This fixes loads like 's1 = load %p (load 1 from %p)' being combined with an
extend into an illegal 's8 = g_extload %p (load 1 from %p)' which doesn't do any
extension, by avoiding touching those < s8 size loads.
This bug was uncovered by a verifier update r351584, which I reverted it to keep
the bots green.
llvm-svn: 352311
The add+and sequence followed by a branch can
happen e.g. when looping over the set bits of an integer:
```
while (x != 0) {
func(x & ~x);
x &= x - 1;
}
```
Reviewed By: ctopper
Differential Revision: https://reviews.llvm.org/D57296
llvm-svn: 352306
def32 here means the producing instruction zeroed bits 63:32. We already do this for zext, but it looks like we can get an and+anyext sometimes.
Spotted in the diffs from D33587.
llvm-svn: 352303
Summary:
lld discards .gnu.linonce.* sections work around a bug in glibc.
https://sourceware.org/bugzilla/show_bug.cgi?id=20543
Unfortunately, the Linux kernel uses a section named
.gnu.linkonce.this_module to store infomation about kernel modules. The
kernel reads data from this section when loading kernel modules, and
errors if it fails to find this section. The current behavior of lld
discards this section when kernel modules are linked, so kernel modules
linked with lld are unloadable by the linux kernel.
The Linux kernel should use a comdat section instead of .gnu.linkonce.
The minimum version of binutils supported by the kernel supports comdat
sections. The kernel is also not relying on the old linkonce behavior;
it seems to have chosen a name that contains a deprecated GNU feature.
Changing the section name now in the kernel would require all kernel
modules to be recompiled to make use of the new section name. Instead,
rather than discarding .gnu.linkonce.*, let's discard the more specific
section name to continue working around the glibc issue while supporting
linking Linux kernel modules.
Link: https://github.com/ClangBuiltLinux/linux/issues/329
Reviewers: pcc, espindola
Reviewed By: pcc
Subscribers: nathanchance, emaste, arichardson, void, srhines
Differential Revision: https://reviews.llvm.org/D57294
llvm-svn: 352302
Bitcast and certain Ptr2Int/Int2Ptr instructions will not alter the
value of their operand and can therefore be looked through when we
determine non-nullness.
Differential Revision: https://reviews.llvm.org/D54956
llvm-svn: 352293
This patch effectively fixes the almost decade old checker naming issue.
The solution is to assert when CheckerManager::getChecker is called on an
unregistered checker, and assert when CheckerManager::registerChecker is called
on a checker that is already registered.
Differential Revision: https://reviews.llvm.org/D55429
llvm-svn: 352292
Fix issue noted in D57281 that only tested the one use for the SDValue (the result flag), not the entire SUB.
I've added the getNode() to make it clearer what is intended than just the -> redirection.
llvm-svn: 352291
As discussed on PR24545, we should try to commute X86::COND_A 'icmp ugt' cases to X86::COND_B 'icmp ult' to more optimally bind the carry flag output to a SBB instruction.
Differential Revision: https://reviews.llvm.org/D57281
llvm-svn: 352289
We often generate X86ISD::SBB(X, 0) for carry flag arithmetic.
I had tried to create test cases for the ADC equivalent (which often uses the same pattern) but haven't managed to find anything yet.
Differential Revision: https://reviews.llvm.org/D57169
llvm-svn: 352288
Unfortunately, up until now, the fact that certain checkers depended on one
another was known, but how these actually unfolded was hidden deep within the
implementation. For example, many checkers (like RetainCount, Malloc or CString)
modelled a certain functionality, and exposed certain reportable bug types to
the user. For example, while MallocChecker models many many different types of
memory handling, the actual "unix.MallocChecker" checker the user was exposed to
was merely and option to this modeling part.
Other than this being an ugly mess, this issue made resolving the checker naming
issue almost impossible. (The checker naming issue being that if a checker
registered more than one checker within its registry function, both checker
object recieved the same name) Also, if the user explicitly disabled a checker
that was a dependency of another that _was_ explicitly enabled, it implicitly,
without "telling" the user, reenabled it.
Clearly, changing this to a well structured, declarative form, where the
handling of dependencies are done on a higher level is very much preferred.
This patch, among the detailed things later, makes checkers declare their
dependencies within the TableGen file Checkers.td, and exposes the same
functionality to plugins and statically linked non-generated checkers through
CheckerRegistry::addDependency. CheckerRegistry now resolves these dependencies,
makes sure that checkers are added to CheckerManager in the correct order,
and makes sure that if a dependency is disabled, so will be every checker that
depends on it.
In detail:
* Add a new field to the Checker class in CheckerBase.td called Dependencies,
which is a list of Checkers.
* Move unix checkers before cplusplus, as there is no forward declaration in
tblgen :/
* Add the following new checkers:
- StackAddrEscapeBase
- StackAddrEscapeBase
- CStringModeling
- DynamicMemoryModeling (base of the MallocChecker family)
- IteratorModeling (base of the IteratorChecker family)
- ValistBase
- SecuritySyntaxChecker (base of bcmp, bcopy, etc...)
- NSOrCFErrorDerefChecker (base of NSErrorChecker and CFErrorChecker)
- IvarInvalidationModeling (base of IvarInvalidation checker family)
- RetainCountBase (base of RetainCount and OSObjectRetainCount)
* Clear up and registry functions in MallocChecker, happily remove old FIXMEs.
* Add a new addDependency function to CheckerRegistry.
* Neatly format RUN lines in files I looked at while debugging.
Big thanks to Artem Degrachev for all the guidance through this project!
Differential Revision: https://reviews.llvm.org/D54438
llvm-svn: 352287
My last patch, D56989, moved the validation of whether a checker exists into
its constructor, but we do support statically linked (and non-plugin) checkers
that were do not have an entry in Checkers.td. However, the handling of this
happens after the creation of the CheckerRegistry object.
This patch fixes this bug by moving even this functionality into
CheckerRegistry's constructor.
llvm-svn: 352284
I added a new enum to CheckerInfo, so we can easily track whether the check is
explicitly enabled, explicitly disabled, or isn't specified in this regard.
Checkers belonging in the latter category may be implicitly enabled through
dependencies in the followup patch. I also made sure that this is done within
CheckerRegisty's constructor, leading to very significant simplifications in
its query-like methods.
Differential Revision: https://reviews.llvm.org/D56989
llvm-svn: 352282
This reduces a bit of duplication between the combining and
lowering places that use it, but the primary motivation is
to make it easier to rearrange the lowering logic and solve
PR40434:
https://bugs.llvm.org/show_bug.cgi?id=40434
llvm-svn: 352280
Since pretty much all methods of CheckerRegistry has AnalyzerOptions as an
argument, it makes sense to just simply require it in it's constructor.
Differential Revision: https://reviews.llvm.org/D56988
llvm-svn: 352279
The actual implementation of unix.API features a dual-checker: two checkers in
one, even though they don't even interact at all. Split them up, as this is a
problem for establishing dependencies.
I added no new code at all, just merely moved it around.
Since the plist files change (and that's a benefit!) this patch isn't NFC.
Differential Revision: https://reviews.llvm.org/D55425
llvm-svn: 352278
Introduce the boolean ento::shouldRegister##CHECKERNAME(const LangOptions &LO)
function very similarly to ento::register##CHECKERNAME. This will force every
checker to implement this function, but maybe it isn't that bad: I saw a lot of
ObjC or C++ specific checkers that should probably not register themselves based
on some LangOptions (mine too), but they do anyways.
A big benefit of this is that all registry functions now register their checker,
once it is called, registration is guaranteed.
This patch is a part of a greater effort to reinvent checker registration, more
info here: D54438#1315953
Differential Revision: https://reviews.llvm.org/D55424
llvm-svn: 352277
Store the controlling expression, the association expressions and the
corresponding TypeSourceInfos as trailing objects.
Additionally use the bit-fields of Stmt to store one SourceLocation,
saving one additional pointer. This saves 3 pointers in total per
GenericSelectionExpr.
Differential Revision: https://reviews.llvm.org/D57104
Reviewed By: aaron.ballman
Reviewers: aaron.ballman, steveire
llvm-svn: 352276
Various cleanups to GenericSelectionExpr factored out of D57104. In particular:
1. Move the friend declaration to the top.
2. Introduce a constant ResultDependentIndex instead of the magic "-1".
3. clang-format
4. Group the member function together so that they can be removed as one block
by D57106.
NFC.
Differential Revision: https://reviews.llvm.org/D57238
Reviewed By: aaron.ballman
llvm-svn: 352275
Summary:
The 512-bit cvt(u)qq2tops, cvt(u)qqtopd, and cvt(u)dqtops intrinsics all have the possibility of taking an explicit rounding mode argument. If the rounding mode is CUR_DIRECTION we'd like to emit a sitofp/uitofp instruction and a select like we do for 256-bit intrinsics.
For cvt(u)qqtopd and cvt(u)dqtops we do this when the form of the software intrinsics that doesn't take a rounding mode argument is used. This is done by using convertvector in the header with the select builtin. But if the explicit rounding mode form of the intrinsic is used and CUR_DIRECTION is passed, we don't do this. We shouldn't have this inconsistency.
For cvt(u)qqtops nothing is done because we can't use the select builtin in the header without avx512vl. So we need to use custom codegen for this.
Even when the rounding mode isn't CUR_DIRECTION we should also use select in IR for consistency. And it will remove another scalar integer mask from our intrinsics.
To accomplish all of these goals I've taken a slightly unusual approach. I've added two new X86 specific intrinsics for sitofp/uitofp with rounding. These intrinsics are variadic on the input and output type so we only need 2 instead of 6. This avoids the need for a switch to map them in CGBuiltin.cpp. We just need to check signed vs unsigned. I believe other targets also use variadic intrinsics like this.
So if the rounding mode is CUR_DIRECTION we'll use an sitofp/uitofp instruction. Otherwise we'll use one of the new intrinsics. After that we'll emit a select instruction if needed.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D56998
llvm-svn: 352267
The IR enforced limit for the address space is 24-bits, but LLT was
only using 23-bits. Additionally, the argument to the constructor was
truncating to 16-bits.
A similar problem still exists for the number of vector elements. The
IR enforces no limit, so if you try to use a vector with > 65535
elements the IRTranslator asserts in the LLT constructor.
llvm-svn: 352264
