llvm-mirror/test/Analysis
Sebastian Pop 93d67e228d DA: remove uses of GEP, only ask SCEV
It's been quite some time the Dependence Analysis (DA) is broken,
as it uses the GEP representation to "identify" multi-dimensional arrays.
It even wrongly detects multi-dimensional arrays in single nested loops:

from test/Analysis/DependenceAnalysis/Coupled.ll, example @couple6
;; for (long int i = 0; i < 50; i++) {
;; A[i][3*i - 6] = i;
;; *B++ = A[i][i];

DA used to detect two subscripts, which makes no sense in the LLVM IR
or in C/C++ semantics, as there are no guarantees as in Fortran of
subscripts not overlapping into a next array dimension:

maximum nesting levels = 1
SrcPtrSCEV = %A
DstPtrSCEV = %A
using GEPs
subscript 0
    src = {0,+,1}<nuw><nsw><%for.body>
    dst = {0,+,1}<nuw><nsw><%for.body>
    class = 1
    loops = {1}
subscript 1
    src = {-6,+,3}<nsw><%for.body>
    dst = {0,+,1}<nuw><nsw><%for.body>
    class = 1
    loops = {1}
Separable = {}
Coupled = {1}

With the current patch, DA will correctly work on only one dimension:

maximum nesting levels = 1
SrcSCEV = {(-2424 + %A)<nsw>,+,1212}<%for.body>
DstSCEV = {%A,+,404}<%for.body>
subscript 0
    src = {(-2424 + %A)<nsw>,+,1212}<%for.body>
    dst = {%A,+,404}<%for.body>
    class = 1
    loops = {1}
Separable = {0}
Coupled = {}

This change removes all uses of GEP from DA, and we now only rely
on the SCEV representation.

The patch does not turn on -da-delinearize by default, and so the DA analysis
will be more conservative in the case of multi-dimensional memory accesses in
nested loops.

I disabled some interchange tests, as the DA is not able to disambiguate
the dependence anymore. To make DA stronger, we may need to
compute a bound on the number of iterations based on the access functions
and array dimensions.

The patch cleans up all the CHECKs in test/Transforms/LoopInterchange/*.ll to
avoid checking for snippets of LLVM IR: this form of checking is very hard to
maintain. Instead, we now check for output of the pass that are more meaningful
than dozens of lines of LLVM IR. Some tests now require -debug messages and thus
only enabled with asserts.

Patch written by Sebastian Pop and Aditya Kumar.

Differential Revision: https://reviews.llvm.org/D35430

llvm-svn: 326837
2018-03-06 21:55:59 +00:00
..
AliasSet Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
AssumptionCache Add files I seem to have dropped in my revert (r290086). 2016-12-19 08:32:13 +00:00
BasicAA Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
BlockFrequencyInfo Revert part of "Cleanup some GraphTraits iteration code" 2017-12-08 22:42:43 +00:00
BranchProbabilityInfo [BPI] Detect branches in loops that make themselves not taken 2018-02-23 17:17:31 +00:00
CallGraph Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
CFLAliasAnalysis [CFLAA] Add missing break; note things are broken. 2017-05-31 02:35:26 +00:00
ConstantFolding Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
CostModel [X86] Add silvermont fp arithmetic cost model tests 2018-03-05 22:13:22 +00:00
Delinearization [SCEV] Assume parameters coming from function calls contain IVs 2017-05-27 15:17:49 +00:00
DemandedBits Fix tests after move to utohexstr. 2017-12-28 17:00:37 +00:00
DependenceAnalysis DA: remove uses of GEP, only ask SCEV 2018-03-06 21:55:59 +00:00
DivergenceAnalysis AMDGPU: Change DivergenceAnalysis for function arguments 2017-04-19 17:42:34 +00:00
DominanceFrontier Another fix the testcase introduced by r261903 - Add the missing matches 2016-02-26 03:41:47 +00:00
Dominators [Dominators] Don't compute DFS InOut numbers eagerly. 2017-06-30 01:28:21 +00:00
GlobalsModRef Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
IVUsers Re-enable "[SCEV] Do not fold dominated SCEVUnknown into AddRecExpr start" 2017-05-26 06:47:04 +00:00
LazyCallGraph [LCG] Spell the printing pass pipeline name for the lazy call graph 2016-03-10 11:24:06 +00:00
LazyValueAnalysis [JumpThreading] Preservation of DT and LVI across the pass 2018-01-12 21:06:48 +00:00
Lint [Lint] Upgrade uses of MemoryIntrinic::getAlignment() to new API. (NFCI) 2018-01-31 16:42:15 +00:00
LoopAccessAnalysis [LAA] Allow more run-time alias checks by coercing pointer expressions to AddRecExprs 2017-09-12 07:48:22 +00:00
LoopInfo [PM] Port LoopInfo to the new pass manager, adding both a LoopAnalysis 2015-01-20 10:58:50 +00:00
MemoryDependenceAnalysis [PM] The assumption cache is fundamentally designed to be self-updating, 2017-01-15 00:26:18 +00:00
MemorySSA Remove alignment argument from memcpy/memmove/memset in favour of alignment attributes (Step 1) 2018-01-19 17:13:12 +00:00
PostDominators [Dominators] Include infinite loops in PostDominatorTree 2017-08-15 18:14:57 +00:00
ProfileSummary Reapply r271728 after adding move cobstructor for ProfileSummaryInfo 2016-06-03 22:54:26 +00:00
RegionInfo [Dominators] Include infinite loops in PostDominatorTree 2017-08-15 18:14:57 +00:00
ScalarEvolution [SCEV] Smart range calculation for SCEVUnknown Phis 2018-03-01 06:56:48 +00:00
ScopedNoAliasAA [ScopedNoAliasAA] Make test basic.ll less confusing 2016-03-07 17:49:10 +00:00
TypeBasedAliasAnalysis [Analysis] Support aggregate access types in TBAA 2018-02-02 14:09:22 +00:00
ValueTracking [ValueTracking] add recursion depth param to matchSelectPattern 2018-01-24 15:20:37 +00:00
alias-analysis-uses.ll Add an "addUsedAAAnalyses" helper function 2016-02-09 01:21:57 +00:00