For now at least, the plan is to keep hlfir.expr usage limited as
sub-expression operand, assignment rhs, and a few other contexts (
e.g. Associate statements). The rest of lowering (statements lowering
in the bridge) will still expect to get and manipulate characters and
arrays in memory. That means that hlfir.expr must be converted to
variable in converter.genExprAddr/converter.genExprBox.
This is done using an hlfir.associate, and generating the related
hlfir.end_associate in the statement context.
hlfir::getFirBase of is updated to avoid bringing in the HLFIR
fir.boxchar/fir.box into FIR when the entity was created with
hlfir::AssociateOp.
Differential Revision: https://reviews.llvm.org/D139328
This only contains the SelectionDAG implementation. GlobalISel to
follow.
The broad approach is:
- Introduce new builtins for 128-bit wide instructions.
- Lower these to @llvm.read_register.i128/@llvm.write_register.i128
- Introduce target-specific ISD nodes which have legal operands (two
i64s rather than an i128). These are named AArch64::{MRRS, MSRR} to
match the instructions they are for. These are a little complex as
they need to match the "shape" of what they're replacing or the
legaliser complains.
- Select these using the existing tryReadRegister/tryWriteRegister to
share the MDString parsing code, and introduce additional code to
ensure these are selected into the right MRRS/MSRR instructions. What
makes this hard is ensuring that the two i64s end up in an XSeqPair
register pair, because SelectionDAG doesn't care that much about
register classes if it can avoid doing so.
The main change to existing code is the reorganisation of
tryReadRegister and tryWriteRegister to try to keep the string parsing
code separate from the instruction creating code.
This also includes the changes to clang to define and use the ACLE
feature macro named `__ARM_FEATURE_SYSREG128`.
Contributors:
Sam Elliott
Lucas Prates
Differential Revision: https://reviews.llvm.org/D139086
This removes our "temporary" hack to assume that readnone/readonly
intrinsics are also willreturn. An explicit willreturn annotation,
usually via default intrinsic attributes, is now required.
Differential Revision: https://reviews.llvm.org/D137630
The test added in https://reviews.llvm.org/D137368 has been failing
on our 32 bit arm bots:
https://lab.llvm.org/buildbot/#/builders/178/builds/3460
You get this for the strings:
<<dead>> 0x883255000000003 [ 10] literal string: Hello, it's me
Instead of the expected:
<<dead>> 0x0000000F [ 3] literal string: Hello, it's me
This is because unlike symbols whose size is a uint64_t, strings
use a StringRef whose size is size_t. size_t changes size between
32 and 64 bit platforms.
This fixes the test by using %z to print the size of the strings,
this works for 32 and 64 bit.
The main motivation for this change is to avoid ambiguity because
mapping symbol names may not be unique across a binary and do not allow uniquely
identifying target address. So that mapping symbols used as branch target
labels make llvm-objdump output less readable.
Another point is that mapping symbols sometimes appear in
non-allocatable sections, like debug info sections which make objdump
output even more confusing.
For example, a small AArch64 executable may contain plenty of `$d[.*]`
symbols and none of them would be useful as a label for resolving
a branch or a memory operand target address:
```
0000000000000254 l .note.ABI-tag 0000000000000000 $d
00000000000008d4 l .eh_frame 0000000000000000 $d
0000000000000868 l .rodata 0000000000000000 $d
0000000000011028 l .data 0000000000000000 $d
0000000000010db8 l .fini_array 0000000000000000 $d
0000000000010db0 l .init_array 0000000000000000 $d
00000000000008e8 l .eh_frame 0000000000000000 $d
0000000000011034 l .bss 0000000000000000 $d
```
Note that GNU objdump doesn't use mapping symbols as branch target
labels for all targets that support such symbols (ARM, AArch64, CSKY).
Differential Revision: https://reviews.llvm.org/D139131
The only way to do this with the current hoisting strategy is by
lowering Affine to Scf first, but that prevents further passes on
Affine.
Differential Revision: https://reviews.llvm.org/D137600
This is a follow-up from discussion in D138412. Sometimes we want to evaluate
the cost of expansion of several SCEVs together with same budget. For example,
if one of them is a bit above cheap limit, and the second one is free, then
we still want to expand. Checking each of them with "cheap" limit is a bit more
pessimistic.
Differential Revision: https://reviews.llvm.org/D138475
Reviewed By: lebedev.ri
This should allow downstream vendors to install multiple LLVM distributions in parallel.
Should we also patch the default values to allow multiple upstream llvm distribution?
Reviewed By: thieta
Differential Revision: https://reviews.llvm.org/D138632
Cannonical frame address after RVV stack adjustment is sp + StackSize +
RVVStackSize * vlenb, and since vlenb is unknown at compile-time (but it
is a constant for particular HW implementation), emit
.cfi_def_cfa_expression so libunwind can read VLENB CSR register at
run-time and obtain correct frame address.
Fixes https://github.com/llvm/llvm-project/issues/58356 (but additional
run-time support for reading CSR may be required)
Differential Revision: https://reviews.llvm.org/D136263
P1787: //CWG600 is resolved by explaining that accessibility affects naming a member in the sense of the ODR.//
Wording: see changes to [class.access] p1 and p4.
Additional references: [[ http://eel.is/c++draft/basic.def.odr#8.sentence-2 | basic.def.odr/8 ]]: //A function is odr-used if it is named by a potentially-evaluated expression or conversion.//
Reviewed By: #clang-language-wg, aaron.ballman
Differential Revision: https://reviews.llvm.org/D139173
P1787: //CWG554 is resolved by using the word “scope” instead of “declarative region”, consistent with its very common use in phrases like “namespace scope”.//
Reviewed By: #clang-language-wg, cor3ntin, aaron.ballman, shafik
Differential Revision: https://reviews.llvm.org/D139172
Unlike projects, runtimes doesn't have a default set of names.
This means you get a leading space at the start, which gets converted
to a ';' giving ";<runtime name>;<runtime name>".
CMake then errors because the "" before the first ';' is treated
as a runtime name and of course it's not a valid name.
Fix this by removing the leading spaces from runtimes before we
insert the ';'.
Reviewed By: ldionne
Differential Revision: https://reviews.llvm.org/D139306
P1787: //CWG405 is resolved by stating that argument-dependent lookup (sometimes) occurs after an ordinary unqualified lookup (making statements like “finding a variable prevents argument-dependent lookup” formally correct).//
Wording: see changes to [basic.lookup.argdep] p1 and p3
This issue seems a duplicate of CWG218, even though it is not officially recognized. A part of a test for CWG218 is reused here, adding cross-references.
Reviewed By: #clang-language-wg, aaron.ballman
Differential Revision: https://reviews.llvm.org/D139095
CodeView doesn't have the ability to represent variables
in other ways than as in registers or memory values, but
LLVM very often transforms simple values into constants,
consider this program:
int f () { int i = 123; return i; }
LLVM will transform `i` into a constant value and just
leave behind a llvm.dbg.value, this can't be represented
as a S_LOCAL record in CodeView. But we can represent it
as a S_CONSTANT record.
This patch checks if the location of a debug value is null,
then we will insert a S_CONSTANT record instead of a S_LOCAL
value with the flag "OptimizedAway".
In lld we then output the S_CONSTANT in the right scope, before
they where always inserted in the global stream, now we check
the scope before inserting it.
This has shown to improve debugging for our developers
internally.
Fixes to llvm/llvm-project#55958
Reviewed By: aganea
Differential Revision: https://reviews.llvm.org/D138995
P1787: // [[ https://wg21.link/cwg952 | CWG952 ]] is resolved by refining the definition of “naming class” per Richard’s suggestion in [[ https://lists.isocpp.org/core/2020/09/9963.php | “CWG1621 and [class.static/2”]].//
Wording:
- [class.static]/2 removed;
- [class.access.base]/5 rephrased.
Currently behavior is the following: unqualified names undergo //unqualified name lookup// [1], which perform //unqualified search// in immediate scope [2]. This scope is the scope the definition of //naming class// [3] refers to. `A::I` is not //accessible// when named in classes `C` and `D` per [3]. In particular, the last item regarding base class ([class.access.base]/5.4) is not applicable, because class `A` is not //accessible// in both classes `C` and `D` per [4].
References:
1. [[ https://eel.is/c++draft/basic.lookup#unqual-4.sentence-2 | basic.lookup.unqual/4 ]]
2. [[ https://eel.is/c++draft/basic.lookup#unqual-3 | basic.lookup.unqual/3 ]]
3. [[ https://eel.is/c++draft/class.access#base-5.sentence-4 | class.access.base/5 ]]
4. [[ https://eel.is/c++draft/class.access#base-4 | class.access.base/4 ]]
Reviewed By: #clang-language-wg, erichkeane, aaron.ballman
Differential Revision: https://reviews.llvm.org/D139326
While we can't use a single BatchAA instance for the entire
MemCpyOpt run without further justification, we can use BatchAA
while performing the queries related to a single instruction
(these will first perform some AA-based checks, and then modify
the IR only afterwards).
The BracketAlignmentStyle BAS_BlockIndent was forcing breaks before a
closing right parenthesis yielding strange-looking results in case of
code structures that have a left parens immediately following a right
parens ")(" such as is seen with indirect function calls via function
pointers and with type casting.
Fixes 57250.
Fixes 58496.
Differential Revision: https://reviews.llvm.org/D137762
Support reading of VLENB (vector byte length) control register, that can be
required for correct unwinding of RVV objects on stack.
Differential Revision: https://reviews.llvm.org/D136264
This is not NFC because the DSE BatchAA is more powerful than the
default one due to EarliestEscape CaptureInfo, so this might
improve results in some cases.
When generating character assignment operations, the generic code
generates some code to handle truncation and padding when the length
differ at runtime. A bypass already exists when the length are compile
time constant and match, but it was not used for the trivial case where
the RHS and LHS length is the same SSA value. In such case, even though,
the length is not know at compile time, it is known to be the same.
This will simplify the code creating character temporaries from a
variable in HLFIR that will use this assignment code.
Note that this probably has little impact on performance (llvm may be clever enough
to later catch that for us). But it makes the generated IR a lot more readable at
little cost.
Differential Revision: https://reviews.llvm.org/D139330
This patch adds `replaceAllUsesExcept` to the rewriter class.
The implementation is copy-pasted from Value + calling
`updateRootInPlace` to notify the listeners about the
corresponding IR changes.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D139382
While MemorySSA use optimization was already using BatchAA, the
publicly exposed MSSA walkers were using plain AAResults. This is
not great, because it is expected that clobber walking will make
repeated AA queries.
This patch makes the clobber API accept a BatchAAResults instance.
The plain APIs are kept as wrappers and will create a BatchAAResults
instance for the duration of the query. In the future, the explicit
BatchAAResults arguments will be used to share AA results across
queries, not just within one query.
Differential Revision: https://reviews.llvm.org/D136164
It seems unnecessarily limiting to disallow recursive partial
inlining, and there are clearly cases where it can benefit
code by avoiding a function call and potentially enabling
other transformations like dead argument elimination
in cases where an argument is only used prior to the early-out
test at the top of the function.
The pass already properly rewrites the recursive calls
within the body of the freshly cloned function, so the only
change here is removing the bail-out when recursion is
detected.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D136383
Msan needs noundef consistency between interface and implementation. If
we call C++ from C we can have noundef on C++ side, and no noundef on
caller C side, noundef implementation will not set TLS for return value,
no noundef caller will expect it. Then we have false reports in msan.
The workaround could be set TLS to zero even for noundef return values.
However if we do that always it will increase binary size by about 10%.
If we do that selectively we need to handle "address is taken"
functions, any non local functions, and probably all function which have
musttail callers. Which is still a lot.
The existing implementation of HasStrictReturn refers to C standard as
the reason not enforcing noundef. I believe it applies only to the case
when return statement is omitted. Testing on Google codebase I never see
such cases, however I've see tens of cases where C code returns actual
uninitialized variables, but we ignore that it because of "omitted
return" case.
So this patch will:
1. fix false-positives with TLS missmatch.
2. detect bugs returning uninitialized variables for C as well.
3. report "omitted return" cases stricter than C, which is already a
warning and very likely a bug in a code anyway.
Reviewed By: kda
Differential Revision: https://reviews.llvm.org/D139296
