Trivial multiplication by zero may survive the worklist. We tried to
reassociate the multiplication with a division instruction, causing us
to divide by zero; bail out instead.
This fixes PR24726.
llvm-svn: 246939
This adds a basic cost model for interleaved-access vectorization (and a better
default for shuffles), and enables interleaved-access vectorization by default.
The relevant difference from the default cost model for interleaved-access
vectorization, is that on PPC, the shuffles that end up being used are *much*
cheaper than modeling the process with insert/extract pairs (which are
quite expensive, especially on older cores).
llvm-svn: 246824
On the A2, with an eye toward QPX unaligned-load merging, we should always use
aggressive interleaving. It is generally superior to only using concatenation
unrolling.
llvm-svn: 246819
Summary:
This function was not taking into account that the
input type could be a vector, and wasn't properly
working for vector types.
This caused an assert when building spec2k6 perlbmk for armv8.
Reviewers: rengolin, mzolotukhin
Subscribers: silviu.baranga, mzolotukhin, rengolin, eugenis, jmolloy, aemerson, llvm-commits
Differential Revision: http://reviews.llvm.org/D12559
llvm-svn: 246759
Summary:
Add a `cleanupendpad` instruction, used to mark exceptional exits out of
cleanups (for languages/targets that can abort a cleanup with another
exception). The `cleanupendpad` instruction is similar to the `catchendpad`
instruction in that it is an EH pad which is the target of unwind edges in
the handler and which itself has an unwind edge to the next EH action.
The `cleanupendpad` instruction, similar to `cleanupret` has a `cleanuppad`
argument indicating which cleanup it exits. The unwind successors of a
`cleanuppad`'s `cleanupendpad`s must agree with each other and with its
`cleanupret`s.
Update WinEHPrepare (and docs/tests) to accomodate `cleanupendpad`.
Reviewers: rnk, andrew.w.kaylor, majnemer
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12433
llvm-svn: 246751
There was infinite loop because it was trying to change assume(true) into
assume(true)
Also added handling when assume(false) appear
http://reviews.llvm.org/D12516
llvm-svn: 246697
After hitting @llvm.assume(X) we can:
- propagate equality that X == true
- if X is icmp/fcmp (with eq operation), and one of operand
is constant we can change all variables with constants in the same BasicBlock
http://reviews.llvm.org/D11918
llvm-svn: 246695
We were bailing to two places if our runtime checks failed. If the initial overflow check failed, we'd go to ScalarPH. If any other check failed, we'd go to MiddleBlock. This caused us to have to have an extra PHI per induction and reduction as the vector loop's exit block was not dominated by its latch.
There's no need to have this behavior - if we just always go to ScalarPH we can get rid of a bunch of complexity.
llvm-svn: 246637
This reduces the complexity of createEmptyBlock() and will open the door to further refactoring.
The test change is simply because we're now constant folding a trivial test.
llvm-svn: 246634
There's no need to widen canonical induction variables. It's just as efficient to create a *new*, wide, induction variable.
Consider, if we widen an indvar, then we'll have to truncate it before its uses anyway (1 trunc). If we create a new indvar instead, we'll have to truncate that instead (1 trunc) [besides which IndVars should go and clean up our mess after us anyway on principle].
This lets us remove a ton of special-casing code.
llvm-svn: 246631
Vectorized loops only ever have one induction variable. All induction PHIs from the scalar loop are rewritten to be in terms of this single indvar.
We were trying very hard to pick an indvar that already existed, even if that indvar wasn't canonical (didn't start at zero). But trying so hard is really fruitless - creating a new, canonical, indvar only results in one extra add in the worst case and that add is trivially easy to push through the PHI out of the loop by instcombine.
If we try and be less clever here and instead let instcombine clean up our mess (as we do in many other places in LV), we can remove unneeded complexity.
llvm-svn: 246630
Summary:
This change turns on by default interleaved access vectorization
for AArch64.
We also clean up some tests which were spedifically enabling this
behaviour.
Reviewers: rengolin
Subscribers: aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D12149
llvm-svn: 246542
Summary:
This change turns on by default interleaved access vectorization on ARM,
as it has shown to be beneficial on ARM.
Reviewers: rengolin
Subscribers: aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D12146
llvm-svn: 246541
Teach FunctionAttr to infer the nonnull attribute on return values of functions which never return a potentially null value. This is done both via a conservative local analysis for the function itself and a optimistic per-SCC analysis. If no function in the SCC returns anything which could be null (other than values from other functions in the SCC), we can conclude no function returned a null pointer. Even if some function within the SCC returns a null pointer, we may be able to locally conclude that some don't.
Differential Revision: http://reviews.llvm.org/D9688
llvm-svn: 246476
If asked to prove a predicate about a value produced by a PHI node, LazyValueInfo was unable to do so even if the predicate was known to be true for each input to the PHI. This prevented JumpThreading from eliminating a provably redundant branch.
The problematic test case looks something like this:
ListNode *p = ...;
while (p != null) {
if (!p) return;
x = g->x; // unrelated
p = p->next
}
The null check at the top of the loop is redundant since the value of 'p' is null checked on entry to the loop and before executing the backedge. This resulted in us a) executing an extra null check per iteration and b) not being able to LICM unrelated loads after the check since we couldn't prove they would execute or that their dereferenceability wasn't effected by the null check on the first iteration.
Differential Revision: http://reviews.llvm.org/D12383
llvm-svn: 246465
Summary:
JumpThreading shouldn't duplicate a convergent call, because that would move a convergent call into a control-inequivalent location. For example,
if (cond) {
...
} else {
...
}
convergent_call();
if (cond) {
...
} else {
...
}
should not be optimized to
if (cond) {
...
convergent_call();
...
} else {
...
convergent_call();
...
}
Test Plan: test/Transforms/JumpThreading/basic.ll
Patch by Xuetian Weng.
Reviewers: resistor, arsenm, jingyue
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12484
llvm-svn: 246415
I'm working on adding !dbg attachments to functions (PR23367), which
we'll use to determine the canonical subprogram for a function (instead
of the `subprograms:` array in the compile units). This updates a few
old tests in preparation.
Transforms/Mem2Reg/ConvertDebugInfo2.ll had an old-style grep+count
based test that would start to fail because I've added an extra line
with `!dbg`. Instead, explicitly `CHECK` for what I think the test
actually cares about.
All three testcases have subprograms with a valid `function:` reference
-- which means my upgrade script will add a `!dbg` attachment -- but
that aren't referenced from any compile unit. I suspect these testcases
were handreduced over-zealously (or have bitrotted?). Add a reference
from the compile unit so that upcoming Verifier checks won't fail here.
llvm-svn: 246351
This reverts isSafeToSpeculativelyExecute's use of ReadNone until we
split ReadNone into two pieces: one attribute which reasons about how
the function reasons about memory and another attribute which determines
how it may be speculated, CSE'd, trap, etc.
llvm-svn: 246331
As a follow-up to r246098, require `DISubprogram` definitions
(`isDefinition: true`) to be 'distinct'. Specifically, add an assembler
check, a verifier check, and bitcode upgrading logic to combat testcase
bitrot after the `DIBuilder` change.
While working on the testcases, I realized that
test/Linker/subprogram-linkonce-weak-odr.ll isn't relevant anymore. Its
purpose was to check for a corner case in PR22792 where two subprogram
definitions match exactly and share the same metadata node. The new
verifier check, requiring that subprogram definitions are 'distinct',
precludes that possibility.
I updated almost all the IR with the following script:
git grep -l -E -e '= !DISubprogram\(.* isDefinition: true' |
grep -v test/Bitcode |
xargs sed -i '' -e 's/= \(!DISubprogram(.*, isDefinition: true\)/= distinct \1/'
Likely some variant of would work for out-of-tree testcases.
llvm-svn: 246327
PR24605 is caused due to an incorrect insert point in instcombine's IR
builder. When simplifying
%t = add X Y
...
%m = icmp ... %t
the replacement for %t should be placed before %t, not before %m, as
there could be a use of %t between %t and %m.
llvm-svn: 246315
Summary:
This patch removes two remaining places where pointer value comparisons
are used to order functions: comparing range annotation metadata, and comparing
block address constants. (These are both rare cases, and so no actual
non-determinism was observed from either case).
The fix for range metadata is simple: the annotation always consists of a pair
of integers, so we just order by those integers.
The fix for block addresses is more subtle. Two constants are the same if they
are the same basic block in the same function, or if they refer to corresponding
basic blocks in each respective function. Note that in the first case, merging
is trivially correct. In the second, the correctness of merging relies on the
fact that the the values of block addresses cannot be compared. This change is
actually an enhancement, as these functions could not previously be merged (see
merge-block-address.ll).
There is still a problem with cross function block addresses, in that constants
pointing to a basic block in a merged function is not updated.
This also more robustly compares floating point constants by all fields of their
semantics, and fixes a dyn_cast/cast mixup.
Author: jrkoenig
Reviewers: dschuff, nlewycky, jfb
Subscribers llvm-commits
Differential revision: http://reviews.llvm.org/D12376
llvm-svn: 246305
handle more allocas with loads past the end of the alloca.
I suspect there are some related crashers with slightly different
patterns, but I'll fix those and add test cases as I find them.
Thanks to David Majnemer for the excellent test case reduction here.
Made this super simple to debug and fix.
llvm-svn: 246289
After hitting @llvm.assume(X) we can:
- propagate equality that X == true
- if X is icmp/fcmp (with eq operation), and one of operand
is constant we can change all variables with constants in the same BasicBlock
http://reviews.llvm.org/D11918
llvm-svn: 246243
Any call which is side effect free is trivially OK to speculate. We
already had similar logic in EarlyCSE and GVN but we were missing it
from isSafeToSpeculativelyExecute.
This fixes PR24601.
llvm-svn: 246232
Constant propagation for single precision math functions (such as
tanf) is already working, but was not enabled. This patch enables
these for many single-precision functions, and adds respective test
cases.
Newly handled functions: acosf asinf atanf atan2f ceilf coshf expf
exp2f fabsf floorf fmodf logf log10f powf sinhf tanf tanhf
llvm-svn: 246194
This patch changes the analysis diagnostics produced when loops with
floating-point recurrences or memory operations are identified. The new messages
say "cannot prove it is safe to reorder * operations; allow reordering by
specifying #pragma clang loop vectorize(enable)". Depending on the type of
diagnostic the message will include additional options such as ffast-math or
__restrict__.
This patch also allows the vectorize(enable) pragma to override the low pointer
memory check threshold. When the hint is given a higher threshold is used.
See the clang patch for the options produced for each diagnostic.
llvm-svn: 246187
Constant propagation for single precision math functions (such as
tanf) is already working, but was not enabled. This patch enables
these for many single-precision functions, and adds respective test
cases.
Newly handled functions: acosf asinf atanf atan2f ceilf coshf expf
exp2f fabsf floorf fmodf logf log10f powf sinhf tanf tanhf
llvm-svn: 246186
Constant propagation for single precision math functions (such as
tanf) is already working, but was not enabled. This patch enables
these for many single-precision functions, and adds respective test
cases.
Newly handled functions: acosf asinf atanf atan2f ceilf coshf expf
exp2f fabsf floorf fmodf logf log10f powf sinhf tanf tanhf
llvm-svn: 246158
Unlike scalar operations, we can perform vector operations on element types that
are smaller than the native integer types. We type-promote scalar operations if
they are smaller than a native type (e.g., i8 arithmetic is promoted to i32
arithmetic on Arm targets). This patch detects and removes type-promotions
within the reduction detection framework, enabling the vectorization of small
size reductions.
In the legality phase, we look through the ANDs and extensions that InstCombine
creates during promotion, keeping track of the smaller type. In the
profitability phase, we use the smaller type and ignore the ANDs and extensions
in the cost model. Finally, in the code generation phase, we truncate the result
of the reduction to allow InstCombine to rewrite the entire expression in the
smaller type.
This fixes PR21369.
http://reviews.llvm.org/D12202
Patch by Matt Simpson <mssimpso@codeaurora.org>!
llvm-svn: 246149
Globals in address spaces other than one may have 0 as a valid address,
so we should not assume that they can be null.
Reviewed by Philip Reames.
llvm-svn: 246137
A release fence acts as a publication barrier for stores within the current thread to become visible to other threads which might observe the release fence. It does not require the current thread to observe stores performed on other threads. As a result, we can allow store-load and load-store forwarding across a release fence.
We do need to make sure that stores before the fence can't be eliminated even if there's another store to the same location after the fence. In theory, we could reorder the second store above the fence and *then* eliminate the former, but we can't do this if the stores are on opposite sides of the fence.
Note: While more aggressive then what's there, this patch is still implementing a really conservative ordering. In particular, I'm not trying to exploit undefined behavior via races, or the fact that the LangRef says only 'atomic' accesses are ordered w.r.t. fences.
Differential Revision: http://reviews.llvm.org/D11434
llvm-svn: 246134
When computing base pointers, we introduce new instructions to propagate the base of existing instructions which might not be bases. However, the algorithm doesn't make any effort to recognize when the new instruction to be inserted is the same as an existing one already in the IR. Since this is happening immediately before rewriting, we don't really have a chance to fix it after the pass runs without teaching loop passes about statepoints.
I'm really not thrilled with this patch. I've rewritten it 4 different ways now, but this is the best I've come up with. The case where the new instruction is just the original base defining value could be merged into the existing algorithm with some complexity. The problem is that we might have something like an extractelement from a phi of two vectors. It may be trivially obvious that the base of the 0th element is an existing instruction, but I can't see how to make the algorithm itself figure that out. Thus, I resort to the call to SimplifyInstruction instead.
Note that we can only adjust the instructions we've inserted ourselves. The live sets are still being tracked in side structures at this point in the code. We can't easily muck with instructions which might be in them. Long term, I'm really thinking we need to materialize the live pointer sets explicitly in the IR somehow rather than using side structures to track them.
Differential Revision: http://reviews.llvm.org/D12004
llvm-svn: 246133
This patch ensures that every analysis diagnostic produced by the vectorizer
will be printed if the loop has a vectorization hint on it. The condition has
also been improved to prevent printing when a disabling hint is specified.
llvm-svn: 246132
As Sanjoy pointed out over in http://reviews.llvm.org/D11819, a switch on an icmp should always be able to become a branch instruction. This patch generalizes that notion slightly to prove that the default case of a switch is unreachable if the cases completely cover all possible bit patterns in the condition. Once that's done, the switch to branch conversion kicks in just fine.
Note: Duplicate case values are disallowed by the LangRef and verifier.
Differential Revision: http://reviews.llvm.org/D11995
llvm-svn: 246125
This was only added to preserve the old ScalarRepl's use of SSAUpdater
which was originally to avoid use of dominance frontiers. Now, we only
need a domtree, and we'll need a domtree right after this pass as well
and so it makes perfect sense to always and only use the dom-tree
powered mem2reg. This was flag-flipper earlier and has stuck reasonably
so I wanted to gut the now-dead code out of SROA before we waste more
time with it. Among other things, this will make passmanager porting
easier.
llvm-svn: 246028
Summary: When comparing basic blocks, there is an additional check that two Value*'s should have the same ID, which interferes with merging equivalent constants of different kinds (such as a ConstantInt and a ConstantPointerNull in the included testcase). The cmpValues function already ensures that the two values in each function are the same, so removing this check should not cause incorrect merging.
Also, the type comparison is redundant, based on reviewing the code and testing on the test suite and several large LTO bitcodes.
Author: jrkoenig
Reviewers: nlewycky, jfb, dschuff
Subscribers: llvm-commits
Differential revision: http://reviews.llvm.org/D12302
llvm-svn: 246001
The loop minimum iterations check below ensures the loop has enough trip count so the generated
vector loop will likely be executed, and it covers the overflow check.
Differential Revision: http://reviews.llvm.org/D12107.
llvm-svn: 245952
It doesn't solve the problem, when for example we load something, and
then assume that it is the same as some constant value, because
globalopt will fail on unknown load instruction. The proposed solution
would be to skip some instructions that we can't evaluate and they are
safe to skip (f.e. load, assume and many others) and see if they are
required to perform optimization (f.e. we don't care about ephemeral
instructions that may appear using @llvm.assume())
http://reviews.llvm.org/D12266
llvm-svn: 245919
Some debug info was drastically out of date, from the days where we used
to emit a list of length one (with a single null entry) rather than an
empty list (or, more recently, no list at all) for list fields that have
no elements.
llvm-svn: 245796
Summary:
Merge functions previously relied on unsigned comparisons of pointer values to
order functions. This caused observable non-determinism in the compiler for
large bitcode programs. Basically, opt -mergefuncs program.bc | md5sum produces
different hashes when run repeatedly on the same machine. Differing output was
observed on three large bitcodes, but it was less frequent on the smallest file.
It is possible that this only manifests on the large inputs, hence remaining
undetected until now.
This patch fixes this by removing (almost, see below) all places where
comparisons between pointers are used to order functions. Most of these changes
are local, but the comparison of global values requires assigning an identifier
to each local in the order it is visited. This is very similar to the way the
comparison function identifies Value*'s defined within a function. Because the
order of visiting the functions and their subparts is deterministic, the
identifiers assigned to the globals will be as well, and the order of functions
will be deterministic.
With these changes, there is no more observed non-determinism. There is also
only minor slowdowns (negligible to 4%) compared to the baseline, which is
likely a result of the fact that global comparisons involve hash lookups and not
just pointer comparisons.
The one caveat so far is that programs containing BlockAddress constants can
still be non-deterministic. It is not clear what the right solution is here. In
particular, even if the global numbers are used to order by function, we still
need a way to order the BasicBlock*'s. Unfortunately, we cannot just bail out
and fail to order the functions or consider them equal, because we require a
total order over functions. Note that programs with BlockAddress constants are
relatively rare, so the impact of leaving this in is minor as long as this pass
is opt-in.
Author: jrkoenig
Reviewers: nlewycky, jfb, dschuff
Subscribers: jevinskie, llvm-commits, chapuni
Differential revision: http://reviews.llvm.org/D12168
llvm-svn: 245762
SCEV expansion can invalidate previously expanded values. For example
in SCEVExpander::ReuseOrCreateCast, if we already have the requested
cast value but it's not at the desired location, a new cast is inserted
and the old cast will be invalidated.
Therefore, when expanding the bounds for the pointers, a later entry can
invalidate the IR value for an earlier one. The fix is to store a value
handle rather than the value itself.
The newly added test has a more detailed description of how the bug
triggers.
This bug can have a negative but potentially highly variable performance
impact in Loop Distribution. Because one of the bound values was
invalidated and is an undef expression now, InstCombine is free to
transform the array overlap check:
Start0 <= End1 && Start1 <= End0
into:
Start0 <= End1
So depending on the runtime location of the arrays, we would detect a
conflict and fall back on the original loop of the versioned loop.
Also tested compile time with SPEC2006 LTO bc files.
llvm-svn: 245760
The original checkin was buggy, this change has a fix.
Original commit message:
[InstCombine] Transform A & (L - 1) u< L --> L != 0
Summary:
This transform is never a pessimization at the IR level (since it
replaces an `icmp` with another), and has potentiall payoffs:
1. It may make the `icmp` fold away or become loop invariant.
2. It may make the `A & (L - 1)` computation dead.
This shows up in Java, in range checks generated by array accesses of
the form `a[i & (a.length - 1)]`.
Reviewers: reames, majnemer
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12210
llvm-svn: 245753
Summary:
This transform is never a pessimization at the IR level (since it
replaces an `icmp` with another), and has potentiall payoffs:
1. It may make the `icmp` fold away or become loop invariant.
2. It may make the `A & (L - 1)` computation dead.
This shows up in Java, in range checks generated by array accesses of
the form `a[i & (a.length - 1)]`.
Reviewers: reames, majnemer
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D12210
llvm-svn: 245635
Summary:
Add an LSR test that exercises isTruncateFree. Without this change, LSR creates
another indvar representing the truncated value.
Reviewers: jholewinski, eliben
Subscribers: jholewinski, llvm-commits
Differential Revision: http://reviews.llvm.org/D12058
llvm-svn: 245611
Instruction::dropUnknownMetadata(KnownSet) is supposed to preserve all
metadata in KnownSet, but the condition for DebugLocs was inverted.
Most users of dropUnknownMetadata() actually worked around this by not
adding LLVMContext::MD_dbg to their list of KnowIDs.
This is now made explicit.
llvm-svn: 245589
Summary: We know that -x & 1 is equivalent to x & 1, avoid using negation for testing if a negative integer is even or odd.
Reviewers: majnemer
Subscribers: junbuml, mssimpso, gberry, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D12156
llvm-svn: 245569
Usually DSE is not supposed to remove lifetime intrinsics, but it's
actually ok to remove them for dead objects in terminating blocks,
because they convey no extra information there. Until we hit a lifetime
start that cannot be removed, that is. Because from that point on the
lifetime intrinsics become interesting again, e.g. for stack coloring.
Reviewers: reames
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11710
llvm-svn: 245542
Here we make ScalarEvolution::isKnownPredicate, indirectly, a little smarter.
Given some relational comparison operator OP, and two AddRec SCEVs, {I,+,S} OP
{J,+,T}, we can reduce this to the comparison I OP J when S == T, both AddRecs
are for the same loop, and both are known not to wrap.
As it turns out, because of the way that backedge-guard expressions can be
leveraged when computing known predicates, this allows indvars to simplify the
if-statement comparison in this loop:
void foo (int *a, int *b, int n) {
for (int i = 0; i < n; ++i) {
if (i > n)
a[i] = b[i] + 1;
}
}
which, somewhat surprisingly, we were not previously optimizing away.
llvm-svn: 245400
After hitting @llvm.assume(X) we can:
- propagate equality that X == true
- if X is icmp/fcmp (with eq operation), and one of operand
is constant we can change all variables with constants in the same BasicBlock
http://reviews.llvm.org/D11918
llvm-svn: 245265
Summary:
Increase the estimated costs for insert/extract element operations on
AArch64. This is motivated by results from benchmarking interleaved
accesses.
Add missing costs for zext/sext/trunc instructions and some integer to
floating point conversions. These costs were previously calculated
by scalarizing these operation and were affected by the cost increase of
the insert/extract element operations.
Reviewers: rengolin
Subscribers: mcrosier, aemerson, rengolin, llvm-commits
Differential Revision: http://reviews.llvm.org/D11939
llvm-svn: 245226
PR24469 resulted because DeleteDeadInstruction in handleNonLocalStoreDeletion was
deleting the next basic block iterator. Fixed the same by resetting the basic block iterator
post call to DeleteDeadInstruction.
llvm-svn: 245195
If we can ignore NaNs, fmin/fmax libcalls can become compare and select
(this is what we turn std::min / std::max into).
This IR should then be optimized in the backend to whatever is best for
any given target. Eg, x86 can use minss/maxss instructions.
This should solve PR24314:
https://llvm.org/bugs/show_bug.cgi?id=24314
Differential Revision: http://reviews.llvm.org/D11866
llvm-svn: 245187
Bitwise arithmetic can obscure a simple sign-test. If replacing the
mask with a truncate is preferable if the type is legal because it
permits us to rephrase the comparison more explicitly.
llvm-svn: 245171
This patch makes the Merge Functions pass faster by calculating and comparing
a hash value which captures the essential structure of a function before
performing a full function comparison.
The hash is calculated by hashing the function signature, then walking the basic
blocks of the function in the same order as the main comparison function. The
opcode of each instruction is hashed in sequence, which means that different
functions according to the existing total order cannot have the same hash, as
the comparison requires the opcodes of the two functions to be the same order.
The hash function is a static member of the FunctionComparator class because it
is tightly coupled to the exact comparison function used. For example, functions
which are equivalent modulo a single variant callsite might be merged by a more
aggressive MergeFunctions, and the hash function would need to be insensitive to
these differences in order to exploit this.
The hashing function uses a utility class which accumulates the values into an
internal state using a standard bit-mixing function. Note that this is a different interface
than a regular hashing routine, because the values to be hashed are scattered
amongst the properties of a llvm::Function, not linear in memory. This scheme is
fast because only one word of state needs to be kept, and the mixing function is
a few instructions.
The main runOnModule function first computes the hash of each function, and only
further processes functions which do not have a unique function hash. The hash
is also used to order the sorted function set. If the hashes differ, their
values are used to order the functions, otherwise the full comparison is done.
Both of these are helpful in speeding up MergeFunctions. Together they result in
speedups of 9% for mysqld (a mostly C application with little redundancy), 46%
for libxul in Firefox, and 117% for Chromium. (These are all LTO builds.) In all
three cases, the new speed of MergeFunctions is about half that of the module
verifier, making it relatively inexpensive even for large LTO builds with
hundreds of thousands of functions. The same functions are merged, so this
change is free performance.
Author: jrkoenig
Reviewers: nlewycky, dschuff, jfb
Subscribers: llvm-commits, aemerson
Differential revision: http://reviews.llvm.org/D11923
llvm-svn: 245140
This seems to only work some of the time. In some situations,
this seems to use a nonsensical type and isn't actually aware of the
memory being accessed. e.g. if branch condition is an icmp of a pointer,
it checks the addressing mode of i1.
llvm-svn: 245137
Summary:
http://reviews.llvm.org/D11212 made Scalar Evolution able to propagate NSW and NUW flags from instructions to SCEVs for add instructions. This patch expands that to sub, mul and shl instructions.
This change makes LSR able to generate pointer induction variables for loops like these, where the index is 32 bit and the pointer is 64 bit:
for (int i = 0; i < numIterations; ++i)
sum += ptr[i - offset];
for (int i = 0; i < numIterations; ++i)
sum += ptr[i * stride];
for (int i = 0; i < numIterations; ++i)
sum += ptr[3 * (i << 7)];
Reviewers: atrick, sanjoy
Subscribers: sanjoy, majnemer, hfinkel, llvm-commits, meheff, jingyue, eliben
Differential Revision: http://reviews.llvm.org/D11860
llvm-svn: 245118
Summary:
This patch implements my promised optimization to reunites certain sexts from
operands after we extract the constant offset. See the header comment of
reuniteExts for its motivation.
One key building block that enables this optimization is Bjarke's poison value
analysis (D11212). That helps to prove "a +nsw b" can't overflow.
Reviewers: broune
Subscribers: jholewinski, sanjoy, llvm-commits
Differential Revision: http://reviews.llvm.org/D12016
llvm-svn: 245003
If <src> is non-zero we can safely set the flag to true, and this
results in less code generated for, e.g. ffs(x) + 1 on FreeBSD.
Thanks to majnemer for suggesting the fix and reviewing.
Code generated before the patch was applied:
0: 0f bc c7 bsf %edi,%eax
3: b9 20 00 00 00 mov $0x20,%ecx
8: 0f 45 c8 cmovne %eax,%ecx
b: 83 c1 02 add $0x2,%ecx
e: b8 01 00 00 00 mov $0x1,%eax
13: 85 ff test %edi,%edi
15: 0f 45 c1 cmovne %ecx,%eax
18: c3 retq
Code generated after the patch was applied:
0: 0f bc cf bsf %edi,%ecx
3: 83 c1 02 add $0x2,%ecx
6: 85 ff test %edi,%edi
8: b8 01 00 00 00 mov $0x1,%eax
d: 0f 45 c1 cmovne %ecx,%eax
10: c3 retq
It seems we can still use cmove and save another 'test' instruction, but
that can be tackled separately.
Differential Revision: http://reviews.llvm.org/D11989
llvm-svn: 244947
We used to be over-conservative about preserving inbounds. Actually, the second
GEP (which applies the constant offset) can inherit the inbounds attribute of
the original GEP, because the resultant pointer is equivalent to that of the
original GEP. For example,
x = GEP inbounds a, i+5
=>
y = GEP a, i // inbounds removed
x = GEP inbounds y, 5 // inbounds preserved
llvm-svn: 244937
DeadStoreElimination does eliminate a store if it stores a value which was loaded from the same memory location.
So far this worked only if the store is in the same block as the load.
Now we can also handle stores which are in a different block than the load.
Example:
define i32 @test(i1, i32*) {
entry:
%l2 = load i32, i32* %1, align 4
br i1 %0, label %bb1, label %bb2
bb1:
br label %bb3
bb2:
; This store is redundant
store i32 %l2, i32* %1, align 4
br label %bb3
bb3:
ret i32 0
}
Differential Revision: http://reviews.llvm.org/D11854
llvm-svn: 244901
Consider this code:
BB:
%i = phi i32 [ 0, %if.then ], [ %c, %if.else ]
%add = add nsw i32 %i, %b
...
In this common case the add can be moved to the %if.else basic block, because
adding zero is an identity operation. If we go though %if.then branch it's
always a win, because add is not executed; if not, the number of instructions
stays the same.
This pattern applies also to other instructions like sub, shl, shr, ashr | 0,
mul, sdiv, div | 1.
Patch by Jakub Kuderski!
llvm-svn: 244887
Most SSE/AVX (non-constant) vector shift instructions only use the lower 64-bits of the 128-bit shift amount vector operand, this patch calls SimplifyDemandedVectorElts to optimize for this.
I had to refactor some of my recent InstCombiner work on the vector shifts to avoid quite a bit of duplicate code, it means that SimplifyX86immshift now (re)decodes the type of shift.
Differential Revision: http://reviews.llvm.org/D11938
llvm-svn: 244872
To be clear: this is an *optimization* not a correctness change.
CodeGenPrep likes to duplicate icmps feeding branch instructions to take advantage of x86's ability to fuze many comparison/branch patterns into a single micro-op and to reduce the need for materializing i1s into general registers. PlaceSafepoints likes to place safepoint polls right at the end of basic blocks (immediately before terminators) when inserting entry and backedge safepoints. These two heuristics interact in a somewhat unfortunate way where the branch terminating the original block will be controlled by a condition driven by unrelocated pointers. This forces the register allocator to keep both the relocated and unrelocated values of the pointers feeding the icmp alive over the safepoint poll.
One simple fix would have been to just adjust PlaceSafepoints to move one back in the basic block, but you can reach similar cases as a result of LICM or other hoisting passes. As a result, doing a post insertion fixup seems to be more robust.
I considered doing this in CodeGenPrep itself, but having to update the live sets of already rewritten safepoints gets complicated fast. In particular, you can't just use def/use information because by moving the icmp, we're extending the live range of it's inputs potentially.
Instead, this patch teaches RewriteStatepointsForGC to make the required adjustments before making the relocations explicit in the IR. This change really highlights the fact that RSForGC is a CodeGenPrep-like pass which is performing target specific lowering. In the long run, we may even want to combine the two though this would require a lot more smarts to be integrated into RSForGC first. We currently rely on being able to run a set of cleanup passes post rewriting because the IR RSForGC generates is pretty damn ugly.
Differential Revision: http://reviews.llvm.org/D11819
llvm-svn: 244821
When rewriting the IR such that base pointers are available for every live pointer, we potentially need to duplicate instructions to propagate the base. The original code had only handled PHI and Select under the belief those were the only instructions which would need duplicated. When I added support for vector instructions, I'd added a collection of hacks for ExtractElement which caught most of the common cases. Of course, I then found the one test case my hacks couldn't cover. :)
This change removes all of the early hacks for extract element. By defining extractelement as a BDV (rather than trying to look through it), we can extend the rewriting algorithm to duplicate the extract as needed. Note that a couple of peephole optimizations were left in for the moment, because while we now handle extractelement as a first class citizen, we're not yet handling insertelement. That change will follow in the near future.
llvm-svn: 244808
As discussed in D11886, this patch moves the SSE/AVX vector blend folding to instcombiner from PerformINTRINSIC_WO_CHAINCombine (which allows us to remove this completely).
InstCombiner already had partial support for this, I just had to add support for zero (ConstantAggregateZero) masks and also the case where both selection inputs were the same (allowing us to ignore the mask).
I also moved all the relevant combine tests into InstCombine/blend_x86.ll
Differential Revision: http://reviews.llvm.org/D11934
llvm-svn: 244723
`InstCombiner::OptimizeOverflowCheck` was asserting an
invariant (operands to binary operations are ordered by decreasing
complexity) that wasn't really an invariant. Fix this by instead having
`InstCombiner::OptimizeOverflowCheck` establish the invariant if it does
not hold.
llvm-svn: 244676
Summary: LowerSwitch crashed with the attached test case after deleting the default block. This happened because the current implementation of deleting dead blocks is wrong. After the default block being deleted, it contains no instruction or terminator, and it should no be traversed anymore. However, since the iterator is advanced before processSwitchInst() function is executed, the block advanced to could be deleted inside processSwitchInst(). The deleted block would then be visited next and crash dyn_cast<SwitchInst>(Cur->getTerminator()) because Cur->getTerminator() returns a nullptr. This patch fixes this problem by recording dead default blocks into a list, and delete them after all processSwitchInst() has been done. It still possible to visit dead default blocks and waste time process them. But it is a compile time issue, and I plan to have another patch to add support to skip dead blocks.
Reviewers: kariddi, resistor, hans, reames
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11852
llvm-svn: 244642
I incorrectly wrote CHECK-NEXT with followin with ':', the check was
ignored by FileCheck.
The non-inbound GEP is folded here because the DataLayout is no longer
optional, the fold was originally guarded with a comment that said:
We need TD information to know the pointer size unless this is inbounds.
Now we always have "TD information" and perform the fold.
Thanks Jonathan Roelofs for noticing.
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 244613
The select pattern recognition in ValueTracking (as used by InstCombine
and SelectionDAGBuilder) only knew about integer patterns. This teaches
it about minimum and maximum operations.
matchSelectPattern() has been extended to return a struct containing the
existing Flavor and a new enum defining the pattern's behavior when
given one NaN operand.
C minnum() is defined to return the non-NaN operand in this case, but
the idiomatic C "a < b ? a : b" would return the NaN operand.
ARM and AArch64 at least have different instructions for these different cases.
llvm-svn: 244580
This patch and a relatec clang patch solve the problem of having to explicitly enable analysis when specifying a loop hint pragma to get the diagnostics. Passing AlwasyPrint as the pass name (see below) causes the front-end to print the diagnostic if the user has specified '-Rpass-analysis' without an '=<target-pass>’. Users of loop hints can pass that compiler option without having to specify the pass and they will get diagnostics for only those loops with loop hints.
llvm-svn: 244555
This patch moves checking the threshold of runtime pointer checks to the vectorization requirements (late diagnostics) and emits a diagnostic that infroms the user the loop would be vectorized if not for exceeding the pointer-check threshold. Clang will also append the options that can be used to allow vectorization.
llvm-svn: 244523
As discussed in D11760, this patch moves the (V)PSRA(WD) arithmetic shift-by-constant folding to InstCombine to match the logical shift implementations.
Differential Revision: http://reviews.llvm.org/D11886
llvm-svn: 244495
This patch moves the verification of fast-math to just before vectorization is done. This way we can tell clang to append the command line options would that allow floating-point commutativity. Specifically those are enableing fast-math or specifying a loop hint.
llvm-svn: 244489
Sometimes interleaving is not beneficial, as determined by the cost-model and sometimes it is disabled by a loop hint (by the user). This patch modifies the diagnostic messages to make it clear why interleaving wasn't done.
llvm-svn: 244485
I looked into adding a warning / error for this to FileCheck, but there doesn't
seem to be a good way to avoid it triggering on the instances of it in RUN lines.
llvm-svn: 244481
This change adds the unroll metadata "llvm.loop.unroll.enable" which directs
the optimizer to unroll a loop fully if the trip count is known at compile time, and
unroll partially if the trip count is not known at compile time. This differs from
"llvm.loop.unroll.full" which explicitly does not unroll a loop if the trip count is not
known at compile time.
The "llvm.loop.unroll.enable" is intended to be added for loops annotated with
"#pragma unroll".
llvm-svn: 244466
The scalarizer can cache incorrect entries when walking up a chain of
insertelement instructions. This occurs when it encounters more than one
instruction that it is not actively searching for, as it unconditionally caches
every element it finds. The fix is to only cache the first element that it
isn't searching for so we don't overwrite correct entries.
Reviewers: hfinkel
Differential Revision: http://reviews.llvm.org/D11559
llvm-svn: 244448
We can only PHI translate instructions. In our attempt to PHI translate
a bitcast, we attempt to translate its operand; however, the operand
might be an argument or a global instead of an instruction. Benignly
bail out when this happens.
This fixes PR24397.
Differential Revision: http://reviews.llvm.org/D11879
llvm-svn: 244418
Summary: llvm::ConstantFoldTerminator function can convert SwitchInst with single case (and default) to a conditional BranchInst. This patch adds support to preserve make.implicit metadata on this conversion.
Reviewers: sanjoy, weimingz, chenli
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D11841
llvm-svn: 244348
This patch fixes the sse2/avx2 vector shift by constant instcombine call to correctly deal with the fact that the shift amount is formed from the entire lower 64-bit and not just the lowest element as it currently assumes.
e.g.
%1 = tail call <4 x i32> @llvm.x86.sse2.psrl.d(<4 x i32> %v, <4 x i32> <i32 15, i32 15, i32 15, i32 15>)
In this case, (V)PSRLD doesn't perform a lshr by 15 but in fact attempts to shift by 64424509455 ((15 << 32) | 15) - giving a zero result.
In addition, this review also recognizes shift-by-zero from a ConstantAggregateZero type (PR23821).
Differential Revision: http://reviews.llvm.org/D11760
llvm-svn: 244341
iisUnmovableInstruction() had a list of instructions hardcoded which are
considered unmovable. The list lacked (at least) an entry for the va_arg
and cmpxchg instructions.
Fix this by introducing a new Instruction::mayBeMemoryDependent()
instead of maintaining another instruction list.
Patch by Matthias Braun <matze@braunis.de>.
Differential Revision: http://reviews.llvm.org/D11577
rdar://problem/22118647
llvm-svn: 244244
Summary: Divide the primitive size in bits by eight so the initial load's alignment is in bytes as expected. Tested with the included unit test.
Reviewers: rengolin, jfb
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11804
llvm-svn: 244229
through PHI nodes across iterations.
This patch teaches the new advanced loop unrolling heuristics to propagate
constants into the loop from the preheader and around the backedge after
simulating each iteration. This lets us brute force solve simple recurrances
that aren't modeled effectively by SCEV. It also makes it more clear why we
need to process the loop in-order rather than bottom-up which might otherwise
make much more sense (for example, for DCE).
This came out of an attempt I'm making to develop a principled way to account
for dead code in the unroll estimation. When I implemented
a forward-propagating version of that it produced incorrect results due to
failing to propagate *cost* between loop iterations through the PHI nodes, and
it occured to me we really should at least propagate simplifications across
those edges, and it is quite easy thanks to the loop being in canonical and
LCSSA form.
Differential Revision: http://reviews.llvm.org/D11706
llvm-svn: 243900
Since r241097, `DIBuilder` has only created distinct `DICompileUnit`s.
The backend is liable to start relying on that (if it hasn't already),
so make uniquable `DICompileUnit`s illegal and automatically upgrade old
bitcode. This is a nice cleanup, since we can remove an unnecessary
`DenseSet` (and the associated uniquing info) from `LLVMContextImpl`.
Almost all the testcases were updated with this script:
git grep -e '= !DICompileUnit' -l -- test |
grep -v test/Bitcode |
xargs sed -i '' -e 's,= !DICompileUnit,= distinct !DICompileUnit,'
I imagine something similar should work for out-of-tree testcases.
llvm-svn: 243885
Remove the fake `DW_TAG_auto_variable` and `DW_TAG_arg_variable` tags,
using `DW_TAG_variable` in their place Stop exposing the `tag:` field at
all in the assembly format for `DILocalVariable`.
Most of the testcase updates were generated by the following sed script:
find test/ -name "*.ll" -o -name "*.mir" |
xargs grep -l 'DILocalVariable' |
xargs sed -i '' \
-e 's/tag: DW_TAG_arg_variable, //' \
-e 's/tag: DW_TAG_auto_variable, //'
There were only a handful of tests in `test/Assembly` that I needed to
update by hand.
(Note: a follow-up could change `DILocalVariable::DILocalVariable()` to
set the tag to `DW_TAG_formal_parameter` instead of `DW_TAG_variable`
(as appropriate), instead of having that logic magically in the backend
in `DbgVariable`. I've added a FIXME to that effect.)
llvm-svn: 243774
The patch changes the SLPVectorizer::vectorizeStores to choose the immediate
succeeding or preceding candidate for a store instruction when it has multiple
consecutive candidates. In this way it has better chance to find more slp
vectorization opportunities.
Differential Revision: http://reviews.llvm.org/D10445
llvm-svn: 243666
Summary:
As added initially, statepoints required their call targets to be a
constant pointer null if ``numPatchBytes`` was non-zero. This turns out
to be a problem ergonomically, since there is no way to mark patchable
statepoints as calling a (readable) symbolic value.
This change remove the restriction of requiring ``null`` call targets
for patchable statepoints, and changes PlaceSafepoints to maintain the
symbolic call target through its transformation.
Reviewers: reames, swaroop.sridhar
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11550
llvm-svn: 243502
Summary:
Make Scalar Evolution able to propagate NSW and NUW flags from instructions to SCEVs in some cases. This is based on reasoning about when poison from instructions with these flags would trigger undefined behavior. This gives a 13% speed-up on some Eigen3-based Google-internal microbenchmarks for NVPTX.
There does not seem to be clear agreement about when poison should be considered to propagate through instructions. In this analysis, poison propagates only in cases where that should be uncontroversial.
This change makes LSR able to create induction variables for expressions like &ptr[i + offset] for loops like this:
for (int i = 0; i < limit; ++i) {
sum += ptr[i + offset];
}
Here ptr is a 64 bit pointer and offset is a 32 bit integer. For NVPTX, LSR currently creates an induction variable for i + offset instead, which is not as fast. Improving this situation is what brings the 13% speed-up on some Eigen3-based Google-internal microbenchmarks for NVPTX.
There are more details in this discussion on llvmdev.
June: http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-June/thread.html#87234
July: http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-July/thread.html#87392
Patch by Bjarke Roune
Reviewers: eliben, atrick, sanjoy
Subscribers: majnemer, hfinkel, jingyue, meheff, llvm-commits
Differential Revision: http://reviews.llvm.org/D11212
llvm-svn: 243460
The 'common' section TLS is not implemented.
Current C/C++ TLS variables are not placed in common section.
DWARF debug info to get the address of TLS variables is not generated yet.
clang and driver changes in http://reviews.llvm.org/D10524
Added -femulated-tls flag to select the emulated TLS model,
which will be used for old targets like Android that do not
support ELF TLS models.
Added TargetLowering::LowerToTLSEmulatedModel as a target-independent
function to convert a SDNode of TLS variable address to a function call
to __emutls_get_address.
Added into lib/Target/*/*ISelLowering.cpp to call LowerToTLSEmulatedModel
for TLSModel::Emulated. Although all targets supporting ELF TLS models are
enhanced, emulated TLS model has been tested only for Android ELF targets.
Modified AsmPrinter.cpp to print the emutls_v.* and emutls_t.* variables for
emulated TLS variables.
Modified DwarfCompileUnit.cpp to skip some DIE for emulated TLS variabls.
TODO: Add proper DIE for emulated TLS variables.
Added new unit tests with emulated TLS.
Differential Revision: http://reviews.llvm.org/D10522
llvm-svn: 243438
Summary:
If a scale or a base register can be rewritten as "Zext({A,+,1})" then
LSR will now consider a formula of that form in its normal cost
computation.
Depends on D9180
Reviewers: qcolombet, atrick
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D9181
llvm-svn: 243348
Summary:
Was D9784: "Remove loop variant range check when induction variable is
strictly increasing"
This change re-implements D9784 with the two differences:
1. It does not use SCEVExpander and does not generate new
instructions. Instead, it does a quick local search for existing
`llvm::Value`s that it needs when modifying the `icmp`
instruction.
2. It is more general -- it deals with both increasing and decreasing
induction variables.
I've added all of the tests included with D9784, and two more.
As an example on what this change does (copied from D9784):
Given C code:
```
for (int i = M; i < N; i++) // i is known not to overflow
if (i < 0) break;
a[i] = 0;
}
```
This transformation produces:
```
for (int i = M; i < N; i++)
if (M < 0) break;
a[i] = 0;
}
```
Which can be unswitched into:
```
if (!(M < 0))
for (int i = M; i < N; i++)
a[i] = 0;
}
```
I went back and forth on whether the top level logic should live in
`SimplifyIndvar::eliminateIVComparison` or be put into its own
routine. Right now I've put it under `eliminateIVComparison` because
even though the `icmp` is not *eliminated*, it no longer is an IV
comparison. I'm open to putting it in its own helper routine if you
think that is better.
Reviewers: reames, nicholas, atrick
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11278
llvm-svn: 243331
Now that we are generating sane codegen for vector sext/zext nodes on SSE targets, this patch uses instcombine to replace the SSE41/AVX2 pmovsx and pmovzx intrinsics with the equivalent native IR code.
Differential Revision: http://reviews.llvm.org/D11503
llvm-svn: 243303
The pointer size of the addrspacecasted pointer might not have matched,
so this would have hit an assert in accumulateConstantOffset.
I think this was here to allow constant folding of a load of an
addrspacecasted constant. Accumulating the offset through the
addrspacecast doesn't make much sense, so something else is necessary
to allow folding the load through this cast.
llvm-svn: 243300
Summary:
Fix the cost of interleaved accesses for ARM/AArch64.
We were calling getTypeAllocSize and using it to check
the number of bits, when we should have called
getTypeAllocSizeInBits instead.
This would pottentially cause the vectorizer to
generate loads/stores and shuffles which cannot
be matched with an interleaved access instruction.
No performance changes are expected for now since
matching/generating interleaved accesses is still
disabled by default.
Reviewers: rengolin
Subscribers: aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D11524
llvm-svn: 243270
r243250 appeared to break clang/test/Analysis/dead-store.c on one of the build
slaves, but I couldn't reproduce this failure locally. Probably a false
positive as I saw this test was broken by r243246 or r243247 too but passed
later without people fixing anything.
llvm-svn: 243253
Summary:
This patch updates TargetTransformInfoImplCRTPBase::getGEPCost to consider
addressing modes. It now returns TCC_Free when the GEP can be completely folded
to an addresing mode.
I started this patch as I refactored SLSR. Function isGEPFoldable looks common
and is indeed used by some WIP of mine. So I extracted that logic to getGEPCost.
Furthermore, I noticed getGEPCost wasn't directly tested anywhere. The best
testing bed seems CostModel, but its getInstructionCost method invokes
getAddressComputationCost for GEPs which provides very coarse estimation. So
this patch also makes getInstructionCost call the updated getGEPCost for GEPs.
This change inevitably breaks some tests because the cost model changes, but
nothing looks seriously wrong -- if we believe the new cost model is the right
way to go, these tests should be updated.
This patch is not perfect yet -- the comments in some tests need to be updated.
I want to know whether this is a right approach before fixing those details.
Reviewers: chandlerc, hfinkel
Subscribers: aschwaighofer, llvm-commits, aemerson
Differential Revision: http://reviews.llvm.org/D9819
llvm-svn: 243250
Summary:
This patch improves trivial loop unswitch.
The current trivial loop unswitch only checks if loop header's terminator contains a trivial unswitch condition. But if the loop header only has one reachable successor (due to intentionally or unintentionally missed code simplification), we should consider the successor as part of the loop header. Therefore, instead of stopping at loop header's terminator, we should keep traversing its successors within loop until reach a *real* conditional branch or switch (whose condition can not be constant folded). This change will enable a single -loop-unswitch pass to unswitch multiple trivial conditions (unswitch one trivial condition could open opportunity to unswitch another one in the same loop), while the old implementation can unswitch only one per pass.
Reviewers: reames, broune
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11481
llvm-svn: 243203
Add a verifier check that `DILocalVariable`s of tag
`DW_TAG_arg_variable` always have a non-zero 'arg:' field, and those of
tag `DW_TAG_auto_variable` always have a zero 'arg:' field. These are
the only configurations that are properly understood by the backend.
(Also, fix the bad examples in LangRef and test/Assembler, and fix the
bug in Kaleidoscope Ch8.)
A large number of testcases seem to have bitrotted their way forward
from some ancient version of the debug info hierarchy that didn't have
`arg:` parameters. If you have out-of-tree testcases that start failing
in the verifier and you don't care enough to get the `arg:` right, you
may have some luck just calling:
sed -e 's/, arg: 0/, arg: 1/'
or some such, but I hand-updated the ones in tree.
llvm-svn: 243183
This patch extend LoopReroll pass to hand the loops which
is similar to the following:
while (len > 1) {
sum4 += buf[len];
sum4 += buf[len-1];
len -= 2;
}
llvm-svn: 243171
The names for instructions inserted were previous dependent on iteration order. By deriving the names from the original instructions, we can avoid instability in tests without resorting to ordered traversals. It also makes the IR mildly easier to read at large scale.
llvm-svn: 243140
Summary:
Resolving a branch allows us to ignore blocks that won't be executed, and thus make our estimate more accurate.
This patch is intended to be applied after D10205 (though it could be applied independently).
Reviewers: chandlerc
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D10206
llvm-svn: 243084
Summary:
Scalarizer has two data structures that hold information about changes
to the function, Gathered and Scattered. These are cleared in finish()
at the end of runOnFunction() if finish() detects any changes to the
function.
However, finish() was checking for changes by only checking if
Gathered was non-empty. The function visitStore() only modifies
Scattered without touching Gathered. As a result, Scattered could have
ended up having stale data if Scalarizer only scalarized store
instructions. Since the data in Scattered is used during the execution
of the pass, this introduced dangling pointer errors.
The fix is to check whether both Scattered and Gathered are empty
before deciding what to do in finish(). This also fixes a problem
where the Function can be modified although the pass returns false.
Reviewers: rnk
Subscribers: rnk, srhines, llvm-commits
Differential Revision: http://reviews.llvm.org/D10459
llvm-svn: 243040
The MSVC ABI requires that we generate an alias for the vtable which
means looking through a GlobalAlias which cannot be overridden improves
our ability to devirtualize.
Found while investigating PR20801.
Patch by Andrew Zhogin!
Differential Revision: http://reviews.llvm.org/D11306
llvm-svn: 242955
Currently, a load from an alloca that is used in as single block and is not preceded
by a store is replaced by undef. This is not always correct if the single block is
inside a loop.
Fix the logic so that:
1) If there are no stores in the block, replace the load with an undef, as before.
2) If there is a store (regardless of where it is in the block w.r.t the load), bail
out, and let the rest of mem2reg handle this alloca.
Patch by: gil.rapaport@intel.com
Differential Revision: http://reviews.llvm.org/D11355
llvm-svn: 242884
Summary: The current code in LoopUnswtich::processCurrentLoop() mixes trivial loop unswitch and non-trivial loop unswitch together. It goes over all basic blocks in the loop and checks if a condition is trivial or non-trivial unswitch condition. However, trivial unswitch condition can only occur in the loop header basic block (where it controls whether or not the loop does something at all). This refactoring separate trivial loop unswitch and non-trivial loop unswitch. Before going over all basic blocks in the loop, it checks if the loop header contains a trivial unswitch condition. If so, unswitch it. Otherwise, go over all blocks like before but don't check trivial condition any more since they are not possible to be in the other blocks. This code has no functionality change.
Reviewers: meheff, reames, broune
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11276
llvm-svn: 242873
types and loads, loads or stores widened past the size of an alloca,
etc.
This started off with a bug report about big-endian behavior with
bitfields and loads and stores to a { i32, i24 } struct. An initial
attempt to fix this was sent for review in D10357, but that didn't
really get to the root of the problem.
The core issue was that canConvertValue and convertValue in SROA were
handling different bitwidth integers by doing a zext of the integer. It
wouldn't do a trunc though, only a zext! This would in turn lead SROA to
form an i24 load from an i24 alloca, zext it to i32, and then use it.
This would at least produce the wrong value for big-endian systems.
One of my many false starts here was to correct the computation for
big-endian systems by shifting. But this doesn't actually work because
the original code has a 64-bit store to the entire 8 bytes, and a 32-bit
load of the last 4 bytes, and because the alloc size is 8 bytes, we
can't lose that last (least significant if bigendian) byte! The real
problem here is that we're forming an i24 load in SROA which is actually
not sufficiently wide to load all of the necessary bits here. The source
has an i32 load, and SROA needs to form that as well.
The straightforward way to do this is to disable the zext logic in
canConvertValue and convertValue, forcing us to actually load all
32-bits. This seems like a really good change, but it in turn breaks
several other parts of SROA.
First in the chain of knock-on failures, we had places where we were
doing integer-widening promotion even though some of the integer loads
or stores extended *past the end* of the alloca's memory! There was even
a comment about preventing this, but it only prevented the case where
the type had a different bit size from its store size. So I added checks
to handle the cases where we actually have a widened load or store and
to avoid trying to special integer widening promotion in those cases.
Second, we actually rely on the ability to promote in the face of loads
past the end of an alloca! This is important so that we can (for
example) speculate loads around PHI nodes to do more promotion. The bits
loaded are garbage, but as long as they aren't used and the alignment is
suitable high (which it wasn't in the test case!) this is "fine". And we
can't stop promoting here, lots of things stop working well if we do. So
we need to add specific logic to handle the extension (and truncation)
case, but *only* where that extension or truncation are over bytes that
*are outside the alloca's allocated storage* and thus totally bogus to
load or store.
And of course, once we add back this correct handling of extension or
truncation, we need to correctly handle bigendian systems to avoid
re-introducing the exact bug that started us off on this chain of misery
in the first place, but this time even more subtle as it only happens
along speculated loads atop a PHI node.
I've ported an existing test for PHI speculation to the big-endian test
file and checked that we get that part correct, and I've added several
more interesting big-endian test cases that should help check that we're
getting this correct.
Fun times.
llvm-svn: 242869
We insert a bitcast which obfuscates the getCalledFunction for the utility
function which looks up attributes from the called function. Loosing ABI
changing parameter attributes is a bad thing.
rdar://21516488
llvm-svn: 242807
A patch by Chakshu Grover!
This patch allows constfolding of trunc,rint,nearbyint,ceil and floor intrinsics using APFloat class.
Differential Revision: http://reviews.llvm.org/D11144
llvm-svn: 242763
Summary:
SpeculativeExecution enables a series straight line optimizations (such
as SLSR and NaryReassociate) on conditional code. For example,
if (...)
... b * s ...
if (...)
... (b + 1) * s ...
speculative execution can hoist b * s and (b + 1) * s from then-blocks,
so that we have
... b * s ...
if (...)
...
... (b + 1) * s ...
if (...)
...
Then, SLSR can rewrite (b + 1) * s to (b * s + s) because after
speculative execution b * s dominates (b + 1) * s.
The performance impact of this change is significant. It speeds up the
benchmarks running EigenFloatContractionKernelInternal16x16
(ba68f42fa6/unsupported/Eigen/CXX11/src/Tensor/TensorContractionCuda.h (cl-526))
by roughly 2%. Some internal benchmarks that have the above code pattern
are improved by up to 40%. No significant slowdowns are observed on
Eigen CUDA microbenchmarks.
Reviewers: jholewinski, broune, eliben
Subscribers: llvm-commits, jholewinski
Differential Revision: http://reviews.llvm.org/D11201
llvm-svn: 242437
Internalizing an individual comdat group member without also internalizing
the other members of the comdat can break comdat semantics. For example,
if a module contains a reference to an internalized comdat member, and the
linker chooses a comdat group from a different object file, this will break
the reference to the internalized member.
This change causes the internalizer to only internalize comdat members if all
other members of the comdat are not externally visible. Once a comdat group
has been fully internalized, there is no need to apply comdat rules to its
members; later optimization passes (e.g. globaldce) can legally drop individual
members of the comdat. So we drop the comdat attribute from all comdat members.
Differential Revision: http://reviews.llvm.org/D10679
llvm-svn: 242423
Self-referential constants containing references to a merged function
no longer cause the MergeFunctions pass to infinite loop. Also adds a
reproduction IR which would otherwise fail, which was isolated from a similar
issue in Chromium.
Author: jrkoenig
Reviewers: nlewycky, jfb
Subscribers: llvm-commits, nlewycky, jfb
Differential Revision: http://reviews.llvm.org/D11208
llvm-svn: 242337
During estimation of unrolling effect we should be able to propagate
constants through casts.
Differential Revision: http://reviews.llvm.org/D10207
llvm-svn: 242257
Sometimes an incidentally created instruction can duplicate a Value used
elsewhere. It then often doesn't end up in the leader table. If it's later
removed, we attempt to remove it from the leader table and segfault.
Instead we should just ignore the removal request, which won't cause any
problems. The reverse situation, where the original instruction is replaced by
the new one (which you might think could leave the leader table empty) cannot
occur, because the incidental instruction will never be found in the first
place.
llvm-svn: 242199
Volatile loads and stores are made visible in global state regardless of
what memory is involved. It is not correct to disregard the ordering
and synchronization scope because it is possible to synchronize with
memory operations performed by hardware.
This partially addresses PR23737.
llvm-svn: 242126
Previously we would refrain from attempting to increase the linkage of
available_externally globals because they were considered weak for the
linker. Now they are treated more like a declaration instead of a weak
definition.
This was causing SSE alignment faults in Chromuim, when some code
assumed it could increase the alignment of a dllimported global that it
didn't control. http://crbug.com/509256
llvm-svn: 242091
This test case was breaking the hexagon elf bot. The failing lines
were actually unnecessary as checking that the store still reads the
correct value demonstrates that everything is working fine now.
llvm-svn: 242073
When spotting that a loop can use ctpop, we were incorrectly replacing all uses of a value with a value derived from ctpop.
The bug here was exposed because we were replacing a use prior to the ctpop with the ctpop value and so we have a use before def, i.e., we changed
%tobool.5 = icmp ne i32 %num, 0
store i1 %tobool.5, i1* %ptr
br i1 %tobool.5, label %for.body.lr.ph, label %for.end
to
store i1 %1, i1* %ptr
%0 = call i32 @llvm.ctpop.i32(i32 %num)
%1 = icmp ne i32 %0, 0
br i1 %1, label %for.body.lr.ph, label %for.end
Even if we inserted the ctpop so that it dominates the store here, that would still be incorrect. The store doesn’t want the result of ctpop.
The fix is very simple, and involves replacing only the branch condition with the ctpop instead of all uses.
Reviewed by Hal Finkel.
llvm-svn: 242068
Enable runtime unrolling for loops with unroll count metadata ("#pragma unroll N")
and a runtime trip count. Also, do not unroll loops with unroll full metadata if the
loop has a runtime loop count. Previously, such loops would be unrolled with a
very large threshold (pragma-unroll-threshold) if runtime unrolled happened to be
enabled resulting in a very large (and likely unwise) unroll factor.
llvm-svn: 242047
Summary:
This at least saves compile time. I also encountered a case where
ephemeral values affect whether other variables are promoted, causing
performance issues. It may be a bug in LSR, but I didn't manage to
reduce it yet. Anyhow, I believe it's in general not worth considering
ephemeral values in LSR.
Reviewers: atrick, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11115
llvm-svn: 242011
There is no suitable basic block to sink instructions in loops without
exits. The only way an instruction in a loop without exits can be used
is as an incoming value to a PHI. In such cases, the incoming block for
the corresponding value is unreachable.
This fixes PR24013.
Differential Revision: http://reviews.llvm.org/D10903
llvm-svn: 241987
After changes in rL231820 loop re-rotation is performed even in -Oz mode. Since loop rotation is disabled for -Oz, it seems loop re-rotation should be disabled too.
Differential Revision: http://reviews.llvm.org/D10961
llvm-svn: 241897
Not doing this can lead to misoptimizations down the line, e.g. because
of range metadata on the replacing load excluding values that are valid
for the load that is being replaced.
llvm-svn: 241886
Summary:
In RewriteLoopExitValues, before expanding out an SCEV expression using
SCEVExpander, try to see if an existing LLVM IR expression already
computes the value we're interested in. If so use that existing
expression.
Apart from reducing IndVars' reliance on the rest of the compilation
pipeline, this also prevents IndVars from concluding some expressions as
"high cost" when they're not. For instance,
`InductiveRangeCheckElimination` often emits code of the following form:
```
len = umin(len_A, len_B)
loop:
...
if (i++ < len)
goto loop
outside_loop:
use(i)
```
`SCEVExpander` refuses to rewrite the use of `i` in `outside_loop`,
since it thinks the value of `i` on loop exit, `len`, is a high cost
expansion since it contains an `umax` in it. With this change,
`IndVars` can see that it can re-use `len` instead of creating a new
expression to compute `umin(len_A, len_B)`.
I considered putting this cleverness in `SCEVExpander`, but I was
worried that it may then have a deterimental effect on other passes
that use it. So I decided it was better to just do this in the one
place where it seems like an obviously good idea, with the intent of
generalizing later if needed.
Reviewers: atrick, reames
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D10782
llvm-svn: 241838
Summary:
Often filter-like loops will do memory accesses that are
separated by constant offsets. In these cases it is
common that we will exceed the threshold for the
allowable number of checks.
However, it should be possible to merge such checks,
sice a check of any interval againt two other intervals separated
by a constant offset (a,b), (a+c, b+c) will be equivalent with
a check againt (a, b+c), as long as (a,b) and (a+c, b+c) overlap.
Assuming the loop will be executed for a sufficient number of
iterations, this will be true. If not true, checking against
(a, b+c) is still safe (although not equivalent).
As long as there are no dependencies between two accesses,
we can merge their checks into a single one. We use this
technique to construct groups of accesses, and then check
the intervals associated with the groups instead of
checking the accesses directly.
Reviewers: anemet
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D10386
llvm-svn: 241673
Summary:
Initially, these intrinsics seemed like part of a family of "frame"
related intrinsics, but now I think that's more confusing than helpful.
Initially, the LangRef specified that this would create a new kind of
allocation that would be allocated at a fixed offset from the frame
pointer (EBP/RBP). We ended up dropping that design, and leaving the
stack frame layout alone.
These intrinsics are really about sharing local stack allocations, not
frame pointers. I intend to go further and add an `llvm.localaddress()`
intrinsic that returns whatever register (EBP, ESI, ESP, RBX) is being
used to address locals, which should not be confused with the frame
pointer.
Naming suggestions at this point are welcome, I'm happy to re-run sed.
Reviewers: majnemer, nicholas
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11011
llvm-svn: 241633
