updated instructions:
pmulld, pmullw, pmulhw, mulsd, mulps, mulpd, divss, divps, divsd, divpd, addpd and subpd.
special optimization case which replaces pmulld with pmullw\pmulhw\pshuf seq.
In case if the real operands bitwidth <= 16.
Differential Revision: https://reviews.llvm.org/D28104
llvm-svn: 291657
The original code considered only v2i64 as slow for this feature. This patch
consider all 128-bit long vector types as slow candidates.
In internal tests, extending this feature to all 128-bit vector types
resulted in an overall improvement of 1% on Exynos M1.
Differential revision: https://reviews.llvm.org/D27998
llvm-svn: 291616
invalid.
This fixes use-after-free bugs that will arise with any interesting use
of SCEV.
I've added a dedicated test that works diligently to trigger these kinds
of bugs in the new pass manager and also checks for them explicitly as
well as triggering ASan failures when things go squirly.
llvm-svn: 291426
The 'fast' costs should only work for shifts by uniform constants (uniform non-constant are lowered using the slow default implementation).
Logical shifts were not taking into account that we must mask the psrlw result, so the costs needed to be doubled.
Added missing AVX2/AVX512BW costs as well.
llvm-svn: 291391
This test case has been reduced from test/Analysis/RegionInfo/mix_1.ll and
provides us with a minimal example of a test case which caused problems while
working on an improved version of the RegionInfo analysis. We upstream this
test case, as it certainly can be helpful in future debugging and optimization
tests.
Test case reduced by Pratik Bhatu <cs12b1010@iith.ac.in>
llvm-svn: 290974
X86 target does not provide any target specific cost calculation for interleave patterns.It uses the common target-independent calculation, which gives very high numbers. As a result, the scalar version is chosen in many cases. The situation on AVX-512 is even worse, since we have 3-src shuffles that significantly reduce the cost.
In this patch I calculate the cost on AVX-512. It will allow to compare interleave pattern with gather/scatter and choose a better solution (PR31426).
* Shiffle-broadcast cost will be changed in Simon's upcoming patch.
Differential Revision: https://reviews.llvm.org/D28118
llvm-svn: 290810
This fixes the issue exposed in PR31393, where we weren't trying
sufficiently hard to diagnose bad TBAA metadata.
This does reduce the variety in the error messages we print out, but I
think the tradeoff of verifying more, simply and quickly overrules the
need for more helpful error messags here.
llvm-svn: 290713
BasicAA in r290603.
I've kept the basic testing in the new PM test file as that also covers
the AAManager invalidation logic. If/when there is a good place for
broader AA testing it could move there.
This test is somewhat unsatisfying as I can't get it to fail even with
ASan outside of explicit checks of the invalidation. Apparently we don't
yet have any test coverage of the BasicAA code paths using either the
domtree or loopinfo -- I made both of them always be null and check-llvm
passed.
llvm-svn: 290612
For vector GEPs, CastGEPIndices can end up in an infinite recursion, because
we compare the vector type to the scalar pointer type, find them different,
and then try to cast a type to itself.
Differential Revision: https://reviews.llvm.org/D28009
llvm-svn: 290260
This patch checks that the SlowMisaligned128Store subtarget feature is set
when penalizing such stores in getMemoryOpCost.
Differential Revision: https://reviews.llvm.org/D27677
llvm-svn: 289845
After r289755, the AssumptionCache is no longer needed. Variables affected by
assumptions are now found by using the new operand-bundle-based scheme. This
new scheme is more computationally efficient, and also we need much less
code...
llvm-svn: 289756
There was an efficiency problem with how we processed @llvm.assume in
ValueTracking (and other places). The AssumptionCache tracked all of the
assumptions in a given function. In order to find assumptions relevant to
computing known bits, etc. we searched every assumption in the function. For
ValueTracking, that means that we did O(#assumes * #values) work in InstCombine
and other passes (with a constant factor that can be quite large because we'd
repeat this search at every level of recursion of the analysis).
Several of us discussed this situation at the last developers' meeting, and
this implements the discussed solution: Make the values that an assume might
affect operands of the assume itself. To avoid exposing this detail to
frontends and passes that need not worry about it, I've used the new
operand-bundle feature to add these extra call "operands" in a way that does
not affect the intrinsic's signature. I think this solution is relatively
clean. InstCombine adds these extra operands based on what ValueTracking, LVI,
etc. will need and then those passes need only search the users of the values
under consideration. This should fix the computational-complexity problem.
At this point, no passes depend on the AssumptionCache, and so I'll remove
that as a follow-up change.
Differential Revision: https://reviews.llvm.org/D27259
llvm-svn: 289755
Summary:
This change adds some verification in the IR verifier around struct path
TBAA metadata.
Other than some basic sanity checks (e.g. we get constant integers where
we expect constant integers), this checks:
- That by the time an struct access tuple `(base-type, offset)` is
"reduced" to a scalar base type, the offset is `0`. For instance, in
C++ you can't start from, say `("struct-a", 16)`, and end up with
`("int", 4)` -- by the time the base type is `"int"`, the offset
better be zero. In particular, a variant of this invariant is needed
for `llvm::getMostGenericTBAA` to be correct.
- That there are no cycles in a struct path.
- That struct type nodes have their offsets listed in an ascending
order.
- That when generating the struct access path, you eventually reach the
access type listed in the tbaa tag node.
Reviewers: dexonsmith, chandlerc, reames, mehdi_amini, manmanren
Subscribers: mcrosier, llvm-commits
Differential Revision: https://reviews.llvm.org/D26438
llvm-svn: 289402
ConstantFolding tried to cast one of the scalar indices to a vector
type. Instead, use the vector type only for the first index (which
is the only one allowed to be a vector) and use its scalar type
otherwise.
Fixes PR31250.
Reviewers: majnemer
Differential Revision: https://reviews.llvm.org/D27389
llvm-svn: 289073
VSX has instructions lxsiwax/lxsdx that can load 32/64 bit value into VSX register cheaply. That patch makes it known to memory cost model, so the vectorization of the test case in pr30990 is beneficial.
Differential Revision: https://reviews.llvm.org/D26713
llvm-svn: 288560
Currently when cost of scalar operations is evaluated the vector type is
used for scalar operations. Patch fixes this issue and fixes evaluation
of the vector operations cost.
Several test showed that vector cost model is too optimistic. It
allowed vectorization of 8 or less add/fadd operations, though scalar
code is faster. Actually, only for 16 or more operations vector code
provides better performance.
Differential Revision: https://reviews.llvm.org/D26277
llvm-svn: 288398
Note that the non-splat lshr+lshr test folded, but that does not
work in general. Something is missing or wrong in computeKnownBits
as the non-splat shl+shl test still shows.
llvm-svn: 288005
Currently LLVM assumes that a pointer addrspacecasted to a different addr space is equivalent to trunc or zext bitwise, which is not true. For example, in amdgcn target, when a null pointer is addrspacecasted from addr space 4 to 0, its value is changed from i64 0 to i32 -1.
This patch teaches LLVM not to assume known bits of addrspacecast instruction to its operand.
Differential Revision: https://reviews.llvm.org/D26803
llvm-svn: 287545
Summary:
This extends FCOPYSIGN support to 512-bit vectors.
I've also added tests to show what the 128-bit and 256-bit cases look like with broadcast loads.
Reviewers: delena, zvi, RKSimon, spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26791
llvm-svn: 287298
Add explicit v16i16/v32i8 ADD/SUB costs, matching the costs of v4i64/v8i32 - they were missing for some reason.
This has side effects on the LV max bandwidth tests (AVX1 now prefers 128-bit vectors vs AVX2 which still prefers 256-bit)
llvm-svn: 286832
If the inrange keyword is present before any index, loading from or
storing to any pointer derived from the getelementptr has undefined
behavior if the load or store would access memory outside of the bounds of
the element selected by the index marked as inrange.
This can be used, e.g. for alias analysis or to split globals at element
boundaries where beneficial.
As previously proposed on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2016-July/102472.html
Differential Revision: https://reviews.llvm.org/D22793
llvm-svn: 286514
These examples are variations that were inspired from a small subgraph taken
from paper.ll which are interesting as they show certain issues with infinite
loops.
llvm-svn: 286450
This patch avoids scalarization of CTLZ by instead expanding to use CTPOP (ref: "Hacker's Delight") when the necessary operations are available.
This also adds the necessary cost models for X86 SSE2 targets (the main beneficiary) to ensure vectorization only happens when its useful.
Differential Revision: https://reviews.llvm.org/D25910
llvm-svn: 286233
There is a bug describing poor cost model for floating point operations:
Bug 29083 - [X86][SSE] Improve costs for floating point operations. This
patch is the second one in series of patches dealing with cost model.
Differential Revision: https://reviews.llvm.org/D25722
llvm-svn: 285564
Summary:
We were trying to add APInt values with different bit sizes after
visiting an addrspacecast instruction which changed the bit width
of the pointer.
Reviewers: majnemer, hfinkel
Subscribers: hfinkel, wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D24774
llvm-svn: 285407
With DQI but without VLX, lower v2i64 and v4i64 MUL operations with v8i64 MUL (vpmullq).
Updated cost table accordingly.
Differential Revision: https://reviews.llvm.org/D26011
llvm-svn: 285304
This reverts commit r285191.
LICM appears to rely on the Alias Set Tracker hitting lifetime markers to prevent
code from being moved outside of the original scope.
llvm-svn: 285227
There are two fixes here: one, AnalyzeUsesOfPointer can't return
false until it has checked all the uses of the pointer. Two, if a
global uses another global, we have to assume the address of the
first global escapes.
Fixes https://llvm.org/bugs/show_bug.cgi?id=30707 .
Differential Revision: https://reviews.llvm.org/D25798
llvm-svn: 285034
We were defaulting to SSE2 costs which weren't taking into account the availability of PBLENDW/PBLENDVB to improve merging of per-element shift results.
llvm-svn: 284939
In BasicAA GEP operand values get adjusted ("wrap-around") based on the
pointersize. Otherwise, in non-64b modes, AA could report false negatives.
However, a wrap-around is valid only for a fully evaluated expression.
It had been introduced to fix an alias problem in
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160118/326163.html.
This commit restricts the wrap-around to constant gep operands only where the
value is known at compile-time.
llvm-svn: 284908
Summary:
When SCEVRewriteVisitor traverses the SCEV DAG, it may visit the same SCEV
multiple times if this SCEV is referenced by multiple other SCEVs. This has
exponential time complexity in the worst case. Memoizing the results will
avoid re-visiting the same SCEV. Add a map to save the results, and override
the visit function of SCEVVisitor. Now SCEVRewriteVisitor only visit each
SCEV once and thus returns the same result for the same input SCEV.
This patch fixes PR18606, PR18607.
Reviewers: Sanjoy Das, Mehdi Amini, Michael Zolotukhin
Differential Revision: https://reviews.llvm.org/D25810
llvm-svn: 284868
When we have a loop with a known upper bound on the number of iterations, and
furthermore know that either the number of iterations will be either exactly
that upper bound or zero, then we can fully unroll up to that upper bound
keeping only the first loop test to check for the zero iteration case.
Most of the work here is in plumbing this 'max-or-zero' information from the
part of scalar evolution where it's detected through to loop unrolling. I've
also gone for the safe default of 'false' everywhere but howManyLessThans which
could probably be improved.
Differential Revision: https://reviews.llvm.org/D25682
llvm-svn: 284818
This is to avoid inlining too many multiplication operands into a SCEV, which could
take exponential time in the worst case.
Reviewers: Sanjoy Das, Mehdi Amini, Michael Zolotukhin
Differential Revision: https://reviews.llvm.org/D25794
llvm-svn: 284784
We weren't accounting for legal types on every subtarget, meaning that many of the costs were using defaults.
We still don't correctly cost (or test) the 512-bit sdiv/udiv by uniform const cases, nor the power-of-2 cases.
llvm-svn: 284744
In loops that look something like
i = n;
do {
...
} while(i++ < n+k);
where k is a constant, the maximum backedge count is k (in fact the backedge
count will be either 0 or k, depending on whether n+k wraps). More generally
for LHS < RHS if RHS-(LHS of first comparison) is a constant then the loop will
iterate either 0 or that constant number of times.
This allows for more loop unrolling with the recent upper bound loop unrolling
changes, and I'm working on a patch that will let loop unrolling additionally
make use of the loop being executed either 0 or k times (we need to retain the
loop comparison only on the first unrolled iteration).
Differential Revision: https://reviews.llvm.org/D25607
llvm-svn: 284465
As discussed on PR28461 we currently miss the chance to lower "fptosi <2 x double> %arg to <2 x i32>" to cvttpd2dq due to its use of illegal types.
This patch adds support for fptosi to 2i32 from both 2f64 and 2f32.
It also recognises that cvttpd2dq zeroes the upper 64-bits of the xmm result (similar to D23797) - we still don't do this for the cvttpd2dq/cvttps2dq intrinsics - this can be done in a future patch.
Differential Revision: https://reviews.llvm.org/D23808
llvm-svn: 284459
Summary: The delinearization algorithm did not consider terms which had an extension without a multiply factor, i.e. a identify factor. We lose cases where size is char type where there will no multiply factor.
Reviewers: sanjoy, grosser
Subscribers: mzolotukhin, Eugene.Zelenko, llvm-commits, mssimpso, sanjoy, grosser
Differential Revision: https://reviews.llvm.org/D16492
llvm-svn: 284378
Summary:
While walking defs of pointer operands we were assuming that the pointer
size would remain constant. This is not true, because addresspacecast
instructions may cast the pointer to an address space with a different
pointer width.
This partial reverts r282612, which was a more conservative solution
to this problem.
Reviewers: reames, sanjoy, apilipenko
Subscribers: wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D24772
llvm-svn: 283557
Summary:
The computeKnownBits and ComputeNumSignBits functions in ValueTracking can now do a simple look-through of ExtractElement.
Reviewers: majnemer, spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D24955
llvm-svn: 283434
Slightly improves the precision of GlobalsAA in certain situations, and
makes the behavior of optimization passes more predictable.
Differential Revision: https://reviews.llvm.org/D24104
llvm-svn: 283165
This should fix:
https://llvm.org/bugs/show_bug.cgi?id=30433
There are a couple of open questions about the codegen:
1. Should we let scalar ops be scalars and avoid vector constant loads/splats?
2. Should we have a pass to combine constants such as the inverted pair that we have here?
Differential Revision: https://reviews.llvm.org/D25165
llvm-svn: 283119
Summary: When identifying cold blocks, consider only the edge to the normal destination if the terminator is InvokeInst and let calcInvokeHeuristics() decide edge weights for the InvokeInst.
Reviewers: mcrosier, hfinkel, davidxl
Subscribers: mcrosier, llvm-commits
Differential Revision: https://reviews.llvm.org/D24868
llvm-svn: 282262
Enhance SCEV to compute the trip count for some loops with unknown stride.
Patch by Pankaj Chawla
Differential Revision: https://reviews.llvm.org/D22377
llvm-svn: 281732
value is a pointer.
This patch is to fix PR30213. When expanding an expr based on ValueOffsetPair,
if the value is of pointer type, we can only create a getelementptr instead
of sub expr.
Differential Revision: https://reviews.llvm.org/D24088
llvm-svn: 281439
Fixed a bug in run-time checks for possible memory conflicts inside loop.
The bug is in Low <-> High boundaries calculation. The High boundary should be calculated as "last memory access pointer + element size".
Differential revision: https://reviews.llvm.org/D23176
llvm-svn: 279930
when unroll runtime iteration loop.
In llvm::UnrollRuntimeLoopRemainder, if the loop to be unrolled is the inner
loop inside a loop nest, the scalar evolution needs to be dropped for its
parent loop which is done by ScalarEvolution::forgetLoop. However, we can
postpone forgetLoop to the end of UnrollRuntimeLoopRemainder so TripCountSC
expansion can still reuse existing value.
Differential Revision: https://reviews.llvm.org/D23572
llvm-svn: 279748
Repeated inserts into AliasSetTracker have quadratic behavior - inserting a
pointer into AST is linear, since it requires walking over all "may" alias
sets and running an alias check vs. every pointer in the set.
We can avoid this by tracking the total number of pointers in "may" sets,
and when that number exceeds a threshold, declare the tracker "saturated".
This lumps all pointers into a single "may" set that aliases every other
pointer.
(This is a stop-gap solution until we migrate to MemorySSA)
This fixes PR28832.
Differential Revision: https://reviews.llvm.org/D23432
llvm-svn: 279274
The patch is to fix the bug in PR28705. It was caused by setting wrong return
value for SCEVExpander::findExistingExpansion. The return values of findExistingExpansion
have different meanings when the function is used in different ways so it is easy to make
mistake. The fix creates two new interfaces to replace SCEVExpander::findExistingExpansion,
and specifies where each interface is expected to be used.
Differential Revision: https://reviews.llvm.org/D22942
llvm-svn: 278161
The fix for PR28705 will be committed consecutively.
In D12090, the ExprValueMap was added to reuse existing value during SCEV expansion.
However, const folding and sext/zext distribution can make the reuse still difficult.
A simplified case is: suppose we know S1 expands to V1 in ExprValueMap, and
S1 = S2 + C_a
S3 = S2 + C_b
where C_a and C_b are different SCEVConstants. Then we'd like to expand S3 as
V1 - C_a + C_b instead of expanding S2 literally. It is helpful when S2 is a
complex SCEV expr and S2 has no entry in ExprValueMap, which is usually caused
by the fact that S3 is generated from S1 after const folding.
In order to do that, we represent ExprValueMap as a mapping from SCEV to
ValueOffsetPair. We will save both S1->{V1, 0} and S2->{V1, C_a} into the
ExprValueMap when we create SCEV for V1. When S3 is expanded, it will first
expand S2 to V1 - C_a because of S2->{V1, C_a} in the map, then expand S3 to
V1 - C_a + C_b.
Differential Revision: https://reviews.llvm.org/D21313
llvm-svn: 278160
Shifts with a uniform but non-constant count were considered very expensive to
vectorize, because the splat of the uniform count and the shift would tend to
appear in different blocks. That made the splat invisible to ISel, and we'd
scalarize the shift at codegen time.
Since r201655, CodeGenPrepare sinks those splats to be next to their use, and we
are able to select the appropriate vector shifts. This updates the cost model to
to take this into account by making shifts by a uniform cheap again.
Differential Revision: https://reviews.llvm.org/D23049
llvm-svn: 277782
As it turns out, modref queries are broken with CFLAA. Specifically,
the data source we were using for determining modref behaviors
explicitly ignores operations on non-pointer values. So, it wouldn't
note e.g. storing an i32 to an i32* (or loading an i64 from an i64*).
It also ignores external function calls, rather than acting
conservatively for them.
(N.B. These operations, where necessary, *are* tracked by CFLAA; we just
use a different mechanism to do so. Said mechanism is relatively
imprecise, so it's unlikely that we can provide reasonably good modref
answers with it as implemented.)
Patch by Jia Chen.
Differential Revision: https://reviews.llvm.org/D22978
llvm-svn: 277366
Summary:
The motivation is the same as in D22141: In order to add the hotness
attribute to optimization remarks we need BFI to be available in all
passes that emit optimization remarks. BFI depends on BPI so unless we
make this lazy as well we would still compute BPI unconditionally.
The solution is to use the new LazyBPI pass in LazyBFI and only compute
BPI when computation of BFI is requested by the client.
I extended the laziness test using a LoopDistribute test to also cover
BPI.
Reviewers: hfinkel, davidxl
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D22835
llvm-svn: 277083
This patch lets CFLAnders respond to mod-ref queries. It also includes
a small bugfix to CFLSteens.
Patch by Jia Chen.
Differential Revision: https://reviews.llvm.org/D22823
llvm-svn: 276939
This change lets us prove things like
"{X,+,10} s< 5000" implies "{X+7,+,10} does not sign overflow"
It does this by replacing replacing getConstantDifference by
computeConstantDifference (which is smarter) in
isImpliedCondOperandsViaRanges.
llvm-svn: 276505
In D12090, the ExprValueMap was added to reuse existing value during SCEV expansion.
However, const folding and sext/zext distribution can make the reuse still difficult.
A simplified case is: suppose we know S1 expands to V1 in ExprValueMap, and
S1 = S2 + C_a
S3 = S2 + C_b
where C_a and C_b are different SCEVConstants. Then we'd like to expand S3 as
V1 - C_a + C_b instead of expanding S2 literally. It is helpful when S2 is a
complex SCEV expr and S2 has no entry in ExprValueMap, which is usually caused
by the fact that S3 is generated from S1 after const folding.
In order to do that, we represent ExprValueMap as a mapping from SCEV to
ValueOffsetPair. We will save both S1->{V1, 0} and S2->{V1, C_a} into the
ExprValueMap when we create SCEV for V1. When S3 is expanded, it will first
expand S2 to V1 - C_a because of S2->{V1, C_a} in the map, then expand S3 to
V1 - C_a + C_b.
Differential Revision: https://reviews.llvm.org/D21313
llvm-svn: 276136
This patch adds costs for the vectorized implementations of CTPOP, the default values were seriously underestimating the cost of these and was encouraging vectorization on targets where serialized use of POPCNT would be much better.
Differential Revision: https://reviews.llvm.org/D22456
llvm-svn: 276104
This patch adds function summary support to CFLAnders. It also comes
with a lot of tests! Woohoo!
Patch by Jia Chen.
Differential Revision: https://reviews.llvm.org/D22450
llvm-svn: 276026
This patch adds proper handling of stratified attributes into our
anders-style CFLAA implementation. It also comes bundled with more
CFLAnders tests. :)
Patch by Jia Chen.
Differential Revision: https://reviews.llvm.org/D22325
llvm-svn: 275604
This adds an incomplete anders-style implementation for CFLAA. It's
incomplete in that it's missing interprocedural analysis, attrs
handling, etc. and that it needs more tests. More tests and features
will be added in future commits.
Patch by Jia Chen.
Differential Revision: https://reviews.llvm.org/D22291
llvm-svn: 275602
Summary:
In preparation for changing GlobalsAA to stop assuming that intrinsics
can't read arbitrary globals, we need to make sure GlobalsAA is querying
function attributes rather than relying on this assumption.
This patch was inspired by: http://reviews.llvm.org/D20206
Reviewers: jmolloy, hfinkel
Subscribers: eli.friedman, llvm-commits
Differential Revision: https://reviews.llvm.org/D21318
llvm-svn: 275433
Summary:
This is necessary for D21771. In order to add the hotness attribute to
optimization remarks we need BFI to be available in all passes that emit
optimization remarks.
However we don't want to pay for computing BFI unless the hotness
attribute is requested.
This is achieved by making BFI lazy at the very high-level through a new
analysis pass -- BFI is not calculated unless requested.
I am adding a test to check the laziness under D21771 where the first
user of the analysis is added.
Reviewers: hfinkel, dexonsmith, davidxl
Subscribers: davidxl, dexonsmith, llvm-commits
Differential Revision: http://reviews.llvm.org/D22141
llvm-svn: 275250
Make some AVX and AVX512 cast costs more precise.
Based on part of a patch by Elena Demikhovsky (D15604).
Differential Revision: http://reviews.llvm.org/D22064
llvm-svn: 275106
For functions which are known to return their argument,
isDereferenceableAndAlignedPointer can examine the argument value.
Differential Revision: http://reviews.llvm.org/D9384
llvm-svn: 275038
When building SCEVs, if a function is known to return its argument, then we can
build the SCEV using the corresponding argument value.
Differential Revision: http://reviews.llvm.org/D9381
llvm-svn: 275037
Motivated by the work on the llvm.noalias intrinsic, teach BasicAA to look
through returned-argument functions when answering queries. This is essential
so that we don't loose all other AA information when supplementing with
llvm.noalias.
Differential Revision: http://reviews.llvm.org/D9383
llvm-svn: 275035
Simon Pilgrim