The Hexagon Vector Loop Carried Reuse pass was allowing reuse between
two shufflevectors with different masks. The reason is that the masks
are not instruction objects, so the code that checks each operand
just skipped over the operands.
This patch fixes the bug by checking if the operands are the same
when they are not instruction objects. If the objects are not the
same, then the code assumes that reuse cannot occur.
Differential Revision: https://reviews.llvm.org/D60019
llvm-svn: 358292
// shuffle (concat X, undef), (concat Y, undef), Mask -->
// concat (shuffle X, Y, Mask0), (shuffle X, Y, Mask1)
The ARM changes with 'vtrn' and narrowed 'vuzp' are improvements.
The x86 changes look neutral or better. There's one test with an
extra instruction, but that could be reversed for a subtarget with
the right attributes. But by default, we want to avoid the 256-bit
op when possible (in my motivating benchmark, a handful of ymm ops
sprinkled into a sequence of xmm ops are triggering frequency
throttling on Haswell resulting in significantly worse perf).
Differential Revision: https://reviews.llvm.org/D60545
llvm-svn: 358291
Use -mlink-builtin-bitcode instead of llvm-link to link
device library so that device library bitcode and user
device code can be compiled in a consistent way.
This is the same approach used by CUDA and OpenMP.
Differential Revision: https://reviews.llvm.org/D60513
llvm-svn: 358290
Systematically add the const-qualified version of children()
to all statement/expression nodes. Previously the const-qualified
variant was only defined for some nodes. NFC.
Patch by: Nicolas Manichon
Differential Revision: https://reviews.llvm.org/D60029
Reviewed By: riccibruno
llvm-svn: 358288
Currently combineHorizontalPredicateResult only handles anyof/allof reduction patterns of legal types, which can be tricky to match as type legalization of bools can introduce bitcasts/truncs/extensions.
This patch extends combineHorizontalPredicateResult to recognise vXi1 bool reductions as well and uses the existing combineBitcastvxi1 helper to create the MOVMSK necessary to then compare the signmask result.
This ensures the accuracy of the reduction costs added in D60403 which assume the MOVMSK generation.
Differential Revision: https://reviews.llvm.org/D60610
llvm-svn: 358286
Disabled by default as this is still an experimental feature.
Reviewed By: thakis
Differential Revision: https://reviews.llvm.org/D59221
llvm-svn: 358285
Statements, expressions and types are not supposed to be copied/moved,
and trying to do so is only going to result in tears. Someone tripped
on this a few days ago on the mailing list. NFC.
Differential Revision: https://reviews.llvm.org/D60123
Reviewed By: aaron.ballman
llvm-svn: 358283
Summary:
We have turned on the flag internally for a while, and we don't receive complains.
Should be good to turn it on now.
If the projects doesn't have .clang-tidy files, no clang-tidy check will
be run.
Reviewers: sammccall
Subscribers: ilya-biryukov, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60560
llvm-svn: 358282
It causes clang to crash while building Chromium. See https://crbug.com/952230
for reproducer.
> The PrologEpilogInserter need to insert a DW_OP_deref_size before
> prepending a memory location expression to an already implicit
> expression to avoid having the existing expression act on the memory
> address instead of the value behind it.
>
> The reason for using DW_OP_deref_size and not plain DW_OP_deref is that
> big-endian targets need to read the right size as simply truncating a
> larger read would yield the wrong result (LSB bytes are not at the lower
> address).
>
> Differential Revision: https://reviews.llvm.org/D59687
llvm-svn: 358281
Summary:
AnnotatingParser::next() is needed to implicitly set TT_BlockComment
versus TT_LineComment. On most other paths through
AnnotatingParser::parseLine(), all tokens are consumed to achieve that.
This change updates one place where this wasn't done.
Contributed by @dchai!
Reviewers: krasimir
Reviewed By: krasimir
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60541
llvm-svn: 358275
Summary:
Last part of re-landing rC356541. Puts TemplateArgumentsList into
responses of the above mentioned two requests.
Reviewers: ioeric, ilya-biryukov
Subscribers: MaskRay, jkorous, arphaman, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D59641
llvm-svn: 358274
Summary:
Part of re-landing rC356541 with D59599. Changes the way we store
template arguments, previous patch was storing them inside Name field of Symbol.
Which was violating the assumption:
```Symbol::Scope+Symbol::Name == clang::clangd::printQualifiedName```
which was made in multiple places inside codebase. This patch instead moves
those arguments into their own field. Currently the field is meant to be
human-readable, can be made structured if need be.
Reviewers: ioeric, ilya-biryukov, gribozavr
Subscribers: MaskRay, jkorous, arphaman, jdoerfert, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D59640
llvm-svn: 358273
Summary:
Prepares ground for printing template arguments as written in the
source code, part of re-landing rC356541 with D59599 applied.
Reviewers: ioeric, ilya-biryukov
Subscribers: mgorny, MaskRay, jkorous, arphaman, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D59639
llvm-svn: 358272
Summary:
Some llc debug options need pass-name as the parameters.
But if we use the pass-name ppc-early-ret, we will get below error:
llc test.ll -stop-after ppc-early-ret
LLVM ERROR: "ppc-early-ret" pass is not registered.
Below pass-names have the pass is not registered error:
ppc-ctr-loops
ppc-ctr-loops-verify
ppc-loop-preinc-prep
ppc-toc-reg-deps
ppc-vsx-copy
ppc-early-ret
ppc-vsx-fma-mutate
ppc-vsx-swaps
ppc-reduce-cr-ops
ppc-qpx-load-splat
ppc-branch-coalescing
ppc-branch-select
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D60248
llvm-svn: 358271
Bug: https://bugs.llvm.org/show_bug.cgi?id=41175
In the bug test case the DSE pass is shortening the range of memory that a
memset is working on. A getelementptr is generated so that the new
starting address can be passed to memset. This instruction was not given
a DebugLoc.
To fix the bug, copy the DebugLoc from the memset instruction.
Patch by Orlando Cazalet-Hyams!
Differential Revision: https://reviews.llvm.org/D60556
llvm-svn: 358270
The PrologEpilogInserter need to insert a DW_OP_deref_size before
prepending a memory location expression to an already implicit
expression to avoid having the existing expression act on the memory
address instead of the value behind it.
The reason for using DW_OP_deref_size and not plain DW_OP_deref is that
big-endian targets need to read the right size as simply truncating a
larger read would yield the wrong result (LSB bytes are not at the lower
address).
Differential Revision: https://reviews.llvm.org/D59687
llvm-svn: 358268
In tutorial "8. Kaleidoscope: Compiling to Object Code" a call to
TargetMachine->addPassesToEmitFile(pass, dest, FileType) is missing
nullptr as its 3rd value.
Patch by Sajjad Heydari!
Differential revision: https://reviews.llvm.org/D60369
llvm-svn: 358267
This test contained an incredibly complicated inferior, but in reality,
all it was testing was that we can backtrace up to main and see main's
arguments.
However, the way this was implemented (setting a breakpoint on a
separate thread) meant that each time the test would run, it would stop
in a different location on the main thread. Most of the time this
location would be deep in some libc function, which meant that the
success of this test depended on our ability to backtrace out of a
random function of the c library that the user happens to have
installed.
This makes the test unpredictable. Backtracing out of a libc function is
an important functionality, but this is not the way to test it. Often it
is not even our fault that we cannot backtrace out because the C library
contains a lot of assembly routines that may not have correct unwind
info associated with them.
For this reason the test has accumulated numerous @expectedFail/Flaky
decorators. In this patch, I replace the inferior with one that does not
depend on libc functions. Instead I create a couple of stack frames of
user code, and have the test verify that. I also simplify the test by
using lldbutil.run_to_source_breakpoint.
llvm-svn: 358266
the MCDwarf.h include.
This removes 50 transitive dependencies for a modification of
MCDwarf.h in a build of llc for a pair of out of line functions
and reduces the build overhead of 'touch MCDwarf.h" by 15% without
impacting test time of check-llvm.
llvm-svn: 358264
Summary:
This patch attempts to solve two issues made this code hard to follow
for me.
The first issue was that a lot of what these visitors do is mutate the
AST. The visitor pattern is not particularly good for that because by
the time you have performed the dynamic type dispatch, it's too late to
go back to the parent node, and change its pointer. The previous code
dealt with that relatively elegantly, but it still meant that one had to
perform manual type checks, which is what the visitor pattern is
supposed to avoid.
The second issue was not being able to return values from the Visit
functions, which meant that one had to store function results in member
variables (a common problem with visitor patterns).
Here, I solve both problems by making the visitor use a type switch
instead of going through double dispatch on the visited object. This
allows one to parameterize the visitor based on the return type and pass
function results as function results. The mutation is fascilitated by
having each Visit function take two arguments -- a reference to the
object itself (with the correct dynamic type), and a reference to the
parent's pointer to this object.
Although this wasn't my explicit goal here, the fact that we're not
using virtual dispatch anymore allows us to make the AST nodes
trivially destructible, which is a good thing, since we were not
destroying them anyway.
Reviewers: aleksandr.urakov, amccarth
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D60410
llvm-svn: 358261
This removes 50 transitive dependencies for a modification of
MCDwarf.h in a build of llc for a single out of line function
and reduces the build overhead by 20% without impacting test
time of check-llvm.
llvm-svn: 358258
If the upper bits of the SHL result aren't used, we might be able to use a narrower shift. For example, on X86 this can turn a 64-bit into 32-bit enabling a smaller encoding.
Differential Revision: https://reviews.llvm.org/D60358
llvm-svn: 358257
Summary:
Some llc debug options need pass-name as the parameters.
But if we use the pass-name ppc-early-ret, we will get below error:
llc test.ll -stop-after ppc-early-ret
LLVM ERROR: "ppc-early-ret" pass is not registered.
Below pass-names have the pass is not registered error:
ppc-ctr-loops
ppc-ctr-loops-verify
ppc-loop-preinc-prep
ppc-toc-reg-deps
ppc-vsx-copy
ppc-early-ret
ppc-vsx-fma-mutate
ppc-vsx-swaps
ppc-reduce-cr-ops
ppc-qpx-load-splat
ppc-branch-coalescing
ppc-branch-select
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D60248
llvm-svn: 358256
This removes 500 transitive dependencies for a modification of
MCDwarf.h in a build of llc for a single out of line function
and reduces the build overhead by more than half without impacting
test time of check-llvm.
llvm-svn: 358255
There are 3 operands of maddld, (add (mul %1, %2), %3) and sometimes
they are constant. If there is constant operand, it takes extra li to
materialize the operand, and one more extra register too. So it's not
profitable to use maddld to optimize mul-add pattern.
Differential Revision: https://reviews.llvm.org/D60181
llvm-svn: 358253
Instead of having an `#if` block in every algorithm using a debug
comparator, this patch introduces the __comp_ref_type trait that
selects __debug_less in debug mode and _Comp& otherwise.
This patch should have no observable functionality change.
llvm-svn: 358252
The typo was introduced to llvm MC in rL204769 (fixed in rL358247) and then to lld.
Also, for relocatable-many-sections.s, the size of .symtab changed at some point and the formula needs update.
llvm-svn: 358248
This special section is named .symtab_shndx, according to gABI Chapter 4
Sections, and the name is used by some other tools. Though the section
type SHT_SYMTAB_SHNDX is what really matters, let's fix the typo
introduced in rL204769 :)
llvm-svn: 358247
Summary:
alloca isn’t auto-init’d right now because it’s a different path in clang that
all the other stuff we support (it’s a builtin, not an expression).
Interestingly, alloca doesn’t have a type (as opposed to even VLA) so we can
really only initialize it with memset.
<rdar://problem/49794007>
Subscribers: jkorous, dexonsmith, cfe-commits, rjmccall, glider, kees, kcc, pcc
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60548
llvm-svn: 358243
