JF Bastien [Wed, 14 Aug 2019 17:39:07 +0000 (17:39 +0000)]
Move to C++14
Summary:
I just bumped the minimum compiler versions to support C++14 in D66188.
Following [our process](http://llvm.org/docs/DeveloperPolicy.html#toolchain) and [our previous agreement](http://lists.llvm.org/pipermail/llvm-dev/2019-January/129452.html), I'm now officially bumping the C++ version to 14 and updating the documentation.
After switching over LLDB's line table parser to libDebugInfo, we
noticed two regressions on the Windows bot. The problem is that when
obtaining a file from the line table prologue, we append paths without
specifying a path style. This leads to incorrect results on Windows for
debug info containing Posix paths:
Thomas Lively [Wed, 14 Aug 2019 16:24:37 +0000 (16:24 +0000)]
[WebAssembly] Stop unrolling SIMD shifts since they are fixed in V8
Summary:
Fixes PR42973. Tests don't change because simd-arith.ll tests behavior
on unimplemented-simd128, which does not include any temporary
workarounds such as the one removed in this revision.
Craig Topper [Wed, 14 Aug 2019 14:52:39 +0000 (14:52 +0000)]
[X86][CostModel] Adjust the costs of ZERO_EXTEND/SIGN_EXTEND with less than 128-bit inputs
Now that we legalize by widening, the element types here won't change. Previously these were modeled as the elements being widened and then the instruction might become an AND or SHL/ASHR pair. But now they'll become something like a ZERO_EXTEND_VECTOR_INREG/SIGN_EXTEND_VECTOR_INREG.
For AVX2, when the destination type is legal its clear the cost should be 1 since we have extend instructions that can produce 256 bit vectors from less than 128 bit vectors. I'm a little less sure about AVX1 costs, but I think the ones I changed were definitely too high, but they might still be too high.
Pavel Labath [Wed, 14 Aug 2019 13:59:04 +0000 (13:59 +0000)]
Revert "raw_ostream: add operator<< overload for std::error_code"
This reverts commit r368849, because it breaks some bots (e.g.
llvm-clang-x86_64-win-fast).
It turns out this is not as NFC as we had hoped, because operator== will
consider two std::error_codes to be distinct even though they both hold
"success" values if they have different categories.
Pavel Labath [Wed, 14 Aug 2019 13:33:28 +0000 (13:33 +0000)]
raw_ostream: add operator<< overload for std::error_code
Summary:
The main motivation for this is unit tests, which contain a large macro
for pretty-printing std::error_code, and this macro is duplicated in
every file that needs to do this. However, the functionality may be
useful elsewhere too.
In this patch I have reimplemented the existing ASSERT_NO_ERROR macros
to reuse the new functionality, but I have kept the macro (as a
one-liner) as it is slightly more readable than ASSERT_EQ(...,
std::error_code()).
Jeremy Morse [Wed, 14 Aug 2019 12:20:02 +0000 (12:20 +0000)]
[DebugInfo] MCP: collect and update DBG_VALUEs encountered in local block
MCP currently uses changeDebugValuesDefReg / collectDebugValues to find
debug users of a register, however those functions assume that all
DBG_VALUEs immediately follow the specified instruction, which isn't
necessarily true. This is going to become very often untrue when we turn
off CodeGenPrepare::placeDbgValues.
Instead of calling changeDebugValuesDefReg on an instruction to change its
debug users, in this patch we instead collect DBG_VALUEs of copies as we
iterate over insns, and update the debug users of copies that are made
dead. This isn't a non-functional change, because MCP will now update
DBG_VALUEs that aren't immediately after a copy, but refer to the same
register. I've hijacked the regression test for PR38773 to test for this
new behaviour, an entirely new test seemed overkill.
George Rimar [Wed, 14 Aug 2019 11:10:11 +0000 (11:10 +0000)]
Recommit r368812 "[llvm/Object] - Convert SectionRef::getName() to return Expected<>"
Changes: no changes. A fix for the clang code will be landed right on top.
Original commit message:
SectionRef::getName() returns std::error_code now.
Returning Expected<> instead has multiple benefits.
For example, it forces user to check the error returned.
Also Expected<> may keep a valuable string error message,
what is more useful than having a error code.
(Object\invalid.test was updated to show the new messages printed.)
This patch makes a change for all users to switch to Expected<> version.
Note: in a few places the error returned was ignored before my changes.
In such places I left them ignored. My intention was to convert the interface
used, and not to improve and/or the existent users in this patch.
(Though I think this is good idea for a follow-ups to revisit such places
and either remove consumeError calls or comment each of them to clarify why
it is OK to have them).
James Henderson [Wed, 14 Aug 2019 10:17:34 +0000 (10:17 +0000)]
[llvm-size][test] Improve llvm-size testing
This patch significantly improves the llvm-size testing. The changes
made are:
1) Change all tests to use yaml2obj instead of assembly or pre-canned
inputs.
2) Move the tests out of the X86 directory, since they don't need to be
there after 1).
3) Increased test coverage.
4) Added comments to explain purpose of tests.
I haven't attempted to add test coverage for all Mach-O related code, as
I am not familiar enough with that file format to be able to.
George Rimar [Wed, 14 Aug 2019 08:46:54 +0000 (08:46 +0000)]
[llvm/Object] - Convert SectionRef::getName() to return Expected<>
SectionRef::getName() returns std::error_code now.
Returning Expected<> instead has multiple benefits.
For example, it forces user to check the error returned.
Also Expected<> may keep a valuable string error message,
what is more useful than having a error code.
(Object\invalid.test was updated to show the new messages printed.)
This patch makes a change for all users to switch to Expected<> version.
Note: in a few places the error returned was ignored before my changes.
In such places I left them ignored. My intention was to convert the interface
used, and not to improve and/or the existent users in this patch.
(Though I think this is good idea for a follow-ups to revisit such places
and either remove consumeError calls or comment each of them to clarify why
it is OK to have them).
George Rimar [Wed, 14 Aug 2019 08:33:26 +0000 (08:33 +0000)]
[llvm-objdump] - Add a relocation-xindex-symbol.test test case.
This rewrites the exitent test case to use YAML instead of the precompiled object
and moves it from test/Object to an appropriate llvm-objdump tests folder.
Dorit Nuzman [Wed, 14 Aug 2019 05:22:20 +0000 (05:22 +0000)]
[LV] Fold-tail flag
This is the compiler-flag equivalent of the Predicate pragma
(https://reviews.llvm.org/D65197), to direct the vectorizer to fold the
remainder-loop into the main-loop using predication.
JF Bastien [Wed, 14 Aug 2019 04:30:51 +0000 (04:30 +0000)]
Remove minimum toolchain soft-error
Summary:
Back in January I changed the minimum toolchain version required to build clang
and LLVM: D57264. Since then we've release LLVM 8, following
[our process](http://llvm.org/docs/DeveloperPolicy.html#toolchain)
it's therefore now a good time to remove the soft-error and officially deprecate
older toolchains. I tried this out last Tursday night to see if any bots
complained, and I saw no complaints. I also manually audited bots and didn't see
any bot that should break, but their toolchain information is unreliable and
some bots are offline.
Once this patch stick we'll move to C++14 as we've
[already agreed](http://lists.llvm.org/pipermail/llvm-dev/2019-January/129452.html).
John McCall [Wed, 14 Aug 2019 03:54:05 +0000 (03:54 +0000)]
Support swifterror in coroutine lowering.
The support for swifterror allocas should work in all lowerings.
The support for swifterror arguments only really works in a lowering
with prototypes where you can ensure that the prototype also has a
swifterror argument; I'm not really sure how it could possibly be
made to work in the switch lowering.
John McCall [Wed, 14 Aug 2019 03:53:17 +0000 (03:53 +0000)]
Extend coroutines to support a "returned continuation" lowering.
A quick contrast of this ABI with the currently-implemented ABI:
- Allocation is implicitly managed by the lowering passes, which is fine
for frontends that are fine with assuming that allocation cannot fail.
This assumption is necessary to implement dynamic allocas anyway.
- The lowering attempts to fit the coroutine frame into an opaque,
statically-sized buffer before falling back on allocation; the same
buffer must be provided to every resume point. A buffer must be at
least pointer-sized.
- The resume and destroy functions have been combined; the continuation
function takes a parameter indicating whether it has succeeded.
- Conversely, every suspend point begins its own continuation function.
- The continuation function pointer is directly returned to the caller
instead of being stored in the frame. The continuation can therefore
directly destroy the frame when exiting the coroutine instead of having
to leave it in a defunct state.
- Other values can be returned directly to the caller instead of going
through a promise allocation. The frontend provides a "prototype"
function declaration from which the type, calling convention, and
attributes of the continuation functions are taken.
- On the caller side, the frontend can generate natural IR that directly
uses the continuation functions as long as it prevents IPO with the
coroutine until lowering has happened. In combination with the point
above, the frontend is almost totally in charge of the ABI of the
coroutine.
- Unique-yield coroutines are given some special treatment.
Joel E. Denny [Wed, 14 Aug 2019 02:56:09 +0000 (02:56 +0000)]
[FileCheck] Move -dump-input diagnostic to first line
Without this patch, `-dump-input` prints a diagnostic at the end of
its marker range. For example:
```
1: Start.
check:1 ^~~~~~
2: Bad.
next:2 X~~~
3: Many lines
next:2 ~~~~~~~~~~
4: of input.
next:2 ~~~~~~~~~
5: End.
next:2 ~~~~ error: no match found
```
This patch moves it to the beginning like this:
```
1: Start.
check:1 ^~~~~~
2: Bad.
next:2 X~~~ error: no match found
3: Many lines
next:2 ~~~~~~~~~~
4: of input.
next:2 ~~~~~~~~~
5: End.
next:2 ~~~~
```
The former somehow looks nicer because the diagnostic doesn't appear
to be somewhere within the marker range. However, the latter is more
practical, especially when the marker range includes the remainder of
a very long dump. First, in the case of an error, this patch enables
me to search the dump for `error:` and usually immediately land where
the detected error began. Second, when trying to follow FileCheck's
logic, it's best to read top down, so this patch enables me to see
each diagnostic as soon as I encounter its marker.
Alex Langford [Tue, 13 Aug 2019 22:16:18 +0000 (22:16 +0000)]
[NFCI] Explicitly provide user-defined constructor for SectionRef
I am changing this to work around an issue that is being hit when
building with clang 3.8. Specifically, clang 3.8 requires that we have a user
defined default constructor for SectionRef for the default initialization of a
const SectionRef.
Jessica Paquette [Tue, 13 Aug 2019 22:14:37 +0000 (22:14 +0000)]
[GlobalISel][NFC] Factor out common target code from GlobalISelEmitterTests
Factor out commonly-used target code from the GlobalISelEmitter tests into
a GlobalISelEmitterCommon.td file. This is tested by the original
GlobalISelEmitter.td test.
This reduces the amount of boilerplate code necessary for tests like this.
Xiangling Liao [Tue, 13 Aug 2019 20:29:01 +0000 (20:29 +0000)]
[AIX]Lowering global address for 32/64bit small/large code models
This patch implements global address lowering for 32/64 bit with small/large code models.
1.For 32bit large code model on AIX, there are newly added pseudo opcode LWZtocL & ADDIStocHA32, the support of which on MC layer will be
provided by future patches.
2.The default code model on AIX should be small code model.
3.Since AIX does not have medium code model, "report_fatal_error" when users specify it.
David Green [Tue, 13 Aug 2019 18:12:08 +0000 (18:12 +0000)]
[ARM] Add MVE beats vector cost model
The MVE architecture has the idea of "beats", where a vector instruction can be
executed over several ticks of the architecture. This adds a similar system
into the Arm backend cost model, multiplying the cost of all vector
instructions by a factor.
This factor essentially becomes the expected difference between scalar code
and vector code, on average. MVE Vector instructions can also overlap so the a
true cost of them is often lower. But equally scalar instructions can in some
situations be dual issued, or have other optimisations such as unrolling or
make use of dsp instructions. The default is chosen as 2. This should not
prevent vectorisation is a most cases (as the vector instructions will still be
doing at least 4 times the work), but it will help prevent over vectorising in
cases where the benefits are less likely.
This adds things so far to the obvious places in ARMTargetTransformInfo, and
updates a few related costs like not treating float instructions as cost 2 just
because they are floats.
Wenlei He [Tue, 13 Aug 2019 17:56:08 +0000 (17:56 +0000)]
[llvm-profdata] Profile dump for compact binary format
Summary: Fix "llvm-profdata show" so it can work with compact binary format profile. The change is to mark all functions "used" so SampleProfileReaderCompactBinary::read will read in all profiles available for dumping. The function names will be MD5 hash for compact binary format.
Steven Wu [Tue, 13 Aug 2019 17:52:21 +0000 (17:52 +0000)]
[AutoUpgrader] Make ArcRuntime Autoupgrader more conservative
Summary:
This is a tweak to r368311 and r368646 which auto upgrades the calls to
objc runtime functions to objc runtime intrinsics, in order to make sure
that the auto upgrader does not trigger with up-to-date bitcode.
It is possible for bitcode that is up-to-date to contain direct calls to
objc runtime function and those are not inserted by compiler as part of
ARC and they should not be upgraded. Now auto upgrader only triggers as
when the old style of ARC marker is used so it is guaranteed that it
won't trigger on update-to-date bitcode.
This also means it won't do this upgrade for bitcode from llvm-8 and
llvm-9, which preserves the behavior of those releases. Ideally they
should be upgraded as well but it is more important to make sure
AutoUpgrader will not trigger on up-to-date bitcode.
Use isGuaranteedToTransferExecutionToSuccessor() instead of
isSafeToSpeculativelyExecute() when seeing whether we can propagate
the information in an assume backwards in isValidAssumeForContext().
The latter is more general - it also allows arbitrary loads/stores -
and is also the condition we want: if our assume is guaranteed to
execute, its condition not holding would be UB.
David Tenty [Tue, 13 Aug 2019 17:04:51 +0000 (17:04 +0000)]
[NFC][AIX] Use assert instead of llvm_unreachable
Addresses post-commit comments on https://reviews.llvm.org/D64825. Use
assert instead of llvm_unreachable to check if invalid csect types are being
generated. Use report_fatal_error on unimplemented XCOFF features.
An incorrect verification error revealed that the list of type tags was
incomplete. This patch adds the missing types by adding a tag kind to
the Dwarf.def file, which is used by the `isType` function.
A test was added for the original verification error.
Momchil Velikov [Tue, 13 Aug 2019 16:12:46 +0000 (16:12 +0000)]
[ARM] Fix encoding of APSR in CLRM instruction
The APSR is encoded by setting bit 15 in the register list of the CLRM
instruction (cf. https://static.docs.arm.com/ddi0553/bh/DDI0553B_h_armv8m_arm.pdf).
Lang Hames [Tue, 13 Aug 2019 16:05:18 +0000 (16:05 +0000)]
[ORC] Refactor definition-generation, add a generator for static libraries.
This patch replaces the JITDylib::DefinitionGenerator typedef with a class of
the same name, and adds support for attaching a sequence of DefinitionGeneration
objects to a JITDylib.
This patch also adds a new definition generator,
StaticLibraryDefinitionGenerator, that can be used to add symbols fom a static
library to a JITDylib. An object from the static library will be added (via
a supplied ObjectLayer reference) whenever a symbol from that object is
referenced.
To enable testing, lli is updated to add support for the --extra-archive option
when running in -jit-kind=orc-lazy mode.
Matt Arsenault [Tue, 13 Aug 2019 15:34:38 +0000 (15:34 +0000)]
GlobalISel: Change representation of shuffle masks
Currently shufflemasks get emitted as any other constant, and you end
up with a bunch of virtual registers of G_CONSTANT with a
G_BUILD_VECTOR. The AArch64 selector then asserts on anything that
doesn't fit this pattern. This isn't an ideal representation, and
should avoid legalization and have fewer opportunities for a
representational error.
Rather than invent a new shuffle mask operand type, similar to what
ShuffleVectorSDNode does, just track the original IR Constant mask
operand. I don't completely like the idea of adding another link to
the IR, but MIR is already quite dependent on IR constants already,
and this will allow sharing the shuffle mask utility functions with
the IR.
Roman Lebedev [Tue, 13 Aug 2019 14:57:37 +0000 (14:57 +0000)]
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case)
Summary:
This implements an optimization described in Hacker's Delight 10-17:
when `C` is constant, the result of `X % C == 0` can be computed
more cheaply without actually calculating the remainder.
The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479.
One huge caveat: this signed case is only valid for positive divisors.
While we can freely negate negative divisors, we can't negate `INT_MIN`,
so for now if `INT_MIN` is encountered, we bailout.
As a follow-up, it should be possible to handle that more gracefully
via extra `and`+`setcc`+`select`.
This passes llvm's test-suite, and from cursory(!) cross-examination
the folds (the assembly) match those of GCC, and manual checking via alive
did not reveal any issues (other than the `INT_MIN` case)
The comment initially matched the code, but the code was incorrect
and was fixed after the initial revert back back when it was introduced,
but the comment was never updated.
Summary:
Given a pattern like:
```
%old_cmp1 = icmp slt i32 %x, C2
%old_replacement = select i1 %old_cmp1, i32 %target_low, i32 %target_high
%old_x_offseted = add i32 %x, C1
%old_cmp0 = icmp ult i32 %old_x_offseted, C0
%r = select i1 %old_cmp0, i32 %x, i32 %old_replacement
```
it can be rewritten as more canonical pattern:
```
%new_cmp1 = icmp slt i32 %x, -C1
%new_cmp2 = icmp sge i32 %x, C0-C1
%new_clamped_low = select i1 %new_cmp1, i32 %target_low, i32 %x
%r = select i1 %new_cmp2, i32 %target_high, i32 %new_clamped_low
```
Iff `-C1 s<= C2 s<= C0-C1`
Also, `ULT` predicate can also be `UGE`; or `UGT` iff `C0 != -1` (+invert result)
Also, `SLT` predicate can also be `SGE`; or `SGT` iff `C2 != INT_MAX` (+invert result)
If `C1 == 0`, then all 3 instructions must be one-use; else at most either `%old_cmp1` or `%old_x_offseted` can have extra uses.
NOTE: if we could reuse `%old_cmp1` as one of the comparisons we'll have to build, this could be less limiting.
So there are two icmp's, each one with 3 predicate variants, so there are 9 fold variants:
Roman Lebedev [Tue, 13 Aug 2019 12:49:06 +0000 (12:49 +0000)]
[InstCombine] foldXorOfICmps(): don't give up on non-single-use ICmp's if all users are freely invertible
Summary:
This is rather unconventional..
As the comment there says, we don't have much folds for xor-of-icmps,
we try to turn them into an and-of-icmps, for which we have plenty of folds.
But if the ICmp we need to invert is not single-use - we give up.
As discussed in https://reviews.llvm.org/D65148#1603922,
we may have a non-canonical CLAMP pattern, with bit match and
select-of-threshold that we'll potentially clamp.
As it can be seen in `canonicalize-clamp-with-select-of-constant-threshold-pattern.ll`,
out of all 8 variations of the pattern, only two are **not** canonicalized into
the variant with and+icmp instead of bit math.
The reason is because the ICmp we need to invert is not single-use - we give up.
We indeed can't perform this fold at will, the general rule is that
we should not increase instruction count in InstCombine,
But we wouldn't end up increasing instruction count if we can adapt every other
user to the inverted value. This way the `not` we create **will** get folded,
and in the end the instruction count did not increase.
For that, of course, we need to look at the users of a Value,
which is again rather unconventional for InstCombine :S
Thus i'm proposing to be a little bit more insistive in `foldXorOfICmps()`.
The alternatives would be to not create that `not`, but add duplicate code to
manually invert all users; or to add some even less general combine to handle
some more specific pattern[s].
projects / llvm / log