Move the Factory.h header file to flang/Optimizer/Builder.
This patch is part of the upstreaming effort from fir-dev branch.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D115801
Use the "pure" attribute (or "readonly") for the vload, vload_half and
vloada_half builtins.
Reviewed By: svenvh
Differential Revision: https://reviews.llvm.org/D110742
This allows the pass to participate in progressive lowering
and it also allows us to write tests better.
Along the way, cleaned up the tests.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D115756
We got an unintended consequence of the optimizer getting smarter when
compiling in a non-standard mode, and there's no good way to inhibit
those optimizations at a later stage. The test is based on an example
linked from D92270.
We allow the "no-strict-float-cast-overflow" exception to normal C
cast rules to preserve legacy code that does not expect overflowing
casts from FP to int to produce UB. See D46236 for details.
Differential Revision: https://reviews.llvm.org/D115804
This patch adds support for prologue and epilogue generation for the z/OS target under the XPLINK64 ABI for functions with a stack size of less than 1048576 bytes (huge stack frames).
Reviewed By: uweigand
Differential Revision: https://reviews.llvm.org/D114457
Summary:
A new option exec-on-ir-changed is defined that allows one to specify an
exe that is called after each pass in the opt pipeline that changes the IR.
The exec-on-ir-change=exe option saves the IR in a temporary file and calls exe
with the name of the file and the name of the pass that just changed it after
each pass alters the IR. exe is also called with the initial IR. This
can be used, for example, to determine which pass corrupts the IR by having
exe as a script that calls llc and runs a test to see after which pass the
results change. The print-changed filtering options are respected. Note that
this is only supported with the new pass manager.
Author: Jamie Schmeiser <schmeise@ca.ibm.com>
Reviewed By: aeubanks (Arthur Eubanks)
Differential Revision: https://reviews.llvm.org/D110776
Currently, lld is marked as a dependency on Windows in
cross-project-tests/CMakeLists.txt which means CMake will fail if lld isn't
enabled. The idea of the cross-project-tests is that tests that don't have
their dependencies met should just be unsupported.
Remove the depenency from the CMake step and check whether Dexter's
platform-specific dependencies have been met in
cross-project-tests/lit.cfg.py. If the dependencies are met then add 'dexter'
to the available_features, otherwise don't and the dexter tests will be
"UNSUPPORTED".
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D115872
GCC currently does not allow `__builtin_strlen()` during constant evaluation. This PR adds a workaround in `std::char_traits<char>::length()`
Reviewed By: Quuxplusone, ldionne, #libc, Mordante
Spies: Mordante, libcxx-commits
Differential Revision: https://reviews.llvm.org/D115795
This change adds an option to disable warnings from the
cppcoreguidelines-narrowing-conversions check on integer to floating-
point conversions which may be narrowing.
An example of a case where this might be useful:
```
std::vector<double> v = {1, 2, 3, 4};
double mean = std::accumulate(v.cbegin(), v.cend(), 0.0) / v.size();
```
The conversion from std::size_t to double is technically narrowing on
64-bit systems, but v almost certainly does not have enough elements
for this to be a problem.
This option would allow the cppcoreguidelines-narrowing-conversions
check to be enabled on codebases which might otherwise turn it off
because of cases like the above.
When parsing the following construct, we parse it as an erroneous
deduction guide declaration and correctly diagnose the issues with it.
template<class> struct B;
struct A { B() noexcept(false); };
However, we then go on to finish late parsing the declaration and this
expects that what we've parsed is a CXXMethodDecl. A
CXXDeductionGuideDecl is not a CXXMethodDecl (it's a FunctionDecl), and
so we assert on the cast.
This fixes the crash by switching from cast<> to dyn_cast<> and not
setting up a "this" scope when the declaration is not a CXXMethodDecl.
This fixes PR49735.
I've found my recent ventures into the swig land painful because
of the strange way they are formatted. This patch attempts to alleviate
future headaches by formatting these files into something resembling the
normal llvm style.
Unfortunately, completely formatting these files automatically does not
work because clang format gets confused by swigs % syntax, so I have
employed a hybrid approach where I formatted blocks of c++ code with
clang-format and then manually massaged the code until it looked
reasonable (and compiled).
I don't expect these files to remain perfectly formatted (although, if
one's editor is configured to configure the current line/block on
request, one can get pretty good results by using it judiciously), but
at least it will prevent the (mangled form of the) old lldb style being
proliferated endlessly.
Differential Revision: https://reviews.llvm.org/D115736
This mostly is the same code that is refactored to decouple the cost and
benefit analysis. The biggest change is top-level function specializeFunctions
that now drives the transformation more like this:
specializeFunctions() {
Cost = getSpecializationCost(F);
calculateGains(F, Cost);
specializeFunction(F);
}
while this is just a restructuring, it helps the functional change in
calculateGains. I.e., we now sort the candidates based on the expected
specialisation gain, which we didn't do before. For this, a book keeping struct
ArgInfo was introduced. If we have a list of N candidates, but we only want
specialise less than N as set by option -func-specialization-max-constants, we
sort the list and discard the candidates that give the least benefit.
Given a formal argument, this change results in selecting the best actual
argument(s). This is NFC'ish in that this shouldn't change the current output
(hence no test change here), but in follow ups starting with D115509, it
should and I want to go one step further and compare all functions and all
arguments, which will mostly build on top of this refactoring and change.
Differential Revision: https://reviews.llvm.org/D115458
This patch extends the GPU kernel outlining pass so that it can take in
an optional data layout specification that will be attached to the GPU
module operation generated. If the data layout specification is not provided
the default data layout is used instead.
Reviewed By: herhut, mehdi_amini
Differential Revision: https://reviews.llvm.org/D115722
This patch updates applyLoopGuards to first collect all conditions and
then applies them in reverse order. This ensures the SCEVs with the
shortest dependency chains are constructed first, limiting the required
stack size.
This fixes a crash reported in D113578.
Note that the order conditions are applied can impact the accuracy of
the result, mostly due to missing min/max simplifications when
constructing SCEVs.
The changed test highlights the impact of the evaluation order. I will
follow up with a SCEV patch to improve min/max simplifications to get
the same results for both orders.
When hitting an else clause the type Stack should be reset to as it was at the start of the if, without taking into account the Type inserted into the Stack during the then branch of the if.
Reviewed By: aardappel
Differential Revision: https://reviews.llvm.org/D115748
D112556 added verification that the live interval for a subreg operand
must have subranges. This patch fixes a corner case, where if all subreg
operands for a particular register are undef uses then no subranges
are required. This matches how LiveIntervalCalc would build the live
intervals in the first place, since an undef use is not considered
to read the register.
Before this patch, CodeGen/AMDGPU/no-remat-indirect-mov.mir would fail
with -early-live-intervals:
# After Live Interval Analysis
...
*** Bad machine code: Live interval for subreg operand has no subranges ***
- function: index_vgpr_waterfall_loop
- basic block: %bb.1 (0x6a9a968) [352B;496B)
- instruction: 432B %24:vgpr_32 = V_MOV_B32_e32 undef %18.sub0:vreg_512, implicit $exec, implicit %18:vreg_512, implicit $m0
- operand 1: undef %18.sub0:vreg_512
Differential Revision: https://reviews.llvm.org/D115360
Having a default value for the lowering strategy of the multi-reduction op has proven
to be unexpected by users. This patch is dropping the default value so that users have
to explicitly choose the lowering strategy to be applied.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D115805
After the switch to the new pass manager, we have observed multiple
instances of catastrophic inlining, where the inliner produces huge
functions with many hundreds of thousands of instructions from small
input IR. We were forced to back out the switch to the new pass
manager for this reason. This patch fixes at least one of the root
cause issues.
LLVM uses a bottom-up inliner, and the fact that functions are processed
bottom-up is not just a question of optimality -- it is an imporant
requirement to prevent runaway inlining. The premise of the current
inlining approach and cost model is that after all calls inside a function
have been inlined, it may get large enough that inlining it into its
callers is no longer considered profitable. This safeguard does not
exist if inlining doesn't happen bottom-up, as inlining the callees,
and their callees, and their callees etc. will always seem individually
profitable, and the inliner can easily flatten the whole call tree.
There are instances where we necessarily have to deviate from bottom-up
inlining: When inlining in an SCC there is no natural "bottom", so
inlining effectively happens top-down. This requires special care,
and the inliner avoids exponential blowup by ensuring that functions
in the SCC grow in a balanced way and will eventually hit the threshold.
However, there is one instance where the inlining advisor explicitly
violates the bottom-up principle: Deferred inlining tries to "defer"
inlining a call if it determines that inlining the caller into all
its call-sites would be more profitable. Something very important to
understand about deferred inlining is that it doesn't make one inlining
choice in place of another -- it effectively chooses to do both. If we
have a call chain A -> B -> C and cost modelling tells us that inlining
B -> C is profitable, but we defer this and instead inline A -> B first,
then we'll now have a call A -> C, and the cost model will (a few special
cases notwithstanding) still tell us that this is profitable. So the end
result is that we inlined *both* B and C, even though under the usual
cost model function B would have been too large to further inline after
C has been integrated into it.
Because deferred inlining violates the bottom-up invariant of the inliner,
it can result in exponential inlining. The exponential-deferred-inlining.ll
test case illustrates this on a simple example (see
https://gist.github.com/nikic/1262b5f7d27278e1b34a190ae10947f5 for a
much more catastrophic case with about 5000x size blowup). If the call
chain A -> B -> C is not a chain but a tree of calls, then we end up
deferring inlining across the tree and end up flattening everything into
the root node.
This patch proposes to address this by disabling deferred inlining
entirely (currently still behind an option). Beyond the issue of
exponential inlining, I don't think that the whole concept makes sense,
at least as long as deferred inlining still ends up inlining both call
edges.
I believe the motivation for having deferred inlining in the first place
is that you might have a small wrapper function with local linkage that
could be eliminated if inlined. This would automatically happen if there
was a single caller, due to the large "last call to local" bonus. However,
this bonus is not extended if there are multiple callers, even if we
would eventually end up inlining into all of them (if the bonus were
extended).
Now, unlike the normal inlining cost model, the deferred inlining cost
model does look at all callers, and will extend the "last call to local"
bonus if it determines that we could inline all of them as long as we
defer the current inlining decision. This makes very little sense.
The "last call to local" bonus doesn't really cost model anything.
It's basically an "infinite" bonus that ensures we always inline the
last call to a local. The fact that it's not literally infinite just
prevents inlining of huge functions, which can easily result in
scalability issues. I very much doubt that it was an intentional
cost-modelling choice to say that getting rid of a small local function
is worth adding 15000 instructions elsewhere, yet this is exactly how
this value is getting used here.
The main alternative I see to complete removal is to change deferred
inlining to an actual either/or decision. That is, to mark deferred
calls as noinline so we're actually trading off one inlining decision
against another, and not just adding a side-channel to the cost model
to do both.
Apart from fixing the catastrophic inlining case, the effect on rustc
is a modest compile-time improvement on average (up to 8% for a
parsing-type crate, where tree-like calls are expected) and pretty
neutral where run-time performance is concerned (mix of small wins
and losses, usually in the sub-1% category).
Differential Revision: https://reviews.llvm.org/D115497
The minimizing and caching filesystem used by the dependency scanner keeps minimized and original files in separate caches.
This setup is not well suited for dealing with files that are sometimes minimized and sometimes not. Such files are being stat-ed and read twice, which is wasteful and also means the two versions of the file can get "out of sync".
This patch squashes the two caches together. When a file is stat-ed or read, its original contents are populated. If a file needs to be minimized, we give the minimizer the already loaded contents instead of reading the file again.
Reviewed By: dexonsmith
Differential Revision: https://reviews.llvm.org/D115346
This adds a GetObjectFileInterface callback member to
StaticLibraryDefinitionGenerator, and adds an optional argument for initializing
that member to StaticLibraryDefinitionGenerator's named constructors. If not
supplied, it will default to getObjectFileInterface from ObjectFileInterface.h.
To enable testing a `-hidden-l<x>` option is added to the llvm-jitlink tool.
This allows archives to be loaded with all contained symbol visibilities demoted
to hidden.
The ObjectLinkingLayer::setOverrideObjectFlagsWithResponsibilityFlags method is
(belatedly) hooked up, and enabled in llvm-jitlink when `-hidden-l<x>` is used
so that the demotion is also applied at symbol resolution time (avoiding any
"mismatched symbol flags" crashes).
After this function call, the LLVM IR would look like the following:
```
if (true)
/* NonVersionedLoop */
else
/* VersionedLoop */
```
Reviewed By: Whitney
Differential Revision: https://reviews.llvm.org/D104631
Weirdly, the opaque pointer compatible variants LLVMConstGEP2 and
LLVMConstInBoundsGEP2 were already declared in the header, but not
actually implemented. This adds the missing implementations and
deprecates the incompatible functions.
Differential Revision: https://reviews.llvm.org/D115787
Store the pointer element type inside LValue so that we can
preserve it when converting it back into an Address. Storing the
pointer element type might not be strictly required here in that
we could probably re-derive it from the QualType (which would
require CGF access though), but storing it seems like the simpler
solution.
The global register case is special and does not store an element
type, as the value is not a pointer type in that case and it's not
possible to create an Address from it.
This is the main remaining part from D103465.
Differential Revision: https://reviews.llvm.org/D115791
When have Zbs extension, we could use bexti to fold (and (not (srl X, C)), 1) to (xor (bexti X, C), 1).
Differential Revision: https://reviews.llvm.org/D115629
Everyone uses -l -L instead of the long option counterparts.
Make help messages attach to -L -l and (--reproduce) use them for response.txt
command line options.
Valentin Clement