As we traverse the CFG backwards, we could end up reaching unreachable
blocks. For unreachable blocks, we won't have computed post order
numbers and because DomAccess is reachable, unreachable blocks cannot be
on any path from it.
This fixes a crash with unreachable blocks.
Add -fpch-instantiate-templates which makes template instantiations be
performed already in the PCH instead of it being done in every single
file that uses the PCH (but every single file will still do it as well
in order to handle its own instantiations). I can see 20-30% build
time saved with the few tests I've tried.
The change may reorder compiler output and also generated code, but
should be generally safe and produce functionally identical code.
There are some rare cases that do not compile with it,
such as test/PCH/pch-instantiate-templates-forward-decl.cpp. If
template instantiation bailed out instead of reporting the error,
these instantiations could even be postponed, which would make them
work.
Enable this by default for clang-cl. MSVC creates PCHs by compiling
them using an empty .cpp file, which means templates are instantiated
while building the PCH and so the .h needs to be self-contained,
making test/PCH/pch-instantiate-templates-forward-decl.cpp to fail
with MSVC anyway. So the option being enabled for clang-cl matches this.
Differential Revision: https://reviews.llvm.org/D69585
If a collection of interconnected phi nodes is only ever loaded, stored
or bitcast then we can convert the whole set to the bitcast type,
potentially helping to reduce the number of register moves needed as the
phi's are passed across basic block boundaries. This has to be done in
CodegenPrepare as it naturally straddles basic blocks.
The alorithm just looks from phi nodes, looking at uses and operands for
a collection of nodes that all together are bitcast between float and
integer types. We record visited phi nodes to not have to process them
more than once. The whole subgraph is then replaced with a new type.
Loads and Stores are bitcast to the correct type, which should then be
folded into the load/store, changing it's type.
This comes up in the biquad testcase due to the way MVE needs to keep
values in integer registers. I have also seen it come up from aarch64
partner example code, where a complicated set of sroa/inlining produced
integer phis, where float would have been a better choice.
I also added undef and extract element handling which increased the
potency in some cases.
This adds it with an option that defaults to off, and disabled for 32bit
X86 due to potential issues around canonicalizing NaNs.
Differential Revision: https://reviews.llvm.org/D81827
GetUnderlyingObject() (and by required symmetry
DecomposeGEPExpression()) will call SimplifyInstruction() on the
passed value if other checks fail. This simplification is very
expensive, but has little effect in practice. This patch removes
the SimplifyInstruction call(), and replaces it with a check for
single-argument phis (which can occur in canonical IR in LCSSA
form), which is the only useful simplification case I was able to
identify.
At O3 the geomean CTMark improvement is -1.7%. The largest
improvement is SPASS with ThinLTO at -6%.
In test-suite, I see only two tests with a hash difference and
no code size difference (PAQ8p, Ptrdist), which indicates that
the simplification only ends up being useful very rarely. (I would
have liked to figure out which simplification is responsible here,
but wasn't able to spot it looking at transformation logs.)
The AMDGPU test case that is update was using two selects with
undef condition, in which case GetUnderlyingObject will return
the first select operand as the underlying object. This will of
course not happen with non-undef conditions, so this was not
testing anything realistic. Additionally this illustrates potential
unsoundness: While GetUnderlyingObject will pick the first operand,
the select might be later replaced by the second operand, resulting
in inconsistent assumptions about the undef value.
Differential Revision: https://reviews.llvm.org/D82261
The outputs between the direct ast-dump test and the ast-dump test after
deserialization should match modulo a few differences.
For hand-written tests, strip the "<undeserialized declarations>"s and
the "imported"s with sed.
For tests generated with "make-ast-dump-check.sh", regenerate the output.
Part 3/n.
The outputs between the direct ast-dump test and the ast-dump test after
deserialization should match modulo a few differences.
For hand-written tests, strip the "<undeserialized declarations>"s and
the "imported"s with sed.
For tests generated with "make-ast-dump-check.sh", regenerate the
output.
Part 2/n.
The serialization of ConstantExpr has currently a number of problems:
- Some fields are just not serialized (ConstantExprBits.APValueKind and
ConstantExprBits.IsImmediateInvocation).
- ASTStmtReader::VisitConstantExpr forgets to add the trailing APValue
to the list of objects to be destroyed when the APValue needs cleanup.
While we are at it, bring the serialization of ConstantExpr more in-line
with what is done with the other expressions by doing the following NFCs:
- Get rid of ConstantExpr::DefaultInit. It is better to not initialize
the fields of an empty ConstantExpr since this will allow msan to
detect if a field was not deserialized.
- Move the initialization of the fields of ConstantExpr to the constructor;
ConstantExpr::Create allocates the memory and ConstantExpr::ConstantExpr
is responsible for the initialization.
Review after commit since this is a straightforward mechanical fix
similar to the other serialization fixes.
Pulled out from the ongoing work on D66004, currently we don't do a good job of simplifying variable shuffle masks that have already lowered to constant pool entries.
This patch adds SimplifyDemandedVectorEltsForTargetShuffle (a custom x86 helper) to first try SimplifyDemandedVectorElts (which we already do) and then constant pool simplification to help mark undefined elements.
To prevent lowering/combines infinite loops, we only handle basic constant pool loads instead of creating new BUILD_VECTOR nodes for lowering - e.g. we don't try to convert them to broadcast/vzext_load - there might be some benefit to this but if so I'd rather we come up with some way to reuse existing code than reimplement a lot of BUILD_VECTOR code.
Differential Revision: https://reviews.llvm.org/D81791
Summary: The patch D81022 seems to break the indentation of the `cleanupIR()` function. This patch fixes this problem
Reviewers: jdoerfert, sstefan1, uenoku
Reviewed By: jdoerfert
Subscribers: hiraditya, uenoku, kuter, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82260
Summary:
Add call site location info into inline remarks so we can differentiate inline sites.
This can be useful for inliner tuning. We can also reconstruct full hierarchical inline
tree from parsing such remarks. The messege of inline remark is also tweaked so we can
differentiate SampleProfileLoader inline from CGSCC inline.
Reviewers: wmi, davidxl, hoy
Subscribers: hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D82213
This patch implements builtins for the following prototypes:
```
vector signed char vec_clrl (vector signed char a, unsigned int n);
vector unsigned char vec_clrl (vector unsigned char a, unsigned int n);
vector signed char vec_clrr (vector signed char a, unsigned int n);
vector signed char vec_clrr (vector unsigned char a, unsigned int n);
```
Differential Revision: https://reviews.llvm.org/D81707
We need to set the cpu_vendor to a non-zero value to indicate
that we already called __cpu_indicator_init once.
This should only happen on a 386 or 486 CPU.
The bridge uses internal boxes of related ssa-values to track all the
information associated with a Fortran variable. Variables may have a
location and a value, but may also carry other properties such as rank,
shape, LEN parameters, etc. in Fortran.
Differential revision: https://reviews.llvm.org/D82228
The previous tests apparently missed a few code branches in DumpDataExtractor
code. Also renames the 'test_instruction' which had the same name as another
test (and Python therefore ignored the test entirely).
Summary: Add the --hot-func-list feature to llvm-profdata show for sample profiles. This feature prints a list of hot functions whose max sample count are above the 99% threshold, with their numbers of total samples, total samples percentage, max samples, entry samples, and their function names.
Reviewers: wmi, hoyFB, wenlei
Reviewed By: wmi
Subscribers: hoyFB, wenlei, llvm-commits, weihe
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81800
This prevents us from creating temporary PoisoningVHs and
AssertingVHs while performing hashmap lookups. As such, it only
matters in assertion-enabled builds.
In C++17 the operand(s) of an overloaded operator are sequenced as for
the corresponding built-in operator when the overloaded operator is
called with the operator notation ([over.match.oper]p2).
Reported in PR35340.
Differential Revision: https://reviews.llvm.org/D81330
Reviewed By: rsmith
When building runtimes, the compiler name (e.g. clang, clang-cl) is set based on
`CMAKE_SYSTEM_NAME` passed to `llvm_ExternalProject_Add()` through `CMAKE_ARGS` argument.
This mechanism doesn't work well if the target is Windows host.
`runtime_default_target()`/`builtin_default_target()` doesn't provide a way
to specify `CMAKE_SYSTEM_NAME` and doesn't set it either.
This patch appends variables specified in `RUNTIMES_CMAKE_ARGS`/`BUILTINS_CMAKE_ARGS`
to `CMAKE_ARGS` argument of `llvm_ExternalProject_Add()` in the case of called
from `runtime_default_target()`/`builtin_default_target()` thus in particular
it allows passing CMAKE_SYSTEM_NAME whenever it is required.
Reviewed By: phosek, compnerd, plotfi
Differential Revision: https://reviews.llvm.org/D81877
Florian Hahn