This patch applies the restrictions on the number of domain conjuncts
also to the domain parts of piecewise affine expressions we generate.
To this end the wording is change slightly. It was needed to support
complex additions featuring zext-instructions but it also fixes PR27045.
lnt profitable runs reports only little changes that might be noise:
Compile Time:
Polybench/[...]/2mm +4.34%
SingleSource/[...]/stepanov_container -2.43%
Execution Time:
External/[...]/186_crafty -2.32%
External/[...]/188_ammp -1.89%
External/[...]/473_astar -1.87%
llvm-svn: 264514
This fixes PR27035. While we now exclude MemIntrinsics from the
polyhedral model if they would access "null" we could exploit this
even more, e.g., remove all parameter combinations that would lead to
the execution of this statement from the context.
llvm-svn: 264284
This might be useful to evaluate the benefit of us handling modref funciton
calls. Also, a new bug that was triggered by modref function calls was
recently reported http://llvm.org/PR27035. To ensure the same issue does not
cause troubles for other people, we temporarily disable this until the bug
is resolved.
llvm-svn: 264140
ISL can conclude additional conditions on parameters from restrictions
on loop variables. Such conditions persist when leaving the loop and the
loop variable is projected out. This results in a narrower domain for
exiting the loop than entering it and is logically impossible for
non-infinite loops.
We fix this by not adding a lower bound i>=0 when constructing BB
domains, but defer it to when also the upper bound it computed, which
was done redundantly even before this patch.
This reduces the number of LNT fails with -polly-process-unprofitable
-polly-position=before-vectorizer from 8 to 6.
llvm-svn: 264118
We bail out if current scop has a complex control flow as this could lead to
building of large domain conditions. This is to reduce compile time. This
addresses r26382.
Contributed-by: Chris Jenneisch <chrisj@codeaurora.org>
Differential Revision: http://reviews.llvm.org/D18362
llvm-svn: 264105
Affine branches are fully modeled and regenerated from the polyhedral domain and
consequently do not require any input conditions to be propagated.
llvm-svn: 263678
This should fix PR19422.
Thanks to Jeremy Huddleston Sequoia for reporting this.
Thanks to Roman Gareev for his investigation and the reduced test case.
llvm-svn: 262612
In order to speed up compile time and to avoid random timeouts we now
separately track assumptions and restrictions. In this context
assumptions describe parameter valuations we need and restrictions
describe parameter valuations we do not allow. During AST generation
we create a runtime check for both, whereas the one for the
restrictions is negated before a conjunction is build.
Except the In-Bounds assumptions we currently only track restrictions.
Differential Revision: http://reviews.llvm.org/D17247
llvm-svn: 262328
Check the ModRefBehaviour of functions in order to decide whether or
not a call instruction might be acceptable.
Differential Revision: http://reviews.llvm.org/D5227
llvm-svn: 261866
From now on we bail only if a non-trivial alias group contains a non-affine
access, not when we discover aliasing and non-affine accesses are allowed.
llvm-svn: 261863
This patch adds support for memcpy, memset and memmove intrinsics. They are
represented as one (memset) or two (memcpy, memmove) memory accesses in the
polyhedral model. These accesses have an access range that describes the
summarized effect of the intrinsic, i.e.,
memset(&A[i], '$', N);
is represented as a write access from A[i] to A[i+N].
Differential Revision: http://reviews.llvm.org/D5226
llvm-svn: 261489
To support non-aligned accesses we introduce a virtual element size
for arrays that divides each access function used for this array. The
adjustment of the access function based on the element size of the
array was therefore moved after this virtual element size was
determined, thus after all accesses have been created.
Differential Revision: http://reviews.llvm.org/D17246
llvm-svn: 261226
So far we separated constant factors from multiplications, however,
only when they are at the outermost level of a parameter SCEV. Now,
we also separate constant factors from the parameter SCEV if the
outermost expression is a SCEVAddRecExpr. With the changes to the
SCEVAffinator we can now improve the extractConstantFactor(...)
function at will without worrying about any other code part. Thus,
if needed we can implement a more comprehensive
extractConstantFactor(...) function that will traverse the SCEV
instead of looking only at the outermost level.
Four test cases were affected. One did not change much and the other
three were simplified.
llvm-svn: 260859
We now distinguish invariant loads to the same memory location if they
have different types. This will cause us to pre-load an invariant
location once for each type that is used to access it. However, we can
thereby avoid invalid casting, especially if an array is accessed
though different typed/sized invariant loads.
This basically reverts the changes in r260023 but keeps the test
cases.
llvm-svn: 260045
We also disable this feature by default, as there are still some issues in
combination with invariant load hoisting that slipped through my initial
testing.
llvm-svn: 260025
Always use access-instruction pointer type to load the invariant values.
Otherwise mismatches between ScopArrayInfo element type and memory access
element type will result in invalid casts. These type mismatches are after
r259784 a lot more common and also arise with types of different size, which
have not been handled before.
Interestingly, this change actually simplifies the code, as we now have only
one code path that is always taken, rather then a standard code path for the
common case and a "fixup" code path that replaces the standard code path in
case of mismatching types.
llvm-svn: 260009
The previously implemented approach is to follow value definitions and
create write accesses ("push defs") while searching for uses. This
requires the same relatively validity- and requirement conditions to be
replicated at multiple locations (PHI instructions, other instructions,
uses by PHIs).
We replace this by iterating over the uses in a SCoP ("pull in
requirements"), and add writes only when at least one read has been
added. It turns out to be simpler code because each use is only iterated
over once and writes are added for the first access that reads it. We
need another iteration to identify escaping values (uses not in the
SCoP), which also makes the difference between such accesses more
obvious. As a side-effect, the order of scalar MemoryAccess can change.
Differential Revision: http://reviews.llvm.org/D15706
llvm-svn: 259987
This allows code such as:
void multiple_types(char *Short, char *Float, char *Double) {
for (long i = 0; i < 100; i++) {
Short[i] = *(short *)&Short[2 * i];
Float[i] = *(float *)&Float[4 * i];
Double[i] = *(double *)&Double[8 * i];
}
}
To model such code we use as canonical element type of the modeled array the
smallest element type of all original array accesses, if type allocation sizes
are multiples of each other. Otherwise, we use a newly created iN type, where N
is the gcd of the allocation size of the types used in the accesses to this
array. Accesses with types larger as the canonical element type are modeled as
multiple accesses with the smaller type.
For example the second load access is modeled as:
{ Stmt_bb2[i0] -> MemRef_Float[o0] : 4i0 <= o0 <= 3 + 4i0 }
To support code-generating these memory accesses, we introduce a new method
getAccessAddressFunction that assigns each statement instance a single memory
location, the address we load from/store to. Currently we obtain this address by
taking the lexmin of the access function. We may consider keeping track of the
memory location more explicitly in the future.
We currently do _not_ handle multi-dimensional arrays and also keep the
restriction of not supporting accesses where the offset expression is not a
multiple of the access element type size. This patch adds tests that ensure
we correctly invalidate a scop in case these accesses are found. Both types of
accesses can be handled using the very same model, but are left to be added in
the future.
We also move the initialization of the scop-context into the constructor to
ensure it is already available when invalidating the scop.
Finally, we add this as a new item to the 2.9 release notes
Reviewers: jdoerfert, Meinersbur
Differential Revision: http://reviews.llvm.org/D16878
llvm-svn: 259784
We support now code such as:
void multiple_types(char *Short, char *Float, char *Double) {
for (long i = 0; i < 100; i++) {
Short[i] = *(short *)&Short[2 * i];
Float[i] = *(float *)&Float[4 * i];
Double[i] = *(double *)&Double[8 * i];
}
}
To support such code we use as element type of the modeled array the smallest
element type of all original array accesses. Accesses with larger types are
modeled as multiple accesses with the smaller type.
For example the second load access is modeled as:
{ Stmt_bb2[i0] -> MemRef_Float[o0] : 4i0 <= o0 <= 3 + 4i0 }
To support jscop-rewritable memory accesses we need each statement instance to
only be assigned a single memory location, which will be the address at which
we load the value. Currently we obtain this address by taking the lexmin of
the access function. We may consider keeping track of the memory location more
explicitly in the future.
llvm-svn: 259587
For schedule generation we assumed that the reverse post order traversal used by
the domain generation is sufficient, however it is not. Once a loop is
discovered, we have to completely traverse it, before we can generate the
schedule for any block/region that is only reachable through a loop exiting
block.
To this end, we add a "loop stack" that will keep track of loops we
discovered during the traversal but have not yet traversed completely.
We will never visit a basic block (or region) outside the most recent
(thus smallest) loop in the loop stack but instead queue such blocks
(or regions) in a waiting list. If the waiting list is not empty and
(might) contain blocks from the most recent loop in the loop stack the
next block/region to visit is drawn from there, otherwise from the
reverse post order iterator.
We exploit the new property of loops being always completed before additional
loops are processed, by removing the LoopSchedules map and instead keep all
information in LoopStack. This clarifies that we indeed always only keep a
stack of in-process loops, but will never keep incomplete schedules for an
arbitrary set of loops. As a result, we can simplify some of the existing code.
This patch also adds some more documentation about how our schedule construction
works.
This fixes http://llvm.org/PR25879
This patch is an modified version of Johannes Doerfert's initial fix.
Differential Revision: http://reviews.llvm.org/D15679
llvm-svn: 259354
Before adding a MK_Value READ MemoryAccess, check whether the read is
necessary or synthesizable. Synthesizable values are later generated by
the SCEVExpander and therefore do not need to be transferred
explicitly. This can happen because the check for synthesizability has
presumbly been forgotten in the case where a phi's incoming value has
been defined in a different statement.
Differential Revision: http://reviews.llvm.org/D15687
llvm-svn: 258998
Ensure that there is at most one phi write access per PHINode and
ScopStmt. In particular, this would be possible for non-affine
subregions with multiple exiting blocks. We replace multiple MAY_WRITE
accesses by one MUST_WRITE access. The written value is constructed
using a PHINode of all exiting blocks. The interpretation of the PHI
WRITE's "accessed value" changed from the incoming value to the PHI like
for PHI READs since there is no unique incoming value.
Because region simplification shuffles around PHI nodes -- particularly
with exit node PHIs -- the PHINodes at analysis time does not always
exist anymore in the code generation pass. We instead remember the
incoming block/value pair in the MemoryAccess.
Differential Revision: http://reviews.llvm.org/D15681
llvm-svn: 258809
Ensure there is at most one write access per definition of an
llvm::Value. Keep track of already created value write access by using
a (dense) map.
Replace addValueWriteAccess by ensureValueStore which can be uses more
liberally without worrying to add redundant accesses. It will be used,
e.g. in a logical correspondant for value reads -- ensureValueReload --
to ensure that the expected definition has been written when loading it.
Differential Revision: http://reviews.llvm.org/D15483
llvm-svn: 258807
The test case we look at does not necessarily require irreducible control flow,
but a normal loop is sufficient to create a non-affine region containing more
than one basic block that dominates the exit node. We replace this irreducible
control flow with a normal loop for the following reasons:
1) This is easier to understand
2) We will subsequently commit a patch that ensures Polly does not process
irreducible control flow.
Within non-affine regions, we could possibly handle irreducible control flow.
llvm-svn: 258496
ISL 0.16 will change how sets are printed which breaks 117 unit tests
that text-compare printed sets. This patch re-formats most of these unit
tests using a script and small manual editing on top of that. When
actually updating ISL, most work is done by just re-running the script
to adapt to the changed output.
Some tests that compare IR and tests with single CHECK-lines that can be
easily updated manually are not included here.
The re-format script will also be committed afterwards. The per-test
formatter invocation command lines options will not be added in the near
future because it is ad hoc and would overwrite the manual edits.
Ideally it also shouldn't be required anymore because ISL's set printing
has become more stable in 0.16.
Differential Revision: http://reviews.llvm.org/D16095
llvm-svn: 257851
Call assumeNoOutOfBound only in updateDimensionality to process situations
when new dimensions are added and new bounds checks are required.
Contributed-by: Tobias Grosser, Gareev Roman
llvm-svn: 257170
Scops that contain many complex branches are likely to result in complex domain
conditions that consist of a large (> 100) number of conjucts. Transforming
such domains is expensive and unlikely to result in efficient code. To avoid
long compile times we detect this case and skip such scops. In the future we may
improve this by either using non-affine subregions to hide such complex
condition structures or by exploiting in certain cases properties (e.g.,
dominance) that allow us to construct the domains of a scop in a way that
results in a smaller number improving conjuncts.
Example of a code that results in complex iteration spaces:
loop.header
/ | \ \
A0 A2 A4 \
\ / \ / \
A1 A3 \
/ \ / \ |
B0 B2 B4 |
\ / \ / |
B1 B3 ^
/ \ / \ |
C0 C2 C4 |
\ / \ / /
C1 C3 /
\ / /
loop backedge
llvm-svn: 256123
This reverts commit r255471.
Johannes raised in the post-commit review of r255471 the concern that PHI
writes in non-affine regions with two exiting blocks are not really MUST_WRITE,
but we just know that at least one out of the set of all possible PHI writes
will be executed. Modeling all PHI nodes as MUST_WRITEs is probably save, but
adding the needed documentation for such a special case is probably not worth
the effort. Michael will be proposing a new patch that ensures only a single
PHI_WRITE is created for non-affine regions, which - besides other benefits -
should also allow us to use a single well-defined MUST_WRITE for such PHI
writes.
(This is not a full revert, but the condition and documentation have been
slightly extended)
llvm-svn: 255503
LLVM's IR guarantees that a value definition occurs before any use, and
also the value of a PHI must be one of the incoming values, "written"
in one of the incoming blocks. Hence, such writes are never conditional
in the context of a non-affine subregion.
llvm-svn: 255471
When introducing separate control flow for the original and optimized code we
introduce now a special 'ExitingBlock':
\ /
EnteringBB
|
SplitBlock---------\
_____|_____ |
/ EntryBB \ StartBlock
| (region) | |
\_ExitingBB_/ ExitingBlock
| |
MergeBlock---------/
|
ExitBB
/ \
This 'ExitingBlock' contains code such as the final_reloads for scalars, which
previously were just added to whichever statement/loop_exit/branch-merge block
had been generated last. Having an explicit basic block makes it easier to
find these constructs when looking at the CFG.
llvm-svn: 255107
gfortran (and fortran in general?) does not compute the address of an array
element directly from the array sizes (e.g., %s0, %s1), but takes first the
maximum of the sizes and 0 (e.g., max(0, %s0)) before multiplying the resulting
value with the per-dimension array subscript expressions. To successfully
delinearize index expressions as we see them in fortran, we first filter 'smax'
expressions out of the SCEV expression, use them to guess array size parameters
and only then continue with the existing delinearization.
llvm-svn: 253995
Trying to build up access functions for any of these blocks is likely to fail,
as error blocks may contain invalid/non-representable instructions, and blocks
dominated by error blocks may reference such instructions, which wil also cause
failures. As all of these blocks are anyhow assumed to not be executed, we can
just remove them early on.
This fixes http://llvm.org/PR25596
llvm-svn: 253818
Only when we check for wrapping we want to use the store size, for all
other cases we use the alloc size now.
Suggested by: Tobias Grosser <tobias@grosser.es>
llvm-svn: 252941
If an llvm.assume dominates the SCoP entry block and the assumed condition
can be expressed as an affine inequality we will now add it to the context.
Differential Revision: http://reviews.llvm.org/D14413
llvm-svn: 252851
Error blocks may contain arbitrary instructions, among them some which we can
not modeled correctly. As we do not generate ScopStmts for error blocks anyhow
there is no point in trying to generate access functions for them.
This fixes llvm.org/PR25494
llvm-svn: 252794
For complex inputs our current approach of construction the boundary context
may in rare cases become computationally so expensive that it is better to
abort. This change adds a compute out check that bounds the compuations we
spend on boundary context construction and bails out if this limit is reached.
We can probably make our boundary construction algorithm more efficient, but
this requires some more investigation and probably also some additional changes
to isl. Until these have been added, we bound the compile time to ensure our
buildbots are green.
llvm-svn: 252758
In certain rare cases (mostly -polly-process-unprofitable on large sequences
of conditions - often without any loop), we see some compile-time timeouts due
to the construction of an overly complex assumption context. This change limits
the number of disjuncts to 150 (adjustable), to prevent us from creating
assumptions contexts that are too large for even the compilation to finish.
The limit has been choosen as large as possible to make sure we do not
unnecessarily drop test coverage. If such cases also appear in
-polly-process-unprofitable=false mode we may need to think about this again,
as the current limitations may still allow assumptions that are way to complex
to be checked profitably at run-time.
There is also certainly room for improvement regarding how (and how efficient)
we construct an assumed context, but this requires some more thinking.
This completes llvm.org/PR25458
llvm-svn: 252750
Thinking more about the last commit I came to realize that for testing the
new functionality it is sufficient to verify that the iteration domains
we construct for a simple test case do not contain any of the complexity that
caused compile time issues for larger inputs.
llvm-svn: 252714
Previously, we just skipped error blocks during scop construction. With
this change we make sure we can construct domains for error blocks such that
these domains can be forwarded to subsequent basic blocks.
This change ensures that basic blocks that post-dominate and are dominated by
a basic block that branches to an error condition have the very same iteration
domain as the branching basic block. Before, this change we would construct
a domain that excludes all error conditions. Such domains could become _very_
complex and were undesirable to build.
Another solution would have been to drop these constraints using a
dominance/post-dominance check instead of modeling the error blocks. Such
a solution could also work in case of unreachable statements or infinite
loops in the scop. However, as we currently (to my believe incorrectly) model
unreachable basic blocks in the post-dominance tree, such a solution is not
yet feasible and requires first a change to LLVM's post-dominance tree
construction.
This commit addresses the most sever compile time issue reported in:
http://llvm.org/PR25458
llvm-svn: 252713
Especially for structs, the SAI object of a base pointer does not
describe all the types that the user might expect when he loads from
that base pointer. While we will still cast integers and pointers we
will now reload the value with the correct type if floating point and
non-floating point values are involved. However, there are now TODOs
where we use bitcasts instead of a proper conversion or reloading.
This fixes bug 25479.
llvm-svn: 252706
We now create all invariant equivalence classes for required invariant loads
instead of creating them on-demand. This way we can check if a parameter
references an invariant load that is actually not executed and was therefor
not materialized. If that happens the parameter is not materialized either.
This fixes bug 25469.
llvm-svn: 252701
In case we also model scalar reads it can happen that a pointer appears in both
a scalar read access as well as the base pointer of an array access. As this
is a little surprising, we add a specific test case to document this behaviour.
To my understanding it should be OK to have a read from an array A[] and
read/write accesses to A[...]. isl is treating these arrays as unrelated as
their dimensionality differs. This seems to be correct as A[] remains constant
throughout the execution of the scop and is not affected by the reads/writes to
A[...]. If this causes confusion, it might make sense to make this behaviour
more obvious by using different names (e.g., A_scalar[], A[...]).
llvm-svn: 252615
Memory references are now printed as follows:
Old New
Scalars: i64 MemRef_val[*] i64 MemRef_val;
Arrays: i64 MemRef_A[*][%m][%o][8] i64 MemRef_A[*][%m][%o];
We do not print any more information about the element size in the type. Such
information has already been available in a comment after the scalar/array
declaration. It was redundant and did not match well with what people were used
from C.
llvm-svn: 252602
If a SCoP contains error blocks we cannot use the domain constraints
to simplify the assumptions as the domain is already influenced by the
assumptions we took. Before this patch we did that and some assumptions
became self-fulfilling as they were implied by the domain constraints.
llvm-svn: 252424
Even if a scalar and memory access have the same base pointer, we cannot use
one SAI object as the type but also the number of dimensions are wrong. For
the attached test case this caused a crash in the invariant load hoisting,
though it could cause various other problems too.
This fixes bug 25428 and a execution time bug in MallocBench/cfrac.
Reported-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
llvm-svn: 252422
An incoming value from a block the is not inside the scop is an
external use, even if the phi is inside the scop. A previous fix in
r251208 did not apply if the phi is inside a non-affine subregion. We
move the check for this phi case before the non-affine subregion check.
llvm-svn: 252157
We do not need to model read-only statements in the SCoP as they will
not cause any side effects that are visible to the outside anyway.
Removing them should safe us time and might even simplify the ASTs we
generate.
Differential Revision: http://reviews.llvm.org/D14272
llvm-svn: 251948
ScalarEvolution doesn't allow the operands of an AddRec to be variant in the
loop of the AddRec. When we rewrite parameter SCEVs it might seem like the
new SCEV violates this property and ScalarEvolution will trigger an
assertion. To avoid this we move the start part out of an AddRec when we
rewrite it, thus avoid the operands to be possibly variant completely.
llvm-svn: 251945
If a base pointer load is preloaded, we have change the base pointer of
the derived SAI. However, as the derived SAI relationship is is
coarse grained, we need to check if we actually preloaded the base
pointer or a different element of the base pointer SAI array.
llvm-svn: 251881
In some cases different memory accesses access the very same array using a
different multi-dimensional array layout where the same dimensions have
different sizes. Instead of asserting when encountering this issue, we
gracefully bail out for this scop.
This fixes llvm.org/PR25252
llvm-svn: 251791
of the Region are external.
During code generation we split off the parts of the PHI nodes in the entry
block, which have incoming blocks that are not part of the region. As these
split-off PHI nodes then are external uses, we consequently also need to model
these uses in ScopInfo.
llvm-svn: 251208
There are several different kinds of constants that could occur in a
branch condition, however we can only handle the most interesting one
namely constant integers. To this end we have to treat others as
non-affine.
This fixes bug 25244.
llvm-svn: 250669
We build the schedule based on a traversal of the region and accumulate
information for each loop in it. The total schedule is associated with the
loop surrounding the SCoP, though it can happen that there are blocks in the
SCoP which are part of loops that are only partially in the SCoP. Instead of
associating information with them (they are not part of the SCoP and
consequently are not modeled) we have to associate the schedule information
with the surrounding loop if any.
This fixes bug 25240.
llvm-svn: 250668
Sorting is replaced by a demand driven code generation that will pre-load a
value when it is needed or, if it was not needed before, at some point
determined by the order of invariant accesses in the program. Only in very
little cases this demand driven pre-loading will kick in, though it will
prevent us from generating faulty code. An example where it is needed is
shown in:
test/ScopInfo/invariant_loads_complicated_dependences.ll
Invariant loads that appear in parameters but are not on the top-level (e.g.,
the parameter is not a SCEVUnknown) will now be treated correctly.
Differential Revision: http://reviews.llvm.org/D13831
llvm-svn: 250655
Polly can now be used as a analysis only tool as long as the code
generation is disabled. However, we do not have an alternative to the
independent blocks pass in place yet, though in the relevant cases
this does not seem to impact the performance much. Nevertheless, a
virtual alternative that allows the same transformations without
changing the input region will follow shortly.
llvm-svn: 250652
Accesses for exit node phis will be handled separately by
buildPHIAccesses if there is more than one exiting edge,
buildScalarDependences does not need to create additional SCALAR
accesses.
This is a corrected version of r250517, which was reverted in r250607.
Differential Revision: http://reviews.llvm.org/D13848
llvm-svn: 250622
We also allow such products for cases where 'Parameter' is loaded within the
scop, but where we can dynamically verify that the value of 'Parameter' remains
unchanged during the execution of the scop.
This change relies on Polly's new RequiredILS tracking infrastructure recently
contributed by Johannes.
llvm-svn: 250019
The domain generation can handle lazy && and || by default but eager
evaluated expressions were dismissed as non-affine. With this patch we
will allow arbitrary combinations of and/or bit-operations in the
conditions of branches.
Differential Revision: http://reviews.llvm.org/D13624
llvm-svn: 249971
If a (assumed) invariant location is loaded multiple times we
generated a parameter for each location. However, this caused compile
time problems for several benchmarks (e.g., 445_gobmk in SPEC2006 and
BT in the NAS benchmarks). Additionally, the code we generate is
suboptimal as we preload the same location multiple times and perform
the same checks on all the parameters that refere to the same value.
With this patch we consolidate the invariant loads in three steps:
1) During SCoP initialization required invariant loads are put in
equivalence classes based on their pointer operand. One
representing load is used to generate a parameter for the whole
class, thus we never generate multiple parameters for the same
location.
2) During the SCoP simplification we remove invariant memory
accesses that are in the same equivalence class. While doing so
we build the union of all execution domains as it is only
important that the location is at least accessed once.
3) During code generation we only preload one element of each
equivalence class with the unified execution domain. All others
are mapped to that preloaded value.
Differential Revision: http://reviews.llvm.org/D13338
llvm-svn: 249853
Drop an unused flag polly-allow-non-scev-backedge-taken-count and also
its occurrences from the tests.
Contributed-by: Chris Jenneisch <chrisj@codeaurora.org>
Differential Revision: http://reviews.llvm.org/D13400
llvm-svn: 249675
This replaces the support for user defined error functions by a
heuristic that tries to determine if a call to a non-pure function
should be considered "an error". If so the block is assumed not to be
executed at runtime. While treating all non-pure function calls as
errors will allow a lot more regions to be analyzed, it will also
cause us to dismiss a lot again due to an infeasible runtime context.
This patch tries to limit that effect. A non-pure function call is
considered an error if it is executed only in conditionally with
regards to a cheap but simple heuristic.
llvm-svn: 249611
This patch allows invariant loads to be used in the SCoP description,
e.g., as loop bounds, conditions or in memory access functions.
First we collect "required invariant loads" during SCoP detection that
would otherwise make an expression we care about non-affine. To this
end a new level of abstraction was introduced before
SCEVValidator::isAffineExpr() namely ScopDetection::isAffine() and
ScopDetection::onlyValidRequiredInvariantLoads(). Here we can decide
if we want a load inside the region to be optimistically assumed
invariant or not. If we do, it will be marked as required and in the
SCoP generation we bail if it is actually not invariant. If we don't
it will be a non-affine expression as before. At the moment we
optimistically assume all "hoistable" (namely non-loop-carried) loads
to be invariant. This causes us to expand some SCoPs and dismiss them
later but it also allows us to detect a lot we would dismiss directly
if we would ask e.g., AliasAnalysis::canBasicBlockModify(). We also
allow potential aliases between optimistically assumed invariant loads
and other pointers as our runtime alias checks are sound in case the
loads are actually invariant. Together with the invariant checks this
combination allows to handle a lot more than LICM can.
The code generation of the invariant loads had to be extended as we
can now have dependences between parameters and invariant (hoisted)
loads as well as the other way around, e.g.,
test/Isl/CodeGen/invariant_load_parameters_cyclic_dependence.ll
First, it is important to note that we cannot have real cycles but
only dependences from a hoisted load to a parameter and from another
parameter to that hoisted load (and so on). To handle such cases we
materialize llvm::Values for parameters that are referred by a hoisted
load on demand and then materialize the remaining parameters. Second,
there are new kinds of dependences between hoisted loads caused by the
constraints on their execution. If a hoisted load is conditionally
executed it might depend on the value of another hoisted load. To deal
with such situations we sort them already in the ScopInfo such that
they can be generated in the order they are listed in the
Scop::InvariantAccesses list (see compareInvariantAccesses). The
dependences between hoisted loads caused by indirect accesses are
handled the same way as before.
llvm-svn: 249607
This single option replaces -polly-detect-unprofitable and -polly-no-early-exit
and is supposed to be the only option that disables compile-time heuristics that
aim to bail out early on scops that are believed to not benefit from Polly
optimizations.
Suggested-by: Johannes Doerfert
llvm-svn: 249426
These flags are now always passed to all tests and need to be disabled if
not needed. Disabling these flags, rather than passing them to almost all
tests, significantly simplfies our RUN: lines.
llvm-svn: 249422
Polly's profitability heuristic saves compile time by skipping trivial scops or
scops were we know no good optimization can be applied. For almost all our tests
this heuristic makes little sense as we aim for minimal test cases when testing
functionality. Hence, in almost all cases this heuristic is better be disabled.
In preparation of disabling Polly's compile time heuristic by default in the
test suite we first explicitly enable it in the couple of test cases that really
use it (or run with/without heuristic side-by-side).
llvm-svn: 249418
This test case was XFAILed under the assumption Polly is unable to detect the
scop. However, disabling Polly's profitability heuristics is sufficient to
detect this scop.
llvm-svn: 249414
A statement with an empty domain complicates the invariant load
hoisting and does not help any subsequent analysis or transformation.
In fact it might introduce parameter dimensions or increase the
schedule dimensionality. To this end, we remove statements with an
empty domain early in the SCoP simplification.
llvm-svn: 249276
We have to skip accesses in non-affine subregions during hoisting as
they might not be executed under the same condition as the entry of
the non-affine subregion.
llvm-svn: 249139
This moves the construction of ScopStmt to the beginning of the
ScopInfo pass. The late creation was a result of the earlier separation
of ScopInfo and TempScopInfo. This will avoid introducing more
ScopStmt-like maps in future commits. The AccFuncMap will also be
removed in some future commit. DomainMap might also be included into
ScopStmt.
The order in which ScopStmt are created changes and initially creates
empty statements that are removed in a simplification.
Differential Revision: http://reviews.llvm.org/D13341
llvm-svn: 249132
When error blocks are not terminated by an unreachable they have successors
that might only be reachable via error blocks. Additionally, branches in
error blocks are not checked during SCoP detection, thus we might not be able
to handle them. With this patch we do not try to model error block exit
conditions. Anything that is only reachable via error blocks is ignored too,
as it will not be executed in the optimized version of the SCoP anyway.
llvm-svn: 249099
The user can provide function names with
-polly-error-functions=name1,name2,name3
that will be treated as error functions. Any call to them is assumed
not to be executed.
This feature is mainly for developers to play around with the new
"error block" feature.
llvm-svn: 249098
As a first step in the direction of assumed invariant loads (loads
that are not written in some context) we now detect and hoist
definitively invariant loads. These invariant loads will be preloaded
in the code generation and used in the optimized version of the SCoP.
If the load is only conditionally executed the preloaded version will
also only be executed under the same condition, hence we will never
access memory that wouldn't have been accessed otherwise. This is also
the most distinguishing feature to licm.
As hoisting can make statements empty we will simplify the SCoP and
remove empty statements that would otherwise cause artifacts in the
code generation.
Differential Revision: http://reviews.llvm.org/D13194
llvm-svn: 248861
This patch allows switch instructions with affine conditions in the
SCoP. Also switch instructions in non-affine subregions are allowed.
Both did not require much changes to the code, though there was some
refactoring needed to integrate them without code duplication.
In the llvm-test suite the number of profitable SCoPs increased from
135 to 139 but more importantly we can handle more benchmarks and user
inputs without preprocessing.
Differential Revision: http://reviews.llvm.org/D13200
llvm-svn: 248701
The new domain construction algorithm now correctly models this test case (and
derives an empty run-time condition). Add this test case to ensure we do not
regress.
llvm-svn: 248669
When the whole SCoP is a non-affine region we need to use the
surrounding loop in the construction of the schedule as that is
the one that will be looked up after the schedule generation.
This fixes bug 24947
llvm-svn: 248667
When recovering multi-dimensional memory accesses, it may happen that different
accesses to the same base array are recovered with different dimensionality.
This patch ensures that the dimensionalities are unified by adding zero valued
dimensions to acesses with lower dimensionality. When starting to model
fixed-size arrays as multi-dimensional in 247906, this has not been taken
care of.
llvm-svn: 248662
This change addresses three issues:
- Read only scalars that enter a PHI node through an edge that comes from
outside the scop are not modeled any more, as such PHI nodes will always
be initialized to this initial value right before the SCoP is entered.
- For PHI nodes that depend on a scalar value that is defined outside the
scop, but where the scalar values is passed through an edge that itself
comes from a BB that is part of the region, we introduce in this basic
block a read of the out-of-scop value to ensure it's value is available
to write it into the PHI alloc location.
- Read only uses of scalars by PHI nodes are ignored in the general read only
handling code, as they are taken care of by the general PHI node modeling
code.
llvm-svn: 248535
After the merge of TempScopInfo into ScopInfo the analysis output
remained because of the existing unit tests. These remains are removed
and the units tests converted to match the equivalent output of
ScopInfo's analysis output. The unit tests are also moved into the
directory of ScopInfo tests.
Differential Revision: http://reviews.llvm.org/D13116
llvm-svn: 248485
A missing return statement that previously did not have a visibly negative
effect caused after some data-structure changes in r248024 multi-dimensional
accesses to be modeled both multi-dimensional as well as linearized. This
commit adds the missing return to avoid the incorrect double modeling as
well as the compile time increases it caused.
llvm-svn: 248171
If we encounter a <nsw> tagged AddRec for a loop we know the trip count of
that loop has to be bounded or the semantics is undefined anyway. Hence, we
only need to add unbounded assumptions if no such AddRec is known.
llvm-svn: 248128
So far we ignored the unbounded parts of the iteration domain, however
we need to assume they do not occure at all to remain sound if they do.
llvm-svn: 248126
We now add loop carried information during the second traversal of the
region instead of in a intermediate step in-between. This makes the
generation simpler, removes code and should even be faster.
llvm-svn: 248125
In order to allow multiple back edges we:
- compute the conditions under which each back edge is taken
- build the union over all these conditions, thus the condition that
any back edge is taken
- apply the same logic to the union we applied to a single back edge
llvm-svn: 248120
All MemoryAccess objects will be owned by ScopInfo::AccFuncMap which
previously stored the IRAccess objects. Instead of creating new
MemoryAccess objects, the already created ones are reused, but their
order might be different now. Some fields of IRAccess and MemoryAccess
had the same meaning and are merged.
This is the last step of fusioning TempScopInfo.{h|cpp} and
ScopInfo.{h.cpp}. Some refactoring might still make sense.
Differential Revision: http://reviews.llvm.org/D12843
llvm-svn: 248024
If the GEP instructions give us enough insights, model scalar accesses as
multi-dimensional (and generate the relevant run-time checks to ensure
correctness). This will allow us to simplify the dependence computation in
a subsequent commit.
llvm-svn: 247906
This will allow to generate non-wrap assumptions for integer expressions
that are part of the SCoP. We compare the common isl representation of
the expression with one computed with modulo semantic. For all parameter
combinations they are not equal we can have integer overflows.
The nsw flags are respected when the modulo representation is computed,
nuw and nw flags are ignored for now.
In order to not increase compile time to much, the non-wrap assumptions
are collected in a separate boundary context instead of the assumed
context. This helps compile time as the boundary context can become
complex and it is therefor not advised to use it in other operations
except runtime check generation. However, the assumed context is e.g.,
used to tighten dependences. While the boundary context might help to
tighten the assumed context it is doubtful that it will help in practice
(it does not effect lnt much) as the boundary (or no-wrap assumptions)
only restrict the very end of the possible value range of parameters.
PET uses a different approach to compute the no-wrap context, though lnt runs
have shown that this version performs slightly better for us.
llvm-svn: 247732
At some point we build loop trip counts using this method. It was replaced by
a simpler trick that works only for affine (e.g., not modulo) constraints and
relies on the removal of unbounded parts. In order to allow modulo constrains
again we go back to the former, more accurate method.
llvm-svn: 247540
Hoist runtime checks in the loop nest if they guard an "error" like event.
Such events are recognized as blocks with an unreachable terminator or a call
to the ubsan function that deals with out of bound accesses. Other "error"
events can be added easily.
We will ignore these blocks when we detect/model/optmize and code generate SCoPs
but we will make sure that they would not have been executed using the assumption
framework.
llvm-svn: 247310
As we do not rely on ScalarEvolution any more we do not need to get
the backedge taken count. Additionally, our domain generation handles
everything that is affine and has one latch and our ScopDetection will
over-approximate everything else.
This change will therefor allow loops with:
- one latch
- exiting conditions that are affine
Additionally, it will not check for structured control flow anymore.
Hence, loops and conditionals are not necessarily single entry single
exit regions any more.
Differential Version: http://reviews.llvm.org/D12758
llvm-svn: 247289
The TempScopInfo (-polly-analyze-ir) pass is removed and its work taken
over by ScopInfo (-polly-scops). Several tests depend on
-polly-analyze-ir and use -polly-scops instead which for the moment
prints the output of both passes. This again is not expected by some
other tests, especially those with negative searches, which have been
adapted.
Differential Version: http://reviews.llvm.org/D12694
llvm-svn: 247288
This patch replaces the last legacy part of the domain generation, namely the
ScalarEvolution part that was used to obtain loop bounds. We now iterate over
the loops in the region and propagate the back edge condition to the header
blocks. Afterwards we propagate the new information once through the whole
region. In this process we simply ignore unbounded parts of the domain and
thereby assume the absence of infinite loops.
+ This patch already identified a couple of broken unit tests we had for
years.
+ We allow more loops already and the step to multiple exit and multiple back
edges is minimal.
+ It allows to model the overflow checks properly as we actually visit
every block in the SCoP and know where which condition is evaluated.
- It is currently not compatible with modulo constraints in the
domain.
Differential Revision: http://reviews.llvm.org/D12499
llvm-svn: 247279
The support for modulo expressions is not comlete and makes the new
domain generation harder. As the currently broken domain generation
needs to be replaced, we will first swap in the new, fixed domain
generation and make it compatible with the modulo expressions later.
llvm-svn: 247278
The support for pointer expressions is broken as it can only handle
some patterns in the IslExprBuilder. We should to treat pointers in
expressions the same as integers at some point and revert this patch.
llvm-svn: 247147
While we do not need to model PHI nodes in the region exit (as it is not part
of the SCoP), we need to prepare for the case that the exit block is split in
code generation to create a single exiting block. If this will happen, hence
if the region did not have a single exiting block before, we will model the
operands of the PHI nodes as escaping scalars in the SCoP.
Differential Revision: http://reviews.llvm.org/D12051
llvm-svn: 247078
Instead of having two separate options
-polly-detect-scops-in-functions-without-loops and
-polly-detect-scops-in-regions-without-loops we now just use
-polly-detect-unprofitable to force the detection of scops ignoring any compile
time saving bailout heuristics.
llvm-svn: 247057
This case probably does not happen for LLVM generated code that is in loop
simplify form, but Polly does support such kind of loops. This commit ensures we
have test coverage as well.
llvm-svn: 246543
Code generation currently does not expect unbounded loops. When
using ISL to compute the loop trip count, if we find that the
iteration domain remains unbounded, we invalidate the Scop by
creating an infeasible context.
Contributed-by: Matthew Simpson <mssimpso@codeaurora.org>
This fixes PR24634.
Differential Revision: http://reviews.llvm.org/D12493
llvm-svn: 246477
While ignoring read-only scalar dependences it was not necessary to consider
store instructins, but as store instructions can be the target of a scalar
read-only dependency we need to consider them for the construction of scalar
read-only dependences.
llvm-svn: 246429
In order to compute domain conditions for conditionals we will now
traverse the region in the ScopInfo once and build the domains for
each block in the region. The SCoP statements can then use these
constraints when they build their domain.
The reason behind this change is twofold:
1) This removes a big chunk of preprocessing logic from the
TempScopInfo, namely the Conditionals we used to build there.
Additionally to moving this logic it is also simplified. Instead
of walking the dominance tree up for each basic block in the
region (as we did before), we now traverse the region only
once in order to collect the domain conditions.
2) This is the first step towards the isl based domain creation.
The second step will traverse the region similar to this step,
however it will propagate back edge conditions. Once both are in
place this conditional handling will allow multiple exit loops
additional logic.
Reviewers: grosser
Differential Revision: http://reviews.llvm.org/D12428
llvm-svn: 246398
If a region does not have more than one loop, we do not identify it as
a Scop in ScopDetection. The main optimizations Polly is currently performing
(tiling, preparation for outer-loop vectorization and loop fusion) are unlikely
to have a positive impact on individual loops. In some cases, Polly's run-time
alias checks or conditional hoisting may still have a positive impact, but those
are mostly enabling transformations which LLVM already performs for individual
loops. As we do not focus on individual loops, we leave them untouched to not
introduce compile time regressions and execution time noise. This results in
good compile time reduction (oourafft: -73.99%, smg2000: -56.25%).
Contributed-by: Pratik Bhatu <cs12b1010@iith.ac.in>
Reviewers: grosser
Differential Revision: http://reviews.llvm.org/D12268
llvm-svn: 246161
Use ISL to compute the loop trip count when scalar evolution is unable to do
so.
Contributed-by: Matthew Simpson <mssimpso@codeaurora.org>
Differential Revision: http://reviews.llvm.org/D9444
llvm-svn: 246142
If nothing is executed we can bail out early. Otherwise we can use the
constraints that ensure at least one statement is executed for
simplification.
llvm-svn: 245585
We will record if a SAI is the base of another SAI or derived from it.
This will allow to reason about indirect base pointers later on and
allows a clearer picture of indirection also in the SCoP dump.
llvm-svn: 245584
This option allows the user to provide additional information about parameter
values as an isl_set. To specify that N has the value 1024, we can provide
the context -polly-context='[N] -> {: N = 1024}'.
llvm-svn: 245175
Besides other changes this version of isl contains a fundamental fix to memory
corruption issues we have seen with imath-32 backed isl_ints.
This update also contains a fix that ensures that the schedule-tree based
version of isl's dependence analysis takes the domain of the schedule into
account.
llvm-svn: 244585
Even though read-only accesses to scalars outside of a scop do not need to be
modeled to derive valid transformations or to generate valid sequential code,
but information about them is useful when we considering memory footprint
analysis and/or kernel offloading.
llvm-svn: 243981
Summary:
When translating PHI nodes into memory dependences during code generation we
require two kinds of memory. 'Normal memory' as for all scalar dependences and
'PHI node memory' to store the incoming values of the PHI node. With this
patch we now mark and track these two kinds of memories, which we previously
incorrectly marked as a single memory object.
Being aware of PHI node storage makes code generation easier, as we do not need
to guess what kind of storage a scalar reference requires. This simplifies the
code nicely.
Reviewers: jdoerfert
Subscribers: pollydev, llvm-commits
Differential Revision: http://reviews.llvm.org/D11554
llvm-svn: 243420
These test cases check whether Polly still gives the same results if
LICM runs before. Currently, it does not and therefore these cases are
expected fails.
llvm-svn: 243037
Instead of flat schedules, we now use so-called schedule trees to represent the
execution order of the statements in a SCoP. Schedule trees make it a lot easier
to analyze, understand and modify properties of a schedule, as specific nodes
in the tree can be choosen and possibly replaced.
This patch does not yet fully move our DependenceInfo pass to schedule trees,
as some additional performance analysis is needed here. (In general schedule
trees should be faster in compile-time, as the more structured representation
is generally easier to analyze and work with). We also can not yet perform the
reduction analysis on schedule trees.
For more information regarding schedule trees, please see Section 6 of
https://lirias.kuleuven.be/handle/123456789/497238
llvm-svn: 242130
Named isl sets can generally have any name if they remain within Polly, but only
certain strings can be parsed by isl. The new names we create ensure that we
can always copy-past isl strings from Polly to other isl tools, e.g. for
debugging.
llvm-svn: 241787
This is very preliminary support, but it seems to work for the most common case.
When observing more/different test cases, we can work on generalizing this.
llvm-svn: 240955
This removes old code that has been disabled since several weeks and was hidden
behind the flags -disable-polly-intra-scop-scalar-to-array=false and
-polly-model-phi-nodes=false. Earlier, Polly used to translate scalars and
PHI nodes to single element arrays, as this avoided the need for their special
handling in Polly. With Johannes' patches adding native support for such scalar
references to Polly, this code is not needed any more. After this commit both
-polly-prepare and -polly-independent are now mostly no-ops. Only a couple of
simple transformations still remain, but they are scheduled for removal too.
Thanks again to Johannes Doerfert for his nice work in making all this code
obsolete.
llvm-svn: 240766
Remainder operations with constant divisor can be modeled as quasi-affine
expression. This patch adds support for detecting and modeling them. We also
add a test that ensures they are correctly code generated.
This patch was extracted from a larger patch contributed by Johannes Doerfert
in http://reviews.llvm.org/D5293
llvm-svn: 240518
While looking through the test cases I realized we did not have a CHECK line
for a duplicate memory access which we may want to eliminate later. To ensure
we do not have (or later introduce) unnecessary memory accesses, we now tighten
the test cases to look for such a pattern (and add the CHECK: line that shows
the redundant memory access).
llvm-svn: 238227
This ensures we pass all tests independently of how we set the options
-disable-polly-intra-scop-scalar-to-array and -polly-model-phi-nodes.
(At least if we enable both or disable both. Enabling them individually makes
little sense, as they will hopefully disappear soon anyhow).
llvm-svn: 238087
Being here, we extend the interface to return the element type and not a pointer
to the element type. We also provide a function to get the size (in bytes) of
the elements stored in this array.
We currently still store the element size as an innermost dimension in
ScopArrayInfo, which is somehow inconsistent and should be addressed in future
patches.
llvm-svn: 237779
I just learned that target triples prevent test cases to be run on other
architectures. Polly test cases are until now sufficiently target independent
to not require any target triples. Hence, we drop them.
llvm-svn: 235384
In Polly we used both the term 'scattering' and the term 'schedule' to describe
the execution order of a statement without actually distinguishing between them.
We now uniformly use the term 'schedule' for the execution order. This
corresponds to the terminology of isl.
History: CLooG introduced the term scattering as the generated code can be used
as a sequential execution order (schedule) or as a parallel dimension
enumerating different threads of execution (placement). In Polly and/or isl the
term placement was never used, but we uniformly refer to an execution order as a
schedule and only later introduce parallelism. When doing so we do not talk
about about specific placement dimensions.
llvm-svn: 235380
This will allow the ScopInfo to build the polyhedral representation for
non-affine regions that contain loops. Such loops are basically not visible
in the SCoP representation. Accesses that are variant in such loops are
therefor represented as non-affine accesses.
Differential Revision: http://reviews.llvm.org/D8153
llvm-svn: 234713
This allows us to delinerize code such as:
A[][n]
for (i
for (j
A[i][n-j-1] = ...
which would previously have been delinearize to an access A[i+1][-j-1].
To recover the correct access we apply the piecewise expression:
{ A[i][j] -> A[i-1][i+N]: i < 0; A[i][j] -> A[i][i]: i >= 0}
This approach generalizes to higher dimensions.
llvm-svn: 233566
This will strip the constant factor of a parameter befor we add it to
the SCoP. As a result the access functions are simplified, e.g., for
the attached test case.
llvm-svn: 233501
This test case was supposed to test the range analysis but it became just
another delinearization test case after enabling delinearization.
Suggested-by: Johannes Doerfert
llvm-svn: 231599
If a scalar was defined and used only in a non-affine subregion we do
not need to model the accesses. However, if the scalar was defined
inside the region and escapes the region we have to model the access.
The same is true if the scalar was defined outside and used inside the
region.
llvm-svn: 230960
This allows us to model non-affine regions in the SCoP representation.
SCoP statements can now describe either basic blocks or non-affine
regions. In the latter case all accesses in the region are accumulated
for the statement and write accesses, except in the entry, have to be
marked as may-write.
Differential Revision: http://reviews.llvm.org/D7846
llvm-svn: 230329
Scops that only read seem generally uninteresting and scops that only write are
most likely initializations where there is also little to optimize. To not
waste compile time we bail early.
Differential Revision: http://reviews.llvm.org/D7735
llvm-svn: 229820
These write are important as they will force the scheduling and code
generation of an otherwise trivial statement and also impose an order of
execution needed to guarantee the correct final value for a scalar in a loop.
Added test case modeled after ClamAV/clamscan.
llvm-svn: 228847
This allows us to model PHI nodes in the polyhedral description
without demoting them. The modeling however will result in the
same accesses as the demotion would have introduced.
Differential Revision: http://reviews.llvm.org/D7415
llvm-svn: 228433
The max loop depth was incorrectly computed for scops that contain a
block from a loop but do not contain the entire loop. We need to
check that the full loop is contained in the region when computing
the max loop depth.
These scops occur when a region containing an inner loop is expanded
to include some blocks from the outer loop, but it cannot be fully
expanded to contain the outer loop because the region containing the
outer loop is invalid.
Differential Revision: http://reviews.llvm.org/D6913
llvm-svn: 225812
This support is still incomplete and consequently hidden behind a switch that
needs to be enabled. One problem is ATM that we incorrectly interpret very large
unsigned values as negative values even if used in an unsigned comparision.
llvm-svn: 225480
Schedule dimensions that have the same constant value accross all statements do
not carry any information, but due to the increased dimensionality of the
schedule cost compile time. To not pay this cost, we remove constant dimensions
if possible.
llvm-svn: 225067
Isl now specifically marks modulo operations that are compared against zero.
They can be implemented with the C/LLVM remainder operation.
We also update a couple of test cases where the output of isl has slightly
changed.
llvm-svn: 223607
SCEV based code generation has been the default for two weeks after having
been tested for a long time. We now drop the support the non-scev-based code
generation.
llvm-svn: 222978
In case a GEP instruction references into a fixed size array e.g., an access
A[i][j] into an array A[100x100], LLVM-IR does not guarantee that the subscripts
always compute values that are within array bounds. We now derive the set of
parameter values for which all accesses are within bounds and add the assumption
that the scop is only every executed with this set of parameter values.
Example:
void foo(float A[][20], long n, long m {
for (long i = 0; i < n; i++)
for (long j = 0; j < m; j++)
A[i][j] = ...
This loop yields out-of-bound accesses if m is at least 20 and at the same time
at least one iteration of the outer loop is executed. Hence, we assume:
n <= 0 or m <= 20.
Doing so simplifies the dependence analysis problem, allows us to perform
more optimizations and generate better code.
TODO: The location where the GEP instruction is executed is not necessarily the
location where the memory is actually accessed. As a result scanning for GEP[s]
is imprecise. Even though this is not a correctness problem, this imprecision
may result in missed optimizations or non-optimal run-time checks.
In polybench where this mismatch between parametric loop bounds and fixed size
arrays is common, we see with this patch significant reductions in compile time
(up to 50%) and execution time (up to 70%). We see two significant compile time
regressions (fdtd-2d, jacobi-2d-imper), and one execution time regression
(trmm). Both regressions arise due to additional optimizations that have been
enabled by this patch. They can be addressed in subsequent commits.
http://reviews.llvm.org/D6369
llvm-svn: 222754
This patch changes the RegionSet type used in ScopDetection from a
std::set to a llvm::SetVector. The reason for the change is to
ensure deterministic output when printing the result of the
analysis. We had a windows buildbot failure for the modified test
because the output was coming in a different order.
Only one test case needed to be modified for this change. We could
use CHECK-DAG directives instead of CHECK in the analysis test cases
because the actual order of scops does not matter, but I think that
change should be done in a separate patch that modifies all the
appliciable tests. I simply modified the test to reflect the
expected deterministic output.
Differential Revision: http://reviews.llvm.org/D5897
llvm-svn: 220423
We use a parametric abstraction of the domain to split alias groups
if accesses cannot be executed under the same parameter evaluation.
The two test cases check that we can remove alias groups if the
pointers which might alias are never accessed under the same parameter
evaluation and that the minimal/maximal accesses are not global but
with regards to the parameter evaluation.
Differential Revision: http://reviews.llvm.org/D5436
llvm-svn: 218758
If there are multiple read only base addresses in an alias group
we can split it into multiple alias groups each with only one
read only access. This way we might reduce the number of
comparisons significantly as it grows linear in the number of
alias groups but exponential in their size.
Differential Revision: http://reviews.llvm.org/D5435
llvm-svn: 218757
If too many parameters are involved in accesses used to create RTCs
we might end up with enormous compile times and RTC expressions.
The reason is that the lexmin/lexmax is dependent on all these
parameters and isl might need to create a case for every "ordering"
of them (e.g., p0 <= p1 <= p2, p1 <= p0 <= p2, ...).
The exact number of parameters allowed in accesses is defined by the
command line option -polly-rtc-max-parameters=XXX and set by default
to 8.
Differential Revision: http://reviews.llvm.org/D5500
llvm-svn: 218566
This change will build all alias groups (minimal/maximal accesses
to possible aliasing base pointers) we have to check before
we can assume an alias free environment. It will also use these
to create Runtime Alias Checks (RTC) in the ISL code generation
backend, thus allow us to optimize SCoPs despite possibly aliasing
pointers when this backend is used.
This feature will be enabled for the isl code generator, e.g.,
--polly-code-generator=isl, but disabled for:
- The cloog code generator (still the default).
- The case delinearization is enabled.
- The case non-affine accesses are allowed.
llvm-svn: 218046
Even though we previously correctly detected the multi-dimensional access
pattern for accesses with a certain base address, we only delinearized
non-affine accesses to this address. Affine accesses have not been touched and
remained as single dimensional accesses. The result was an inconsistent
description of accesses to the same array, with some being one dimensional and
some being multi-dimensional.
This patch ensures that all accesses are delinearized with the same
dimensionality as soon as a single one of them has been detected as non-affine.
While writing this patch, it became evident that the options
-polly-allow-nonaffine and -polly-detect-keep-going have not been properly
supported in case delinearization has been turned on. This patch adds relevant
test coverage and addresses these issues as well. We also added some more
documentation to the functions that are modified in this patch.
This fixes llvm.org/PR20123
Differential Revision: http://reviews.llvm.org/D5329
llvm-svn: 217728
We now verify that such functions are correctly detected even in combination
with delinearization. This change is added to ensure we have good test coverage
for the subsequent delinearization fix.
We also remove unnecessary instructions from the test case.
llvm-svn: 217664
In Polly we used to have a mix of test cases, some that used 'opt %s' and others
that used 'opt < %s'. We now change all to use 'opt < %s'. Piping in test files
is preferable as it does prevent temporary files to be written to disk. This
brings us in line with what is usus in LLVM.
llvm-svn: 216816
This reverts commit 215684. The intention of the commit is great, but
unfortunately it seems to be the cause of 14 LNT test suite failures:
http://lab.llvm.org:8011/builders/perf-x86_64-penryn-O3-polly/builds/116
To make our buildbots and performance testers green until this issue is solved,
we temporarily revert this commit.
llvm-svn: 215816
The support is limited to signed modulo access and condition
expressions with a constant right hand side, e.g., A[i % 2] or
A[i % 9]. Test cases are modified according to this new feature and
new test cases are added.
Differential Revision: http://reviews.llvm.org/D4843
llvm-svn: 215684
There is no needed for neither 1-dimensional nor higher dimensional arrays to
require positive offsets in the outermost array dimension.
We originally introduced this assumption with the support for delinearizing
multi-dimensional arrays.
llvm-svn: 214665
As our delinearization works optimistically, we need in some cases run-time
checks that verify our optimistic assumptions. A simple example is the
following code:
void foo(long n, long m, long o, double A[n][m][o]) {
for (long i = 0; i < 100; i++)
for (long j = 0; j < 150; j++)
for (long k = 0; k < 200; k++)
A[i][j][k] = 1.0;
}
After clang linearized the access to A and we delinearized it again to
A[i][j][k] we need to ensure that we do not access the delinearized array
out of bounds (this information is not available in LLVM-IR). Hence, we
need to verify the following constraints at run-time:
CHECK: Assumed Context:
CHECK: [o, m] -> { : m >= 150 and o >= 200 }
llvm-svn: 212198
This change is particularly useful in the code generation as we need
to know which binary operator/identity element we need to combine/initialize
the privatization locations.
+ Print the reduction type for each memory access
+ Adjusted the test cases to comply with the new output format and
to test for the right reduction type
llvm-svn: 212126
Iterate over all store memory accesses and check for valid binary reduction
candidate loads by following the operands of the stored value. For each
candidate pair we check if they have the same base address and there are no
other accesses which may overlap with them. This ensures that no intermediate
value can escape into other memory locations or is overwritten at some point.
+ 17 test cases for reduction detection and reduction dependency modeling
llvm-svn: 211957
Due to bad habit we sometimes used a variable %defaultOpts that listed
a set of passes commonly run to prepare for Polly. None of these test cases
actually needs special preparation and only two of them need the 'basicaa' to
be scheduled. Scheduling the required alias analysis explicitly makes the test
cases clearer.
llvm-svn: 211671
We had a set of test cases that have been incomplete and XFAILED. This patch
completes a couple of the interesting ones and removes the ones which seem
redundant or not sufficiently reduced to be useful.
llvm-svn: 211670
We do this currently only for test cases where we have integer offsets that
clearly access array dimensions out-of-bound.
-; for (long i = 0; i < n; i++)
-; for (long j = 0; j < m; j++)
-; for (long k = 0; k < o; k++)
+; for (long i = 0; i < n - 3; i++)
+; for (long j = 4; j < m; j++)
+; for (long k = 0; k < o - 7; k++)
; A[i+3][j-4][k+7] = 1.0;
This will be helpful if we later want to simplify the access functions under the
assumption that they do not access memory out of bounds.
llvm-svn: 210179
Without this patch, the testcase would fail on the delinearization of the second
array:
; void foo(long n, long m, long o, double A[n][m][o]) {
; for (long i = 0; i < n; i++)
; for (long j = 0; j < m; j++)
; for (long k = 0; k < o; k++) {
; A[i+3][j-4][k+7] = 1.0;
; A[i][0][k] = 2.0;
; }
; }
; CHECK: [n, m, o] -> { Stmt_for_body6[i0, i1, i2] -> MemRef_A[3 + i0, -4 + i1, 7 + i2] };
; CHECK: [n, m, o] -> { Stmt_for_body6[i0, i1, i2] -> MemRef_A[i0, 0, i2] };
Here is the output of FileCheck on the testcase without this patch:
; CHECK: [n, m, o] -> { Stmt_for_body6[i0, i1, i2] -> MemRef_A[i0, 0, i2] };
^
<stdin>:26:2: note: possible intended match here
[n, m, o] -> { Stmt_for_body6[i0, i1, i2] -> MemRef_A[o0] };
^
It is possible to find a good delinearization for A[i][0][k] only in the context
of the delinearization of both array accesses.
There are two ways to delinearize together all array subscripts touching the
same base address: either duplicate the code from scop detection to first gather
all array references and then run the delinearization; or as implemented in this
patch, use the same delinearization info that we computed during scop detection.
llvm-svn: 210117
We do not have a use for this information at the moment. If we need this at some
point, the "instruction -> access" mapping needs to be enhanced as a single
instruction could then possibly perform multiple accesses.
This patch allows us to build the polyhedral information for scops with scalar
dependences.
llvm-svn: 201815
In rare cases the modification of one scop can effect the validity of other
scops, as code generation of an earlier scop may make the scalar evolution
functions derived for later scops less precise. The example that triggered this
patch was a scop that contained an 'or' expression as follows:
%add13710 = or i32 %j.19, 1
--> {(1 + (4 * %l)),+,2}<nsw><%for.body81>
Scev could only analyze the 'or' as it knew %j.19 is a multiple of 2. This
information was not available after the first scop was code generated (or
independent-blocks was run on it) and SCEV could not derive a precise SCEV
expression any more. This means we could not any more code generate this SCoP.
My current understanding is that there is always the risk that an earlier code
generation change invalidates later scops. As the example we have seen here is
difficult to avoid, we use this occasion to guard us against all such
invalidations.
This patch "solves" this issue by verifying right before we start working on
a detected scop, if this scop is in fact still valid. This adds a certain
overhead. However the verification we run is anyways very fast and secondly
it is only run on detected scops. So the overhead should not be very large. As
a later optimization we could detect scops only on demand, such that we need
to run scop-detections always only a single time.
This should fix the single last failure in the LLVM test-suite for the new
scev-based code generation.
llvm-svn: 201593
SCoP invariant parameters with the different start value would deter parameter
sharing. For example, when compiling the following C code:
void foo(float *input) {
for (long j = 0; j < 8; j++) {
// SCoP begin
for (long i = 0; i < 8; i++) {
float x = input[j * 64 + i + 1];
input[j * 64 + i] = x * x;
}
}
}
Polly would creat two parameters for these memory accesses:
p_0: {0,+,256}
p_2: {4,+,256}
[j * 64 + i + 1] => MemRef_input[o0] : 4o0 = p_1 + 4i0
[j * 64 + i] => MemRef_input[o0] : 4o0 = p_0 + 4i0
These parameters only differ from start value. To enable parameter sharing,
we split the start value from SCEVAddRecExpr, so they would share a single
parameter that always has zero start value:
p0: {0,+,256}<%for.cond1.preheader>
[j * 64 + i + 1] => MemRef_input[o0] : 4o0 = 4 + p_1 + 4i0
[j * 64 + i] => MemRef_input[o0] : 4o0 = p_0 + 4i0
Such translation can make the polly-dependence much faster.
Contributed-by: Star Tan <tanmx_star@yeah.net>
llvm-svn: 187728
Previously this happend to work for integers up to i64, but we got it wrong
for larger numbers. Fix this and add test cases to verify this keeps working.
Reported by: Sven Verdoolaege <skimo at kotnet dot org>
llvm-svn: 183986
We do not only need to understand that 'k * p' is a parameter expression, but
also need to store this expression in the set of parameters. Before this patch
we wrongly stored the two individual parameters %k and %p.
Reported by: Sebastian Pop <spop@codeaurora.org>
llvm-svn: 179485
In my previous commits I failed to realise that my new requires lines fully
disabled these tests. We now properly check if we are in an asserts build and
only disable the tests if assertions are not available.
Reported-by: Sean Silva <silvas@purdue.edu>
llvm-svn: 176900
Johannes Doerfert