I've been looking at missed vectorizations in one codebase.
One particular thing that stands out is that some of the loops
reach vectorizer in a rather mangled form, with weird PHI's,
and some of the loops aren't even in a rotated form.
After taking a more detailed look, that happened because
the loop's headers were too big by then. It is evident that
SimplifyCFG's common code hoisting transform is at fault there,
because the pattern it handles is precisely the unrotated
loop basic block structure.
Surprizingly, `SimplifyCFGOpt::HoistThenElseCodeToIf()` is enabled
by default, and is always run, unlike it's friend, common code sinking
transform, `SinkCommonCodeFromPredecessors()`, which is not enabled
by default and is only run once very late in the pipeline.
I'm proposing to harmonize this, and disable common code hoisting
until //late// in pipeline. Definition of //late// may vary,
here currently i've picked the same one as for code sinking,
but i suppose we could enable it as soon as right after
loop rotation happens.
Experimentation shows that this does indeed unsurprizingly help,
more loops got rotated, although other issues remain elsewhere.
Now, this undoubtedly seriously shakes phase ordering.
This will undoubtedly be a mixed bag in terms of both compile- and
run- time performance, codesize. Since we no longer aggressively
hoist+deduplicate common code, we don't pay the price of said hoisting
(which wasn't big). That may allow more loops to be rotated,
so we pay that price. That, in turn, that may enable all the transforms
that require canonical (rotated) loop form, including but not limited to
vectorization, so we pay that too. And in general, no deduplication means
more [duplicate] instructions going through the optimizations. But there's still
late hoisting, some of them will be caught late.
As per benchmarks i've run {F12360204}, this is mostly within the noise,
there are some small improvements, some small regressions.
One big regression i saw i fixed in rG8d487668d09fb0e4e54f36207f07c1480ffabbfd, but i'm sure
this will expose many more pre-existing missed optimizations, as usual :S
llvm-compile-time-tracker.com thoughts on this:
http://llvm-compile-time-tracker.com/compare.php?from=e40315d2b4ed1e38962a8f33ff151693ed4ada63&to=c8289c0ecbf235da9fb0e3bc052e3c0d6bff5cf9&stat=instructions
* this does regress compile-time by +0.5% geomean (unsurprizingly)
* size impact varies; for ThinLTO it's actually an improvement
The largest fallout appears to be in GVN's load partial redundancy
elimination, it spends *much* more time in
`MemoryDependenceResults::getNonLocalPointerDependency()`.
Non-local `MemoryDependenceResults` is widely-known to be, uh, costly.
There does not appear to be a proper solution to this issue,
other than silencing the compile-time performance regression
by tuning cut-off thresholds in `MemoryDependenceResults`,
at the cost of potentially regressing run-time performance.
D84609 attempts to move in that direction, but the path is unclear
and is going to take some time.
If we look at stats before/after diffs, some excerpts:
* RawSpeed (the target) {F12360200}
* -14 (-73.68%) loops not rotated due to the header size (yay)
* -272 (-0.67%) `"Number of live out of a loop variables"` - good for vectorizer
* -3937 (-64.19%) common instructions hoisted
* +561 (+0.06%) x86 asm instructions
* -2 basic blocks
* +2418 (+0.11%) IR instructions
* vanilla test-suite + RawSpeed + darktable {F12360201}
* -36396 (-65.29%) common instructions hoisted
* +1676 (+0.02%) x86 asm instructions
* +662 (+0.06%) basic blocks
* +4395 (+0.04%) IR instructions
It is likely to be sub-optimal for when optimizing for code size,
so one might want to change tune pipeline by enabling sinking/hoisting
when optimizing for size.
Reviewed By: mkazantsev
Differential Revision: https://reviews.llvm.org/D84108
TODO
* PrintIRInstrumentation and TimePassesHandler would be using this new callback.
* "Running pass" logging will also be moved to use this callback.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D84772
It was unclear what `isa` was supposed to mean so we did not provide any
traits for this context selector. With this patch we will allow *any*
string or identifier. We use the target attribute and target info to
determine if the trait matches. In other words, we will check if the
provided value is a target feature that is available (at the call site).
Fixes PR46338
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D83281
This patch teaches SCEVExpander to directly preserve LCSSA.
As it is currently, SCEV does not look through PHI nodes in loops,
as it might break LCSSA form. Once SCEVExpander can preserve
LCSSA form, it should be safe for SCEV to look through PHIs.
To preserve LCSSA form, this patch uses formLCSSAForInstructions
on operands of newly created instructions, if the definition is inside
a different loop than the new instruction.
The final value we return from expandCodeFor may also need LCSSA
phis, depending on the insert point. As no user for it exists there yet,
create a temporary instruction at the insert point, which can be passed
to formLCSSAForInstructions. This temporary instruction is removed
after LCSSA construction.
Reviewed By: mkazantsev
Differential Revision: https://reviews.llvm.org/D71538
A list of target features is disabled when there is no hardware
floating-point support. This is the case when one of the following
options is passed to clang:
- -mfloat-abi=soft
- -mfpu=none
This option list is missing, however, the extension "+nofp" that can be
specified in -march flags, such as "-march=armv8-a+nofp".
This patch also disables unsupported target features when nofp is passed
to -march.
Differential Revision: https://reviews.llvm.org/D82948
Currently, getCastInstrCost has limited information about the cast it's
rating, often just the opcode and types. Sometimes there is a context
instruction as well, but it isn't trustworthy: for instance, when the
vectorizer is rating a plan, it calls getCastInstrCost with the old
instructions when, in fact, it's trying to evaluate the cost of the
instruction post-vectorization. Thus, the current system can get the
cost of certain casts incorrect as the correct cost can vary greatly
based on the context in which it's used.
For example, if the vectorizer queries getCastInstrCost to evaluate the
cost of a sext(load) with tail predication enabled, getCastInstrCost
will think it's free most of the time, but it's not always free. On ARM
MVE, a VLD2 group cannot be extended like a normal VLDR can. Similar
situations can come up with how masked loads can be extended when being
split.
To fix that, this path adds a new parameter to getCastInstrCost to give
it a hint about the context of the cast. It adds a CastContextHint enum
which contains the type of the load/store being created by the
vectorizer - one for each of the types it can produce.
Original patch by Pierre van Houtryve
Differential Revision: https://reviews.llvm.org/D79162
cmake was still considering the empty value of ${fake_version_inc}
even if it was not defined.
Reviewed By: vsapsai
Differential Revision: https://reviews.llvm.org/D82847
We can implement find_first_unset_in() in the same function
if every BitWord we use is first flipped.
Differential Revision: https://reviews.llvm.org/D84717
Instead, pattern match extends of extract_subvectors to generate
widening operations. Since extract_subvector is not a legal node, this
is implemented via a custom combine that recognizes extract_subvector
nodes before they are legalized. The combine produces custom ISD nodes
that are later pattern matched directly, just like the intrinsic was.
Also removes the clang builtins for these operations since the
instructions can now be generated from portable code sequences.
Differential Revision: https://reviews.llvm.org/D84556
Substitutions are already reported in the diagnostics appearing before
the input dump in the case of failed directives, and they're reported
in traces (produced by `-vv -dump-input=never`) in the case of
successful directives. However, those reports are not always
convenient to view while investigating the input dump, so this patch
adds the substitution report to the input dump too. For example:
```
$ cat check
CHECK: hello [[WHAT:[a-z]+]]
CHECK: [[VERB]] [[WHAT]]
$ FileCheck -vv -DVERB=goodbye check < input |& tail -8
<<<<<<
1: hello world
check:1 ^~~~~~~~~~~
2: goodbye word
check:2'0 X~~~~~~~~~~~ error: no match found
check:2'1 with "VERB" equal to "goodbye"
check:2'2 with "WHAT" equal to "world"
>>>>>>
```
Without this patch, the location reported for a substitution for a
directive match is the directive's full match range. This location is
misleading as it implies the substitution itself matches that range.
This patch changes the reported location to just the match range start
to suggest the substitution is known at the start of the match. (As
in the above example, input dumps don't mark any range for
substitutions. The location info in that case simply identifies the
right line for the annotation.)
Reviewed By: mehdi_amini, thopre
Differential Revision: https://reviews.llvm.org/D83650
Summary:
Simplify ChildrenGetter to a simple wrapper around a GraphDiff call.
GraphDiff already handles nullptr in children, so the special casing in
clang can also be removed.
Reviewers: kuhar, dblaikie
Subscribers: llvm-commits, cfe-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D84713
The refactoring encapsulates frequency calculation in MachineBlockFrequencyInfo,
and renames the API to clarify its motivation. It should clarify
frequencies may not be reset 'freely' by users of the analysis, as the
API serves as a partial update to avoid a full analysis recomputation.
Differential Revision: https://reviews.llvm.org/D84427
This patch adds support for emitting custom range list content.
We are able to handcraft a custom range list via the following syntax.
```
debug_rnglists:
- Lists:
- Entries:
- Operator: DW_RLE_startx_endx
Values: [ 0x1234, 0x1234 ]
- Content: '1234567890abcdef'
- Content: 'abcdef1234567890'
```
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D84618
This adds the `ShType` key similar to others `Sh*` keys we have.
My use case is the following. Imagine we have a `SHT_SYMTAB_SHNDX`
section and want to hide it from a dumper. The natural way would be to
do something like:
```
- Name: .symtab_shndx
Type: [[TYPE=SHT_SYMTAB_SHNDX]]
Entries: [ 0, 1 ]
```
and then change the TYPE from `SHT_SYMTAB_SHNDX` to something else,
for example to `SHT_PROGBITS`.
But we have a problem: regular sections does not have `Entries` key,
so yaml2obj will be unable to produce a section.
The solution is to introduce a `ShType` key to override the final type.
This is not the first time I am facing the need to change the type. I
was able to invent workarounds or solved issues differently in the past,
but finally came to conclusion that we just should support the `ShType`.
Differential revision: https://reviews.llvm.org/D84738
The print() function in the `AbstractAttribute` structure overrides
the function in the `AADepGraphNode`, so we need to mark it as
override.
This should fix a buildbot failure introduced by 5ee07dc.
In order to facilitate review of D79485 here is a small NFC change which restructures code around handling of SCCs in BPI.
Reviewed By: davidxl
Differential Revision: https://reviews.llvm.org/D84514
This patch added dependency graph to the attributor so that we can dump the dependencies between AAs more easily. We can also apply general graph algorithms to the graph, making it easier for us to create deep wrappers.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D78861
This is recommit of f51bc4fb60, reverted in 8577595e03, because
the function `flock` is not available on Solaris. In this variant
`flock` was replaced with `fcntl`, which is a POSIX function.
New functions `lockFile`, `tryLockFile` and `unlockFile` implement
simple file locking. They lock or unlock entire file. This must be
enough to support simulataneous writes to log files in parallel builds.
Differential Revision: https://reviews.llvm.org/D78896
PGO profile is usually more precise than sample profile. However, PGO profile
needs to be collected from loadtest and loadtest may not be representative
enough to the production workload. Sample profile collected from production
can be used as a supplement -- for functions cold in loadtest but warm/hot
in production, we can scale up the related function in PGO profile if the
function is warm or hot in sample profile.
The implementation contains changes in compiler side and llvm-profdata side.
Given an instr profile and a sample profile, for a function cold in PGO
profile but warm/hot in sample profile, llvm-profdata will either mark
all the counters in the profile to be -1 or scale up the max count in the
function to be above hot threshold, depending on the zero counter ratio in
the profile. The assumption is if there are too many counters being zero
in the function profile, the profile is more likely to cause harm than good,
then llvm-profdata will mark all the counters to be -1 indicating the
function is hot but the profile is unaccountable. In compiler side, if a
function profile with all -1 counters is seen, the function entry count will
be set to be above hot threshold but its internal profile will be dropped.
In the long run, it may be useful to let compiler support using PGO profile
and sample profile at the same time, but that requires more careful design
and more substantial changes to make two profiles work seamlessly. The patch
here serves as a simple intermediate solution.
Differential Revision: https://reviews.llvm.org/D81981
Summary:
Refactor Succ/Pred maps to have a single map lookup when constructing
children. The preivous desing made sense when used by GraphTraits.
This more closely matches the previous approach in DomTree.
Reviewers: dblaikie
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84567
Summary:
Use getChildren() method in GraphDiff instead of GraphTraits.
This simplifies the code and allows for refactorigns inside GraphDiff.
All usecase need not have a light-weight/copyable range.
Clean GraphTraits implementation.
Reviewers: dblaikie
Subscribers: hiraditya, llvm-commits, george.burgess.iv
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84562
Summary:
This seems obvious in hindsight, but the result is surprising.
I've measured compile-time of `-openmpopt` pass standalone
on RawSpeed unity build, and while there is some OpenMP stuff,
most is not OpenMP. But nonetheless the pass does a lot of costly
preparations before ever trying to look for OpenMP stuff in SCC.
Numbers (n=25): 0.094624s -> 0.005976s, an -93.68% improvement, or ~16x
Reviewers: jdoerfert
Reviewed By: jdoerfert
Subscribers: yaxunl, hiraditya, guansong, llvm-commits, sstefan1
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84689
I have introduced a new TargetFrameLowering query function:
isStackIdSafeForLocalArea
that queries whether or not it is safe for objects of a given stack
id to be bundled into the local area. The default behaviour is to
always bundle regardless of the stack id, however for AArch64 this is
overriden so that it's only safe for fixed-size stack objects.
There is future work here to extend this algorithm for multiple local
areas so that SVE stack objects can be bundled together and accessed
from their own virtual base-pointer.
Differential Revision: https://reviews.llvm.org/D83859
Subclasses will commonly gather that information from a remote during
construction, in which case they won't have meaningful values to pass to
TargetProcessControl's constructor.
Adds a range-based version of `std::move`, the version that moves a range, not the one that creates r-value references.
Reviewed By: dblaikie, gamesh411
Differential Revision: https://reviews.llvm.org/D83902
This is the part of the patch that's moving the Updates to a CFGDiff
object. Splitting off from the clean-up work merging the two branches when BUI is null.
Differential Revision: https://reviews.llvm.org/D77341
This is the first of two patches to address PR46753. We basically allow
mem2reg to promote allocas that are used in doppable instructions, for
now that means `llvm.assume`. The uses of the alloca (or a bitcast or
zero offset GEP from there) are replaced by `undef` in the droppable
instructions.
Reviewed By: Tyker
Differential Revision: https://reviews.llvm.org/D83976
This patch removes the OMP_DIRECTIVE definition from OMPKinds.def since they
are now defined in OMP.td and OMP_DIRECTIVE is not used anymore in the code.
Reviewed By: jdenny
Differential Revision: https://reviews.llvm.org/D84329
Common up some existing MBB name printing logic into a single place.
Note that basic block dumping now prints the same set of attributes as
the MIRPrinter.
Change-Id: I8f022bbd922e831bc96d63143d7472c03282530b
Differential Revision: https://reviews.llvm.org/D83253
PassManager.h is one of the top headers in the ClangBuildAnalyzer frontend worst offenders list.
This exposes a large number of implicit dependencies on various forward declarations/includes in other headers that need addressing.
Rather than handling zlib handling manually, use find_package from CMake
to find zlib properly. Use this to normalize the LLVM_ENABLE_ZLIB,
HAVE_ZLIB, HAVE_ZLIB_H. Furthermore, require zlib if LLVM_ENABLE_ZLIB is
set to YES, which requires the distributor to explicitly select whether
zlib is enabled or not. This simplifies the CMake handling and usage in
the rest of the tooling.
This is a reland of abb0075 with all followup changes and fixes that
should address issues that were reported in PR44780.
Differential Revision: https://reviews.llvm.org/D79219
Widen or narrow a type to a type with the same scalar size as
another. This can be used to force G_PTR_ADD/G_PTRMASK's scalar
operand to match the bitwidth of the pointer type. Use this to
disallow narrower types for G_PTRMASK.
It is not useful to report WidenScalar for a pointer value, so always
report a scalar value with the target size. This allows using this to
clamp the scalar operand to the pointer size in operations like
G_PTR_ADD or G_PTRMASK.
The std::function itself was tested, not the result of the actual
query. It seems like there should be a warning for this. Inline the
check to avoid this.
This patch makes ownership of the JITLinkMemoryManager by ObjectLinkingLayer
optional: the layer can still own the memory manager but no longer has to.
Evevntually we want to move to a state where ObjectLinkingLayer never owns its
memory manager. For now allowing optional ownership makes it easier to develop
classes that can dynamically use either RTDyldObjectLinkingLayer, which owns
its memory managers, or ObjectLinkingLayer (e.g. LLJIT).
Rather than handling zlib handling manually, use find_package from CMake
to find zlib properly. Use this to normalize the LLVM_ENABLE_ZLIB,
HAVE_ZLIB, HAVE_ZLIB_H. Furthermore, require zlib if LLVM_ENABLE_ZLIB is
set to YES, which requires the distributor to explicitly select whether
zlib is enabled or not. This simplifies the CMake handling and usage in
the rest of the tooling.
This is a reland of abb0075 with all followup changes and fixes that
should address issues that were reported in PR44780.
Differential Revision: https://reviews.llvm.org/D79219
This adds the llvm.abs(), llvm.umin(), llvm.umax(), llvm.smin(),
and llvm.smax() intrinsics specified in D81829. For SelectionDAG,
the ISD opcodes and all the legalization and lowering already exist,
so this just wires them up to the intrinsic in the SDAG builder and
adds rudimentary tests. For GlobalISel only the min/max intrinsics
are wired up, as llvm.abs() will require the addition of a G_ABS op,
and corresponding legalization support.
Differential Revision: https://reviews.llvm.org/D84125
Clarify the relation between a block's BlockFrequency and the
getEntryFreq() API, and added an API for the relatively common case of
finding a block's frequency relative to the entrypoint.
Added / moved some comments to header.
Differential Revision: https://reviews.llvm.org/D84357
This patch refactors `emitDebugInfo()` to make the length field be
inferred from its content. Besides, the `Visitor` class is removed in
this patch. The original `Visitor` class helps us determine an
appropriate length and emit the .debug_info section. These two
processes can be merged into one process. Besides, the length field
should be inferred when it's missing rather than when it's zero.
Reviewed By: jhenderson, labath
Differential Revision: https://reviews.llvm.org/D84008
Currently there are plenty of instructions that SCEVExpander creates but
does not track as created. IRBuilder allows specifying a callback
whenever an instruction is inserted. Use this to call
rememberInstruction automatically for each created instruction.
There are still a few rememberInstruction calls remaining, because in
some cases Inst::Create functions are used to construct instructions.
Suggested by @lebedev.ri in D75980.
Reviewers: mkazantsev, reames, sanjoy.google, lebedev.ri
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D84326
Add support in LegalizerHelper for lowering G_SADDSAT etc. either
using add/subtract-with-overflow or using max/min instructions.
Enable this lowering for AMDGPU so it can be tested. The legalization
rules are still approximate and skips out on using the clamp bit to
treat these as legal, which has never been used before. This also
doesn't yet try to deal with expanding SALU cases.
Summary:
This is the next patch of [[ https://reviews.llvm.org/D76210 | D76210 ]].
This patch made a map in `InformationCache` for caching results.
Reviewers: jdoerfert, sstefan1, uenoku, homerdin, baziotis
Reviewed By: jdoerfert
Subscribers: hiraditya, uenoku, kuter, bbn, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83246
This reapplies commit d4020ef7c4, reverted in ac0edc5588 because it
broke build of LLDB. This commit contains appropriate changes for LLDB.
The original commit message is below.
Documentation on CreateProcessW states that maximal size of command line
is 32767 characters including ternimation null character. In the
function llvm::sys::commandLineFitsWithinSystemLimits this limit was set
to 32768. As a result if command line was exactly 32768 characters long,
a response file was not created and CreateProcessW was called with
too long command line.
Differential Revision: https://reviews.llvm.org/D83772
The revert was a misfire.
Remove the temporary flag PGSOIRPassOrTestOnly and the guard code which was used
for the staged rollout. This is a cleanup (NFC) as it's now false by default.
Differential Revision: https://reviews.llvm.org/D84057
This patch refactors the range list table to hold both the range list
table and the location list table.
Reviewed By: jhenderson, labath
Differential Revision: https://reviews.llvm.org/D84239
Summary:
This patch fix a problem where clause needed to be in the allowed set even
they were in the required set. A required clause is allowed obvisouly. This allow
to remove the duplicate in OMP.td
Reviewers: kiranchandramohan, DavidTruby, richard.barton.arm, jdoerfert, sscalpone, kiranktp, ichoyjx
Reviewed By: kiranchandramohan
Subscribers: yaxunl, guansong, sstefan1, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84353
TPCDynamicLibrarySearchGenerator uses a TargetProcessControl instance to
load libraries and search for symbol addresses in a target process. It
can be used in place of a DynamicLibrarySearchGenerator to enable
target-process agnostic lookup.
This reverts commit 4a539faf74.
There is a __llvm_profile_instrument_range related crash in PGO-instrumented clang:
```
(gdb) bt
llvm::ConstantRange const&, llvm::APInt const&, unsigned int, bool) ()
llvm::ScalarEvolution::getRangeForAffineAR(llvm::SCEV const*, llvm::SCEV
const*, llvm::SCEV const*, unsigned int) ()
```
(The body of __llvm_profile_instrument_range is inlined, so we can only find__llvm_profile_instrument_target in the trace)
```
23│ 0x000055555dba0961 <+65>: nopw %cs:0x0(%rax,%rax,1)
24│ 0x000055555dba096b <+75>: nopl 0x0(%rax,%rax,1)
25│ 0x000055555dba0970 <+80>: mov %rsi,%rbx
26│ 0x000055555dba0973 <+83>: mov 0x8(%rsi),%rsi # %rsi=-1 -> SIGSEGV
27│ 0x000055555dba0977 <+87>: cmp %r15,(%rbx)
28│ 0x000055555dba097a <+90>: je 0x55555dba0a76 <__llvm_profile_instrument_target+342>
```
Previously, the vins*vlx instructions were incorrectly defined with i64 as the
second argument. This patches fixes this issue by correcting the second argument
of the vins*vlx instructions/intrinsics to be i32.
Differential Revision: https://reviews.llvm.org/D84277
- Update documentation to clarify that `}` does not need to be doubled up.
- Update `EscapedBrace` test case to test this behavior
Differential Revision: https://reviews.llvm.org/D83888
This patch includes the supporting code that enables always
instrumenting the function entry block by default.
This patch will NOT the default behavior.
It adds a variant bit in the profile version, adds new directives in
text profile format, and changes llvm-profdata tool accordingly.
This patch is a split of D83024 (https://reviews.llvm.org/D83024)
Many test changes from D83024 are also included.
Differential Revision: https://reviews.llvm.org/D84261
This reverts commit e64afefdf8. It caused
a PGO bootstrapped clang to crash on many source files.
`__llvm_profile_instrument_range` seems to trigger a null pointer dereference.
Call stack:
__llvm_profile_instrument_range
llvm::APInt::udiv(llvm::APInt const&) const
getRangeForAffineARHelper
The implementation of the xvtlsbb builtins/intrinsics were not correct as the
intrinsics previously used i1 as an argument type. This patch changes the i1
argument type used in these intrinsics to be i32 instead, as having the second
as an i1 can lead to issues in the backend.
Differential Revision: https://reviews.llvm.org/D84291
We can just use the definition from config.h. This means we need to move
a few lines around in CMakeLists.txt - the TF_AOT detection needs to be
before the spot we process the config.h.cmake files.
Differential Revision: https://reviews.llvm.org/D84349
There's no reason to involve the hassle of a virtual method targets
have to override for a simple boolean.
Not sure exactly what's going on with Mips, but it seems to define its
own totally separate handler classes.
This was structured in a way that implied every split argument is in
memory, or in registers. It is possible to pass an original argument
partially in registers, and partially in memory. Transpose the logic
here to only consider a single piece at a time. Every individual
CCValAssign should be treated independently, and any merge to original
value needs to be handled later.
This is in preparation for merging some preprocessing hacks in the
AMDGPU calling convention lowering into the generic code.
I'm also not sure what the correct behavior for memlocs where the
promoted size is larger than the original value. I've opted to clamp
the memory access size to not exceed the value register to avoid the
explicit trunc/extend/vector widen/vector extract instruction. This
happens for AMDGPU for i8 arguments that end up stack passed, which
are promoted to i16 (I think this is a preexisting DAG bug though, and
they should not really be promoted when in memory).
(This reverts commit a5e0194709, and
corrects author).
Rename the pass to be able to extend it to function properties other than inliner features.
Reviewed By: mtrofin
Differential Revision: https://reviews.llvm.org/D82044
Rename the pass to be able to extend it to function properties other than inliner features.
Reviewed By: mtrofin
Differential Revision: https://reviews.llvm.org/D82044
Explain why you can only get a cached analysis result, not compute one
on the fly.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D84259
A linker optimization is available on PowerPC for GOT indirect PCRelative loads.
The idea is that we can mark a usual GOT indirect load:
pld 3, vec@got@pcrel(0), 1
lwa 3, 4(3)
With a relocation to say that if we don't need to go through the GOT we can let
the linker further optimize this and replace a load with a nop.
pld 3, vec@got@pcrel(0), 1
.Lpcrel1:
.reloc .Lpcrel1-8,R_PPC64_PCREL_OPT,.-(.Lpcrel1-8)
lwa 3, 4(3)
This patch adds the logic that allows the compiler to add the R_PPC64_PCREL_OPT.
Reviewers: nemanjai, lei, hfinkel, sfertile, efriedma, tstellar, grosbach
Reviewed By: nemanjai
Differential Revision: https://reviews.llvm.org/D79864
Summary:
AIX assembly's .set directive is not usable for aliasing purpose.
We need to use extra-label-at-defintion strategy to generate symbol
aliasing on AIX.
Reviewed By: DiggerLin, Xiangling_L
Differential Revision: https://reviews.llvm.org/D83252
For a long time, the InstCombine pass handled target specific
intrinsics. Having target specific code in general passes was noted as
an area for improvement for a long time.
D81728 moves most target specific code out of the InstCombine pass.
Applying the target specific combinations in an extra pass would
probably result in inferior optimizations compared to the current
fixed-point iteration, therefore the InstCombine pass resorts to newly
introduced functions in the TargetTransformInfo when it encounters
unknown intrinsics.
The patch should not have any effect on generated code (under the
assumption that code never uses intrinsics from a foreign target).
This introduces three new functions:
TargetTransformInfo::instCombineIntrinsic
TargetTransformInfo::simplifyDemandedUseBitsIntrinsic
TargetTransformInfo::simplifyDemandedVectorEltsIntrinsic
A few target specific parts are left in the InstCombine folder, where
it makes sense to share code. The largest left-over part in
InstCombineCalls.cpp is the code shared between arm and aarch64.
This allows to move about 3000 lines out from InstCombine to the targets.
Differential Revision: https://reviews.llvm.org/D81728
We don't need the StringsAndChecksumsRef forward declaration as we have to include StringsAndChecksums.h.
We don't need DebugSubsectionRecord.h and we forward declare all referenced classes.
We don't need to include cstdint as we don't use any stdint types.
Summary:
This patch reduces file size in debug builds by dropping variable locations a
debugger user will not see.
After building the debug entity history map we loop through it. For each
variable we look at each entry. If the entry opens a location range which does
not intersect any of the variable's scope's ranges then we mark it for removal.
After visiting the entries for each variable we also mark any clobbering
entries which will no longer be referenced for removal, and then finally erase
the marked entries. This all requires the ability to query the order of
instructions, so before this runs we number them.
Tests:
Added llvm/test/DebugInfo/X86/trim-var-locs.mir
Modified llvm/test/DebugInfo/COFF/register-variables.ll
Branch folding merges the tails of if.then and if.else into if.else. Each
blocks' debug-locations point to different scopes so when they're merged we
can't use either. Because of this the variable 'c' ends up with a location
range which doesn't cover any instructions in its scope; with the patch
applied the location range is dropped and its flag changes to IsOptimizedOut.
Modified llvm/test/DebugInfo/X86/live-debug-variables.ll
Modified llvm/test/DebugInfo/ARM/PR26163.ll
In both tests an out of scope location is now removed. The remaining location
covers the entire scope of the variable allowing us to emit it as a single
location.
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D82129
When the compiler generates a GOT indirect load it must generate two loads. One
that loads the address of the element from the GOT and a second to load the
actual element based on the address just loaded from the GOT. However, the
linker can optimize these two loads into one load if it knows that it is safe
to do so. The compiler can tell the linker that the optimization is safe
by using the R_PPC64_PCREL_OPT relocation.
This patch extends the .reloc directive to allow the following setup
pld 3, vec@got@pcrel(0), 1
.Lpcrel1=.-8
... More instructions possible here ...
.reloc .Lpcrel1,R_PPC64_PCREL_OPT,.-.Lpcrel1
lwa 3, 4(3)
Reviewers: nemanjai, lei, hfinkel, sfertile, efriedma, tstellar, grosbach, MaskRay
Reviewed By: nemanjai, MaskRay
Differential Revision: https://reviews.llvm.org/D79625
-Use the actual sect/offset to keep track of symbols in the cache so they don't get created multiple times with different addresses.
-Remove getSymTag from PDBFunctionSymbol/PDBPublicSymbol because it's already implemented in the base class
-Merge the symbolizer test files for DIA and native, since the tests are the same.
-Implement getCompilandId for NativeLineNumber
Reviewed By: amccarth
Differential Revision: https://reviews.llvm.org/D84208
There are a few questionable things about this intrinsic and existing
DAG implementation. For some reason the intrinsic hardcodes the second
operand to be scalar-only i32, and SelectionDAG builder makes a
legalization decision based on whether the operand is constant.
Outside of compiler-rt (where it's arguably an anti-pattern too),
LLVM tries to keep its build files as simple as possible. See e.g.
llvm/docs/SupportLibrary.rst, "Code Organization".
Differential Revision: https://reviews.llvm.org/D84243
OptNoneInstrumentation is part of StandardInstrumentations. It skips
functions (or loops) that are marked optnone.
The feature of skipping optional passes for optnone functions under NPM
is gated on a -enable-npm-optnone flag. Currently it is by default
false. That is because we still need to mark all required passes to be
required. Otherwise optnone functions will start having incorrect
semantics. After that is done in following changes, we can remove the
flag and always enable this.
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D83519
Documentation on CreateProcessW states that maximal size of command line
is 32767 characters including ternimation null character. In the
function llvm::sys::commandLineFitsWithinSystemLimits this limit was set
to 32768. As a result if command line was exactly 32768 characters long,
a response file was not created and CreateProcessW was called with
too long command line.
Differential Revision: https://reviews.llvm.org/D83772
In addition, move the definition of the class into the Debugify.h,
so we can use it from different levels.
The motivation for this is D82547.
Differential Revision: https://reviews.llvm.org/D83391
Replace `std::is_same<X, std::random_access_iterator_tag>` with `std::is_base_of<std::random_access_iterator_tag, X>` in STLExtra algos.
This doesn't have too much impact on LLVM internally as no structs derive from it.
However external projects embedding LLVM may use `std::contiguous_iterator_tag` which should be considered by these algorithms.
As well as any other potential tags people want to define derived from `std::random_access_iterator_tag`
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D84141
Try to make the behavior more consistent with getGCDType, and bias
towards returning something closer to the source type whenever there's
an ambiguity.
Try harder to find a canonical unmerge type when trying to cover the
desired target type. Handle finding a compatible unmerge type for two
vectors with different element types. This will return the largest
multiple of the source vector element that will evenly divide the
target vector type.
Also make the handling mixing scalars and vectors, and prefer the
source element type as the unmerge target type.
PTX does not support negative values in .bNN data directives and we must
typecast such values to unsigned before printing them.
MCAsmInfo can now specify whether such casting is necessary for particular
target.
Differential Revision: https://reviews.llvm.org/D83423
Remove the temporary flag PGSOIRPassOrTestOnly and the guard code which was used
for the staged rollout. This is a cleanup (NFC) as it's now false by default.
Differential Revision: https://reviews.llvm.org/D84057
Summary:
This is the InlineAdvisor used in 'development' mode. It enables two
scenarios:
- loading models via a command-line parameter, thus allowing for rapid
training iteration, where models can be used for the next exploration
phase without requiring recompiling the compiler. This trades off some
compilation speed for the added flexibility.
- collecting training logs, in the form of tensorflow.SequenceExample
protobufs. We generate these as textual protobufs, which simplifies
generation and testing. The protobufs may then be readily consumed by a
tensorflow-based training algorithm.
To speed up training, training logs may also be collected from the
'default' training policy. In that case, this InlineAdvisor does not
use a model.
RFC: http://lists.llvm.org/pipermail/llvm-dev/2020-April/140763.html
Reviewers: jdoerfert, davidxl
Subscribers: mgorny, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83733
This patch adds a TileInfo abstraction and utilities to
create a 3-level loop nest for tiling.
Reviewers: anemet
Reviewed By: anemet
Differential Revision: https://reviews.llvm.org/D77550
The PC Relative code now allows for calls that are marked with the relocation
R_PPC64_REL24_NOTOC. This indicates that the caller does not have a valid TOC
pointer in R2 and does not require R2 to be restored after the call.
This patch is added to support local calls to callees that require a TOC
Reviewed By: sfertile, MaskRay, nemanjai, stefanp
Differential Revision: https://reviews.llvm.org/D83504
Its effect could be achieved by
`-stop-after`,`-print-after`,`-print-after-all`. But a few tests need to
print MIR after ISel which could not be done with
`-print-after`/`-stop-after` since isel pass does not have commandline name.
That's the reason `--print-machineinstrs` is downgraded to
`--print-after-isel` in this patch. `--print-after-isel` could be
removed after we switch to new pass manager since isel pass would have a
commandline text name to use `print-after` or equivalent switches.
The motivation of this patch is to reduce tests dependency on
would-be-deprecated feature.
Reviewed By: arsenm, dsanders
Differential Revision: https://reviews.llvm.org/D83275
Summary:
This support is needed for the Fortran array variables with pointer/allocatable
attribute. This support enables debugger to identify the status of variable
whether that is currently allocated/associated.
for pointer array (before allocation/association)
without DW_AT_associated
(gdb) pt ptr
type = integer (140737345375288:140737354129776)
(gdb) p ptr
value requires 35017956 bytes, which is more than max-value-size
with DW_AT_associated
(gdb) pt ptr
type = integer (:)
(gdb) p ptr
$1 = <not associated>
for allocatable array (before allocation)
without DW_AT_allocated
(gdb) pt arr
type = integer (140737345375288:140737354129776)
(gdb) p arr
value requires 35017956 bytes, which is more than max-value-size
with DW_AT_allocated
(gdb) pt arr
type = integer, allocatable (:)
(gdb) p arr
$1 = <not allocated>
Testing
- unit test cases added
- check-llvm
- check-debuginfo
Reviewed By: aprantl
Differential Revision: https://reviews.llvm.org/D83544
This allows tracking the in-memory type of a pointer argument to a
function for ABI purposes. This is essentially a stripped down version
of byval to remove some of the stack-copy implications in its
definition.
This includes the base IR changes, and some tests for places where it
should be treated similarly to byval. Codegen support will be in a
future patch.
My original attempt at solving some of these problems was to repurpose
byval with a different address space from the stack. However, it is
technically permitted for the callee to introduce a write to the
argument, although nothing does this in reality. There is also talk of
removing and replacing the byval attribute, so a new attribute would
need to take its place anyway.
This is intended avoid some optimization issues with the current
handling of aggregate arguments, as well as fixes inflexibilty in how
frontends can specify the kernel ABI. The most honest representation
of the amdgpu_kernel convention is to expose all kernel arguments as
loads from constant memory. Today, these are raw, SSA Argument values
and codegen is responsible for turning these into loads.
Background:
There currently isn't a satisfactory way to represent how arguments
for the amdgpu_kernel calling convention are passed. In reality,
arguments are passed in a single, flat, constant memory buffer
implicitly passed to the function. It is also illegal to call this
function in the IR, and this is only ever invoked by a driver of some
kind.
It does not make sense to have a stack passed parameter in this
context as is implied by byval. It is never valid to write to the
kernel arguments, as this would corrupt the inputs seen by other
dispatches of the kernel. These argumets are also not in the same
address space as the stack, so a copy is needed to an alloca. From a
source C-like language, the kernel parameters are invisible.
Semantically, a copy is always required from the constant argument
memory to a mutable variable.
The current clang calling convention lowering emits raw values,
including aggregates into the function argument list, since using
byval would not make sense. This has some unfortunate consequences for
the optimizer. In the aggregate case, we end up with an aggregate
store to alloca, which both SROA and instcombine turn into a store of
each aggregate field. The optimizer never pieces this back together to
see that this is really just a copy from constant memory, so we end up
stuck with expensive stack usage.
This also means the backend dictates the alignment of arguments, and
arbitrarily picks the LLVM IR ABI type alignment. By allowing an
explicit alignment, frontends can make better decisions. For example,
there's real no advantage to an aligment higher than 4, so a frontend
could choose to compact the argument layout. Similarly, there is a
high penalty to using an alignment lower than 4, so a frontend could
opt into more padding for small arguments.
Another design consideration is when it is appropriate to expose the
fact that these arguments are all really passed in adjacent
memory. Currently we have a late IR optimization pass in codegen to
rewrite the kernel argument values into explicit loads to enable
vectorization. In most programs, unrelated argument loads can be
merged together. However, exposing this property directly from the
frontend has some disadvantages. We still need a way to track the
original argument sizes and alignments to report to the driver. I find
using some side-channel, metadata mechanism to track this
unappealing. If the kernel arguments were exposed as a single buffer
to begin with, alias analysis would be unaware that the padding bits
betewen arguments are meaningless. Another family of problems is there
are still some gaps in replacing all of the available parameter
attributes with metadata equivalents once lowered to loads.
The immediate plan is to start using this new attribute to handle all
aggregate argumets for kernels. Long term, it makes sense to migrate
all kernel arguments, including scalars, to be passed indirectly in
the same manner.
Additional context is in D79744.
This patch adds a new variant of the matrix lowering pass that only does
a minimal lowering and only depends on TTI. The main purpose of this pass
is to have a pass with minimal dependencies to run as part of the backend
pipeline.
At the moment, the only difference to the regular lowering pass is that it
does not support remarks. But in subsequent patches add support for tiling
to the lowering pass which will require more analysis, which we do not want
to run in the backend, as the lowering should happen in the middle-end in
practice and running it in the backend is mostly for convenience when
running llc.
Reviewers: anemet, Gerolf, efriedma, hfinkel
Reviewed By: anemet
Differential Revision: https://reviews.llvm.org/D76867
Add narrowScalarFor action.
Add narrow scalar for typeIndex == 0 for G_FPTOSI/G_FPTOUI.
Legalize using narrowScalarFor as s16->s32 G_FPTOSI/G_FPTOUI
followed by s32->s64 G_SEXT/G_ZEXT.
Differential Revision: https://reviews.llvm.org/D84010
There is a strange "feature" of the code: it handles all relocations as `Elf_Rela`.
For handling `Elf_Rel` it converts them to `Elf_Rela` and passes `bool IsRela` to
specify the real type everywhere.
A related issue is that the
`decode_relrs` helper in lib/Object has to return `Expected<std::vector<Elf_Rela>>`
because of that, though it could return a vector of `Elf_Rel`.
I think we should just start using templates for relocation types, it makes the code
cleaner and shorter. This patch does it.
Differential revision: https://reviews.llvm.org/D83871
Common code sinking is already guarded with a (with default-off!) flag,
so add a flag for hoisting, too.
D84108 will hopefully make hoisting off-by-default too.
This patch implements the .debug_rnglists section. We are able to
produce the .debug_rnglists section by the following syntax.
```
debug_rnglists:
- Format: DWARF32 ## Optional
Length: 0x1234 ## Optional
Version: 5 ## Optional
AddressSize: 0x08 ## Optional
SegmentSelectorSize: 0x00 ## Optional
OffsetEntryCount: 2 ## Optional
Offsets: [1, 2] ## Optional
Lists:
- Entries:
- Operator: DW_RLE_base_address
Values: [ 0x1234 ]
```
The generated .debug_rnglists is verified by llvm-dwarfdump, except for
the operator DW_RLE_startx_endx, since llvm-dwarfdump doesn't support
it.
Reviewed By: jhenderson
Differential Revision: https://reviews.llvm.org/D83624
This patch
- adds `canCreateUndefOrPoison`
- refactors `canCreatePoison` so it can deal with constantexprs
`canCreateUndefOrPoison` will be used at D83926.
Reviewed By: nikic, jdoerfert
Differential Revision: https://reviews.llvm.org/D84007
Matt Arsenault