interleave-group
The vectorizer currently does not attempt to create interleave-groups that
contain predicated loads/stores; predicated strided accesses can currently be
vectorized only using masked gather/scatter or scalarization. This patch makes
predicated loads/stores candidates for forming interleave-groups during the
Loop-Vectorizer's analysis, and adds the proper support for masked-interleave-
groups to the Loop-Vectorizer's planning and transformation stages. The patch
also extends the TTI API to allow querying the cost of masked interleave groups
(which each target can control); Targets that support masked vector loads/
stores may choose to enable this feature and allow vectorizing predicated
strided loads/stores using masked wide loads/stores and shuffles.
Reviewers: Ayal, hsaito, dcaballe, fhahn, javed.absar
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D53011
llvm-svn: 344472
In the motivating case from PR35681 and represented by the macro-fuse-cmp test:
https://bugs.llvm.org/show_bug.cgi?id=35681
...there's a 37 -> 31 byte size win for the loop because we eliminate the big base
address offsets.
SPEC2017 on Ryzen shows no significant perf difference.
Differential Revision: https://reviews.llvm.org/D42607
llvm-svn: 324289
All these headers already depend on CodeGen headers so moving them into
CodeGen fixes the layering (since CodeGen depends on Target, not the
other way around).
llvm-svn: 318490
- Targets that want to support memcmp expansions now return the list of
supported load sizes.
- Expansion codegen does not assume that all power-of-two load sizes
smaller than the max load size are valid. For examples, this is not the
case for x86(32bit)+sse2.
Fixes PR34887.
llvm-svn: 316905
Summary:
Right now there are two functions with the same name, one does the work
and the other one returns true if expansion is needed. Rename
TargetTransformInfo::expandMemCmp to make it more consistent with other
members of TargetTransformInfo.
Remove the unused Instruction* parameter.
Differential Revision: https://reviews.llvm.org/D38165
llvm-svn: 314096
This is intended to be a superset of the functionality from D31037 (EarlyCSE) but implemented
as an independent pass, so there's no stretching of scope and feature creep for an existing pass.
I also proposed a weaker version of this for SimplifyCFG in D30910. And I initially had almost
this same functionality as an addition to CGP in the motivating example of PR31028:
https://bugs.llvm.org/show_bug.cgi?id=31028
The advantage of positioning this ahead of SimplifyCFG in the pass pipeline is that it can allow
more flattening. But it needs to be after passes (InstCombine) that could sink a div/rem and
undo the hoisting that is done here.
Decomposing remainder may allow removing some code from the backend (PPC and possibly others).
Differential Revision: https://reviews.llvm.org/D37121
llvm-svn: 312862
SLP vectorizer supports horizontal reductions for Add/FAdd binary
operations. Patch adds support for horizontal min/max reductions.
Function getReductionCost() is split to getArithmeticReductionCost() for
binary operation reductions and getMinMaxReductionCost() for min/max
reductions.
Patch fixes PR26956.
Differential revision: https://reviews.llvm.org/D27846
llvm-svn: 312791
Summary:
We add the precise cache sizes and associativity for the following Intel
architectures:
- Penry
- Nehalem
- Westmere
- Sandy Bridge
- Ivy Bridge
- Haswell
- Broadwell
- Skylake
- Kabylake
Polly uses since several months a performance model for BLAS computations that
derives optimal cache and register tile sizes from cache and latency
information (based on ideas from "Analytical Modeling Is Enough for High-Performance BLIS", by Tze Meng Low published at TOMS 2016).
While bootstrapping this model, these target values have been kept in Polly.
However, as our implementation is now rather mature, it seems time to teach
LLVM itself about cache sizes.
Interestingly, L1 and L2 cache sizes are pretty constant across
micro-architectures, hence a set of architecture specific default values
seems like a good start. They can be expanded to more target specific values,
in case certain newer architectures require different values. For now a set
of Intel architectures are provided.
Just as a little teaser, for a simple gemm kernel this model allows us to
improve performance from 1.2s to 0.27s. For gemm kernels with less optimal
memory layouts even larger speedups can be reported.
Reviewers: Meinersbur, bollu, singam-sanjay, hfinkel, gareevroman, fhahn, sebpop, efriedma, asb
Reviewed By: fhahn, asb
Subscribers: lsaba, asb, pollydev, llvm-commits
Differential Revision: https://reviews.llvm.org/D37051
llvm-svn: 311647
Store operation takes 2 UOps on X86 processors. The exact cost calculation affects several optimization passes including loop unroling.
This change compensates performance degradation caused by https://reviews.llvm.org/D34458 and shows improvements on some benchmarks.
Differential Revision: https://reviews.llvm.org/D35888
llvm-svn: 311285
The root cause of reverting was fixed - PR33514.
Summary:
The patch makes instruction count the highest priority for
LSR solution for X86 (previously registers had highest priority).
Reviewers: qcolombet
Differential Revision: http://reviews.llvm.org/D30562
From: Evgeny Stupachenko <evstupac@gmail.com>
<evgeny.v.stupachenko@intel.com>
llvm-svn: 310289
The cost of an interleaved access was only implemented for AVX512. For other
X86 targets an overly conservative Base cost was returned, resulting in
avoiding vectorization where it is actually profitable to vectorize.
This patch starts to add costs for AVX2 for most prominent cases of
interleaved accesses (stride 3,4 chars, for now).
Note1: Improvements of up to ~4x were observed in some of EEMBC's rgb
workloads; There is also a known issue of 15-30% degradations on some of these
workloads, associated with an interleaved access followed by type
promotion/widening; the resulting shuffle sequence is currently inefficient and
will be improved by a series of patches that extend the X86InterleavedAccess pass
(such as D34601 and more to follow).
Note 2: The costs in this patch do not reflect port pressure penalties which can
be very dominant in the case of interleaved accesses since most of the shuffle
operations are restricted to a single port. Further tuning, that may incorporate
these considerations, will be done on top of the upcoming improved shuffle
sequences (that is, along with the abovementioned work to extend
X86InterleavedAccess pass).
Differential Revision: https://reviews.llvm.org/D34023
llvm-svn: 306238
There are a couple of potential improvements as seen in the IR and asm:
1. We're unnecessarily extending to a larger type to compare values.
2. The codegen for (select cond, 1, -1) could avoid a cmov.
(or we could change the order of the compares, so we have a select with 0 operand)
llvm-svn: 305802
This seems to be interacting badly with ASan somehow, causing false reports of
heap-buffer overflows: PR33514.
> Summary:
> The patch makes instruction count the highest priority for
> LSR solution for X86 (previously registers had highest priority).
>
> Reviewers: qcolombet
>
> Differential Revision: http://reviews.llvm.org/D30562
>
> From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 305720
Summary:
The patch makes instruction count the highest priority for
LSR solution for X86 (previously registers had highest priority).
Reviewers: qcolombet
Differential Revision: http://reviews.llvm.org/D30562
From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 304824
Summary:
Expanding the loop idiom test for memcpy to also recognize
unordered atomic memcpy. The only difference for recognizing
an unordered atomic memcpy and instead of a normal memcpy is
that the loads and/or stores involved are unordered atomic operations.
Background: http://lists.llvm.org/pipermail/llvm-dev/2017-May/112779.html
Patch by Daniel Neilson!
Reviewers: reames, anna, skatkov
Reviewed By: reames, anna
Subscribers: llvm-commits, mzolotukhin
Differential Revision: https://reviews.llvm.org/D33243
llvm-svn: 304806
getArithmeticInstrCost(), getShuffleCost(), getCastInstrCost(),
getCmpSelInstrCost(), getVectorInstrCost(), getMemoryOpCost(),
getInterleavedMemoryOpCost() implemented.
Interleaved access vectorization enabled.
BasicTTIImpl::getCastInstrCost() improved to check for legal extending loads,
in which case the cost of the z/sext instruction becomes 0.
Review: Ulrich Weigand, Renato Golin.
https://reviews.llvm.org/D29631
llvm-svn: 300052
Summary:
LSV wants to know the maximum size that can be loaded to a vector register.
On X86, this always matches the maximum register width. Implement this
accordingly and add a test to make sure that LSV can vectorize up to the
maximum permissible width on X86.
Reviewers: delena, arsenm
Reviewed By: arsenm
Subscribers: wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D31504
llvm-svn: 299589
getIntrinsicInstrCost() used to only compute scalarization cost based on types.
This patch improves this so that the actual arguments are checked when they are
available, in order to handle only unique non-constant operands.
Tests updates:
Analysis/CostModel/X86/arith-fp.ll
Transforms/LoopVectorize/AArch64/interleaved_cost.ll
Transforms/LoopVectorize/ARM/interleaved_cost.ll
The improvement in getOperandsScalarizationOverhead() to differentiate on
constants made it necessary to update the interleaved_cost.ll tests even
though they do not relate to intrinsics.
Review: Hal Finkel
https://reviews.llvm.org/D29540
llvm-svn: 297705
Refactoring to remove duplications of this method.
New method getOperandsScalarizationOverhead() that looks at the present unique
operands and add extract costs for them. Old behaviour was to just add extract
costs for one operand of the type always, which still happens in
getArithmeticInstrCost() if no operands are provided by the caller.
This is a good start of improving on this, but there are more places
that can be improved by using getOperandsScalarizationOverhead().
Review: Hal Finkel
https://reviews.llvm.org/D29017
llvm-svn: 293155
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
This code seems to be target dependent which may not be the same for all targets.
Passed the decision whether the given stride is complex or not to the target by sending stride information via SCEV to getAddressComputationCost instead of 'IsComplex'.
Specifically at X86 targets we dont see any significant address computation cost in case of the strided access in general.
Differential Revision: https://reviews.llvm.org/D27518
llvm-svn: 291106
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
All of these existed because MSVC 2013 was unable to synthesize default
move ctors. We recently dropped support for it so all that error-prone
boilerplate can go.
No functionality change intended.
llvm-svn: 284721
This reverts commit r278048. Something changed between the last time I
built this--it takes awhile on my ridiculously slow and ancient
computer--and now that broke this.
llvm-svn: 278053
Summary:
Based on two patches by Michael Mueller.
This is a target attribute that causes a function marked with it to be
emitted as "hotpatchable". This particular mechanism was originally
devised by Microsoft for patching their binaries (which they are
constantly updating to stay ahead of crackers, script kiddies, and other
ne'er-do-wells on the Internet), but is now commonly abused by Windows
programs to hook API functions.
This mechanism is target-specific. For x86, a two-byte no-op instruction
is emitted at the function's entry point; the entry point must be
immediately preceded by 64 (32-bit) or 128 (64-bit) bytes of padding.
This padding is where the patch code is written. The two byte no-op is
then overwritten with a short jump into this code. The no-op is usually
a `movl %edi, %edi` instruction; this is used as a magic value
indicating that this is a hotpatchable function.
Reviewers: majnemer, sanjoy, rnk
Subscribers: dberris, llvm-commits
Differential Revision: https://reviews.llvm.org/D19908
llvm-svn: 278048
The cost is calculated for all X86 targets. When gather/scatter instruction
is not supported we calculate the cost of scalar sequence.
Differential revision: http://reviews.llvm.org/D15677
llvm-svn: 256519
When the target does not support these intrinsics they should be converted to a chain of scalar load or store operations.
If the mask is not constant, the scalarizer will build a chain of conditional basic blocks.
I added isLegalMaskedGather() isLegalMaskedScatter() APIs.
Differential Revision: http://reviews.llvm.org/D13722
llvm-svn: 251237
Originally I planned to use the same interface for masked gather/scatter and set isConsecutive to "false" in this case.
Now I'm implementing masked gather/scatter and see that the interface is inconvenient. I want to add interfaces isLegalMaskedGather() / isLegalMaskedScatter() instead of using the "Consecutive" parameter in the existing interfaces.
Differential Revision: http://reviews.llvm.org/D13850
llvm-svn: 250686
rather than 'unsigned' for their costs.
For something like costs in particular there is a natural "negative"
value, that of savings or saved cost. As a consequence, there is a lot
of code that subtracts or creates negative values based on cost, all of
which is prone to awkwardness or bugs when dealing with an unsigned
type. Similarly, we *never* want these values to wrap, as that would
cause Very Bad code generation (likely percieved as an infinite loop as
we try to emit over 2^32 instructions or some such insanity).
All around 'int' seems a much better fit for these basic metrics. I've
added asserts to ensure that at least the TTI interface never returns
negative numbers here. If we ever have a use case for negative numbers,
we can remove this, but this way a bug where someone used '-1' to
produce a 'very large' cost will be caught by the assert.
This passes all tests, and is also UBSan clean.
No functional change intended.
Differential Revision: http://reviews.llvm.org/D11741
llvm-svn: 244080
DataLayout is no longer optional. It was initialized with or without
a DataLayout, and the DataLayout when supplied could have been the
one from the TargetMachine.
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Reviewers: echristo
Subscribers: jholewinski, llvm-commits, rafael, yaron.keren
Differential Revision: http://reviews.llvm.org/D11021
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 241774
This checks subtarget feature compatibility for inlining by verifying
that the callee is a strict subset of the caller's features. This includes
the cpu as part of the subtarget we can get via the incoming functions as
the backend takes CPUs as feature sets.
This allows us to inline things like:
int foo() { return baz(); }
int __attribute__((target("sse4.2"))) bar() {
return foo();
}
so that generic code can be inlined into specialized functions.
llvm-svn: 241221
The patch disabled unrolling in loop vectorization pass when VF==1 on x86 architecture,
by setting MaxInterleaveFactor to 1. Unrolling in loop vectorization pass may introduce
the cost of overflow check, memory boundary check and extra prologue/epilogue code when
regular unroller will unroll the loop another time. Disable it when VF==1 remove the
unnecessary cost on x86. The same can be done for other platforms after verifying
interleaving/memory bound checking to be not perf critical on those platforms.
Differential Revision: http://reviews.llvm.org/D9515
llvm-svn: 236613
now that we have a correct and cached subtarget specific to the
function.
Also, finish providing a cached per-function subtarget in the core
LLVMTargetMachine -- that layer hadn't switched over yet.
The only use of the TargetMachine was to re-lookup a subtarget for
a particular function to work around the fact that TTI was immutable.
Now that it is per-function and we haved a cached subtarget, use it.
This still leaves a few interfaces with real warts on them where we were
passing Function objects through the TTI interface. I'll remove these
and clean their usage up in subsequent commits now that this isn't
necessary.
llvm-svn: 227738
intermediate TTI implementation template and instead query up to the
derived class for both the TargetMachine and the TargetLowering.
Most of the derived types had a TLI cached already and there is no need
to store a less precisely typed target machine pointer.
This will in turn make it much cleaner to look up the TLI via
a per-function subtarget instead of the generic subtarget, and it will
pave the way toward pulling the subtarget used for unroll preferences
into the same form once we are *always* using the function to look up
the correct subtarget.
llvm-svn: 227737
TargetIRAnalysis access path directly rather than implementing getTTI.
This even removes getTTI from the interface. It's more efficient for
each target to just register a precise callback that creates their
specific TTI.
As part of this, all of the targets which are building their subtargets
individually per-function now build their TTI instance with the function
and thus look up the correct subtarget and cache it. NVPTX, R600, and
XCore currently don't leverage this functionality, but its trivial for
them to add it now.
llvm-svn: 227735
null.
For some reason some of the original TTI code supported a null target
machine. This seems to have been legacy, and I made matters worse when
refactoring this code by spreading that pattern further through the
various targets.
The TargetMachine can't actually be null, and it doesn't make sense to
support that use case. I've now consistently removed it and removed all
of the code trying to cope with that situation. This is probably good,
as several targets *didn't* cope with it being null despite the null
default argument in their constructors. =]
llvm-svn: 227734
base which it adds a single analysis pass to, to instead return the type
erased TargetTransformInfo object constructed for that TargetMachine.
This removes all of the pass variants for TTI. There is now a single TTI
*pass* in the Analysis layer. All of the Analysis <-> Target
communication is through the TTI's type erased interface itself. While
the diff is large here, it is nothing more that code motion to make
types available in a header file for use in a different source file
within each target.
I've tried to keep all the doxygen comments and file boilerplate in line
with this move, but let me know if I missed anything.
With this in place, the next step to making TTI work with the new pass
manager is to introduce a really simple new-style analysis that produces
a TTI object via a callback into this routine on the target machine.
Once we have that, we'll have the building blocks necessary to accept
a function argument as well.
llvm-svn: 227685
Dorit Nuzman