The existing code assumed fcmp to always be an Instruction, but it can also be a ConstExpr.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D115450
There is a pointer to the DataLayout in SelectionDAGBuilder called
'DL' that is hardly ever used. In most cases the code seems to just
use `DAG.getDataLayout()` instead. Given that DL is also often used
as a shadowed variable for the debug location it seems sensible to
just kill off the few remaining uses and be consistent with the rest
of the code.
Differential Revision: https://reviews.llvm.org/D114451
After D113888 / 32b6c17b29079e7d the MMO size of a masked loads/store is
unknown. When we are converting back to a standard load/store because
the mask is known all ones, we can refine that to the correct size from
the size of the vector being loaded/stored.
Differential Revision: https://reviews.llvm.org/D114582
This patch extends LLVM IR to add metadata that can be used to emit macho files with two build version load commands.
It utilizes "darwin.target_variant.triple" and "darwin.target_variant.SDK Version" metadata names for that,
which will be set by a future patch in clang.
MachO uses two build version load commands to represent an object file / binary that is targeting both the macOS target,
and the Mac Catalyst target. At runtime, a dynamic library that supports both targets can be loaded from either a native
macOS or a Mac Catalyst app on a macOS system. We want to add support to this to upstream to LLVM to be able to build
compiler-rt for both targets, to finish the complete support for the Mac Catalyst platform, which is right now targetable
by upstream clang, but the compiler-rt bits aren't supported because of the lack of this multiple build version support.
Differential Revision: https://reviews.llvm.org/D112189
Add the calculation of a score, which will be used during ML training. The
score qualifies the quality of a regalloc policy, and is independent of
what we train (currently, just eviction), or the regalloc algo itself.
We can then use scores to guide training (which happens offline), by
formulating a reward based on score variation - the goal being lowering
scores (currently, that reward is percentage reduction relative to
Greedy's heuristic)
Currently, we compute the score by factoring different instruction
counts (loads, stores, etc) with the machine basic block frequency,
regardless of the instructions' provenance - i.e. they could be due to
the regalloc policy or be introduced previously. This is different from
RAGreedy::reportStats, which accummulates the effects of the allocator
alone. We explored this alternative but found (at least currently) that
the more naive alternative introduced here produces better policies. We
do intend to consolidate the two, however, as we are actively
investigating improvements to our reward function, and will likely want
to re-explore scoring just the effects of the allocator.
In either case, we want to decouple score calculation from allocation
algorighm, as we currently evaluate it after a few more passes after
allocation (also, because score calculation should be reusable
regardless of allocation algorithm).
We intentionally accummulate counts independently because it facilitates
per-block reporting, which we found useful for debugging - for instance,
we can easily report the counts indepdently, and then cross-reference
with perf counter measurements.
Differential Revision: https://reviews.llvm.org/D115195
A new basic block ordering improving existing MachineBlockPlacement.
The algorithm tries to find a layout of nodes (basic blocks) of a given CFG
optimizing jump locality and thus processor I-cache utilization. This is
achieved via increasing the number of fall-through jumps and co-locating
frequently executed nodes together. The name follows the underlying
optimization problem, Extended-TSP, which is a generalization of classical
(maximum) Traveling Salesmen Problem.
The algorithm is a greedy heuristic that works with chains (ordered lists)
of basic blocks. Initially all chains are isolated basic blocks. On every
iteration, we pick a pair of chains whose merging yields the biggest increase
in the ExtTSP value, which models how i-cache "friendly" a specific chain is.
A pair of chains giving the maximum gain is merged into a new chain. The
procedure stops when there is only one chain left, or when merging does not
increase ExtTSP. In the latter case, the remaining chains are sorted by
density in decreasing order.
An important aspect is the way two chains are merged. Unlike earlier
algorithms (e.g., based on the approach of Pettis-Hansen), two
chains, X and Y, are first split into three, X1, X2, and Y. Then we
consider all possible ways of gluing the three chains (e.g., X1YX2, X1X2Y,
X2X1Y, X2YX1, YX1X2, YX2X1) and choose the one producing the largest score.
This improves the quality of the final result (the search space is larger)
while keeping the implementation sufficiently fast.
Differential Revision: https://reviews.llvm.org/D113424
In the style of D113888, this patch updates the various VP memory
operations (load, store, gather, scatter) to use UnknownSize. This is
for the same reason as for masked loads and stores: the number of
elements accessed is not generally known at compile time.
This is somewhat pessimistic in the sense that we may still find
un-canonicalized intrinsics featuring both an all-true mask and an EVL
equal to the vector size. Arguably those should be canonicalized before
the SelectionDAG, so those have been left for future work.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D115036
This patch fixes a case where the 'align' parameter attribute on the
pointer operands to llvm.vp.load and llvm.vp.store was being dropped
during the conversion to the SelectionDAG. The default alignment
equal to the ABI type alignment of the vector type was kept. It also
updates the documentation to reflect the fact that the parameter
attribute is now properly supported.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D114422
LLVM loops cannot represent irreducible structures in the CFG. This
change introduce the concept of cycles as a generalization of loops,
along with a CycleInfo analysis that discovers a nested
hierarchy of such cycles. This is based on Havlak (1997), Nesting of
Reducible and Irreducible Loops.
The cycle analysis is implemented as a generic template and then
instatiated for LLVM IR and Machine IR. The template relies on a new
GenericSSAContext template which must be specialized when used for
each IR.
This review is a restart of an older review request:
https://reviews.llvm.org/D83094
Original implementation by Nicolai Hähnle <nicolai.haehnle@amd.com>,
with recent refactoring by Sameer Sahasrabuddhe <sameer.sahasrabuddhe@amd.com>
Differential Revision: https://reviews.llvm.org/D112696
DebugLoc is cheap to move, passing it by-val rather than const ref to
take advantage of the fact that it is consumed that way by the
MachineInstr ctor, which creates some optimization oportunities.
Differential Revision: https://reviews.llvm.org/D115208
The Pre-RA VLIWMachineScheduler used by Hexagon is a relatively generic
implementation that would make sense to use on other VLIW targets.
This commit lifts those classes into their own header/source file with the
root VLIWMachineScheduler. I chose this path rather than adding the
strategy et al. into MachineScheduler to avoid bloating the file with other
implementations.
Target-specific behaviors have been captured and replicated through
function overloads.
- Added an overloadable DFAPacketizer creation member function. This is
mainly done for our downstream, which has the capability to override
the DFAPacketizer with custom implementations. This is an upstreamable
TODO on our end. Currently, it always returns the result of
TargetInstrInfo::CreateTargetScheduleState
- Added an extra helper which returns the number of instructions in the
current packet. This is used in our downstream, and may be useful
elsewhere.
- Placed the priority heuristic values into the ConvergingVLIWscheduler
class instead of defining them as local statics in the implementation
- Added a overridable helper in ConvergingVLIWScheduler so that targets
can create their own VLIWResourceModel
Differential Revision: https://reviews.llvm.org/D113150
A new basic block ordering improving existing MachineBlockPlacement.
The algorithm tries to find a layout of nodes (basic blocks) of a given CFG
optimizing jump locality and thus processor I-cache utilization. This is
achieved via increasing the number of fall-through jumps and co-locating
frequently executed nodes together. The name follows the underlying
optimization problem, Extended-TSP, which is a generalization of classical
(maximum) Traveling Salesmen Problem.
The algorithm is a greedy heuristic that works with chains (ordered lists)
of basic blocks. Initially all chains are isolated basic blocks. On every
iteration, we pick a pair of chains whose merging yields the biggest increase
in the ExtTSP value, which models how i-cache "friendly" a specific chain is.
A pair of chains giving the maximum gain is merged into a new chain. The
procedure stops when there is only one chain left, or when merging does not
increase ExtTSP. In the latter case, the remaining chains are sorted by
density in decreasing order.
An important aspect is the way two chains are merged. Unlike earlier
algorithms (e.g., based on the approach of Pettis-Hansen), two
chains, X and Y, are first split into three, X1, X2, and Y. Then we
consider all possible ways of gluing the three chains (e.g., X1YX2, X1X2Y,
X2X1Y, X2YX1, YX1X2, YX2X1) and choose the one producing the largest score.
This improves the quality of the final result (the search space is larger)
while keeping the implementation sufficiently fast.
Differential Revision: https://reviews.llvm.org/D113424
Expanding on D109750.
Since `DBG_VALUE` instructions have final register validity determined in
`LDVImpl::handleDebugValue`, there is no apparent reason to immediately prune
unused register operands as their defs are erased. Consequently, this renders
`MachineInstr::eraseFromParentAndMarkDBGValuesForRemoval` moot; gaining a
substantial performance improvement.
The only necessary changes involve making relevant passes consider invalid
DBG_VALUE vregs uses as valid.
Reviewed By: MatzeB
Differential Revision: https://reviews.llvm.org/D112852
This patch proposes to move emission of global variables, types,
imported entities, etc from DwarfDebug::beginModule() to DwarfDebug::endModule().
Effectively, this changes nothing but the order of debug entities which
will be as follows:
* subprograms (including related context, local variables/labels,
local imported entities; related types can be created as a part of
the emission of local entities of an abstract subprogram);
* global variables (including related context and types);
* retained types and enums;
* non-local-scoped imported entities;
* basic types;
* other types left (as a part of local variables attributes emission).
Note that the order of emitted compile units may also be changed as now we emit
units that contain subprograms first and then all other non-empty units.
The motivation behind this change is the following:
(1) DwarfDebug::beginModule() is run at the very beginning of backend's pipeline,
from this time IR can be significantly changed by target-specific passes.
If it happens for debug metadata of global entities, those changes will not
be reflected in the emitted DWARF.
(2) imported subprogram names should refer to an abstract subprogram if it exists,
but it isn't known in DwarfDebug::beginModule() (it's possible to make some
guesses based on location info, but it's not quite reliable);
(3) aforementioned entities if they are scoped within a bracketed block
(subject of D113741) couldn't be emitted in DwarfDebug::beginModule()
(they need parent emitted first). Another problem is if to try to gather
some information about local entities and defer their emission
(till subprogram's processing or DwarfDebug::endModule()) all the gathered
details might be irrelevant / invalid by the time the entities are being
emitted (because of (1)).
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D114705
As an extension to D111976, this converts clamp fptosi, clamped between
0 and (2^n)-1 to a fptoui.sat. This can greatly help on targets with
conversions that naturally saturate, such as Arm.
X86 disables the transform as some of the test cases increases in size.
A fptoui.sat necessitates a fp clamp without native support, so there is
little use in converting if the instruction is just going to be
expanded.
Differential Revision: https://reviews.llvm.org/D112428
This reverts commits
* ee691970a9a85470948ada623c31f0ab8773617c (D113741),
* 79d3132998b2828be8f7d2ec411f91fb11b3e01f (D114705)
due to lldb and dexter test failures.
This patch proposes to move emission of global variables, types,
imported entities, etc from DwarfDebug::beginModule() to DwarfDebug::endModule().
Effectively, this changes nothing but the order of debug entities which
will be as follows:
* subprograms (including related context, local variables/labels,
local imported entities; related types can be created as a part of
the emission of local entities of an abstract subprogram);
* global variables (including related context and types);
* retained types and enums;
* non-local-scoped imported entities;
* basic types;
* other types left (as a part of local variables attributes emission).
Note that the order of emitted compile units may also be changed as now we emit
units that contain subprograms first and then all other non-empty units.
The motivation behind this change is the following:
(1) DwarfDebug::beginModule() is run at the very beginning of backend's pipeline,
from this time IR can be significantly changed by target-specific passes.
If it happens for debug metadata of global entities, those changes will not
be reflected in the emitted DWARF.
(2) imported subprogram names should refer to an abstract subprogram if it exists,
but it isn't known in DwarfDebug::beginModule() (it's possible to make some
guesses based on location info, but it's not quite reliable);
(3) aforementioned entities if they are scoped within a bracketed block
(subject of D113741) couldn't be emitted in DwarfDebug::beginModule()
(they need parent emitted first). Another problem is if to try to gather
some information about local entities and defer their emission
(till subprogram's processing or DwarfDebug::endModule()) all the gathered
details might be irrelevant / invalid by the time the entities are being
emitted (because of (1)).
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D114705
Prior to this patch, tail duplication handled debug info poorly -
specifically, debug instructions would be dropped instead of being set
undef, potentially extending the lifetimes of prior debug values that
should be killed. The pass was also very aggressive with dropping debug
info, dropping debug info even when the SSA value it referred to was
still present. This patch attempts to handle debug info more carefully,
checking to see whether each affected debug value can still be live,
setting it undef if not.
Reviewed By: jmorse
Differential Revision: https://reviews.llvm.org/D106875
MVE can treat v16i1, v8i1, v4i1 and v2i1 as different views onto the
same 16bit VPR.P0 register, with v2i1 holding two 8 bit values for the
two halves. This was never treated as a legal type in llvm in the past
as there are not many 64bit instructions and no 64bit compares. There
are a few instructions that could use it though, notably a VSELECT (as
it can handle any size using the underlying v16i8 VPSEL), AND/OR/XOR for
similar reasons, some gathers/scatter and long multiplies and VCTP64
instructions.
This patch goes through and makes v2i1 a legal type, handling all the
cases that fall out of that. It also makes VSELECT legal for v2i64 as a
side benefit. A lot of the codegen changes as a result - usually in way
that is a little better or a little worse, but still expensive. Costs
can change a little too in the process, again in a way that expensive
things remain expensive. A lot of the tests that changed are mainly to
ensure correctness - the code can hopefully be improved in the future
where it comes up in practice.
The intrinsics currently remain using the v4i1 they previously did to
emulate a v2i1. This will be changed in a followup patch but this one
was already large enough.
Differential Revision: https://reviews.llvm.org/D114449
This patch begins extending handling for peeking through bitcast nodes to big-endian targets as well as the existing little-endian case.
Differential Revision: https://reviews.llvm.org/D114676
combinePMULH currently only truncates vXi32/vXi64 multiplies to PMULHW/PMULUW if the source operands are SEXT/ZEXT instructions for a 'free' truncation.
But we can generalize this to any source operand with sufficient leading sign/zero bits that would allow PACKS/PACKUS to be used as a 'cheap' truncation.
This helps us avoid the wider multiplies, in exchange for truncation on both source operands instead of the result.
Differential Revision: https://reviews.llvm.org/D113371
This patch implements a new MachineFunction in the ARM backend for
placing BTI instructions. It is similar to the existing AArch64
aarch64-branch-targets pass.
BTI instructions are inserted into basic blocks that:
- Have their address taken
- Are the entry block of a function, if the function has external
linkage or has its address taken
- Are mentioned in jump tables
- Are exception/cleanup landing pads
Each BTI instructions is placed in the beginning of a BB after the
so-called meta instructions (e.g. exception handler labels).
Each outlining candidate and the outlined function need to be in agreement about
whether BTI placement is enabled or not. If branch target enforcement is
disabled for a function, the outliner should not covertly enable it by emitting
a call to an outlined function, which begins with BTI.
The cost mode of the outliner is adjusted to account for the extra BTI
instructions in the outlined function.
The ARM Constant Islands pass will maintain the count of the jump tables, which
reference a block. A `BTI` instruction is removed from a block only if the
reference count reaches zero.
PAC instructions in entry blocks are replaced with PACBTI instructions (tests
for this case will be added in a later patch because the compiler currently does
not generate PAC instructions).
The ARM Constant Island pass is adjusted to handle BTI
instructions correctly.
Functions with static linkage that don't have their address taken can
still be called indirectly by linker-generated veneers and thus their
entry points need be marked with BTI or PACBTI.
The changes are tested using "LLVM IR -> assembly" tests, jump tables
also have a MIR test. Unfortunately it is not possible add MIR tests
for exception handling and computed gotos because of MIR parser
limitations.
This patch is part of a series that adds support for the PACBTI-M extension of
the Armv8.1-M architecture, as detailed here:
https://community.arm.com/arm-community-blogs/b/architectures-and-processors-blog/posts/armv8-1-m-pointer-authentication-and-branch-target-identification-extension
The PACBTI-M specification can be found in the Armv8-M Architecture Reference
Manual:
https://developer.arm.com/documentation/ddi0553/latest
The following people contributed to this patch:
- Mikhail Maltsev
- Momchil Velikov
- Ties Stuij
Reviewed By: ostannard
Differential Revision: https://reviews.llvm.org/D112426
The REM DAG combine uses the visitDivLike functions to try and get an
optimized DIV node to provide better codegen, however in some cases this
visitDivLike call ends up in the BuildSDIVPow2 target hook, which in
turn sometimes will return the same node passed in to indicate not to
change it. The REM DAG combine does not anticipate this and creates a
cycle in the DAG because of it.
Fix this by ensuring any such optimized div node returned is distinct
from the node being combined.
Differential Revision: https://reviews.llvm.org/D114716
