The large code model allows code and data segments to exceed 2GB, which
means that some symbol references may require a displacement that cannot
be encoded as a displacement from RIP. The large PIC model even relaxes
the assumption that the GOT itself is within 2GB of all code. Therefore,
we need a special code sequence to materialize it:
.LtmpN:
leaq .LtmpN(%rip), %rbx
movabsq $_GLOBAL_OFFSET_TABLE_-.LtmpN, %rax # Scratch
addq %rax, %rbx # GOT base reg
From that, non-local references go through the GOT base register instead
of being PC-relative loads. Local references typically use GOTOFF
symbols, like this:
movq extern_gv@GOT(%rbx), %rax
movq local_gv@GOTOFF(%rbx), %rax
All calls end up being indirect:
movabsq $local_fn@GOTOFF, %rax
addq %rbx, %rax
callq *%rax
The medium code model retains the assumption that the code segment is
less than 2GB, so calls are once again direct, and the RIP-relative
loads can be used to access the GOT. Materializing the GOT is easy:
leaq _GLOBAL_OFFSET_TABLE_(%rip), %rbx # GOT base reg
DSO local data accesses will use it:
movq local_gv@GOTOFF(%rbx), %rax
Non-local data accesses will use RIP-relative addressing, which means we
may not always need to materialize the GOT base:
movq extern_gv@GOTPCREL(%rip), %rax
Direct calls are basically the same as they are in the small code model:
They use direct, PC-relative addressing, and the PLT is used for calls
to non-local functions.
This patch adds reasonably comprehensive testing of LEA, but there are
lots of interesting folding opportunities that are unimplemented.
I restricted the MCJIT/eh-lg-pic.ll test to Linux, since the large PIC
code model is not implemented for MachO yet.
Differential Revision: https://reviews.llvm.org/D47211
llvm-svn: 335508
We canonicalize to select with a zext-add and either zext-sub or sext-sub,
so this shows a pattern that's not conforming to the general trend.
llvm-svn: 335506
Pass -enable-linkonceodr-outlining by default when LTO is enabled.
The outliner shouldn't compete with any sort of linker deduplication
on linkonceodr functions when LTO is enabled. Therefore, this behaviour
should be the default.
llvm-svn: 335504
The expected behaviour of command-line flags to clang is to have
the last of -m(whatever) and -mno-(whatever) win. The outliner
didn't do that. This fixes that and updates the test.
llvm-svn: 335503
With the static tables sorted we can binary search them directly for reg->mem lookups. This removes 6 DenseMaps that had to be created when X86InstrInfo is constructed.
We still have one Mem->Reg DenseMap for the reverse direction. This is created just as before by walking the reg->mem arrays to populate it.
Differential Revision: https://reviews.llvm.org/D48527
llvm-svn: 335501
* Move `REQUIRES:` line to the top
* llvm-mc ... -o %t -> llvm-mc ... -o %t.o
* Don't check "TEXT" "DATA" columns (they are bfd-style names that do
not fit into llvm well) in llvm-objdump output
llvm-svn: 335498
This removes debug locations from ConstantSDNode and ConstantSDFPNode.
When this kind of node is materialized we no longer create a line table
entry which jumps back to the constant's first point of use. This makes
single-stepping behavior smoother, and it matches the model used by IR,
where Constants have no locations. See this thread for more context:
http://lists.llvm.org/pipermail/llvm-dev/2018-June/124164.html
I'd like to handle constant BuildVectorSDNodes and to try to eliminate
passing SDLocs to SelectionDAG::getConstant*() in follow-up commits.
Differential Revision: https://reviews.llvm.org/D48468
llvm-svn: 335497
Summary:
This change renames the Backend and BackendPrinter to Pipeline and PipelinePrinter respectively.
Variables and comments have also been updated to reflect this change.
The reason for this rename, is to be slightly more correct about what MCA is modeling. MCA models a Pipeline, which implies some logical sequence of stages.
Reviewers: andreadb, courbet, RKSimon
Reviewed By: andreadb, courbet
Subscribers: mgorny, javed.absar, tschuett, gbedwell, llvm-commits
Differential Revision: https://reviews.llvm.org/D48496
llvm-svn: 335496
There are quite a few if statements that enumerate all these cases. It gets
even worse in our fork of LLVM where we also have a Triple::cheri (which
is mips64 + CHERI instructions) and we had to update all if statements that
check for Triple::mips64 to also handle Triple::cheri. This patch helps to
reduce our diff to upstream and should also make some checks more readable.
Reviewed By: atanasyan
Differential Revision: https://reviews.llvm.org/D48548
llvm-svn: 335493
Summary:
Tools that reformat code often call `getStyle` to decide the format style
to use on a certain source file. In practice, "file" style is widely used. As a
result, many tools hardcode "file" when calling `getStyle`, which makes it hard
to control the default style in tools across a codebase when needed. This change
introduces a `DefaultFormatStyle` constant (default to "file" in upstream), which
can be modified downstream if wanted, so that all users/tools built from the same
source tree can have a consistent default format style.
This also adds an DefaultFallbackStyle that is recommended to be used by tools and can be modified downstream.
Reviewers: sammccall, djasper
Reviewed By: sammccall
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D48492
llvm-svn: 335492
Note a normal select test is not currently possible because this
relies on input registers tracked in SIMachineFunctionInfo which
are not currently serializable in MIR, but this does work end-to-end
from the IR.
llvm-svn: 335490
I thought I fixed this in r308673, but that fix was
very broken. The assumption that any frame index can be used
in place of another was more widespread than I realized.
Even when stack slot sharing was disabled, this was still
replacing frame index uses with a different ID with a different
stack slot.
Really fix this by doing the coloring per-stack ID, so all of
the coloring logically done in a separate namespace. This is a lot
simpler than trying to figure out how to change the color if
the stack ID is different.
llvm-svn: 335488
If a function has sample to use, but cannot use them because of no debug
information, currently a warning will be issued to inform the missing
opportunity.
This warning assumes the binary generating the profile and the binary using
the profile are similar enough. It is not always the case. Sometimes even
if the binaries are not quite similar, we may still get some benefit by
using sampleFDO. In those cases, we may still want to apply sampleFDO but
not want to see a lot of such warnings pop up.
The patch adds an option for the warning.
Differential Revision: https://reviews.llvm.org/D48510
llvm-svn: 335484
ICF is able to merge sections which relocations referring regular input sections
or mergeable sections, so it handles InputSection and MergeInputSection cases.
The following "return false" line which is executed in case of another type
of the sections is uncovered by our test cases:
https://github.com/llvm-mirror/lld/blob/master/ELF/ICF.cpp#L285
Patch fixes code coverage for this place.
llvm-svn: 335482
We can prove that some delinearized subscripts do not wrap around to become
negative by the fact that they are from inbound geps of load/store locations.
This helps improve the delinearisation in cases where we can't prove that they
are non-negative from SCEV alone.
Differential Revision: https://reviews.llvm.org/D48481
llvm-svn: 335481
Summary:
With this patch when any `FunctionDecl` of a redeclaration chain is imported
then we bring in the whole declaration chain. This involves functions and
function template specializations. Also friend functions are affected. The
chain is imported as it is in the "from" tu, the order of the redeclarations
are kept. I also changed the lookup logic in order to find friends, but first
making them visible in their declaration context. We may have long
redeclaration chains if all TU contains the same prototype, but our
measurements shows no degradation in time of CTU analysis (Tmux, Xerces,
Bitcoin, Protobuf). Also, as further work we could squash redundant
prototypes, but first ensure that functionality is working properly; then
should we optimize.
This may seem like a huge patch, sorry about that. But, most of the changes are
new tests, changes in the production code is not that much. I also tried to
create a smaller patch which does not affect specializations, but that patch
failed to pass some of the `clang-import-test`s because there we import
function specializations. Also very importantly, we can't just change the
import of `FunctionDecl`s without changing the import of function template
specializations because they are handled as `FunctionDecl`s.
Reviewers: a.sidorin, r.stahl, xazax.hun, balazske
Subscribers: rnkovacs, dkrupp, cfe-commits
Differential Revision: https://reviews.llvm.org/D47532
llvm-svn: 335480
This should avoid relying on the pointee type
to get the alignment, particularly since pointee
types are supposed to be removed at some point.
Also fixes not getting the alignment for unsized types.
llvm-svn: 335478
pseudo_barrier_wait() begins by decrementing an atomic variable. Since
these are always_inline in libc++, there is no line table anchor to
break on before we decrement it. This meant that on gcc we stopped after
the variable has been decremented, which meant that thread2 could have
exited, violating the test setup. On clang this wasn't a problem
because it generated some line table entries for the do{}while(0) loop
in the macro, so we still ended up stopping, before we touched the
variable.
I fix this by adding a dummy statement before the pseudo_barrier_wait()
command and setting the breakpoint there.
llvm-svn: 335476
This test case check that ICF does not merge 2 sections which relocations
efer to symbols that live in sections of the different types
(regular input section and mergeable input sections in this case).
It covers the following line of code, which was uncovered previously:
https://github.com/llvm-mirror/lld/blob/master/ELF/ICF.cpp#L271
llvm-svn: 335475
Not only should SafepointIRVerifier ignore unreachable blocks (as suggested in https://reviews.llvm.org/D47011) but it also has to ignore dead blocks.
In @test2 (see the new tests):
br i1 true, label %right, label %left
left:
...
right:
...
merge:
%val = phi i8 addrspace(1)* [ ..., %left ], [ ..., %right ]
use %val
both left and right branches are reachable.
If they collide then SafepointIRVerifier reports an error.
Because of the foldable branch condition GVN finds the left branch dead and removes the phi node entry that merges values from right and left. Then the use comes from the right branch. This results in no collision.
So, SafepointIRVerifier ends up in different results depending on either GVN is run or not.
To solve this issue this patch adds Dead Block detection to SafepointIRVerifier which can ignore dead blocks while validating IR. The Dead Block detection algorithm is taken from GVN but modified to not split critical edges. That is needed to keep CFG unchanged by SafepointIRVerifier.
Patch by Yevgeny Rouban.
Reviewed By: anna, apilipenko, DaniilSuchkov
Differential Revision: https://reviews.llvm.org/D47441
llvm-svn: 335473
Summary:
Patch fixes several problems in the implementation of NVPTX RTL.
1. Detection of the last iteration for loops with static scheduling, no chunks.
2. Fixes reductions for the serialized parallel constructs.
3. Fixes handling of the barriers.
Reviewers: grokos
Reviewed By: grokos
Subscribers: Hahnfeld, guansong, openmp-commits
Differential Revision: https://reviews.llvm.org/D48480
llvm-svn: 335469
There's probably a better solution, but adding spaces
in the IR vector examples sidesteps the problem without
uglifying the plain text.
llvm-svn: 335468
With MSVC, PCH files are created along with an object file that needs to
be linked into the final library or executable. That object file
contains the code generated when building the headers. In particular, it
will include definitions of inline dllexport functions, and because they
are emitted in this object file, other files using the PCH do not need
to emit them. See the bug for an example.
This patch makes clang-cl match MSVC's behaviour in this regard, causing
significant compile-time savings when building dlls using precompiled
headers.
For example, in a 64-bit optimized shared library build of Chromium with
PCH, it reduces the binary size and compile time of
stroke_opacity_custom.obj from 9315564 bytes to 3659629 bytes and 14.6
to 6.63 s. The wall-clock time of building blink_core.dll goes from
38m41s to 22m33s. ("user" time goes from 1979m to 1142m).
Differential Revision: https://reviews.llvm.org/D48426
llvm-svn: 335466
Summary:
This ensures that the snippet always sees the same values for registers,
making measurements reproducible.
This will also allow exploring different values.
Reviewers: gchatelet
Subscribers: tschuett, llvm-commits
Differential Revision: https://reviews.llvm.org/D48542
llvm-svn: 335465
Summary:
In order to avoid build failures on MS, we use -fms-compatibility too in
the tests which use the TestBase. Moved the family of `testImport`
functions under a test fixture class, so we can use parameterized tests.
Refactored `testImport` and `testImportSequence`, because `for` loops over
the different compiler options is no longer needed, that is handeld by
the test framework via parameters from now on.
Reviewers: a.sidorin, r.stahl, xazax.hun
Subscribers: rnkovacs, dkrupp, cfe-commits
Differential Revision: https://reviews.llvm.org/D47367
llvm-svn: 335464
Code is dead. We use only InputSections when building the list of
sections elegible for the ICF:
https://github.com/llvm-mirror/lld/blob/master/ELF/ICF.cpp#L439
And 'isEligible' filters out SyntheticSections as well for us.
That way the only Kind we have in the Sections vector is SectionBase::Regular,
so we do not need to check sections kind at all, it is always the same.
llvm-svn: 335460
Summary:
In C++ code snippets of the form `@field` are common. This makes clang-format
keep them together in text protos, whereas before it would break them.
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D48543
llvm-svn: 335459
Sanjay Patel