Now, that we have funnel shift intrinsics, it should be safe to convert this form of rotate to it.
In the worst case (a target that doesn't have rotate instructions), we will expand this into a
branch-less sequence of ALU ops (neg/and/and/lshr/shl/or) in the backend, so it's still very
likely to be a perf improvement over the original code.
The motivating source code pattern for this is shown in:
https://bugs.llvm.org/show_bug.cgi?id=34924
Background:
I looked at several different options before deciding where to try this - instcombine, simplifycfg,
CGP - because it doesn't fit cleanly anywhere AFAIK.
The backend (CGP, SDAG, GlobalIsel?) is too late for what we're trying to accomplish. We want to
have the IR converted before we reach things like vectorization because the reduced code can make a
loop much simpler to transform.
Technically, this could be included in instcombine, but it's a large pattern match that includes
control-flow, so it just felt wrong to stuff into there (although I have a draft of that patch).
Similarly, this could be part of simplifycfg, but all of this pattern matching is a stretch.
So we're left with our relatively new dumping ground for homeless transforms: aggressive-instcombine.
This only runs at -O3, but that seems like a reasonable limitation given that source code has many
options to avoid this pattern (including the recently added clang intrinsics for rotates).
I'm including a PhaseOrdering test because we require the teamwork of 3 passes (aggressive-instcombine,
instcombine, simplifycfg) to get this into the minimal IR form that we want. That test shows a bug
with the new pass manager that's independent of this change (but it will be masked if we canonicalize
harder to funnel shift intrinsics in instcombine).
Differential Revision: https://reviews.llvm.org/D55604
llvm-svn: 349396
The problem is shown specifically for a case with vector multiply here:
https://bugs.llvm.org/show_bug.cgi?id=40032
...and this might mask the original backend bug for ARM shown in:
https://bugs.llvm.org/show_bug.cgi?id=39967
As the test diffs here show, we were (and probably still aren't) doing
these kinds of transforms in a principled way. We are producing more or
equal wide instructions than we started with in some cases, so we still
need to restrict/correct other transforms from overstepping.
If there are perf regressions from this change, we can either carve out
exceptions to the general IR rules, or improve the backend to do these
transforms when we know the transform is profitable. That's probably
similar to a change like D55448.
Differential Revision: https://reviews.llvm.org/D55744
llvm-svn: 349389
If a saturating add/sub has one constant operand, then we can
determine the possible range of outputs it can produce, and simplify
an icmp comparison based on that.
The implementation is based on a similar existing mechanism for
simplifying binary operator + icmps.
Differential Revision: https://reviews.llvm.org/D55735
llvm-svn: 349369
If a saturating add/sub with a constant operand is compared to
another constant, we should be able to determine that the condition
is always true/false in some cases (but currently don't).
llvm-svn: 349261
The current code relies on LeaderUseCount to determine if we can remove
an SSA copy, but in that the LeaderUseCount does not refer to the SSA
copy. If a SSA copy is a dominating leader, we use the operand as dominating
leader instead. This means we removed a user of a ssa copy and we should
decrement its use count, so we can remove the ssa copy once it becomes dead.
Fixes PR38804.
Reviewers: efriedma, davide
Reviewed By: davide
Differential Revision: https://reviews.llvm.org/D51595
llvm-svn: 349217
When converting dbg.declares, if the described value is a [s|z]ext,
refer to the ext directly instead of referring to its operand.
This fixes a narrowing bug (the debugger got the sign of a variable
wrong, see llvm.org/PR35400).
The main reason to refer to the ext's operand was that an optimization
may remove the ext itself, leading to a dropped variable. Now that
InstCombine has been taught to use replaceAllDbgUsesWith (r336451), this
is less of a concern. Other passes can/should adopt this API as needed
to fix dropped variable bugs.
Differential Revision: https://reviews.llvm.org/D51813
llvm-svn: 349214
Optimization transformations are intentionally disabled by the 'optnone'
function attribute. Therefore do not warn if transformation metadata is
still present.
Using the legacy pass manager structure, the `skipFunction` method takes
care for the optnone attribute (already called before this patch). For
the new pass manager, there is no equivalent, so we check for the
'optnone' attribute manually.
Differential Revision: https://reviews.llvm.org/D55690
llvm-svn: 349184
The `changeToCall` function did not preserve the invoke's metadata.
Currently, there is probably no metadata that depends on being applied
on a CallInst or InvokeInst. Therefore we can replace the instruction's
metadata.
This fixes http://llvm.org/PR39994
Suggested-by: Moritz Kreutzer <moritz.kreutzer@siemens.com>
Differential Revision: https://reviews.llvm.org/D55666
llvm-svn: 349170
ProfileSampleAccurate is used to indicate the profile has exact match to the
code to be optimized.
Previously ProfileSampleAccurate is handled in ProfileSummaryInfo::isColdCallSite
and ProfileSummaryInfo::isColdBlock. A better solution is to initialize function
entry count to 0 when ProfileSampleAccurate is true, so we don't have to handle
ProfileSampleAccurate in multiple places.
Differential Revision: https://reviews.llvm.org/D55660
llvm-svn: 349088
Currently memcpyopt optimizes cases like
memset(a, byte, N);
memcpy(b, a, M);
to
memset(a, byte, N);
memset(b, byte, M);
if M <= N. Often this allows further simplifications down the line,
which drop the first memset entirely.
This patch extends this optimization for the case where M > N, but we
know that the bytes a[N..M] are undef due to alloca/lifetime.start.
This situation arises relatively often for Rust code, because Rust does
not initialize trailing structure padding and loves to insert redundant
memcpys. This also fixes https://bugs.llvm.org/show_bug.cgi?id=39844.
The previous version of this patch did not perform dependency checking
properly: While the dependency is checked at the position of the memset,
the used size must be that of the memcpy. Previously the size of the
memset was used, which missed modification in the region
MemSetSize..CopySize, resulting in miscompiles. The added tests cover
variations of this issue.
Differential Revision: https://reviews.llvm.org/D55120
llvm-svn: 349078
The actual type of the first argument of the @dbg intrinsic
doesn't really matter as we're setting it to `undef`, but the
bitcode reader is picky about `void` types.
llvm-svn: 349069
For anyone curious, the first test example is illustrative of a real code idiom produced by branching on the result of a three way comparison.
llvm-svn: 348997
When loops are deleted, we don't keep track of variables modified inside
the loops, so the DI will contain the wrong value for these.
e.g.
int b() {
int i;
for (i = 0; i < 2; i++)
;
patatino();
return a;
-> 6 patatino();
7 return a;
8 }
9 int main() { b(); }
(lldb) frame var i
(int) i = 0
We mark instead these values as unavailable inserting a
@llvm.dbg.value(undef to make sure we don't end up printing an incorrect
value in the debugger. We could consider doing something fancier,
for, e.g. constants, in the future.
PR39868.
rdar://problem/46418795)
Differential Revision: https://reviews.llvm.org/D55299
llvm-svn: 348988
This fixes https://bugs.llvm.org/show_bug.cgi?id=39908.
The evaluateGEPOffsetExpression() function simplifies GEP offsets for
use in comparisons against zero, basically by converting X*Scale+Offset==0
to X+Offset/Scale==0 if Scale divides Offset. However, before this is done,
Offset is masked down to the pointer size. This results in incorrect
results for negative Offsets, because we basically end up dividing the
32-bit offset *zero* extended to 64-bit bits (rather than sign extended).
Fix this by explicitly sign extending the truncated value.
Differential Revision: https://reviews.llvm.org/D55449
llvm-svn: 348987
As mentioned in D55604, there are 2 bugs here:
1. The new pass manager is speculating wildly by default.
2. The old pass manager is not converting this to funnel shift.
llvm-svn: 348980
Indices for getelementptr can be signed so we should use
getMinSignedBits instead of getActiveBits here. The function later calls
getSExtValue to get the int64_t value, which also checks
getMinSignedBits.
This fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=11647.
Reviewers: mssimpso, efriedma, davide
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D55536
llvm-svn: 348957
When multiple loop transformation are defined in a loop's metadata, their order of execution is defined by the order of their respective passes in the pass pipeline. For instance, e.g.
#pragma clang loop unroll_and_jam(enable)
#pragma clang loop distribute(enable)
is the same as
#pragma clang loop distribute(enable)
#pragma clang loop unroll_and_jam(enable)
and will try to loop-distribute before Unroll-And-Jam because the LoopDistribute pass is scheduled after UnrollAndJam pass. UnrollAndJamPass only supports one inner loop, i.e. it will necessarily fail after loop distribution. It is not possible to specify another execution order. Also,t the order of passes in the pipeline is subject to change between versions of LLVM, optimization options and which pass manager is used.
This patch adds 'followup' attributes to various loop transformation passes. These attributes define which attributes the resulting loop of a transformation should have. For instance,
!0 = !{!0, !1, !2}
!1 = !{!"llvm.loop.unroll_and_jam.enable"}
!2 = !{!"llvm.loop.unroll_and_jam.followup_inner", !3}
!3 = !{!"llvm.loop.distribute.enable"}
defines a loop ID (!0) to be unrolled-and-jammed (!1) and then the attribute !3 to be added to the jammed inner loop, which contains the instruction to distribute the inner loop.
Currently, in both pass managers, pass execution is in a fixed order and UnrollAndJamPass will not execute again after LoopDistribute. We hope to fix this in the future by allowing pass managers to run passes until a fixpoint is reached, use Polly to perform these transformations, or add a loop transformation pass which takes the order issue into account.
For mandatory/forced transformations (e.g. by having been declared by #pragma omp simd), the user must be notified when a transformation could not be performed. It is not possible that the responsible pass emits such a warning because the transformation might be 'hidden' in a followup attribute when it is executed, or it is not present in the pipeline at all. For this reason, this patche introduces a WarnMissedTransformations pass, to warn about orphaned transformations.
Since this changes the user-visible diagnostic message when a transformation is applied, two test cases in the clang repository need to be updated.
To ensure that no other transformation is executed before the intended one, the attribute `llvm.loop.disable_nonforced` can be added which should disable transformation heuristics before the intended transformation is applied. E.g. it would be surprising if a loop is distributed before a #pragma unroll_and_jam is applied.
With more supported code transformations (loop fusion, interchange, stripmining, offloading, etc.), transformations can be used as building blocks for more complex transformations (e.g. stripmining+stripmining+interchange -> tiling).
Reviewed By: hfinkel, dmgreen
Differential Revision: https://reviews.llvm.org/D49281
Differential Revision: https://reviews.llvm.org/D55288
llvm-svn: 348944
For SampleFDO, when a callsite doesn't appear in the profile, it will not be marked as cold callsite unless the option -profile-sample-accurate is specified.
But profile-sample-accurate doesn't cover function isFunctionColdInCallGraph which is used to decide whether a function should be put into text.unlikely section, so even if the user knows the profile is accurate and specifies profile-sample-accurate, those functions not appearing in the sample profile are still not be put into text.unlikely section right now.
The patch fixes that.
Differential Revision: https://reviews.llvm.org/D55567
llvm-svn: 348940
Without this check, we hit an assertion in getZExtValue, if the constant
value does not fit into an uint64_t.
As getZExtValue returns an uint64_t, should we update
getAggregateElement to take an uin64_t as well?
This fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=6109.
Reviewers: efriedma, craig.topper, spatel
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D55547
llvm-svn: 348906
Summary:
Enable suspend point simplification for cases where:
* coro.save and coro.suspend are in different basic blocks
* where there are intervening intrinsics
Reviewers: modocache, tks2103, lewissbaker
Reviewed By: modocache
Subscribers: EricWF, llvm-commits
Differential Revision: https://reviews.llvm.org/D55160
llvm-svn: 348897
Struct types may have leading zero-size elements like [0 x i32], in
which case the "real" element at offset 0 will not necessarily coincide
with the 0th element of the aggregate. ConstantFoldLoadThroughBitcast()
wants to drill down the element at offset 0, but currently always picks
the 0th aggregate element to do so. This patch changes the code to find
the first non-zero-size element instead, for the struct case.
The motivation behind this change is https://github.com/rust-lang/rust/issues/48627.
Rust is fond of emitting [0 x iN] separators between struct elements to
enforce alignment, which prevents constant folding in this particular case.
The additional tests with [4294967295 x [0 x i32]] check that we don't
end up unnecessarily looping over a large number of zero-size elements
of a zero-size array.
Differential Revision: https://reviews.llvm.org/D55169
llvm-svn: 348895
It's currently not safe to outline landingpad instructions (see
llvm.org/PR39917). Like @llvm.eh.typeid.for, the order and content of
previous landingpad instructions in a function alters the lowering of
subsequent landingpads by renumbering type info ID's. Outlining a
landingpad therefore breaks exception handling & unwinding.
llvm-svn: 348870
call iM movmsk(sext <N x i1> X) --> zext (bitcast <N x i1> X to iN) to iM
This has the potential to create less-than-8-bit scalar types as shown in
some of the test diffs, but it looks like the backend knows how to deal
with that in these patterns. This is the simple part of the fix suggested in:
https://bugs.llvm.org/show_bug.cgi?id=39927
Differential Revision: https://reviews.llvm.org/D55529
llvm-svn: 348862
These involve cases where certain uses are dead by means of having
no demanded bits, even though the used instruction still has demanded
bits when other uses are taken into account. BDCE currently does not
simplify such cases.
llvm-svn: 348861
Summary:
When eliminating a dead argument or return value in a function with
local linkage, all uses, including in dbg.value intrinsics, would be
replaced with null constants. This would mean that, for example for an
integer argument, the debug info would incorrectly express that the
value is 0. Instead, replace all uses with undef to indicate that the
argument/return value is optimized out.
Also, make sure that metadata uses of return values are rewritten even
if there are no non-metadata uses of the value.
As a bit of historical curiosity, the code that emitted null constants
was introduced in the initial check-in of the pass in 2003, before
'undef' values even existed in LLVM.
This fixes PR23260.
Reviewers: dblaikie, aprantl, vsk, djtodoro
Reviewed By: aprantl
Subscribers: llvm-commits
Tags: #debug-info
Differential Revision: https://reviews.llvm.org/D55513
llvm-svn: 348837
Summary: The comment says we need 3 extracts and a select at the end. But didn't we just account for the select in the vector cost above. Aren't we just extracting the single element after taking the min/max in the vector register?
Reviewers: RKSimon, spatel, ABataev
Reviewed By: RKSimon
Subscribers: javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D55480
llvm-svn: 348739
Currently memcpyopt optimizes cases like
memset(a, byte, N);
memcpy(b, a, M);
to
memset(a, byte, N);
memset(b, byte, M);
if M <= N. Often this allows further simplifications down the line,
which drop the first memset entirely.
This patch extends this optimization for the case where M > N, but we
know that the bytes a[N..M] are undef due to alloca/lifetime.start.
This situation arises relatively often for Rust code, because Rust does
not initialize trailing structure padding and loves to insert redundant
memcpys. This also fixes https://bugs.llvm.org/show_bug.cgi?id=39844.
For the implementation, I'm reusing a bit of code for a similar existing
optimization (direct memcpy of undef). I've also added memset support to
MemDepAnalysis GetLocation -- Instead, getPointerDependencyFrom could be
used, but it seems to make more sense to add this to GetLocation and thus
make the computation cachable.
Differential Revision: https://reviews.llvm.org/D55120
llvm-svn: 348645
The splitting pass uses its 'unlikelyExecuted' predicate to statically
decide which blocks are cold.
- Do not treat noreturn calls as if they are cold unless they are actually
marked cold. This is motivated by functions like exit() and longjmp(), which
are not beneficial to outline.
- Do not treat inline asm as an outlining barrier. In practice asm("") is
frequently used to inhibit basic block merging; enabling outlining in this case
results in substantial memory savings.
- Treat invokes of cold functions as cold.
As a drive-by, remove the 'exceptionHandlingFunctions' predicate, because it's
no longer needed. The pass can identify & outline blocks dominated by EH pads,
so there's no need to special-case __cxa_begin_catch etc.
Differential Revision: https://reviews.llvm.org/D54244
llvm-svn: 348640
Algorithm: Identify maximal cold regions and put them in a worklist. If
a candidate region overlaps with another, discard it. While the worklist
is full, remove a single-entry sub-region from the worklist and attempt
to outline it. By the non-overlap property, this should not invalidate
parts of the domtree pertaining to other outlining regions.
Testing: LNT results on X86 are clean. With test-suite + externals, llvm
outlines 134KB pre-patch, and 352KB post-patch (+ ~2.6x). The file
483.xalancbmk/src/Constants.cpp stands out as an extreme case where llvm
outlines over 100 times in some functions (mostly EH paths). There was
not a significant performance impact pre vs. post-patch.
Differential Revision: https://reviews.llvm.org/D53887
llvm-svn: 348639
We were overcounting the number of arithmetic operations needed at each level before we reach a legal type. We were using the full vector type for that level, but we are going to split the input vector at that level in half. So the effective arithmetic operation cost at that level is half the width.
So for example on 8i32 on an sse target. Were were calculating the cost of an 8i32 op which is likely 2 for basic integer. Then after the loop we count 2 more v4i32 ops. For a total arith cost of 4. But if you look at the assembly there would only be 3 arithmetic ops.
There are still more bugs in this code that I'm going to work on next. The non pairwise code shouldn't count extract subvectors in the loop. There are no extracts, the types are split in registers. For pairwise we need to use 2 two src permute shuffles.
Differential Revision: https://reviews.llvm.org/D55397
llvm-svn: 348621
DemandedBits and BDCE currently only support scalar integers. This
patch extends them to also handle vector integer operations. In this
case bits are not tracked for individual vector elements, instead a
bit is demanded if it is demanded for any of the elements. This matches
the behavior of computeKnownBits in ValueTracking and
SimplifyDemandedBits in InstCombine.
Unlike the previous iteration of this patch, getDemandedBits() can now
again be called on arbirary (sized) instructions, even if they don't
have integer or vector of integer type. (For vector types the size of the
returned mask will now be the scalar size in bits though.)
The added LoopVectorize test case shows a case which triggered an
assertion failure with the previous attempt, because getDemandedBits()
was called on a pointer-typed instruction.
Differential Revision: https://reviews.llvm.org/D55297
llvm-svn: 348602
This patch introduces a new instinsic `@llvm.experimental.widenable_condition`
that allows explicit representation for guards. It is an alternative to using
`@llvm.experimental.guard` intrinsic that does not contain implicit control flow.
We keep finding places where `@llvm.experimental.guard` is not supported or
treated too conservatively, and there are 2 reasons to that:
- `@llvm.experimental.guard` has memory write side effect to model implicit control flow,
and this sometimes confuses passes and analyzes that work with memory;
- Not all passes and analysis are aware of the semantics of guards. These passes treat them
as regular throwing call and have no idea that the condition of guard may be used to prove
something. One well-known place which had caused us troubles in the past is explicit loop
iteration count calculation in SCEV. Another example is new loop unswitching which is not
aware of guards. Whenever a new pass appears, we potentially have this problem there.
Rather than go and fix all these places (and commit to keep track of them and add support
in future), it seems more reasonable to leverage the existing optimizer's logic as much as possible.
The only significant difference between guards and regular explicit branches is that guard's condition
can be widened. It means that a guard contains (explicitly or implicitly) a `deopt` block successor,
and it is always legal to go there no matter what the guard condition is. The other successor is
a guarded block, and it is only legal to go there if the condition is true.
This patch introduces a new explicit form of guards alternative to `@llvm.experimental.guard`
intrinsic. Now a widenable guard can be represented in the CFG explicitly like this:
%widenable_condition = call i1 @llvm.experimental.widenable.condition()
%new_condition = and i1 %cond, %widenable_condition
br i1 %new_condition, label %guarded, label %deopt
guarded:
; Guarded instructions
deopt:
call type @llvm.experimental.deoptimize(<args...>) [ "deopt"(<deopt_args...>) ]
The new intrinsic `@llvm.experimental.widenable.condition` has semantics of an
`undef`, but the intrinsic prevents the optimizer from folding it early. This form
should exploit all optimization boons provided to `br` instuction, and it still can be
widened by replacing the result of `@llvm.experimental.widenable.condition()`
with `and` with any arbitrary boolean value (as long as the branch that is taken when
it is `false` has a deopt and has no side-effects).
For more motivation, please check llvm-dev discussion "[llvm-dev] Giving up using
implicit control flow in guards".
This patch introduces this new intrinsic with respective LangRef changes and a pass
that converts old-style guards (expressed as intrinsics) into the new form.
The naming discussion is still ungoing. Merging this to unblock further items. We can
later change the name of this intrinsic.
Reviewed By: reames, fedor.sergeev, sanjoy
Differential Revision: https://reviews.llvm.org/D51207
llvm-svn: 348593
The current algorithm that collects live/dead/inloop blocks relies on some invariants
related to RPO and PO traversals. In particular, the important fact it requires is that
the only loop's latch is the first block in PO traversal. It also relies on fact that during
RPO we visit all prececessors of a block before we visit this block (backedges ignored).
If a loop has irreducible non-loop cycle inside, both these assumptions may break.
This patch adds detection for this situation and prohibits the terminator folding
for loops with irreducible CFG.
We can in theory support this later, for this some algorithmic changes are needed.
Besides, irreducible CFG is not a frequent situation and we can just don't bother.
Thanks @uabelho for finding this!
Differential Revision: https://reviews.llvm.org/D55357
Reviewed By: skatkov
llvm-svn: 348567
DemandedBits and BDCE currently only support scalar integers. This
patch extends them to also handle vector integer operations. In this
case bits are not tracked for individual vector elements, instead a
bit is demanded if it is demanded for any of the elements. This matches
the behavior of computeKnownBits in ValueTracking and
SimplifyDemandedBits in InstCombine.
The getDemandedBits() method can now only be called on instructions that
have integer or vector of integer type. Previously it could be called on
any sized instruction (even if it was not particularly useful). The size
of the return value is now always the scalar size in bits (while
previously it was the type size in bits).
Differential Revision: https://reviews.llvm.org/D55297
llvm-svn: 348549
Partial Redundancy Elimination of GEPs prevents CodeGenPrepare from
sinking the addressing mode computation of memory instructions back
to its uses. The problem comes from the insertion of PHIs, which
confuse CGP and make it bail.
I've autogenerated the check lines of an existing test and added a
store instruction to demonstrate the motivation behind this change.
The store is now using the gep instead of a phi.
Differential Revision: https://reviews.llvm.org/D55009
llvm-svn: 348496
This reverts commit r348457.
The original commit causes clang to crash when doing an instrumented
build with a new pass manager. Reverting to unbreak our integrate.
llvm-svn: 348484
I was finally able to quantify what i thought was missing in the fix,
it was vector constants. If we have a scalar (and %x, -1),
it will be instsimplified before we reach this code,
but if it is a vector, we may still have a -1 element.
Thus, we want to avoid the fold if *at least one* element is -1.
Or in other words, ignoring the undef elements, no sign bits
should be set. Thus, m_NonNegative().
A follow-up for rL348181
https://bugs.llvm.org/show_bug.cgi?id=39861
llvm-svn: 348462
This patch teaches LoopSimplifyCFG to delete loop blocks that have
become unreachable after terminator folding has been done.
Differential Revision: https://reviews.llvm.org/D54023
Reviewed By: anna
llvm-svn: 348457
This change caused SEGVs in instcombine. (The r347934 change seems to me to be a
precipitating cause, not a root cause. Details are on the llvm-commits thread
for r347934.)
llvm-svn: 348426
Extracting from a splat constant is always handled by InstSimplify.
Move the test for this from InstCombine to InstSimplify to make
sure that stays true.
llvm-svn: 348423
Treat terminators which resume exception propagation as returning instructions
(at least, for the purposes of marking outlined functions `noreturn`). This is
to avoid inserting traps after calls to outlined functions which unwind.
rdar://46129950
llvm-svn: 348404
Summary: debug intrinsics might be marked norecurse to enable the caller function to be norecurse and optimized if needed. This avoids code gen optimisation differences when -g is used, as in globalOpt.cpp:processInternalGlobal checks.
Reviewers: chandlerc, jmolloy, aprantl
Reviewed By: aprantl
Subscribers: aprantl, llvm-commits
Differential Revision: https://reviews.llvm.org/D55187
llvm-svn: 348381
The tests here are based on the motivating cases from D54827.
More background:
1. We don't get these cases in general with SimplifyCFG because the root
of the pattern match is an icmp, not a branch. I'm not sure how often
we encounter this pattern vs. the seemingly more likely case with
branches, but I don't see evidence to leave the minimal pattern
unoptimized.
2. This has a chance of increasing compile-time because we're using a
ValueTracking call to handle the match. The motivating cases could be
handled with a simpler pair of calls to isImpliedTrueByMatchingCmp/
isImpliedFalseByMatchingCmp, but I saw that we have a more
comprehensive wrapper around those, so we might as well use it here
unless there's evidence that it's significantly slower.
3. Ideally, we'd handle the fold to constants in InstSimplify, but as
with the existing code here, we could extend this to handle cases
where the result is not a constant, but a new combined predicate.
That would mean splitting the logic across the 2 passes and possibly
duplicating the pattern-matching cost.
4. As mentioned in D54827, this seems like the kind of thing that should
be handled in Correlated Value Propagation, but that pass is currently
limited to dealing with instructions with constant operands, so extending
this bit of InstCombine is the smallest/easiest way to get these patterns
optimized.
llvm-svn: 348367
Ideally, we would fold all of these in InstSimplify in a
similar way to rL347896, but this is a bit awkward when
we're trying to simplify a compare directly because the
ValueTracking API expects the compare as an input, but
in InstSimplify, we just have the operands of the compare.
Given that we can do transforms besides just simplifications,
we might as well just extend the code in InstCombine (which
already does simplifications with constant operands).
llvm-svn: 348312
If a PHI node out of extracted region has multiple incoming values from it,
split this PHI on two parts. First PHI has incomings only from region and
extracts with it (they are placed to the separate basic block that added to the
list of outlined), and incoming values in original PHI are replaced by first
PHI. Similar solution is already used in CodeExtractor for PHIs in entry block
(severSplitPHINodes method). It covers PR39433 bug.
Patch by Sergei Kachkov!
Differential Revision: https://reviews.llvm.org/D55018
llvm-svn: 348205
When we have a shuffle that extends a source vector with undefs
and then do some binop on that, we must make sure that the extra
elements remain undef with that binop if we reverse the order of
the binop and shuffle.
'or' is probably the easiest example to show the bug because
'or C, undef --> -1' (not undef). But there are other
opcode/constant combinations where this is true as shown by
the 'shl' test.
llvm-svn: 348191
These are the baseline tests for D54827.
Patch based on code originally written by: @yinyuefengyi (luo xionghu)
Differential Revision: https://reviews.llvm.org/D54994
llvm-svn: 348151
Add '-k 1' to 'sort -b' calls in SimpleLoopUnswitch tests, as required
for sort implementation on NetBSD. The '-b' modifier is ineffective
if specified without any key. Per the manpage:
Note that the -b option has no effect unless key fields are specified.
Differential Revision: https://reviews.llvm.org/D55168
llvm-svn: 348097
We were adding the entire scalarization extraction cost for reductions, which returns the total cost of extracting every element of a vector type.
For reductions we don't need to do this - we just need to extract the 0'th element after the reduction pattern has completed.
Fixes PR37731
Rebased and reapplied after being reverted in rL347541 due to PR39774 - which was fixed by D54955/rL347759 and D55017/rL347997
Differential Revision: https://reviews.llvm.org/D54585
llvm-svn: 348076
Extend ssub.sat(X, C) -> sadd.sat(X, -C) canonicalization to also
support non-splat vector constants. This is done by generalizing
the implementation of the isNotMinSignedValue() helper to return
true for constants that are non-splat, but don't contain any
signed min elements.
Differential Revision: https://reviews.llvm.org/D55011
llvm-svn: 348072
Summary:
Follow up to D54270, which allowed importing of var args functions
unless they called va_start. As pointed out in the post-commit comments
on that patch, the inliner can handle functions that call va_start in
certain situations as well. Go ahead and enable importing of all var
args functions. Measurements on a large binary show that this increases
imports and binary size by an insignificant amount.
Reviewers: davidxl
Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith, llvm-commits
Differential Revision: https://reviews.llvm.org/D54607
llvm-svn: 348068
These tests should probably go under a separate test file because they
should fold with just -constprop, but they're similar to the scalar
tests already in here.
llvm-svn: 348045
Summary:
When mem2reg inserts phi nodes in blocks with unreachable predecessors,
it adds undef operands for those incoming edges. When there are
multiple such predecessors, the order is currently based on the address
of the BasicBlocks. This change fixes that by using the BBNumbers in
the sort/search predicates, as is done elsewhere in mem2reg to ensure
determinism.
Also adds a testcase with a bunch of unreachable preds, which
(nodeterministically) fails without the fix.
Reviewers: majnemer
Reviewed By: majnemer
Subscribers: mgrang, llvm-commits
Differential Revision: https://reviews.llvm.org/D55077
llvm-svn: 348024
Summary:
An additional fix for PR39774. Need to update the references for the
RedcutionRoot instruction when it is replaced during the vectorization
phase to avoid compiler crash on reduction vectorization.
Reviewers: RKSimon, spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D55017
llvm-svn: 347997
Adding a new reduction pattern match for vectorizing code similar
to TSVC s3111:
for (int i = 0; i < N; i++)
if (a[i] > b)
sum += a[i];
This patch adds support for fadd, fsub and fmull, as well as multiple
branches and different (but compatible) instructions (ex. add+sub) in
different branches.
The difference from the previous patch(https://reviews.llvm.org/D49168)
is as follows:
- Added check of fast-math property of fp-instruction to the
previous patch
- Fix/add some pattern for if-reduction.ll
Differential Revision: https://reviews.llvm.org/D54464
Patch by Takahiro Miyoshi <takahiro.miyoshi@linaro.org>
and Masakazu Ueno <masakazu.ueno@linaro.org>
llvm-svn: 347989
Terminator folding transform lacks MemorySSA update for memory Phis,
while they exist within MemorySSA analysis. They need exactly the same
type of updates as regular Phis. Failing to update them properly ends up
with inconsistent MemorySSA and manifests in various assertion failures.
This patch adds Memory Phi updates to this transform.
Thanks to @jonpa for finding this!
Differential Revision: https://reviews.llvm.org/D55050
Reviewed By: asbirlea
llvm-svn: 347979
r320789 suppressed moving the insertion point of SCEV expressions with
dev/rem operations to the loop header in non-loop-invariant situations.
This, and similar, hoisting is also unsafe in the loop-invariant case,
since there may be a guard against a zero denominator. This is an
adjustment to the fix of r320789 to suppress the movement even in the
loop-invariant case.
This fixes PR30806.
Differential Revision: https://reviews.llvm.org/D54713
llvm-svn: 347934
Also revert fix r347876
One of the buildbots was reporting a failure in some relevant tests that I can't
repro or explain at present, so reverting until I can isolate.
llvm-svn: 347911
This is an almost direct move of the functionality from InstCombine to
InstSimplify. There's no reason not to do this in InstSimplify because
we never create a new value with this transform.
(There's a question of whether any dominance-based transform belongs in
either of these passes, but that's a separate issue.)
I've changed 1 of the conditions for the fold (1 of the blocks for the
branch must be the block we started with) into an assert because I'm not
sure how that could ever be false.
We need 1 extra check to make sure that the instruction itself is in a
basic block because passes other than InstCombine may be using InstSimplify
as an analysis on values that are not wired up yet.
The 3-way compare changes show that InstCombine has some kind of
phase-ordering hole. Otherwise, we would have already gotten the intended
final result that we now show here.
llvm-svn: 347896
This commit caused a large compile-time slowdown in some cases when NDEBUG is
off due to the dominator tree verification it added. Fix this by only doing
dominator tree and loop info verification when something has been hoisted.
Differential Revision: https://reviews.llvm.org/D52827
llvm-svn: 347889
Summary:
When splitting musttail calls, the split blocks' original terminators
get removed; inform the DTU when this happens.
Also add a testcase that fails an assertion in the DTU without this fix.
Reviewers: fhahn, junbuml
Reviewed By: fhahn
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D55027
llvm-svn: 347872
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
llvm-svn: 347871
This reverts commits r347776 and r347778.
The first one, r347776, caused significant compile time regressions
for certain input files, see PR39836 for details.
llvm-svn: 347867
Lack of an attribute means that the function hasn't been checked for what vector width it requires. So if the caller or the callee doesn't have the attribute we should make sure the combined function after inlining does not have the attribute.
If the caller already doesn't have the attribute we can just avoid adding it. Otherwise if the callee doesn't have the attribute just remove the caller's attribute.
llvm-svn: 347841
Packing the flags into one bitcode word will save effort in
adding new flags in the future.
Differential Revision: https://reviews.llvm.org/D54755
llvm-svn: 347806
In PR39807 we incorrectly handle circumstances where calls are common'd
from conditional blocks into the parent BB. Calls that can be inlined
must always have DebugLocs, however we strip them during commoning, which
the IR verifier asserts on.
Fix this by using applyMergedLocation: it will perform the same DebugLoc
stripping of conditional Locs, but will also generate an unknown location
DebugLoc that satisfies the requirement for inlinable calls to always have
locations.
Some of the prior logic for selecting a DebugLoc is now likely redundant;
I'll generate a follow-up to remove it (involves editing more regression
tests).
Differential Revision: https://reviews.llvm.org/D54997
llvm-svn: 347782
This commit caused failures because it failed to correctly handle cases where
we hoist a phi, then hoist a use of that phi, then have to rehoist that use. We
need to make sure that we rehoist the use to _after_ the hoisted phi, which we
do by always rehoisting to the immediate dominator instead of just rehoisting
everything to the original preheader.
An option is also added to control whether control flow is hoisted, which is
off in this commit but will be turned on in a subsequent commit.
Differential Revision: https://reviews.llvm.org/D52827
llvm-svn: 347776
Combine
sat(sat(X + C1) + C2) -> sat(X + (C1+C2))
and
sat(sat(X - C1) - C2) -> sat(X - (C1+C2))
if the sign of C1 and C2 matches.
In the unsigned case we can compute C1+C2 with saturating arithmetic,
and InstSimplify will reduce this just to the saturation value. For
the signed case, we cannot perform the simplification if the result
of the addition overflows.
This change is part of https://reviews.llvm.org/D54534.
llvm-svn: 347773
Canonicalize ssub.sat(X, C) to ssub.sat(X, -C) if C is constant and
not signed minimum. This will help further optimizations to apply.
This change is part of https://reviews.llvm.org/D54534.
llvm-svn: 347772
Always-overflow was already determined for unsigned addition, but
not subtraction. This patch establishes parity.
This allows us to perform some additional simplifications for
signed saturating subtractions.
This change is part of https://reviews.llvm.org/D54534.
llvm-svn: 347771
If ValueTracking can determine that the add/sub can newer overflow,
replace it with the corresponding nuw/nsw add/sub.
Additionally, for the unsigned case, if ValueTracking determines
that the add/sub always overflows, replace the result with the
saturation value.
This change is part of https://reviews.llvm.org/D54534.
llvm-svn: 347770
If a saturating add intrinsic has one constant argument, make sure
it is on the RHS. This will simplify further transformations.
This change is part of https://reviews.llvm.org/D54534.
llvm-svn: 347769
Summary:
If the original reduction root instruction was vectorized, it might be
removed from the tree. It means that the insertion point may become
invalidated and the whole vectorization of the reduction leads to the
incorrect output result.
The ReductionRoot instruction must be marked as externally used so it
could not be removed. Otherwise it might cause inconsistency with the
cost model and we may end up with too optimistic optimization.
Reviewers: RKSimon, spatel, hfinkel, mkuper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54955
llvm-svn: 347759
InlineCost also treats them as free and the current implementation
can cause assertion failures if PHI nodes are moved outside the region
from entry BBs to the region.
It also updates the code to use the instructionsWithoutDebug iterator.
Reviewers: davidxl, davide, vsk, graham-yiu-huawei
Reviewed By: davidxl
Differential Revision: https://reviews.llvm.org/D54748
llvm-svn: 347683
It fixes a bug that doesn't update Phi inputs of the only live successor that
is in the list of block's successors more than once.
Thanks @uabelho for finding this.
Differential Revision: https://reviews.llvm.org/D54849
Reviewed By: anna
llvm-svn: 347640
Summary:
Removing ncompatible attributes at indirect-call promoted callsites, not removing it results in
at least a IR verification error.
Reviewers: davidxl, xur, mssimpso
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54913
llvm-svn: 347605
This reverts commit r346970.
It was causing PR39774, a crash in slp-vectorizer on a rather simple loop
with just a bunch of 'and's in the body.
llvm-svn: 347541
OriginalOp of a Predicate refers to the original IR value,
before renaming. While solving in IPSCCP, we have to use
the operand of the ssa_copy instead, to avoid missing
updates for nested conditions on the same IR value.
Fixes PR39772.
llvm-svn: 347524
Support funnel shifts in InstCombine demanded bits simplification.
If the shift amount is constant, we can determine both the demanded
bits of the operands, as well as the known bits of the result.
If one of the operands has no demanded bits, it will be replaced
by undef and the funnel shift will be simplified into a simple shift
due to the simplifications added in D54778.
Differential Revision: https://reviews.llvm.org/D54869
llvm-svn: 347515
This changeset is modeled after Intel's submission for SVML. It enables
trigonometry functions vectorization via SLEEF: http://sleef.org/.
* A new vectorization library enum is added to TargetLibraryInfo.h: SLEEF.
* A new option is added to TargetLibraryInfoImpl - ClVectorLibrary: SLEEF.
* A comprehensive test case is included in this changeset.
* In a separate changeset (for clang), a new vectorization library argument is
added to -fveclib: -fveclib=SLEEF.
Trigonometry functions that are vectorized by sleef:
acos
asin
atan
atanh
cos
cosh
exp
exp2
exp10
lgamma
log10
log2
log
sin
sinh
sqrt
tan
tanh
tgamma
Patch by Stefan Teleman
Differential Revision: https://reviews.llvm.org/D53927
llvm-svn: 347510
The following simplifications are implemented:
* `fshl(X, 0, C) -> shl X, C%BW`
* `fshl(X, undef, C) -> shl X, C%BW` (assuming undef = 0)
* `fshl(0, X, C) -> lshr X, BW-C%BW`
* `fshl(undef, X, C) -> lshr X, BW-C%BW` (assuming undef = 0)
* `fshr(X, 0, C) -> shl X, (BW-C%BW)`
* `fshr(X, undef, C) -> shl X, BW-C%BW` (assuming undef = 0)
* `fshr(0, X, C) -> lshr X, C%BW`
* `fshr(undef, X, C) -> lshr, X, C%BW` (assuming undef = 0)
The simplification is only performed if the shift amount C is constant,
because we can explicitly compute C%BW and BW-C%BW in this case.
Differential Revision: https://reviews.llvm.org/D54778
llvm-svn: 347505
When removing edges, we also update Phi inputs and may end up removing
a Phi if it has only one input. We should not do it for edges that leave the current
loop because these Phis are LCSSA Phis and need to be preserved.
Thanks @dmgreen for finding this!
Differential Revision: https://reviews.llvm.org/D54841
llvm-svn: 347484
The MergeFunctions pass was originally intended to emit aliases
instead of thunks where possible (unnamed_addr). However, for a
long time this functionality was behind a flag hardcoded to false,
bitrotted and was eventually removed in r309313.
Originally the functionality was first disabled in r108417 due to
lack of support for aliases in Mach-O. I believe that this is no
longer the case nowadays, but not really familiar with this area.
In the interest of being conservative, this patch reintroduces the
aliasing functionality behind a default disabled -mergefunc-use-aliases
flag.
Differential Revision: https://reviews.llvm.org/D53285
llvm-svn: 347407
LVI was symbolically executing binary operators only when the RHS was
constant, missing the case where we have a ConstantRange for the RHS,
but not an actual constant. Tested using check-all and by
bootstrapping. Compile time is not impacted measurably.
Differential Revision: https://reviews.llvm.org/D19859
llvm-svn: 347379
These are part of D54666, so adding them here before the patch to
show the baseline (currently unoptimized) results.
Patch by: @nikic (Nikita Popov)
llvm-svn: 347331
This patch fixes PR39695.
The original LoopSink only considers memory alias in loop body. But PR39695 shows that instructions following sink candidate in preheader should also be checked. This is a conservative patch, it simply adds whole preheader block to alias set. It may lose some optimization opportunity, but I think that is very rare because: 1 in the most common case st/ld to the same address, the load should already be optimized away. 2 usually preheader is not very large.
Differential Revision: https://reviews.llvm.org/D54659
llvm-svn: 347325
This patch fixes the issue noticed in D54532.
The problem is that cst_pred_ty-based matchers like m_Zero() currently do not match
scalar undefs (as expected), but *do* match vector undefs. This may lead to optimization
inconsistencies in rare cases.
There is only one existing test for which output changes, reverting the change from D53205.
The reason here is that vector fsub undef, %x is no longer matched as an m_FNeg(). While I
think that the new output is technically worse than the previous one, it is consistent with
scalar, and I don't think it's really important either way (generally that undef should have
been folded away prior to reassociation.)
I've also added another test case for this issue based on InstructionSimplify. It took some
effort to find that one, as in most cases undef folds are either checked first -- and in the
cases where they aren't it usually happens to not make a difference in the end. This is the
only case I was able to come up with. Prior to this patch the test case simplified to undef
in the scalar case, but zeroinitializer in the vector case.
Patch by: @nikic (Nikita Popov)
Differential Revision: https://reviews.llvm.org/D54631
llvm-svn: 347318
The initial version of patch lacked Phi nodes updates in destinations of removed
edges. This version contains this update and tests on this situation.
Differential Revision: https://reviews.llvm.org/D54021
llvm-svn: 347289
Summary:
Currently, when vectorizing stores to uniform addresses, the only
instance we prevent vectorization is if there are multiple stores to the
same uniform address causing an unsafe dependency.
This patch teaches LAA to avoid vectorizing loops that have an unsafe
cross-iteration dependency between a load and a store to the same uniform address.
Fixes PR39653.
Reviewers: Ayal, efriedma
Subscribers: rkruppe, llvm-commits
Differential Revision: https://reviews.llvm.org/D54538
llvm-svn: 347220
The general approach taken is to make note of loop invariant branches, then when
we see something conditional on that branch, such as a phi, we create a copy of
the branch and (empty versions of) its successors and hoist using that.
This has no impact by itself that I've been able to see, as LICM typically
doesn't see such phis as they will have been converted into selects by the time
LICM is run, but once we start doing phi-to-select conversion later it will be
important.
Differential Revision: https://reviews.llvm.org/D52827
llvm-svn: 347190
This patch introduces infrastructure and the simplest case for constant-folding
of branch and switch instructions within loop into unconditional branches.
It is useful as a cleanup for such passes as loop unswitching that sometimes
produce such branches.
Only the simplest case supported in this patch: after the folding, no block
should become dead or stop being part of the loop. Support for more
sophisticated cases will go separately in follow-up patches.
Differential Revision: https://reviews.llvm.org/D54021
Reviewed By: anna
llvm-svn: 347183
Fix all of the missing debug location errors in CVP found by debugify.
This includes the missing-location-after-udiv-truncation case described
in llvm.org/PR38178.
llvm-svn: 347147
We need to control exponential behavior of loop-unswitch so we do not get
run-away compilation.
Suggested solution is to introduce a multiplier for an unswitch cost that
makes cost prohibitive as soon as there are too many candidates and too
many sibling loops (meaning we have already started duplicating loops
by unswitching).
It does solve the currently known problem with compile-time degradation
(PR 39544).
Tests are built on top of a recently implemented CHECK-COUNT-<num>
FileCheck directives.
Reviewed By: chandlerc, mkazantsev
Differential Revision: https://reviews.llvm.org/D54223
llvm-svn: 347097
We were adding the entire scalarization extraction cost for reductions, which returns the total cost of extracting every element of a vector type.
For reductions we don't need to do this - we just need to extract the 0'th element after the reduction pattern has completed.
Fixes PR37731
Differential Revision: https://reviews.llvm.org/D54585
llvm-svn: 346970
There's a potential miscompile here. It's unlikely in the real
world because this transform is guarded with shouldChangeType(),
but this test file doesn't include a standard data-layout for
some reason (despite including a custom 1), so we can see the bug.
llvm-svn: 346966
This is a problem seen in common rotate idioms as noted in:
https://bugs.llvm.org/show_bug.cgi?id=34924
Note that we are not canonicalizing standard IR (shifts and logic) to the intrinsics yet.
(Although I've written this before...) I think this is the last step before we enable
that transform. Ie, we could regress code by doing that transform without this
simplification in place.
In PR34924, I questioned whether this is a valid transform for target-independent IR,
but I convinced myself this is ok. If we're speculating a funnel shift by turning cmp+br
into select, then SimplifyCFG has already determined that the transform is justified.
It's possible that SimplifyCFG is not taking into account profile or other metadata,
but if that's true, then it's a bug independent of funnel shifts.
Also, we do have CGP code to restore a guard like this around an intrinsic if it can't
be lowered cheaply. But that isn't necessary for funnel shift because the default
expansion in SelectionDAGBuilder includes this same cmp+select.
Differential Revision: https://reviews.llvm.org/D54552
llvm-svn: 346960
A call to @llvm.trap can be expected to be cold (i.e. unlikely to be
reached in a normal program execution).
Outlining paths which unconditionally trap is an important memory
saving. As the hot/cold splitting pass (imho) should not treat all
noreturn calls as cold, explicitly mark @llvm.trap cold so that it can
be outlined.
Split out of https://reviews.llvm.org/D54244.
Differential Revision: https://reviews.llvm.org/D54329
llvm-svn: 346885
Summary:
Previously we marked all vararg functions as non-inlinable in the
function summary, which prevented their importing. However, the
corresponding inliner restriction was loosened in r321940/r342675
to only apply to functions calling va_start. Adjust the summary
flag computation to match.
Reviewers: davidxl
Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith, llvm-commits
Differential Revision: https://reviews.llvm.org/D54270
llvm-svn: 346883
Summary:
These asserts are based on the assumption that the order of true/false operands in a select and those in the compare would always be the same.
This fixes PR39595.
Reviewers: craig.topper, spatel, dmgreen
Reviewed By: craig.topper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54359
llvm-svn: 346874
Regenerate this test using update_test_checks.py in preparation for an
upcomming commit, to make it not depend on the names of instructions.
llvm-svn: 346869
The shift amount of a funnel shift is modulo the scalar bitwidth:
http://llvm.org/docs/LangRef.html#llvm-fshl-intrinsic
...so we can use demanded bits analysis on that operand to simplify it
when we have a power-of-2 bitwidth.
This is another step towards canonicalizing {shift/shift/or} to the
intrinsics in IR.
Differential Revision: https://reviews.llvm.org/D54478
llvm-svn: 346814
LoopUtils.cpp contains a utility that splits an loop exit block, so that the new block contains only edges coming from the loop. In the case of nested loops, the exit path for the inner loop might also be the back-edge of the outer loop. The new block which is inserted on this path, is now a latch for the outer loop, and it needs to hold the loop metadata for the outer loop. (The test case gives a more concrete view of the situation.)
Patch by Chang Lin (clin1)
Differential Revision: https://reviews.llvm.org/D53876
llvm-svn: 346810
The cmp+branch variant of this pattern is shown in:
https://bugs.llvm.org/show_bug.cgi?id=34924
...and as discussed there, we probably can't transform
that without a rotate intrinsic. We do have that now
via funnel shift, but we're not quite ready to
canonicalize IR to that form yet. The case with 'select'
should already be transformed though, so that's this patch.
The sequence with negation followed by masking is what we
use in the backend and partly in clang (though that part
should be updated).
https://rise4fun.com/Alive/TplC
%cmp = icmp eq i32 %shamt, 0
%sub = sub i32 32, %shamt
%shr = lshr i32 %x, %shamt
%shl = shl i32 %x, %sub
%or = or i32 %shr, %shl
%r = select i1 %cmp, i32 %x, i32 %or
=>
%neg = sub i32 0, %shamt
%masked = and i32 %shamt, 31
%maskedneg = and i32 %neg, 31
%shl2 = lshr i32 %x, %masked
%shr2 = shl i32 %x, %maskedneg
%r = or i32 %shl2, %shr2
llvm-svn: 346807
The IEEE-754 Standard makes it clear that fneg(x) and
fsub(-0.0, x) are two different operations. The former is a bitwise
operation, while the latter is an arithmetic operation. This patch
creates a dedicated FNeg IR Instruction to model that behavior.
Differential Revision: https://reviews.llvm.org/D53877
llvm-svn: 346774
This patch updates DuplicateInstructionsInSplitBetween to update a DTU
instead of applying updates to the DT directly.
Given that there only are 2 users, also updated them in this patch to
avoid churn.
I slightly moved the code in CallSiteSplitting around to reduce the
places where we have to pass in DTU. If necessary, I could split those
changes in a separate patch.
This fixes missing DT updates when dealing with musttail calls in
CallSiteSplitting, by using DTU->deleteBB.
Reviewers: junbuml, kuhar, NutshellySima, indutny, brzycki
Reviewed By: NutshellySima
llvm-svn: 346769
This is a longer variant for the pattern handled in
rL346713
This one includes zexts.
Eventually, we should canonicalize all rotate patterns
to the funnel shift intrinsics, but we need a bit more
infrastructure to make sure the vectorizers handle those
intrinsics as well as the shift+logic ops.
https://rise4fun.com/Alive/FMn
Name: narrow rotateright
%neg = sub i8 0, %shamt
%rshamt = and i8 %shamt, 7
%rshamtconv = zext i8 %rshamt to i32
%lshamt = and i8 %neg, 7
%lshamtconv = zext i8 %lshamt to i32
%conv = zext i8 %x to i32
%shr = lshr i32 %conv, %rshamtconv
%shl = shl i32 %conv, %lshamtconv
%or = or i32 %shl, %shr
%r = trunc i32 %or to i8
=>
%maskedShAmt2 = and i8 %shamt, 7
%negShAmt2 = sub i8 0, %shamt
%maskedNegShAmt2 = and i8 %negShAmt2, 7
%shl2 = lshr i8 %x, %maskedShAmt2
%shr2 = shl i8 %x, %maskedNegShAmt2
%r = or i8 %shl2, %shr2
llvm-svn: 346716
The sub-pattern for the shift amount in a rotate can take on
several different forms, and there's apparently no way to
canonicalize those without seeing the entire rotate sequence.
This is the form noted in:
https://bugs.llvm.org/show_bug.cgi?id=39624https://rise4fun.com/Alive/qnT
%zx = zext i8 %x to i32
%maskedShAmt = and i32 %shAmt, 7
%shl = shl i32 %zx, %maskedShAmt
%negShAmt = sub i32 0, %shAmt
%maskedNegShAmt = and i32 %negShAmt, 7
%shr = lshr i32 %zx, %maskedNegShAmt
%rot = or i32 %shl, %shr
%r = trunc i32 %rot to i8
=>
%truncShAmt = trunc i32 %shAmt to i8
%maskedShAmt2 = and i8 %truncShAmt, 7
%shl2 = shl i8 %x, %maskedShAmt2
%negShAmt2 = sub i8 0, %truncShAmt
%maskedNegShAmt2 = and i8 %negShAmt2, 7
%shr2 = lshr i8 %x, %maskedNegShAmt2
%r = or i8 %shl2, %shr2
llvm-svn: 346713
This just identifies the intrinsics as candidates for vectorization.
It does not mean we will attempt to vectorize under normal conditions
(the test file is forcing vectorization).
The cost model must be fixed to show that the transform is profitable
in general.
Allowing vectorization with these intrinsics is required to avoid
potential regressions from canonicalizing to the intrinsics from
generic IR:
https://bugs.llvm.org/show_bug.cgi?id=37417
llvm-svn: 346661
This patch relaxes overconservative checks on whether or not we could write
memory before we execute an instruction. This allows us to hoist guards out of
loops even if they are not in the header block.
Differential Revision: https://reviews.llvm.org/D50891
Reviewed By: fedor.sergeev
llvm-svn: 346643
ComputeValueKnownInPredecessors has a "visited" set to prevent infinite
loops, since a value can be visited more than once. However, the
implementation didn't prevent the algorithm from taking exponential
time. Instead of removing elements from the RecursionSet one at a time,
we should keep around the whole set until
ComputeValueKnownInPredecessors finishes, then discard it.
The testcase is synthetic because I was having trouble effectively
reducing the original. But it's basically the same idea.
Instead of failing, we could theoretically cache the result instead.
But I don't think it would help substantially in practice.
Differential Revision: https://reviews.llvm.org/D54239
llvm-svn: 346562
After D45330, Dominators are required for IPSCCP and can be preserved.
This patch preserves DominatorTreeAnalysis in the new pass manager. AFAIK the legacy pass manager cannot preserve function analysis required by a module analysis.
Reviewers: davide, dberlin, chandlerc, efriedma, kuhar, NutshellySima
Reviewed By: chandlerc, kuhar, NutshellySima
Differential Revision: https://reviews.llvm.org/D47259
llvm-svn: 346486
We can stop recording conditions once we reached the immediate dominator
for the block containing the call site. Conditions in predecessors of the
that node will be the same for all paths to the call site and splitting
is not beneficial.
This patch makes CallSiteSplitting dependent on the DT anlysis. because
the immediate dominators seem to be the easiest way of finding the node
to stop at.
I had to update some exiting tests, because they were checking for
conditions that were true/false on all paths to the call site. Those
should now be handled by instcombine/ipsccp.
Reviewers: davide, junbuml
Reviewed By: junbuml
Differential Revision: https://reviews.llvm.org/D44627
llvm-svn: 346483
This patch adds logic to detect reductions across the inner and outer
loop by following the incoming values of PHI nodes in the outer loop. If
the incoming values take part in a reduction in the inner loop or come
from outside the outer loop, we found a reduction spanning across inner
and outer loop.
With this change, ~10% more loops are interchanged in the LLVM
test-suite + SPEC2006.
Fixes https://bugs.llvm.org/show_bug.cgi?id=30472
Reviewers: mcrosier, efriedma, karthikthecool, davide, hfinkel, dmgreen
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D43245
llvm-svn: 346438
Summary:
This fixes PR 37422
In ELF, non-weak symbols can also be non-prevailing. In this particular
PR, the __llvm_profile_* symbols are non-prevailing but weren't getting
dropped - causing multiply-defined errors with lld.
Also add a test, strong_non_prevailing.ll, to ensure that multiple
copies of a strong symbol are dropped.
To fix the test regressions exposed by this fix,
- do not mark prevailing copies for symbols with 'appending' linkage.
There's no one prevailing copy for such symbols.
- fix the prevailing version in dead-strip-fulllto.ll
- explicitly pass exported symbols to llvm-lto in fumcimport.ll and
funcimport_var.ll
Reviewers: tejohnson, pcc
Subscribers: mehdi_amini, inglorion, eraman, steven_wu, dexonsmith,
dang, srhines, llvm-commits
Differential Revision: https://reviews.llvm.org/D54125
llvm-svn: 346436
Summary:
When the 3rd argument to these intrinsics is zero, lowering them
to shift instructions produces poison values, since we end up with
shift amounts equal to the number of bits in the shifted value. This
means we can only lower these intrinsics if we can prove that the
3rd argument is not zero.
Reviewers: arsenm
Reviewed By: arsenm
Subscribers: bnieuwenhuizen, jvesely, wdng, nhaehnle, llvm-commits
Differential Revision: https://reviews.llvm.org/D53739
llvm-svn: 346422
This eliminates the outlining penalty for llvm.trap/unreachable, because
callers no longer have to emit cleanup/ret instructions after calling an
outlined `noreturn` function.
rdar://45523626
llvm-svn: 346421
The patch has been reverted because it ended up prohibiting propagation
of a constant to exit value. For such values, we should skip all checks
related to hard uses because propagating a constant is always profitable.
Differential Revision: https://reviews.llvm.org/D53691
llvm-svn: 346397
LSR reassociates constants as unfolded offsets when the constants fit as
immediate add operands, which currently prevents such constants from being
combined later with loop invariant registers.
This patch modifies GenerateCombinations() to generate a second formula which
includes the unfolded offset in the combined loop-invariant register.
This commit fixes a bug in the original patch (committed at r345114, reverted
at r345123).
Differential Revision: https://reviews.llvm.org/D51861
llvm-svn: 346390
Summary:
MergeFunctions currently tries to process strong functions before
weak functions, because weak functions can simply call strong
functions, while a strong/weak function cannot call a weak function
(a backing strong function is needed).
This patch additionally tries to process external functions before
local functions, because we definitely have to keep the external
function, but may be able to drop the local one (and definitely
can if it is also unnamed_addr).
Unfortunately, this exposes an existing bug in the implementation:
The FnTree and FNodesInTree structures can currently go out of
sync in the case where two weak functions are merged, because the
function in FnTree/FNodesInTree is RAUWed. This leaves it behind in
FnTree (this is intended, as it is the strong backing function which
should be used for further merges), while it is replaced in
FNodesInTree (this is not intended).
This is fixed by switching FNodesInTree from using a ValueMap to
using a DenseMap of AssertingVH.
This exposes another minor issue: Currently FNodesInTree is not
cleared after MergeFunctions finishes running. Currently, this is
potentially dangerous (e.g. if something else wants to RAUW a function
with a non-function), but at the very least it is unnecessary/inefficient.
After the change to use AssertingVH it becomes more problematic,
because there are certainly passes that remove functions.
This issue is fixed by clearing FNodesInTree at the end of the pass.
Reviewers: jfb, whitequark
Reviewed By: whitequark
Subscribers: rkruppe, llvm-commits
Differential Revision: https://reviews.llvm.org/D53271
llvm-svn: 346386
Summary:
For unnamed_addr functions we RAUW instead of only replacing direct callers. However, functions in which replacements were performed currently are not added back to the worklist, resulting in missed merging opportunities.
Fix this by calling removeUsers() prior to RAUW.
Reviewers: jfb, whitequark
Reviewed By: whitequark
Subscribers: rkruppe, llvm-commits
Differential Revision: https://reviews.llvm.org/D53262
llvm-svn: 346385
If all the edge counts for a function are zero, skip count population and
annotation, as nothing will happen. This can save some compile time.
Differential Revision: https://reviews.llvm.org/D54212
llvm-svn: 346370
Sanjay Patel