This should really, really fix PR13916. For real this time. The
underlying bug is... a bit more subtle than I had imagined.
The setup is a code pattern that leads to an @llvm.memcpy call with two
equal pointers to an alloca in the source and dest. Now, not any pattern
will do. The alloca needs to be formed just so, and both pointers should
be wrapped in different bitcasts etc. When this precise pattern hits,
a funny sequence of events transpires. First, we correctly detect the
potential for overlap, and correctly optimize the memcpy. The first
time. However, we do simplify the set of users of the alloca, and that
causes us to run the alloca back through the SROA pass in case there are
knock-on simplifications. At this point, a curious thing has happened.
If we happen to have an i8 alloca, we have direct i8 pointer values. So
we don't bother creating a cast, we rewrite the arguments to the memcpy
to dircetly refer to the alloca.
Now, in an unrelated area of the pass, we have clever logic which
ensures that when visiting each User of a particular pointer derived
from an alloca, we only visit that User once, and directly inspect all
of its operands which refer to that particular pointer value. However,
the mechanism used to detect memcpy's with the potential to overlap
relied upon getting visited once per *Use*, not once per *User*. This is
always true *unless* the same exact value is both source and dest. It
turns out that almost nothing actually produces that pattern though.
We can hand craft test cases that more directly test this behavior of
course, and those are included. Also, note that there is a significant
missed optimization here -- we prove in many cases that there is
a non-volatile memcpy call with identical source and dest addresses. We
shouldn't prevent splitting the alloca in that case, and in fact we
should just remove such memcpy calls eagerly. I'll address that in
a subsequent commit.
llvm-svn: 164669
only a missed optimization opportunity if the store is over-aligned, but a
miscompile if the store's new type has a higher natural alignment than the
memcpy did. Fixes PR13920!
llvm-svn: 164641
reason we were getting two of the same alloca is because of a memmove/memcpy
which had the same alloca in both the src and dest. Now we detect that case
directly. This has the same testcase as before, but fixes a clang test
CodeGenObjC/exceptions.m which runs clang -O2.
llvm-svn: 164636
Chandler, it's not obvious that it's okay that this alloca gets into the list
twice to begin with. Please review and see whether this is the fix you really
want, but I wanted to get a fix checked in quickly.
llvm-svn: 164634
to chains or cycles between PHIs and/or selects. Also add a couple of
really nice test cases reduced from Kostya's reports in PR13905 and
PR13906. Both are fixed by this patch.
llvm-svn: 164596
David (I think), but I would appreciate folks verifying that this fixes
the big crasher.
I'm still working on a reduced test case, but because this was causing
problems I wanted to get the fix checked in quickly.
llvm-svn: 164585
integer promotion analogous to vector promotion. When there is an
integer alloca being accessed both as its integer type and as a narrower
integer type, promote the narrower access to "insert" and "extract" the
smaller integer from the larger one, and make the integer alloca
a candidate for promotion.
In the new formulation, we don't care about target legal integer or use
thresholds to control things. Instead, we only perform this promotion to
an integer type which the frontend has already emitted a load or store
for. This bounds the scope and prevents optimization passes from
coalescing larger and larger entities into a single integer.
llvm-svn: 164479
across the uses of the alloca. It's entirely possible for negative
numbers to come up here, and in some rare cases simply doing the 2's
complement arithmetic isn't the correct decision. Notably, we can't zext
the index of the GEP. The definition of GEP is that these offsets are
sign extended or truncated to the size of the pointer, and then wrapping
2's complement arithmetic used.
This patch fixes an issue that comes up with *no* input from the
buildbots or bootstrap afaict. The only place where it manifested,
disturbingly, is Clang's own regression test suite. A reduced and
targeted collection of tests are added to cope with this. Note that I've
tried to pin down the potential cases of overflow, but may have missed
some cases. I've tried to add a few cases to test this, but its hard
because LLVM has quite limited support for >64bit constructs.
llvm-svn: 164475
selects with a constant condition. This resulted in the operands
remaining live through the SROA rewriter. Most of the time, this just
caused some dead allocas to persist and get zapped by later passes, but
in one case found by Joerg, it caused a crash when we tried to *promote*
the alloca despite it having this dead use. We already have the
mechanisms in place to handle this, just wire select up to them.
llvm-svn: 164427
This is a follow-up from r163302, which added a transformation to
SimplifyCFG that turns some switches into loads from lookup tables.
It was pointed out that some targets, such as GPUs and deeply embedded
targets, might not find this appropriate, but SimplifyCFG doesn't have
enough information about the target to decide this.
This patch adds the reverse transformation to CodeGenPrep: it turns
loads from lookup tables back into switches for targets where we do not
build jump tables (assuming these are also the targets where lookup
tables are inappropriate).
Hopefully we will eventually get to have target information in
SimplifyCFG, and then this CodeGenPrep transformation can be removed.
llvm-svn: 164206
from the dragonegg build bots when we turned on the full version of the
pass. Included a much reduced test case for this pesky bug, despite
bugpoint's uncooperative behavior.
Also, I audited all the similar code I could find and didn't spot any
other cases where this mistake cropped up.
llvm-svn: 164178
working on FCA splitting. Instead of refusing to form a common type when
there are uses of a subsection of the alloca as well as a use of the
entire alloca, just skip the subsection uses and continue looking for
a whole-alloca use with a type that we can use.
This produces slightly prettier IR I think, and also fixes the other
failure in the test.
llvm-svn: 164146
FCAs. This is essential in order to promote allocas that are used in
struct returns by frontends like Clang. The FCA load would block the
rest of the pass from firing, resulting is significant regressions with
the bullet benchmark in the nightly test suite.
Thanks to Duncan for repeated discussions about how best to do this, and
to both him and Benjamin for review.
This appears to have blocked many places where the pass tries to fire,
and so I'm expect somewhat different results with this fix added.
As with the last big patch, I'm including a change to enable the SROA by
default *temporarily*. Ben is going to remove this as soon as the LNT
bots pick up the patch. I'm just trying to get a round of LNT numbers
from the stable machines in the lab.
NOTE: Four clang tests are expected to fail in the brief window where
this is enabled. Sorry for the noise!
llvm-svn: 164119
partition use lists a bit. No functionality changed.
These visitors are actually visiting a tuple of a Use and an offset into
the alloca. However, we use the InstVisitor to handle the dispatch over
the users, and so the Use and Offset are stored in class member
variables and set just before each call to visit(). This is fairly
awkward and makes the functions a bit harder to read, but its the only
real option we have until InstVisitor can be rewritten to use variadic
templates.
However, this pattern shouldn't be followed on the helper member
functions where there is no interface constraint from the visitor. We
already were passing the instruction as a normal parameter rather than
use the Use to get at it, start passing the offset as well. This will
become more important in subsequent patches as the offset will in some
cases change while visiting a single instruction.
llvm-svn: 164003
new one, and add support for running the new pass in that mode and in
that slot of the pass manager. With this the new pass can completely
replace the old one within the pipeline.
The strategy for enabling or disabling the SSAUpdater logic is to do it
by making the requirement of the domtree analysis optional. By default,
it is required and we get the standard mem2reg approach. This is usually
the desired strategy when run in stand-alone situations. Within the
CGSCC pass manager, we disable requiring of the domtree analysis and
consequentially trigger fallback to the SSAUpdater promotion.
In theory this would allow the pass to re-use a domtree if one happened
to be available even when run in a mode that doesn't require it. In
practice, it lets us have a single pass rather than two which was
simpler for me to wrap my head around.
There is a hidden flag to force the use of the SSAUpdater code path for
the purpose of testing. The primary testing strategy is just to run the
existing tests through that path. One notable difference is that it has
custom code to handle lifetime markers, and one of the tests has been
enhanced to exercise that code.
This has survived a bootstrap and the test suite without serious
correctness issues, however my run of the test suite produced *very*
alarming performance numbers. I don't entirely understand or trust them
though, so more investigation is on-going.
To aid my understanding of the performance impact of the new SROA now
that it runs throughout the optimization pipeline, I'm enabling it by
default in this commit, and will disable it again once the LNT bots have
picked up one iteration with it. I want to get those bots (which are
much more stable) to evaluate the impact of the change before I jump to
any conclusions.
NOTE: Several Clang tests will fail because they run -O3 and check the
result's order of output. They'll go back to passing once I disable it
again.
llvm-svn: 163965
* wrap code blocks in \code ... \endcode;
* refer to parameter names in paragraphs correctly (\arg is not what most
people want -- it starts a new paragraph);
* use \param instead of \arg to document parameters in order to be consistent
with the rest of the codebase.
llvm-svn: 163902
pointless checks in here, bad asserts, and just confusing code. I've
also added a bit more to the comment to clarify what this function is
really trying to do as it was not obvious to Duncan when studying it.
Thanks to Duncan for helping me dig through the issue.
No real functionality changed here in practical cases, and certainly no
test case. This is just cleanup spotted by inspection.
llvm-svn: 163897
inspection by Duncan during review. My suspicion is that we would still
have returned 0 anyways in this case, but doing it sooner is better.
llvm-svn: 163895
deeply suspicious and likely to go away eventually. Also fix a bogus
comment about one of the checks in the vector GEP analysis. Based on
review from Duncan.
llvm-svn: 163894
Originally I had anticipated needing to thread this through more bits of
the SROA pass itself, but that ended up not happening. In the end, this
is a much simpler way to manange the variable.
llvm-svn: 163893
This is essentially a ground up re-think of the SROA pass in LLVM. It
was initially inspired by a few problems with the existing pass:
- It is subject to the bane of my existence in optimizations: arbitrary
thresholds.
- It is overly conservative about which constructs can be split and
promoted.
- The vector value replacement aspect is separated from the splitting
logic, missing many opportunities where splitting and vector value
formation can work together.
- The splitting is entirely based around the underlying type of the
alloca, despite this type often having little to do with the reality
of how that memory is used. This is especially prevelant with unions
and base classes where we tail-pack derived members.
- When splitting fails (often due to the thresholds), the vector value
replacement (again because it is separate) can kick in for
preposterous cases where we simply should have split the value. This
results in forming i1024 and i2048 integer "bit vectors" that
tremendously slow down subsequnet IR optimizations (due to large
APInts) and impede the backend's lowering.
The new design takes an approach that fundamentally is not susceptible
to many of these problems. It is the result of a discusison between
myself and Duncan Sands over IRC about how to premptively avoid these
types of problems and how to do SROA in a more principled way. Since
then, it has evolved and grown, but this remains an important aspect: it
fixes real world problems with the SROA process today.
First, the transform of SROA actually has little to do with replacement.
It has more to do with splitting. The goal is to take an aggregate
alloca and form a composition of scalar allocas which can replace it and
will be most suitable to the eventual replacement by scalar SSA values.
The actual replacement is performed by mem2reg (and in the future
SSAUpdater).
The splitting is divided into four phases. The first phase is an
analysis of the uses of the alloca. This phase recursively walks uses,
building up a dense datastructure representing the ranges of the
alloca's memory actually used and checking for uses which inhibit any
aspects of the transform such as the escape of a pointer.
Once we have a mapping of the ranges of the alloca used by individual
operations, we compute a partitioning of the used ranges. Some uses are
inherently splittable (such as memcpy and memset), while scalar uses are
not splittable. The goal is to build a partitioning that has the minimum
number of splits while placing each unsplittable use in its own
partition. Overlapping unsplittable uses belong to the same partition.
This is the target split of the aggregate alloca, and it maximizes the
number of scalar accesses which become accesses to their own alloca and
candidates for promotion.
Third, we re-walk the uses of the alloca and assign each specific memory
access to all the partitions touched so that we have dense use-lists for
each partition.
Finally, we build a new, smaller alloca for each partition and rewrite
each use of that partition to use the new alloca. During this phase the
pass will also work very hard to transform uses of an alloca into a form
suitable for promotion, including forming vector operations, speculating
loads throguh PHI nodes and selects, etc.
After splitting is complete, each newly refined alloca that is
a candidate for promotion to a scalar SSA value is run through mem2reg.
There are lots of reasonably detailed comments in the source code about
the design and algorithms, and I'm going to be trying to improve them in
subsequent commits to ensure this is well documented, as the new pass is
in many ways more complex than the old one.
Some of this is still a WIP, but the current state is reasonbly stable.
It has passed bootstrap, the nightly test suite, and Duncan has run it
successfully through the ACATS and DragonEgg test suites. That said, it
remains behind a default-off flag until the last few pieces are in
place, and full testing can be done.
Specific areas I'm looking at next:
- Improved comments and some code cleanup from reviews.
- SSAUpdater and enabling this pass inside the CGSCC pass manager.
- Some datastructure tuning and compile-time measurements.
- More aggressive FCA splitting and vector formation.
Many thanks to Duncan Sands for the thorough final review, as well as
Benjamin Kramer for lots of review during the process of writing this
pass, and Daniel Berlin for reviewing the data structures and algorithms
and general theory of the pass. Also, several other people on IRC, over
lunch tables, etc for lots of feedback and advice.
llvm-svn: 163883
* wrap code blocks in \code ... \endcode;
* refer to parameter names in paragraphs correctly (\arg is not what most
people want -- it starts a new paragraph).
llvm-svn: 163790
pointers-to-strong-pointers may be in play. These can lead to retains and
releases happening in unstructured ways, foiling the optimizer. This fixes
rdar://12150909.
llvm-svn: 163180
- CodeGenPrepare pass for identifying div/rem ops
- Backend specifies the type mapping using addBypassSlowDivType
- Enabled only for Intel Atom with O2 32-bit -> 8-bit
- Replace IDIV with instructions which test its value and use DIVB if the value
is positive and less than 256.
- In the case when the quotient and remainder of a divide are used a DIV
and a REM instruction will be present in the IR. In the non-Atom case
they are both lowered to IDIVs and CSE removes the redundant IDIV instruction,
using the quotient and remainder from the first IDIV. However,
due to this optimization CSE is not able to eliminate redundant
IDIV instructions because they are located in different basic blocks.
This is overcome by calculating both the quotient (DIV) and remainder (REM)
in each basic block that is inserted by the optimization and reusing the result
values when a subsequent DIV or REM instruction uses the same operands.
- Test cases check for the presents of the optimization when calculating
either the quotient, remainder, or both.
Patch by Tyler Nowicki!
llvm-svn: 163150
Scan the body of the loop and find instructions that may trap.
Use this information when deciding if it is safe to hoist or sink instructions.
Notice that we can optimize the search of instructions that may throw in the case of nested loops.
rdar://11518836
llvm-svn: 163132
For example, the ARM target does not have efficient ISel handling for vector
selects with scalar conditions. This patch adds a TLI hook which allows the
different targets to report which selects are supported well and which selects
should be converted to CF duting codegen prepare.
llvm-svn: 163093
We update until we hit a fixpoint. This is probably slow but also
slightly simplifies the code. It should also fix the occasional
invalid domtrees observed when building with expensive checking.
I couldn't find a case where this had a measurable slowdown, but
if someone finds a pathological case where it does we may have
to find a cleverer way of updating dominators here.
Thanks to Duncan for the test case.
llvm-svn: 163091
The old PHI updating code in loop-rotate was replaced with SSAUpdater a while
ago, it has no problems with comples PHIs. What had to be fixed is detecting
whether a loop was already rotated and updating dominators when multiple exits
were present.
This change increases overall code size a bit, mostly due to additional loop
unrolling opportunities. Passes test-suite and selfhost with -verify-dom-info.
Fixes PR7447.
Thanks to Andy for the input on the domtree updating code.
llvm-svn: 162912
This disables malloc-specific optimization when -fno-builtin (or -ffreestanding)
is specified. This has been a problem for a long time but became more severe
with the recent memory builtin improvements.
Since the memory builtin functions are used everywhere, this required passing
TLI in many places. This means that functions that now have an optional TLI
argument, like RecursivelyDeleteTriviallyDeadFunctions, won't remove dead
mallocs anymore if the TLI argument is missing. I've updated most passes to do
the right thing.
Fixes PR13694 and probably others.
llvm-svn: 162841
No intended behavior change. This was introduced in r162023. With the fixed
algorithm a Release build of ARMInstPrinter.cpp goes from 16s to 10s on a
2011 MBP.
llvm-svn: 162559
optimizations are guarded by the -enable-double-float-shrink LLVM option.
Last bit of PR13574. Patch by Weiming Zhao <weimingz@codeaurora.org>.
llvm-svn: 162368
This optimization is really just replacing allocas wholesale with
globals, there is no scalarization.
The underlying motivation for this patch is to simplify the SROA pass
and focus it on splitting and promoting allocas.
llvm-svn: 162271
where some fact lake a=b dominates a use in a phi, but doesn't dominate the
basic block itself.
This feature could also be implemented by splitting critical edges, but at least
with the current algorithm reasoning about the dominance directly is faster.
The time for running "opt -O2" in the testcase in pr10584 is 1.003 times slower
and on gcc as a single file it is 1.0007 times faster.
llvm-svn: 162023
and allow some optimizations to turn conditional branches into unconditional.
This commit adds a simple control-flow optimization which merges two consecutive
basic blocks which are connected by a single edge. This allows the codegen to
operate on larger basic blocks.
rdar://11973998
llvm-svn: 161852
multiple scalar promotions on a single loop. This also has the effect of
preserving the order of stores sunk out of loops, which is aesthetically
pleasing, and it happens to fix the testcase in PR13542, though it doesn't
fix the underlying problem.
llvm-svn: 161459
GetBestDestForJumpOnUndef() assumes there is at least 1 successor, which isn't
true if the block ends in an indirect branch with no successors. Fix this by
bailing out earlier in this case.
llvm-svn: 160546
Fixes PR13371: indvars pass incorrectly substitutes 'undef' values.
I do not like this fix. It's needed until/unless the meaning of undef
changes. It attempts to be complete according to the IR spec, but I
don't have much confidence in the implementation given the difficulty
testing undefined behavior. Worse, this invalidates some of my
hard-fought work on indvars and LSR to optimize pointer induction
variables. It results benchmark regressions, which I'll track
internally. On x86_64 no LTO I see:
-3% huffbench
-3% 400.perlbench
-8% fhourstones
My only suggestion for recovering is to change the meaning of
undef. If we could trust an arbitrary instruction to produce a some
real value that can be manipulated (e.g. incremented) according to
non-undef rules, then this case could be easily handled with SCEV.
llvm-svn: 160421
This places limits on CollectSubexprs to constrains the number of
reassociation possibilities. It limits the recursion depth and skips
over chains of nested recurrences outside the current loop.
Fixes PR13361. Although underlying SCEV behavior is still potentially bad.
llvm-svn: 160340
All SCEV expressions used by LSR formulae must be safe to
expand. i.e. they may not contain UDiv unless we can prove nonzero
denominator.
Fixes PR11356: LSR hoists UDiv.
llvm-svn: 160205
This was always part of the VMCore library out of necessity -- it deals
entirely in the IR. The .cpp file in fact was already part of the VMCore
library. This is just a mechanical move.
I've tried to go through and re-apply the coding standard's preferred
header sort, but at 40-ish files, I may have gotten some wrong. Please
let me know if so.
I'll be committing the corresponding updates to Clang and Polly, and
Duncan has DragonEgg.
Thanks to Bill and Eric for giving the green light for this bit of cleanup.
llvm-svn: 159421
include/llvm/Analysis/DebugInfo.h to include/llvm/DebugInfo.h.
The reasoning is because the DebugInfo module is simply an interface to the
debug info MDNodes and has nothing to do with analysis.
llvm-svn: 159312
before the expression root. Any existing operators that are changed to use one
of them needs to be moved between it and the expression root, and recursively
for the operators using that one. When I rewrote RewriteExprTree I accidentally
inverted the logic, resulting in the compacting going down from operators to
operands rather than up from operands to the operators using them, oops. Fix
this, resolving PR12963.
llvm-svn: 159265
- simplifycfg: invoke undef/null -> unreachable
- instcombine: invoke new -> invoke expect(0, 0) (an arbitrary NOOP intrinsic; only done if the allocated memory is unused, of course)
- verifier: allow invoke of intrinsics (to make the previous step work)
llvm-svn: 159146
- provide more extensive set of functions to detect library allocation functions (e.g., malloc, calloc, strdup, etc)
- provide an API to compute the size and offset of an object pointed by
Move a few clients (GVN, AA, instcombine, ...) to the new API.
This implementation is a lot more aggressive than each of the custom implementations being replaced.
Patch reviewed by Nick Lewycky and Chandler Carruth, thanks.
llvm-svn: 158919
Dynamic GEPs created by SROA needed to insert extra "i32 0"
operands to index through structs and arrays to get to the
vector being indexed.
llvm-svn: 158590
For non-address users, Base and Scaled registers are not specially
associated to fit an address mode, so SCEVExpander should apply normal
expansion rules. Otherwise we may sink computation into inner loops
that have already been optimized.
llvm-svn: 158537
example degenerate phi nodes and binops that use themselves in unreachable code.
Thanks to Charles Davis for the testcase that uncovered this can of worms.
llvm-svn: 158508
since then the entire expression must equal zero (similarly for other operations
with an absorbing element). With this in place a bunch of reassociate code for
handling constants is dead since it is all taken care of when linearizing. No
intended functionality change.
llvm-svn: 158398
POD type, causing memory corruption when mapping to APInts with bitwidth > 64.
Merge another crash testcase into crash.ll while there.
llvm-svn: 158369
topologies, it is quite possible for a leaf node to have huge multiplicity, for
example: x0 = x*x, x1 = x0*x0, x2 = x1*x1, ... rapidly gives a value which is x
raised to a vast power (the multiplicity, or weight, of x). This patch fixes
the computation of weights by correctly computing them no matter how big they
are, rather than just overflowing and getting a wrong value. It turns out that
the weight for a value never needs more bits to represent than the value itself,
so it is enough to represent weights as APInts of the same bitwidth and do the
right overflow-avoiding dance steps when computing weights. As a side-effect it
reduces the number of multiplies needed in some cases of large powers. While
there, in view of external uses (eg by the vectorizer) I made LinearizeExprTree
static, pushing the rank computation out into users. This is progress towards
fixing PR13021.
llvm-svn: 158358
problem was that by moving instructions around inside the function, the pass
could accidentally move the iterator being used to advance over the function
too. Fix this by only processing the instruction equal to the iterator, and
leaving processing of instructions that might not be equal to the iterator
to later (later = after traversing the basic block; it could also wait until
after traversing the entire function, but this might make the sets quite big).
Original commit message:
Grab-bag of reassociate tweaks. Unify handling of dead instructions and
instructions to reoptimize. Exploit this to more systematically eliminate
dead instructions (this isn't very useful in practice but is convenient for
analysing some testcase I am working on). No need for WeakVH any more: use
an AssertingVH instead.
llvm-svn: 158226
can move instructions within the instruction list. If the instruction just
happens to be the one the basic block iterator is pointing to, and it is
moved to a different basic block, then we get into an infinite loop due to
the iterator running off the end of the basic block (for some reason this
doesn't fire any assertions). Original commit message:
Grab-bag of reassociate tweaks. Unify handling of dead instructions and
instructions to reoptimize. Exploit this to more systematically eliminate
dead instructions (this isn't very useful in practice but is convenient for
analysing some testcase I am working on). No need for WeakVH any more: use
an AssertingVH instead.
llvm-svn: 158199
instructions to reoptimize. Exploit this to more systematically eliminate
dead instructions (this isn't very useful in practice but is convenient for
analysing some testcase I am working on). No need for WeakVH any more: use
an AssertingVH instead.
llvm-svn: 158073
replacement to make it at least as generic as the instruction being replaced.
This includes:
* dropping nsw/nuw flags
* getting the least restrictive tbaa and fpmath metadata
* merging ranges
Fixes PR12979.
llvm-svn: 157958
- compute size & offset at the same time. The side-effects of this are that we now support negative GEPs. It's now approaching a phase that it can be reused by other passes (e.g., lowering of the objectsize intrinsic)
- use APInt throughout to handle wrap-arounds
- add support for PHI instrumentation
- add a cache (required for recursive PHIs anyway)
- remove hoisting support for now, since it was wrong in a few cases
sorry for the churn here.. tests will follow soon.
llvm-svn: 157775
- hoist checks out of loops where SCEV is smart enough
- add additional statistics to measure how much we loose for not supporting interprocedural and pointers loaded from memory
llvm-svn: 157649
making it stronger and more sane.
Delete the code from tblgen that produced the old code.
Besides being a path forward in intrinsic sanity, this also eliminates a bunch of
machine generated code that was compiled into Function.o
llvm-svn: 157545
then it doesn't alter the instructions composing it, however it would continue
to move the instructions to just before the expression root. Ensure it doesn't
move them either, so now it really does nothing if there is nothing to do. That
commit also ensured that nsw etc flags weren't cleared if the expression was not
being changed. Tweak this a bit so that it doesn't clear flags on the initial
part of a computation either if that part didn't change but later bits did.
llvm-svn: 157518
with arbitrary topologies (previously it would give up when hitting a diamond
in the use graph for example). The testcase from PR12764 is now reduced from
a pile of additions to the optimal 1617*%x0+208. In doing this I changed the
previous strategy of dropping all uses for expression leaves to one of dropping
all but one use. This works out more neatly (but required a bunch of tweaks)
and is also safer: some recently fixed bugs during recursive linearization were
because the linearization code thinks it completely owns a node if it has no uses
outside the expression it is linearizing. But if the node was also in another
expression that had been linearized (and thus all uses of the node from that
expression dropped) then the conclusion that it is completely owned by the
expression currently being linearized is wrong. Keeping one use from within each
linearized expression avoids this kind of mistake.
llvm-svn: 157467
leader table. That's because it wasn't expecting instructions to turn up as
leader for a value number that is not its own, but equality propagation could
create this situation. One solution is to have the leader table use a WeakVH
but this slows down GVN by about 5%. Instead just have equality propagation not
add instructions to the leader table, only constants and arguments. In theory
this might cause GVN to run more (each time it changes something it runs again)
but it doesn't seem to occur enough to cause a slow down.
llvm-svn: 157251
so that it can be reused in MemCpyOptimizer. This analysis is needed to remove
an unnecessary memcpy when returning a struct into a local variable.
rdar://11341081
PR12686
llvm-svn: 156776
replace the operands of expressions with only one use with undef and generate
a new expression for the original without using RAUW to update the original.
Thus any copies of the original expression held in a vector may end up
referring to some bogus value - and using a ValueHandle won't help since there
is no RAUW. There is already a mechanism for getting the effect of recursion
non-recursively: adding the value to be recursed on to RedoInsts. But it wasn't
being used systematically. Have various places where recursion had snuck in at
some point use the RedoInsts mechanism instead. Fixes PR12169.
llvm-svn: 156379
The primitive conservative heuristic seems to give a slight overall
improvement while not regressing stuff. Make it available to wider
testing. If you notice any speed regressions (or significant code
size regressions) let me know!
llvm-svn: 156258
This came up when a change in block placement formed a cmov and slowed down a
hot loop by 50%:
ucomisd (%rdi), %xmm0
cmovbel %edx, %esi
cmov is a really bad choice in this context because it doesn't get branch
prediction. If we emit it as a branch, an out-of-order CPU can do a better job
(if the branch is predicted right) and avoid waiting for the slow load+compare
instruction to finish. Of course it won't help if the branch is unpredictable,
but those are really rare in practice.
This patch uses a dumb conservative heuristic, it turns all cmovs that have one
use and a direct memory operand into branches. cmovs usually save some code
size, so we disable the transform in -Os mode. In-Order architectures are
unlikely to benefit as well, those are included in the
"predictableSelectIsExpensive" flag.
It would be better to reuse branch probability info here, but BPI doesn't
support select instructions currently. It would make sense to use the same
heuristics as the if-converter pass, which does the opposite direction of this
transform.
Test suite shows a small improvement here and there on corei7-level machines,
but the actual results depend a lot on the used microarchitecture. The
transformation is currently disabled by default and available by passing the
-enable-cgp-select2branch flag to the code generator.
Thanks to Chandler for the initial test case to him and Evan Cheng for providing
me with comments and test-suite numbers that were more stable than mine :)
llvm-svn: 156234
minor behavior changes with this, but nothing I have seen evidence of in
the wild or expect to be meaningful. The real goal is unifying our logic
and simplifying the interfaces. A summary of the changes follows:
- Make 'callIsSmall' actually accept a callsite so it can handle
intrinsics, and simplify callers appropriately.
- Nuke a completely bogus declaration of 'callIsSmall' that was still
lurking in InlineCost.h... No idea how this got missed.
- Teach the 'isInstructionFree' about the various more intelligent
'free' heuristics that got added to the inline cost analysis during
review and testing. This mostly surrounds int->ptr and ptr->int casts.
- Switch most of the interesting parts of the inline cost analysis that
were essentially computing 'is this instruction free?' to use the code
metrics routine instead. This way we won't keep duplicating logic.
All of this is motivated by the desire to allow other passes to compute
a roughly equivalent 'cost' metric for a particular basic block as the
inline cost analysis. Sadly, re-using the same analysis for both is
really messy because only the actual inline cost analysis is ever going
to go to the contortions required for simplification, SROA analysis,
etc.
llvm-svn: 156140
Allow the "SplitCriticalEdge" function to split the edge to a landing pad. If
the pass is *sure* that it thinks it knows what it's doing, then it may go ahead
and specify that the landing pad can have its critical edge split. The loop
unswitch pass is one of these passes. It will split the critical edges of all
edges coming from a loop to a landing pad not within the loop. Doing so will
retain important loop analysis information, such as loop simplify.
llvm-svn: 155817
The required checks are moved to ChainInstruction() itself and the
policy decisions are moved to IVChain::isProfitableInc().
Also cache the ExprBase in IVChain to avoid frequent recomputations.
No functional change intended.
llvm-svn: 155676
elements to minimize the number of multiplies required to compute the
final result. This uses a heuristic to attempt to form near-optimal
binary exponentiation-style multiply chains. While there are some cases
it misses, it seems to at least a decent job on a very diverse range of
inputs.
Initial benchmarks show no interesting regressions, and an 8%
improvement on SPASS. Let me know if any other interesting results (in
either direction) crop up!
Credit to Richard Smith for the core algorithm, and helping code the
patch itself.
llvm-svn: 155616
If the loop contains invoke instructions, whose unwind edge escapes the loop,
then don't try to unswitch the loop. Doing so may cause the unwind edge to be
split, which not only is non-trivial but doesn't preserve loop simplify
information.
Fixes PR12573
llvm-svn: 154987
This introduces a threshold of 200 IV Users, which is very
conservative but should be sufficient to avoid serious compile time
sink or stack overflow. The llvm test-suite with LTO never exceeds 190
users per loop.
The bug doesn't relate to a specific type of loop. Checking in an
arbitrary giant loop as a unit test would be silly.
Fixes rdar://11262507.
llvm-svn: 154983
also fix SimplifyLibCalls to use TLI rather than compile-time conditionals to enable optimizations on floor, ceil, round, rint, and nearbyint
llvm-svn: 154960