Summary:
Existing heuristic uses the ratio between the function entry
frequency and the loop invocation frequency to find cold loops. However,
even if the loop executes frequently, if it has a small trip count per
each invocation, vectorization is not beneficial. On the other hand,
even if the loop invocation frequency is much smaller than the function
invocation frequency, if the trip count is high it is still beneficial
to vectorize the loop.
This patch uses estimated trip count computed from the profile metadata
as a primary metric to determine coldness of the loop. If the estimated
trip count cannot be computed, it falls back to the original heuristics.
Reviewers: Ayal, mssimpso, mkuper, danielcdh, wmi, tejohnson
Reviewed By: tejohnson
Subscribers: tejohnson, mzolotukhin, llvm-commits
Differential Revision: https://reviews.llvm.org/D32451
llvm-svn: 305729
The recommit fixes two bugs: The first one is to use CurrentBlock instead of
PREInstr's Parent as param of performScalarPREInsertion because the Parent
of a clone instruction may be uninitialized. The second one is stop PRE when
CurrentBlock to its predecessor is a backedge and an operand of CurInst is
defined inside of CurrentBlock. The same value defined inside of loop in last
iteration can not be regarded as available.
Right now scalarpre doesn't have phi-translate support, so it will miss some
simple pre opportunities. Like the following testcase, current scalarpre cannot
recognize the last "a * b" is fully redundent because a and b used by the last
"a * b" expr are both defined by phis.
long a[100], b[100], g1, g2, g3;
__attribute__((pure)) long goo();
void foo(long a, long b, long c, long d) {
g1 = a * b;
if (__builtin_expect(g2 > 3, 0)) {
a = c;
b = d;
g2 = a * b;
}
g3 = a * b; // fully redundant.
}
The patch adds phi-translate support in scalarpre. This is only a temporary
solution before the newpre based on newgvn is available.
Differential Revision: https://reviews.llvm.org/D32252
llvm-svn: 305578
Summary: Fixes an issue using RegisterStandardPasses from a statically linked object before PassManagerBuilder::addGlobalExtension is called from a dynamic library.
Reviewers: efriedma, theraven
Reviewed By: efriedma
Subscribers: mehdi_amini, mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D33515
llvm-svn: 305303
1. When there is no perfect iteration order, we can't let phi nodes
put themselves in terms of things that come later in the iteration
order, or we will endlessly cycle (the normal RPO algorithm clears the
hashtable to avoid this issue).
2. We are sometimes erasing the wrong expression (causing pessimism)
because our equality says loads and stores are the same.
We introduce an exact equality function and use it when erasing to
make sure we erase only identical expressions, not equivalent ones.
llvm-svn: 304807
I did this a long time ago with a janky python script, but now
clang-format has built-in support for this. I fed clang-format every
line with a #include and let it re-sort things according to the precise
LLVM rules for include ordering baked into clang-format these days.
I've reverted a number of files where the results of sorting includes
isn't healthy. Either places where we have legacy code relying on
particular include ordering (where possible, I'll fix these separately)
or where we have particular formatting around #include lines that
I didn't want to disturb in this patch.
This patch is *entirely* mechanical. If you get merge conflicts or
anything, just ignore the changes in this patch and run clang-format
over your #include lines in the files.
Sorry for any noise here, but it is important to keep these things
stable. I was seeing an increasing number of patches with irrelevant
re-ordering of #include lines because clang-format was used. This patch
at least isolates that churn, makes it easy to skip when resolving
conflicts, and gets us to a clean baseline (again).
llvm-svn: 304787
Summary:
This problem stems from the fact that instructions are allocated using new
in LLVM, i.e. there is no relationship that can be derived by just looking
at the pointer value.
This interface dispatches to appropriate dominance check given 2 instructions,
i.e. in case the instructions are in the same basic block, ordered basicblock
(with instruction numbering and caching) are used. Otherwise, dominator tree
is used.
This is a preparation patch for https://reviews.llvm.org/D32720
Reviewers: dberlin, hfinkel, davide
Subscribers: davide, mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D33380
llvm-svn: 304764
This was rL304226, reverted in 304228 due to a clang assertion failure
on the build bots. That problem should have been addressed by clang
commit rL304470.
llvm-svn: 304488
Replace GVFlags::LiveRoot with GVFlags::Live and use that instead of
all the DeadSymbols sets. This is refactoring in order to make
liveness information available in the RegularLTO pipeline.
llvm-svn: 304466
Summary: Also see D33429 for other ThinLTO + New PM related changes.
Reviewers: davide, chandlerc, tejohnson
Subscribers: mehdi_amini, Prazek, cfe-commits, inglorion, llvm-commits, eraman
Differential Revision: https://reviews.llvm.org/D33525
llvm-svn: 304378
Summary:
In rL302576, DISubprograms gained the constraint that a !dbg attachments to functions must
have a 1:1 mapping to DISubprograms. As part of that change, the function cloning support
was adjusted to attempt to enforce this invariant during cloning. However, there
were several problems with the implementation. Part of these were fixed in rL304079.
However, there was a more fundamental problem with these changes, namely that it
bypasses the matadata value map, causing the cloned metadata to be a mix of metadata
pointing to the new suprogram (where manual code was added to fix those up) and the
old suprogram (where this was not the case). This mismatch could cause a number of
different assertion failures in the DWARF emitter. Some of these are given at
https://github.com/JuliaLang/julia/issues/22069, but some others have been observed
as well. Attempt to rectify this by partially reverting the manual DI metadata fixup,
and instead using the standard value map approach. To retain the desired semantics
of not duplicating the compilation unit and inlined subprograms, explicitly freeze
these in the value map.
Reviewers: dblaikie, aprantl, GorNishanov, echristo
Reviewed By: aprantl
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D33655
llvm-svn: 304226
The recommit is to fix a bug about ExtractValue and InsertValue ops. For those
ops, some varargs inside GVN::Expression are not value numbers but raw index
numbers. It is wrong to do phi-translate for raw index numbers, and the fix is
to stop doing that.
Right now scalarpre doesn't have phi-translate support, so it will miss some
simple pre opportunities. Like the following testcase, current scalarpre cannot
recognize the last "a * b" is fully redundent because a and b used by the last
"a * b" expr are both defined by phis.
long a[100], b[100], g1, g2, g3;
__attribute__((pure)) long goo();
void foo(long a, long b, long c, long d) {
g1 = a * b;
if (__builtin_expect(g2 > 3, 0)) {
a = c;
b = d;
g2 = a * b;
}
g3 = a * b; // fully redundant.
}
The patch adds phi-translate support in scalarpre. This is only a temporary
solution before the newpre based on newgvn is available.
Differential Revision: https://reviews.llvm.org/D32252
llvm-svn: 304050
Right now scalarpre doesn't have phi-translate support, so it will miss some
simple pre opportunities. Like the following testcase, current scalarpre cannot
recognize the last "a * b" is fully redundent because a and b used by the last
"a * b" expr are both defined by phis.
long a[100], b[100], g1, g2, g3;
__attribute__((pure)) long goo();
void foo(long a, long b, long c, long d) {
g1 = a * b;
if (__builtin_expect(g2 > 3, 0)) {
a = c;
b = d;
g2 = a * b;
}
g3 = a * b; // fully redundant.
}
The patch adds phi-translate support in scalarpre. This is only a temporary
solution before the newpre based on newgvn is available.
Differential Revision: https://reviews.llvm.org/D32252
llvm-svn: 303923
This patch provides an initial prototype for a pass that sinks instructions based on GVN information, similar to GVNHoist. It is not yet ready for commiting but I've uploaded it to gather some initial thoughts.
This pass attempts to sink instructions into successors, reducing static
instruction count and enabling if-conversion.
We use a variant of global value numbering to decide what can be sunk.
Consider:
[ %a1 = add i32 %b, 1 ] [ %c1 = add i32 %d, 1 ]
[ %a2 = xor i32 %a1, 1 ] [ %c2 = xor i32 %c1, 1 ]
\ /
[ %e = phi i32 %a2, %c2 ]
[ add i32 %e, 4 ]
GVN would number %a1 and %c1 differently because they compute different
results - the VN of an instruction is a function of its opcode and the
transitive closure of its operands. This is the key property for hoisting
and CSE.
What we want when sinking however is for a numbering that is a function of
the *uses* of an instruction, which allows us to answer the question "if I
replace %a1 with %c1, will it contribute in an equivalent way to all
successive instructions?". The (new) PostValueTable class in GVN provides this
mapping.
This pass has some shown really impressive improvements especially for codesize already on internal benchmarks, so I have high hopes it can replace all the sinking logic in SimplifyCFG.
Differential revision: https://reviews.llvm.org/D24805
llvm-svn: 303850
This patch builds over https://reviews.llvm.org/rL303349 and replaces
the use of the condition only if it is safe to do so.
We should not blindly RAUW the condition if experimental.guard or assume
is a use of that
condition. This is because LVI may have used the guard/assume to
identify the
value of the condition, and RUAWing will fold the guard/assume and uses
before the guards/assumes.
Reviewers: sanjoy, reames, trentxintong, mkazantsev
Reviewed by: sanjoy, reames
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D33257
llvm-svn: 303633
This reapplies commit r303438 modified to not verify cross-imported
bitcode in FunctionImporter.
rdar://problem/31233625
Differential Revision: https://reviews.llvm.org/D33370
llvm-svn: 303470
Refactor the strlen optimization code to work for both strlen and wcslen.
This especially helps with programs in the wild where people pass
L"string"s to const std::wstring& function parameters and the wstring
constructor gets inlined.
This also fixes a lingerind API problem/bug in getConstantStringInfo()
where zeroinitializers would always give you an empty string (without a
length) back regardless of the actual length of the initializer which
did not work well in the TrimAtNul==false causing the PR mentioned
below.
Note that the fixed getConstantStringInfo() needed fixes to SelectionDAG
memcpy lowering and may lead to some cases for out-of-bounds
zeroinitializer accesses not getting optimized anymore. So some code
with UB may produce out of bound memory reads now instead of just
producing zeros.
The refactoring "accidentally" fixes http://llvm.org/PR32124
Differential Revision: https://reviews.llvm.org/D32839
llvm-svn: 303461
This is a complicated bug involving two issues:
1. What do we do with phi nodes when we prove all arguments are not
live?
2. When is it safe to use value leaders to determine if we can ignore
an argumnet?
llvm-svn: 303453
The approach I followed was to emit the remark after getTreeCost concludes
that SLP is profitable. I initially tried emitting them after the
vectorizeRootInstruction calls in vectorizeChainsInBlock but I vaguely
remember missing a few cases for example in HorizontalReduction::tryToReduce.
ORE is placed in BoUpSLP so that it's available from everywhere (notably
HorizontalReduction::tryToReduce).
We use the first instruction in the root bundle as the locator for the remark.
In order to get a sense how far the tree is spanning I've include the size of
the tree in the remark. This is not perfect of course but it gives you at
least a rough idea about the tree. Then you can follow up with -view-slp-tree
to really see the actual tree.
llvm-svn: 302811
This pass uses a new target hook to decide whether or not to expand a particular
intrinsic to the shuffevector sequence.
Differential Revision: https://reviews.llvm.org/D32245
llvm-svn: 302631
This change is required because the notion of count is different for
sample profiling and getProfileCount will need to determine the
underlying profile type.
Differential revision: https://reviews.llvm.org/D33012
llvm-svn: 302597
- This change allows targets to opt-in to using them instead of the log2
shufflevector algorithm.
- The SLP and Loop vectorizers have the common code to do shuffle reductions
factored out into LoopUtils, and now have a unified interface for generating
reductions regardless of the preference of the target. LoopUtils now uses TTI
to determine what kind of reductions the target wants to handle.
- For CodeGen, basic legalization support is added.
Differential Revision: https://reviews.llvm.org/D30086
llvm-svn: 302514
Use a combination of !associated, comdat, @llvm.compiler.used and
custom sections to allow dead stripping of globals and their asan
metadata. Sometimes.
Currently this works on LLD, which supports SHF_LINK_ORDER with
sh_link pointing to the associated section.
This also works on BFD, which seems to treat comdats as
all-or-nothing with respect to linker GC. There is a weird quirk
where the "first" global in each link is never GC-ed because of the
section symbols.
At this moment it does not work on Gold (as in the globals are never
stripped).
This is a second re-land of r298158. This time, this feature is
limited to -fdata-sections builds.
llvm-svn: 301587
Currently, this pass only focuses on *trivial* loop unswitching. At that
reduced problem it remains significantly better than the current loop
unswitch:
- Old pass is worse than cubic complexity. New pass is (I think) linear.
- New pass is much simpler in its design by focusing on full unswitching. (See
below for details on this).
- New pass doesn't carry state for thresholds between pass iterations.
- New pass doesn't carry state for correctness (both miscompile and
infloop) between pass iterations.
- New pass produces substantially better code after unswitching.
- New pass can handle more trivial unswitch cases.
- New pass doesn't recompute the dominator tree for the entire function
and instead incrementally updates it.
I've ported all of the trivial unswitching test cases from the old pass
to the new one to make sure that major functionality isn't lost in the
process. For several of the test cases I've worked to improve the
precision and rigor of the CHECKs, but for many I've just updated them
to handle the new IR produced.
My initial motivation was the fact that the old pass carried state in
very unreliable ways between pass iterations, and these mechansims were
incompatible with the new pass manager. However, I discovered many more
improvements to make along the way.
This pass makes two very significant assumptions that enable most of these
improvements:
1) Focus on *full* unswitching -- that is, completely removing whatever
control flow construct is being unswitched from the loop. In the case
of trivial unswitching, this means removing the trivial (exiting)
edge. In non-trivial unswitching, this means removing the branch or
switch itself. This is in opposition to *partial* unswitching where
some part of the unswitched control flow remains in the loop. Partial
unswitching only really applies to switches and to folded branches.
These are very similar to full unrolling and partial unrolling. The
full form is an effective canonicalization, the partial form needs
a complex cost model, cannot be iterated, isn't canonicalizing, and
should be a separate pass that runs very late (much like unrolling).
2) Leverage LLVM's Loop machinery to the fullest. The original unswitch
dates from a time when a great deal of LLVM's loop infrastructure was
missing, ineffective, and/or unreliable. As a consequence, a lot of
complexity was added which we no longer need.
With these two overarching principles, I think we can build a fast and
effective unswitcher that fits in well in the new PM and in the
canonicalization pipeline. Some of the remaining functionality around
partial unswitching may not be relevant today (not many test cases or
benchmarks I can find) but if they are I'd like to add support for them
as a separate layer that runs very late in the pipeline.
Purely to make reviewing and introducing this code more manageable, I've
split this into first a trivial-unswitch-only pass and in the next patch
I'll add support for full non-trivial unswitching against a *fixed*
threshold, exactly like full unrolling. I even plan to re-use the
unrolling thresholds, as these are incredibly similar cost tradeoffs:
we're cloning a loop body in order to end up with simplified control
flow. We should only do that when the total growth is reasonably small.
One of the biggest changes with this pass compared to the previous one
is that previously, each individual trivial exiting edge from a switch
was unswitched separately as a branch. Now, we unswitch the entire
switch at once, with cases going to the various destinations. This lets
us unswitch multiple exiting edges in a single operation and also avoids
numerous extremely bad behaviors, where we would introduce 1000s of
branches to test for thousands of possible values, all of which would
take the exact same exit path bypassing the loop. Now we will use
a switch with 1000s of cases that can be efficiently lowered into
a jumptable. This avoids relying on somehow forming a switch out of the
branches or getting horrible code if that fails for any reason.
Another significant change is that this pass actively updates the CFG
based on unswitching. For trivial unswitching, this is actually very
easy because of the definition of loop simplified form. Doing this makes
the code coming out of loop unswitch dramatically more friendly. We
still should run loop-simplifycfg (at the least) after this to clean up,
but it will have to do a lot less work.
Finally, this pass makes much fewer attempts to simplify instructions
based on the unswitch. Something like loop-instsimplify, instcombine, or
GVN can be used to do increasingly powerful simplifications based on the
now dominating predicate. The old simplifications are things that
something like loop-instsimplify should get today or a very, very basic
loop-instcombine could get. Keeping that logic separate is a big
simplifying technique.
Most of the code in this pass that isn't in the old one has to do with
achieving specific goals:
- Updating the dominator tree as we go
- Unswitching all cases in a switch in a single step.
I think it is still shorter than just the trivial unswitching code in
the old pass despite having this functionality.
Differential Revision: https://reviews.llvm.org/D32409
llvm-svn: 301576
Commits were:
"Use WeakVH instead of WeakTrackingVH in AliasSetTracker's UnkownInsts"
"Add a new WeakVH value handle; NFC"
"Rename WeakVH to WeakTrackingVH; NFC"
The changes assumed pointers are 8 byte aligned on all architectures.
llvm-svn: 301429
Summary:
I plan to use WeakVH to mean "nulls itself out on deletion, but does
not track RAUW" in a subsequent commit.
Reviewers: dblaikie, davide
Reviewed By: davide
Subscribers: arsenm, mehdi_amini, mcrosier, mzolotukhin, jfb, llvm-commits, nhaehnle
Differential Revision: https://reviews.llvm.org/D32266
llvm-svn: 301424
