generated for targets lacking that instruction)	2014-06-24 20:05:18 +00:00
fast-isel-crash.ll
fast-isel-ext.ll
fast-isel-fold.ll
fast-isel-GEP-coalesce.ll
fast-isel-indirectbr.ll
fast-isel-load-store.ll
fast-isel-redefinition.ll
fast-isel-ret.ll
fast-isel-shifter.ll
fastisel-gep-promote-before-add.ll
fcpsgn.ll
float-asmprint.ll
float-to-int.ll	[PowerPC] Make use of VSX f64 <-> i64 conversion instructions	2014-03-23 05:35:00 +00:00
floatPSA.ll
fma.ll
fnabs.ll
fneg.ll
fold-li.ll
fold-zero.ll
fp_to_uint.ll
fp-branch.ll
fp-int-fp.ll
fpcopy.ll
frame-size.ll
frameaddr.ll
Frames-alloca.ll	[PowerPC] Allow stack frames without parameter save area	2014-06-23 13:47:52 +00:00
Frames-large.ll	[PowerPC] Allow stack frames without parameter save area	2014-06-23 13:47:52 +00:00
Frames-leaf.ll
Frames-small.ll	[PowerPC] Allow stack frames without parameter save area	2014-06-23 13:47:52 +00:00
frounds.ll
fsel.ll
fsl-e500mc.ll
fsl-e5500.ll
fsqrt.ll
func-addr.ll	[PPC64] Fix PR19893 - improve code generation for local function addresses	2014-06-16 21:36:02 +00:00
glob-comp-aa-crash.ll
hello-reloc.s	llvm-readobj: fix MachO relocatoin printing a bit.	2014-07-04 10:57:56 +00:00
hello.ll
hidden-vis-2.ll
hidden-vis.ll
i1-to-double.ll
i32-to-float.ll	[PowerPC] Make use of VSX f64 <-> i64 conversion instructions	2014-03-23 05:35:00 +00:00
i64_fp_round.ll
i64_fp.ll
i64-to-float.ll	[PowerPC] Make use of VSX f64 <-> i64 conversion instructions	2014-03-23 05:35:00 +00:00
i128-and-beyond.ll
iabs.ll
ifcvt.ll
illegal-element-type.ll
in-asm-f64-reg.ll
indexed-load.ll	[PowerPC] PR19796: Also match ISD::TargetConstant in isIntS16Immediate	2014-05-20 17:20:34 +00:00
indirectbr.ll
inlineasm-copy.ll
inlineasm-i64-reg.ll
int-fp-conv-0.ll
int-fp-conv-1.ll
inverted-bool-compares.ll
isel-rc-nox0.ll
isel.ll
ispositive.ll
itofp128.ll
jaggedstructs.ll
LargeAbsoluteAddr.ll
lbzux.ll
lha.ll
lit.local.cfg	Reduce verbiage of lit.local.cfg files	2014-06-09 22:42:55 +00:00
load-constant-addr.ll
load-shift-combine.ll
long-compare.ll
longdbl-truncate.ll
lsa.ll
lsr-postinc-pos.ll
mask64.ll
mature-mc-support.ll
mcm-1.ll
mcm-2.ll
mcm-3.ll
mcm-4.ll
mcm-5.ll
mcm-6.ll
mcm-7.ll
mcm-8.ll
mcm-9.ll
mcm-10.ll	[PowerPC] Implement some additional TLI callbacks	2014-04-12 21:52:38 +00:00
mcm-11.ll	[PowerPC] Implement some additional TLI callbacks	2014-04-12 21:52:38 +00:00
mcm-12.ll
mcm-default.ll
mcm-obj-2.ll	[PowerPC] Implement some additional TLI callbacks	2014-04-12 21:52:38 +00:00
mcm-obj.ll
mem_update.ll
mem-rr-addr-mode.ll
misched-inorder-latency.ll
misched.ll
mul-neg-power-2.ll
mul-with-overflow.ll
mulhs.ll
mulli64.ll
mult-alt-generic-powerpc64.ll
mult-alt-generic-powerpc.ll
named-reg-alloc-r0.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r1-64.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r1.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r2-64.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r2.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r13-64.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
named-reg-alloc-r13.ll	[PowerPC] Add global named register support	2014-05-11 19:29:11 +00:00
neg.ll
negctr.ll
no-dead-strip.ll
novrsave.ll
optcmp.ll
or-addressing-mode.ll
popcnt.ll
ppc32-i1-vaarg.ll
ppc32-pic.ll	[PowerPC] 32-bit ELF PIC support	2014-07-18 23:29:49 +00:00
ppc32-vacopy.ll
ppc64-32bit-addic.ll
ppc64-abi-extend.ll
ppc64-align-long-double.ll
ppc64-altivec-abi.ll	[PowerPC] Fix on-stack AltiVec arguments with 64-bit SVR4	2014-06-23 12:36:34 +00:00
ppc64-byval-align.ll	[PowerPC] Fix testcase regression	2014-07-07 19:41:54 +00:00
ppc64-calls.ll	[PowerPC] Simplify and improve loading into TOC register	2014-06-18 17:52:49 +00:00
ppc64-crash.ll
ppc64-cyclecounter.ll
ppc64-linux-func-size.ll
ppc64-prefetch.ll
ppc64-smallarg.ll	[PowerPC] Fix small argument stack slot offset for LE	2014-06-20 16:34:05 +00:00
ppc64-toc.ll
ppc64-vaarg-int.ll
ppc64-zext.ll
ppc64le-aggregates.ll	[PowerPC] ELFv2 aggregate passing support	2014-07-21 00:13:26 +00:00
ppc64le-calls.ll	[PowerPC] ELFv2 stack space reduction	2014-07-20 23:43:15 +00:00
ppc64le-crsave.ll	[PowerPC] ELFv2 explicit CFI for CR fields	2014-07-21 00:03:18 +00:00
ppc64le-localentry.ll	[PowerPC] ELFv2 function call changes	2014-07-20 23:31:44 +00:00
ppc64le-smallarg.ll	[PowerPC] ELFv2 stack space reduction	2014-07-20 23:43:15 +00:00
ppc440-fp-basic.ll
ppc440-msync.ll
ppc-prologue.ll
ppc-vaarg-agg.ll
ppcf128-1-opt.ll
ppcf128-1.ll
ppcf128-2.ll
ppcf128-3.ll
ppcf128-4.ll
ppcf128-endian.ll	Fix ppcf128 component access on little-endian systems	2014-07-03 15:06:47 +00:00
pr3711_widen_bit.ll
pr12757.ll
pr13641.ll
pr13891.ll
pr15031.ll
pr15359.ll
pr15630.ll
pr15632.ll
pr16556-2.ll
pr16556.ll
pr16573.ll
pr17168.ll
pr17354.ll
private.ll
pwr3-6x.ll
pwr7-gt-nop.ll
quadint-return.ll
r31.ll
recipest.ll
reg-coalesce-simple.ll
reg-names.ll
reloc-align.ll
remap-crash.ll
remat-imm.ll
resolvefi-basereg.ll	[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex	2014-06-27 13:04:12 +00:00
resolvefi-disp.ll	[PowerPC] Fix invalid displacement created by LocalStackAlloc	2014-07-11 17:19:31 +00:00
retaddr.ll
return-val-i128.ll
rlwimi2.ll
rlwimi3.ll
rlwimi-and.ll
rlwimi-commute.ll
rlwimi-dyn-and.ll	[PowerPC] Fix rlwimi isel when mask is not constant	2014-04-13 17:10:58 +00:00
rlwimi-keep-rsh.ll
rlwimi.ll
rlwinm2.ll
rlwinm.ll
rotl-2.ll
rotl-64.ll
rotl.ll
rounding-ops.ll
rs-undef-use.ll
s000-alias-misched.ll
sdag-ppcf128.ll
sections.ll	[PowerPC] 32-bit ELF PIC support	2014-07-18 23:29:49 +00:00
select_lt0.ll
select-cc.ll
set0-v8i16.ll
setcc_no_zext.ll
seteq-0.ll
shift128.ll
shl_elim.ll
shl_sext.ll
sign_ext_inreg1.ll
sj-ctr-loop.ll
sjlj.ll
small-arguments.ll
spill-nor0.ll
splat-bug.ll	[PATCH] Correct type used for VADD_SPLAT optimization on PowerPC	2014-05-27 15:57:51 +00:00
srl-mask.ll
stack-protector.ll
stack-realign.ll	[PowerPC] 32-bit ELF PIC support	2014-07-18 23:29:49 +00:00
std-unal-fi.ll
stdux-constuse.ll
stfiwx-2.ll
stfiwx.ll
store-load-fwd.ll
store-update.ll
structsinmem.ll
structsinregs.ll
stubs.ll
stwu8.ll
stwu-gta.ll
stwux.ll
sub-bv-types.ll
subc.ll
subsumes-pred-regs.ll	[SDAG] Make the DAGCombine worklist not grow endlessly due to duplicate	2014-07-23 07:08:53 +00:00
svr4-redzone.ll	[PowerPC] Allow stack frames without parameter save area	2014-06-23 13:47:52 +00:00
tailcall1-64.ll
tailcall1.ll
tailcallpic1.ll
tls-pic.ll
tls.ll
trampoline.ll
unal4-std.ll
unal-altivec2.ll
unal-altivec.ll
unaligned.ll
unsafe-math.ll
unwind-dw2-g.ll
unwind-dw2.ll
vaddsplat.ll
varargs-struct-float.ll	[PowerPC] ELFv2 aggregate passing support	2014-07-21 00:13:26 +00:00
varargs.ll
vcmp-fold.ll
vec_auto_constant.ll
vec_br_cmp.ll
vec_buildvector_loadstore.ll
vec_call.ll
vec_cmp.ll	Fix typos	2014-06-19 19:41:26 +00:00
vec_constants.ll
vec_conv.ll
vec_extload.ll
vec_fmuladd.ll
vec_fneg.ll
vec_insert.ll
vec_misaligned.ll	[PPC64LE] Generate correct code for unaligned little-endian vector loads	2014-06-09 22:00:52 +00:00
vec_mul.ll	[PPC64LE] Generate correct little-endian code for v16i8 multiply	2014-06-09 16:06:29 +00:00
vec_perf_shuffle.ll
vec_rounding.ll
vec_select.ll
vec_shift.ll
vec_shuffle_le.ll	[PPC64LE] Recognize shufflevector patterns for little endian	2014-06-10 14:35:01 +00:00
vec_shuffle.ll
vec_splat_constant.ll
vec_splat.ll
vec_sqrt.ll
vec_vrsave.ll
vec_zero.ll
vec-abi-align.ll
vector-identity-shuffle.ll
vector.ll
vperm-instcombine.ll	[PPC64LE] Add test case for r210282 commit	2014-06-05 22:57:38 +00:00
vperm-lowering.ll	[PPC64LE] Fix lowering of BUILD_VECTOR and SHUFFLE_VECTOR for little endian	2014-06-06 14:06:26 +00:00
vrsave-spill.ll
vrspill.ll
vsx-args.ll	[PowerPC] v2[fi]64 need to be explicitly passed in VSX registers	2014-03-28 19:58:11 +00:00
vsx-fma-m.ll	[PowerPC] Add subregister classes for f64 VSX values	2014-03-29 05:29:01 +00:00
vsx-self-copy.ll	[PowerPC] Use a small cleanup pass to remove VSX self copies	2014-03-27 23:12:31 +00:00
vsx-spill.ll	[PowerPC] Add subregister classes for f64 VSX values	2014-03-29 05:29:01 +00:00
vsx.ll	[PowerPC] Fix FrameIndex handling in SelectAddressRegImm	2014-07-20 22:26:40 +00:00
vtable-reloc.ll	[ppc64] Avoid copy relocs in named rodata sections	2014-03-14 12:45:22 +00:00
weak_def_can_be_hidden.ll
zero-not-run.ll