Summary:
this reduces significantly the number of assumes generated without aftecting too much
the information that is preserved. this improves the compile-time cost
of enable-knowledge-retention significantly.
Reviewers: jdoerfert, sstefan1
Reviewed By: jdoerfert
Subscribers: hiraditya, asbirlea, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79650
Summary:
Extend StackLifetime with option to calculate liveliness
where alloca is only considered alive on basic block entry
if all non-dead predecessors had it alive at terminators.
Depends on D82043.
Reviewers: eugenis
Reviewed By: eugenis
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82124
Adds aarch64-sve-vector-bits-{min,max} to allow the size of SVE
data registers (in bits) to be specified. This allows the code
generator to make assumptions it normally couldn't. As a starting
point this information is used to mark fixed length vector types
that can fit within the specified size as legal.
Reviewers: rengolin, efriedma
Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D80384
For a loop, a join block is a block that is reachable along multiple
disjoint paths from the exiting block of a loop. If the exit condition
of the loop is divergent, then such join blocks must also be marked
divergent. This currently fails in some cases because not all join
blocks are identified correctly.
The workaround is to conservatively mark every join block of any
branch (not necessarily the exiting block of a loop) as divergent.
https://bugs.llvm.org/show_bug.cgi?id=46372
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D81806
Summary:
Assume all usages of this function are explicitly fixed-width operations
and cast to FixedVectorType
Reviewers: efriedma, sdesmalen, c-rhodes, majnemer, dblaikie
Reviewed By: sdesmalen
Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D80262
Values that are uniform within a loop but appear divergent to uses
outside the loop are "tainted" so that such uses are marked
divergent. But if such a use is a branch, then it's divergence needs
to be propagated. The simplest way to do that is to put the branch
back in the main worklist so that it is processed appropriately.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D81822
Summary:
this reduces significantly the number of assumes generated without aftecting too much
the information that is preserved. this improves the compile-time cost
of enable-knowledge-retention significantly.
Reviewers: jdoerfert, sstefan1
Reviewed By: jdoerfert
Subscribers: hiraditya, asbirlea, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79650
Have BasicTTI call the base implementation so that both agree on the
default behaviour, which the default being a cost of '1'. This has
required an X86 specific implementation as it seems to be very
reliant on those instructions being free. Changes are also made to
AMDGPU so that their implementations distinguish between cost kinds,
so that the unrolling isn't affected. PowerPC also has its own
implementation to prevent changes to the reg-usage vectorizer test.
The cost model test changes now reflect that ret instructions are not
generally free.
Differential Revision: https://reviews.llvm.org/D79164
Summary:
ThinLTO linking runs dataflow processing on collected
function parameters. Then StackSafetyGlobalInfoWrapperPass
in ThinLTO backend will run as usual looking up to external
symbol in the summary if needed.
Depends on D80985.
Reviewers: eugenis, pcc
Reviewed By: eugenis
Subscribers: inglorion, hiraditya, steven_wu, dexonsmith, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D81242
We don't need process paramenters which marked as
byval as we are not going to pass interested allocas
without copying.
If we pass value into byval argument, we just handle that
as Load of corresponding type and stop that branch of analysis.
Summary:
There was this comment here previously:
```
- // It is currently not resolved how to interpret NSW for left
- // shift by BitWidth - 1, so we avoid applying flags in that
- // case. Remove this check (or this comment) once the situation
- // is resolved. See
- // http://lists.llvm.org/pipermail/llvm-dev/2015-April/084195.html
- // and http://reviews.llvm.org/D8890 .
```
But langref was fixed in rL286785, and the behavior is pretty obvious:
http://volta.cs.utah.edu:8080/z/MM4WZP
^ nuw can always be propagated. nsw can be propagated if
either nuw is specified, or the shift is by *less* than bitwidth-1.
This mimics similar D81189 Reassociate change, alive2 is happy about that one.
I'm not sure `NUW` isn't being printed, but that seems unrelated.
Reviewers: mkazantsev, reames, sanjoy, nlopes, craig.topper, efriedma
Reviewed By: efriedma
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81243
Allow InvokeInst to have the second optional prof branch weight for
its unwind branch. InvokeInst is a terminator with two successors.
It might have its unwind branch taken many times. If so
the BranchProbabilityInfo unwind branch heuristic can be inaccurate.
This patch allows a higher accuracy calculated with both branch
weights set.
Changes:
- A new section about InvokeInst is added to
the BranchWeightMetadata page. It states the old information that
missed in the doc and adds new about the second branch weight.
- Verifier is changed to allow either 1 or 2 branch weights
for InvokeInst.
- A new test is written for BranchProbabilityInfo to demonstrate
the main improvement of the simple fix in calcMetadataWeights().
- Several new testcases are created for Inliner. Those check that
both weights are accounted for invoke instruction weight
calculation.
- PGOUseFunc::setBranchWeights() is fixed to be applicable to
InvokeInst.
Reviewers: davidxl, reames, xur, yamauchi
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D80618
This lets us to remove !stack-safe metadata and
better controll when to perform StackSafety
analysis.
Reviewers: eugenis
Subscribers: hiraditya, steven_wu, dexonsmith, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D80771
When fixing probability of unreachable edges in
BranchProbabilityInfo::calcMetadataWeights() proportionally distribute
remainder probability over the reachable edges. The old implementation
distributes the remainder probability evenly.
See examples in the fixed tests.
Reviewers: yamauchi, ebrevnov
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D80611
Currently, BasicAA does not exploit information about value ranges of
indexes. For example, consider the 2 pointers %a = %base and
%b = %base + %stride below, assuming they are used to access 4 elements.
If we know that %stride >= 4, we know the accesses do not alias. If
%stride is a constant, BasicAA currently gets that. But if the >= 4
constraint is encoded using an assume, it misses the NoAlias.
This patch extends DecomposedGEP to include an additional MinOtherOffset
field, which tracks the constant offset similar to the existing
OtherOffset, which the difference that it also includes non-negative
lower bounds on the range of the index value. When checking if the
distance between 2 accesses exceeds the access size, we can use this
improved bound.
For now this is limited to using non-negative lower bounds for indices,
as this conveniently skips cases where we do not have a useful lower
bound (because it is not constrained). We potential miss out in cases
where the lower bound is constrained but negative, but that can be
exploited in the future.
Reviewers: sanjoy, hfinkel, reames, asbirlea
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D76194
This add constant folding for all the integer vector reduce intrinsics,
providing that the argument is a constant vector. zeroinitializer always
produces 0 for all intrinsics, and other values can be handled with
APInt operators.
Differential Revision: https://reviews.llvm.org/D80516
We've started (D80598) the process of migrating away from the inline operand lists in statepoints to using explicit operand bundles. Update a few tests to reflect the new preference. More to come, these were simply the ones outside any obvious grouping.
We need to process only parameters. Allocas access can be calculated
afterwards.
Also don't create fake function for aliases and just resolve them on
initialization.
Recommitting part of "[CostModel] Unify Intrinsic Costs."
de71def3f5
Now that the 'free' intrinsic information has been sunk to the lowest
level, query the base implementation in BasicTTI before doing
anything else. I suspect this is the change that was causing the main
changes, particularly the large effects on debug builds.
Differential Revision: https://reviews.llvm.org/D80012
Summary:
Users of SCEV reasonably assume that multiplication of two constant
SCEVs will in turn be constant.
However, that is not always the case:
First, we can get here with reached depth limit, and will create
MultExpr SCEV `C1 * C2` and cache it.
Then, we can get here with the same operands, but with small depth
level. But this time we will find existing MultExpr SCEV and return
it, instead of expected constant SCEV.
This patch changes getMultExpr to not apply depth limit to all constant
operands expression, allowing them to be folded.
Reviewers: reames, mkazantsev
Subscribers: hiraditya, javed.absar, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79893
This has not been implemented by any backends which appear to cover
the functionality through getCastInstrCost. Sink what there is in the
default implementation into BasicTTI.
Differential Revision: https://reviews.llvm.org/D78922
Hide the method that allows setting probability for particular edge
and introduce a public method that sets probabilities for all
outgoing edges at once.
Setting individual edge probability is error prone. More over it is
difficult to check that the total probability is 1.0 because there is
no easy way to know when the user finished setting all
the probabilities.
Related bug is fixed in BranchProbabilityInfo::calcMetadataWeights().
Changing unreachable branch probabilities to raw(1) and distributing
the rest (oldProbability - raw(1)) over the reachable branches could
introduce total probability inaccuracy bigger than 1/numOfBranches.
Reviewers: yamauchi, ebrevnov
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79396
If we don't know anything about the alignment of a pointer, Align(1) is
still correct: all pointers are at least 1-byte aligned.
Included in this patch is a bugfix for an issue discovered during this
cleanup: pointers with "dereferenceable" attributes/metadata were
assumed to be aligned according to the type of the pointer. This
wasn't intentional, as far as I can tell, so Loads.cpp was fixed to
stop making this assumption. Frontends may need to be updated. I
updated clang's handling of C++ references, and added a release note for
this.
Differential Revision: https://reviews.llvm.org/D80072
After D76797 the dominator tree is no longer used in LVI, so we
can remove it as a pass dependency, and also get rid of the
dominator tree enabling/disabling logic in JumpThreading.
Apart from cleaning up the code, this also clarifies LVI
cache consistency, in that the LVI cache can no longer
depend on whether the DT was or wasn't enabled due to
pending DT updates at any given time.
Differential Revision: https://reviews.llvm.org/D76985
This is D77454, except for stores. All the infrastructure work was done
for loads, so the remaining changes necessary are relatively small.
Differential Revision: https://reviews.llvm.org/D79968
The "null-pointer-is-valid" attribute needs to be checked by many
pointer-related combines. To make the check more efficient, convert
it from a string into an enum attribute.
In the future, this attribute may be replaced with data layout
properties.
Differential Revision: https://reviews.llvm.org/D78862
We need to use it to handle <16 x double> indirect indexes
in the AMDGPU BE.
The only visible change from adding it is in ARM cost model.
To me it looks reasonable. With doubling a vector size it
quadruples the cost up to the size 8 and then it did only
double it. Now it also quadruples, which seems a logical
progression to me.
Actual AMDGPU code is to follow, this is a common part, plus
load/store legalization in the AMDGPU BE not to break what
works now.
Differential Revision: https://reviews.llvm.org/D79952
For IR generated by a compiler, this is really simple: you just take the
datalayout from the beginning of the file, and apply it to all the IR
later in the file. For optimization testcases that don't care about the
datalayout, this is also really simple: we just use the default
datalayout.
The complexity here comes from the fact that some LLVM tools allow
overriding the datalayout: some tools have an explicit flag for this,
some tools will infer a datalayout based on the code generation target.
Supporting this properly required plumbing through a bunch of new
machinery: we want to allow overriding the datalayout after the
datalayout is parsed from the file, but before we use any information
from it. Therefore, IR/bitcode parsing now has a callback to allow tools
to compute the datalayout at the appropriate time.
Not sure if I covered all the LLVM tools that want to use the callback.
(clang? lli? Misc IR manipulation tools like llvm-link?). But this is at
least enough for all the LLVM regression tests, and IR without a
datalayout is not something frontends should generate.
This change had some sort of weird effects for certain CodeGen
regression tests: if the datalayout is overridden with a datalayout with
a different program or stack address space, we now parse IR based on the
overridden datalayout, instead of the one written in the file (or the
default one, if none is specified). This broke a few AVR tests, and one
AMDGPU test.
Outside the CodeGen tests I mentioned, the test changes are all just
fixing CHECK lines and moving around datalayout lines in weird places.
Differential Revision: https://reviews.llvm.org/D78403
Fix the assumption that all bitcasts of the same type sizes are free.
We now only assume that bitcasts between ints and ptrs of the same
size are free. This allows TTImpl to just call the concrete
implementation of getCastInstrCost.
Differential Revision: https://reviews.llvm.org/D78918
Summary:
This patch makes propagatesPoison be more accurate by returning true on
more bin ops/unary ops/casts/etc.
The changed test in ScalarEvolution/nsw.ll was introduced by
a19edc4d15 .
IIUC, the goal of the tests is to show that iv.inc's SCEV expression still has
no-overflow flags even if the loop isn't in the wanted form.
It becomes more accurate with this patch, so think this is okay.
Reviewers: spatel, lebedev.ri, jdoerfert, reames, nikic, sanjoy
Reviewed By: spatel, nikic
Subscribers: regehr, nlopes, efriedma, fhahn, javed.absar, llvm-commits, hiraditya
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78615
- Specifically check for sext/zext users which have 'long' form NEON
instructions.
- Add more entries to the table for sext/zexts so that we can report
more accurately the number of vmovls required for NEON.
- Pass the instruction to the pass implementation.
Differential Revision: https://reviews.llvm.org/D79561
Summary: with this patch the assume salvageKnowledge will not generate assume if all knowledge is already available in an assume with valid context. assume bulider can also in some cases update an existing assume with better information.
Reviewers: jdoerfert
Reviewed By: jdoerfert
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78014
Summary:
If the only use of a value is a start or end lifetime intrinsic then mark the intrinsic as trivially dead. This should allow for that value to then be removed as well.
Currently, this only works for allocas, globals, and arguments.
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79355
No changes relative to last time, but after a mitigation for
an AMDGPU regression landed.
---
If SimplifyInstruction() does not succeed in simplifying the
instruction, it will compute the known bits of the instruction
in the hope that all bits are known and the instruction can be
folded to a constant. I have removed a similar optimization
from InstCombine in D75801, and would like to drop this one as well.
On average, we spend ~1% of total compile-time performing this
known bits calculation. However, if we introduce some additional
statistics for known bits computations and how many of them succeed
in simplifying the instruction we get (on test-suite):
instsimplify.NumKnownBits: 216
instsimplify.NumKnownBitsComputed: 13828375
valuetracking.NumKnownBitsComputed: 45860806
Out of ~14M known bits calculations (accounting for approximately
one third of all known bits calculations), only 0.0015% succeed in
producing a constant. Those cases where we do succeed to compute
all known bits will get folded by other passes like InstCombine
later. On test-suite, only lencod.test and GCC-C-execute-pr44858.test
show a hash difference after this change. On lencod we see an
improvement (a loop phi is optimized away), on the GCC torture
test a regression (a function return value is determined only
after IPSCCP, preventing propagation from a noinline function.)
There are various regressions in InstSimplify tests. However, all
of these cases are already handled by InstCombine, and corresponding
tests have already been added there.
Differential Revision: https://reviews.llvm.org/D79294
Summary: When phi translation cannot be performed, be conservative and make the MemoryLocation unknown.
Reviewers: george.burgess.iv
Subscribers: Prazek, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79386
If SimplifyInstruction() does not succeed in simplifying the
instruction, it will compute the known bits of the instruction
in the hope that all bits are known and the instruction can be
folded to a constant. I have removed a similar optimization
from InstCombine in D75801, and would like to drop this one as well.
On average, we spend ~1% of total compile-time performing this
known bits calculation. However, if we introduce some additional
statistics for known bits computations and how many of them succeed
in simplifying the instruction we get (on test-suite):
instsimplify.NumKnownBits: 216
instsimplify.NumKnownBitsComputed: 13828375
valuetracking.NumKnownBitsComputed: 45860806
Out of ~14M known bits calculations (accounting for approximately
one third of all known bits calculations), only 0.0015% succeed in
producing a constant. Those cases where we do succeed to compute
all known bits will get folded by other passes like InstCombine
later. On test-suite, only lencod.test and GCC-C-execute-pr44858.test
show a hash difference after this change. On lencod we see an
improvement (a loop phi is optimized away), on the GCC torture
test a regression (a function return value is determined only
after IPSCCP, preventing propagation from a noinline function.)
There are various regressions in InstSimplify tests. However, all
of these cases are already handled by InstCombine, and corresponding
tests have already been added there.
Differential Revision: https://reviews.llvm.org/D79294
If both the source and the destination need to be split then the two halves of the split operation are completely independent and don't need to be split or joined. So we don't need to assess a cost for the split or join.
Differential Revision: https://reviews.llvm.org/D79111
Also fix some cost tables for vXi1 types to match the costs entries for the types they will be promoted to.
Differential Revision: https://reviews.llvm.org/D79045
vpermw is 2 uops. vpermt2b/vpermt2w are two shuffle uops and a port 015 uop. Weirdly vpermb is a single uop.
This patch bumps the cost to 2 for these operations. Maybe should go to 3 for the vpermt2*, but I've started conservative.
I've also removed a few entries that were now the same as earlier subtargets or that I didn't think we really did. Like I don't think we extend v32i8 to v32i16, shuffle, and then truncate.
Differential Revision: https://reviews.llvm.org/D79148
Summary:
By design 'BranchProbabilityInfo:: getEdgeProbability(const BasicBlock *Src, const BasicBlock *Dst) const' should return sum of probabilities over all edges from Src to Dst. Current implementation is buggy and returns 1/num_of_successors if probabilities are not explicitly set.
Note current implementation of BPI printing has an issue as well and annotates each edge with sum of probabilities over all ages from one basic block to another. That's why 30% probability reported (instead of 10%) in the lit test. This is not urgent issue since only printing is affected.
Note also current implementation assumes that either all or none edges have probabilities set. This is not the only place which uses such assumption. At least we should assert that in verifier. In addition we can think on a more robust API of BPI which would prevent situations.
Reviewers: skatkov, yrouban, taewookoh
Reviewed By: skatkov
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79071
Summary:
A valid DominatorTree is needed to do PhiTranslation.
Before this patch, a MemoryUse could be optimized to an access outside a loop, while the address it loads from is modified in the loop.
This can lead to a miscompile.
Reviewers: george.burgess.iv
Subscribers: Prazek, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79068
We generate much better code these days than we used to. And we use the same sequence for AVX1 and AVX2 for these
For v4i64->v4i32 we generate:
vextractf128 xmm1, ymm0, 1
vshufps xmm0, xmm0, xmm1, 136 # xmm0 = xmm0[0,2],xmm1[0,2]
And for v8i64->v8i32 we generate:
vperm2f128 ymm2, ymm0, ymm1, 49 # ymm2 = ymm0[2,3],ymm1[2,3]
vinsertf128 ymm0, ymm0, xmm1, 1
vshufps ymm0, ymm0, ymm2, 136 # ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
Differential Revision: https://reviews.llvm.org/D79109
The improvements to the x86 vector insert/extract element costs in D74976 resulted in the estimated costs for vector initialization and scalarization increasing higher than should be expected. This is particularly noticeable on pre-SSE4 targets where the available of legal INSERT_VECTOR_ELT ops is more limited.
This patch does 2 things:
1 - it implements X86TTIImpl::getScalarizationOverhead to more accurately represent the typical costs of a ISD::BUILD_VECTOR pattern.
2 - it adds a DemandedElts mask to getScalarizationOverhead to permit the SLP's BoUpSLP::getGatherCost to be rewritten to use it directly instead of accumulating raw vector insertion costs.
This fixes PR45418 where a v4i8 (zext'd to v4i32) was no longer vectorizing.
A future patch should extend X86TTIImpl::getScalarizationOverhead to tweak the EXTRACT_VECTOR_ELT scalarization costs as well.
Reviewed By: @craig.topper
Differential Revision: https://reviews.llvm.org/D78216
I've modified isTruncateFree to get an accurate cost for types that need to be split. I'm planning to look into fixing it for all vectors, but need more cost cleanups first.
Differential Revision: https://reviews.llvm.org/D78973
All avx512 truncate instructions except vXi64->vXi32 are 2 uops
on port 5. So raise their costs to 2. Except when we have an
earlier faster sequence like pshufb for 128 bit input vectors.
Add a lower cost of 3 v16i16->v16i8 with avx512f where we can
extend to v16i32 then truncate. And a cost of 2 for avx512bw with
and without avx512vl. There we can use vpmovwb with either a ymm
or zmm input. Both of these beat masking, splitting, and using
packuswb which is our avx/avx2 codegen.
We aren't properly costing extends from k-registers. I also added
command lines without avx512bw to be able to show all the different
extending strategies we have.
Previously, we just always returned 1. But that ignores that we have to do the operation for each subvector or scalar component.
Differential Revision: https://reviews.llvm.org/D78824
Summary:
This is RFC for fixes in poison-related functions of ValueTracking.
These functions assume that a value can be poison bitwisely, but the semantics
of bitwise poison is not clear at the moment.
Allowing a value to have bitwise poison adds complexity to reasoning about
correctness of optimizations.
This patch makes the analysis functions simply assume that a value is
either fully poison or not, which has been used to understand the correctness
of a few previous optimizations.
The bitwise poison semantics seems to be only used by these functions as well.
In terms of implementation, using value-wise poison concept makes existing
functions do more precise analysis, which is what this patch contains.
Reviewers: spatel, lebedev.ri, jdoerfert, reames, nikic, nlopes, regehr
Reviewed By: nikic
Subscribers: fhahn, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78503
Summary:
This is RFC for fixes in poison-related functions of ValueTracking.
These functions assume that a value can be poison bitwisely, but the semantics
of bitwise poison is not clear at the moment.
Allowing a value to have bitwise poison adds complexity to reasoning about
correctness of optimizations.
This patch makes the analysis functions simply assume that a value is
either fully poison or not, which has been used to understand the correctness
of a few previous optimizations.
The bitwise poison semantics seems to be only used by these functions as well.
In terms of implementation, using value-wise poison concept makes existing
functions do more precise analysis, which is what this patch contains.
Reviewers: spatel, lebedev.ri, jdoerfert, reames, nikic, nlopes, regehr
Reviewed By: nikic
Subscribers: fhahn, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78503
This will allow us to use the datalayout to disambiguate other
constructs in IR, like load alignment. Split off from D78403.
Differential Revision: https://reviews.llvm.org/D78413
This moves v32i16/v64i8 to a model consistent with how we
treat integer types with avx1.
This does change the ABI for types vXi16/vXi8 vectors larger than
512 bits to pass in multiple zmms instead of multiple ymms. We'd
already hacked some code to make v64i8/v32i16 pass in zmm.
Cost model is still a bit of a mess. In some place I tried to
match existing behavior. But really we need to account for
splitting and concating costs. Cost model for shuffles is
especially pessimistic.
Differential Revision: https://reviews.llvm.org/D76212
Summary:
Don't attempt to analyze the decomposed GEP for scalable type.
GEP index scale is not compile-time constant for scalable type.
Be conservative, return MayAlias.
Explicitly call TypeSize::getFixedSize() to assert on places where
scalable type doesn't make sense.
Add unit tests to check functionality of -basicaa for scalable type.
This patch is needed for D76944.
Reviewers: sdesmalen, efriedma, spatel, bjope, ctetreau
Reviewed By: efriedma
Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D77828
If we're inserting into v2i8/v4i8/v8i8/v2i16/v4i16 style sub-128bit vectors ensure we don't use the SK_PermuteTwoSrc cost of the legalized value type - this is a followup to rG12c629ec6c59 which added equivalent sub-128bit shuffle costs
This is similar to what I recently did for getArithmeticReductionCost.
I'm trying to account for the narrowing from 512->256->128 as we go.
I've also added a new helper method getMinMaxCost that tries to
handle the cases where we have native min/max instructions and
fall back to cmp+select when we don't.
Differential Revision: https://reviews.llvm.org/D76634
v2i8/v4i8/v8i8 + v2i16/v4i16 all show up in vectorizer code and by just using the legalized types (v16i8/v8i16) we're highly exaggerating the actual cost of the shuffle.
For the PHI node
%1 = phi [%A, %entry], [%X, %latch]
it is incorrect to use SCEV of backedge val %X as an exit value
of PHI unless %X is loop invariant.
This is because exit value of %1 is value of %X at one-before-last
iteration of the loop.
Reviewed By: Meinersbur
Differential Revision: https://reviews.llvm.org/D73181
Add a new llvm.amdgcn.ballot intrinsic modeled on the ballot function
in GLSL and other shader languages. It returns a bitfield containing the
result of its boolean argument in all active lanes, and zero in all
inactive lanes.
This is intended to replace the existing llvm.amdgcn.icmp and
llvm.amdgcn.fcmp intrinsics after a suitable transition period.
Use the new intrinsic in the atomic optimizer pass.
Differential Revision: https://reviews.llvm.org/D65088
This patch attempts to more accurately model the reduction of
power of 2 vectors of types we natively support. This takes into
account the narrowing of vectors that occur as we go from 512
bits to 256 bits, to 128 bits. It also takes into account the use
of wider elements in the shuffles for the first 2 steps of a
reduction from 128 bits. And uses a v8i16 shift for the final step
of vXi8 reduction.
The default implementation uses the legalized type for the arithmetic
for all levels. And uses the single source permute cost of the
legalized type for all levels. This penalizes things like
lack of v16i8 pshufb on pre-sse3 targets and the splitting and
joining that needs to be done for integer types on AVX1. We never
need v16i8 shuffle for a reduction and we only need split AVX1 ops
when type the type wide and needs to be split. I think we're still
over costing splits and joins for AVX1, but we're closer now.
I've also removed all pairwise special casing because I don't
think we ever want to generate that on X86. I've also adjusted
the add handling to more accurately account for any type splitting
that occurs before we reach a legal type.
Differential Revision: https://reviews.llvm.org/D76478
Summary:
Avoid blind cast from Operator to ExtractElementInst in
computeKnownBitsFromOperator. This resulted in some crashes
in downstream fuzzy testing. Instead we use getOperand directly
on the Operator when accessing the vector/index operands.
Haven't seen any problems with InsertElement and ShuffleVector,
but I believe those could be used in constant expressions as well.
So the same kind of fix as for ExtractElement was also applied for
InsertElement.
When it comes to ShuffleVector we now simply bail out if a dynamic
cast of the Operator to ShuffleVectorInst fails. I've got no
reproducer indicating problems for ShuffleVector, and a fix would be
slightly more complicated as getShuffleDemandedElts is involved.
Reviewers: RKSimon, nikic, spatel, efriedma
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76564
Previously we multiplied the cost for the table entries by the number of splits needed. But that implies that each split goes through a reduction to scalar independently. I think what really happens is that the we AND/OR the split pieces until we're down to a single value with a legal type and then do special reduction sequence on that.
So to model that this patch takes the number of splits minus one multiplied by the cost of a AND/OR at the legal element count and adds that on top of the table lookup.
Differential Revision: https://reviews.llvm.org/D76400
This is fixing up various places that use the implicit
TypeSize->uint64_t conversion.
The new overloads in MemoryLocation.h are already used in various places
that construct a MemoryLocation from a TypeSize, including MemorySSA.
(They were using the implicit conversion before.)
Differential Revision: https://reviews.llvm.org/D76249
Summary:
Run StackSafetyAnalysis at the end of the IR pipeline and annotate
proven safe allocas with !stack-safe metadata. Do not instrument such
allocas in the AArch64StackTagging pass.
Reviewers: pcc, vitalybuka, ostannard
Reviewed By: vitalybuka
Subscribers: merge_guards_bot, kristof.beyls, hiraditya, cfe-commits, gilang, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D73513
Because we have to use a ConstantExpr at some point, the canonical form
isn't set in stone, but this seems reasonable.
The pretty sizeof(<vscale x 4 x i32>) dumping is a relic of ancient
LLVM; I didn't have to touch that code. :)
Differential Revision: https://reviews.llvm.org/D75887
Summary:
Do not iterate on scalable vector. Also do not return constant scalable vector
from ConstantInt::get().
Fix result type by using getElementCount() instead of getNumElements().
Reviewers: sdesmalen, efriedma, apazos, huntergr, willlovett
Reviewed By: efriedma
Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73753
Refines the gather/scatter cost model, but also changes the TTI
function getIntrinsicInstrCost to accept an additional parameter
which is needed for the gather/scatter cost evaluation.
This did require trivial changes in some non-ARM backends to
adopt the new parameter.
Extending gathers and truncating scatters are now priced cheaper.
Differential Revision: https://reviews.llvm.org/D75525
If a call argument has the "returned" attribute, we can simplify
the call to the value of that argument. The "-inst-simplify" pass
already handled this for the constant integer argument case via
known bits, which is invoked in SimplifyInstruction. However,
non-constant (or non-int) arguments are not handled at all right now.
This addresses one of the regressions from D75801.
Differential Revision: https://reviews.llvm.org/D75815
Summary: This patch adds an analysis pass to collect loop nests and
summarize properties of the nest (e.g the nest depth, whether the nest
is perfect, what's the innermost loop, etc...).
The motivation for this patch was discussed at the latest meeting of the
LLVM loop group (https://ibm.box.com/v/llvm-loop-nest-analysis) where we
discussed
the unimodular loop transformation framework ( “A Loop Transformation
Theory and an Algorithm to Maximize Parallelism”, Michael E. Wolf and
Monica S. Lam, IEEE TPDS, October 1991). The unimodular framework
provides a convenient way to unify legality checking and code generation
for several loop nest transformations (e.g. loop reversal, loop
interchange, loop skewing) and their compositions. Given that the
unimodular framework is applicable to perfect loop nests this is one
property of interest we expose in this analysis. Several other utility
functions are also provided. In the future other properties of interest
can be added in a centralized place.
Authored By: etiotto
Reviewer: Meinersbur, bmahjour, kbarton, Whitney, dmgreen, fhahn,
reames, hfinkel, jdoerfert, ppc-slack
Reviewed By: Meinersbur
Subscribers: bryanpkc, ppc-slack, mgorny, hiraditya, llvm-commits
Tag: LLVM
Differential Revision: https://reviews.llvm.org/D68789
Summary: This patch adds an analysis pass to collect loop nests and
summarize properties of the nest (e.g the nest depth, whether the nest
is perfect, what's the innermost loop, etc...).
The motivation for this patch was discussed at the latest meeting of the
LLVM loop group (https://ibm.box.com/v/llvm-loop-nest-analysis) where we
discussed
the unimodular loop transformation framework ( “A Loop Transformation
Theory and an Algorithm to Maximize Parallelism”, Michael E. Wolf and
Monica S. Lam, IEEE TPDS, October 1991). The unimodular framework
provides a convenient way to unify legality checking and code generation
for several loop nest transformations (e.g. loop reversal, loop
interchange, loop skewing) and their compositions. Given that the
unimodular framework is applicable to perfect loop nests this is one
property of interest we expose in this analysis. Several other utility
functions are also provided. In the future other properties of interest
can be added in a centralized place.
Authored By: etiotto
Reviewer: Meinersbur, bmahjour, kbarton, Whitney, dmgreen, fhahn,
reames, hfinkel, jdoerfert, ppc-slack
Reviewed By: Meinersbur
Subscribers: bryanpkc, ppc-slack, mgorny, hiraditya, llvm-commits
Tag: LLVM
Differential Revision: https://reviews.llvm.org/D68789
This tries to improve the accuracy of extract/insert element costs by accounting for subvector extraction/insertion for >128-bit vectors and the shuffling of elements to/from the 0'th index.
It also adds INSERTPS for f32 types and PINSR/PEXTR costs for integer types (at the moment we assume the same cost as MOVD/MOVQ - which isn't always true).
Differential Revision: https://reviews.llvm.org/D74976
Tyker