which have successfully round-tripped through the combine phase, and use
this to ensure all operands to DAG nodes are visited by the combiner,
even if they are only added during the combine phase.
This is critical to have the combiner reach nodes that are *introduced*
during combining. Previously these would sometimes be visited and
sometimes not be visited based on whether they happened to end up on the
worklist or not. Now we always run them through the combiner.
This fixes quite a few bad codegen test cases lurking in the suite while
also being more principled. Among these, the TLS codegeneration is
particularly exciting for programs that have this in the critical path
like TSan-instrumented binaries (although I think they engineer to use
a different TLS that is faster anyways).
I've tried to check for compile-time regressions here by running llc
over a merged (but not LTO-ed) clang bitcode file and observed at most
a 3% slowdown in llc. Given that this is essentially a worst case (none
of opt or clang are running at this phase) I think this is tolerable.
The actual LTO case should be even less costly, and the cost in normal
compilation should be negligible.
With this combining logic, it is possible to re-legalize as we combine
which is necessary to implement PSHUFB formation on x86 as
a post-legalize DAG combine (my ultimate goal).
Differential Revision: http://reviews.llvm.org/D4638
llvm-svn: 213898
In order to enable the preservation of noalias function parameter information
after inlining, and the representation of block-level __restrict__ pointer
information (etc.), additional kinds of aliasing metadata will be introduced.
This metadata needs to be carried around in AliasAnalysis::Location objects
(and MMOs at the SDAG level), and so we need to generalize the current scheme
(which is hard-coded to just one TBAA MDNode*).
This commit introduces only the necessary refactoring to allow for the
introduction of other aliasing metadata types, but does not actually introduce
any (that will come in a follow-up commit). What it does introduce is a new
AAMDNodes structure to hold all of the aliasing metadata nodes associated with
a particular memory-accessing instruction, and uses that structure instead of
the raw MDNode* in AliasAnalysis::Location, etc.
No functionality change intended.
llvm-svn: 213859
The target-independent DAGcombiner will generate:
asr w1, X, #31 w1 = splat sign bit.
add X, X, w1, lsr #28 X = X + 0 or pow2-1
asr w0, X, asr #4 w0 = X/pow2
However, the add + shifts is expensive, so generate:
add w0, X, 15 w0 = X + pow2-1
cmp X, wzr X - 0
csel X, w0, X, lt X = (X < 0) ? X + pow2-1 : X;
asr w0, X, asr 4 w0 = X/pow2
llvm-svn: 213758
insertions.
The old behavior could cause arbitrarily bad memory usage in the DAG
combiner if there was heavy traffic of adding nodes already on the
worklist to it. This commit switches the DAG combine worklist to work
the same way as the instcombine worklist where we null-out removed
entries and only add new entries to the worklist. My measurements of
codegen time shows slight improvement. The memory utilization is
unsurprisingly dominated by other factors (the IR and DAG itself
I suspect).
This change results in subtle, frustrating churn in the particular order
in which DAG combines are applied which causes a number of minor
regressions where we fail to match a pattern previously matched by
accident. AFAICT, all of these should be using AddToWorklist to directly
or should be written in a less brittle way. None of the changes seem
drastically bad, and a few of the changes seem distinctly better.
A major change required to make this work is to significantly harden the
way in which the DAG combiner handle nodes which become dead
(zero-uses). Previously, we relied on the ability to "priority-bump"
them on the combine worklist to achieve recursive deletion of these
nodes and ensure that the frontier of remaining live nodes all were
added to the worklist. Instead, I've introduced a routine to just
implement that precise logic with no indirection. It is a significantly
simpler operation than that of the combiner worklist proper. I suspect
this will also fix some other problems with the combiner.
I think the x86 changes are really minor and uninteresting, but the
avx512 change at least is hiding a "regression" (despite the test case
being just noise, not testing some performance invariant) that might be
looked into. Not sure if any of the others impact specific "important"
code paths, but they didn't look terribly interesting to me, or the
changes were really minor. The consensus in review is to fix any
regressions that show up after the fact here.
Thanks to the other reviewers for checking the output on other
architectures. There is a specific regression on ARM that Tim already
has a fix prepped to commit.
Differential Revision: http://reviews.llvm.org/D4616
llvm-svn: 213727
'Worklist' consistently rather than a deeply confusing mixture of
'WorkList' and 'Worklist'.
Notably, the very 'WorkList' of the DAG combiner was exposed to target
specific DAG combines under an interface 'AddToWorklist' which was
implemented by in turn calling 'AddToWorkList' in the combiner. This has
sent me circling with the wrong case in grep one too many times.
I chose to normalize on 'Worklist' because that one won the grep-vote
for llvm/lib/... by a hundered hits or so, and it is used in places
relatively "canonical" such as InstCombine's Worklist. Let's all jsut
pick this casing, whether "correct", "good", or "bad" and be
consistent...
llvm-svn: 213506
stack, filter all handle nodes from the DAG combiner worklist.
This will also handle cases where other handle nodes might be
(erroneously) added to the worklist and then cause bugs and explosions
when deleted. For example, when running the legalizer within the DAG
combiner, there are times when other handle nodes are used and can end
up here.
llvm-svn: 213505
Canonicalize shuffles according to rules:
* shuffle(A, shuffle(A, B)) -> shuffle(shuffle(A,B), A)
* shuffle(B, shuffle(A, B)) -> shuffle(shuffle(A,B), B)
* shuffle(B, shuffle(A, Undef)) -> shuffle(shuffle(A, Undef), B)
This patch helps identifying more shuffle pairs that could be combined reusing
the already existing rules in the DAGCombiner.
Added new test 'combine-vec-shuffle-5.ll' to verify that the canonicalized
shuffles are now folded into a single shuffle node by the DAGCombiner.
Added more test cases to 'combine-vec-shuffle-4.ll'.
llvm-svn: 213504
It turns out that in most cases (the main exception being i1-related
types) once these operations are formed we cannot separate them and
the targets end up having to deal with them whether they want to or
not.
This is not a good situation, and a more reasonable default can be
formed by ackowledging this and having targets leave them as Legal.
Only x86 seems to be affected (other targets don't even try marking
the operation Expand).
Mostly there's no visible change here yet, but it will be useful to
have truly expanded EXTLOADS for MVT::f16 softening support.
llvm-svn: 213162
This patch adds two new rules to the DAGCombiner:
1. shuffle (shuffle A, Undef, M0), B, M1 -> shuffle A, B, M2
2. shuffle (shuffle A, Undef, M0), A, M1 -> shuffle A, Undef, M2
We only do this if the combined shuffle is legal for the target.
Example:
;;
define <4 x float> @test(<4 x float> %a, <4 x float> %b) {
%1 = shufflevector <4 x float> %a, <4 x float> undef, <4 x i32><i32 6, i32 0, i32 1, i32 7>
%2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32><i32 1, i32 2, i32 4, i32 5>
ret <4 x i32> %2
}
;;
(using llc -mcpu=corei7 -march=x86-64)
Before, the x86 backend generated:
pshufd $120, %xmm0, %xmm0
shufps $-108, %xmm0, %xmm1
movaps %xmm1, %xmm0
Now the x86 backend generates:
movsd %xmm1, %xmm0
llvm-svn: 213069
This patch fixes a crasher in method 'DAGCombiner::visitOR' due to an invalid
call to method 'isShuffleMaskLegal'. On x86, method 'isShuffleMaskLegal'
always expects a legal vector value type in input.
With this patch, we immediately check if the input OR dag node has a legal
vector type; we only try to fold a OR dag node into a single shufflevector
if we know that the resulting shuffle will have a legal type.
This is to avoid calling method 'isShuffleMaskLegal' on a potentially
illegal vector value type.
Added a new test-case to file 'CodeGen/X86/combine-or.ll' to verify that
DAGCombiner doesn't crash in the attempt to check/combine an OR between shuffles
with illegal types.
llvm-svn: 213020
This patch teaches the DAGCombiner how to fold a pair of shuffles
according to rules:
1. shuffle(shuffle A, B, M0), B, M1) -> shuffle(A, B, M2)
2. shuffle(shuffle A, B, M0), A, M1) -> shuffle(A, B, M3)
The new rules would only trigger if the resulting shuffle has legal type and
legal mask.
Added test 'combine-vec-shuffle-3.ll' to verify that DAGCombiner correctly
folds shuffles on x86 when the resulting mask is legal. Also added some negative
cases to verify that we avoid introducing illegal shuffles.
llvm-svn: 213001
Verify that DAGCombiner does not crash when trying to fold a pair of shuffles
according to rule (added at r212539):
(shuffle (shuffle A, Undef, M0), Undef, M1) -> (shuffle A, Undef, M2)
The DAGCombiner avoids folding shuffles if the resulting shuffle dag node
is not legal for the target. That means, the resulting shuffle must have
legal type and legal mask.
Before, the DAGCombiner only called method
'TargetLowering::isShuffleMaskLegal' to check if it was "safe" to fold according
to the above-mentioned rule. However, this caused a crash in the x86 backend
since method 'isShuffleMaskLegal' always expects to be called on a
legal vector type.
llvm-svn: 212915
This patch teaches the DAGCombiner how to fold shuffles according to the
following new rules:
1. shuffle(shuffle(x, y), undef) -> x
2. shuffle(shuffle(x, y), undef) -> y
3. shuffle(shuffle(x, y), undef) -> shuffle(x, undef)
4. shuffle(shuffle(x, y), undef) -> shuffle(y, undef)
The backend avoids to combine shuffles according to rules 3. and 4. if
the resulting shuffle does not have a legal mask. This is to avoid introducing
illegal shuffles that are potentially expanded into a sub-optimal sequence of
target specific dag nodes during vector legalization.
Added test case combine-vec-shuffle-2.ll to verify that we correctly triggers
the new rules when combining shuffles.
llvm-svn: 212748
Summary:
On MIPS32r6/MIPS64r6, floating point comparisons return 0 or -1 but integer
comparisons return 0 or 1.
Updated the various uses of getBooleanContents. Two simplifications had to be
disabled when float and int boolean contents differ:
- ScalarizeVecRes_VSELECT except when the kind of boolean contents is trivially
discoverable (i.e. when the condition of the VSELECT is a SETCC node).
- visitVSELECT (select C, 0, 1) -> (xor C, 1).
Come to think of it, this one could test for the common case of 'C'
being a SETCC too.
Preserved existing behaviour for all other targets and updated the affected
MIPS32r6/MIPS64r6 tests. This also fixes the pi benchmark where the 'low'
variable was counting in the wrong direction because it thought it could simply
add the result of the comparison.
Reviewers: hfinkel
Reviewed By: hfinkel
Subscribers: hfinkel, jholewinski, mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D4389
llvm-svn: 212697
tracks which elements of the build vector are in fact undef.
This should make actually inpsecting them (likely in my next patch)
reasonably pretty. Also makes the output parameter optional as it is
clear now that *most* users are happy with undefs in their splats.
llvm-svn: 212581
This patch teaches how to fold a shuffle according to rule:
shuffle (shuffle (x, undef, M0), undef, M1) -> shuffle(x, undef, M2)
We do this only if the resulting mask M2 is legal; this is to avoid introducing
illegal shuffles that are potentially expanded into a sub-optimal sequence
of target specific dag nodes.
This patch has the advantage of being target independent, since it works on ISD
nodes. Therefore, all targets (not only x86) can take advantage of this rule.
The idea behind this patch is that most shuffle pairs can be safely combined
before we run the legalizer on vector operations. This allows us to
combine/simplify dag nodes earlier in the process and not only immediately
before instruction selection stage.
That said. This patch is not meant to replace any existing target specific
combine rules; backends might still introduce new shuffles during legalization
stage. Also, this rule is very simple and avoids to aggressively optimize
shuffles.
llvm-svn: 212539
nodes about whether they are splats. This is factored out and improved
from r212324 which got reverted as it was far too aggressive. The new
API should help more conservatively handle buildvectors that are
a mixture of splatted and undef values.
No functionality change at this point. The hope is to slowly
re-introduce the undef-tolerant optimization of splats, but each time
being forced to make a concious decision about how to handle the undefs
in a way that doesn't lead to contradicting assumptions about the
collapsed value.
Hal has pointed out in discussions that this may not end up being the
desired API and instead it may be more convenient to get a mask of the
undef elements or something similar. I'm starting simple and will expand
the API as I adapt actual callers and see exactly what they need.
llvm-svn: 212514
lanes in vector splats.
The core problem here is that undef lanes can't *unilaterally* be
considered to contribute to splats. Their handling needs to be more
cautious. There is also a reported failure of the nightly testers
(thanks Tobias!) that may well stem from the same core issue. I'm going
to fix this theoretical issue, factor the APIs a bit better, and then
verify that I don't see anything bad with Tobias's reduction from the
test suite before recommitting.
Original commit message for r212324:
[x86] Generalize BuildVectorSDNode::getConstantSplatValue to work for
any constant, constant FP, or undef splat and to tolerate any undef
lanes in a splat, then replace all uses of isSplatVector in X86's
lowering with it.
This fixes issues where undef lanes in an otherwise splat vector would
prevent the splat logic from firing. It is a touch more awkward to use
this interface, but it is much more accurate. Suggestions for better
interface structuring welcome.
With this fix, the code generated with the widening legalization
strategy for widen_cast-4.ll is *dramatically* improved as the special
lowering strategies for a v16i8 SRA kick in even though the high lanes
are undef.
We also get a slightly different choice for broadcasting an aligned
memory location, and use vpshufd instead of vbroadcastss. This looks
like a minor win for pipelining and domain crossing, but a minor loss
for the number of micro-ops. I suspect its a wash, but folks can
easily tweak the lowering if they want.
llvm-svn: 212475
any constant, constant FP, or undef splat and to tolerate any undef
lanes in a splat, then replace all uses of isSplatVector in X86's
lowering with it.
This fixes issues where undef lanes in an otherwise splat vector would
prevent the splat logic from firing. It is a touch more awkward to use
this interface, but it is much more accurate. Suggestions for better
interface structuring welcome.
With this fix, the code generated with the widening legalization
strategy for widen_cast-4.ll is *dramatically* improved as the special
lowering strategies for a v16i8 SRA kick in even though the high lanes
are undef.
We also get a slightly different choice for broadcasting an aligned
memory location, and use vpshufd instead of vbroadcastss. This looks
like a minor win for pipelining and domain crossing, but a minor loss
for the number of micro-ops. I suspect its a wash, but folks can easily
tweak the lowering if they want.
llvm-svn: 212324
The PowerPC 128-bit long double data type (ppcf128 in LLVM) is in fact a
pair of two doubles, where one is considered the "high" or
more-significant part, and the other is considered the "low" or
less-significant part. When a ppcf128 value is stored in memory or a
register pair, the high part always comes first, i.e. at the lower
memory address or in the lower-numbered register, and the low part
always comes second. This is true both on big-endian and little-endian
PowerPC systems. (Similar to how with a complex number, the real part
always comes first and the imaginary part second, no matter the byte
order of the system.)
This was implemented incorrectly for little-endian systems in LLVM.
This commit fixes three related issues:
- When printing an immediate ppcf128 constant to assembler output
in emitGlobalConstantFP, emit the high part first on both big-
and little-endian systems.
- When lowering a ppcf128 type to a pair of f64 types in SelectionDAG
(which is used e.g. when generating code to load an argument into a
register pair), use correct low/high part ordering on little-endian
systems.
- In a related issue, because lowering ppcf128 into a pair of f64 must
operate differently from lowering an int128 into a pair of i64,
bitcasts between ppcf128 and int128 must not be optimized away by the
DAG combiner on little-endian systems, but must effect a word-swap.
Reviewed by Hal Finkel.
llvm-svn: 212274
The SelectionDAG bad a special case for ISD::SELECT_CC, where it would
allow targets to specify:
setOperationAction(ISD::SELECT_CC, MVT::Other, Expand);
to indicate that they wanted to expand ISD::SELECT_CC for all types.
This wasn't applied correctly everywhere, and it makes writing new
DAG patterns with ISD::SELECT_CC difficult.
llvm-svn: 210541
This patch adds new target specific combine rules to identify horizontal
add/sub idioms from BUILD_VECTOR dag nodes.
This patch also teaches the DAGCombiner how to canonicalize sequences of
insert_vector_elt dag nodes according to the following rule:
(insert_vector_elt (insert_vector_elt A, I0), I1) ->
(insert_vecto_elt (insert_vector_elt A, I1), I0)
This new canonicalization rule only triggers if the inner insert_vector
dag node has exactly one use; also, both indices must be known constants,
and I1 < I0.
This last rule made it possible to write a simpler algorithm to identify
horizontal add/sub patterns because now we don't have to worry about the
ordering of insert_vector_elt dag nodes.
llvm-svn: 210477
This patch modifies SelectionDAGBuilder to construct SDNodes with associated
NoSignedWrap, NoUnsignedWrap and Exact flags coming from IR BinaryOperator
instructions.
Added a new SDNode type called 'BinaryWithFlagsSDNode' to allow accessing
nsw/nuw/exact flags during codegen.
Patch by Marcello Maggioni.
llvm-svn: 210467
This patch teaches the backend how to simplify/canonicalize dag node
sequences normally introduced by the backend when promoting certain dag nodes
with illegal vector type.
This patch adds two new combine rules:
1) fold (shuffle (bitcast (BINOP A, B)), Undef, <Mask>) ->
(shuffle (BINOP (bitcast A), (bitcast B)), Undef, <Mask>)
2) fold (BINOP (shuffle (A, Undef, <Mask>)), (shuffle (B, Undef, <Mask>))) ->
(shuffle (BINOP A, B), Undef, <Mask>).
Both rules are only triggered on the type-legalized DAG.
In particular, rule 1. is a target specific combine rule that attempts
to sink a bitconvert into the operands of a binary operation.
Rule 2. is a target independet rule that attempts to move a shuffle
immediately after a binary operation.
llvm-svn: 209930
Summary:
If both vector args to vselect are concat_vectors and the condition is
constant and picks half a vector from each argument, convert the vselect
into a concat_vectors.
Added a test.
The ConvertSelectToConcatVector is assuming it doesn't get vselects with
arguments of, for example, <undef, undef, true, true>. Those get taken
care of in the checks above its call.
Reviewers: nadav, delena, grosbach, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D3916
llvm-svn: 209929
An address only use of an extract element of a load can be simplified to a
load. Without this the result of the extract element is spilled to the
stack so that an address is available.
llvm-svn: 209788
This reverts r208640 (I've just XFAILed the test) because it broke ppc64/Linux
self-hosting. Because nearly every regression test triggers a segfault, I hope
this will be easy to fix.
llvm-svn: 209747
Right now the load may not get DCE'd because of the side-effect of updating
the base pointer.
This can happen if we lower a read-modify-write of an illegal larger type
(e.g. i48) such that the modification only affects one of the subparts (the
lower i32 part but not the higher i16 part). See the testcase.
In order to spot the dead load we need to revisit it when SimplifyDemandedBits
decided that the value of the load is masked off. This is the
CommitTargetLoweringOpt piece.
I checked compile time with ARM64 by sending SPEC bitcode files through llc.
No measurable change.
Fixes <rdar://problem/16031651>
llvm-svn: 208640
For pattern like ((x >> C1) & Mask) << C2, DAG combiner may convert it
into (x >> (C1-C2)) & (Mask << C2), which makes pattern matching of ubfx
more difficult.
For example:
Given
%shr = lshr i64 %x, 4
%and = and i64 %shr, 15
%arrayidx = getelementptr inbounds [8 x [64 x i64]]* @arr, i64 0, %i64 2, i64 %and
%0 = load i64* %arrayidx
With current shift folding, it takes 3 instrs to compute base address:
lsr x8, x0, #1
and x8, x8, #0x78
add x8, x9, x8
If using ubfx, it only needs 2 instrs:
ubfx x8, x0, #4, #4
add x8, x9, x8, lsl #3
This fixes bug 19589
llvm-svn: 207702
Otherwise the legalizer would just scalarize everything. Support for
mulhi in the targets isn't that great yet so on most targets we get
exactly the same scalarized output. Add a test for x86 vector udiv.
I had to disable the mulhi nodes on ARM because there aren't any patterns
for it. As far as I know ARM has instructions for getting the high part of
a multiply so this should be fixed.
llvm-svn: 207315
define below all header includes in the lib/CodeGen/... tree. While the
current modules implementation doesn't check for this kind of ODR
violation yet, it is likely to grow support for it in the future. It
also removes one layer of macro pollution across all the included
headers.
Other sub-trees will follow.
llvm-svn: 206837
This particular DAG combine is designed to kick in when both ConstantFPs will
end up being loaded via a litpool, however those nodes have a semi-legal
status, dictated by isFPImmLegal so in some cases there wouldn't have been a
litpool in the first place. Don't try to be clever in those circumstances.
Picked up while merging some AArch64 tests.
llvm-svn: 206365
We had disabled use of TBAA during CodeGen (even when otherwise using AA)
because the ptrtoint/inttoptr used by CGP for address sinking caused BasicAA to
miss basic type punning that it should catch (and, thus, we'd fail to override
TBAA when we should).
However, when AA is in use during CodeGen, CGP now uses normal GEPs and
bitcasts, instead of ptrtoint/inttoptr, when doing address sinking. As a
result, BasicAA should be able to make us do the right thing in the face of
type-punning, and it seems safe to enable use of TBAA again. self-hosting seems
fine on PPC64/Linux on the P7, with TBAA enabled and -misched=shuffle.
Note: We still don't update TBAA when merging stack slots, although because
BasicAA should now catch all such cases, this is no longer a blocking issue.
Nevertheless, I plan to commit code to deal with this properly in the near
future.
llvm-svn: 206093
sign/zero/any extensions. However a few places were not checking properly the
property of the load and were turning an indexed load into a regular extended
load. Therefore the indexed value was lost during the process and this was
triggering an assertion.
<rdar://problem/16389332>
llvm-svn: 205923
When the loop vectorizer vectorizes code that uses the loop induction variable,
we often end up with IR like this:
%b1 = insertelement <2 x i32> undef, i32 %v, i32 0
%b2 = shufflevector <2 x i32> %b1, <2 x i32> undef, <2 x i32> zeroinitializer
%i = add <2 x i32> %b2, <i32 2, i32 3>
If the add in this example is not legal (as is the case on PPC with VSX), it
will be scalarized, and we'll end up with a number of extract_vector_elt nodes
with the vector shuffle as the input operand, and that vector shuffle is fed by
one or more build_vector nodes. By the time that vector operations are
expanded, visitEXTRACT_VECTOR_ELT will not create new extract_vector_elt by
looking through the vector shuffle (to make sure that no illegal operations are
created), and so the extract_vector_elt -> vector shuffle -> build_vector is
never simplified to an operand of the build vector.
By looking at build_vectors through a shuffle we fix this particular situation,
preventing a vector from being built, only to be deconstructed again (for the
scalarized add) -- an expensive proposition when this all needs to be done via
the stack. We probably want a more comprehensive fix here where we look back
recursively through any shuffles to any build_vectors or scalar_to_vectors,
etc. but that can come later.
llvm-svn: 205179
This patch renames method 'isConstantSplat' as 'getConstantSplatValue'
(mainly for consistency reasons), and rewrites its logic to ensure
that we always perform a legal 'cast<ConstantSDNode>'.
Added test shift-combine-crash.ll to verify that DAGCombiner no longer crashes with an assertion failure in the attempt to simplify a vector shift by a vector of all undef counts.
llvm-svn: 204536
This is already done for shifts. Allow it for rotations as well. E.g.:
(rotl:i32 x, (trunc (and y, 31))) -> (rotl:i32 x, (and (trunc y), 31))
Use the newly factored-out distributeTruncateThroughAnd.
With this patch and some X86.td tweaks we should be able to remove redundant
masking of the rotation amount like in the example above. HW implicitly
performs this masking.
The testcase will be added as part of the X86 patch.
llvm-svn: 203316
This is the new idiom:
x<<(y&31) | x>>((0-y)&31)
which is recognized as:
x ROTL (y&31)
The change refines matchRotateSub. In
Neg & (OpSize - 1) == (OpSize - Pos) & (OpSize - 1), if Pos is
Pos' & (OpSize - 1) we can just use Pos' instead of Pos.
llvm-svn: 203315
Slightly change the wording in the function comment. Originally, it can be
misunderstood as we turned the input into two subsequent rotates.
Better connect the comment which talks about Mask and the code which used
LoBits. Renamed variable to MaskLoBits.
llvm-svn: 203314
This patch teaches the DAGCombiner how to fold a binary OR between two
shufflevector into a single shuffle vector when possible.
The rules are:
1. fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf A, B, Mask1)
2. fold (or (shuf A, V_0, MA), (shuf B, V_0, MB)) -> (shuf B, A, Mask2)
The DAGCombiner can take advantage of the fact that OR is commutative and
compute two possible shuffle masks (Mask1 and Mask2) for the resulting
shuffle node.
Before folding a dag according to either rule 1 or 2, DAGCombiner verifies
that the resulting shuffle mask is legal for the target.
DAGCombiner would firstly try to fold according to 1.; If not possible
then it will try to fold according to 2.
If both Mask1 and Mask2 are illegal then we conservatively don't fold
the OR instruction.
llvm-svn: 203156
Currently this code is duplicated across visitSHL, visitSRA and visitSRL. The
plan is to add rotates as clients to this new function.
There is no functional change intended here.
llvm-svn: 202908
This extract-and-trunc vector optimization cannot work for i1 values as
currently implemented, and so I'm disabling this for now for i1 values. In the
future, this can be fixed properly.
Soon I'll commit support for i1 CR bit tracking in the PowerPC backend, and
this will be covered by one of the existing regression tests.
llvm-svn: 202449
shifted mask rather than masking and shifting separately.
The patch adds this transformation to the DAGCombiner:
(shl (and (setcc:i8v16 ...) N01C) N1C) -> (and (setcc:i8v16 ...) N01C<<N1C)
<rdar://problem/16054492>
Patch by Adam Nemet <anemet@apple.com>
llvm-svn: 201906
BUILD_VECTOR nodes, e.g.:
(concat_vectors (BUILD_VECTOR a1, a2, a3, a4), (BUILD_VECTOR b1, b2, b3, b4))
->
(BUILD_VECTOR a1, a2, a3, a4, b1, b2, b3, b4)
This fixes an issue with AVX, where a sequence was not recognized as a 256-bit
vbroadcast due to the concat_vectors.
llvm-svn: 201158
During DAGCombine visitShiftByConstant assumes that certain binary operations
with only constant operands can always be folded successfully. This is no longer
true when the constant is opaque. This commit fixes visitShiftByConstant by not
performing the optimization for opaque constants. Otherwise we would end up in
an infinite DAGCombine loop.
llvm-svn: 200900
when the input is a concat_vectors and the insert replaces one of the
concat halves:
Lower half: fold (insert_subvector (concat_vectors X, Y), Z) ->
(concat_vectors Z, Y)
Upper half: fold (insert_subvector (concat_vectors X, Y), Z) ->
(concat_vectors X, Z)
This can be seen with the following IR:
define <8 x float> @lower_half(<4 x float> %v1, <4 x float> %v2, <4 x
float> %v3) {
%1 = shufflevector <4 x float> %v1, <4 x float> %v2, <8 x i32> <i32
0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%2 = tail call <8 x float> @llvm.x86.avx.vinsertf128.ps.256(<8 x
float> %1, <4 x float> %v3, i8 0)
The vinsertf128 intrinsic is converted into an insert_subvector node
in SelectionDAGBuilder.cpp.
Using AVX, without the patch this generates two vinsertf128 instructions:
vinsertf128 $1, %xmm1, %ymm0, %ymm0
vinsertf128 $0, %xmm2, %ymm0, %ymm0
With the patch this is optimized into:
vinsertf128 $1, %xmm1, %ymm2, %ymm0
Patch by Robert Lougher.
llvm-svn: 200506
Make sure that we don't introduce illegal build_vector dag nodes
when trying to fold a sign_extend of a build_vector.
This fixes a regression introduced by r200234.
Added test CodeGen/X86/fold-vector-sext-crash.ll
to verify that llc no longer crashes with an assertion failure
due to an illegal build_vector of type MVT::v4i64.
Thanks to Ilia Filippov for spotting this regression and for
providing a reproducible test case.
llvm-svn: 200313
Also update the comment, since it actually produces a
select (setcc) instead of select_cc.
It was checking and using the setcc result type for the
type of the sext, instead of the type of the compared items.
In my problem case, the sext was to i32 and was used as the setcc type,
but the expected type was i64.
No test since I haven't been able to hit the problem with
this on any in-tree targets.
llvm-svn: 200249
This patch teaches the DAGCombiner how to fold a sext/aext/zext dag node when
the operand in input is a build vector of constants (or UNDEFs).
The inability to fold a sext/zext of a constant build_vector was the root
cause of some pcg bugs affecting vselect expansion on x86-64 with AVX support.
Before this change, the DAGCombiner only knew how to fold a sext/zext/aext of a
ConstantSDNode.
llvm-svn: 200234
Issue outcomes from DAGCombiner::MergeConsequtiveStores, more precisely from
mem-ops sequence sorting.
Consider, how MergeConsequtiveStores works for next example:
store i8 1, a[0]
store i8 2, a[1]
store i8 3, a[1] ; a[1] again.
return ; DAG starts here
1. Method will collect all the 3 stores.
2. It sorts them by distance from the base pointer (farthest with highest
index).
3. It takes first consecutive non-overlapping stores and (if possible) replaces
them with a single store instruction.
The point is, we can't determine here which 'store' instruction
would be the second after sorting ('store 2' or 'store 3').
It happens that 'store 3' would be the second, and 'store 2' would be the third.
So after merging we have the next result:
store i16 (1 | 3 << 8), base ; is a[0] but bit-casted to i16
store i8 2, a[1]
So actually we swapped 'store 3' and 'store 2' and got wrong contents in a[1].
Fix: In sort routine just also take into account mem-op sequence number.
llvm-svn: 200201
There are currently two issues, of which I currently know, that prevent TBAA
from being correctly usable in CodeGen:
1. Stack coloring does not update TBAA when merging allocas. This is easy
enough to fix, but is not the largest problem.
2. CGP inserts ptrtoint/inttoptr pairs when sinking address computations.
Because BasicAA does not handle inttoptr, we'll often miss basic type punning
idioms that we need to catch so we don't miscompile real-world code (like LLVM).
I don't yet have a small test case for this, but this fixes self hosting a
non-asserts build of LLVM on PPC64 when using -enable-aa-sched-mi and -misched=shuffle.
llvm-svn: 200093
This option (which is !NDEBUG only) allows restricting the use of alias
analysis in DAGCombiner to a specific function. This has proved extremely
valuable to isolating bugs related to this feature, and mirrors the
misched-only-func option provided by the new instruction scheduler.
llvm-svn: 200088
This commit caused -Woverloaded-virtual warnings. The two new
TargetTransformInfo::getIntImmCost functions were only added to the superclass,
and to the X86 subclass. The other targets were not updated, and the
warning highlighted this by pointing out that e.g. ARMTTI::getIntImmCost was
hiding the two new getIntImmCost variants.
We could pacify the warning by adding "using TargetTransformInfo::getIntImmCost"
to the various subclasses, or turning it off, but I suspect that it's wrong to
leave the functions unimplemnted in those targets. The default implementations
return TCC_Free, which I don't think is right e.g. for ARM.
llvm-svn: 200058
Retry commit r200022 with a fix for the build bot errors. Constant expressions
have (unlike instructions) module scope use lists and therefore may have users
in different functions. The fix is to simply ignore these out-of-function uses.
llvm-svn: 200034
DAGCombiner::GatherAllAliases, which is only used when AA used is enabled
during DAGCombine, had a fundamentally incorrect assumption for which this
change compensates. GatherAllAliases, which is used to find aliasing
predecessor chain nodes (so that a better chain can be selected for a load or
store to enable subsequent optimizations) assumed that walking up the chain
would always catch all possibly-aliasing loads and stores. This is not true: To
really find all aliases, we also need to search for aliases through the value
operand of a store, etc. Consider the following situation:
Token1 = ...
L1 = load Token1, %52
S1 = store Token1, L1, %51
L2 = load Token1, %52+8
S2 = store Token1, L2, %51+8
Token2 = Token(S1, S2)
L3 = load Token2, %53
S3 = store Token2, L3, %52
L4 = load Token2, %53+8
S4 = store Token2, L4, %52+8
If we search for aliases of S3 (which loads address %52), and we look only
through the chain, then we'll miss the trivial dependence on L1 (which loads
from %52). We then might change all loads and stores to use Token1 as their
chain operand, which could result in copying %53 into %52 before copying
%52 into %51 (which should happen first).
The problem is, however, that searching for such data dependencies can become
expensive, and the cost is not directly related to the chain depth. Instead,
we'll rule out such configurations by insisting that we've visited all chain
users (except for users of the original chain, which is not necessary). When
doing this, we need to look through nodes we don't care about (otherwise,
things like register copies will interfere with trivial use cases).
Unfortunately, I don't have a small test case for this problem. Creating the
underlying situation is not hard (a pair of memcpys will do it), but arranging
for the default instruction schedule to be incorrect is very fragile.
This unbreaks self hosting on PPC64 when using
-mllvm -combiner-global-alias-analysis -mllvm -combiner-alias-analysis.
llvm-svn: 200033
These transformations obviously won't work for indexed (pre/post-inc) loads and
stores. In practice, I'm not sure there is any benefit to enabling them for
indexed nodes because other transformations that these might enable likely also
won't handle indexed nodes.
I don't have an in-tree test case that hits this problem, but an upcoming bug
fix will make it much more likely.
llvm-svn: 200023
This pass identifies expensive constants to hoist and coalesces them to
better prepare it for SelectionDAG-based code generation. This works around the
limitations of the basic-block-at-a-time approach.
First it scans all instructions for integer constants and calculates its
cost. If the constant can be folded into the instruction (the cost is
TCC_Free) or the cost is just a simple operation (TCC_BASIC), then we don't
consider it expensive and leave it alone. This is the default behavior and
the default implementation of getIntImmCost will always return TCC_Free.
If the cost is more than TCC_BASIC, then the integer constant can't be folded
into the instruction and it might be beneficial to hoist the constant.
Similar constants are coalesced to reduce register pressure and
materialization code.
When a constant is hoisted, it is also hidden behind a bitcast to force it to
be live-out of the basic block. Otherwise the constant would be just
duplicated and each basic block would have its own copy in the SelectionDAG.
The SelectionDAG recognizes such constants as opaque and doesn't perform
certain transformations on them, which would create a new expensive constant.
This optimization is only applied to integer constants in instructions and
simple (this means not nested) constant cast experessions. For example:
%0 = load i64* inttoptr (i64 big_constant to i64*)
Reviewed by Eric
llvm-svn: 200022
This fixes a regression intruced by r199135.
Revision 199135 tried to simplify part of the logic in method
DAGCombiner::SimplifyVBinOp introducing calls to method BuildVectorSDNode::isConstant().
However, that revision wrongly changed the check performed by method
SimplifyVBinOp to identify dag nodes that can be folded.
Before revision 199135, that method only tried to simplify vector binary operations
if both operands were build_vector of Constant/ConstantFP/Undef only.
After revision 199135, method SimplifyVBinop tried to
simplify also vector binary operations with only one constant operand.
This fixes the problem restoring the old behavior of SimplifyVBinOp.
llvm-svn: 199328
This commit teaches DAG to reassociate vector ops, which in turn enables
constant folding of vector op chains that appear later on during custom lowering
and DAG combine.
Reviewed by Andrea Di Biagio
llvm-svn: 199135
At the moment we expect rotates to have the form:
(or (shl X, Y), (shr X, Z))
where Y == bitsize(X) - Z or Z == bitsize(X) - Y. This form means that
the (or ...) is undefined for Y == 0 or Z == 0. This undefinedness can
be avoided by using Y == (C * bitsize(X) - Z) & (bitsize(X) - 1) or
Z == (C * bitsize(X) - Y) & (bitsize(X) - 1) for any integer C
(including 0, the most natural choice).
llvm-svn: 198861
InstCombine converts (sub 32, (add X, C)) into (sub 32-C, X),
so a rotate left of a 32-bit Y by X+C could appear as either:
(or (shl Y, (add X, C)), (shr Y, (sub 32, (add X, C))))
without InstCombine or:
(or (shl Y, (add X, C)), (shr Y, (sub 32-C, X)))
with it.
We already matched the first form. This patch handles the second too.
llvm-svn: 198860
There is a wrong assumption that the vector element type and the
type of each ConstantSDNode in the build_vector were the same.
However, when promoting the integer operand of a legally typed
build_vector, the operand type and the vector element type do not
need to be the same
(See method 'DAGTypeLegalizer::PromoteIntOp_BUILD_VECTOR' in
LegalizeIntegerTypes.cpp).
in AArch64 backend, the following dag sequence:
C0: i1 = Constant<0>
C1: i1 = Constant<-1>
V: v8i1 = BUILD_VECTOR C1, C1, C0, C0, C0, C0, C0, C0
is type-legalized into:
NewC0: i32 = Constant<0>
NewC1: i32 = Constant<1>
V: v8i8 = BUILD_VECTOR NewC1, NewC1, NewC0, NewC0, NewC0, NewC0, NewC0, NewC0
Forcing a getZeroExtend to VTBits to ensure that the new constant
is correctly.
llvm-svn: 198582
For AArch64 backend, if DAGCombiner see "sext(setcc)", it will
combine them together to a single setcc with extended value type.
Then if it see "zext(setcc)", it assumes setcc is Vxi1, and try to
create "(and (vsetcc), (1, 1, ...)". While setcc isn't Vxi1,
DAGcombiner will create wrong node and get wrong code emitted.
llvm-svn: 198190
ConstantSDNodes (or UNDEFs) into a simple BUILD_VECTOR.
For example, given the following sequence of dag nodes:
i32 C = Constant<1>
v4i32 V = BUILD_VECTOR C, C, C, C
v4i32 Result = SIGN_EXTEND_INREG V, ValueType:v4i1
The SIGN_EXTEND_INREG node can be folded into a build_vector since
the vector in input is a BUILD_VECTOR of constants.
The optimized sequence is:
i32 C = Constant<-1>
v4i32 Result = BUILD_VECTOR C, C, C, C
llvm-svn: 198084
DAGCombiner could fold (truncate (load)) -> smaller load if the original
load was the width of the truncation result or wider. This patch extends
it to handle cases where the original load was narrower (and so the
extension type stays the same).
llvm-svn: 197030
Summary:
Moved the requirement for SelectionDAG::getConstant() to return legally
typed nodes slightly earlier. There were two optional DAGCombine passes
that were missed out and were required to produce type-legal DAGs.
Simplified a code-path in tryFoldToZero() to use SelectionDAG::getConstant().
This provides support for both promoted and expanded vector types whereas the
previous code only supported promoted vector types.
Fixes a "Type for zero vector elements is not legal" assertion detected by
an llvm-stress generated test.
Reviewers: resistor
CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D2251
llvm-svn: 195635
This patch is a rewrite of the original patch commited in r194542. Instead of
relying on the type legalizer to do the splitting for us, we now peform the
splitting ourselves in the DAG combiner. This is necessary for the case where
the vector mask is a legal type after promotion and still wouldn't require
splitting.
Patch by: Juergen Ributzka
NOTE: This is a candidate for the 3.4 branch.
llvm-svn: 195397
This is to avoid this transformation in some cases:
fold (conv (load x)) -> (load (conv*)x)
On architectures that don't natively support some vector
loads efficiently casting the load to a smaller vector of
larger types and loading is more efficient.
Patch by Micah Villmow.
llvm-svn: 194783
This patch reapplies r193676 with an additional fix for the Hexagon backend. The
SystemZ backend has already been fixed by r194148.
The Type Legalizer recognizes that VSELECT needs to be split, because the type
is to wide for the given target. The same does not always apply to SETCC,
because less space is required to encode the result of a comparison. As a result
VSELECT is split and SETCC is unrolled into scalar comparisons.
This commit fixes the issue by checking for VSELECT-SETCC patterns in the DAG
Combiner. If a matching pattern is found, then the result mask of SETCC is
promoted to the expected vector mask type for the given target. Now the type
legalizer will split both VSELECT and SETCC.
This allows the following X86 DAG Combine code to sucessfully detect the MIN/MAX
pattern. This fixes PR16695, PR17002, and <rdar://problem/14594431>.
Reviewed by Nadav
llvm-svn: 194542
The Type Legalizer recognizes that VSELECT needs to be split, because the type
is to wide for the given target. The same does not always apply to SETCC,
because less space is required to encode the result of a comparison. As a result
VSELECT is split and SETCC is unrolled into scalar comparisons.
This commit fixes the issue by checking for VSELECT-SETCC patterns in the DAG
Combiner. If a matching pattern is found, then the result mask of SETCC is
promoted to the expected vector mask type for the given target. This mask has
usually the same size as the VSELECT return type (except for Intel KNL). Now the
type legalizer will split both VSELECT and SETCC.
This allows the following X86 DAG Combine code to sucessfully detect the MIN/MAX
pattern. This fixes PR16695, PR17002, and <rdar://problem/14594431>.
Reviewed by Nadav
llvm-svn: 193676
Making useAA() default to true for SystemZ showed that the combiner alias
analysis wasn't handling volatile accesses. This hit many of the SystemZ
tests, but I arbitrarily picked one for the purpose of this patch.
llvm-svn: 193518
Most SelectionDAG code drops the TBAA info when creating a new form of a
load and store (e.g. during legalization, or when converting a plain
load to an extending one). This patch tries to catch all cases where
the TBAA information can legitimately be carried over.
The patch adds alternative forms of getLoad() and getExtLoad() that take
a MachineMemOperand instead of individual fields. (The corresponding
getTruncStore() already exists.) The idea is to use the MachineMemOperand
forms when all fields are carried over (size, pointer info, isVolatile,
isNonTemporal, alignment and TBAA info). If some adjustment is being
made, e.g. to narrow the load, then we still pass the individual fields
but also pass the TBAA info.
llvm-svn: 193517
This optimization is not SSE specific so I am moving it to DAGco.
The new scalar_to_vector dag node exposed a missing pattern in the AArch64 target that I needed to add.
llvm-svn: 193393
When canonicalizing dags according to the rule
(shl (zext (shr X, c1) ), c1) ==> (zext (shl (shr X, c1), c1))
remember to add the new shl dag to the DAGCombiner worklist of nodes.
If we don't explicitly add it to the worklist of nodes to visit, we
may not trigger later on the rule that folds the shift left + logical
shift right into a AND instruction with bitmask.
llvm-svn: 192883
This happens e.g. with <2 x i64> -1 on x86_32. It cannot be generated directly
because i64 is illegal. It would be nice if getNOT would handle this
transparently, but I don't see a way to generate a legal constant there right
now. Fixes PR17487.
llvm-svn: 192795
This should fix the buildbots.
Original commit message:
[DAGCombiner] Slice a big load in two loads when the element are next to each
other in memory and the target has paired load and performs post-isel loads
combining.
E.g., this optimization will transform something like this:
a = load i64* addr
b = trunc i64 a to i32
c = lshr i64 a, 32
d = trunc i64 c to i32
into:
b = load i32* addr1
d = load i32* addr2
Where addr1 = addr2 +/- sizeof(i32), if the target supports paired load and
performs post-isel loads combining.
One should overload TargetLowering::hasPairedLoad to provide this information.
The default is false.
<rdar://problem/14477220>
llvm-svn: 192476
other in memory and the target has paired load and performs post-isel loads
combining.
E.g., this optimization will transform something like this:
a = load i64* addr
b = trunc i64 a to i32
c = lshr i64 a, 32
d = trunc i64 c to i32
into:
b = load i32* addr1
d = load i32* addr2
Where addr1 = addr2 +/- sizeof(i32), if the target supports paired load and
performs post-isel loads combining.
One should overload TargetLowering::hasPairedLoad to provide this information.
The default is false.
<rdar://problem/14477220>
llvm-svn: 192471
DAGCombiner::visitFP_EXTEND will apply the following transformation:
fold (fpext (load x)) -> (fpext (fptrunc (extload x)))
but the implementation does not handle indexed loads (pre/post inc.), but did
not specifically ignore them either (unlike for extending loads, which it
already ignored), causing an assert when the transformation was applied to an
indexed load. This is the minimal fix for correctness (causing the
transformation to be skipped for indexed loads).
Unfortunately, I don't have an in-tree test case.
llvm-svn: 191989
This change fixes the problem reported in pr17380 and re-add the dagcombine
transformation ensuring that the value types are always legal if the
transformation is triggered after Legalization took place.
Added the test case from pr17380.
llvm-svn: 191509
(shl (zext (shr A, X)), X) => (zext (shl (shr A, X), X)).
The rule only triggers when there are no other uses of the
zext to avoid materializing more instructions.
This helps the DAGCombiner understand that the shl/shr
sequence can then be converted into an and instruction.
llvm-svn: 191393
The Type Legalizer recognizes that VSELECT needs to be split, because the type
is to wide for the given target. The same does not always apply to SETCC,
because less space is required to encode the result of a comparison. As a result
VSELECT is split and SETCC is unrolled into scalar comparisons.
This commit fixes the issue by checking for VSELECT-SETCC patterns in the DAG
Combiner. If a matching pattern is found, then the result mask of SETCC is
promoted to the expected vector mask for the given target. This mask has usually
te same size as the VSELECT return type (except for Intel KNL). Now the type
legalizer will split both VSELECT and SETCC.
This allows the following X86 DAG Combine code to sucessfully detect the MIN/MAX
pattern. This fixes PR16695, PR17002, and <rdar://problem/14594431>.
llvm-svn: 191130
C-like languages promote types like unsigned short to unsigned int before
performing an arithmetic operation. Currently the rotate matcher in the
DAGCombiner does not consider this situation.
This commit extends the DAGCombiner in the way that the pattern
(or (shl ([az]ext x), (*ext y)), (srl ([az]ext x), (*ext (sub 32, y))))
is folded into
([az]ext (rotl x, y))
The matching is restricted to aext and zext because in this cases the upper
bits are either undefined or known. Test case is included.
This fixes PR16726.
llvm-svn: 191049
C-like languages promote types like unsigned short to unsigned int before
performing an arithmetic operation. Currently the rotate matcher in the
DAGCombiner does not consider this situation.
This commit extends the DAGCombiner in the way that the pattern
(or (shl ([az]ext x), (*ext y)), (srl ([az]ext x), (*ext (sub 32, y))))
is folded into
([az]ext (rotl x, y))
The matching is restricted to aext and zext because in this cases the upper
bits are either undefined or known. Test case is included.
This fixes PR16726.
llvm-svn: 191045
DAGCombiner::isAlias can be called with SrcValue1 or SrcValue2 null, and we
can't use AA in this case (if we try, then the casting code in AA will assert).
llvm-svn: 190763
This uses the TargetSubtargetInfo::useAA() function to control the defaults of
the -combiner-alias-analysis and -combiner-global-alias-analysis options.
llvm-svn: 189564
We want to convert code like (or (srl N, 8), (shl N, 8)) into (srl (bswap N),
const), but this is only valid if the bits above 16 on the source pattern are
0, the checks we were doing on this were slightly wrong before.
llvm-svn: 189348
If we have a binary operation like ISD:ADD, we can set the result type
equal to the result type of one of its operands rather than using
TargetLowering::getPointerTy().
Also, any use of DAG.getIntPtrConstant(C) as an operand for a binary
operation can be replaced with:
DAG.getConstant(C, OtherOperand.getValueType());
llvm-svn: 189227
The small utility function that pattern matches Base + Index +
Offset patterns for loads and stores fails to recognize the base
pointer for loads/stores from/into an array at offset 0 inside a
loop. As a result DAGCombiner::MergeConsecutiveStores was not able
to merge all stores.
This commit fixes the issue by adding an additional pattern match
and also a test case.
Reviewer: Nadav
llvm-svn: 188936
A common idiom is to use zero and all-ones as sentinal values and to
check for both in a single conditional ("x != 0 && x != (unsigned)-1").
That generates code, for i32, like:
testl %edi, %edi
setne %al
cmpl $-1, %edi
setne %cl
andb %al, %cl
With this transform, we generate the simpler:
incl %edi
cmpl $1, %edi
seta %al
Similar improvements for other integer sizes and on other platforms. In
general, combining the two setcc instructions into one is better.
rdar://14689217
llvm-svn: 188315
This virtual function can be implemented by targets to specify the type
to use for the index operand of INSERT_VECTOR_ELT, EXTRACT_VECTOR_ELT,
INSERT_SUBVECTOR, EXTRACT_SUBVECTOR. The default implementation returns
the result from TargetLowering::getPointerTy()
The previous code was using TargetLowering::getPointerTy() for vector
indices, because this is guaranteed to be legal on all targets. However,
using TargetLowering::getPointerTy() can be a problem for targets with
pointer sizes that differ across address spaces. On such targets,
when vectors need to be loaded or stored to an address space other than the
default 'zero' address space (which is the address space assumed by
TargetLowering::getPointerTy()), having an index that
is a different size than the pointer can lead to inefficient
pointer calculations, (e.g. 64-bit adds for a 32-bit address space).
There is no intended functionality change with this patch.
llvm-svn: 187748
This patch prevents the following combine when the input vector is used more
than once.
insert_vector_elt (build_vector elt0, ..., eltN), NewEltIdx, idx
=>
build_vector elt0, ..., NewEltIdx, ..., eltN
The reasons are:
- Building a vector may be expensive, so try to reuse the existing part of a
vector instead of creating a new one (think big vectors).
- elt0 to eltN now have two users instead of one. This may prevent some other
optimizations.
llvm-svn: 187396
in-tree implementations of TargetLoweringBase::isFMAFasterThanMulAndAdd in
order to resolve the following issues with fmuladd (i.e. optional FMA)
intrinsics:
1. On X86(-64) targets, ISD::FMA nodes are formed when lowering fmuladd
intrinsics even if the subtarget does not support FMA instructions, leading
to laughably bad code generation in some situations.
2. On AArch64 targets, ISD::FMA nodes are formed for operations on fp128,
resulting in a call to a software fp128 FMA implementation.
3. On PowerPC targets, FMAs are not generated from fmuladd intrinsics on types
like v2f32, v8f32, v4f64, etc., even though they promote, split, scalarize,
etc. to types that support hardware FMAs.
The function has also been slightly renamed for consistency and to force a
merge/build conflict for any out-of-tree target implementing it. To resolve,
see comments and fixed in-tree examples.
llvm-svn: 185956
When folding sub x, x (and other similar constructs), where x is a vector, the
result is a vector of zeros. After type legalization, make sure that the input
zero elements have a legal type. This type may be larger than the result's
vector element type.
This was another bug found by llvm-stress.
llvm-svn: 185949
ReduceLoadWidth unconditionally drops extensions from loads. Limit it to the
case when all of the bits the extension would otherwise produce are dropped by
the shrink. It would be possible to shrink the load in more cases by merging
the extensions, but this isn't trivial and a very rare case. I left a TODO for
that case.
Fixes PR16551.
llvm-svn: 185755
DAGCombiner was counting all uses of a load node when considering whether it's
worth combining into a zextload. Really, it wants to ignore the chain and just
count real uses.
rdar://problem/13896307
llvm-svn: 185419
This already helps SSE2 x86 a lot because it lacks an efficient way to
represent a vector select. The long term goal is to enable the backend to match
a canonicalized pattern into a single instruction (e.g. vabs or pabs).
llvm-svn: 180597
This exposed an issue with PowerPC AltiVec where it appears it was setting the wrong vector boolean contents. The included change
fixes the PowerPC tests, and was OK'd by Hal.
llvm-svn: 180129
This pattern occurs in SROA output due to the way vector arguments are lowered
on ARM.
The testcase from PR15525 now compiles into this, which is better than the code
we got with the old scalarrepl:
_Store:
ldr.w r9, [sp]
vmov d17, r3, r9
vmov d16, r1, r2
vst1.8 {d16, d17}, [r0]
bx lr
Differential Revision: http://llvm-reviews.chandlerc.com/D647
llvm-svn: 179106
This is helps on architectures where i8,i16 are not legal but we have byte, and
short loads/stores. Allowing us to merge copies like the one below on ARM.
copy(char *a, char *b, int n) {
do {
int t0 = a[0];
int t1 = a[1];
b[0] = t0;
b[1] = t1;
radar://13536387
llvm-svn: 178546
We would also like to merge sequences that involve a variable index like in the
example below.
int index = *idx++
int i0 = c[index+0];
int i1 = c[index+1];
b[0] = i0;
b[1] = i1;
By extending the parsing of the base pointer to handle dags that contain a
base, index, and offset we can handle examples like the one above.
The dag for the code above will look something like:
(load (i64 add (i64 copyfromreg %c)
(i64 signextend (i8 load %index))))
(load (i64 add (i64 copyfromreg %c)
(i64 signextend (i32 add (i32 signextend (i8 load %index))
(i32 1)))))
The code that parses the tree ignores the intermediate sign extensions. However,
if there is a sign extension it needs to be on all indexes.
(load (i64 add (i64 copyfromreg %c)
(i64 signextend (add (i8 load %index)
(i8 1))))
vs
(load (i64 add (i64 copyfromreg %c)
(i64 signextend (i32 add (i32 signextend (i8 load %index))
(i32 1)))))
radar://13536387
llvm-svn: 178483
- Handle the case where the result of 'insert_subvect' is bitcasted
before 'extract_subvec'. This removes the redundant insertf128/extractf128
pair on unaligned 256-bit vector load/store on vectors of non 64-bit integer.
llvm-svn: 177945
For instance, following transformation will be disabled:
x + x + x => 3.0f * x;
The problem of these transformations is that it introduces a FP constant, which
following Instruction-Selection pass cannot handle.
Reviewed by Nadav, thanks a lot!
rdar://13445387
llvm-svn: 177933
LegalizeDAG.cpp uses the value of the comparison operands when checking
the legality of BR_CC, so DAGCombiner should do the same.
v2:
- Expand more BR_CC value types for NVPTX
v3:
- Expand correct BR_CC value types for Hexagon, Mips, and XCore.
llvm-svn: 176694
A legal BUILD_VECTOR goes in and gets constant folded into another legal
BUILD_VECTOR so we don't lose any legality here. The problematic PPC
optimization that made this check necessary was fixed recently.
llvm-svn: 175759
(2xi32) (truncate ((2xi64) bitcast (buildvector i32 a, i32 x, i32 b, i32 y)))
can be folded into a (2xi32) (buildvector i32 a, i32 b).
Such a DAG would cause uneccessary vdup instructions followed by vmovn
instructions.
We generate this code on ARM NEON for a setcc olt, 2xf64, 2xf64. For example, in
the vectorized version of the code below.
double A[N];
double B[N];
void test_double_compare_to_double() {
int i;
for(i=0;i<N;i++)
A[i] = (double)(A[i] < B[i]);
}
radar://13191881
Fixes bug 15283.
llvm-svn: 175670
DAGCombiner::ReduceLoadWidth was converting (trunc i32 (shl i64 v, 32))
into (shl i32 v, 32) into undef. To prevent this, check the shift count
against the final result size.
Patch by: Kevin Schoedel
Reviewed by: Nadav Rotem
llvm-svn: 174972
Sorry for the lack of a test case. I tried writing one for i386 as i know selects are illegal on this target, but they are actually considered legal by isel and expanded later.
I can't see any targets to trigger this, but checking for the legality of a node before forming it is general goodness.
llvm-svn: 174934
Previously, even when a pre-increment load or store was generated,
we often needed to keep a copy of the original base register for use
with other offsets. If all of these offsets are constants (including
the offset which was combined into the addressing mode), then this is
clearly unnecessary. This change adjusts these other offsets to use the
new incremented address.
llvm-svn: 174746
base point of a load, and the overall alignment of the load. This caused infinite loops in DAG combine with the
original application of this patch.
ORIGINAL COMMIT LOG:
When the target-independent DAGCombiner inferred a higher alignment for a load,
it would replace the load with one with the higher alignment. However, it did
not place the new load in the worklist, which prevented later DAG combines in
the same phase (for example, target-specific combines) from ever seeing it.
This patch corrects that oversight, and updates some tests whose output changed
due to slightly different DAGCombine outputs.
llvm-svn: 174431
it would replace the load with one with the higher alignment. However, it did
not place the new load in the worklist, which prevented later DAG combines in
the same phase (for example, target-specific combines) from ever seeing it.
This patch corrects that oversight, and updates some tests whose output changed
due to slightly different DAGCombine outputs.
llvm-svn: 174343
The optimization handles esoteric cases but adds a lot of complexity both to the X86 backend and to other backends.
This optimization disables an important canonicalization of chains of SEXT nodes and makes SEXT and ZEXT asymmetrical.
Disabling the canonicalization of consecutive SEXT nodes into a single node disables other DAG optimizations that assume
that there is only one SEXT node. The AVX mask optimizations is one example. Additionally this optimization does not update the cost model.
llvm-svn: 172968
The included test case is derived from one of the GCC compatibility tests.
The problem arises after the selection DAG has been converted to type-legalized
form. The combiner first sees a 64-bit load that can be converted into a
pre-increment form. The original load feeds into a SRL that isolates the
upper 32 bits of the loaded doubleword. This looks like an opportunity for
DAGCombiner::ReduceLoadWidth() to replace the 64-bit load with a 32-bit load.
However, this transformation is not valid, as the replacement load is not
a pre-increment load. The pre-increment load produces an extra result,
which feeds a subsequent add instruction. The replacement load only has
one result value, and this value is propagated to all uses of the pre-
increment load, including the add. Because the add is looking for the
second result value as its operand, it ends up attempting to add a constant
to a token chain, resulting in a crash.
So the patch simply disables this transformation for any load with more than
two result values.
llvm-svn: 172480
It cahced XOR's operands before calling visitXOR() but failed to update the
operands when visitXOR changed the XOR node.
rdar://12968664
llvm-svn: 171999
peculiar headers under include/llvm.
This struct still doesn't make a lot of sense, but it makes more sense
down in TargetLowering than it did before.
llvm-svn: 171739
DAGCombiner::reduceBuildVecConvertToConvertBuildVec() was making two
mistakes:
1. It was checking the legality of scalar INT_TO_FP nodes and then generating
vector nodes.
2. It was passing the result value type to
TargetLoweringInfo::getOperationAction() when it should have been
passing the value type of the first operand.
llvm-svn: 171420
into their new header subdirectory: include/llvm/IR. This matches the
directory structure of lib, and begins to correct a long standing point
of file layout clutter in LLVM.
There are still more header files to move here, but I wanted to handle
them in separate commits to make tracking what files make sense at each
layer easier.
The only really questionable files here are the target intrinsic
tablegen files. But that's a battle I'd rather not fight today.
I've updated both CMake and Makefile build systems (I think, and my
tests think, but I may have missed something).
I've also re-sorted the includes throughout the project. I'll be
committing updates to Clang, DragonEgg, and Polly momentarily.
llvm-svn: 171366
directly.
This is in preparation for removing the use of the 'Attribute' class as a
collection of attributes. That will shift to the AttributeSet class instead.
llvm-svn: 171253
try to reduce the width of this load, and would end up transforming:
(truncate (lshr (sextload i48 <ptr> as i64), 32) to i32)
to
(truncate (zextload i32 <ptr+4> as i64) to i32)
We lost the sext attached to the load while building the narrower i32
load, and replaced it with a zext because lshr always zext's the
results. Instead, bail out of this combine when there is a conflict
between a sextload and a zext narrowing. The rest of the DAG combiner
still optimize the code down to the proper single instruction:
movswl 6(...),%eax
Which is exactly what we wanted. Previously we read past the end *and*
missed the sign extension:
movl 6(...), %eax
llvm-svn: 169802
Sooooo many of these had incorrect or strange main module includes.
I have manually inspected all of these, and fixed the main module
include to be the nearest plausible thing I could find. If you own or
care about any of these source files, I encourage you to take some time
and check that these edits were sensible. I can't have broken anything
(I strictly added headers, and reordered them, never removed), but they
may not be the headers you'd really like to identify as containing the
API being implemented.
Many forward declarations and missing includes were added to a header
files to allow them to parse cleanly when included first. The main
module rule does in fact have its merits. =]
llvm-svn: 169131
checks to avoid performing compile-time arithmetic on PPCDoubleDouble.
Now that APFloat supports arithmetic on PPCDoubleDouble, those checks
are no longer needed, and we can treat the type like any other.
llvm-svn: 166958
- If more than 1 elemennts are defined and target supports the vectorized
conversion, use the vectorized one instead to reduce the strength on
conversion operation.
llvm-svn: 166546
- Folding (trunc (concat ... X )) to (concat ... (trunc X) ...) is valid
when '...' are all 'undef's.
- r166125 relies on this transformation.
llvm-svn: 166155
- If the extracted vector has the same type of all vectored being concatenated
together, it should be simplified directly into v_i, where i is the index of
the element being extracted.
llvm-svn: 166125
This class is used by LSR and a number of places in the codegen.
This is the first step in de-coupling LSR from TLI, and creating
a new interface in between them.
llvm-svn: 165455
multiple stores with a single load. We create the wide loads and stores (and their chains)
before we remove the scalar loads and stores and fix the DAG chain. We attempted to merge
loads with a different chain. When that happened, the assumption that it is safe to RAUW
broke and a cycle was introduced.
llvm-svn: 165148
is not profitable in many cases because modern processors perform multiple stores
in parallel and merging stores prior to merging requires extra work. We handle two main cases:
1. Store of multiple consecutive constants:
q->a = 3;
q->4 = 5;
In this case we store a single legal wide integer.
2. Store of multiple consecutive loads:
int a = p->a;
int b = p->b;
q->a = a;
q->b = b;
In this case we load/store either ilegal vector registers or legal wide integer registers.
llvm-svn: 165125
because moden processos can store multiple values in parallel, and preparing the consecutive store requires
some work. We only handle these cases:
1. Consecutive stores where the values and consecutive loads. For example:
int a = p->a;
int b = p->b;
q->a = a;
q->b = b;
2. Consecutive stores where the values are constants. Foe example:
q->a = 4;
q->b = 5;
llvm-svn: 164910
buildbots. Original commit message:
A DAGCombine optimization for merging consecutive stores. This optimization is not profitable in many cases
because moden processos can store multiple values in parallel, and preparing the consecutive store requires
some work. We only handle these cases:
1. Consecutive stores where the values and consecutive loads. For example:
int a = p->a;
int b = p->b;
q->a = a;
q->b = b;
2. Consecutive stores where the values are constants. Foe example:
q->a = 4;
q->b = 5;
llvm-svn: 164890
because moden processos can store multiple values in parallel, and preparing the consecutive store requires
some work. We only handle these cases:
1. Consecutive stores where the values and consecutive loads. For example:
int a = p->a;
int b = p->b;
q->a = a;
q->b = b;
2. Consecutive stores where the values are constants. Foe example:
q->a = 4;
q->b = 5;
llvm-svn: 164885
by xoring the high-bit. This fails if the source operand is a vector because we need to negate
each of the elements in the vector.
Fix rdar://12281066 PR13813.
llvm-svn: 163802
The DAGCombiner tries to optimise a BUILD_VECTOR by checking if it
consists purely of get_vector_elts from one or two source vectors. If
so, it either makes a concat_vectors node or a shufflevector node.
However, it doesn't check the element type width of the underlying
vector, so if you have this sequence:
Node0: v4i16 = ...
Node1: i32 = extract_vector_elt Node0
Node2: i32 = extract_vector_elt Node0
Node3: v16i8 = BUILD_VECTOR Node1, Node2, ...
It will attempt to:
Node0: v4i16 = ...
NewNode1: v16i8 = concat_vectors Node0, ...
Where this is actually invalid because the element width is completely
different. This causes an assertion failure on DAG legalization stage.
Fix:
If output item type of BUILD_VECTOR differs from input item type.
Make concat_vectors based on input element type and then bitcast it to the output vector type. So the case described above will transformed to:
Node0: v4i16 = ...
NewNode1: v8i16 = concat_vectors Node0, ...
NewNode2: v16i8 = bitcast NewNode1
llvm-svn: 162195
Add a micro-optimization to getNode of CONCAT_VECTORS when both operands are undefs.
Can't find a testcase for this because VECTOR_SHUFFLE already handles undef operands, but Duncan suggested that we add this.
Together with Michael Kuperstein <michael.m.kuperstein@intel.com>
llvm-svn: 160229
multiple scalars and insert them into a vector. Next, we shuffle the elements
into the correct places, as before.
Also fix a small dagcombine bug in SimplifyBinOpWithSameOpcodeHands, when the
migration of bitcasts happened too late in the SelectionDAG process.
llvm-svn: 159991
boolean flag to an enum: { Fast, Standard, Strict } (default = Standard).
This option controls the creation by optimizations of fused FP ops that store
intermediate results in higher precision than IEEE allows (E.g. FMAs). The
behavior of this option is intended to match the behaviour specified by a
soon-to-be-introduced frontend flag: '-ffuse-fp-ops'.
Fast mode - allows formation of fused FP ops whenever they're profitable.
Standard mode - allow fusion only for 'blessed' FP ops. At present the only
blessed op is the fmuladd intrinsic. In the future more blessed ops may be
added.
Strict mode - allow fusion only if/when it can be proven that the excess
precision won't effect the result.
Note: This option only controls formation of fused ops by the optimizers. Fused
operations that are explicitly requested (e.g. FMA via the llvm.fma.* intrinsic)
will always be honored, regardless of the value of this option.
Internally TargetOptions::AllowExcessFPPrecision has been replaced by
TargetOptions::AllowFPOpFusion.
llvm-svn: 158956
This patch adds DAG combines to form FMAs from pairs of FADD + FMUL or
FSUB + FMUL. The combines are performed when:
(a) Either
AllowExcessFPPrecision option (-enable-excess-fp-precision for llc)
OR
UnsafeFPMath option (-enable-unsafe-fp-math)
are set, and
(b) TargetLoweringInfo::isFMAFasterThanMulAndAdd(VT) is true for the type of
the FADD/FSUB, and
(c) The FMUL only has one user (the FADD/FSUB).
If your target has fast FMA instructions you can make use of these combines by
overriding TargetLoweringInfo::isFMAFasterThanMulAndAdd(VT) to return true for
types supported by your FMA instruction, and adding patterns to match ISD::FMA
to your FMA instructions.
llvm-svn: 158757
When a combine twiddles an extract_vector, care should be take to preserve
the type of the index operand. No luck extracting a reasonable testcase,
unfortunately.
rdar://11391009
llvm-svn: 156419
Instead of passing listener pointers to RAUW, let SelectionDAG itself
keep a linked list of interested listeners.
This makes it possible to have multiple listeners active at once, like
RAUWUpdateListener was already doing. It also makes it possible to
register listeners up the call stack without controlling all RAUW calls
below.
DAGUpdateListener uses an RAII pattern to add itself to the SelectionDAG
list of active listeners.
llvm-svn: 155248
Fix a dagcombine optimization which assumes that the vsetcc result type is always
of the same size as the compared values. This is ture for SSE/AVX/NEON but not
for all targets.
llvm-svn: 154490
when -ffast-math, i.e. don't just always do it if the reciprocal can
be formed exactly. There is already an IR level transform that does
that, and it does it more carefully.
llvm-svn: 154296
shuffle node because it could introduce new shuffle nodes that were not
supported efficiently by the target.
2. Add a more restrictive shuffle-of-shuffle optimization for cases where the
second shuffle reverses the transformation of the first shuffle.
llvm-svn: 154266
reciprocal if converting to the reciprocal is exact. Do it even if inexact
if -ffast-math. This substantially speeds up ac.f90 from the polyhedron
benchmarks.
llvm-svn: 154265
This allows us to keep passing reduced masks to SimplifyDemandedBits, but
know about all the bits if SimplifyDemandedBits fails. This allows instcombine
to simplify cases like the one in the included testcase.
llvm-svn: 154011
Do not try to optimize swizzles of shuffles if the source shuffle has more than
a single user, except when the source shuffle is also a swizzle.
llvm-svn: 153864
(i16 load $addr+c*sizeof(i16)) and replace uses of (i32 vextract) with the
i16 load. It should issue an extload instead: (i32 extload $addr+c*sizeof(i16)).
rdar://11035895
llvm-svn: 152675
v8i8 -> v8i32 on AVX machines. The codegen often scalarizes ANY_EXTEND nodes.
The DAGCombiner has two optimizations that can mitigate the problem. First,
if all of the operands of a BUILD_VECTOR node are extracted from an ZEXT/ANYEXT
nodes, then it is possible to create a new simplified BUILD_VECTOR which uses
UNDEFS/ZERO values to eliminate the scalar ZEXT/ANYEXT nodes.
Second, another dag combine optimization lowers BUILD_VECTOR into a shuffle
vector instruction.
In the case of zext v8i8->v8i32 on AVX, a value in an XMM register is to be
shuffled into a wide YMM register.
This patch modifes the second optimization and allows the creation of
shuffle vectors even when the newly generated vector and the original vector
from which we extract the values are of different types.
llvm-svn: 150340
In this patch we optimize this pattern and convert the sequence into extract op of a narrow type.
This allows the BUILD_VECTOR dag optimizations to construct efficient shuffle operations in many cases.
llvm-svn: 149692
overly conservative. It was concerned about cases where it would prohibit
folding simple [r, c] addressing modes. e.g.
ldr r0, [r2]
ldr r1, [r2, #4]
=>
ldr r0, [r2], #4
ldr r1, [r2]
Change the logic to look for such cases which allows it to form indexed memory
ops more aggressively.
rdar://10674430
llvm-svn: 148086
detect a pattern which can be implemented with a small 'shl' embedded in
the addressing mode scale. This happens in real code as follows:
unsigned x = my_accelerator_table[input >> 11];
Here we have some lookup table that we look into using the high bits of
'input'. Each entity in the table is 4-bytes, which means this
implicitly gets turned into (once lowered out of a GEP):
*(unsigned*)((char*)my_accelerator_table + ((input >> 11) << 2));
The shift right followed by a shift left is canonicalized to a smaller
shift right and masking off the low bits. That hides the shift right
which x86 has an addressing mode designed to support. We now detect
masks of this form, and produce the longer shift right followed by the
proper addressing mode. In addition to saving a (rather large)
instruction, this also reduces stalls in Intel chips on benchmarks I've
measured.
In order for all of this to work, one part of the DAG needs to be
canonicalized *still further* than it currently is. This involves
removing pointless 'trunc' nodes between a zextload and a zext. Without
that, we end up generating spurious masks and hiding the pattern.
llvm-svn: 147936
a combined-away node and the result of the combine isn't substantially
smaller than the input, it's just canonicalized. This is the first part
of a significant (7%) performance gain for Snappy's hot decompression
loop.
llvm-svn: 147604
undefined result. This adds new ISD nodes for the new semantics,
selecting them when the LLVM intrinsic indicates that the undef behavior
is desired. The new nodes expand trivially to the old nodes, so targets
don't actually need to do anything to support these new nodes besides
indicating that they should be expanded. I've done this for all the
operand types that I could figure out for all the targets. Owners of
various targets, please review and let me know if any of these are
incorrect.
Note that the expand behavior is *conservatively correct*, and exactly
matches LLVM's current behavior with these operations. Ideally this
patch will not change behavior in any way. For example the regtest suite
finds the exact same instruction sequences coming out of the code
generator. That's why there are no new tests here -- all of this is
being exercised by the existing test suite.
Thanks to Duncan Sands for reviewing the various bits of this patch and
helping me get the wrinkles ironed out with expanding for each target.
Also thanks to Chris for clarifying through all the discussions that
this is indeed the approach he was looking for. That said, there are
likely still rough spots. Further review much appreciated.
llvm-svn: 146466
change, now you need a TargetOptions object to create a TargetMachine. Clang
patch to follow.
One small functionality change in PTX. PTX had commented out the machine
verifier parts in their copy of printAndVerify. That now calls the version in
LLVMTargetMachine. Users of PTX who need verification disabled should rely on
not passing the command-line flag to enable it.
llvm-svn: 145714
Conservatively returns zero when the GV does not specify an alignment nor is it
initialized. Previously it returns ABI alignment for type of the GV. However, if
the type is a "packed" type, then the under-specified alignments is attached to
the load / store instructions. In that case, the alignment of the type cannot be
trusted.
rdar://10464621
llvm-svn: 145300
than ABI alignment. These are loads / stores from / to "packed" data structures.
Their alignments are intentionally under-specified.
rdar://10301431
llvm-svn: 145273
Add support for trimming constants to GetDemandedBits. This fixes some funky
constant generation that occurs when stores are expanded for targets that don't
support unaligned stores natively.
llvm-svn: 144102
When this field is true it means that the load is from constant (runt-time or compile-time) and so can be hoisted from loops or moved around other memory accesses
llvm-svn: 144100
svn r139159 caused SelectionDAG::getConstant() to promote BUILD_VECTOR operands
with illegal types, even before type legalization. For this testcase, that led
to one BUILD_VECTOR with i16 operands and another with promoted i32 operands,
which triggered the assertion.
llvm-svn: 142370
with a vector condition); such selects become VSELECT codegen nodes.
This patch also removes VSETCC codegen nodes, unifying them with SETCC
nodes (codegen was actually often using SETCC for vector SETCC already).
This ensures that various DAG combiner optimizations kick in for vector
comparisons. Passes dragonegg bootstrap with no testsuite regressions
(nightly testsuite as well as "make check-all"). Patch mostly by
Nadav Rotem.
llvm-svn: 139159
hasPredecessorHelper function allows predecessors to be cached to speed up
repeated invocations. This fixes PR10186.
X.isPredecessorOf(Y) now just calls Y.hasPredecessor(X)
Y.hasPredecessor(X) calls Y.hasPredecessorHelper(X, Visited, Worklist) with
empty Visited and Worklist sets (i.e. no caching over invocations).
Y.hasPredecessorHelper(X, Visited, Worklist) caches search state in Visited
and Worklist to speed up repeated calls. The Visited set is searched for X
before going to the worklist to further search the DAG if necessary.
llvm-svn: 134592
1. (((x) & 0xFF00) >> 8) | (((x) & 0x00FF) << 8)
=> (bswap x) >> 16
2. ((x&0xff)<<8)|((x&0xff00)>>8)|((x&0xff000000)>>8)|((x&0x00ff0000)<<8))
=> (rotl (bswap x) 16)
This allows us to eliminate most of the def : Pat patterns for ARM rev16
revsh instructions. It catches many more cases for ARM and x86.
rdar://9609108
llvm-svn: 133503
GetDemandBits (which must operate on the vector element type).
Fix the a usage of getZeroExtendInReg which must also be done on scalar types.
llvm-svn: 133052
converted to add x,x if x is a undef. add undef, undef does not guarantee
that the resulting low order bit is zero.
Fixes <rdar://problem/9453156> and <rdar://problem/9487392>.
llvm-svn: 133022
The potential DAGCombine which enforces this more generally messes up some other very fragile patterns, so I'm leaving that alone, at least for now.
llvm-svn: 132809
If there is a store after the load node, then there is a chain, which means
that there is another user. Thus, asking hasOneUser would fail. Instead we
ask hasNUsesOfValue on the 'data' value.
llvm-svn: 131183
transformations in target-specific DAG combines without causing DAGCombiner to
delete the same node twice. If you know of a better way to avoid this (see my
next patch for an example), please let me know.
llvm-svn: 128758
1. Inform users of ADDEs with two 0 operands that it never sets carry
2. Fold other ADDs or ADDCs into the ADDE if possible
It would be neat if we could do the same thing for SETCC+ADD eventually, but we can't do that in target independent code.
llvm-svn: 126557
Limit the folding of any_ext and sext into the load operation to scalars.
Limit the active-bits trunc optimization to scalars.
Document vector trunc and vector sext in LangRef.
Similar to commit 126080 (for enabling zext).
llvm-svn: 126424
The DAGCombiner folds the zext into complex load instructions. This patch
prevents this optimization on vectors since none of the supported targets
knows how to perform load+vector_zext in one instruction.
llvm-svn: 126080
transformation if we can't legally create a build vector of the correct
type. Check that we can make the transformation first, and add a TODO to
refactor this code with similar cases.
Fixes: PR9223 and rdar://9000350
llvm-svn: 125631
generating i8 shift amounts for things like i1024 types. Add
an assert in getNode to prevent this from occuring in the future,
fix the buggy transformation, revert my previous patch, and
document this gotcha in ISDOpcodes.h
llvm-svn: 125465
The DAGCombiner created illegal BUILD_VECTOR operations.
The patch added a check that either illegal operations are
allowed or that the created operation is legal.
llvm-svn: 125435
The bug happens when the DAGCombiner attempts to optimize one of the patterns
of the SUB opcode. It tries to create a zero of type v2i64. This type is legal
on 32bit machines, but the initializer of this vector (i64) is target dependent.
Currently, the initializer attempts to create an i64 zero constant, which fails.
Added a flag to tell the DAGCombiner to create a legal zero, if we require that
the pass would generate legal types.
llvm-svn: 125391
the load, then it may be legal to transform the load and store to integer
load and store of the same width.
This is done if the target specified the transformation as profitable. e.g.
On arm, this can transform:
vldr.32 s0, []
vstr.32 s0, []
to
ldr r12, []
str r12, []
rdar://8944252
llvm-svn: 124708
This happens all the time when a smul is promoted to a larger type.
On x86-64 we now compile "int test(int x) { return x/10; }" into
movslq %edi, %rax
imulq $1717986919, %rax, %rax
movq %rax, %rcx
shrq $63, %rcx
sarq $34, %rax <- used to be "shrq $32, %rax; sarl $2, %eax"
addl %ecx, %eax
This fires 96 times in gcc.c on x86-64.
llvm-svn: 124559
This happens e.g. for code like "X - X%10" where we lower the modulo operation
to a series of multiplies and shifts that are then subtracted from X, leading to
this missed optimization.
llvm-svn: 124532
loads properly. We miscompiled the testcase into:
_test: ## @test
movl $128, (%rdi)
movzbl 1(%rdi), %eax
ret
Now we get a proper:
_test: ## @test
movl $128, (%rdi)
movsbl (%rdi), %eax
movzbl %ah, %eax
ret
This fixes PR8757.
llvm-svn: 122392
count operand. These should be the same but apparently are
not always, and this is cleaner anyway. This improves the
code in an existing test.
llvm-svn: 122354
BUILD_VECTOR operands where the element type is not legal. I had previously
changed this code to insert TRUNCATE operations, but that was just wrong.
llvm-svn: 122102
when the wider type is legal. This allows us to compile:
define zeroext i16 @test1(i16 zeroext %x) nounwind {
entry:
%div = udiv i16 %x, 33
ret i16 %div
}
into:
test1: # @test1
movzwl 4(%esp), %eax
imull $63551, %eax, %eax # imm = 0xF83F
shrl $21, %eax
ret
instead of:
test1: # @test1
movw $-1985, %ax # imm = 0xFFFFFFFFFFFFF83F
mulw 4(%esp)
andl $65504, %edx # imm = 0xFFE0
movl %edx, %eax
shrl $5, %eax
ret
Implementing rdar://8760399 and example #4 from:
http://blog.regehr.org/archives/320
We should implement the same thing for [su]mul_hilo, but I don't
have immediate plans to do this.
llvm-svn: 121696
zextOrTrunc(), and APSInt methods extend(), extOrTrunc() and new method
trunc(), to be const and to return a new value instead of modifying the
object in place.
llvm-svn: 121120
if the extension types were not the same. The result was that if you
fed a select with sext and zext loads, as in the testcase, then it
would get turned into a zext (or sext) of the select, which is wrong
in the cases when it should have been an sext (resp. zext). Reported
and diagnosed by Sebastien Deldon.
llvm-svn: 119728
// %a = ...
// %b = and i32 %a, 2
// %c = srl i32 %b, 1
// brcond i32 %c ...
//
// into
//
// %a = ...
// %b = and i32 %a, 2
// %c = setcc eq %b, 0
// brcond %c ...
Make sure it restores local variable N1, which corresponds to the condition operand if it fails to match.
This apparently breaks TCE but since that backend isn't in the tree I don't have a test for it.
llvm-svn: 115571
CombinerAA cannot assume that different FrameIndex's never alias, but can instead use
MachineFrameInfo to get the actual offsets of these slots and check for actual aliasing.
This fixes CodeGen/X86/2010-02-19-TailCallRetAddrBug.ll and CodeGen/X86/tailcallstack64.ll
when CombinerAA is enabled, modulo a different register allocation sequence.
llvm-svn: 114348
there are clearly no stores between the load and the store. This fixes
this miscompile reported as PR7833.
This breaks the test/CodeGen/X86/narrow_op-2.ll optimization, which is
safe, but awkward to prove safe. Move it to X86's README.txt.
llvm-svn: 112861
I am assured by people more knowledgeable than me that there are no rounding issues in eliminating this.
This fixed <rdar://problem/8197504>.
llvm-svn: 108639
disabled and then never turned back on again. Adjust some tests, one because
this change avoids an unnecessary instruction, and the other to make it
continue testing what it was intended to test.
llvm-svn: 107941
atomic intrinsics, either because the use locking instructions for the
atomics, or because they perform the locking directly. Add support in the
DAG combiner to fold away the fences.
llvm-svn: 106630
Mon Ping provided; unfortunately bugpoint failed to
reduce it, but I think it's important to have a test for
this in the suite. 8023512.
llvm-svn: 104624
so that it will continue to test what it was meant to test when I commit a
separate change for better support of BUILD_VECTOR and VECTOR_SHUFFLE for Neon.
Fix a DAG combiner crash exposed by this test change.
llvm-svn: 104380
The trouble arises when the result of a vector cmp + sext is then and'ed with all ones. Instcombine will turn it into a vector cmp + zext, dag combiner will miss turning it into a vsetcc and hell breaks loose after that.
Teach dag combine to turn a vector cpm + zest into a vsetcc + and 1. This fixes rdar://7923010.
llvm-svn: 104094
This doesn't occur much at all, it only seems to formed in the case
when the trunc optimization kicks in due to phase ordering. In that
case it is saves a few bytes on x86-32.
llvm-svn: 101350
a load/or/and/store sequence into a narrower store when it is
safe. Daniel tells me that clang will start producing this sort
of thing with bitfields, and this does trigger a few dozen times
on 176.gcc produced by llvm-gcc even now.
This compiles code like CodeGen/X86/2009-05-28-DAGCombineCrash.ll
into:
movl %eax, 36(%rdi)
instead of:
movl $4294967295, %eax ## imm = 0xFFFFFFFF
andq 32(%rdi), %rax
shlq $32, %rcx
addq %rax, %rcx
movq %rcx, 32(%rdi)
and each of the testcases into a single store. Each of them used
to compile into craziness like this:
_test4:
movl $65535, %eax ## imm = 0xFFFF
andl (%rdi), %eax
shll $16, %esi
addl %eax, %esi
movl %esi, (%rdi)
ret
llvm-svn: 101343
1. Makes it possible to lower with floating point loads and stores.
2. Avoid unaligned loads / stores unless it's fast.
3. Fix some memcpy lowering logic bug related to when to optimize a
load from constant string into a constant.
4. Adjust x86 memcpy lowering threshold to make it more sane.
5. Fix x86 target hook so it uses vector and floating point memory
ops more effectively.
rdar://7774704
llvm-svn: 100090
long test(long x) { return (x & 123124) | 3; }
Currently compiles to:
_test:
orl $3, %edi
movq %rdi, %rax
andq $123127, %rax
ret
This is because instruction and DAG combiners canonicalize
(or (and x, C), D) -> (and (or, D), (C | D))
However, this is only profitable if (C & D) != 0. It gets in the way of the
3-addressification because the input bits are known to be zero.
llvm-svn: 97616
dragonegg self-host build. I reverted 96640 in order to revert
96556 (96640 goes on top of 96556), but it also looks like with
both of them applied the breakage happens even earlier. The
symptom of the 96556 miscompile is the following crash:
llvm[3]: Compiling AlphaISelLowering.cpp for Release build
cc1plus: /home/duncan/tmp/tmp/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:4982: void llvm::SelectionDAG::ReplaceAllUsesWith(llvm::SDNode*, llvm::SDNode*, llvm::SelectionDAG::DAGUpdateListener*): Assertion `(!From->hasAnyUseOfValue(i) || From->getValueType(i) == To->getValueType(i)) && "Cannot use this version of ReplaceAllUsesWith!"' failed.
Stack dump:
0. Running pass 'X86 DAG->DAG Instruction Selection' on function '@_ZN4llvm19AlphaTargetLowering14LowerOperationENS_7SDValueERNS_12SelectionDAGE'
g++: Internal error: Aborted (program cc1plus)
This occurs when building LLVM using LLVM built by LLVM (via
dragonegg). Probably LLVM has miscompiled itself, though it
may have miscompiled GCC and/or dragonegg itself: at this point
of the self-host build, all of GCC, LLVM and dragonegg were built
using LLVM. Unfortunately this kind of thing is extremely hard
to debug, and while I did rummage around a bit I didn't find any
smoking guns, aka obviously miscompiled code.
Found by bisection.
r96556 | evancheng | 2010-02-18 03:13:50 +0100 (Thu, 18 Feb 2010) | 5 lines
Some dag combiner goodness:
Transform br (xor (x, y)) -> br (x != y)
Transform br (xor (xor (x,y), 1)) -> br (x == y)
Also normalize (and (X, 1) == / != 1 -> (and (X, 1)) != / == 0 to match to "test on x86" and "tst on arm"
r96640 | evancheng | 2010-02-19 01:34:39 +0100 (Fri, 19 Feb 2010) | 16 lines
Transform (xor (setcc), (setcc)) == / != 1 to
(xor (setcc), (setcc)) != / == 1.
e.g. On x86_64
%0 = icmp eq i32 %x, 0
%1 = icmp eq i32 %y, 0
%2 = xor i1 %1, %0
br i1 %2, label %bb, label %return
=>
testl %edi, %edi
sete %al
testl %esi, %esi
sete %cl
cmpb %al, %cl
je LBB1_2
llvm-svn: 96672
type is the same as the element type of the vector. EXTRACT_VECTOR_ELT can
be used to extended the width of an integer type. This fixes a bug for
Generic/vector-casts.ll on a ppc750.
llvm-svn: 94990
Instcombine does this but apparently there are situations where this pattern will escape the optimizer and / or created by isel. Here is a case that's seen in JavaScriptCore:
%t1 = sub i32 0, %a
%t2 = add i32 %t1, -1
The dag combiner pattern: ((c1-A)+c2) -> (c1+c2)-A
will fold it to -1 - %a.
llvm-svn: 93773
really does need to be a vector type, because
TargetLowering::getOperationAction for SIGN_EXTEND_INREG uses that type,
and it needs to be able to distinguish between vectors and scalars.
Also, fix some more issues with legalization of vector casts.
llvm-svn: 93043
When folding a and(any_ext(load)) both the any_ext and the
load have to have only a single use.
This removes the anyext-uses.ll testcase which started failing
because it is unreduced and unclear what it is testing.
llvm-svn: 92950
(OP (trunc x), (trunc y)) -> (trunc (OP x, y))
Unfortunately this simple change causes dag combine to infinite looping. The problem is the shrink demanded ops optimization tend to canonicalize expressions in the opposite manner. That is badness. This patch disable those optimizations in dag combine but instead it is done as a late pass in sdisel.
This also exposes some deficiencies in dag combine and x86 setcc / brcond lowering. Teach them to look pass ISD::TRUNCATE in various places.
llvm-svn: 92849
Fold (zext (and x, cst)) -> (and (zext x), cst)
DAG combiner likes to optimize expression in the other way so this would end up cause an infinite looping.
llvm-svn: 91574
1. Only perform (zext (shl (zext x), y)) -> (shl (zext x), y) when y is a constant. This makes sure it remove at least one zest.
2. If the shift is a left shift, make sure the original shift cannot shift out bits.
llvm-svn: 91399
unconditional branches or fallthroghes. Instcombine/SimplifyCFG
should be simplifying branches with known conditions.
This fixes some problems caused by these transformations not
updating the MachineBasicBlock CFG.
llvm-svn: 89017
1. Switch from an std::set to a SmallPtrSet for visited chain nodes.
2. Do not force the recursive flattening of token factor nodes, regardless of
use count.
3. Immediately process newly created TokenFactor nodes.
Also, improve combiner-aa by teaching it that loads to non-overlapping offsets
of relatively aligned objects cannot alias.
These changes result in a >5x speedup for combiner-aa on most testcases.
llvm-svn: 81816
MERGE_VALUES nodes. Replacing the result values with the
operands in one MERGE_VALUES node may cause another
MERGE_VALUES node be CSE'd with the first one, and bring
its uses along, so that the first one isn't dead, as this
code expects. Fix this by iterating until the node is
really dead. This fixes PR4699.
llvm-svn: 78619
Chandler Carruth