Johannes Doerfert [Mon, 12 Dec 2022 23:17:27 +0000 (15:17 -0800)]
[Attributor] Introduce AA[Intra/Inter]Reachability
We had two AAs for reachability but it was very cumbersome to extend
them. We also had some fallback to use LLVM-core mechanisms and cache
the result. The new design shares the query code and interface nicely
between AAIntraFnReachability and AAInterFnReachability.
As part of the rewrite we also added the ExclusionSet to the queries.
Johannes Doerfert [Wed, 14 Dec 2022 03:36:45 +0000 (19:36 -0800)]
[OpenMP][FIX] Remove accidental and somewhat random change
Chuanqi Xu [Wed, 14 Dec 2022 03:28:06 +0000 (11:28 +0800)]
[NFC] [C++20] [Modules] Add more lambda tests
Add more lambda tests in modules. This is useful when we started to work
on https://github.com/llvm/llvm-project/issues/57222.
Tulio Magno Quites Machado Filho [Wed, 14 Dec 2022 02:59:20 +0000 (10:59 +0800)]
Warn about unsupported ibmlongdouble
When -mabi=ieeelongdouble is enabled by default, libc++ does not support
-mabi=ibmlongdouble.
Reviewed By: qiucf
Differential Revision: https://reviews.llvm.org/D139450
Johannes Doerfert [Tue, 13 Dec 2022 22:35:50 +0000 (14:35 -0800)]
[OpenMP][FIX] Remove unsound reasoning about written to values
Even if a value is for sure written we need to visit the call sites as
they might end up inside the function that reads and writes the value.
In a follow up we can introduce correct reasoning to avoid the backwards
traversal in this case and instead check if any call site between the
write and the read might reach a potential write we want to exclude.
Johannes Doerfert [Tue, 13 Dec 2022 19:11:55 +0000 (11:11 -0800)]
[NFC] Rerun update test checks on Attributor and OpenMP-Opt tests
Johannes Doerfert [Tue, 13 Dec 2022 18:11:18 +0000 (10:11 -0800)]
[OpenMP][NFCI] Remove effectively dead code in clang and the runtime
Differential Revision: https://reviews.llvm.org/D136903
jacquesguan [Mon, 12 Dec 2022 08:58:04 +0000 (16:58 +0800)]
[mlir][Vector] Use llvm::zip to avoid assertion failed.
This patch fixes the issue https://github.com/llvm/llvm-project/issues/59455.
We could omit the un-changed dimensions in offsets and sizes, so llvm::zip_equal would fail in this case.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D139815
Mike Urbach [Wed, 7 Dec 2022 02:30:56 +0000 (19:30 -0700)]
[mlir][Python] Add a simple PyOpOperand iterator for PyValue uses.
This adds a simple PyOpOperand based on MlirOpOperand, which can has
properties for the owner op and operation number.
This also adds a PyOpOperandIterator that defines methods for __iter__
and __next__ so PyOpOperands can be iterated over using the the
MlirOpOperand C API.
Finally, a uses psuedo-container is added to PyValue so the uses can
generically be iterated.
Depends on D139596
Reviewed By: stellaraccident, jdd
Differential Revision: https://reviews.llvm.org/D139597
gonglingqin [Wed, 14 Dec 2022 01:54:26 +0000 (09:54 +0800)]
[OpenMP][Test] Make the output error message consistent with the comment
Modify the error message output of affinity/kmp-affinity.c and
affinity/omp-places.c.
Differential Revision: https://reviews.llvm.org/D139803
Jon Chesterfield [Wed, 14 Dec 2022 02:02:58 +0000 (02:02 +0000)]
[openmp][amdgpu] Let fine grain and kernarg pools differ
Yeting Kuo [Tue, 13 Dec 2022 09:41:46 +0000 (17:41 +0800)]
[VP][RISCV] Add vp.ctpop and RISC-V support.
The patch also adds expandVPCTPOP in TargetLowering to expand VP_CTPOP nodes.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D139920
Rob Suderman [Wed, 14 Dec 2022 01:19:04 +0000 (17:19 -0800)]
[mlir][tosa] Fix padding for tosa.conv2d and tosa.depthwise_conv2d decomposition
Decomposition did not take padding into account when decomposing into fully
connected operation.
Reviewed By: NatashaKnk
Differential Revision: https://reviews.llvm.org/D139500
Kazu Hirata [Wed, 14 Dec 2022 01:36:56 +0000 (17:36 -0800)]
[flang] Fix a warning
This patch fixes:
flang/lib/Lower/PFTBuilder.cpp:1042:6: error: function 'dumpScope'
is not needed and will not be emitted
[-Werror,-Wunneeded-internal-declaration]
Kazu Hirata [Wed, 14 Dec 2022 01:25:00 +0000 (17:25 -0800)]
[IPO] Fix a warning
This patch fixes:
llvm/lib/Transforms/IPO/AttributorAttributes.cpp:1028:21: error:
unused function 'operator<<' [-Werror,-Wunused-function]
Roy Sundahl [Fri, 2 Dec 2022 21:13:19 +0000 (13:13 -0800)]
[ubsan][test] Be more specific than "error:" with implicit-check-not
The string "error:" is too general and reports false positive from system
messages that can occur on stderr when using network devices for testing.
The string "runtime error:" is sufficiently unique for these connection
errors to pass through w/o a false positive.
rdar://
100564373
Reviewed By: thetruestblue
Differential Revision: https://reviews.llvm.org/D139230
Pengxuan Zheng [Tue, 13 Dec 2022 20:35:11 +0000 (12:35 -0800)]
[lld-macho][test][nfc] Update stabs.s to use touch -d instead of -t
The test currently uses `touch -t` (e.g., `env TZ=UTC touch -t
"
197001010000.16"`) to set file timestamps. However, this does not seem to set
the time zone correctly in a singularity container. While this is probably a
bug/limitation of the singularity container, but we can instead use `touch
-d` (e.g., `touch -d "1970-01-01 00:00:16 UTC"`) to achieve the same result
without relying on a fix from singularity. Thoughts?
Reviewed By: int3, #lld-macho
Differential Revision: https://reviews.llvm.org/D139980
Nicole Rabjohn [Wed, 14 Dec 2022 00:54:52 +0000 (19:54 -0500)]
[CMake] Setting the LLVM_TARGET_TRIPLE macro based on the LLVM_DEFAULT_TARGET_TRIPLE
After D137870, LLVM_TARGET_TRIPLE is no longer defined on the runtime
path into compiler-rt. This patch creates a common block of code to set
LLVM_TARGET_TRIPLE equal to the default for both the llvm- and runtime-
paths.
Differential Revision: https://reviews.llvm.org/D138864
Haowei Wu [Wed, 14 Dec 2022 00:38:12 +0000 (16:38 -0800)]
Revert "Reland "[compiler-rt][test] Heed COMPILER_RT_DEBUG when compiling unittests""
This reverts commit
255c3e3dcb06299aa2365f70817322a8a381c351, which
breaks Msan-x86_64-Test.
Xiang Li [Sat, 10 Dec 2022 04:34:27 +0000 (20:34 -0800)]
[NFC] [DirectX backend] copy dxil-dis as exe file on windows
Without .exe in the name on windows, the test not work.
Differential Revision: https://reviews.llvm.org/D139765
Kirill Stoimenov [Tue, 13 Dec 2022 23:03:48 +0000 (23:03 +0000)]
[LSAN][NFC] Eliminated GetThreadRegistryLocked from the LSAN interface to avoid the need to implement it in HWASAN.
Reviewed By: vitalybuka
Differential Revision: https://reviews.llvm.org/D139976
Kazu Hirata [Tue, 13 Dec 2022 23:44:30 +0000 (15:44 -0800)]
[mlir] Fix a warning
This patch fixes:
mlir/lib/Dialect/Linalg/Transforms/DataLayoutPropagation.cpp:52:42:
error: 'None' is deprecated: Use std::nullopt
instead. [-Werror,-Wdeprecated-declarations]
V Donaldson [Tue, 13 Dec 2022 22:55:46 +0000 (14:55 -0800)]
[flang] Fix build failure
Bill Wendling [Tue, 13 Dec 2022 23:06:29 +0000 (15:06 -0800)]
[X86] Don't zero out %eax if both %al and %ah are used
The iterator over super and sub registers doesn't include both 8-bit
registers in its list. So if both registers are used and only one of
them is live on return, then we need to make sure that the other 8-bit
register is also marked as live and not zeroed out.
Reviewed By: nickdesaulniers
Differential Revision: https://reviews.llvm.org/D139679
Koakuma [Tue, 13 Dec 2022 22:58:54 +0000 (17:58 -0500)]
[SPARC] Add GNU syntax extensions of WRPR instruction
Add two GNU extension forms of wrpr:
wrpr %reg, %rd -> wrpr %reg, %g0, %rd
wrpr imm, %rd -> wrpr %g0, imm, %rd
Note: since the semantics of wrpr %rs1, %rs2/imm, %rd is %rd <- (%rs1 xor %rs2/imm),
in case (1) it technically doesn't matter whether the %g0 is placed as the first or
second operand, however, I'm placing it as the second operand to match the encoding
of GNU as.
This fixes the wrpr instruction part of bug #40792
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139535
Cyndy Ishida [Tue, 13 Dec 2022 20:39:11 +0000 (12:39 -0800)]
[llvm][Object] set SF_Hidden flag for MachO files
Reviewed By: pete, ributzka
Differential Revision: https://reviews.llvm.org/D139862
Jason Molenda [Tue, 13 Dec 2022 22:50:52 +0000 (14:50 -0800)]
Skip TestEarlyProcessLaunch.py w/ system debugserver
This test depends on having a new packet supported by debugserver;
skip it until we have a system debugserver installed on the CI bots
with this change.
Roman Lebedev [Tue, 13 Dec 2022 20:53:08 +0000 (23:53 +0300)]
[NFC][X86][Codegen] Fix test coverage for AVX512 i1 mask replication
At least some of the seemingly-bad codegen was because we really need to
consume the replicated i1 mask by the mask-consuming instruction
if we want faithful codegen for the replication shuffle.
But some of the odd codegen is still there.
Richard Smith [Tue, 13 Dec 2022 22:41:12 +0000 (14:41 -0800)]
Don't try to look up a name containing a dependent type.
Template instantiation can create names that are still dependent, such
as `operator T`. Don't assume that they can be looked up immediately,
and instead defer lookup for such names until we know what `T` is.
Vasileios Porpodas [Tue, 13 Dec 2022 21:47:25 +0000 (13:47 -0800)]
[NFC] Cleanup: Remove uses of BasicBlock::getInstList().
This is part of a series of patches that aim at making Function::getInstList() private.
Differential Revision: https://reviews.llvm.org/D139971
Koakuma [Tue, 13 Dec 2022 22:25:42 +0000 (17:25 -0500)]
[SPARC] Mark the %g0 register as constant & use it to materialize zeros
Materialize zeros by copying from %g0, which is now marked as constant.
This makes it possible for some common operations (like integer negation) to be
performed in fewer instructions.
This continues @arichardson's patch at D132561.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D138887
Hanhan Wang [Wed, 7 Dec 2022 23:47:06 +0000 (15:47 -0800)]
[mlir][tensor][linalg] Enhance pack op propagation across generic ops.
Considering the case that generic + pack (with outer_dim_perms), the
truth is that it is equipvelent to generic + pack + transpose. There are
two steps to bubble up the pack op accross the generic op.
Step 1. swap generic + pack -> pack + generic.
In this step, we can bind the packing information to dimensions of
iteration domain. With the information, we can pack the operands with
corresponding data tile sizes; the packed inner dimensions will be
appended to the indexing_maps. Note that the outer dimensions of
indexing maps are not changed at all.
Step 2. Fold the transpose into generic op.
The step two is just updating the indexing map, so we do not have to
handle outer_dim_perms anymore.
There could be step 3 to extract the transpose op out (i.e., generic ->
transpose + generic), then we can fold the transpose into the pack op.
This step is not done in the revision.
Co-authored-by: Lorenzo Chelini <l.chelini@icloud.com>
Reviewed By: chelini
Differential Revision: https://reviews.llvm.org/D139680
Sanjay Patel [Tue, 13 Dec 2022 21:34:35 +0000 (16:34 -0500)]
[ValueTracking] peek through extends in haveNoCommonBitsSet (2nd try)
The 1st try was not clean because a portion of the code diff
made it into the pre-commit patch to add tests. This should
be the same end result without the muddied code diff.
Original commit message:
In cases with matching extends, this allows changing an 'add'
into an 'or' and narrowing the 'or' which then simplifies to
a constant.
In cases with opposite extends, we just convert to an 'or'
currently, but that could be reduced too.
https://alive2.llvm.org/ce/z/fTHzdb
Keith Smiley [Tue, 13 Dec 2022 18:57:24 +0000 (10:57 -0800)]
[lld-macho][test] Add test case for section ordering
This covers the special section ordering handling. This chooses to fake the sections vs creating them organically which is probably a bit easier to maintain.
Differential Revision: https://reviews.llvm.org/D139959
Nuno Lopes [Tue, 13 Dec 2022 21:45:12 +0000 (21:45 +0000)]
[llvm][tests] Don't run -O2 on instcombine test [NFC]
Kirill Stoimenov [Tue, 13 Dec 2022 20:38:32 +0000 (20:38 +0000)]
[HWASAN][NFC] Added implementation for the most of the allocator related functions from the LSAN interface.
Reviewed By: vitalybuka
Differential Revision: https://reviews.llvm.org/D139964
bixia1 [Tue, 13 Dec 2022 19:38:32 +0000 (11:38 -0800)]
[mlir][sparse] Make some integration tests run with vectorization.
Reviewed By: Peiming
Differential Revision: https://reviews.llvm.org/D139887
Keith Smiley [Tue, 13 Dec 2022 04:46:51 +0000 (20:46 -0800)]
[lld-macho] Fix __objc_stubs ordering
In the case of large binaries previously you could end up getting
relocation failures. This mirrors ld64's ordering of these sections. I'm
not sure this solves all cases but it should help in some.
Fixes https://github.com/llvm/llvm-project/issues/58298
Differential Revision: https://reviews.llvm.org/D139909
Sanjay Patel [Tue, 13 Dec 2022 21:18:52 +0000 (16:18 -0500)]
[InstCombine] add tests for add-of-extends; NFC (2nd try)
The 1st attempt (
c8cba0bc4a8c9f ) accidentally picked up a code diff.
Sanjay Patel [Tue, 13 Dec 2022 21:11:33 +0000 (16:11 -0500)]
Revert "[InstCombine] add tests for add-of-extends; NFC"
This reverts commit
c8cba0bc4a8c9f4f3f10e17f601ed924dfb82bef.
An unintended code change snuck into this (was supposed to just add tests).
Sanjay Patel [Tue, 13 Dec 2022 21:09:50 +0000 (16:09 -0500)]
Revert "[ValueTracking] peek through extends in haveNoCommonBitsSet"
This reverts commit
7520d187cf0dedcf5085f71bc1a5472c75cc8dbb.
The pre-commit patch that was supposed to NFC add tests for this
patch picked up some code that should not be here. I'll clean up
and re-commit.
V Donaldson [Mon, 12 Dec 2022 22:20:06 +0000 (14:20 -0800)]
[flang] Submodules
A submodule is a program unit that may contain the implementions of procedures
declared in an ancestor module or submodule.
Processing for the equivalence groups and variables declared in a submodule
scope is similar to existing processing for the equivalence groups and
variables in module and procedure scopes. However, module and procedure scopes
are tied directly to code in the Pre-FIR Tree (PFT), whereas processing for a
submodule must have access to an ancestor module scope that is guaranteed
to be present in a .mod file, but is not guaranteed to be in the PFT. This
difference is accommodated by tying processing directly to a front end scope.
Function scopes that can be processed on the fly are done that way; the
resulting variable information is never stored. Module and submodule scopes
whose symbol information may be needed during lowering of any number of module
procedures are instead cached on first use, and reused as needed.
These changes are a direct extension of current code. All module and submodule
variables in scope are processed, whether referenced or not. A possible
alternative would be to instead process symbols only when first used. While
this could ultimately be beneficial, such an approach must account for the
presence of equivalence groups. That information is not currently available
for on-the-fly variable processing.
Some additional changes are needed to include submodules in places where
modules must be considered, and to include separate module procedures in
places where other subprogram variants are considered. There is also a fix
for a bug involving the use of variables in an equivalence group in a
namelist group, which also involves scope processing code.
Sanjay Patel [Tue, 13 Dec 2022 20:16:47 +0000 (15:16 -0500)]
[ValueTracking] peek through extends in haveNoCommonBitsSet
In cases with matching extends, this allows changing an 'add'
into an 'or' and narrowing the 'or' which then simplifies to
a constant.
In cases with opposite extends, we just convert to an 'or'
currently, but that could be reduced too.
https://alive2.llvm.org/ce/z/fTHzdb
Sanjay Patel [Tue, 13 Dec 2022 18:47:33 +0000 (13:47 -0500)]
[InstCombine] add tests for add-of-extends; NFC
Michael Jones [Mon, 7 Nov 2022 17:47:58 +0000 (09:47 -0800)]
[libc] add scanf int conversions
This patch adds the integer conversions %d/i/u/o/x/X to scanf as well as
unit tests to check their correctness.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D138880
Michael Jones [Fri, 18 Nov 2022 22:19:30 +0000 (14:19 -0800)]
[libc] fix scanf error handling
Scanf is supposed to return EOF when it fails to make any conversions
and there is an input failure. Previously it would return EOF on a
matching failure, which may be an input failure but can also be a
parsing error.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D139891
chenglin.bi [Tue, 13 Dec 2022 20:42:09 +0000 (04:42 +0800)]
[InstCombine] Fold logic-and/logic-or by distributive laws
X && Z || Y && Z --> (X || Y) && Z
https://alive2.llvm.org/ce/z/nM6kZb
(X || Z) && (Y || Z) --> (X && Y) || Z
https://alive2.llvm.org/ce/z/_EWLRR
Fix: https://github.com/llvm/llvm-project/issues/53861
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D139408
serge-sans-paille [Tue, 13 Dec 2022 13:54:11 +0000 (14:54 +0100)]
Use different export file when linking with dylib
This fixes compilation of clang with lld, as reported in
https://reviews.llvm.org/D135402.
Basically some symbols are not available when not linking statically.
Differential Revision: https://reviews.llvm.org/D139932
bixia1 [Mon, 12 Dec 2022 21:39:55 +0000 (13:39 -0800)]
[mlir][crunner] Add support for invoking std::sort.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D139880
Cabrera, Anthony [Tue, 13 Dec 2022 19:48:25 +0000 (14:48 -0500)]
[flang] emit warning when encountering a non-variable actual argument when its corresponding dummy argument expects a VOLATILE variable
This patch implements @klausler's suggestion in `llvm-project` [issue #58973](https://github.com/llvm/llvm-project/issues/58973); encountering the `VOLATILE` attribute should produce a __warning__, not a __fatal error__.
When tested on the following Fortran program `snem0601_012_.f90`:
```fortran
module mod
contains
subroutine sub(m6,error)
integer,intent(inout) :: error
integer,volatile :: m6
if (any ((/m6/).ne.(/6/))) &
& then
error = 1
end if
end subroutine
end module
program fe1nvol12
use mod
integer :: error = 0
call sub(6,error)
if (error .ne. 0) then
print *,'NG: snem0601_012'
end if
print *,'pass: snem0601_012'
end program fe1nvol12
```
the following output is produced:
```bash
$ flang-new -fc1 snem0601_012_.f90
/noback/93u/Sandbox/issue_58973_volatile_dummy_arg/snem0601_012_.f90:21:12: warning: actual argument associated with VOLATILE dummy argument 'm6=' is not a variable
call sub(6,error)
^
```
Reviewed By: clementval, klausler
Differential Revision: https://reviews.llvm.org/D139134
Jordan Rupprecht [Tue, 13 Dec 2022 20:20:21 +0000 (12:20 -0800)]
[bazel] Add __support_compiler_features dep in more places
Philip Reames [Tue, 13 Dec 2022 20:01:32 +0000 (12:01 -0800)]
[RISCV] Reuse VL (if non-zero) when building single element vector for start of reduction chain
This is an alternative patch on a path to D137530.
The basic problem being tackled here is that we need to place a scalar into lane 0 of a vector register before our reduction instructions. Since we only care about lane 0 of the vector, we can use any VL >= 1 provided that the total amount of work performed matches the work performed for a VL=1.
This change does not contain the logic from D137530 to perform the insert at the original VT, and then extract down to LMUL1. That turns out to be a good choice, as discussion in this review has indicated there are issues around LMUL2 and above with our representation of vmv.s.x. We'd also need to be careful with the splat logic for the same reasons.
The only potentially concerning codegen change I spot here is that we stop using a broadcast load (for VL=1) and instead do a scalar load and insert. I think this is probably reasonable; if reviewers disagree, I can investigate using a broadcast load which writes to the undef lanes. If we want to do that, we should do it for VECTOR_INSERT_ELT as well, so that'll end up as it's own patch series.
Differential Revision: https://reviews.llvm.org/D139656
Slava Zakharin [Tue, 13 Dec 2022 17:52:08 +0000 (09:52 -0800)]
[mlir][math] Added math::FPowI conversion to calls of outlined implementations.
Power functions are implemented as linkonce_odr scalar functions
for FPowI operations met in a module.
Vector form of FPowI is linearized into a sequence of calls
of the scalar functions.
Option {min-width-of-fpowi-exponent} controls which FPowI operations
are converted by MathToFuncs: if the width of the exponent's integer
type is less than the specified value, then the operation is not converted.
Flang will specify {min-width-of-fpowi-exponent=33} to make sure that
math::FPowI operations with exponent wider than 32 bits will be converted
by MathToFuncs, and operations with more narrow exponent will be left
for MathToLLVM to convert them to LLVM::PowIOp.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D139804
Josh Stone [Sat, 10 Dec 2022 01:06:04 +0000 (17:06 -0800)]
[X86] Fix cmp order in probing BuildStackAlignAND
Due to reversed arguments, the loop start was almost always skipping the
whole loop, since FinalStackProbed is probably less than StackPtr for
large alignments. The intent was to skip the loop if the first sub on
StackPtr made it less than FinalStackProbed already, so flip it.
Reviewed By: serge-sans-paille
Differential Revision: https://reviews.llvm.org/D139756
Craig Topper [Tue, 13 Dec 2022 20:02:10 +0000 (12:02 -0800)]
[InstCombine] Support vectors in simplifyIntToPtrRoundTripCast.
This code compares getPointerTypeSizeInBits and getTypeSizeInBits.
getPointerTypeSizeInBits contains a call to getScalarType while
getTypeSizeInBits does not. This makes the code incorrect for vectors.
For scalable vectors this caused a warning about a scalable TypeSize
being converted to unsigned.
Switch to DL.getTypeSizeInBits for the pointers too. This should
work since inttoptr/ptrtoint can't change the number of elements.
This was suggested by @nikic in D139911.
Fixes PR59480.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D139962
Vasileios Porpodas [Tue, 13 Dec 2022 01:03:08 +0000 (17:03 -0800)]
[NFC] Cleanup: Remove Function::getBasicBlockList() when not required.
This is part of a series of patches that aim at making Function::getBasicBlockList() private.
Differential Revision: https://reviews.llvm.org/D139910
Jason Molenda [Tue, 13 Dec 2022 19:37:56 +0000 (11:37 -0800)]
Launch state discoverable in Darwin, use for SafeToCallFunctions
The dynamic linker on Darwin, dyld, can provide status of
the process state for a few significant points early on,
most importantly, when libSystem has been initialized and it
is safe to call functions behind the scenes. Pipe this
information up from debugserver to DynamicLoaderMacOS, for
the DynamicLoader::IsFullyInitialized() method, then have
Thread::SafeToCallFunctions use this information. Finally,
for the two utility functions in the AppleObjCRuntimeV2
LanguageRuntime plugin that I was fixing, call this method
before running our utility functions to collect the list of
objc classes registered in the runtime.
User expressions will still be allowed to run any time -
we assume the user knows what they are doing - but these
two additional utility functions that they are unaware of
will be limited by this state.
Differential Revision: https://reviews.llvm.org/D139054
rdar://
102436092
can probably make function calls.
Jakub Kuderski [Tue, 13 Dec 2022 19:33:22 +0000 (14:33 -0500)]
Reland "[mlir][arith] Add canonicalization patterns for 'mul*i_extended'"
- Add a fold for `mulsi_extended(x, 1)`
- Add folds to demote wide integer multiplication to `mul*i_extended` when the result is shifted
and truncated: `trunci(shrui(mul(*ext(x), *ext(y)), c)) -> mul*i_extended(x, y)`
Reviewed By: Mogball, jpienaar
Differential Revision: https://reviews.llvm.org/D139778
Philip Reames [Tue, 13 Dec 2022 19:09:06 +0000 (11:09 -0800)]
[RISCV] Use make_range instead of iterator_range for code from
8e6c3094
Jordan fixed this once in 4f9d069, but using make_range is more idiomatic than my accidental iterator_range usage, even with the template type to fix the warning.
Rahman Lavaee [Tue, 13 Dec 2022 17:52:05 +0000 (09:52 -0800)]
Revert "[Propeller] Use Fixed MBB ID instead of volatile MachineBasicBlock::Number."
This reverts commit
6015a045d768feab3bae9ad9c0c81e118df8b04a.
Differential Revision: https://reviews.llvm.org/D139952
Dave Lee [Fri, 9 Dec 2022 03:53:33 +0000 (19:53 -0800)]
[lldb] Remove basestring references in examples (NFC)
`basestring` is Python 2 only.
Mike Urbach [Wed, 7 Dec 2022 20:03:24 +0000 (13:03 -0700)]
[mlir][Python] Add `__hash__` implementation for Block.
This allows us to hash Blocks and use them in sets or parts of larger
hashable objects. The implementation is the same as other core IR
constructs: the C API object's pointer is hashed.
Differential Revision: https://reviews.llvm.org/D139599
Jordan Rupprecht [Tue, 13 Dec 2022 18:57:37 +0000 (10:57 -0800)]
Jordan Rupprecht [Tue, 13 Dec 2022 18:50:20 +0000 (10:50 -0800)]
[NFC] Specify template type to fix -Wctad-qmaybe-unsupported
Steven Wu [Tue, 13 Dec 2022 18:42:04 +0000 (10:42 -0800)]
Workaround an assertion failure during module build
After the change in https://reviews.llvm.org/D131858, clang cannot
bootstrap itself with modules due to assertion failure:
(lvaluePath->getType() == elemTy && "Unexpected type reference!")
Workaround the assertion by converting some of the includes into forward
declares.
Reviewed By: arphaman
Differential Revision: https://reviews.llvm.org/D139956
Jakub Kuderski [Tue, 13 Dec 2022 18:40:18 +0000 (13:40 -0500)]
Revert "[mlir][sparse] Make some integration tests run with vectorization."
This reverts commit
2d7e3ec6b5e3d99de6ce79049d4bef96213109ee.
This broke buildbots [1] and I can also reproduce this locally.
[1] https://lab.llvm.org/buildbot#builders/61/builds/36953
Roman Lebedev [Tue, 13 Dec 2022 18:19:24 +0000 (21:19 +0300)]
[NFC][MC] `MCInst`: `Operands` small size optimization: store 10, not 8, inline `MCOperand`
This improves the torture test of
```
./bin/llvm-exegesis -mcpu=znver3 -mode=inverse_throughput --opcode-index=-1 --benchmarks-file=/dev/null --dump-object-to-disk=0 --measurements-print-progress --skip-measurements
```
from ~2m16s to ~2min07s, and has the following effect on memory:
```
heaptrack stats:
allocations:
100828624 ->
77362343 (-23.2%)
leaked allocations: 1128
temporary allocations:
24911300 -> 1576308 (-93.7%) !!!
peak heap memory consumption:
78.2MB after 02.121s -> 76.4MB after 01.985s (-2.3%)
peak RSS (including heaptrack overhead):
193.4MB -> 192.6MB (-0.4%)
```
The reasoning is that having more Operands than the SSO is costly,
because we go to global allocator, but having larger SSO is fine,
even if it's not always needed, because MCInst is hopefully pool-allocated.
I'm not sure who is the code owner of this component.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D139882
Roman Lebedev [Tue, 13 Dec 2022 17:09:21 +0000 (20:09 +0300)]
[NFC][Costmodel][X86] Replication shuffle: AVX512F can promote i1 to i32.
As the added codegen test coverage shows,
there isn't that much difference between AVX512DQI and
baseline AVX512F codegen, DQI added `vpmovm2d`/`vpmovd2m`,
but with just the Foundation we can use `vpternlogd`/`vptestmd`
to do the same.
Roman Lebedev [Tue, 13 Dec 2022 17:57:56 +0000 (20:57 +0300)]
[x86][Costmodel] AVX512VL: add missing costs for v8 i1<->i32 casts
This would come up as a regression in the follow-up Replication-of-i1 patch.
https://godbolt.org/z/fxr9Mzssr
Roman Lebedev [Tue, 13 Dec 2022 16:36:36 +0000 (19:36 +0300)]
[NFC][X86][Codegen] Add codegen test coverage for i1 mask replication (AVX512 only)
Apparently i didn't add it when adding cost model coverage?
Joshua Cao [Tue, 13 Dec 2022 17:37:18 +0000 (09:37 -0800)]
[LoopFusion] sink second loop PHIs
Fixes https://github.com/llvm/llvm-project/issues/59023
PHI nodes that are in the second loop only have the first loop as its
predecessor. These PHI nodes should be sunk to the end of the fused
loop. If the second loop uses the PHI, then the loops cannot be fused.
I don't think this should happen in typical compilation workflows.
The PHI will be in a dedicated exit block of the first loop following
LCSSA transformations.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D139812
Jakub Kuderski [Tue, 13 Dec 2022 18:04:37 +0000 (13:04 -0500)]
[mlir][arith] Clean up arith canonicalization td rules
- Use `ConstantLikeMatcher` instead of matching `Arith_ConstantOp`
- Reflow some defs to fit the line limit
Suggested by @jpienaar and @Mogball in https://reviews.llvm.org/D139778.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D139941
bixia1 [Mon, 12 Dec 2022 23:15:46 +0000 (15:15 -0800)]
[mlir][sparse] Make some integration tests run with vectorization.
Reviewed By: Peiming
Differential Revision: https://reviews.llvm.org/D139887
Alexey Bataev [Thu, 8 Dec 2022 18:04:46 +0000 (10:04 -0800)]
[SLP][NFC]Initial redesign of ShuffleInstructionBuilder, NFC.
The patch redesigns ShuffleInstructionBuilder so it could later be used
for reshuffling of the buildvector sequences and vectorized parts of
externally used scalars. Also will allow to generalize cost model for
the gathers/buildvectors.
Part of D110978.
Differential Revision: https://reviews.llvm.org/D139718
Matthias Kramm [Tue, 13 Dec 2022 17:34:58 +0000 (18:34 +0100)]
[mlir] Implement backward dataflow.
This enables interprocedural lifeness analysis, very busy expression
analysis, etc.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D138935
Arthur Eubanks [Tue, 13 Dec 2022 01:16:11 +0000 (17:16 -0800)]
[opt] Do not add verify pass at beginning of pipeline
We've already verified the input module manually in opt so this is redundant.
Reviewed By: bjope
Differential Revision: https://reviews.llvm.org/D139899
Arthur Eubanks [Tue, 13 Dec 2022 01:16:11 +0000 (17:16 -0800)]
[StandardInstrumentations] Handle initial IR before checking if IR is interesting
Otherwise we end up in situations where we print thre initial IR after we've already printed some changed IR.
Reviewed By: jamieschmeiser
Differential Revision: https://reviews.llvm.org/D139898
Craig Topper [Tue, 13 Dec 2022 17:09:42 +0000 (09:09 -0800)]
[RISCV] Use reduction result type for EXTRACT_VECTOR_ELT in lowerReductionSeq.
Remove the call to getSExtOrTrunc.
Reduction ISD nodes produce a scalar result and that result is
allowed to be larger than the vector element type due to type
legalization. This is the same rule we allow for EXTRACT_VECTOR_ELT
for the same reason.
We can copy the result type over from the reduction node to
EXTRACT_VECTOR_ELT.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D139757
Craig Topper [Tue, 13 Dec 2022 17:00:45 +0000 (09:00 -0800)]
[RISCV] Add support for call returns to RISCVSExtWRemoval.
The implementation is inspired by code from PPCInstrInfo.
We look for a copy from X10(A0) preceded by an ADJCALLSTACKUP and
a PseudoCall. We use the PseudoCall to look up the IR function being
called to find it's return attributes.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D139808
Philip Reames [Tue, 13 Dec 2022 17:02:01 +0000 (09:02 -0800)]
[RISCV] Use lowerScalarInsert when folding op into reduction [nfc]
This doesn't cause any functional change since this is being applied to a insert generated by the same routine. This is mostly about consolidating the logic for vmv.s.x into one place to simplify future changes.
Philip Reames [Tue, 13 Dec 2022 16:47:49 +0000 (08:47 -0800)]
[RISCV] Use lowerScalarInsert in lowerReductionSeq [nfc]
Use the newly introduced helper routine. At the moment, this generates the same code (at this call site!) since LMUL is restricted to LMUL1 or less, and VL is hard coded to 1. In a future patch, I will loosen the second part.
Kirill Stoimenov [Tue, 6 Dec 2022 22:26:41 +0000 (22:26 +0000)]
[HWASAN] Implemented LsanMetadata in HWASAN.
Please note that that this code is not used. It will be connected at some point when the rest of the code is ready.
Reviewed By: vitalybuka
Differential Revision: https://reviews.llvm.org/D139464
Alex Brachet [Tue, 13 Dec 2022 17:01:07 +0000 (17:01 +0000)]
[llvm-driver] Just use argv[0]'s filename for finding tool
Usually we want the stem of argv[0] so something like clang-15
will correctly be identified as clang. For lld however, ld.lld
or ld-link would have a stem of just ld, so we also want to
use the full filename. The bug previously was that we were using
all of argv[0] so if you use a full path that happens to include a tool
name then that could be found first. This was the case in 2 stage
build where the binaries are stored in "tools/clang/stage2-bins/"
so some tools would end up as clang and not their intended tool.
Sameer Sahasrabuddhe [Mon, 12 Dec 2022 08:06:45 +0000 (13:36 +0530)]
[AAPointerInfo] track multiple constant offsets for each use
An expression of the form `gep(base, select(pred, const1, const2))` can result
in a set of offsets instead of just one. PointerInfo can now track these sets
instead of conservatively modeling them as Unknown. In general, AAPointerInfo
now uses AAPotentialConstantValues to examine the operands of the GEP.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D138646
Alex Brachet [Tue, 13 Dec 2022 16:56:08 +0000 (16:56 +0000)]
[CMake] Add missing tablegen deps
LLVM GN Syncbot [Tue, 13 Dec 2022 16:27:58 +0000 (16:27 +0000)]
[gn build] Port
233ca84a25a7
Matt Arsenault [Wed, 30 Nov 2022 04:27:26 +0000 (23:27 -0500)]
InstCombine: Add baseline tests for and/or of fcmp to class combine
This is for patterns like !isfinite || zero.
Jakub Kuderski [Tue, 13 Dec 2022 16:05:47 +0000 (11:05 -0500)]
Revert "[mlir][arith] Add canonicalization patterns for 'mul*i_extended'"
This reverts commit
834c17f618ce87b14446e42250d924b8d5f01abe.
Revert to properly address post-commit comments by @jpienaar
under https://reviews.llvm.org/D139778.
Pavel Kopyl [Tue, 13 Dec 2022 15:31:17 +0000 (18:31 +0300)]
[NVPTX] Backend support for variadic functions
This patch adds lowering for function calls with variadic number of
arguments as well as enables support for the following
instructions/intrinsics:
- va_arg
- va_start
- va_end
- va_copy
Note that this patch doesn't intent to include clang's support for
variadic functions for CUDA.
According to the docs:
PTX version 6.0 supports passing unsized array parameter to a
function which can be used to implement variadic functions. [0]
The last parameter in the parameter list may be a .param array of
type .b8 with no size specified. It is used to pass an arbitrary
number of parameters to the function packed into a single array
object.
When calling a function with such an unsized last argument, the last
argument may be omitted from the call instruction if no parameter is
passed through it. Accesses to this array parameter must be within
the bounds of the array. The result of an access is undefined if no
array was passed, or if the access was outside the bounds of the
actual array being passed. [1]
Note that aggregates passed by value as variadic arguments are not
currently supported.
[0] https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#variadic-functions
[1] https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#kernel-and-function-directives-func
Differential Revision: https://reviews.llvm.org/D138531
Guillaume Chatelet [Tue, 13 Dec 2022 16:05:35 +0000 (16:05 +0000)]
[libc][bazel] Add missing dependency in string_memory_utils
Nikita Popov [Tue, 13 Dec 2022 15:56:59 +0000 (16:56 +0100)]
[ThinLTO] Convert tests to opaque pointers (NFC)
Philip Reames [Tue, 13 Dec 2022 15:54:46 +0000 (07:54 -0800)]
[RISCV] Use vmv.v.i for insertion into lane 0 of undef vector when profitable
If we're initializing lane 0 of an undef vector, we can optionally write to other lanes of the vector. Doing so may require additional work, so we don't want to e.g. always use a splat. However, since we don't have an immediate form of vmv.s.x it's useful to use a vmv.v.i if the work required is expected to be equal in practice. We restrict this to when LMUL <= 1 to a) prevent doing additional work at higher LMULs, and b) avoid overconstraining the register allocator.
At the moment, the new utility is only used by one case in INSERT_VECTOR_ELT lowering. My expectation is that we will reuse this in a couple other places, but each of those deserve individual review.
This change is inspired by D137530, but is not directly related to it. I vaguely remember we discussed the tradeoffs of using vmv.v.i in another recent review, but couldn't find it.
Differential Revision: https://reviews.llvm.org/D139648
Philip Reames [Tue, 13 Dec 2022 15:47:53 +0000 (07:47 -0800)]
[RISCV][InsertVSETVLI] Reverse traversal order of block in post pass [nfc]
his unblocks a following change to be more sophisticated during post pass rewriting.
Review wise, I basically just want a second set of eyes. This change should be straight forward, but since it took me an embarrassing number of attempts to get make check to pass. Let's make sure I'm not missing yet another cornercase.
Differential Revision: https://reviews.llvm.org/D139877
Matt Arsenault [Thu, 17 Nov 2022 06:12:22 +0000 (22:12 -0800)]
InstCombine: Fold logic of fp_classes together
Move logical operators on pairs of llvm.is.fpclass on the same value
into the test mask of a single is_fpclass.
or (class x, mask0), (class x, mask1) -> class x, (mask0 | mask1)
and (class x, mask0), (class x, mask1) -> class x, (mask0 & mask1)
xor (class x, mask0), (class x, mask1) -> class x, (mask0 ^ mask1)
The and/or cases should appear frequently in the builtin math
libraries; haven't seen the xor case but handle it for completeness.
Matt Arsenault [Thu, 1 Dec 2022 15:30:36 +0000 (10:30 -0500)]
InstCombine: Perform simpler test of APFloat first
Jakub Kuderski [Tue, 13 Dec 2022 15:49:14 +0000 (10:49 -0500)]
[mlir][arith] Add canonicalization patterns for 'mul*i_extended'
- Add a fold for `mulsi_extended(x, 1)`
- Add folds to demote wide integer multiplication to `mul*i_extended` when the result is shifted
and truncated: `trunci(shrui(mul(*ext(x), *ext(y)), c)) -> mul*i_extended(x, y)`
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D139778
Philip Reames [Tue, 13 Dec 2022 15:31:05 +0000 (07:31 -0800)]
[llvm-stress] Support scalable vectors and out of bounds indices to vector ops
This change does two things.
First, it supports scalable vectors in llvm-stress output. There's an off-by-default flag -enable-scalable-vectors which generates a scalable vector with 50% probability when a random vector type is selected. Because of the existing implementation, this tends to skew an entire test towards either scalable or fixed since we most select initial values, not random types.
Second, it selects a random value for the index of an insertelement or extractelement. As a result, out of bounds indices can not appear for fixed lengh vectors. Previously, it chose a random constant which was, by construction, in range.
Pierre van Houtryve [Tue, 13 Dec 2022 15:20:42 +0000 (10:20 -0500)]
[AMDGPU] Add `.workgroup_processor_mode` to v5 MD
Adds Workgroup Processor Mode (WGP) to the HSA Metadata for Code Object v5/GFX10+.
The field is already present as an asm directive and in the compute program resource register but is also needed in the MD.
Reviewed By: kzhuravl
Differential Revision: https://reviews.llvm.org/D139931
Nikita Popov [Tue, 13 Dec 2022 15:22:27 +0000 (16:22 +0100)]
[Linker] Convert some tests to opaque pointers (NFC)
Pierre van Houtryve [Tue, 13 Dec 2022 14:59:20 +0000 (09:59 -0500)]
[AMDGPU] Add bf16 storage support
- [Clang] Declare AMDGPU target as supporting BF16 for storage-only purposes on amdgcn
- Add Sema & CodeGen tests cases.
- Also add cases that D138651 would have covered as this patch replaces it.
- [AMDGPU] Add BF16 storage-only support
- Support legalization/dealing with bf16 operations in DAGIsel.
- bf16 as a type remains illegal and is represented as i16 for storage purposes.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139398