This is a bug noted in the recent D72733 and seen
in the similar transform just above the changed source code.
I added tests with illegal types and zexts to show the bug -
we could transform legal phi ops to illegal, etc. I did not add
tests with trunc because we won't see any diffs on those patterns.
That is because InstCombiner::SliceUpIllegalIntegerPHI() appears to
do those transforms independently of datalayout. It can also create
more casts than are present in existing code.
There are some existing regression tests that do not include a
datalayout that would be altered by this fix. I assumed that the
lack of a datalayout in those regression files is an oversight, so
I added the minimal layout (make i32 legal) necessary to preserve
behavior on those tests.
Differential Revision: https://reviews.llvm.org/D73907
Under MVE, we do not have any lowering for fminimum, which a
vector_reduce_fmin without NoNan will be expanded into. As with the
other recent patches, force this to expand in the pre-isel pass. Note
that Neon lowering would be OK because the scalar fminimum uses the
vector VMIN instruction, but is probably better to just rely on the
scalar operations, which is what is done here.
Also fixes what appears to be the reversal of INF vs -INF in the
vector_reduce_fmin widening code.
This ports the existing case for G_XOR from `getTestBitOperand` in
AArch64ISelLowering into GlobalISel.
The idea is to flip between TBZ and TBNZ while walking through G_XORs.
Let's say we have
```
tbz (xor x, c), b
```
Let's say the `b`-th bit in `c` is 1. Then
- If the `b`-th bit in `x` is 1, the `b`-th bit in `(xor x, c)` is 0.
- If the `b`-th bit in `x` is 0, then the `b`-th bit in `(xor x, c)` is 1.
So, then
```
tbz (xor x, c), b == tbnz x, b
```
Let's say the `b`-th bit in `c` is 0. Then
- If the `b`-th bit in `x` is 1, the `b`-th bit in `(xor x, c)` is 1.
- If the `b`-th bit in `x` is 0, then the `b`-th bit in `(xor x, c)` is 0.
So, then
```
tbz (xor x, c), b == tbz x, b
```
Differential Revision: https://reviews.llvm.org/D73929
This implements the following optimization:
```
(tbz (shl x, c), b) -> (tbz x, b-c)
```
Which appears in `getTestBitOperand` in AArch64ISelLowering.cpp.
If we test bit `b` of `shl x, c`, we can fold away the `shl` by looking `c` bits
to the right of `b` in `x` when this fits in the type. So, we can just test the
`b-c`th bit.
Differential Revision: https://reviews.llvm.org/D73924
Summary:
The AIX assembler .space directive can't take a second non-zero argument to fill
with. But LLVM emitFill currently assumes it can. We add a flag to the AsmInfo
to check if non-zero fill is supported, and if we can't zerofill non-zero values
we just splat the .byte directives.
Reviewers: stevewan, sfertile, DiggerLin, jasonliu, Xiangling_L
Reviewed By: jasonliu
Subscribers: Xiangling_L, wuzish, nemanjai, hiraditya, kbarton, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73554
Given
```
tb(n)z (and x, m), b
```
Where the `b`-th bit of `m` is 1,
```
tb(n)z (and x, m), b == tb(n)z x, b
```
So, we can walk past a `G_AND` in this case.
Also add test/CodeGen/AArch64/GlobalISel/opt-fold-and-tbz-tbnz.mir to test this.
Differential Revision: https://reviews.llvm.org/D73790
convertPtrAddToAdd improved overall code size and quality by a significant amount,
but on -O0 we generate some cross-class copies due to the fact that we emitted
G_PTRTOINT and G_INTTOPTR around the G_ADD. Unfortunately at -O0 we don't run any
register coalescing, so these cross class copies end up escaping as moves, and
we ended up regressing 3 benchmarks on CTMark (though still a winner overall).
This patch changes the lowering to instead directly emit the G_ADD into the
destination register, and then force changes the dest LLT to s64 from p0. This
should be ok, as all uses of the register should now be selected and therefore
the LLT doesn't matter for the users. It does however matter for the importer
patterns, which will fail to select a G_ADD if there's a p0 LLT.
I'm not able to get rid of the G_PTRTOINT on the source yet however. We can't
use the same trick of breaking the type system since that could break the
selection of the defining instruction. Thus with -O0 we still end up with a
cross class copy on source.
Code size improvements on -O0:
Program baseline new diff
test-suite :: CTMark/Bullet/bullet.test 965520 949164 -1.7%
test-suite...TMark/7zip/7zip-benchmark.test 1069456 1052600 -1.6%
test-suite...ark/tramp3d-v4/tramp3d-v4.test 1213692 1199804 -1.1%
test-suite...:: CTMark/sqlite3/sqlite3.test 421680 419736 -0.5%
test-suite...-typeset/consumer-typeset.test 837076 833380 -0.4%
test-suite :: CTMark/lencod/lencod.test 799712 796976 -0.3%
test-suite...:: CTMark/ClamAV/clamscan.test 688264 686132 -0.3%
test-suite :: CTMark/kimwitu++/kc.test 1002344 999648 -0.3%
test-suite...Mark/mafft/pairlocalalign.test 422296 421768 -0.1%
test-suite :: CTMark/SPASS/SPASS.test 656792 656532 -0.0%
Geomean difference -0.6%
Differential Revision: https://reviews.llvm.org/D73910
Start using a new strategy with a combination of merge and unmerges.
This allows scalarizing before lowering, which in cases like
<2 x s128> avoids producing giant illegal shifts.
Followup to D73135. If the target doesn't have hard float (default
for ARM), then we assert when trying to soften the result of vector
reduction intrinsics. This patch marks these for expansion as well.
(A bit odd to use vectors on a target without hard float ... but
that's where you end up if you expose target-independent vector types.)
Differential Revision: https://reviews.llvm.org/D73854
We have to be careful in SimplifyDemandedBits with loads in case we attempt to combine back to a constant (which then gets turned into a constant pool load again), but we can at least set the upper KnownBits for a ZEXTLoad to zero.
These can be lowered to code sequences using CMPFP and CMPFPE which then get
selected to VCMP and VCMPE. The implementation isn't fully correct, as the chain
operand isn't handled correctly, but resolving that looks like it would involve
changes around FPSCR-handling instructions and how the FPSCR is modelled.
The fp-intrinsics test was already testing some of this but as the entire test
was being XFAILed it wasn't noticed. Un-XFAIL the test and instead leave the
cases where we aren't generating the right instruction sequences as FIXME.
Differential Revision: https://reviews.llvm.org/D73194
Summary:
In big-endian MVE, the simple vector load/store instructions (i.e.
both contiguous and non-widening) don't all store the bytes of a
register to memory in the same order: it matters whether you did a
VSTRB.8, VSTRH.16 or VSTRW.32. Put another way, the in-register
formats of different vector types relate to each other in a different
way from the in-memory formats.
So, if you want to 'bitcast' or 'reinterpret' one vector type as
another, you have to carefully specify which you mean: did you want to
reinterpret the //register// format of one type as that of the other,
or the //memory// format?
The ACLE `vreinterpretq` intrinsics are specified to reinterpret the
register format. But I had implemented them as LLVM IR bitcast, which
is specified for all types as a reinterpretation of the memory format.
So a `vreinterpretq` intrinsic, applied to values already in registers,
would code-generate incorrectly if compiled big-endian: instead of
emitting no code, it would emit a `vrev`.
To fix this, I've introduced a new IR intrinsic to perform a
register-format reinterpretation: `@llvm.arm.mve.vreinterpretq`. It's
implemented by a trivial isel pattern that expects the input in an
MQPR register, and just returns it unchanged.
In the clang codegen, I only emit this new intrinsic where it's
actually needed: I prefer a bitcast wherever it will have the right
effect, because LLVM understands bitcasts better. So we still generate
bitcasts in little-endian mode, and even in big-endian when you're
casting between two vector types with the same lane size.
For testing, I've moved all the codegen tests of vreinterpretq out
into their own file, so that they can have a different set of RUN
lines to check both big- and little-endian.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D73786
Summary:
These instructions generate a vector of consecutive elements starting
from a given base value and incrementing by 1, 2, 4 or 8. The `wdup`
versions also wrap the values back to zero when they reach a given
limit value. The instruction updates the scalar base register so that
another use of the same instruction will continue the sequence from
where the previous one left off.
At the IR level, I've represented these instructions as a family of
target-specific intrinsics with two return values (the constructed
vector and the updated base). The user-facing ACLE API provides a set
of intrinsics that throw away the written-back base and another set
that receive it as a pointer so they can update it, plus the usual
predicated versions.
Because the intrinsics return two values (as do the underlying
instructions), the isel has to be done in C++.
This is the first family of MVE intrinsics that use the `imm_1248`
immediate type in the clang Tablegen framework, so naturally, I found
I'd given it the wrong C integer type. Also added some tests of the
check that the immediate has a legal value, because this is the first
time those particular checks have been exercised.
Finally, I also had to fix a bug in MveEmitter which failed an
assertion when I nested two `seq` nodes (the inner one used to extract
the two values from the pair returned by the IR intrinsic, and the
outer one put on by the predication multiclass).
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D73357
Summary:
The unpredicated case of this is trivial: the clang codegen just makes
a vector splat of the input, and LLVM isel is already prepared to
handle that. For the predicated version, I've generated a `select`
between the same vector splat and the `inactive` input parameter, and
added new Tablegen isel rules to match that pattern into a predicated
`MVE_VDUP` instruction.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D73356
Summary:
A Copy with a source that is zeros is the same as a Set of zeros.
This fixes the invariant that SrcAlign should always be non-null.
Reviewers: courbet
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73791
Summary:
D68092 introduced a new SIRemoveShortExecBranches optimization pass and
broke some graphics shaders. The problem is that it was removing
branches over KILL pseudo instructions, and the fix is to explicitly
check for that in mustRetainExeczBranch.
Reviewers: critson, arsenm, nhaehnle, cdevadas, hakzsam
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73771
We don't need tests for truncating the result. There's nothing
special about those truncates.
We can test llrint/llround for 64-bit and 32-bit targets in the same file.
Same with lrint/lround with i32 result result. lrint/lround with
64-bit result should only occur on a 64-bit target.
Add some missing tests for f80 conversions.
The code paths in the absence of TargetMachine, TargetLowering or
TargetRegisterInfo are poorly tested. As rL285987 said, requiring
TargetPassConfig allows us to delete many (untested) checks littered
everywhere.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D73754
This reverts commit e34801c8e6df and the followup due to multiple
problems.
I've tried to keep the tests and RDA parts where possible, as those
still seem useful.
The current FirstMI.getDebugLoc() is actually null in almost all cases.
If it isn't, the generated .loc will be considered initial. The .loc
will have the prologue_end flag and terminate the prologue prematurely.
Also use an overload of BuildMI that will not prepend
PATCHABLE_FUNCTION_ENTRY to a MachineInstr bundle.
Summary:
Virtual registers that are undef have an empty LiveInterval at this
point, which means beginIndex() and endIndex() cannot be used. We
only need those indices to determine the range in which to scan for
affected other NSA instructions, and undef operands cannot contribute
to that range.
Reviewers: arsenm, rampitec, mareko
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D73831
We were checking that the original Value * for the compare operands
were null. But that can never happen.
I believe we intended to check for 0 registers here instead.
Fixes PR44749.
This is based on this llvm-dev thread http://lists.llvm.org/pipermail/llvm-dev/2019-December/137521.html
The current strategy for f16 is to promote type to float every except where the specific width is required like loads, stores, and bitcasts. This results in rounding occurring in odd places instead of immediately after arithmetic operations. This interacts in weird ways with the __fp16 type in clang which is a storage only type where arithmetic is always promoted to float. InstCombine can remove some fpext/fptruncs around such arithmetic and turn it into arithmetic on half. This wouldn't be so bad if SelectionDAG was able to put those fpext/fpround back in when it promotes.
It is also not obvious how to handle to make the existing strategy work with STRICT fp. We need to use STRICT versions of the conversions which require chain operands. But if the conversions are created for a bitcast, there is no place to get an appropriate chain from.
This patch implements a different strategy where conversions are emitted directly around arithmetic operations. And otherwise its passed around as an i16 including in arguments and return values. This can result in more conversions between arithmetic operations, but is closer to matching the IR the frontend generates for __fp16. And it will allow us to use the chain from constrained arithmetic nodes to link the STRICT_FP_TO_FP16/STRICT_FP16_TO_FP that will need to be added. I've set it up so that each target can opt into the new behavior. Converting all the targets myself was more than I was able to handle.
Differential Revision: https://reviews.llvm.org/D73749
This fixes legalizations of global stores > 128-bits. It seems work is
needed on how this split actually occurs. For example, we get the
right code for s160, with an s128 and s32 load, but get 5 s32 loads
for <5 x s32>.
When you encounter a G_TRUNC, you are moving from a larger type to a smaller
type.
Asking for the i-th bit on a larger value is the same as asking for the i-th
bit on a smaller value.
So, we should always be able to walk through G_TRUNC when computing the bit
for a TB(N)Z.
Differential Revision: https://reviews.llvm.org/D73748
