We have SelectionDAG patterns for 8 & 16-bit atomic operations, but they
assume the value types will have been legalized to 32-bits. So this adds
the ability to widen them to both AArch64 & generic GISel
infrastructure.
Although this combine checks that there's no load folding barriers between
the loads that it's trying to merge, it was inserting the load at the
MIRBuilder's default insertion point, which is the G_OR use inst.
This was causing a miscompile in the test suite's
SingleSource/Regression/C/gcc-c-torture/execute/GCC-C-execute-bswap-2
Differential Revision: https://reviews.llvm.org/D106251
llvm::KnownBits::byteSwap() and reverse() don't modify in-place, so
we weren't actually computing anything. This was causing a miscompile on an
arm64 stage2 bootstrap clang build.
s56 stores are broken down into s32 + s24 stores. During this step
both of those new stores use an anyextended s64 value, resulting in
truncating stores. With s56, the s24 requires another lower step to
make it legal, and we were crashing because we didn't expect non-pow-2
stores to also be truncating as well.
Differential Revision: https://reviews.llvm.org/D106183
Since we're still building on top of the MVT based infrastructure, we
need to track the pointer type/address space on the side so we can end
up with the correct pointer LLTs when interpreting CCValAssigns.
This adds some level of type safety, allows helper functions to be added for
specific opcodes for free, and also allows us to succinctly check for class
membership with the usual dyn_cast/isa/cast functions.
To start off with, add variants for the different load/store operations with some
places using it.
Differential Revision: https://reviews.llvm.org/D105751
This fixes not respecting signext/zeroext in these cases. In the
anyext case, this avoids a larger merge with undef and should be a
better canonical form.
This should also handle this if a merge is needed, but I'm not aware
of a case where that can happen. In a future change this will also
allow AMDGPU to drop some custom code without introducing regressions.
Generalize the existing eq/ne case using `extractParts`. The original code only
handled narrowings for types of width 2n->n. This generalization allows for any
type that can be broken down by `extractParts`.
General overview is:
- Loop over each narrow-sized part and do exactly what the 2-register case did.
- Loop over the leftover-sized parts and do the same thing
- Widen the leftover-sized XOR results to the desired narrow size
- OR that all together and then do the comparison against 0 (just like the old
code)
This shows up a lot when building clang for AArch64 using GlobalISel, so it's
worth fixing. For the sake of simplicity, this doesn't handle the non-eq/ne
case yet.
Also remove the code in this case that notifies the observer; we're just going
to delete MI anyway so talking to the observer shouldn't be necessary.
Differential Revision: https://reviews.llvm.org/D105161
The original motivation for this was to implement moreElementsVector of shuffles
on AArch64, which resulted in complex sequences of artifacts like unmerge(unmerge(concat...))
which the combiner couldn't handle. It seemed here that the better option,
instead of writing ever-more-complex combines, was to have a way to find
the original "non-artifact" source registers for a given definition, walking
through arbitrary expressions of unmerge/concat/insert. As long as the bits
aren't extended or truncated, this is a pretty simple algorithm that avoids
the need for lots of combines and instead jumps straight to the final result
we want.
I've only used this new technique in 2 places within tryCombineUnmerge, using it
in more general situations resulted in infinite loops in AMDGPU. So for now
it's used when we would otherwise fail to combine and that seems to work.
In order to support looking through G_INSERTs, I also had to add it as an
artifact in isArtifact(), which caused a whole lot of issues in tests. AMDGPU
started infinite looping since full legalization of G_INSERT doensn't seem to
be there. To work around this, I've temporarily added a CLI option to use the
old behaviour so that the MIR tests will still run and terminate.
Other minor changes include no longer making >128b G_MERGE/UNMERGE legal.
We never had isel support for that anyway and it was a remnant of the legacy
legalizer rules. However being legal prevented the combiner from checking if it
was dead and deleting them.
Differential Revision: https://reviews.llvm.org/D104355
`LegalizerHelper::insertParts` uses `extractGCDType` on registers split into
a desired type and a smaller leftover type. This is used to populate a list
of registers. Each register in the list will have the same type as returned by
`extractGCDType`.
If we have
- `ResultTy` = s792
- `PartTy` = s64
- `LeftoverTy` = s24
When we call `extractGCDType`, we'll end up with two different types appended
to the list:
Part: gcd(792, 64, 24) => s8
Leftover: gcd(792, 24, 24) => s24
When this happens, we'll hit an assert while trying to build a G_MERGE_VALUES.
This patch changes the code for the leftover type so that we reuse the GCD from
the desired type.
e.g.
Leftover: gcd(792, 8, 24) => s8
https://llvm.godbolt.org/z/137Kqxj6j
Differential Revision: https://reviews.llvm.org/D105674
SelectionDAG's equivalents in ISD::InputArg/OutputArg track the
original argument index. Mips relies on this, and its currently
reinventing its own parallel CallLowering infrastructure which tracks
these indexes on the side. Add this to help move towards deleting the
custom mips handling.
This patch prevents GlobalISel from optimizing out redundant branch
instructions when compiling without optimizations.
The motivating example is code like the following common pattern in
Swift, where users expect to be able to set a breakpoint on the early
exit:
public func f(b: Bool) {
guard b else {
return // I would like to set a breakpoint here.
}
...
}
The patch modifies two places in GlobalISEL: The first one is in
IRTranslator.cpp where the removal of redundant branches is made
conditional on the optimization level. The second one is in
AArch64InstructionSelector.cpp where an -O0 *only* optimization is
being removed.
Disabling these optimizations increases code size at -O0 by
~8%. However, doing so improves debuggability, and debug builds are
the primary reason why developers compile without optimizations. We
thus concluded that this is the right trade-off.
rdar://79515454
Differential Revision: https://reviews.llvm.org/D105238
We're trying to match a few pointer computation patterns here for
re-association opportunities.
1) Isolating a constant operand to be on the RHS, e.g.:
G_PTR_ADD(BASE, G_ADD(X, C)) -> G_PTR_ADD(G_PTR_ADD(BASE, X), C)
2) Folding two constants in each sub-tree as long as such folding
doesn't break a legal addressing mode.
G_PTR_ADD(G_PTR_ADD(BASE, C1), C2) -> G_PTR_ADD(BASE, C1+C2)
AArch64 code size improvements on CTMark with -Os:
Program before after diff
pairlocalalign 251048 251044 -0.0%
consumer-typeset 421820 421812 -0.0%
kc 431348 431320 -0.0%
SPASS 413404 413300 -0.0%
clamscan 384396 384220 -0.0%
tramp3d-v4 370640 370412 -0.1%
lencod 432096 431772 -0.1%
bullet 479400 478796 -0.1%
sqlite3 288504 288072 -0.1%
7zip-benchmark 573796 570768 -0.5%
Geomean difference -0.1%
Differential Revision: https://reviews.llvm.org/D105069
In `IRTranslator::translateGetElementPtr`, when we run into a vector gep with
some scalar operands, we try to normalize those operands using
`buildSplatVector`.
This is fine except for when the getelementptr has a <1 x N> type. In that case
it is treated as a scalar. If we run into one of these then every call to
```
// With VectorWidth = 1
LLT::fixed_vector(VectorWidth, PtrTy)
```
will assert.
Here's an example (equivalent to the added testcase):
https://godbolt.org/z/hGsTnMYdW
To get around this, this patch adds a variable, `WantSplatVector`, which
is true when our vector type ought to actually be represented using a vector.
When it's false, we'll translate as a scalar. This checks if `VectorWidth > 1`.
This fixes this bug:
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=35496
Differential Revision: https://reviews.llvm.org/D105316
This enables proper lowering of non-byte sized loads. We still aren't
faithfully preserving memory types everywhere, so the legality checks
still only consider the size.
Previously we didn't preserve the memory type and had to blindly
interpret a number of bytes. Now that non-byte memory accesses are
representable, we can handle these correctly.
Ported from DAG version (minus some weird special case i1 legality
checking which I don't fully understand, and we don't have a way to
query for)
For now, this is NFC and the test changes are placeholders. Since the
legality queries are still relying on byte-flattened memory sizes, the
legalizer can't actually see these non-byte accesses. This keeps this
change self contained without merging it with the larger patch to
switch to LLT memory queries.
GlobalISel is relying on regular MachineMemOperands to track all of
the memory properties of accesses. Just the raw byte size is
insufficent to disambiguate all situations. For example, if we need to
split an unaligned extending load, we need to know the number of bits
in the original source value and can't infer it from the result
type. This is also a problem for extending vector loads.
This does decrease the maximum representable size from the full
uint64_t bytes to a maximum of 16-bits. No in tree testcases hit this,
other than places using UINT64_MAX for unknown sizes. This may be an
issue for G_MEMCPY and co., although they can just use unknown size
for large static sizes. This also has potential for backend abuse by
relying on the type when it really shouldn't be relevant after
selection.
This does not include the necessary MIR printer/parser changes to
represent this.
This patch relands https://reviews.llvm.org/D104454, but fixes some failing
builds on Mac OS which apparently has a different definition for size_t,
that caused 'ambiguous operator overload' for the implicit conversion
of TypeSize to a scalar value.
This reverts commit b732e6c9a8.
Adds legalizer, register bank select, and instruction
select support for G_SBFX and G_UBFX. These opcodes generate
scalar or vector ALU bitfield extract instructions for
AMDGPU. The instructions allow both constant or register
values for the offset and width operands.
The 32-bit scalar version is expanded to a sequence that
combines the offset and width into a single register.
There are no 64-bit vgpr bitfield extract instructions, so the
operations are expanded to a sequence of instructions that
implement the operation. If the width is a constant,
then the 32-bit bitfield extract instructions are used.
Moved the AArch64 specific code for creating G_SBFX to
CombinerHelper.cpp so that it can be used by other targets.
Only bitfield extracts with constant offset and width values
are handled currently.
Differential Revision: https://reviews.llvm.org/D100149
To reflect that the size may be scalable, a TypeSize is returned
instead of an unsigned. In places where the result is used,
it currently relies on an implicit cast of TypeSize -> uint64_t,
which asserts that the type is not scalable.
This patch is NFC for fixed-width vectors.
Reviewed By: aemerson
Differential Revision: https://reviews.llvm.org/D104454
This also adds new interfaces for the fixed- and scalable case:
* LLT::fixed_vector
* LLT::scalable_vector
The strategy for migrating to the new interfaces was as follows:
* If the new LLT is a (modified) clone of another LLT, taking the
same number of elements, then use LLT::vector(OtherTy.getElementCount())
or if the number of elements is halfed/doubled, it uses .divideCoefficientBy(2)
or operator*. That is because there is no reason to specifically restrict
the types to 'fixed_vector'.
* If the algorithm works on the number of elements (as unsigned), then
just use fixed_vector. This will need to be fixed up in the future when
modifying the algorithm to also work for scalable vectors, and will need
then need additional tests to confirm the behaviour works the same for
scalable vectors.
* If the test used the '/*Scalable=*/true` flag of LLT::vector, then
this is replaced by LLT::scalable_vector.
Reviewed By: aemerson
Differential Revision: https://reviews.llvm.org/D104451
Since this method can apply to cmpxchg operations, make sure it's clear
what value we're actually retrieving. This will help ensure we don't
accidentally ignore the failure ordering of cmpxchg in the future.
We could potentially introduce a getOrdering() method on AtomicSDNode
that asserts the operation isn't cmpxchg, but not sure that's
worthwhile.
Differential Revision: https://reviews.llvm.org/D103338
This only applies to FastIsel. GlobalIsel seems to sidestep
the issue.
This fixes https://bugs.llvm.org/show_bug.cgi?id=46996
One of the things we do in llvm is decide if a type needs
consecutive registers. Previously, we just checked if it
was an array or not.
(plus an SVE specific check that is not changing here)
This causes some confusion when you arbitrary IR like:
```
%T1 = type { double, i1 };
define [ 1 x %T1 ] @foo() {
entry:
ret [ 1 x %T1 ] zeroinitializer
}
```
We see it is an array so we call CC_AArch64_Custom_Block
which bails out when it sees the i1, a type we don't want
to put into a block.
This leaves the location of the double in some kind of
intermediate state and leads to odd codegen. Which then crashes
the backend because it doesn't know how to implement
what it's been asked for.
You get this:
```
renamable $d0 = FMOVD0
$w0 = COPY killed renamable $d0
```
Rather than this:
```
$d0 = FMOVD0
$w0 = COPY $wzr
```
The backend knows how to copy 64 bit to 64 bit registers,
but not 64 to 32. It can certainly be taught how but the real
issue seems to be us even trying to assign a register block
in the first place.
This change makes the logic of
AArch64TargetLowering::functionArgumentNeedsConsecutiveRegisters
a bit more in depth. If we find an array, also check that all the
nested aggregates in that array have a single member type.
Then CC_AArch64_Custom_Block's assumption of a type that looks
like [ N x type ] will be valid and we get the expected codegen.
New tests have been added to exercise these situations. Note that
some of the output is not ABI compliant. The aim of this change is
to simply handle these situations and not to make our processing
of arbitrary IR ABI compliant.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D104123
<string> is currently the highest impact header in a clang+llvm build:
https://commondatastorage.googleapis.com/chromium-browser-clang/llvm-include-analysis.html
One of the most common places this is being included is the APInt.h header, which needs it for an old toString() implementation that returns std::string - an inefficient method compared to the SmallString versions that it actually wraps.
This patch replaces these APInt/APSInt methods with a pair of llvm::toString() helpers inside StringExtras.h, adjusts users accordingly and removes the <string> from APInt.h - I was hoping that more of these users could be converted to use the SmallString methods, but it appears that most end up creating a std::string anyhow. I avoided trying to use the raw_ostream << operators as well as I didn't want to lose having the integer radix explicit in the code.
Differential Revision: https://reviews.llvm.org/D103888
G_INSERT legalization is incomplete and doesn't work very
well. Instead try to use sequences of G_MERGE_VALUES/G_UNMERGE_VALUES
padding with undef values (although this can get pretty large).
For the case of load/store narrowing, this is still performing the
load/stores in irregularly sized pieces. It might be cleaner to split
this down into equal sized pieces, and rely on load/store merging to
optimize it.
When narrowing G_ADD and G_SUB, handle types that aren't a multiple of
the type we're narrowing to. This allows us to handle types like s96
on 64 bit targets.
Note that the test here has a couple of dead instructions because of
the way the setup legalizes. I wasn't able to come up with a way to
write this test that avoids that easily.
Differential Revision: https://reviews.llvm.org/D97811
When narrowing G_INSERT, handle types that aren't a multiple of the
type we're narrowing to. This comes up if we're narrowing something
like an s96 to fit in 64 bit registers and also for non-byte multiple
packed types if they come up.
This implementation handles these cases by extending the extra bits to
the narrow size and truncating the result back to the destination
size.
Differential Revision: https://reviews.llvm.org/D97791
It's still in use in a few places so we can't delete it yet but there's not
many at this point.
Differential Revision: https://reviews.llvm.org/D103352
Also changes the fewerElements helper to use the lookthrough constant helper
instead of m_ICst, since m_ICst doesn't look through extends.
Differential Revision: https://reviews.llvm.org/D103227
Adjusting the load register type is a widenScalar type action, not a
lowering. lowerLoad should be reserved for operations that change the
memory access size, such as unaligned load decomposition. With this
trying to adjust the register type, it was hard to avoid infinite
loops in the legalizer. Adds a bandaid to avoid regressing a few
AArch64 tests, but I'm not sure what the exact condition is and
there's probably a cleaner way to do this.
For AMDGPU this regresses handling of some cases for unaligned loads,
but the way this is currently working is a pretty ugly hack.
Thhis is a port from the DAG legalization. We're still missing some of the
canonicalizations of shuffles but it's a start.
Differential Revision: https://reviews.llvm.org/D102828
There were a bunch of lost debug location remarks that show up when legalizing
tail calls on AArch64.
This would happen because we drop the return in the block where we emit the
tail call. So, we end up dropping the debug location, which makes the
LostDebugLocObserver report a missing debug location.
Although it's *true* that we lose these debug locations, this isn't
a particularly useful remark. We expect to drop these debug locations when
emitting tail calls. Suppressing remarks in this case is preferable, since the
amount of noise could hide actual debug location related bugs.
To do this, I just plumbed the LostDebugLocObserver through the relevant
LegalizerHelper functions. This is the only case I can think of where we need
the LostDebugLocObserver in the LegalizerHelper. So, rather than storing it
in the LegalizerHelper proper and mucking around with the constructors, I
figured it'd be cleanest to take the simplest path for now.
This clears up ~20 noisy lost debug location remarks on CTMark in AArch64 at
-Os.
Differential Revision: https://reviews.llvm.org/D103128
The function `reduceOperationWidth` helps to legalize a vector
operation either by narrowing its type or by scalarizing the
operation itself. It currently supports instructions with one result.
This patch, in addition allows the same for instructions with two
results (for instance, G_SDIVREM).
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D100725
Currently, variadic dbg.values (i.e. those using a DIArgList as part of
their location) are not handled properly by FastISel or GlobalISel, and
will produce invalid DBG_VALUE instructions if they encounter them. This
patch fixes this issue by emitting undef DBG_VALUE instructions for
variadic dbg.values, so that no incorrect instruction is produced and
any prior variable location is terminated.
This is simply a quick-fix to prevent errors; a correct implementation
should come later for these ISel pipelines to ensure that we do not drop
debug information unnecessarily.
Differential Revision: https://reviews.llvm.org/D102500
Use existing KnownBits helpers from KnownBits.h to simplify G_ICMPs.
E.g.
x == x -> true
x != x -> false
load(x) > 1 -> true (when the load is known to be greater than 1)
And so on.
Differential Revision: https://reviews.llvm.org/D102542
This extends any frame record created in the function to include that
parameter, passed in X22.
The new record looks like [X22, FP, LR] in memory, and FP is stored with 0b0001
in bits 63:60 (CodeGen assumes they are 0b0000 in normal operation). The effect
of this is that tools walking the stack should expect to see one of three
values there:
* 0b0000 => a normal, non-extended record with just [FP, LR]
* 0b0001 => the extended record [X22, FP, LR]
* 0b1111 => kernel space, and a non-extended record.
All other values are currently reserved.
If compiling for arm64e this context pointer is address-discriminated with the
discriminator 0xc31a and the DB (process-specific) key.
There is also an "i8** @llvm.swift.async.context.addr()" intrinsic providing
front-ends access to this slot (and forcing its creation initialized to nullptr
if necessary).
I've taken the following steps to add unwinding support from inline assembly:
1) Add a new `unwind` "attribute" (like `sideeffect`) to the asm syntax:
```
invoke void asm sideeffect unwind "call thrower", "~{dirflag},~{fpsr},~{flags}"()
to label %exit unwind label %uexit
```
2.) Add Bitcode writing/reading support + LLVM-IR parsing.
3.) Emit EHLabels around inline assembly lowering (SelectionDAGBuilder + GlobalISel) when `InlineAsm::canThrow` is enabled.
4.) Tweak InstCombineCalls/InlineFunction pass to not mark inline assembly "calls" as nounwind.
5.) Add clang support by introducing a new clobber: "unwind", which lower to the `canThrow` being enabled.
6.) Don't allow unwinding callbr.
Reviewed By: Amanieu
Differential Revision: https://reviews.llvm.org/D95745
Currently the ValueHandler handles both selecting the type and
location for arguments, as well as inserting instructions needed to
handle them. Split this so that the determination of the argument
handling is independent of the function state. Currently the checks
for tail call compatibility do not follow the full assignment logic,
so it misses cases where arguments require nontrivial legalization.
This should help avoid targets ending up in a buggy state where the
argument evaluation may change in different contexts.
For contiguous ranges we drop the last bit-test case but in doing so we skip
adding the new MBB PHI edges to the list of replacement PHI edges, and as a
result we incorrectly omit them in the G_PHI in finishPendingPhis().
Was found when bootstrapping clang with -O3 and GlobalISel enabled on Apple Silicon.
A ConstantAggregateZero may be created from a scalable vector type.
However, it still assumed fixed number of elements when queried for
them. This patch changes ConstantAggregateZero to correctly report its
element count.
This change fixes a couple of issues. Firstly, it fixes a crash in
Constant::getUniqueValue when called on a scalable-vector
zeroinitializer constant.
Secondly, it fixes a latent bug in GlobalISel's IRTranslator in which
translating a scalable-vector zeroinitializer would hit the assertion in
ConstantAggregateZero::getNumElements when casting to a FixedVectorType,
rather than reporting an error more gracefully. This is currently
hypothetical as the IRTranslator has deeper issues preventing the use of
scalable vector types.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D102082
The function template `CallLowering::setArgFlags` is invoked both
for arguments and return values. In the latter case, it calls
`getParamStackAlign` with argument index `~0u`. Nothing wrong
happens now, as the argument is safely incremented back to 0
inside `getParamStackAlign` (the type is `unsigned`), but in
principle it's fragile and may become incorrect.
Differential Revision: https://reviews.llvm.org/D102004
Unfortunately the current call lowering code is built on top of the
legacy MVT/DAG based code. However, GlobalISel was not using it the
same way. In short, the DAG passes legalized types to the assignment
function, and GlobalISel was passing the original raw type if it was
simple.
I do believe the DAG lowering is conceptually broken since it requires
picking a type up front before knowing how/where the value will be
passed. This ends up being a problem for AArch64, which wants to pass
i1/i8/i16 values as a different size if passed on the stack or in
registers.
The argument type decision is split across 3 different places which is
hard to follow. SelectionDAG builder uses
getRegisterTypeForCallingConv to pick a legal type, tablegen gives the
illusion of controlling the type, and the target may have additional
hacks in the C++ part of the call lowering. AArch64 hacks around this
by not using the standard AnalyzeFormalArguments and special casing
i1/i8/i16 by looking at the underlying type of the original IR
argument.
I believe people have generally assumed the calling convention code is
processing the original types, and I've discovered a number of dead
paths in several targets.
x86 actually relies on the opposite behavior from AArch64, and relies
on x86_32 and x86_64 sharing calling convention code where the 64-bit
cases implicitly do not work on x86_32 due to using the pre-legalized
types.
AMDGPU targets without legal i16/f16 have always used a broken ABI
that promotes to i32/f32. GlobalISel accidentally fixed this to be the
ABI we should have, but this fixes it so we're using the worse ABI
that is compatible with the DAG. Ideally we would fix the DAG to match
the old GlobalISel behavior, but I don't wish to fight that battle.
A new native GlobalISel call lowering framework should let the target
process the incoming types directly.
CCValAssigns select a "ValVT" and "LocVT" but the meanings of these
aren't entirely clear. Different targets don't use them consistently,
even within their own call lowering code. My current belief is the
intent was "ValVT" is supposed to be the legalized value type to use
in the end, and and LocVT was supposed to be the ABI passed type
(which is also legalized).
With the default CCState::Analyze functions always passing the same
type for these arguments, these only differ when the TableGen part of
the lowering decide to promote the type from one legal type to
another. AArch64's i1/i8/i16 hack ends up inverting the meanings of
these values, so I had to add an additional hack to let the target
interpret how large the argument memory is.
Since targets don't consistently interpret ValVT and LocVT, this
doesn't produce quite equivalent code to the initial DAG
lowerings. I've opted to consistently interpret LocVT as the in-memory
size for stack passed values, and ValVT as the register type to assign
from that memory. We therefore produce extending loads directly out of
the IRTranslator, whereas the DAG would emit regular loads of smaller
values. This will also produce loads/stores that are wider than the
argument value if the allocated stack slot is larger (and there will
be undef padding bytes). If we had the optimizations to reduce
load/stores based on truncated values, this wouldn't produce a
different end result.
Since ValVT/LocVT are more consistently interpreted, we now will emit
more G_BITCASTS as requested by the CCAssignFn. For example AArch64
was directly assigning types to some physical vector registers which
according to the tablegen spec should have been casted to a vector
with a different element type.
This also moves the responsibility for inserting
G_ASSERT_SEXT/G_ASSERT_ZEXT from the target ValueHandlers into the
generic code, which is closer to how SelectionDAGBuilder works.
I had to xfail an x86 test since I don't see a quick way to fix it
right now (I filed bug 50035 for this). It's broken independently of
this change, and only triggers since now we end up with more ands
which hit the improperly handled selection pattern.
I also observed that FP arguments that need promotion (e.g. f16 passed
as f32) are broken, and use regular G_TRUNC and G_ANYEXT.
TLDR; the current call lowering infrastructure is bad and nobody has
ever understood how it chooses types.
Fix a bug where buildZExtInReg will create and use a new register instead of using the register from parameter DstOp Res.
Reviewed By: arsenm, foad
Differential Revision: https://reviews.llvm.org/D101871
This is a compile time optimization. DILocation:get() is expensive to call, and
we were calling it to create a line zero debug loc for *every* instruction we
translated. We only really need to do this just before we build constants in the
entry block, so I moved this code there. This reduces the LLVM -O0 codegen time
of sqlite3 IR by around 0.7% instructions executed and by about ~2% in CPU time.
We can probably do better with a more involved change, since the reason we need
to create one for each new constant is because we're using the debug scope and
inlined-at loc. If we just use a single instruction's scope and drop the
inlined-at, we can just cache these and have them be free.
CommandLine.h is indirectly included in ~50% of TUs when building
clang, and VirtualFileSystem.h is large.
(Already remarked by jhenderson on D70769.)
No behavior change.
Differential Revision: https://reviews.llvm.org/D100957
This is currently built on top of the SelectionDAG call lowering, but
does not use it the same way. SelectionDAG passes legalized types to
the assignment functions, and the tablegenerated assignment functions
may change the value types expected for registers. This does not
change the types used, just moves the register creation to help fix
this in the future.
Defer the register creation until after all of the assignment
decisions have been made. This will also help have correct tail call
compatibility checking in a future change. Currently it does not work
as expected for any arguments split across multiple registers.
It turns out we actually import a bunch of selection code for intrinsics. The
imported code checks that the register banks on the G_INTRINSIC instruction
are correct. If so, it goes ahead and selects it.
This adds code to AArch64RegisterBankInfo to allow us to correctly determine
register banks on intrinsics which have known register bank constraints.
For now, this only handles @llvm.aarch64.neon.uaddlv. This is necessary for
porting AArch64TargetLowering::LowerCTPOP.
Also add a utility for getting the intrinsic ID from a G_INTRINSIC instruction.
This seems a little nicer than having to know about how intrinsic instructions
are structured.
Differential Revision: https://reviews.llvm.org/D100398
When we pass a AArch64 Homogeneous Floating-Point
Aggregate (HFA) argument with increased alignment
requirements, for example
struct S {
__attribute__ ((__aligned__(16))) double v[4];
};
Clang uses `[4 x double]` for the parameter, which is passed
on the stack at alignment 8, whereas it should be at
alignment 16, following Rule C.4 in
AAPCS (https://github.com/ARM-software/abi-aa/blob/master/aapcs64/aapcs64.rst#642parameter-passing-rules)
Currently we don't have a way to express in LLVM IR the
alignment requirements of the function arguments. The align
attribute is applicable to pointers only, and only for some
special ways of passing arguments (e..g byval). When
implementing AAPCS32/AAPCS64, clang resorts to dubious hacks
of coercing to types, which naturally have the needed
alignment. We don't have enough types to cover all the
cases, though.
This patch introduces a new use of the stackalign attribute
to control stack slot alignment, when and if an argument is
passed in memory.
The attribute align is left as an optimizer hint - it still
applies to pointer types only and pertains to the content of
the pointer, whereas the alignment of the pointer itself is
determined by the stackalign attribute.
For byval arguments, the stackalign attribute assumes the
role, previously perfomed by align, falling back to align if
stackalign` is absent.
On the clang side, when passing arguments using the "direct"
style (cf. `ABIArgInfo::Kind`), now we can optionally
specify an alignment, which is emitted as the new
`stackalign` attribute.
Patch by Momchil Velikov and Lucas Prates.
Differential Revision: https://reviews.llvm.org/D98794
GCC warning:
```
/llvm-project/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp: In member function ‘bool llvm::CombinerHelper::matchFunnelShiftToRotate(llvm::MachineInstr&)’:
/llvm-project/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp:3882:35: warning: ?: using integer constants in boolean context, the expression will always evaluate to ‘true’ [-Wint-in-bool-context]
3882 | Opc == TargetOpcode::G_FSHL ? TargetOpcode::G_ROTL : TargetOpcode::G_ROTR;
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
This patch adds 3 methods, one for power-of-2 vectors which use tree
reductions using vector ops, before a final reduction op. For non-pow-2
types it generates multiple narrow reductions and combines the values with
scalar ops.
Differential Revision: https://reviews.llvm.org/D97163
Basically a port of isBitfieldExtractOpFromSExtInReg in AArch64ISelDAGToDAG.
This is only done post-legalization for now. Once the legalizer knows how to
decompose these back into shifts, this requirement can probably be removed.
Differential Revision: https://reviews.llvm.org/D99230
Currently needsStackRealignment returns false if canRealignStack returns false.
This means that the behavior of needsStackRealignment does not correspond to
it's name and description; a function might need stack realignment, but if it
is not possible then this function returns false. Furthermore,
needsStackRealignment is not virtual and therefore some backends have made use
of canRealignStack to indicate whether a function needs stack realignment.
This patch attempts to clarify the situation by separating them and introducing
new names:
- shouldRealignStack - true if there is any reason the stack should be
realigned
- canRealignStack - true if we are still able to realign the stack (e.g. we
can still reserve/have reserved a frame pointer)
- hasStackRealignment = shouldRealignStack && canRealignStack (not target
customisable)
Targets can now override shouldRealignStack to indicate that stack realignment
is required.
This change will make it easier in a future change to handle the case where we
need to realign the stack but can't do so (for example when the register
allocator creates an aligned spill after the frame pointer has been
eliminated).
Differential Revision: https://reviews.llvm.org/D98716
Change-Id: Ib9a4d21728bf9d08a545b4365418d3ffe1af4d87
Darwin platforms for both AArch64 and X86 can provide optimized `bzero()`
routines. In this case, it may be preferable to use `bzero` in place of a
memset of 0.
This adds a G_BZERO generic opcode, similar to G_MEMSET et al. This opcode can
be generated by platforms which may want to use bzero.
To emit the G_BZERO, this adds a pre-legalize combine for AArch64. The
conditions for this are largely a port of the bzero case in
`AArch64SelectionDAGInfo::EmitTargetCodeForMemset`.
The only difference in comparison to the SelectionDAG code is that, when
compiling for minsize, this will fire for all memsets of 0. The original code
notes that it's not beneficial to do this for small memsets; however, using
bzero here will save a mov from wzr. For minsize, I think that it's preferable
to prioritise omitting the mov.
This also fixes a bug in the libcall legalization code which would delete
instructions which could not be legalized. It also adds a check to make sure
that we actually get a libcall name.
Code size improvements (Darwin):
- CTMark -Os: -0.0% geomean (-0.1% on pairlocalalign)
- CTMark -Oz: -0.2% geomean (-0.5% on bullet)
Differential Revision: https://reviews.llvm.org/D99358
This may occur when swifterror codegen in the translator generates these,
but we shouldn't try to handle them since they should have regclasses anyway.
rdar://75784009
Differential Revision: https://reviews.llvm.org/D99287
byval requires an implicit copy between the caller and callee such
that the callee may write into the stack area without it modifying the
value in the parent. Previously, this was passing through the raw
pointer value which would break if the callee wrote into it.
Most of the time, this copy can be optimized out (however we don't
have the optimization SelectionDAG does yet).
This will trigger more fallbacks for AMDGPU now, since we don't have
legalization for memcpy yet (although we should stop using byval
anyway).
These are pseudos without any users, so DCE was killing them in the combiner.
Marking them as having side effects doesn't seem quite right since they don't.
Gives a nice 0.3% geomean size win on CTMark -Os.
Differential Revision: https://reviews.llvm.org/D98811
The previous technique relied on early-exiting the legalizer predicate
initialization, leaving an empty rule table. That causes a fallback
for most instructions, but some have legacy rules defined like G_ZEXT
which can try continue, but then crash.
We should fall back earlier, in the translator, to avoid this issue.
Differential Revision: https://reviews.llvm.org/D98730
byval arguments need to be assumed writable. Only implicitly stack
passed arguments which aren't addressable in the IR can be assumed
immutable.
Mips is still broken since for some reason its doing its own thing
with the ValueHandlers (and x86 doesn't actually handle byval
arguments now, although some of the code is there).
This was essentially ignoring byval and treating them as a pointer
argument which needed to be loaded from. This should copy the frame
index value to the virtual register, not insert a load from the frame
index into the pointer value.
For AMDGPU, this was producing a load from the byval pointer argument,
to a pointer used for the byval arguments. I do not understand how
AArch64 managed to work before since it appears to be similarly
broken.
We could also change the ValueHandler API to avoid the extra copy from
the frame index, since currently it returns a new register.
I believe there is still an issue with outgoing byval arguments. These
should have a copy inserted in case the callee decided to overwrite
the memory.
For attribute sets, the return index is at 0, and arguments start at
1. getParamAlignment adds the offset of 1, so we need to convert from
attribute index back to IR index.
It is good to have a combined `divrem` instruction when the
`div` and `rem` are computed from identical input operands.
Some targets can lower them through a single expansion that
computes both division and remainder. It effectively reduces
the number of instructions than individually expanding them.
Reviewed By: arsenm, paquette
Differential Revision: https://reviews.llvm.org/D96013
If every element is extracted from a G_BUILD_VECTOR, pass through the source
registers. This is different to the extract(build_vector) combine because this
one tolerates multiple users as long as they're exhaustive.
Differential Revision: https://reviews.llvm.org/D97890
If we have
```
%vec = G_BUILD_VECTOR %reg, %reg, ..., %reg
```
Then lower it to
```
%vec = G_DUP %reg
```
Also update the selector to handle constant splats on G_DUP.
This will not combine when the splat is all zeros or ones. Tablegen-imported
patterns rely on these being G_BUILD_VECTOR.
Minor code size improvements on CTMark at -Os.
Also adds some utility functions to make it a bit easier to recognize splats,
and an AArch64-specific splat helper.
Differential Revision: https://reviews.llvm.org/D97731
This is recommit of 4c8fb7ddd6.
MIR in one unit test had mismatched types.
For vectors we consider a bit as known if it is the same for all demanded
vector elements (all elements by default). KnownBits BitWidth for vector
type is size of vector element. Add support for G_BUILD_VECTOR.
This allows combines of urem_pow2_to_mask in pre-legalizer combiner.
Differential Revision: https://reviews.llvm.org/D96122
For vectors we consider a bit as known if it is the same for all demanded
vector elements (all elements by default). KnownBits BitWidth for vector
type is size of vector element. Add support for G_BUILD_VECTOR.
This allows combines of urem_pow2_to_mask in pre-legalizer combiner.
Differential Revision: https://reviews.llvm.org/D96122
Refactor insertion of the asserting ops. This enables using them for
AMDGPU.
This code should essentially be the same for every target. Mips, X86
and ARM all have different code there now, but this seems to be an
accident. The assignment functions are called with different types
than they would be in the DAG, so this is all likely an assortment of
hacks to get around that.
This merges more AMDGPU ABI lowering code into the generic call
lowering. Start cleaning up by factoring away more of the pack/unpack
logic into the buildCopy{To|From}Parts functions. These could use more
improvement, and the SelectionDAG versions are significantly more
complex, and we'll eventually have to emulate all of those cases too.
This is mostly NFC, but does result in some minor instruction
reordering. It also removes some of the limitations with mismatched
sizes the old code had. However, similarly to the merge on the input,
this is forcing gfx6/gfx7 to use the gfx8+ ABI (which is what we
actually want, but SelectionDAG is stuck using the weird emergent
ABI).
This also changes the load/store size for stack passed EVTs for
AArch64, which makes it consistent with the DAG behavior.
To do this while supporting the existing functionality in SelectionDAG of using
PGO info, we add the ProfileSummaryInfo and LazyBlockFrequencyInfo analysis
dependencies to the instruction selector pass.
Then, use the predicate to generate constant pool loads for f32 materialization,
if we're targeting optsize/minsize.
Differential Revision: https://reviews.llvm.org/D97732
The current narrowing code for G_PHI can only handle the case
where the size is a multiple of the narrow size. If this is not
the case, fall back to SDAG instead of asserting.
Original patch by shepmaster.
Differential Revision: https://reviews.llvm.org/D92446
Generic code should probably not introduce G_INSERT/G_EXTRACT. The
mirror unpackRegs should also be removed, but AMDGPU still has a use
remaining which needs to be fixed.
I copied the nearly identical function from AArch64 into AMDGPU, so
fix this duplication.
Mips and X86 have their own more exotic versions which should be
removed. However replacing those is better left for a separate patch
since it requires other changes to avoid regressions.
We don't yet have working codegen for the resulting unmerges, and if
we did it would probably be horrible.
Differential Revision: https://reviews.llvm.org/D97035
AMDGPU currently has a lot of pre-processing code to pre-split
argument types into 32-bit pieces before passing it to the generic
code in handleAssignments. This is a bit sloppy and also requires some
overly fancy iterator work when building the calls. It's better if all
argument marshalling code is handled directly in
handleAssignments. This handles more situations like decomposing large
element vectors into sub-element sized pieces.
This should mostly be NFC, but does change the generated code by
shifting where the initial argument packing instructions are placed. I
think this is nicer looking, since it now emits the packing code
directly after the relevant copies, rather than after the copies for
the remaining arguments.
This doubles down on gfx6/gfx7 using the gfx8+ ABI for 16-bit
types. This is ultimately the better option, but incompatible with the
DAG. Fixing this requires more work, especially for f16.
Same implementation as G_SEXT_INREG.
Add a testcase to combine-sext-inreg for a concrete example, and a testcase
to KnownBitsTest.
Differential Revision: https://reviews.llvm.org/D96897
This adds a G_ASSERT_SEXT opcode, similar to G_ASSERT_ZEXT. This instruction
signifies that an operation was already sign extended from a smaller type.
This is useful for functions with sign-extended parameters.
E.g.
```
define void @foo(i16 signext %x) {
...
}
```
This adds verifier, regbankselect, and instruction selection support for
G_ASSERT_SEXT equivalent to G_ASSERT_ZEXT.
Differential Revision: https://reviews.llvm.org/D96890
The API is a bit awkward since you need to index into an array in the
passed struct. I guess an alternative would be to pass all of the
individual fields.
The implementation for vectors is broken and doesn't seem to be used by
anything. Explicitly remove support for them, they can be added again
later when they're properly implemented.
Reviewed By: aemerson
Differential Revision: https://reviews.llvm.org/D95699
This combine tries to do inter-block hoisting of extends of G_PHIs, into the
originating blocks of the phi's incoming value. The idea is to expose further
optimization opportunities that are normally obscured by the PHI.
Some basic heuristics, and a target hook for AArch64 is added, to allow tuning.
E.g. if the extend is used by a G_PTR_ADD, it doesn't perform this combine
since it may be folded into the addressing mode during selection.
There are very minor code size improvements on AArch64 -Os, but the real benefit
is that it unlocks optimizations like AArch64 conditional compares on some
benchmarks.
Differential Revision: https://reviews.llvm.org/D95703
Implements same logis as in SelectionDAG.
G_FMINNUM_IEEE and G_FMAXNUM_IEEE are never SNaN by definition and
never NaN when one operand is known non-NaN and other known non-SNaN.
G_FMINNUM and G_FMAXNUM are never NaN/SNaN when one of the operands
is known non-NaN/SNaN.
Differential Revision: https://reviews.llvm.org/D91716
The builder was using the extend user as the insertion point, which meant that
we were incorrectly "moving" the load from its original position, and therefore
could violate memory operation ordering.
This was taking the calling convention from the parent function,
instead of the callee. Avoids regressions in a future patch when the
caller and callee have different type breakdowns.
For some reason AArch64's lowerFormalArguments seems to intentionally
ignore the parent isVarArg.
On AArch64 (which seems to be the only target that supports it), this
attribute allows codegen to avoid saving/restoring the value in x0
across a call.
Gives a 0.1% geomean -Os code size improvement on CTMark.
Differential Revision: https://reviews.llvm.org/D96099
These two cases have identical implementations other than an
unreachable part of `G_ADD` that checks if the scalar we're narrowing
is a vector. Combining them to avoid unnecessary divergence.
If the G_BR + G_BRCOND in this combine use the same MBB, then it will infinite
loop. Don't allow that to happen.
Differential Revision: https://reviews.llvm.org/D95895
When replacing the dst reg with the src reg, we need to make sure that we
propagate the dst reg's register class through to the src.
Otherwise, we aren't meeting the requirements for G_ASSERT_ZEXT, and so the
verifier will fail.
Differential Revision: https://reviews.llvm.org/D95708
I think every target will want to remove these in the same way. Rather than
making them all implement the same code, let's just put this in
InstructionSelect.
Differential Revision: https://reviews.llvm.org/D95652
Remove the call to setFlags in favour of creating the instruction with
the correct flags in the first place, so we don't have to explicitly
notify the observer.
Differential Revision: https://reviews.llvm.org/D95681
This adds generic regbankselect support for G_ASSERT_ZEXT.
It inherits whatever register bank the source was given, always, on all targets.
I think that at the point where we run into these, the source register bank
should be decided.
This also adds some AArch64-specific code which makes sure we can handle
G_ASSERT_ZEXT when deciding on register banks for G_STORE, G_PHI, ... etc.
Differential Revision: https://reviews.llvm.org/D95649
It's the same as the ZEXT/TRUNC case, except SrcBitWidth is given by the
immediate operand.
Update KnownBitsTest.cpp and a MIR test for a concrete example.
Differential Revision: https://reviews.llvm.org/D95566
Treat hint instructions like G_ASSERT_ZEXT like COPY instructions in helpers
which walk through copies.
This ensures that instructions like G_ASSERT_ZEXT won't impact any optimizations
that rely on these helpers.
Differential Revision: https://reviews.llvm.org/D95577
These are widened to a wider UADDE/USUBE, with the overflow value
unused, and with the same synthesis of a new overflow value as for the
O operations.
Reviewed By: paquette
Differential Revision: https://reviews.llvm.org/D95326
This adds a generic opcode which communicates that a type has already been
zero-extended from a narrower type.
This is intended to be similar to AssertZext in SelectionDAG.
For example,
```
%x_was_extended:_(s64) = G_ASSERT_ZEXT %x, 16
```
Signifies that the top 48 bits of %x are known to be 0.
This is useful in cases like this:
```
define i1 @zeroext_param(i8 zeroext %x) {
%cmp = icmp ult i8 %x, -20
ret i1 %cmp
}
```
In AArch64, `%x` must use a 32-bit register, which is then truncated to a 8-bit
value.
If we know that `%x` is already zero-ed out in the relevant high bits, we can
avoid the truncate.
Currently, in GISel, this looks like this:
```
_zeroext_param:
and w8, w0, #0xff ; We don't actually need this!
cmp w8, #236
cset w0, lo
ret
```
While SDAG does not produce the truncation, since it knows that it's
unnecessary:
```
_zeroext_param:
cmp w0, #236
cset w0, lo
ret
```
This patch
- Adds G_ASSERT_ZEXT
- Adds MIRBuilder support for it
- Adds MachineVerifier support for it
- Documents it
It also puts G_ASSERT_ZEXT into its own class of "hint instruction." (There
should be a G_ASSERT_SEXT in the future, maybe a G_ASSERT_ALIGN as well.)
This allows us to skip over hints in the legalizer etc. These can then later
be selected like COPY instructions or removed.
Differential Revision: https://reviews.llvm.org/D95564
Just use the existing `Known.sextInReg` implementation.
- Update KnownBitsTest.cpp.
- Update combine-redundant-and.mir for a more concrete example.
Differential Revision: https://reviews.llvm.org/D95484
The current algorithm just tries to localize defs as far as they can go, and in
the case of G_PHI operands, it clones the def into the predecessor block for
each incoming edge. When multiple edges have the same register value, this can
cause unnecessary code bloat, and inhibit later optimizations.
This change checks if a given phi operand is unique in the phi, if not the
def of that register is not localized to the predecessor.
Differential Revision: https://reviews.llvm.org/D95406
Implement widening for G_SADDO and G_SSUBO.
Add legalize-add/sub tests for narrow overflowing add/sub on AArch64.
Differential Revision: https://reviews.llvm.org/D95034
The widenScalar implementation for signed and unsigned overflowing
operations were very similar: both are checked by truncating the result
and then re-sign/zero-extending it and checking that it matches the
computed operation.
Using a truncate + zero-extend for the unsigned case instead of manually
producing the AND instruction like before leads to an extra copy
instruction during legalization, but this should be harmless.
Differential Revision: https://reviews.llvm.org/D95035
Implement widening for G_SADDO and G_SSUBO. Previously it was only
implemented for G_UADDO and G_USUBO. Also add legalize-add/sub tests for
narrow overflowing add/sub on AArch64.
Differential Revision: https://reviews.llvm.org/D95034
Make this look more like the DAG handling and move to common code.
I also noticed AArch64 seems to not be properly adding the
physreg:virtreg mapping to the function live ins.
If constants are hidden behind G_ANYEXT we can treat them same way as G_SEXT.
For that purpose we extend getConstantVRegValWithLookThrough with option
to handle G_ANYEXT same way as G_SEXT.
Differential Revision: https://reviews.llvm.org/D92219
When constraining an operand register using constrainOperandRegClass(),
the function may emit a COPY in case the provided register class does
not match the current operand register class. However, the operand
itself is not updated to make use of the COPY, thereby resulting in
incorrect code. This patch fixes that bug by updating the machine
operand accordingly.
Reviewed By: dsanders
Differential Revision: https://reviews.llvm.org/D91244
This is a restricted version of the combine in `DAGCombiner::MatchLoadCombine`.
(See D27861)
This tries to recognize patterns like below (assuming a little-endian target):
```
s8* x = ...
s32 val = a[0] | (a[1] << 8) | (a[2] << 16) | (a[3] << 24)
->
s32 val = *((i32)a)
s8* x = ...
s32 val = a[3] | (a[2] << 8) | (a[1] << 16) | (a[0] << 24)
->
s32 val = BSWAP(*((s32)a))
```
(This patch also handles the big-endian target case as well, in which the first
example above has a BSWAP, and the second example above does not.)
To recognize the pattern, this searches from the last G_OR in the expression
tree.
E.g.
```
Reg Reg
\ /
OR_1 Reg
\ /
OR_2
\ Reg
.. /
Root
```
Each non-OR register in the tree is put in a list. Each register in the list is
then checked to see if it's an appropriate load + shift logic.
If every register is a load + potentially a shift, the combine checks if those
loads + shifts, when OR'd together, are equivalent to a wide load (possibly with
a BSWAP.)
To simplify things, this patch
(1) Only handles G_ZEXTLOADs (which appear to be the common case)
(2) Only works in a single MachineBasicBlock
(3) Only handles G_SHL as the bit twiddling to stick the small load into a
specific location
An IR example of this is here: https://godbolt.org/z/4sP9Pj (lifted from
test/CodeGen/AArch64/load-combine.ll)
At -Os on AArch64, this is a 0.5% code size improvement for CTMark/sqlite3,
and a 0.4% improvement for CTMark/7zip-benchmark.
Also fix a bug in `isPredecessor` which caused it to fail whenever `DefMI` was
the first instruction in the block.
Differential Revision: https://reviews.llvm.org/D94350
This fixes double printing of insertion debug messages in the
legalizer.
Try to cleanup usage of observers. Currently the use of observers is
pretty hard to follow and it's not clear what is responsible for
them. Observers are referenced in 3 places:
1. In the MachineFunction
2. In the MachineIRBuilder
3. In the LegalizerHelper
The observers in the MachineFunction and MachineIRBuilder are both
called only on insertions, and are redundant with each other. The
source of the double printing was the same observer was added to both
the MachineFunction, and the MachineIRBuilder. One of these references
needs to be removed. Arguably observers in general should be fully
removed from one or the other, but it may be useful to have a local
observer in the MachineIRBuilder that is not added to the function's
observers. Alternatively, the wrapper observer could manage a local
observer in one place.
The LegalizerHelper only ever calls the observer on changing/changed
instructions, and never insertions. Logically these are two different
types of observers, for changes and for insertions.
Additionally, some places used the GISelObserverWrapper when they only
needed a single observer they could use directly.
Setting the observer in the LegalizerHelper constructor is not
flexible enough if the LegalizerHelper is constructed anywhere outside
the one used by the legalizer. AMDGPU calls the LegalizerHelper in
RegBankSelect, and needs to use a local observer to apply the regbank
to newly created instructions. Currently it accomplishes this by
constructing a local MachineIRBuilder. I'm trying to move the
MachineIRBuilder to be owned/maintained by the RegBankSelect pass
itself, but the locally constructed LegalizerHelper would reset the
observer.
Mips also has a special case use of the LegalizationArtifactCombiner
in applyMappingImpl; I think we do need to run the artifact combiner
during RegBankSelect, but in a more consistent way outside of
applyMappingImpl.
If the return values can't be lowered to registers
SelectionDAG performs the sret demotion. This patch
contains the basic implementation for the same in
the GlobalISel pipeline.
Furthermore, targets should bring relevant changes
during lowerFormalArguments, lowerReturn and
lowerCall to make use of this feature.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D92953
Recently a few patches are made to move towards using select i1 instead of and/or i1 to represent "a && b"/"a || b" in C/C++.
"a && b" in C/C++ does not evaluate b if a is false whereas 'and a, b' in IR evaluates b and uses its result regardless of the result of a.
This is problematic because it can cause miscompilation if b was an erroneous operation (https://llvm.org/pr48353).
In C/C++, the result is simply false because b is not evaluated, but in IR the result is poison.
The discussion at D93065 has more context about this.
This patch makes two branch-splitting optimizations (one in SelectionDAGBuilder, one in CodeGenPrepare) recognize
select form of and/or as well using m_LogicalAnd/Or.
Since it is CodeGen, I think this is semantically ok (at least as safe as what codegen already did).
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D93853
Returning int64_t was arbitrarily limiting for wide integer types, and
the functions should handle the full generality of the IR.
Also changes the full form which returns the originally defined
vreg. Add another wrapper for the common case of just immediately
converting to int64_t (arguably this would be useful for the full
return value case as well).
One possible issue with this change is some of the existing uses did
break without conversion to getConstantVRegSExtVal, and it's possible
some without adequate test coverage are now broken.
This migrates all LLVM (except Kaleidoscope and
CodeGen/StackProtector.cpp) DebugLoc::get to DILocation::get.
The CodeGen/StackProtector.cpp usage may have a nullptr Scope
and can trigger an assertion failure, so I don't migrate it.
Reviewed By: #debug-info, dblaikie
Differential Revision: https://reviews.llvm.org/D93087
The lowering of vector selects needs to first splat the scalar mask into a vector
first.
This was causing a crash when building oggenc in the test suite.
Differential Revision: https://reviews.llvm.org/D91655
It's fairly common to need matchers for a specific constant value, or for
common idioms like finding a negated register.
Add
- `m_SpecificICst`, which returns true when matching a specific value..
- `m_ZeroInt`, which returns true when an integer 0 is matched.
- `m_Neg`, which returns when a register is negated.
Also update a few places which use idioms related to the new matchers.
Differential Revision: https://reviews.llvm.org/D91397
No longer rely on an external tool to build the llvm component layout.
Instead, leverage the existing `add_llvm_componentlibrary` cmake function and
introduce `add_llvm_component_group` to accurately describe component behavior.
These function store extra properties in the created targets. These properties
are processed once all components are defined to resolve library dependencies
and produce the header expected by llvm-config.
Differential Revision: https://reviews.llvm.org/D90848
We have a frequent pattern where we're merging two KnownBits to get the common/shared bits, and I just fell for the gotcha where I tried to use the & operator to merge them........
We can use KnownBitsAnalysis to cover cases when mask is not trivial. It can
also help with cases when mask is not constant but can still be folded into
one. Since 'and' is comutative we should treat both operands as possible
replacements.
Differential Revision: https://reviews.llvm.org/D90674
This sequence of instructions can be simplified if they are single use and
some operands are constants. Additional combines may be applied afterwards.
Differential Revision: https://reviews.llvm.org/D90223
Sequence of same shift instructions with constant operands can be combined into
a single shift instruction.
Differential Revision: https://reviews.llvm.org/D90217
Convert GISelKnownBits.computeKnownBitsImpl shift handling to use the common KnownBits implementations, which makes use of the known leading/trailing bits for shifted values in cases where we don't know the shift amount value, as detailed in https://blog.regehr.org/archives/1709
Differential Revision: https://reviews.llvm.org/D90527
https://reviews.llvm.org/D88060
This adds the following combines
1) build_vector formation from insert_vec_elts
2) insert_vec_elts (build_vector) -> build_vector
In certain places in llvm/lib/CodeGen we were relying upon the TypeSize
comparison operators when in fact the code was only ever expecting
either scalar values or fixed width vectors. I've changed some of these
places to use the equivalent scalar operator.
Differential Revision: https://reviews.llvm.org/D88482
In order to prevent the ExpandReductions pass from expanding some intrinsics
before they get to codegen, I had to add a -disable-expand-reductions flag
for testing purposes.
Differential Revision: https://reviews.llvm.org/D89028
https://reviews.llvm.org/D88865
This adds a single combine for GlobalISel to fold:
ptradd (inttoptr C1) C2
Into:
C1 + C2
Additionally, a small test for AArch64 is added.
Patch by pnappa.
When the first operand is a null pointer we can avoid making a G_PTR_ADD and
make a G_INTTOPTR with the offset operand.
This helps us avoid making add with 0 later on for targets such as AMDGPU.
Differential Revision: https://reviews.llvm.org/D87140
If the known shift amount is bigger than or equal to the bitwidth of the type of the value to be shifted,
the result is target dependent, so don't try to infer any bits.
This fixes a crash we've seen in one of our internal test suites.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D89232
The IRTranslator depends on the branch probability info pass when the
optimization level is different than None and it depends all the time on
the StackProtector pass.
We have to explicitly call out pass dependencies otherwise the pass manager
may not be able to schedule the IRTranslator.
Before this patch, we were lucky because previous passes depend on the branch
probability info pass (like the Global Variable Optimization) and the stack
protector pass is initialized in initializeCodeGen.
However, if the target has a custom pipeline without any passes like Global
Variable Optimization, the pipeline creation will fail, at least because of
the branch probability info pass dependency (it is unlikely that
initializeCodeGen is not called).
This patch adds the missing dependencies to the IRTranslator.
Differential Revision: https://reviews.llvm.org/D89063
If a CSEMIRBuilder query hits the instruction at the current insert point,
move insert point ahead one so that subsequent uses of the builder don't end up with
uses before defs.
This fix also shows that AMDGPU was also affected by this bug often, but got away
with it because it was using a G_IMPLICIT_DEF before the use.
Differential Revision: https://reviews.llvm.org/D88605
Before, for each original argument i, ValNo was set to i + PartIdx, but
ValNo is intended to reflect the index of the value before splitting.
Hence, ValNo should always be set to i and not consider the PartIdx.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D86511
During lowering of G_UMULO and friends, the previous code moved the builder's
insertion point to be after the legalizing instruction. When that happened, if
there happened to be a "G_CONSTANT i32 0" immediately after, the CSEMIRBuilder
would try to find that constant during the buildConstant(zero) call, and since
it dominates itself would return the iterator unchanged, even though the def
of the constant was *after* the current insertion point. This resulted in the
compare being generated *before* the constant which it was using.
There's no need to modify the insertion point before building the mul-hi or
constant. Delaying moving the insert point ensures those are built/CSEd before
the G_ICMP is built.
Fixes PR47679
Differential Revision: https://reviews.llvm.org/D88514
Fix creation of illegal unmerge when widen was requested to a type which
is not a multiple of the destination type. E.g. when trying to widen
an s48 unmerge to s64 the existing code would create an illegal unmerge
from s64 to s48.
Instead, create further unmerges to a GCD type, then use this to remerge
these intermediate results to the actual destinations.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D88422
When we see this:
```
%and = G_AND %x, %y
%xor = G_XOR %and, %y
```
Produce this:
```
%not = G_XOR %x, -1
%new_and = G_AND %not, %y
```
as long as we are guaranteed to eliminate the original G_AND.
Also matches all commuted forms. E.g.
```
%and = G_AND %y, %x
%xor = G_XOR %y, %and
```
will be matched as well.
Differential Revision: https://reviews.llvm.org/D88104
Previously, if a floating-point type was legal, but FNEG wasn't legal,
we would use FSUB. Instead, we should use integer ops, to preserve the
semantics. (Alternatively, there's a compiler-rt call we could use, but
there isn't much reason to use that.)
It turns out we actually are still using this obscure codepath in a few
cases: on some targets, we have "legal" floating-point types that don't
actually support any floating-point operations. In particular, ARM and
AArch64 are using this path.
The implementation for SelectionDAG is pretty simple because we can
reuse the infrastructure from FCOPYSIGN.
See also 9a3dc3e, the corresponding change to type legalization.
Also includes a "bonus" change to STRICT_FSUB legalization, so we can
lower a STRICT_FSUB to a float libcall.
Includes the changes to both LegalizeDAG and GlobalISel so we don't have
inconsistent results in the future.
Fixes https://bugs.llvm.org/show_bug.cgi?id=46792 .
Differential Revision: https://reviews.llvm.org/D84287
The shift amount type does not necessarily match the result type. This
was inserting a trunc from s32 to s32, which asserted. Just preserve
the original shift amount type which can be legalized later.
For <8 x s32> = fptrunc <8 x s64> the fewerElementsVector action tries to break
down the source vector into the final source vectors of <2 x s64> using unmerge.
This fixes a crash due to using the wrong number of elements for the breakdown
type.
Also add some legalizer tests for explicitly G_FPTRUNC which we didn't have.
Differential Revision: https://reviews.llvm.org/D87814
https://reviews.llvm.org/D86393
Patch adds five new `GICombinerRules`, one for each of the following unary
FP instrs: `G_FNEG`, `G_FABS`, `G_FPTRUNC`, `G_FSQRT`, and `G_FLOG2`. The
combine rules perform the FP operation on the constant operand and replace
the original instr with the result. Patch additionally adds new combiner
tests for the AArch64 target to test these new combiner rules.
test/CodeGen/AArch64/GlobalISel/combine-trunc.mir was failing
due to the different order for evaluating function arguments.
This patch updates the related code to fix the issue.
https://reviews.llvm.org/D87668
Patch adds two new GICombinerRules, one for G_MUL(X, 1) and another for G_MUL(X, -1).
G_MUL(X, 1) is an identity combine, and G_MUL(X, -1) gets replaced with G_SUB(0, X).
Patch additionally adds new combiner tests for the AArch64 target to test these
new combiner rules, as well as updates AMDGPU GISel tests.
Patch by mkitzan
Update TargetMachine.Options with function attributes before we start
to generate MIR instructions. This allows access to correct function
attributes via TargetMachine.Options (it used to access attributes of
the function that was translated first).
This affects some existing tests with "no-nans-fp-math" attribute.
Follow-up on D87456.
Differential Revision: https://reviews.llvm.org/D87511
Add a combiner helper that replaces G_UNMERGE where all the destination lanes
are dead except the first one with a G_TRUNC.
Differential Revision: https://reviews.llvm.org/D87174
Add a combiner helper that replaces G_UNMERGE of big constants into direct
use of smaller constants.
Differential Revision: https://reviews.llvm.org/D87166
https://reviews.llvm.org/D87554
Patch adds one new GICombinerRule for G_FABS. The combine rule folds G_FABS(G_FABS(X)) to G_FABS(X).
Patch additionally adds new combiner tests for the AArch64 target to test this new combiner rule.
Patch by mkitzan.
Add the matching and applying function to the combiner helper for
G_UNMERGE_VALUES(G_MERGE_VALUES).
This combine also supports any merge-like input nodes, like G_BUILD_VECTORS
and is robust against bitcasts in between int unmerge and merge nodes.
When the input type of the merge node and the output type of the unmerge
node are not the same, but the sizes are, the combine still applies but
creates bitcasts between the sources and the destinations instead of
reusing the destinations directly.
Long term, the artifact combiner should probably reuse that helper, but
as of today, it doesn't use any outside helper, so I kept it this way.
Differential Revision: https://reviews.llvm.org/D87117
Check for NoNaNsFPMath function attribute in isKnownNeverSNaN.
Function attributes are in held in 'TargetMachine.Options'.
Among other things, this allows selection of some patterns imported
in D87351 since G_FCANONICALIZE is not generated when isKnownNeverSNaN
returns true in lowerFMinNumMaxNum.
However we notice some incorrect results since function attributes are
not correctly written in TargetMachine.Options when next function is
processed. Take a look at @v_test_no_global_nnans_med3_f32_pat0_srcmod0,
it has "no-nans-fp-math"="false" but TargetMachine.Options still has it
set to true since first function in test file had this attribute set to
true. This will be fixed in D87511.
Differential Revision: https://reviews.llvm.org/D87456
Halide users reported this here: https://llvm.org/pr46176
I reported the issue to MSVC here:
https://developercommunity.visualstudio.com/content/problem/1179643/msvc-copies-overaligned-non-trivially-copyable-par.html
This codepath is apparently not covered by LLVM's unit tests, so I added
coverage in a unit test.
If we want to support this configuration going forward, it means that is
in general not safe to pass a SmallVector<T, N> by value if alignof(T)
is greater than 4. This doesn't appear to come up often because passing
a SmallVector by value is inefficient and not idiomatic: it copies the
inline storage. In this case, the SmallVector<LLT,4> is captured by
value by a lambda, and the lambda is passed by value into std::function,
and that's how we hit the bug.
Differential Revision: https://reviews.llvm.org/D87475
We weren't using this before, so none of the MachineFunction CFG edges had the
branch probability information added. As a result, block placement later in the
pipeline was flying blind.
This is enabled only with optimizations enabled like SelectionDAG.
Differential Revision: https://reviews.llvm.org/D86824
This is a port of the functionality from SelectionDAG, which tries to find
a tree of conditions from compares that are then combined using OR or AND,
before using that result as the input to a branch. Instead of naively
lowering the code as is, this change converts that into a sequence of
conditional branches on the sub-expressions of the tree.
Like SelectionDAG, we re-use the case block codegen functionality from
the switch lowering utils, which causes us to generate some different code.
The result of which I've tried to mitigate in earlier combine patches.
Differential Revision: https://reviews.llvm.org/D86665
This combine previously tried to take sequences like:
%cond = G_ICMP pred, a, b
G_BRCOND %cond, %truebb
G_BR %falsebb
%truebb:
...
%falsebb:
...
and by inverting the compare predicate and swapping branch targets, delete the
G_BR and instead have a single conditional branch to the falsebb. Since in an
earlier patch we have a combine to fold not(icmp) into just an inverted icmp,
we don't need this combine to do as much. This patch instead generalizes the
combine by just looking for:
G_BRCOND %cond, %truebb
G_BR %falsebb
%truebb:
...
%falsebb:
...
and then inverting the condition using a not (xor). The xor can be folded away
in a separate combine. This change also lets us avoid some optimization code
in the IRTranslator.
I also think that deleting G_BRs in the combiner is unnecessary. That's
something that targets can decide to do at selection time and could simplify
generic code in future.
Differential Revision: https://reviews.llvm.org/D86664
The post-index matcher, before it queries the target legality, walks uses
of some instructions which in pathological cases can be massive. Since
no targets actually support indexed loads yet, disable this to stop wasting
compile time on something which is going to fail anyway.
Unwinders may only preserve the lower 64bits of Neon and SVE registers,
as only the registers in the base ABI are guaranteed to be preserved
over the exception edge. The caller will need to preserve additional
registers for when the call throws an exception and the unwinder has
tried to recover state.
For e.g.
svint32_t bar(svint32_t);
svint32_t foo(svint32_t x, bool *err) {
try { bar(x); } catch (...) { *err = true; }
return x;
}
`z0` needs to be spilled before the call to `bar(x)` and reloaded before
returning from foo, as the exception handler may have clobbered z0.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D84737
This is needed for an upcoming change to how we translate conditional branches
which might generate these.
Differential Revision: https://reviews.llvm.org/D86383
With gcc 6.3.0, I hit the following compilation error:
../lib/CodeGen/GlobalISel/Combiner.cpp: In member function
‘bool llvm::Combiner::combineMachineInstrs(llvm::MachineFunction&,
llvm::GISelCSEInfo*)’:
../lib/CodeGen/GlobalISel/Combiner.cpp:156:54: error: suggest parentheses
around ‘&&’ within ‘||’ [-Werror=parentheses]
assert(!CSEInfo || !errorToBool(CSEInfo->verify()) &&
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
"CSEInfo is not consistent. Likely missing calls to "
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"observer on mutations");
Fix the code as suggested by the compiler.
https://reviews.llvm.org/D83833
Patch adds two new GICombinerRules for G_SELECT. The rules include:
combining selects with undef comparisons into their first selectee value,
and to combine away selects with constant comparisons. Patch additionally
adds a new combiner test for the AArch64 target to test these new G_SELECT
combiner rules and the existing select_same_val combiner rule.
Patch by mkitzan
https://reviews.llvm.org/D86676
Sometimes we can have the following code
x:gpr(s32) = G_OP
Say we build G_OP2 to the same x and then delete the previous instruction. Using something like
Register X = ...;
auto NewMIB = CSEBuilder.buildOp2(X, ... args);
Currently there's a mismatch in how NewMIB is profiled and inserted into the CSEMap (ie it doesn't consider register bank/register class along with type).Unify the profiling by refactoring and calling the common method.
This was found by turning on the CSEInfo::verify in at the end of each of our GISel passes which turns inconsistent state/non determinism in CSEing into crashes which likely usually indicates missing calls to Observer on mutations (the most common case). Here non determinism usually means not cseing sometimes, but almost never about producing incorrect code.
Also this patch adds this verification at the end of the combiners as well.
AArch64, X86 and Mips currently directly consumes these and custom
lowering to produce a libcall, but really these should follow the
normal legalization process through the libcall/lower action.
This produces less work for addressing mode matching. I think this is
safe since I don't think machine IR is supposed to give the same
aliasing properties as getelementptr in the IR.
shl ([sza]ext x, y) => zext (shl x, y).
Turns expensive 64 bit shifts into 32 bit if it does not overflow the
source type:
This is a port of an AMDGPU DAG combine added in
5fa289f0d8. InstCombine does this
already, but we need to do it again here to apply it to shifts
introduced for lowered getelementptrs. This will help matching
addressing modes that use 32-bit offsets in a future patch.
TableGen annoyingly assumes only a single match data operand, so
introduce a reusable struct. However, this still requires defining a
separate GIMatchData for every combine which is still annoying.
Adds a morally equivalent function to the existing
getShiftAmountTy. Without this, we would have to do try to repeatedly
query the legalizer info and guess at what type to use for the shift.
Known bits for G_ANYEXT was incorrectly using KnownBits::zext, causing
us to treat the high bits as zero even though they're (by definition)
unknown.
Differential Revision: https://reviews.llvm.org/D86323
The check for the landingpad instructions was overly restrictive. In optimimized builds PHI nodes can appear
before the landingpad instructions, resulting in a fallback to SelectionDAG.
This change relaxes the check to allow PHI nodes.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D86141
If we have a mask, and a value x, where (x & mask) == x, we can drop the AND
and just use x.
This is about a 0.4% geomean code size improvement on CTMark at -O3 for AArch64.
In AArch64, this is most useful post-legalization. Patterns like this often
show up when legalizing s1s, which must be extended to larger types.
e.g.
```
%cmp:_(s32) = G_ICMP ...
%and:_(s32) = G_AND %cmp, 1
```
Since G_ICMP only produces a single bit, there's no reason to mask it with the
G_AND.
Differential Revision: https://reviews.llvm.org/D85463
It's annoying to have to maintain multiple, nearly identical chains of if
statements which all set the same attributes.
Add a helper function, `addFlagsUsingAttrFn` which performs the attribute
setting.
Then, use wrappers for that function in `lowerCall` and `setArgFlags`.
(Note that the flag-setting code in `setArgFlags` was missing the returned
attribute. There's no selection for this yet, so no test. It's an example of
the kind of thing this lets us avoid, though.)
Differential Revision: https://reviews.llvm.org/D86159
Similar to this commit:
faf8065a99
Testcase is pretty much the same as
test/CodeGen/AArch64/tailcall-explicit-sret.ll
Except it uses i64 (since we don't handle the i1024 return values yet), and
doesn't have indirect tail call testcases (because we can't translate those
yet).
Differential Revision: https://reviews.llvm.org/D86148
This is restricted to single use loads, which if we fold to sextloads we can
find more optimal addressing modes on AArch64.
This also fixes an overload the MachineFunction::getMachineMemOperand() method
which was incorrectly using the MF alignment instead of the MMO alignment.
Differential Revision: https://reviews.llvm.org/D85966
By detecting this sign extend pattern early, we can uncover opportunities for
more optimizations.
Differential Revision: https://reviews.llvm.org/D85965
We weren't looking through the parameters on calls at all.
E.g., say you had
```
declare i32 @zext(i32 zeroext %x)
...
%y = call i32 @zext(i32 %something)
...
```
At the point of the call, we wouldn't know that the %something should have the
zeroext attribute.
This sets flags in about the same way as
TargetLoweringBase::ArgListEntry::setAttributes.
Differential Revision: https://reviews.llvm.org/D86125
Some of the lower implementations were relying on this, however the
type was not set depending on which form .lower* helper form you were
using. For instance, if you used an unconditonal lower(), the type was
never set. Most of the lower actions do not benefit from a type
parameter, and just expand in terms of the original operation's types.
However, some lowerings could benefit from an additional type hint to
combine a promotion and an expansion. An example of this is for
add/sub sat. The DAG integer legalization tries to use smarter
expansions directly when promoting the integer type, and doesn't
always produce the same instruction with a wider type.
Treat this as an optional hint argument, that only means something for
specific lower actions. It may be useful to generalize this mechanism
to pass a full list of type indexes and desired types, but I haven't
run into a case like that yet.
These should really match either G_BUILD_VECTOR or
G_BUILD_VECTOR_TRUNC, but there doesn't seem to be an existing
mechanism for matching alternative opcodes. There is GIM_SwitchOpcode,
but it seems to assume it's oly only used for matcher optimization.
I could also omit any opcode check and rely on the matcher directly
checking the opcode, but the table optimizer currently assumes there
has to be an opcode check.
Also doesn't try to handle undef elements like the DAG version.
This is mostly a straight port from SelectionDAG. We re-use the actual bit-test
analysis part from SwitchLoweringUtils, which was factored out earlier to
support jump-tables.
Differential Revision: https://reviews.llvm.org/D85233
This implements
```
(logic_op (op x...), (op y...)) -> (op (logic_op x, y))
```
when `op` is an extend, a shift, or an and.
This is similar to `DAGCombiner::hoistLogicOpWithSameOpcodeHands`
(with a bunch of missing cases, e.g. G_TRUNC, G_BITCAST, etc.)
This is implemented so it works both pre and post-legalization.
This also adds a general way to add a series of instructions in a combine.
(`applyBuildInstructionSteps`).
Differential Revision: https://reviews.llvm.org/D85050
This mirrors the support for the equivalent extracts. This also
creates a huge mess that would be greatly improved if we had any bit
operation combines.
Use the same basic strategy as LegalizeVectorTypes. Try to index into
smaller pieces if there's a constant index, and otherwise fall back to
a stack temporary.
Add given input and mark it as tied.
Doesn't create additional copy compared to
matching input constraint to virtual register.
Differential Revision: https://reviews.llvm.org/D85122
Get the argument register and ensure there's a copy to the virtual
register. AMDGPU and AArch64 have similarish code to get the livein
value, and I also want to use this in multiple places.
This is a bit more aggressive about setting the register class than
the original function, but that's probably OK.
I think we're missing a few verifier checks for function live ins. I
noticed AArch64's calling convention code is not actually adding
liveins to functions, only the entry block (which apparently might not
matter that much?). There should probably be a verifier check that
entry block live ins are also live into the function. We also might
need a verifier check that the copy to the livein virtual register is
in the entry block.
This one is pretty easy and shrinks the list of unhandled
intrinsics. I'm not sure how relevant the insert point is. Using the
insert position of EntryBuilder will place this after
constants. SelectionDAG seems to end up emitting these after argument
copies and before anything else, but I don't think it really
matters. This also ends up emitting these in the opposite order from
SelectionDAG, but I don't think that matters either.
This also needs a fix to stop the later passes dropping this as a dead
instruction. DeadMachineInstructionElim's version of isDead special
cases LOCAL_ESCAPE for some reason, and I'm not sure why it's excluded
from MachineInstr::isLabel (or why isDead doesn't check it).
I also noticed DeadMachineInstructionElim never considers inline asm
as dead, but GlobalISel will drop asm with no constraints.
This patch stops unconditionally transforming FSUB(-0, X) into an FNEG(X) while building the MIR.
This corresponds with the SelectionDAGISel change in D84056.
Differential Revision: https://reviews.llvm.org/D85139
Use pad with undef and unmerge with unused results. This is annoyingly
similar to several other places in LegalizerHelper, but they're all
slightly different.
For AMDGPU, vectors with elements < 32 bits should be indexed in
32-bit elements and the desired bits extracted from there. For
elements > 64-bits, these should be reduce to 64/32 elements to enable
the normal dynamic indexing paths.
In the dynamic index cases, this produces shorter code most of the
time. This does immediately regress the constant index cases, but this
should be fixed once we have the most basic of shift combines.
The element size > 64 case is pretty much ported from the exisiting
DAG implementation for extract element promote. The increasing element
size case is new.
https://reviews.llvm.org/D84909
Patch adds two new GICombinerRules, one for G_INTTOPTR and one for
G_PTRTOINT. The G_INTTOPTR elides ptr2int(int2ptr(x)) to a copy of x, if
the cast is within the same address space. The G_PTRTOINT elides
int2ptr(ptr2int(x)) to a copy of x. Patch additionally adds new combiner
tests for the AArch64 target to test these new combiner rules.
Patch by mkitzan
I still think it's highly questionable that we have two intrinsics
with identical behavior and only vary by the name of the libcall used
if it happens to be lowered that way, but try to reduce the feature
delta between SDAG and GlobalISel for recently added intrinsics. I'm
not sure which opcode should be considered the canonical one, but
lower roundeven back to round.
In future, we'd like to use the perfect-shuffle mechanism to deal with these
shuffle permutations. For now, this improves performance by avoiding the
super-expensive const-pool load + tbl instruction.
Differential Revision: https://reviews.llvm.org/D84866
I think these were added as a workaround for SelectionDAG lacking half
legalization support in the past. I think they should probably be
removed from the IR, but clang does still have a target control to
emit these instead of the native half fpext/fptrunc.
Amara Emerson