Replace three "strip & accumulate" implementations with a single one
This patch replaces the three almost identical "strip & accumulate"
implementations for constant pointer offsets with a single one,
combining the respective functionalities. The old interfaces are kept
for now.
[NFC] Adjust "invalid.ll.bc" tests to check for AttrKind #255 not #63
We are about to add enum attributes with AttrKind numbers >= 63. This
means we cannot use AttrKind #63 to test for an invalid attribute number
in the RAW format anymore. This patch changes the number of an invalid
attribute to #255. There is no change to the character of the tests.
[X86] Don't convert 8 or 16 bit ADDs to LEAs on Atom in FixupLEAPass.
We use the functions that convert to three address to do the
conversion, but changing an 8 or 16 bit will cause it to create
a virtual register. This can't be done after register allocation
where this pass runs.
I've switched the pass completely to a white list of instructions
that can be converted to LEA instead of a blacklist that was
incorrect. This will avoid surprises if we enhance the three
address conversion function to include additional instructions
in the future.
This patch doesn't work with binaries built w/ `--emit-relocs`, e.g.
```
$ echo 'int main() { return 0; }' | clang -Wl,--emit-relocs -x c - -o foo && llvm-objcopy --strip-unneeded foo
llvm-objcopy: error: 'foo': not stripping symbol '__gmon_start__' because it is named in a relocation
```
...then flipping the operands in the compare instruction can allow using a subtract that sets compare flags.
Motivated by diffs in D58875 - not sure if this changes anything there,
but this seems like a good thing independent of that.
There's a more involved version of this transform already in IR (in instcombine
although that seems misplaced to me) - see "swapMayExposeCSEOpportunities()".
David Tenty [Wed, 10 Jul 2019 22:13:55 +0000 (22:13 +0000)]
[NFC]Fix IR/MC depency issue for function descriptor SDAG implementation
Summary: llvm/IR/GlobalValue.h can't be included in MC, that creates a circular dependency between MC and IR libraries. This circular dependency is causing an issue for build system that enforce layering.
[AArch64][GlobalISel] Optimize compare and branch cases with G_INTTOPTR and unknown values.
Since we have distinct types for pointers and scalars, G_INTTOPTRs can sometimes
obstruct attempts to find constant source values. These usually come about when
try to do some kind of null pointer check. Teaching getConstantVRegValWithLookThrough
about this operation allows the CBZ/CBNZ optimization to catch more cases.
This change also improves the case where we can't find a constant source at all.
Previously we would emit a cmp, cset and tbnz for that. Now we try to just emit
a cmp and conditional branch, saving an instruction.
The cumulative code size improvement of this change plus D64354 is 5.5% geomean
on arm64 CTMark -O0.
[GlobalISel][AArch64][NFC] Use getDefIgnoringCopies from Utils where we can
There are a few places where we walk over copies throughout
AArch64InstructionSelector.cpp. In Utils, there's a function that does exactly
this which we can use instead.
Note that the utility function works with the case where we run into a COPY
from a physical register. We've run into bugs with this a couple times, so using
it should defend us from similar future bugs.
Also update opt-fold-compare.mir to show that we still handle physical registers
properly.
Michael Berg [Wed, 10 Jul 2019 18:23:26 +0000 (18:23 +0000)]
Move three folds for FADD, FSUB and FMUL in the DAG combiner away from Unsafe to more aligned checks that reflect context
Summary: Unsafe does not map well alone for each of these three cases as it is missing NoNan context when accessed directly with clang. I have migrated the fold guards to reflect the expectations of handing nan and zero contexts directly (NoNan, NSZ) and some tests with it. Unsafe does include NSZ, however there is already precedent for using the target option directly to reflect that context.
David Greene [Wed, 10 Jul 2019 18:07:01 +0000 (18:07 +0000)]
[System Model] [TTI] Update cache and prefetch TTI interfaces
Rework the TTI cache and software prefetching APIs to prepare for the
introduction of a general system model. Changes include:
- Marking existing interfaces const and/or override as appropriate
- Adding comments
- Adding BasicTTIImpl interfaces that delegate to a subtarget
implementation
- Adding a default "no information" subtarget implementation
Only a handful of targets use these interfaces currently: AArch64,
Hexagon, PPC and SystemZ. AArch64 already has a custom subtarget
implementation, so its custom TTI implementation is migrated to use
the new facilities in BasicTTIImpl to invoke its custom subtarget
implementation. The custom TTI implementations continue to exist for
the other targets with this change. They are not moved over to
subtarget-based implementations.
The end goal is to have the default subtarget implementation defer to
the system model defined by the target. With this change, the default
subtarget implementation essentially returns "no information" for
these interfaces. None of the existing users of TTI will hit that
implementation because they define their own custom TTI
implementations and won't use the BasicTTIImpl implementations.
Once system models are in place for the targets that use these
interfaces, their custom TTI implementations can be removed.
Don Hinton [Wed, 10 Jul 2019 17:57:05 +0000 (17:57 +0000)]
Recommit "[CommandLine] Remove OptionCategory and SubCommand caches from the Option class."
Previously reverted in 364141 due to buildbot breakage, and fixed here
by making GeneralCategory global a ManagedStatic.
Summary:
This change processes `OptionCategory`s and `SubCommand`s as they
are seen instead of caching them in the Option class and processing
them later. Doing so simplifies the work needed to be done by the Global
parser and significantly reduces the size of the Option class to a mere 64
bytes.
Removing the `OptionCategory` cache saved 24 bytes, and removing
the `SubCommand` cache saved an additional 48 bytes, for a total of a
72 byte reduction.
[LoopRotate + MemorySSA] Keep an <instruction-cloned instruction> map.
Summary:
The map kept in loop rotate is used for instruction remapping, in order
to simplify the clones of instructions. Thus, if an instruction can be
simplified, its simplified value is placed in the map, even when the
clone is added to the IR. MemorySSA in contrast needs to know about that
clone, so it can add an access for it.
To resolve this: keep a different map for MemorySSA.
Lang Hames [Wed, 10 Jul 2019 17:24:24 +0000 (17:24 +0000)]
[ORC] Add custom IR compiler configuration to LLJITBuilder to enable obj caches.
LLJITBuilder now has a setCompileFunctionCreator method which can be used to
construct a CompileFunction for the LLJIT instance being created. The motivating
use-case for this is supporting ObjectCaches, which can now be set up at
compile-function construction time. To demonstrate this an example project,
LLJITWithObjectCache, is included.
[CodeExtractor] Fix sinking of allocas with multiple bitcast uses (PR42451)
An alloca which can be sunk into the extraction region may have more
than one bitcast use. Move these uses along with the alloca to prevent
use-before-def.
Use bool() consistently to get boolean value of Error, Optional and
Expected types in EXPECT calls. While static_cast is used in all cases
but one, bool provides more clarity and makes more sense as a new
default.
Indeed, not all of these patterns are canonical.
But since this fold will only produce a single instruction
i'm really interested in handling even uncanonical patterns.
Other than these 6 patterns, i can't think of any other
reasonable variants right now, although i'm sure they exist.
For now let's start with patterns where both shift amounts are variable,
with trivial constant "offset" between them, since i believe this is
both simplest to handle and i think this is most common.
But again, there are likely other variants where we could use
ValueTracking/ConstantRange to handle more cases.
Jay Foad [Wed, 10 Jul 2019 14:53:47 +0000 (14:53 +0000)]
[AMDGPU] Allow abs/neg source modifiers on v_cndmask_b32
Summary:
D59191 added support for these modifiers in the assembler and
disassembler. This patch just teaches instruction selection that it can
use them.
Summary:
This patch simplifies 2 aspects in the FileCheckNumericVariable code.
First, setValue() method is turned into a void function since being
called only on undefined variable is an invariant and is now asserted
rather than returned. This remove the assert from the callers.
Second, clearValue() method is also turned into a void function since
the only caller does not check its return value since it may be trying
to clear the value of variable that is already cleared without this
being noteworthy.
Summary:
The value of the FileCheckNumericVariable class instance representing
the @LINE numeric variable is set and cleared respectively before and
after substitutions are made, if any. However, when a substitution
fails, the value is not cleared. This causes the next substitution of
@LINE later on to give the wrong value since setValue is a nop if the
value is already set. This is what caused failures after commit r365249.
Simon Pilgrim [Wed, 10 Jul 2019 11:26:57 +0000 (11:26 +0000)]
[X86] EltsFromConsecutiveLoads - store Loads on a per-element basis. NFCI.
Cache the LoadSDNode nodes so we can easily map to/from the element index instead of packing them together - this will be useful for future patches for PR16739 etc.
Nikola Prica [Wed, 10 Jul 2019 11:17:48 +0000 (11:17 +0000)]
[ELF] Loose a condition for relocation with a symbol
Deleted code was introduced as a work around for a bug in the gold linker
(http://sourceware.org/PR16794). Test case that was given as a reason for
this part of code, the one on previous link, now works for the gold.
This condition is too strict and when a code is compiled with debug info
it forces generation of numerous relocations with symbol for architectures
that do not have relocation addend.
Simon Pilgrim [Wed, 10 Jul 2019 10:46:36 +0000 (10:46 +0000)]
[X86][SSE] EltsFromConsecutiveLoads - add basic dereferenceable support
This patch checks to see if the vector element loads are based off a dereferenceable pointer that covers the entire vector width, in which case we don't need to have element loads at both extremes of the vector width - just the start (base pointer) of it.
[SimpleLoopUnswitch] Add a test case exposing a bug
This test exposes a bug in SimpleLoopUnswitch that leads to a crash on
assert(SuccessorsCount > 1 && "Cannot unswitch a condition without multiple distinct successors!");
when SimpleLoopUnswitch considers unswitching of a loop by a switch with one successor.
Consolidate llvm::BWH_* statics into an enum to fix
module build issues. This fixes the LLVM_Bitcode module,
getting rid of -Wmodules-ambiguous-internal-linkage.
[Support] Move llvm::MemoryBuffer to sys::fs::file_t
Summary:
On Windows, Posix integer file descriptors are a compatibility layer
over native file handles provided by the C runtime. There is a hard
limit on the maximum number of file descriptors that a process can open,
and the limit is 8192. LLD typically doesn't run into this limit because
it opens input files, maps them into memory, and then immediately closes
the file descriptor. This prevents it from running out of FDs.
For various reasons, I'd like to open handles to every input file and
keep them open during linking. That requires migrating MemoryBuffer over
to taking open native file handles instead of integer FDs.
Matt Arsenault [Tue, 9 Jul 2019 23:34:29 +0000 (23:34 +0000)]
GlobalISel: Implement lower for G_FCOPYSIGN
In SelectionDAG AMDGPU treated these as legal, but this was mostly
because the bitcasts required for FP types were painful. Theoretically
the bitpattern should eventually match to bfi, so don't bother trying
to get the patterns to import.
[docs][Remarks] Add documentation for remarks in LLVM
This adds documentation that describes remarks in LLVM.
It aims at explaining what remarks are, how to enable them, and what
users can do with the different modes.
It lists all the available flags in LLVM (excluding clang), and
describes the expected YAML structure as well as the tools that support
the YAML format today.
For a given set of live values, the spill cost will always be the
same for each call. Compute the cost once and multiply it by the
number of calls.
(I'm not sure this spill cost modeling makes sense if there are
multiple calls, as the spill cost will likely be shared across
calls in that case. But that's how it currently works.)
hwasan: Improve precision of checks using short granule tags.
A short granule is a granule of size between 1 and `TG-1` bytes. The size
of a short granule is stored at the location in shadow memory where the
granule's tag is normally stored, while the granule's actual tag is stored
in the last byte of the granule. This means that in order to verify that a
pointer tag matches a memory tag, HWASAN must check for two possibilities:
* the pointer tag is equal to the memory tag in shadow memory, or
* the shadow memory tag is actually a short granule size, the value being loaded
is in bounds of the granule and the pointer tag is equal to the last byte of
the granule.
Pointer tags between 1 to `TG-1` are possible and are as likely as any other
tag. This means that these tags in memory have two interpretations: the full
tag interpretation (where the pointer tag is between 1 and `TG-1` and the
last byte of the granule is ordinary data) and the short tag interpretation
(where the pointer tag is stored in the granule).
When HWASAN detects an error near a memory tag between 1 and `TG-1`, it
will show both the memory tag and the last byte of the granule. Currently,
it is up to the user to disambiguate the two possibilities.
Because this functionality obsoletes the right aligned heap feature of
the HWASAN memory allocator (and because we can no longer easily test
it), the feature is removed.
Also update the documentation to cover both short granule tags and
outlined checks.
Philip Reames [Tue, 9 Jul 2019 19:59:39 +0000 (19:59 +0000)]
[PoisonChecking] Flesh out complete todo list for full coverage
Note: I don't actually plan to implement all of the cases at the moment, I'm just documenting them for completeness. There's a couple of cases left which are practically useful for me in debugging loop transforms, and I'll probably stop there for the moment.
[X86][AMDGPU][DAGCombiner] Move call to allowsMemoryAccess into isLoadBitCastBeneficial/isStoreBitCastBeneficial to allow X86 to bypass it
Basically the problem is that X86 doesn't set the Fast flag from
allowsMemoryAccess on certain CPUs due to slow unaligned memory
subtarget features. This prevents bitcasts from being folded into
loads and stores. But all vector loads and stores of the same width
are the same cost on X86.
This patch merges the allowsMemoryAccess call into isLoadBitCastBeneficial to allow X86 to skip it.
Philip Reames [Tue, 9 Jul 2019 19:26:12 +0000 (19:26 +0000)]
[PoisonCheker] Support for out of bounds operands on shifts + insert/extractelement
These are sources of poison which don't come from flags, but are clearly documented in the LangRef. Left off support for scalable vectors for the moment, but should be easy to add if anyone is interested.
Sean Fertile [Tue, 9 Jul 2019 19:21:01 +0000 (19:21 +0000)]
Boilerplate for producing XCOFF object files from the PowerPC backend.
Stubs out a number of the classes needed to produce a new object file format
(XCOFF) for the powerpc-aix target. For testing input is an empty module which
produces an object file with just a file header.
Bob Haarman [Tue, 9 Jul 2019 18:50:55 +0000 (18:50 +0000)]
[ThinLTO] only emit used or referenced CFI records to index
Summary: We emit CFI_FUNCTION_DEFS and CFI_FUNCTION_DECLS to
distributed ThinLTO indices to implement indirect function call
checking. This change causes us to only emit entries for functions
that are either defined or used by the module we're writing the index
for (instead of all functions in the combined index), which can make
the indices substantially smaller.
Philip Reames [Tue, 9 Jul 2019 18:49:29 +0000 (18:49 +0000)]
Add a transform pass to make the executable semantics of poison explicit in the IR
Implements a transform pass which instruments IR such that poison semantics are made explicit. That is, it provides a (possibly partial) executable semantics for every instruction w.r.t. poison as specified in the LLVM LangRef. There are obvious parallels to the sanitizer tools, but this pass is focused purely on the semantics of LLVM IR, not any particular source language.
The target audience for this tool is developers working on or targetting LLVM from a frontend. The idea is to be able to take arbitrary IR (with the assumption of known inputs), and evaluate it concretely after having made poison semantics explicit to detect cases where either a) the original code executes UB, or b) a transform pass introduces UB which didn't exist in the original program.
At the moment, this is mostly the framework and still needs to be fleshed out. By reusing existing code we have decent coverage, but there's a lot of cases not yet handled. What's here is good enough to handle interesting cases though; for instance, one of the recent LFTR bugs involved UB being triggered by integer induction variables with nsw/nuw flags would be reported by the current code.
(See comment in PoisonChecking.cpp for full explanation and context)
projects / llvm / log