This is a restricted version of the combine in `DAGCombiner::MatchLoadCombine`.
(See D27861)
This tries to recognize patterns like below (assuming a little-endian target):
```
s8* x = ...
s32 val = a[0] | (a[1] << 8) | (a[2] << 16) | (a[3] << 24)
->
s32 val = *((i32)a)
s8* x = ...
s32 val = a[3] | (a[2] << 8) | (a[1] << 16) | (a[0] << 24)
->
s32 val = BSWAP(*((s32)a))
```
(This patch also handles the big-endian target case as well, in which the first
example above has a BSWAP, and the second example above does not.)
To recognize the pattern, this searches from the last G_OR in the expression
tree.
E.g.
```
Reg Reg
\ /
OR_1 Reg
\ /
OR_2
\ Reg
.. /
Root
```
Each non-OR register in the tree is put in a list. Each register in the list is
then checked to see if it's an appropriate load + shift logic.
If every register is a load + potentially a shift, the combine checks if those
loads + shifts, when OR'd together, are equivalent to a wide load (possibly with
a BSWAP.)
To simplify things, this patch
(1) Only handles G_ZEXTLOADs (which appear to be the common case)
(2) Only works in a single MachineBasicBlock
(3) Only handles G_SHL as the bit twiddling to stick the small load into a
specific location
An IR example of this is here: https://godbolt.org/z/4sP9Pj (lifted from
test/CodeGen/AArch64/load-combine.ll)
At -Os on AArch64, this is a 0.5% code size improvement for CTMark/sqlite3,
and a 0.4% improvement for CTMark/7zip-benchmark.
Also fix a bug in `isPredecessor` which caused it to fail whenever `DefMI` was
the first instruction in the block.
Differential Revision: https://reviews.llvm.org/D94350
Original patch by @rogfer01.
This patch adds support for sign-, zero-, and any-extension from
scalable mask vector types to integer vector types, as well as
truncation in the opposite direction.
Authored-by: Roger Ferrer Ibanez <rofirrim@gmail.com>
Co-Authored-by: Fraser Cormack <fraser@codeplay.com>
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D94590
On z/OS, the error message "EDC5111I Permission denied." is not matched correctly in lit tests. This patch updates the check expression to match successfully.
Differential Revision: https://reviews.llvm.org/D94432
This fixes llvm.org/PR48554
Some test cases had to be updated because the hash function for
union_maps have been changed which affects the output order.
This is mostly SEO so that the new API can take over the old API when people
search for the different SB* classes. Sadly epydoc decided to throw in a -class
prefix behind all the class file names, so we can't just overwrite the old files
with the newly generated ones.
Reviewed By: JDevlieghere
Differential Revision: https://reviews.llvm.org/D94900
We have no lowering for VSELECT vXi1, vXi1, vXi1, so mark them as
expanded to turn them into a series of logical operations.
Differential Revision: https://reviews.llvm.org/D94946
Enums and constants are currently missing in the new LLDB Python API docs.
In theory we could just let them be autogenerated like the SB API classes, but sadly the generated documentation
suffers from a bunch of problems. Most of these problems come from the way SWIG is representing enums, which is
done by translating every single enum case into its own constant. This has a bunch of nasty effects:
* Because SWIG throws away the enum types, we can't actually reference the enum type itself in the API. Also because automodapi is impossible to script, this can't be fixed in post (at least without running like sed over the output files).
* The lack of enum types also causes that every enum *case* has its own full doc page. Having a full doc page that just shows a single enum case is pointless and it really slows down sphinx.
* There is no SWIG code for the enums, so there is also no place to write documentation strings for them. Also there is no support for copying the doxygen strings (which would be in the wrong format, but better than nothing) for enums (let alone our defines), so we can't really document all this code.
* Because the enum cases are just forwards to the native lldb module (which we mock), automodapi actually takes the `Mock` docstrings and adds it to every single enum case.
I don't see any way to solve this via automodapi or SWIG. The most reasonable way to solve this is IMHO to write a simple Clang tool
that just parses our enum/constant headers and emits an *.rst file that we check in. This way we can do all the LLDB-specific enum case and constant
grouping that we need to make a readable documentation page.
As we're without any real documentation until I get around to write that tool, I wrote a doc page for the enums/constants as a stop gap measure.
Most of this is done by just grepping our enum header and then manually cleaning up all the artifacts and copying the few doc strings we have.
Reviewed By: JDevlieghere
Differential Revision: https://reviews.llvm.org/D94959
isFixedFormSuffix and isFreeFormSuffix should be defined in
flangFrontend rather than flangFrontendTool library. That's for 2
reasons:
* these methods are used in flangFrontend rather than flangFrontendTool
* flangFrontendTool depends on flangFrontend
As mentioned in the post-commit review for D94228, without this change
shared library builds fail.
Differential Revision: https://reviews.llvm.org/D94968
* This allows us to hoist trait level information for regions and sized-variadic to class level attributes (_ODS_REGIONS, _ODS_OPERAND_SEGMENTS, _ODS_RESULT_SEGMENTS).
* Eliminates some splicey python generated code in favor of a native helper for it.
* Makes it possible to implement custom, variadic and region based builders with one line of python, without needing to manually code access to the segment attributes.
* Needs follow-on work for region based callbacks and support for SingleBlockImplicitTerminator.
* A follow-up will actually add ODS support for generating custom Python builders that delegate to this new method.
* Also includes the start of an e2e sample for constructing linalg ops where this limitation was discovered (working progressively through this example and cleaning up as I go).
Differential Revision: https://reviews.llvm.org/D94738
Split impliesPoison into two recursive walks, one over V, the
other over ValAssumedPoison. This allows us to reason about poison
implications in a number of additional cases that are important
in practice. This is a generalized form of D94859, which handles
the cmp to cmp implication in particular.
Differential Revision: https://reviews.llvm.org/D94866
This commit adds a new trait that can be attached to ops that have
signed semantics.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D94896
This patch factors out the "VLMax" operand passed to most
scalable-vector ISel patterns into a property of each VType.
This is seen as a preparatory change to allow RVV in the future to
more easily support fixed-length vector types with constrained vector
lengths, with the AVL operand set to the length of the fixed-length
vector. It has no effect on the scalable code generation path.
Reviewed By: HsiangKai
Differential Revision: https://reviews.llvm.org/D94594
This adds some basic MVE sadd_sat/ssub_sat/uadd_sat/usub_sat costs,
based on when the instruction is legal. With smaller than legal types
that are promoted we generate shr(qadd(shl, shl)), so the cost is 4
appropriately.
Differential Revision: https://reviews.llvm.org/D94958
The TableGen emitter for directives has two slots for flangClass information and this was mainly
to be able to keep up with the legacy openmp parser at the time. Now that all clauses are encapsulated in
AccClause or OmpClause, these two strings are not necessary anymore and were the the source of couple
of problem while working with the generic structure checker for OpenMP.
This patch remove the flangClassValue string from DirectiveBase.td and use the string flangClass as the
placeholder for the encapsulated class.
Reviewed By: sameeranjoshi
Differential Revision: https://reviews.llvm.org/D94821
When performing peephole optimization to simplify the code, after removing
passed FPSP/XSRSP instruction we will set any uses of that FRSP/XSRSP to the
source of the FRSP/XSRSP.
We are finding the machine instruction using virtual register holding FRSP/XSRSP
results by searching all following instructions and encountering an issue
that the first use of the virtual register is a debug MI causing:
1. virtual register in the debug MI removed unexpectedly.
2. virtual register used in non-debug MI not replaced with the source of
FRSP/XSRSP. which stays in a undef status.
This patch fix the issue by only searching non-debug machine instruction using
virtual register holding FRSP/XSRSP results when the vr only has one non debug
usage.
Differential Revisien: https://reviews.llvm.org/D94711
Reviewed by: nemanjai
cmake_minimum_required(VERSION) calls cmake_policy(VERSION),
which sets all policies up to VERSION to NEW.
LLVM started requiring CMake 3.13 last year, so we can remove
a bunch of code setting policies prior to 3.13 to NEW as it
no longer has any effect.
Reviewed By: phosek, #libunwind, #libc, #libc_abi, ldionne
Differential Revision: https://reviews.llvm.org/D94374
Previously we did not record local class declarations. Now with features like
findImplementation and typeHierarchy, we have a need to index such local
classes to accurately report subclasses and implementations of methods.
Performance testing results:
- No changes in indexing timing.
- No significant change in memory usage.
- **1%** increase in #relations.
- **0.17%** increase in #refs.
- **0.22%** increase #symbols.
**New index stats**
Time to index: **4:13 min**
memory usage **543MB**
number of symbols: **521.5K**
number of refs: **8679K**
number of relations: **49K**
**Base Index stats**
Time to index: **4:15 min**
memory usage **542MB**
number of symbols: **520K**
number of refs: **8664K**
number of relations: **48.5K**
Fixes: https://github.com/clangd/clangd/issues/644
Differential Revision: https://reviews.llvm.org/D94785
83daa49758 made loop-rotate more conservative in the presence of
function calls in the prepare-for-lto stage. The code did not properly
account for calls that are no actual function calls, like calls to
intrinsics. This patch updates the code to ensure only calls that are
lowered to actual calls are considered inline candidates.
Add Semantic checks for OpenMP 4.5 - 2.7.4 Workshare Construct.
- The structured block in a workshare construct may consist of only
scalar or array assignments, forall or where statements,
forall, where, atomic, critical or parallel constructs.
- All array assignments, scalar assignments, and masked array
assignments must be intrinsic assignments.
- The construct must not contain any user defined function calls unless
the function is ELEMENTAL.
Test cases : omp-workshare03.f90, omp-workshare04.f90, omp-workshare05.f90
Resolve test cases (omp-workshare01.f90 and omp-workshare02.f90) marked as XFAIL
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D93091
This CPU supports all v8.5a features except BTI, and so identifies as v8.5a to
Clang. A bit weird, but the best way for things like xnu to detect the new
features it cares about.
In prehistorical times, AffineApplyOp was allowed to produce multiple values.
This allowed the creation of intricate SSA use-def chains.
AffineApplyNormalizer was originally introduced as a means of reusing the AffineMap::compose method to write SSA use-def chains.
Unfortunately, symbols that were produced by an AffineApplyOp needed to be promoted to dims and reordered for the mathematical composition to be valid.
Since then, single result AffineApplyOp became the law of the land but the original assumptions were not revisited.
This revision revisits these assumptions and retires AffineApplyNormalizer.
Differential Revision: https://reviews.llvm.org/D94920
Such files (Thin-%%%%%%.tmp.o) are supposed to be deleted immediately
after they're used (either by renaming or deletion). However, we've seen
instances on Windows where this doesn't happen, probably due to the
filesystem being flaky. This is effectively a resource leak which has
prevented us from using the ThinLTO cache on Windows.
Since those temporary files are in the thinlto cache directory which we
prune periodically anyway, allowing them to be pruned too seems like a
tidy way to solve the problem.
Differential revision: https://reviews.llvm.org/D94962
This patch updates the llvm module map to reflect changes made in
`24672ddea3c97fd1eca3e905b23c0116d7759ab8` and fixes the module builds
(`-DLLVM_ENABLE_MODULES=On`).
Signed-off-by: Med Ismail Bennani <medismail.bennani@gmail.com>
Add the following standard predefinitions that f18 supports:
* `__flang__`,
* `__flang_major__`,
* `__flang_minor__`,
* `__flang_patchlevel__`
Summary of changes:
- Populate Fortran::parser::Options#predefinitions with the default
supported predefinitions
Differential Revision: https://reviews.llvm.org/D94516
Tweak dexter-tests/memvars/inline-escaping-function.c added in D94761
(b7e516202e) by adding a 'param' use after the merge point. The test XFAILS
with and without this change, but without it the test looks very similar to
memvars/unused-merged-value.c. The test now demonstrates the problem more
clearly.
Currently the new flang driver always runs in free form mode. This patch
adds support for fixed form mode detection based on the file extensions.
Like `f18`, `flang-new` will treat files ending with ".f", ".F" and
".ff" as fixed form. Additionally, ".for", ".FOR", ".fpp" and ".FPP"
file extensions are recognised as fixed form files. This is consistent
with gfortran [1]. In summary, files with the following extensions are
treated as fixed-form:
* ".f", ".F", ".ff", ".for", ".FOR", ".fpp", ".FPP"
For consistency with flang/test/lit.cfg.py and f18, this patch also adds
support for the following file extensions:
* ".ff", ".FOR", ".for", ".ff90", ".fpp", ".FPP"
This is added in flang/lib/Frontend/FrontendOptions.cpp. Additionally,
the following extensions are included:
* ".f03", ".F03", ".f08", ".F08"
This is for compatibility with gfortran [1] and other popular Fortran
compilers [2].
NOTE: internally Flang will only differentiate between fixed and free
form files. Currently Flang does not support switching between language
standards, so in this regard file extensions are irrelevant. More
specifically, both `file.f03` and `file.f18` are represented with
`Language::Fortran` (as opposed to e.g. `Language::Fortran03`).
Summary of changes:
- Set Fortran::parser::Options::sFixedForm according to the file type
- Add isFixedFormSuffix and isFreeFormSuffix helper functions to
FrontendTool/Utils.h
- Change FrontendOptions::GetInputKindForExtension to support the missing
file extensions that f18 supports and some additional ones
- FrontendActionTest.cpp is updated to make sure that the test input is
treated as free-form
[1] https://gcc.gnu.org/onlinedocs/gfortran/GNU-Fortran-and-GCC.html
[2] https://github.com/llvm/llvm-project/blob/master/flang/docs/OptionComparison.md#notes
Differential Revision: https://reviews.llvm.org/D94228
This was already done in SemaTemplateInstantiateDecl.cpp, but not in
SemaTemplateInstantiate.cpp.
Anecdotally I've seen some clangd crashes where coredumps point to this
being a problem, but I cannot reproduce this so far.
Differential Revision: https://reviews.llvm.org/D94933
On z/OS, the following error message is not matched correctly in lit tests. This patch updates the CHECK expression to match the end period successfully.
```
EDC5129I No such file or directory.
```
Differential Revision: https://reviews.llvm.org/D94239
This patch computes the cost for vector.reduce<operand> for scalable vectors.
The cost is split into two parts: the legalization cost and the horizontal
reduction.
Differential Revision: https://reviews.llvm.org/D93639
These dexter tests illustrate PR48719, the summary of which is:
Sometimes we insert dbg.values for merged values (PHIs) when promoting
variables, sometimes we don't. Sometimes there is no PHI because the merged
value is never used. It doesn't matter because LiveDebugValues understands these
merged values (implicit or otherwise) and correctly updates the debug
info. Importantly, these merged variable values (which may or may not exist as
PHIs, and may or not be represented with dbg.values) are //always// implicitly
defined by the combination of incoming edges and the incoming variable locations
along those edges by virtue of LiveDebugValues existing. Unfortunately, it is
possible to mess with the CFG and remove / move these edges before
LiveDebugValues runs. In this case our debug info model only works when the
merged value is tracked by a dbg.value. Currently, this is only done rigorously
for variables which are A) promoted in the first round of mem2reg and B) are
used after the merge point.
As an example, compile the following source with -O3 -g and step through with a
debugger. You will see parama=5 throughout the function fun which is incorrect -
we expect to see param=20 after the conditional assignment.
__attribute__((optnone))
void esc(int* p) {}
__attribute__((optnone))
void fluff() {}
__attribute__((noinline))
int fun(int parama, int paramb) {
if (parama)
parama = paramb;
fluff(); // DexLabel('s0')
esc(¶ma);
return 0;
}
int main() {
return fun(5, 20);
}
1. parama is escaped by esc(¶ma) so it is not promoted by
SROA/mem2reg (failing condition "A" above).
2. InstCombine's LowerDbgDeclare converts the dbg.declare to a set of
dbg.values (tracking the stored SSA values).
3. InstCombine replaces the two stores to parama's alloca (the initial
parameter register store in entry and the assignment in if.then) with a
PHI+store in the common sucessor.
4. SimplifyCFG folds the blocks together and converts the PHI to a
select.
The debug info is not updated to account for the merged value in the successor
prior to SimplifyCFG when it exists as a PHI, or during when it becomes a
select.
As with D89543, which added some dexter tests for escaped locals, the idea is
to build a set of source-level tests which highlights existing issues and
might be useful in evaluating a new debug info model.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D94761
Add support for option -I in the new Flang driver. This will allow for
included headers and module files in other directories, as the default
search path is currently the working folder. The behaviour of this is
consistent with the current f18 driver, where the current folder (i.e.
".") has the highest priority followed by the order of '-I's taking
priority from first to last.
Summary of changes:
- Add SearchDirectoriesFromDashI to PreprocessorOptions, to be forwarded
into the parser's searchDirectories
- Add header files and non-functional module files to be used in
regression tests. The module files are just text files and are used to
demonstrated that paths specified with `-I` are taken into account when
searching for .mod files.
Differential Revision: https://reviews.llvm.org/D93453
If a srl doesn't introduce any sign bits into the truncated result, then replace with a sra to let us use a PACKSS truncation - fixes a regression noticed in D56387 on pre-SSE41 targets that don't have PACKUSDW.
This caused a miscompile in Chromium, see comments on the codereview for
discussion and pointer to a reproducer.
> InstCombine already performs a fold where X == Y ? f(X) : Z is
> transformed to X == Y ? f(Y) : Z if f(Y) simplifies. However,
> if f(X) only has one use, then we can always directly replace the
> use inside the instruction. To actually be profitable, limit it to
> the case where Y is a non-expr constant.
>
> This could be further extended to replace uses further up a one-use
> instruction chain, but for now this only looks one level up.
>
> Among other things, this also subsumes D94860.
>
> Differential Revision: https://reviews.llvm.org/D94862
This also reverts the follow-up
a003f26539cf4db744655e76c41f4c4a8913f116:
> [llvm] Prevent infinite loop in InstCombine of select statements
>
> This fixes an issue where the RHS and LHS the comparison operation
> creating the predicate were swapped back and forth forever.
>
> Differential Revision: https://reviews.llvm.org/D94934
D84108 exposed a bad interaction between inlining and loop-rotation
during regular LTO, which is causing notable regressions in at least
CINT2006/473.astar.
The problem boils down to: we now rotate a loop just before the vectorizer
which requires duplicating a function call in the preheader when compiling
the individual files ('prepare for LTO'). But this then prevents further
inlining of the function during LTO.
This patch tries to resolve this issue by making LoopRotate more
conservative with respect to rotating loops that have inline-able calls
during the 'prepare for LTO' stage.
I think this change intuitively improves the current situation in
general. Loop-rotate tries hard to avoid creating headers that are 'too
big'. At the moment, it assumes all inlining already happened and the
cost of duplicating a call is equal to just doing the call. But with LTO,
inlining also happens during full LTO and it is possible that a previously
duplicated call is actually a huge function which gets inlined
during LTO.
From the perspective of LV, not much should change overall. Most loops
calling user-provided functions won't get vectorized to start with
(unless we can infer that the function does not touch memory, has no
other side effects). If we do not inline the 'inline-able' call during
the LTO stage, we merely delayed loop-rotation & vectorization. If we
inline during LTO, chances should be very high that the inlined code is
itself vectorizable or the user call was not vectorizable to start with.
There could of course be scenarios where we inline a sufficiently large
function with code not profitable to vectorize, which would have be
vectorized earlier (by scalarzing the call). But even in that case,
there probably is no big performance impact, because it should be mostly
down to the cost-model to reject vectorization in that case. And then
the version with scalarized calls should also not be beneficial. In a way,
LV should have strictly more information after inlining and make more
accurate decisions (barring cost-model issues).
There is of course plenty of room for things to go wrong unexpectedly,
so we need to keep a close look at actual performance and address any
follow-up issues.
I took a look at the impact on statistics for
MultiSource/SPEC2000/SPEC2006. There are a few benchmarks with fewer
loops rotated, but no change to the number of loops vectorized.
Reviewed By: sanwou01
Differential Revision: https://reviews.llvm.org/D94232
This patch adds a new test case which depends on AArch64 SVE support and
dynamic resize capability enabled. It created two seperate threads which
have different values of sve registers and SVE vector granule at various
points during execution.
We test that LLDB is doing the size and offset updates properly for all
of the threads including the main thread and when we VG is updated using
prctl call or by 'register write vg' command the appropriate changes are
also update in register infos.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D82866
Jessica Paquette