Matt Arsenault [Sat, 20 Mar 2021 16:53:58 +0000 (12:53 -0400)]
AMDGPU/GlobalISel: Enable CSE in pre-legalizer combiner
Simon Pilgrim [Sun, 21 Mar 2021 14:00:59 +0000 (14:00 +0000)]
[DAG] Limit (sext_in_reg (zero_extend_vector_inreg x)) to exact sign extension
As commented by @craig.topper on rG1ba5c550d418, we can't guarantee that we'll be extending zero bits, just sign bit. So, revert to the old code for zero_extend_vector_inreg cases.
Jez Ng [Sun, 21 Mar 2021 05:10:04 +0000 (01:10 -0400)]
[lld-macho][nfc] Format Options.td
Summary: A good chunk of it was mis-indented. Fixed by using the
formatting settings from llvm/utils/vim.
Simon Pilgrim [Sun, 21 Mar 2021 12:22:51 +0000 (12:22 +0000)]
[X86][AVX] ComputeNumSignBitsForTargetNode - add X86ISD::VBROADCAST handling for scalar sources
The target shuffle code handles vector sources, but X86ISD::VBROADCAST can also accept a scalar source for splatting.
Added as an extension to PR49658
Simon Pilgrim [Sun, 21 Mar 2021 12:08:53 +0000 (12:08 +0000)]
[X86] Add 'mulhs' variant of PR49658 test case
David Green [Sun, 21 Mar 2021 12:00:06 +0000 (12:00 +0000)]
[ARM] VINS f16 pattern
This adds an extra pattern for inserting an f16 into a odd vector lane
via an VINS. If the dual-insert-lane pattern does not happen to apply,
this can help with some simple cases.
Differential Revision: https://reviews.llvm.org/D95471
luxufan [Fri, 19 Mar 2021 09:02:28 +0000 (17:02 +0800)]
[RISCV] remove redundant instruction when eliminate frame index
The reason for generating mv a0, a0 instruction is when the stack object offset is large then int<12>. To deal this situation, in the elimintateFrameIndex function, it will
create a virtual register, which needs the register scavenger to scavenge it. If the machine instruction that contains the stack object and the opcode is ADDI(the addi
was generated by frameindexNode), and then this instruction's destination register was the same as the register that was generated by the register scavenger, then the
mv a0, a0 was generated. So to eliminnate this instruction, in the eliminateFrameIndex function, if the instrution opcode is ADDI, then the virtual register can't be created.
Differential Revision: https://reviews.llvm.org/D92479
Simon Pilgrim [Sun, 21 Mar 2021 10:40:57 +0000 (10:40 +0000)]
[X86][AVX] computeKnownBitsForTargetNode - add X86ISD::VBROADCAST handling for scalar sources
The target shuffle code handles vector sources, but X86ISD::VBROADCAST can also accept a scalar source for splatting.
Suggested by @craig.topper on PR49658
Simon Pilgrim [Sun, 21 Mar 2021 10:16:55 +0000 (10:16 +0000)]
[X86] Add PR49658 test case
Simon Pilgrim [Sun, 21 Mar 2021 09:57:20 +0000 (09:57 +0000)]
[X86] computeKnownBitsForTargetNode - add X86ISD::PMULUDQ handling
Reuse the existing KnownBits multiplication code to handle what is effectively a ISD::UMUL_LOHI varient
Fangrui Song [Sun, 21 Mar 2021 07:56:03 +0000 (00:56 -0700)]
[Driver] Linux.cpp: add -internal-isystem lib/../$triple/include
With this change, for `#include <ar.h>`, `clang --target=aarch64-linux-gnu`
will read `/usr/lib/gcc/aarch64-linux-gnu/10/../../../../aarch64-linux-gnu/include/ar.h`
(on Debian gcc->gcc-cross)
instead of `/usr/include/ar.h`. Some glibc headers (e.g. gnu/stubs.h) are different across architectures.
Fangrui Song [Sun, 21 Mar 2021 04:37:49 +0000 (21:37 -0700)]
[Driver] Gnu.cpp: drop an unneeded special rule related to sysroot
Fangrui Song [Sun, 21 Mar 2021 04:32:55 +0000 (21:32 -0700)]
[Driver] Gnu.cpp: drop an unneeded special rule related to sysroot
Seem unnecessary to diverge from GCC here.
Beside, lib/../$OSLibDir can be considered closer to the GCC
installation then the system root. The comment should not apply.
Fangrui Song [Sun, 21 Mar 2021 03:12:45 +0000 (20:12 -0700)]
[Driver] Gnu.cpp: remove unneeded -L detection hack for -mx32
Removing the hack actually improves our compatibility with gcc -mx32.
Fangrui Song [Sun, 21 Mar 2021 01:56:40 +0000 (18:56 -0700)]
[Driver] Gnu.cpp: remove unneeded -L detection for libc++
If clang is installed in the system, the other -L suffice;
otherwise $ccc_install_dir/../lib below suffices.
Fangrui Song [Sun, 21 Mar 2021 01:50:14 +0000 (18:50 -0700)]
[Driver] Gnu.cpp: remove unneeded -L lib/gcc/$triple/$version/../../../$triple
After path resolution, it duplicates a subsequent -L entry. The entry below
(lib/gcc/$triple/$version/../../../../$OSLibDir) usually does not exist (e.g.
Arch Linux; Debian cross gcc). When it exists, it typically just has ld.so (e.g.
Debian native gcc) which cannot cause collision. Removing the -L (similar to
reordering it) is therefore justified.
Craig Topper [Sun, 21 Mar 2021 00:43:30 +0000 (17:43 -0700)]
[RISCV] Add test case to show a case where (mul (and X, 0xffffffff), (and Y, 0xffffffff)) optimization does not improve code.
If the mul add two users, one of which was a sext.w, the mul
would also be selected to a MULW before our pattern runs. This
causes the ANDs to now be used by the already selected MULW and
the mul we still need to select. They are unneeded on the MULW
since MULW only reads the lower bits. So they get selected to
SLLI+SRLI for the MULW use. The use for the
(mul (and X, 0xffffffff), (and Y, 0xffffffff)) manages to reuse
the SLLI.
The end result is increased register pressure and no improvement
to how soon we can start the MULW.
Chris Lattner [Sat, 20 Mar 2021 04:22:15 +0000 (21:22 -0700)]
[Canonicalizer] Process regions top-down instead of bottom up & reuse existing constants.
This reapplies b5d9a3c / https://reviews.llvm.org/D98609 with a one line fix in
processExistingConstants to skip() when erasing a constant we've already seen.
Original commit message:
1) Change the canonicalizer to walk the function in top-down order instead of
bottom-up order. This composes well with the "top down" nature of constant
folding and simplification, reducing iterations and re-evaluation of ops in
simple cases.
2) Explicitly enter existing constants into the OperationFolder table before
canonicalizing. Previously we would "constant fold" them and rematerialize
them, wastefully recreating a bunch fo constants, which lead to pointless
memory traffic.
Both changes together provide a 33% speedup for canonicalize on some mid-size
CIRCT examples.
One artifact of this change is that the constants generated in normal pattern
application get inserted at the top of the function as the patterns are applied.
Because of this, we get "inverted" constants more often, which is an aethetic
change to the IR but does permute some testcases.
Differential Revision: https://reviews.llvm.org/D99006
Andrew Litteken [Sat, 20 Mar 2021 23:03:02 +0000 (18:03 -0500)]
Revert "[IRSim] Adding basic implementation of llvm-sim."
Causing build errors on the Windows Buildbots.
This reverts commit
5155dff2784a47583d432d796b7cf47a0bed9f20.
Jessica Clarke [Sat, 20 Mar 2021 22:35:40 +0000 (22:35 +0000)]
[RISCV] Update comment in RISCVInstrInfoM.td
Missed in
07ed62b7d551.
Craig Topper [Sat, 20 Mar 2021 22:14:46 +0000 (15:14 -0700)]
[RISCV] Disable (mul (and X, 0xffffffff), (and Y, 0xffffffff)) optimization when Zba is enabled.
This optimization is trying to save SRLI instructions needed to
implement the ANDs. If we have zext.w we won't save anything.
Because we don't check that the multiply is the only user of the
AND we might even increase instruction count.
Craig Topper [Sat, 20 Mar 2021 22:09:15 +0000 (15:09 -0700)]
[RISCV] Add Zba command lines to xaluo.ll. NFC
Some of the patterns end up with 32 to 64 bit zero extends on RV64
which can be handled by zext.w.
Fangrui Song [Sat, 20 Mar 2021 22:24:02 +0000 (15:24 -0700)]
[test] Delete "-internal-isystem" "/usr/local/include"
Craig Topper [Sat, 20 Mar 2021 19:34:06 +0000 (12:34 -0700)]
[RISCV] Add isel pattern to optimize (mul (and X, 0xffffffff), (and Y, 0xffffffff)) on RV64
This patterns computes the full 64 bit product of a 32x32 unsigned
multiply. This requires a two pairs of SLLI+SRLI to zero the
upper 32 bits of the inputs.
We can do better than this by using two SLLI to move the lower
bits to the upper bits then use MULHU to compute the product. This
is the high half of a full 64x64 product. Since we put 32 0s in the lower
bits of the inputs we know the 128-bit product will have zeros in the
lower 64 bits. So the upper 64 bits, which MULHU computes, will contain
the original 64 bit product we were after.
The same trick would work for (mul (sext_inreg X, i32), (sext_inreg Y, i32))
using MULHS, but sext_inreg is sext.w which is already one instruction so we
wouldn't save anything.
Differential Revision: https://reviews.llvm.org/D99026
Andrew Litteken [Thu, 17 Sep 2020 20:43:40 +0000 (15:43 -0500)]
[IRSim] Adding basic implementation of llvm-sim.
This is a similarity visualization tool that accepts a Module and
passes it to the IRSimilarityIdentifier. The resulting SimilarityGroups
are output in a JSON file.
Tests are found in test/tools/llvm-sim and check for the file not found,
a bad module, and that the JSON is created correctly.
Reviewers: paquette, jroelofs, MaskRay
Recommit of:
15645d044bcfe2a0f63156048b302f997a717688 to fix linking
errors.
Differential Revision: https://reviews.llvm.org/D86974
Jinsong Ji [Sat, 20 Mar 2021 03:48:48 +0000 (03:48 +0000)]
[AIX] Update rpath for BUILD_SHARED_LIBS
BUILD_SHARED_LIBS build llvm component as shared library,
which can reduce the size a lot.
Normally, the binary use ORIGIN../lib to load component libraries,
unfortunatly, ORIGIN is not supported by AIX ld.
We hardcoded the build lib and install lib path in rpath for now
to enable BUILD_SHARED_LIBS build.
Understand that this is not perfect solution,
we can update this when we find better solution.
Reviewed By: hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D98901
Fangrui Song [Sat, 20 Mar 2021 20:24:49 +0000 (13:24 -0700)]
[test] Fix Driver/gcc-toolchain.cpp if CLANG_DEFAULT_RTLIB is compiler-rt
Sanjay Patel [Sat, 20 Mar 2021 18:45:56 +0000 (14:45 -0400)]
[BranchProbability] move options for 'likely' and 'unlikely'
This makes the settings available for use in other passes by housing
them within the Support lib, but NFC otherwise.
See D98898 for the proposed usage in SimplifyCFG
(where this change was originally included).
Differential Revision: https://reviews.llvm.org/D98945
Jez Ng [Sat, 20 Mar 2021 05:03:50 +0000 (01:03 -0400)]
[lld-macho] Minor touch-up to objc.s
Stephen Kelly [Wed, 17 Mar 2021 23:22:31 +0000 (23:22 +0000)]
[AST] Ensure that an empty json file is generated if compile errors
Differential Revision: https://reviews.llvm.org/D98827
Fangrui Song [Sat, 20 Mar 2021 18:06:44 +0000 (11:06 -0700)]
[test] Fix Driver/gcc-toolchain.cpp if CLANG_DEFAULT_CXX_STDLIB is libc++
Fangrui Song [Sat, 20 Mar 2021 17:36:51 +0000 (10:36 -0700)]
[VE] Fix types of multiclass template arguments in TableGen files
There were not properly checked before `[TableGen] Improve handling of template arguments`.
Fangrui Song [Sat, 20 Mar 2021 16:57:05 +0000 (09:57 -0700)]
Revert "Revert "[Driver] Drop obsoleted Ubuntu 11.04 gcc detection""
This reverts commit
243333ef3ec6c1e3910eb442177c2e2e927e6a87.
Vaivaswatha Nagaraj [Fri, 19 Mar 2021 14:05:13 +0000 (19:35 +0530)]
[OCaml] Add (get/set)_module_identifer functions
Also:
- Fix a bug that crept in when fixing a buildbot failure in
https://github.com/llvm/llvm-project/commit/
f7be9db6220cb39f0eaa12d2af3abedf0d86c303
- Use mlsize_t for cstr_to_string as that is what
caml_alloc_string specifies.
Differential Revision: https://reviews.llvm.org/D98851
David Zarzycki [Sat, 20 Mar 2021 11:52:08 +0000 (07:52 -0400)]
[lit] Sort testing summary output
As fallout from from the record-and-reorder work, people asked that the
summary output be sorted to aid diffing.
David Zarzycki [Sat, 20 Mar 2021 11:29:01 +0000 (07:29 -0400)]
Revert "[Driver] Drop obsoleted Ubuntu 11.04 gcc detection"
This reverts commit
bdf39e6b0ed4b41a1842ac0193f30a726f8d9f63.
The change is failing on Fedora 33 (x86-64).
Nathan James [Sat, 20 Mar 2021 10:59:36 +0000 (10:59 +0000)]
[clang-tidy] Fix bugprone-terminating-continue when continue appears inside a switch
Don't emit a warning if the `continue` appears in a switch context as changing it to `break` will break out of the switch rather than a do loop containing the switch.
Fixes https://llvm.org/PR49492.
Reviewed By: aaron.ballman
Differential Revision: https://reviews.llvm.org/D98338
Butygin [Fri, 12 Mar 2021 14:39:43 +0000 (17:39 +0300)]
[mlir] Additional folding for SelectOp
* Fold SelectOp when both true and false args are same SSA value
* Fold some cmp + select patterns
Differential Revision: https://reviews.llvm.org/D98576
Jeroen Dobbelaere [Sat, 20 Mar 2021 10:37:09 +0000 (11:37 +0100)]
Revert of D49126 [PredicateInfo] Use custom mangling to support ssa_copy with unnamed types.
Now that intrinsic name mangling can cope with unnamed types, the custom name mangling in PredicateInfo (introduced by D49126) can be removed.
(See D91250, D48541)
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D91661
Wang, Pengfei [Sat, 20 Mar 2021 04:55:46 +0000 (12:55 +0800)]
[X86] Fix a bug when calculating the ldtilecfg insertion points.
The BB we initialized the ldtilecfg is special. We don't need to check
if its predecessor BBs need to insert ldtilecfg for calls.
We reused the flag HasCallBeforeAMX, so that the predecessors won't be
added to CfgNeedInsert.
This case happens only when the entry BB is in a loop. We need to hoist
the first tile config point out of the loop in future.
Reviewed By: LuoYuanke
Differential Revision: https://reviews.llvm.org/D98845
Butygin [Fri, 12 Mar 2021 14:39:43 +0000 (17:39 +0300)]
[mlir] Canonicalize IfOp with trivial `then` and `else` bodies to list of SelectOp's
* Do we need a threshold on maximum number of Yeild arguments processed (maximum number of SelectOp's to be generated)?
* Had to modify some old IfOp tests to not get optimized by this pattern
Differential Revision: https://reviews.llvm.org/D98592
Juneyoung Lee [Sat, 20 Mar 2021 17:14:06 +0000 (02:14 +0900)]
[CFLGraph] Fix a crash due to missing handling of freeze
https://reviews.llvm.org/D85534#2636321
Fangrui Song [Sat, 20 Mar 2021 06:50:22 +0000 (23:50 -0700)]
[Driver][test] Fix gcc-toolchain.cpp on non-x86_64
joker881 [Sat, 20 Mar 2021 04:37:30 +0000 (12:37 +0800)]
Title: Remove a redundant parameter in clang/unittests/AST/CMakeLists.txt Reviewed by: MaskRay Differential Revision: https://reviews.llvm.org/D98922
Fangrui Song [Sat, 20 Mar 2021 06:23:28 +0000 (23:23 -0700)]
[Driver] Drop obsoleted Ubuntu 11.04 gcc detection
It has a very broken gcc installation path (usr/lib/i386-linux-gnu/gcc/i686-linux-gnu).
Fangrui Song [Sat, 20 Mar 2021 05:50:35 +0000 (22:50 -0700)]
[Driver] Delete toplevel i386-gnu/gcc detection in favor of i386-gnu alias triple detection
This is used by hurd.c (usr/lib/gcc/i386-gnu/4.6.0) but we can leverage
the existing alias triple detection.
Fangrui Song [Sat, 20 Mar 2021 05:45:50 +0000 (22:45 -0700)]
[Driver] Drop unneeded $triple/gcc/$triple detection
Shao-Ce Sun [Sat, 20 Mar 2021 05:43:07 +0000 (13:43 +0800)]
[NFC][ValueTypes] Align code by column
Adjusted some whitespaces.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D98975
Rob Suderman [Sat, 20 Mar 2021 00:47:39 +0000 (17:47 -0700)]
[mlir][tosa] Add tosa.reverse lowering to linalg.generic
Reverse lowers to a linalg.generic op by reversing the read order
in the index map.
Differential Revision: https://reviews.llvm.org/D98997
Carl Ritson [Sat, 20 Mar 2021 04:38:26 +0000 (13:38 +0900)]
[AMDGPU] Add MDT update missing from D98915
Siva Chandra [Sat, 20 Mar 2021 04:12:33 +0000 (04:12 +0000)]
[libc] Use add_library in add_entrypoint_library instead of invoking ar.
Mehdi Amini [Sat, 20 Mar 2021 01:23:12 +0000 (01:23 +0000)]
Update syntax for amx.tile_muli to use two Unit attr to mark the zext case
This makes the annotation tied to the operand and the use of a keyword
more explicit/readable on what it means.
Differential Revision: https://reviews.llvm.org/D99001
Nemanja Ivanovic [Sat, 20 Mar 2021 03:52:40 +0000 (22:52 -0500)]
[PowerPC][NFC] Do not produce i64 constants in 32-bit mode
There are some instances where we produce constants of type MVT::i64
unconditionally in the target DAG combines. This is not actually
valid in 32-bit mode.
Craig Topper [Sat, 20 Mar 2021 03:39:48 +0000 (20:39 -0700)]
[RISCV] Rename WriteShift/ReadShift scheduler classes to WriteShiftImm/ReadShiftImm. Move variable shifts from WriteIALU/ReadIALU to new WriteShiftReg/ReadShiftReg.
Previously only immediate shifts were in WriteShift. Register
shifts were grouped with IALU. Seems likely that immediate shifts
would be as fast or faster than register shifts. And that immediate
shifts wouldn't be any faster than IALU. So if any deserved to be in
their own group it should be register shifts not immediate shifts.
Rather than try to flip them let's just add more granularity
and give each kind their own class. I've used new names for both to
make them unambiguous and to force any downstream implementations to
be forced to put correct information in their scheduler models.
Reviewed By: evandro
Differential Revision: https://reviews.llvm.org/D98911
Fangrui Song [Sat, 20 Mar 2021 03:35:59 +0000 (20:35 -0700)]
[Driver] Stop searching other prefixes once a GCC installation is found in one prefix
so that when --sysroot is specified, the detected GCC installation will not be
overridden by another from /usr which happens to have a larger version.
This behavior is particularly inconvenient when the system has a larger version
GCC while the user wants to try out an older sysroot.
Delete some tests from linux-ld.c which overlap with cross-linux.c
Carl Ritson [Sat, 20 Mar 2021 02:27:52 +0000 (11:27 +0900)]
[AMDGPU] Rename SIInsertSkips Pass
Pass no longer handles skips. Pass now removes unnecessary
unconditional branches and lowers early termination branches.
Hence rename to SILateBranchLowering.
Move code to handle returns to epilog from SIPreEmitPeephole
into SILateBranchLowering. This means SIPreEmitPeephole only
contains optional optimisations, and all required transforms
are in SILateBranchLowering.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D98915
Carl Ritson [Sat, 20 Mar 2021 01:29:08 +0000 (10:29 +0900)]
[AMDGPU] Merge SIRemoveShortExecBranches into SIPreEmitPeephole
SIRemoveShortExecBranches is an optimisation so fits well in the
context of SIPreEmitPeephole.
Test changes relate to early termination from kills which have now
been lowered prior to considering branches for removal.
As these use s_cbranch the execz skips are now retained instead.
Currently either behaviour is valid as kill with EXEC=0 is a nop;
however, if early termination is used differently in future then
the new behaviour is the correct one.
Reviewed By: foad
Differential Revision: https://reviews.llvm.org/D98917
Lang Hames [Sat, 20 Mar 2021 02:13:50 +0000 (19:13 -0700)]
[llvm-jitlink] Scan input files for first object to determine triple.
The previous logic would crash if the first input file was an archive rather
than an object.
Senran Zhang [Sat, 20 Mar 2021 02:09:11 +0000 (19:09 -0700)]
[Utils][vim] Highlight `poison` keyword
Reviewed By: awarzynski, MaskRay
Differential Revision: https://reviews.llvm.org/D98927
Stella Laurenzo [Sat, 20 Mar 2021 01:44:51 +0000 (18:44 -0700)]
[mlir][python] Adapt to `segment_sizes` attribute type change.
* Broken by https://reviews.llvm.org/rG1a75be0023cd80fd8560d689999a63d4368c90e6
Lang Hames [Sat, 20 Mar 2021 00:31:29 +0000 (17:31 -0700)]
[JITLink] Remove redundant local variable definitions from a unit test.
Stella Laurenzo [Fri, 19 Mar 2021 22:43:42 +0000 (15:43 -0700)]
[mlir][python] Function decorator for capturing a FuncOp from a python function.
* Moves this out of a test case where it was being developed to good effect and generalizes it.
* Having tried a number of things like this, I think this balances concerns reasonably well.
Differential Revision: https://reviews.llvm.org/D98989
Carl Ritson [Sat, 20 Mar 2021 01:03:26 +0000 (10:03 +0900)]
[AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles
Add code so duplication index register changes can be removed from
inside bundles.
Reviewed By: rampitec, foad
Differential Revision: https://reviews.llvm.org/D98940
Anshil Gandhi [Sat, 20 Mar 2021 00:59:45 +0000 (19:59 -0500)]
[NFC] [PowerPC] Determine Endianness in PPCTargetMachine
The TargetMachine uses the triple to determine endianness. Just
use that logic rather than replicating it in PPCSubtarget.
Differential revision: https://reviews.llvm.org/D98674
River Riddle [Sat, 20 Mar 2021 01:19:16 +0000 (18:19 -0700)]
[mlir][docs] Remove the BuiltinDialect documentation from langref and generate it from ODS
Now that all of the builtin dialect is generated from ODS, its documentation in LangRef can be split out and replaced with references to Dialects/Builtin.md. LangRef is quite crusty right now and should really have a full cleanup done in a followup.
Differential Revision: https://reviews.llvm.org/D98562
Chris Lattner [Sat, 20 Mar 2021 00:57:47 +0000 (17:57 -0700)]
[testsuite] Make testsuite more stable vs canonicalization change. NFC.
Differential Revision: https://reviews.llvm.org/D98998
Jonas Devlieghere [Sat, 20 Mar 2021 00:57:17 +0000 (17:57 -0700)]
Revert "[lldb] Make the API, Shell and Unit tests independent lit test suites"
This reverts commit
6c52d4fd4c24a0cf738e44516ca8378d65dcf019.
Fangrui Song [Sat, 20 Mar 2021 00:47:29 +0000 (17:47 -0700)]
[Driver] Delete compatibility aliases -mpie-copy-relocations and -mno-pie-copy-relocations
They should be unused everywhere.
Dhruva Chakrabarti [Sat, 20 Mar 2021 00:40:37 +0000 (17:40 -0700)]
Empty test commit, verifying commit access
River Riddle [Sat, 20 Mar 2021 00:11:23 +0000 (17:11 -0700)]
[mlir][NFC] Use the native range instead of APInt when computing operand ranges
This removes the need to construct an APInt for each value, given that it is guaranteed to contain 32 bit elements.
BEGIN_PUBLIC
...text exposed to open source public git repo...
END_PUBLIC
River Riddle [Fri, 19 Mar 2021 23:19:23 +0000 (16:19 -0700)]
[mlir] Update `simplifyRegions` to use RewriterBase for erasure notifications
This allows for notifying callers when operations/blocks get erased, which is especially useful for the greedy pattern driver. The current greedy pattern driver "throws away" all information on constants in the operation folder because it doesn't know if they get erased or not. By passing in RewriterBase, we can directly track this and prevent the need for the pattern driver to rediscover all of the existing constants. In some situations this cuts the compile time of the canonicalizer in half.
Differential Revision: https://reviews.llvm.org/D98755
Peter Collingbourne [Fri, 19 Mar 2021 23:23:30 +0000 (16:23 -0700)]
gn build: Unbreak Android cross-compilation.
- D96404 defaulted to libunwind which isn't provided by NDK r21
(or r22), so specify -rtlib=libgcc on non-arm32.
- D97993 means that we need to use --gcc-toolchain instead of -B
to let the driver find libgcc.
Ellis Hoag [Fri, 19 Mar 2021 23:24:16 +0000 (16:24 -0700)]
Port D97640 to llvm/include/llvm/ProfileData/InstrProfData.inc
Differential Revision: https://reviews.llvm.org/D98982
Lang Hames [Fri, 19 Mar 2021 21:54:07 +0000 (14:54 -0700)]
[JITLink] Don't issue lookups for empty symbol sets.
Issuing a lookup for an empty symbol set is legal, but can actually result in
unrelated work being done if there was a work queue left over from the previous
lookup. We can avoid doing this unrelated work (reducing stack depth and
interleaving of debugging output) by not issuing these no-op lookups in the
first place.
Christoffer Lernö [Fri, 19 Mar 2021 22:55:52 +0000 (18:55 -0400)]
Add type attributes to LLVM C API
The LLVM C API is missing type attributes as is needed by attributes
such as sret and byval. This patch adds three missing wrapper
functions.
Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=48249
https://reviews.llvm.org/D97763
Vedant Kumar [Thu, 18 Mar 2021 18:12:17 +0000 (11:12 -0700)]
[lldb/PlatformPOSIX] Change LoadImage default to RTLD_LAZY (reapply)
In general, it seems like the debugger should allow programs to load & run with
libraries as far as possible, instead of defaulting to being super-picky about
unavailable symbols.
This is critical on macOS/Darwin, as libswiftCore.dylib may 1) export a version
symbol using @available markup and then 2) expect that other exported APIs are
only dynamically used once the version symbol is checked. We can't open a
version of the library built with a bleeding-edge SDK on an older OS without
RTLD_LAXY (or pervasive/expensive @available markup added to dyld APIs).
This was previously committed as
cb8c1ee269da and reverted due to
unknown failures on the Linux bots. This version adds additional asserts
to check that the shared objects are where we expect them & that calling
f1() from libt1 produces the expected value. The Linux failure is
tracked by https://bugs.llvm.org/show_bug.cgi?id=49656.
See: https://lists.llvm.org/pipermail/lldb-dev/2021-March/016796.html
Differential Revision: https://reviews.llvm.org/D98879
Fangrui Song [Fri, 19 Mar 2021 22:42:37 +0000 (15:42 -0700)]
[docs] Improve documentation of -B and --gcc-toolchain
Differential Revision: https://reviews.llvm.org/D97902
Fangrui Song [Fri, 19 Mar 2021 22:42:18 +0000 (15:42 -0700)]
[Driver] Suppress GCC detection under -B
In GCC, if `-B $prefix` is specified, `$prefix` is used to find executable files and startup files.
`$prefix/include` is added as an include search directory.
Clang overloads -B with GCC installation detection semantics which make the
behavior less predictable (due to the "largest GCC version wins" rule) and
interact poorly with --gcc-toolchain (--gcc-toolchain can be overridden by -B).
* `clang++ foo.cpp` detects GCC installation under `/usr`.
* `clang++ --gcc-toolchain=Inputs foo.cpp` detects GCC installation under `Inputs`.
* `clang++ -BA --gcc-toolchain=B foo.cpp` detects GCC installation under A and B and the larger version wins. With this patch, only B is used for detection.
* `clang++ -BA foo.cpp` detects GCC installation under `A` and `/usr`, and the larger GCC version wins. With this patch `A` is not used for detection.
This patch changes -B to drop the GCC detection semantics. Its executable
searching semantics are preserved. --gcc-toolchain is the recommended option to
specify the GCC installation detection directory.
(
Note: Clang detects GCC installation in various target dependent directories.
`$sysroot/usr` (sysroot defaults to "") is a common directory used by most targets.
Such a directory is expected to contain something like `lib{,32,64}/gcc{,-cross}/$triple`.
Clang will then construct library/include paths from the directory.
)
Differential Revision: https://reviews.llvm.org/D97993
Vedant Kumar [Fri, 19 Mar 2021 22:26:16 +0000 (15:26 -0700)]
Revert "[lldb/PlatformPOSIX] Change LoadImage default to RTLD_LAZY"
This reverts commit
cb8c1ee269da72eb6e2c18800cd8ab0a74050785. The test
is failing on Debian for unknown reasons.
https://lab.llvm.org/buildbot/#/builders/68/builds/8990
Vedant Kumar [Thu, 18 Mar 2021 18:12:17 +0000 (11:12 -0700)]
[lldb/PlatformPOSIX] Change LoadImage default to RTLD_LAZY
In general, it seems like the debugger should allow programs to load & run with
libraries as far as possible, instead of defaulting to being super-picky about
unavailable symbols.
This is critical on macOS/Darwin, as libswiftCore.dylib may 1) export a version
symbol using @available markup and then 2) expect that other exported APIs are
only dynamically used once the version symbol is checked. We can't open a
version of the library built with a bleeding-edge SDK on an older OS without
RTLD_LAXY (or pervasive/expensive @available markup added to dyld APIs).
See: https://lists.llvm.org/pipermail/lldb-dev/2021-March/016796.html
Differential Revision: https://reviews.llvm.org/D98879
Arthur Eubanks [Fri, 19 Mar 2021 21:30:48 +0000 (14:30 -0700)]
[NewPM] Verify LoopAnalysisResults after a loop pass
All loop passes should preserve all analyses in LoopAnalysisResults. Add
checks for those when the checks are enabled (which is by default with
expensive checks on).
Note that due to PR44815, we don't check LAR's ScalarEvolution.
Apparently calling SE.verify() can change its results.
This is a reland of https://reviews.llvm.org/D98820 which was reverted
due to unacceptably large compile time regressions in normal debug
builds.
Jessica Paquette [Thu, 11 Mar 2021 23:36:01 +0000 (15:36 -0800)]
[GlobalISel] Add G_SBFX + G_UBFX (bitfield extraction opcodes)
There is a bunch of similar bitfield extraction code throughout *ISelDAGToDAG.
E.g, ARMISelDAGToDAG, AArch64ISelDAGToDAG, and AMDGPUISelDAGToDAG all contain
code that matches a bitfield extract from an and + right shift.
Rather than duplicating code in the same way, this adds two opcodes:
- G_UBFX (unsigned bitfield extract)
- G_SBFX (signed bitfield extract)
They work like this
```
%x = G_UBFX %y, %lsb, %width
```
Where `lsb` and `width` are
- The least-significant bit of the extraction
- The width of the extraction
This will extract `width` bits from `%y`, starting at `lsb`. G_UBFX zero-extends
the result, while G_SBFX sign-extends the result.
This should allow us to use the combiner to match the bitfield extraction
patterns rather than duplicating pattern-matching code in each target.
Differential Revision: https://reviews.llvm.org/D98464
River Riddle [Fri, 19 Mar 2021 21:20:14 +0000 (14:20 -0700)]
[mlir][Pass] Coalesce dynamic pass pipelines before running
This was missed when dynamic pass pipelines were added, and is necessary for maximizing the performance/parallelism potential of the pass pipeline.
Fangrui Song [Fri, 19 Mar 2021 21:35:22 +0000 (14:35 -0700)]
[llvm-readobj] Remove legacy GNU_PROPERTY_X86_ISA_1_{NEEDED,USED} and dump new GNU_PROPERTY_X86_ISA_1_{NEEDED,USED}
https://sourceware.org/bugzilla/show_bug.cgi?id=26703 deprecated the
previous GNU_PROPERTY_X86_ISA_1_{CMOV,SSE,*} values (renamed to `COMPAT`)
and added new values.
Since the legacy values are not used by compilers, having dumping support in
llvm-readobj is unnecessary. So just drop the legacy feature.
The new values are used by GCC 11
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97250) `-march=x86-64-v[234]` to
indicate the micro-architecture ISA levels.
Differential Revision: https://reviews.llvm.org/D98818
Jonas Devlieghere [Fri, 19 Mar 2021 21:31:22 +0000 (14:31 -0700)]
[lldb] Update assert.test to be less strict
Be less strict when checking for the assert substring.
Arthur Eubanks [Fri, 19 Mar 2021 21:31:08 +0000 (14:31 -0700)]
Revert "[NewPM] Verify LoopAnalysisResults after a loop pass"
This reverts commit
94c269baf58330a5e303a4f86f64681f2f7a858b.
Still causes too large of compile time regression in normal debug
builds. Will put under expensive checks instead.
Jonas Devlieghere [Fri, 19 Mar 2021 21:20:26 +0000 (14:20 -0700)]
[lldb] Call os_log_fault on lldb_assert
Call `os_log_fault` when an lldb assert fails. We piggyback off
`LLVM_SUPPORT_XCODE_SIGNPOSTS`, which also depends on `os_log`, to avoid
having to introduce another CMake check and corresponding define.
This patch also adds a small test using lldb-test that verifies we abort
with a "regular" assertion when asserts are enabled.
Differential revision: https://reviews.llvm.org/D98987
Ella Ma [Fri, 19 Mar 2021 04:14:13 +0000 (21:14 -0700)]
[llvm] Add assertions for the smart pointers with the possibility to be null in ModuleLazyLoaderCache::operator()
Split from D91844.
The return value of function `ModuleLazyLoaderCache::operator()` in file llvm/tools/llvm-link/llvm-link.cpp. According to the bug report of my static analyzer, the std::function variable `ModuleLazyLoaderCache::createLazyModule` points to function `loadFile`, which may return `nullptr` when error. And the pointer is dereferenced without a check.
Reviewed By: tejohnson
Differential Revision: https://reviews.llvm.org/D97258
Stella Laurenzo [Fri, 19 Mar 2021 18:57:01 +0000 (11:57 -0700)]
NFC: Break up the mlir python bindings into individual sources.
* IRModules.cpp -> (IRCore.cpp, IRAffine.cpp, IRAttributes.cpp, IRTypes.cpp).
* The individual pieces now compile in the 5-15s range whereas IRModules.cpp was starting to approach a minute (didn't capture a before time).
* More fine grained splitting is possible, but this represents the most obvious.
Differential Revision: https://reviews.llvm.org/D98978
Arthur Eubanks [Wed, 17 Mar 2021 21:53:57 +0000 (14:53 -0700)]
[NewPM] Verify LoopAnalysisResults after a loop pass
All loop passes should preserve all analyses in LoopAnalysisResults. Add
checks for those.
Note that due to PR44815, we don't check LAR's ScalarEvolution.
Apparently calling SE.verify() can change its results.
Only verify MSSA when VerifyMemorySSA, normally it's very expensive.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D98820
Butygin [Fri, 12 Mar 2021 14:39:43 +0000 (17:39 +0300)]
[MLIR] Test pattern benefit sorting between operation specific and operation agnostic patterns.
Previously low benefit op-specific patterns never had a chance to match
even if high benefit op-agnostic pattern failed to match.
This was already fixed upstream, this commit just adds testscase
Differential Revision: https://reviews.llvm.org/D98513
Benjamin Kramer [Fri, 19 Mar 2021 19:56:59 +0000 (20:56 +0100)]
[mlir][Linalg] Make LLVM_DEBUG region bigger to avoid warnings in Release builds
Transforms.cpp:586:16: error: unused variable 'v' [-Werror,-Wunused-variable]
for (Value v : operands)
^
Benjamin Kramer [Fri, 19 Mar 2021 19:35:17 +0000 (20:35 +0100)]
[CodeGen] Don't crash on for loops with cond variables and no increment
This looks like an oversight from
a875721d8a2d, creating IR that refers
to `for.inc` even if it doesn't exist.
Differential Revision: https://reviews.llvm.org/D98980
Jim Ingham [Fri, 19 Mar 2021 19:38:41 +0000 (12:38 -0700)]
Revert "Make the stop-on-sharedlibrary-events setting work."
This reverts commit
9406d43138811ac4dfd0ab31434f65a649bc882e.
I messed up a test, and when I got it right it was failing. The changed logic
doesn't work quite right (now the async callback called at sync time is
forcing us to stop. I need to be a little more careful about that.
Jim Ingham [Fri, 19 Mar 2021 19:38:23 +0000 (12:38 -0700)]
Revert "Skip all the tests for Windows."
This reverts commit
a8d62fc8ff1c836e16cfb1a510ee8063ac2652ff.
Sanjay Patel [Fri, 19 Mar 2021 19:09:53 +0000 (15:09 -0400)]
[SLP] remove unnecessary characters in test; NFC
Glitch that crept in with
62f9c3358b81
Sanjay Patel [Fri, 19 Mar 2021 19:05:52 +0000 (15:05 -0400)]
[SLP] add tests for min/max reductions that use intrinsics; NFC
Jim Ingham [Fri, 19 Mar 2021 19:05:16 +0000 (12:05 -0700)]
Skip all the tests for Windows.
Jim Ingham [Fri, 19 Mar 2021 18:56:46 +0000 (11:56 -0700)]
Make the stop-on-sharedlibrary-events setting work.
We weren't taking into account the "m_should_stop" setting that the
synchronous breakpoint callback had already set when we did PerformAction
in the StopInfoBreakpoint. So we didn't obey its instructions when it
told us to stop. Fixed that and added some tests both for when we
just have the setting, and when we have the setting AND other breakpoints
at the shared library load notification breakpoint address.
Differential Revision: https://reviews.llvm.org/D98914
Louis Dionne [Thu, 18 Mar 2021 21:05:25 +0000 (14:05 -0700)]
[libc++] NFCI: Remove dead code in the Lit configuration
I was trying to fix something else and I stumbled upon several methods
that are not used anymore in target_info.py.
Differential Revision: https://reviews.llvm.org/D98896