Iterate through the loop and check that the observable values
produced are the same whether tail predication happens or not.
We want to find out if the tail-predicated version of this loop will
produce the same values as the loop in its original form. For this to
be true, the newly inserted implicit predication must not change the
the (observable) results.
We're doing this because many instructions in the loop will not be
predicated and so the conversion from VPT predication to tail
predication can result in different values being produced, because of
falsely predicated lanes not being updated in the converted form.
A masked load, whether through VPT or tail predication, will write
zeros to any of the falsely predicated bytes. So, from the loads, we
know that the false lanes are zeroed and here we're trying to track
that those false lanes remain zero, or where they change, the
differences are masked away by their user(s).
All MVE loads and stores have to be predicated, so we know that any
load operands, or stored results are equivalent already. Other
explicitly predicated instructions will perform the same operation in
the original loop and the tail-predicated form too. Because of this,
we can insert loads, stores and other predicated instructions into
our KnownFalseZeros set and build from there.
Differential Revision: https://reviews.llvm.org/D75452
This teaches Loop Strength Reduction the details about masked load and
store address operands, so that it can have a better time optimising
them as it would for normal loads and stores.
Differential Revision: https://reviews.llvm.org/D75371
Summary:
The VSHLC instruction performs a left shift of a whole vector register
by an immediate shift count up to 32, shifting in new bits at the low
end from a GPR and delivering the shifted-out bits from the high end
back into the same GPR.
Since the instruction produces two outputs (the shifted vector
register and the output GPR of shifted-out bits), it has to be
instruction-selected in C++ rather than Tablegen.
Reviewers: MarkMurrayARM, dmgreen, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75445
Summary:
These are exactly parallel to the existing `vadciq` intrinsics, which
we implemented last year as part of the original MVE intrinsics
framework setup.
Just like VADC/VADCI, the MVE VSBC/VSBCI instructions deliver two
outputs, both of which the intrinsic exposes: a modified vector
register and a carry flag. So they have to be instruction-selected in
C++ rather than Tablegen. However, in this case, that's trivial: the
same C++ isel routine we already have for VADC works unchanged, and
all we have to do is to pass it a different instruction id.
Reviewers: MarkMurrayARM, dmgreen, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75444
Use MIOperand in collectLocalKilledOperands to make the search
global, as we already have to search for global uses too. This
allows us to delete more dead code when tail predicating.
Differential Revision: https://reviews.llvm.org/D75167
One of the checks has been removed as it seem invalid.
The LoopStep size is always almost a 32-bit.
Differential Revision: https://reviews.llvm.org/D75079
Summary:
These instructions convert a vector of floats to a vector of integers
of the same size, with assorted non-default rounding modes.
Implemented in IR as target-specific intrinsics, because as far as I
can see there are no matches for that functionality in the standard IR
intrinsics list.
Reviewers: MarkMurrayARM, dmgreen, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75255
Summary:
These instructions make a vector of `<4 x float>` by widening every
other lane of a vector of `<8 x half>`.
I wondered about representing these using standard IR, along the lines
of a shufflevector to extract elements of the input into a `<4 x half>`
followed by an `fpext` to turn that into `<4 x float>`. But it looks as
if that would take a lot of work in isel lowering to make it match any
pattern I could sensibly write in Tablegen, and also I haven't been
able to think of any other case where that pattern might be generated
in IR, so there wouldn't be any extra code generation win from doing
it that way.
Therefore, I've just used another target-specific intrinsic. We can
always change it to the other way later if anyone thinks of a good
reason.
(In order to put the intrinsic definition near similar things in
`IntrinsicsARM.td`, I've also lifted the definition of the
`MVEMXPredicated` multiclass higher up the file, without changing it.)
Reviewers: MarkMurrayARM, dmgreen, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75254
Summary:
These instructions work like VMOVN (narrowing a vector of wide values
to half size, and overwriting every other lane of an output register
with the result), except that the narrowing conversion is saturating.
They come in three signedness flavours: signed to signed, unsigned to
unsigned, and signed to unsigned. All are represented in IR by a
target-specific intrinsic that takes two separate 'unsigned' flags.
Reviewers: MarkMurrayARM, dmgreen, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75252
The MVE gather instructions smaller than 32bits zext extend the values
in the offset register, as opposed to sign extending them. We need to
make sure that the code that we select from is suitably extended, which
this patch attempts to fix by tightening up the offset checks.
Differential Revision: https://reviews.llvm.org/D75361
Summary:
There exists no corresponding code change for this commit, and this
commit causes downstream breakages.
This reverts commit 2db5547c01.
Reviewers: samparker
Subscribers: kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75358
This addes extra patterns for the VMLAS MVE instruction, which performs
Qda = Qda * Qn + Rm, a similar pattern to the existing VMLA. The sinking
of splat(Rm) into the loop is already performed, meaning we just need
extra Pat's in tablegen.
Differential Revision: https://reviews.llvm.org/D75115
Ensure that we're recording implicit defs, as well as visiting implicit
uses and implicit defs when we're walking through operands.
Differential Revision: https://reviews.llvm.org/D75185
This will address the issue: P8198 and P8199 (from D73534).
The methods was not handle bundles properly.
Differential Revision: https://reviews.llvm.org/D74904
Under fp16 we optimise the bitcast between a VMOVhr and a CopyToReg via
custom lowering. This rewrites that to be a DAG combine instead, which
helps produce better code in the cases where the bitcast is actaully
legal.
Differential Revision: https://reviews.llvm.org/D72753
Summary:
This commit adds the predicated MVE intrinsics for the same set of
unary operations that I added in their unpredicated forms in
* D74333 (vrint)
* D74334 (vrev)
* D74335 (vclz, vcls)
* D74336 (vmovl)
* D74337 (vmovn)
but since the predicated versions are a lot more similar to each
other, I've kept them all together in a single big patch. Everything
here is done in the standard way we've been doing other predicated
operations: an IR intrinsic called `@llvm.arm.mve.foo.predicated` and
some isel rules that match that alongside whatever they accept for the
unpredicated version of the same instruction.
In order to write the isel rules conveniently, I've refactored the
existing isel rules for the affected instructions into multiclasses
parametrised by a vector-type class, in the usual way. All those
refactorings are intended to leave the existing isel rules unchanged:
the only difference should be that new ones for the predicated
intrinsics are introduced.
The only tiny infrastructure change I needed in this commit was to
change the implementation of `IntrinsicMX` in `arm_mve_defs.td` so
that the records it defines are anonymous rather than named (and use
`NameOverride` to set the output intrinsic name), which allows me to
call it twice in two multiclasses with the same `NAME` without a
tablegen-time error.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75165
This adds infrastructure to print and parse MIR MachineOperand comments.
The motivation for the ARM backend is to print condition code names instead of
magic constants that are difficult to read (for human beings). For example,
instead of this:
dead renamable $r2, $cpsr = tEOR killed renamable $r2, renamable $r1, 14, $noreg
t2Bcc %bb.4, 0, killed $cpsr
we now print this:
dead renamable $r2, $cpsr = tEOR killed renamable $r2, renamable $r1, 14 /* CC::always */, $noreg
t2Bcc %bb.4, 0 /* CC:eq */, killed $cpsr
This shows that MachineOperand comments are enclosed between /* and */. In this
example, the EOR instruction is not conditionally executed (i.e. it is "always
executed"), which is encoded by the 14 immediate machine operand. Thus, now
this machine operand has /* CC::always */ as a comment. The 0 on the next
conditional branch instruction represents the equal condition code, thus now
this operand has /* CC:eq */ as a comment.
As it is a comment, the MI lexer/parser completely ignores it. The benefit is
that this keeps the change in the lexer extremely minimal and no target
specific parsing needs to be done. The changes on the MIPrinter side are also
minimal, as there is only one target hooks that is used to create the machine
operand comments.
Differential Revision: https://reviews.llvm.org/D74306
Change the way that we remove the redundant iteration count code in
the presence of IT blocks. collectLocalKilledOperands has been
introduced to scan an instructions operands, collecting the killed
instructions and then visiting them too. This is used to delete the
code in the preheader which calculates the iteration count. We also
track any IT blocks within the preheader and, if we remove all the
instructions from the IT block, we also remove the IT instruction.
isSafeToRemove is used to remove any redundant uses of the iteration
count within the loop body.
Differential Revision: https://reviews.llvm.org/D74975
Summary:
This patch adds two families of ACLE intrinsics: vqdmullbq and
vqdmulltq (including vector-vector and vector-scalar variants) and the
corresponding LLVM IR intrinsics llvm.arm.mve.vqdmull and
llvm.arm.mve.vqdmull.predicated.
Reviewers: simon_tatham, MarkMurrayARM, dmgreen, ostannard
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74845
Summary:
Some predicated MVE intrinsics return a vector with element size
different from the input vector element size. In this case the
predicate must type correspond to the output vector type.
The following intrinsics use the incorrect predicate type:
* llvm.arm.mve.mull.int.predicated
* llvm.arm.mve.mull.poly.predicated
* llvm.arm.mve.vshll.imm.predicated
This patch fixes the issue.
Reviewers: simon_tatham, dmgreen, ostannard, MarkMurrayARM
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74838
Check that no Q-regs are live out of the loop, unless the instruction
within the loop is predicated on the vctp.
Differential Revision: https://reviews.llvm.org/D72713
Similar to VADDV and VADDLV that have been added recently, this adds
lowering and patterns for VMLAV, VMLAVA, VMLALV and VMLALVA. They
perform the same roles as the add's, just folding a mul into the same
instruction (and so taking two inputs). As such, they need to be lowered
in the same way as the types are often not legal.
Differential Revision: https://reviews.llvm.org/D74390
Following on from the extra VADDV lowering, this extends things to
handle VADDLV which allows summing values into a pair of i32 registers,
together treated as a i64. This needs to be done in DAGCombine too as
the types are otherwise illegal, which is a fairly simple addition on
top of the existing code.
There is also a VADDLVA instruction handled here, that adds the incoming
values from the two general purpose registers. As opposed to the
non-long version where we could just add patterns for add(x, VADDV), the
long version needs to handle this early before the i64 has being split
into too many pieces.
Differential Revision: https://reviews.llvm.org/D74224
We already make use of the VADDV vector reduction instruction for cases
where the input and the output start out at the same type. The MVE
instruction however will sum into an i32, so if we are summing a v16i8
into an i32, we can still use the same instructions. In terms of IR,
this looks like a sext of a legal type (v16i8) into a very illegal type
(v16i32) and a vecreduce.add of that into the result. This means we have
to catch the pattern early in a DAG combine, producing a target VADDVs/u
node, where the signedness is now important.
This is the first part, handling VADDV and VADDVA. There are also
VADDVL/VADDVLA instructions, which are interesting because they sum into
a 64bit value. And VMLAV and VMLALV, which are interesting because they
also do a multiply of two values. It may look a little odd in places as
a result.
On it's own this will probably not do very much, as the vectorizer will
not produce this IR yet.
Differential Revision: https://reviews.llvm.org/D74218
Summary:
This patch adds a new MVE intrinsics family, `vbrsrq`: vector bit
reverse and shift right. The intrinsics are compiled into the VBRSR
instruction. Two new LLVM IR intrinsics were also added: arm.mve.vbrsr
and arm.mve.vbrsr.predicated.
Reviewers: simon_tatham, dmgreen, ostannard, MarkMurrayARM
Reviewed By: simon_tatham
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74721
Summary:
These are in some sense the inverse of vmovl[bt]q: they take a vector
of n wide elements and truncate each to half its width. So they only
write half a vector's worth of output data, and therefore they also
take an 'inactive' parameter to provide the other half of the data in
the output vector. So vmovnb overwrites the even lanes of 'inactive'
with the narrowed values from the main input, and vmovnt overwrites
the odd lanes.
LLVM had existing codegen which generates these MVE instructions in
response to IR that takes two vectors of wide elements, or two vectors
of narrow ones. But in this case, we have one vector of each. So my
clang codegen strategy is to narrow the input vector of wide elements
by simply reinterpreting it as the output type, and then we have two
narrow vectors and can represent the operation as a vector shuffle
that interleaves lanes from both of them.
Even so, not all the cases I needed ended up being selected as a
single MVE instruction, so I've added a couple more patterns that spot
combinations of the 'MVEvmovn' and 'ARMvrev32' SDNodes which can be
generated as a VMOVN instruction with operands swapped.
This commit adds the unpredicated forms only.
Reviewers: dmgreen, miyuki, MarkMurrayARM, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74337
Summary:
These intrinsics take a vector of 2n elements, and return a vector of
n wider elements obtained by sign- or zero-extending every other
element of the input vector. They're represented in IR as a
shufflevector that extracts the odd or even elements of the input,
followed by a sext or zext.
Existing LLVM codegen already matches this pattern and generates the
VMOVLB instruction (which widens the even-index input lanes). But no
existing isel rule was generating VMOVLT, so I've added some. However,
the new rules currently only work in little-endian MVE, because the
pattern they expect from isel lowering includes a bitconvert which
doesn't have the right semantics in big-endian.
The output of one existing codegen test is improved by those new
rules.
This commit adds the unpredicated forms only.
Reviewers: dmgreen, miyuki, MarkMurrayARM, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74336
Summary:
vclzq maps nicely to the existing target-independent @llvm.ctlz IR
intrinsic. But vclsq ('count leading sign bits') has no corresponding
target-independent intrinsic, so I've made up @llvm.arm.mve.vcls.
This commit adds the unpredicated forms only.
Reviewers: dmgreen, miyuki, MarkMurrayARM, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74335
Summary:
This adds the unpredicated forms of six different MVE intrinsics which
all round a vector of floating-point numbers to integer values,
leaving them still in FP format, differing only in rounding mode and
exception settings.
Five of them map to existing target-independent intrinsics in LLVM IR,
such as @llvm.trunc and @llvm.rint. The sixth, mapping to the `vrintn`
instruction, is done by inventing a target-specific intrinsic.
(`vrintn` behaves the same as `vrintx` in terms of the output value:
the side effects on the FPSCR flags are the only difference between
the two. But ACLE specifies separate user-callable intrinsics for the
two, so the side effects matter enough to make sure we generate the
right one of the two instructions in each case.)
Reviewers: dmgreen, miyuki, MarkMurrayARM, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74333
Summary:
This patch adds vector-scalar variants to the following families of
MVE intrinsics:
* vaddq
* vsubq
* vmulq
* vqaddq
* vqsubq
* vhaddq
* vhsubq
* vqdmulhq
* vqrdmulhq
The vector-scalar variants perform a splat operation on the scalar
operand and then perform the same operations as their vector-vector
counterparts. Code generation is done accordingly (using LLVM IR 'insert'
and 'shuffle' operations which are later converted into an ARMvdup
SDNode).
Reviewers: simon_tatham, dmgreen, MarkMurrayARM, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D74620
This adds another pattern to the combiner for a case that we were not handling
to generate the REV16 instruction for ARM/Thumb2 and a bswap+ror on X86.
Differential Revision: https://reviews.llvm.org/D74032
Simon pointed out that this function is doing a bitcast, which can be
incorrect for big endian. That makes the lowering of VMOVN in MVE
wrong, but the function is shared between Neon and MVE so both can
be incorrect.
This attempts to fix things by using the newly added VECTOR_REG_CAST
instead of the BITCAST. As it may now be used on Neon, I've added the
relevant patterns for it there too. I've also added a quick dag combine
for it to remove them where possible.
Differential Revision: https://reviews.llvm.org/D74485
A small IR change in calculating the active lanes resulted in no longer
recognising tail-predication. Now recognise both an 'add' and 'or' in
the expression that calculates the active lanes.
Differential Revision: https://reviews.llvm.org/D74394
Printing floating point number in decimal is inconvenient for humans.
Verbose asm output will print out floating point values in comments, it
helps.
But in lots of cases, users still need additional work to covert the
decimal back to hex or binary to check the bit patterns,
especially when there are small precision difference.
Hexadecimal form is one of the supported form in LLVM IR, and easier for
debugging.
This patch try to print all FP constant in hex form instead.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D73566
Once we have created a tail-predicated hardware-loop, and thus know the number
of elements that are processed, we want to clean-up the iteration count
expression of that loop. In D73682, we bailed the analysis on conditionally
executed instructions. This adds support for IT-blocks, so that we can handle
these cases again. The restriction is that we only support IT blocks containing
1 statement, but that seems to cover most cases and forms of the iteration
count expression.
Differential Revision: https://reviews.llvm.org/D73947
Checking that the use-def chain that performs the loop count
isSafeToRemove is not sufficient because it means that we can
remove register copies that we need to restore lr to its correct
value. This change now prevents the transform from kicking in for the
'remove-elem-moves' test which needs to addressed later on.
Differential Revision: https://reviews.llvm.org/D74037
While validating each MVE instruction, check that all instructions
that touch memory are somehow predicated upon the VCTP.
Differential Revision: https://reviews.llvm.org/D73616
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.
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
This reverts commit e34801c8e6 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 MVE_VLDRWU32_qi_pre gather loads, like the other _pre/_post mve
loads returns the writeback as result 0, the value as result 1. The llvm
ir intrinsic seems to have this the other way around though, and so when
lowering from one to the other we need to switch the first two outputs.
I've also fixed up the types of _pre/_post on normal MVE loads. There we
were already getting the values the right way around, just not for the
types. I don't believe this was causing anything to go wrong, but it was
very confusing to read in the debug output.
Differential Revision: https://reviews.llvm.org/D73370
We had support for runtime trip count values, but not constants, and this adds
supports for that.
And added a minor optimisation while I was add it: don't invoke Cleanup when
there's nothing to clean up.
Differential Revision: https://reviews.llvm.org/D73198
When expanding the LoopStart, we try to remove the iteration count
calculation. However, if part of the calculation was also used to
calculate the number of elements we could end up deleting
instructions that were required to feed DLSTP/WLSTP.
Differential Revision: https://reviews.llvm.org/D73275
The codegen for splitting a llvm.vector.reduction intrinsic into parts
will be better than the codegen for the generic reductions. This will
only directly effect when vectorization factors are specified by the
user.
Also added tests to make sure the codegen for larger reductions is OK.
Differential Revision: https://reviews.llvm.org/D72257
Summary:
Immediate vmvnq is code-generated as a simple vector constant in IR,
and left to the backend to recognize that it can be created with an
MVE VMVN instruction. The predicated version is represented as a
select between the input and the same constant, and I've added a
Tablegen isel rule to turn that into a predicated VMVN. (That should
be better than the previous VMVN + VPSEL: it's the same number of
instructions but now it can fold into an adjacent VPT block.)
The unpredicated forms of VBIC and VORR are done by enabling the same
isel lowering as for NEON, recognizing appropriate immediates and
rewriting them as ARMISD::VBICIMM / ARMISD::VORRIMM SDNodes, which I
then instruction-select into the right MVE instructions (now that I've
also reworked those instructions to use the same MC operand encoding).
In order to do that, I had to promote the Tablegen SDNode instance
`NEONvorrImm` to a general `ARMvorrImm` available in MVE as well, and
similarly for `NEONvbicImm`.
The predicated forms of VBIC and VORR are represented as a vector
select between the original input vector and the output of the
unpredicated operation. The main convenience of this is that it still
lets me use the existing isel lowering for VBICIMM/VORRIMM, and not
have to write another copy of the operand encoding translation code.
This intrinsic family is the first to use the `imm_simd` system I put
into the MveEmitter tablegen backend. So, naturally, it showed up a
bug or two (emitting bogus range checks and the like). Fixed those,
and added a full set of tests for the permissible immediates in the
existing Sema test.
Also adjusted the isel pattern for `vmovlb.u8`, which stopped matching
because lowering started turning its input into a VBICIMM. Now it
recognizes the VBICIMM instead.
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/D72934
The hasSideEffect parameter is usually automatically inferred from
instruction patterns. For some of our MVE instructions, we do not have
patterns though, such as for the pre/post inc loads and stores. This
instead specifies the flag manually on the base MVE_VLDRSTR_base
tablegen class, making sure we get this correct.
This can help with scheduling multiple loads more optimally. Here I've
added a unittest as a more direct form of testing.
Differential Revision: https://reviews.llvm.org/D73117
Extends the gather/scatter pass in MVEGatherScatterLowering.cpp to
enable the transformation of masked scatters into calls to MVE's masked
scatter intrinsic.
Differential Revision: https://reviews.llvm.org/D72856
This patch uses helper function rewriteLoopExitValues that is refactored in
D72602 to rematerialise the iteration count in exit blocks, so that we can
clean-up loop update expressions inside the hardware-loops later in
ARMLowOverheadLoops, which is necessary to get actual performance gains for
tail-predicated loops.
Differential Revision: https://reviews.llvm.org/D72714
This adds Post inc variants of the VLD2/4 and VST2/4 instructions in
MVE. It uses the same mechanism/nodes as Neon, transforming the
intrinsic+add pair into a ARMISD::VLD2_UPD, which gets selected to a
post-inc instruction. The code to do that is mostly taken from the
existing Neon code, but simplified as less variants are needed.
It also fills in some getTgtMemIntrinsic for the arm.mve.vld2/4
instrinsics, which allow the nodes to have MMO's, calculated as the full
length to the memory being loaded/stored.
Differential Revision: https://reviews.llvm.org/D71194
We were previously not necessarily favouring postinc for the MVE loads
and stores, leading to extra code prior to the loop to set up the
preinc. MVE in general can benefit from postinc (as we don't have
unrolled loops), and certain instructions like the VLD2's only post-inc
versions are available.
Differential Revision: https://reviews.llvm.org/D70790
Introduce a method to walk through use-def chains to decide whether
it's possible to remove a given instruction and its users. These
instructions are then stored in a set until the end of the transform
when they're erased. This is now used to perform checks on the
iteration count (LoopDec chain), element count (VCTP chain) and the
possibly redundant iteration count.
As well as being able to remove chains of instructions, we know also
check that the sub feeding the vctp is producing the expected value.
Differential Revision: https://reviews.llvm.org/D71837
Recommitting e93e0d413f after reverting due to test failures, which
will hopefully now be fixed. Original commit message:
After expanding the pseudo instructions, update the liveness info.
We do this in a post-order traversal of the loop, including its
exit blocks and preheader(s).
Differential Revision: https://reviews.llvm.org/D72131
The mve-phireg.ll test no longer really tests what it was added for,
but the original case was fairly complex. I've left the test in as a
general codegen test.
Fix a missing and broken test: 2 VPT blocks predicated on the same VCMP
instruction that can be folded. The problem was that for each VPT block, we
record the predicate statements with a list, but the same instruction was added
twice. Thus, we were running in an assert trying to remove the same instruction
twice. To avoid this the instructions are now recorded with a set.
Differential Revision: https://reviews.llvm.org/D72699
We have a whitelist of instructions that we allow when tail
predicating, since these are trivial ones that we've deemed need no
special handling. Now change ARMLowOverheadLoops to allow the
non-trivial instructions if they're contained within a valid VPT
block. Since a valid block is one that is predicated upon the VCTP so
we know that these non-trivial instructions will still behave as
expected once the implicit predication is used instead.
This also fixes a previous test failure.
Differential Revision: https://reviews.llvm.org/D72509
Use the already provided helper function to get the operand type so
that we can detect whether the vpr is being used as a predicate or
not. Also use existing helpers to get the predicate indices when we
converting the vpt blocks. This enables us to support both types of
vpr predicate operand.
Differential Revision: https://reviews.llvm.org/D72504
Summary:
This patch fixes pr23772 [ARM] r226200 can emit illegal thumb2 instruction: "sub sp, r12, #80".
The violation was that SUB and ADD (reg, immediate) instructions can only write to SP if the source register is also SP. So the above instructions was unpredictable.
To enforce that the instruction t2(ADD|SUB)ri does not write to SP we now enforce the destination register to be rGPR (That exclude PC and SP).
Different than the ARM specification, that defines one instruction that can read from SP, and one that can't, here we inserted one that can't write to SP, and other that can only write to SP as to reuse most of the hard-coded size optimizations.
When performing this change, it uncovered that emitting Thumb2 Reg plus Immediate could not emit all variants of ADD SP, SP #imm instructions before so it was refactored to be able to. (see test/CodeGen/Thumb2/mve-stacksplot.mir where we use a subw sp, sp, Imm12 variant )
It also uncovered a disassembly issue of adr.w instructions, that were only written as SUBW instructions (see llvm/test/MC/Disassembler/ARM/thumb2.txt).
Reviewers: eli.friedman, dmgreen, carwil, olista01, efriedma, andreadb
Reviewed By: efriedma
Subscribers: gbedwell, john.brawn, efriedma, ostannard, kristof.beyls, hiraditya, dmgreen, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70680
Due to the current way that we collect predicated instructions, we
can't easily handle vpsel in tail predicated loops. There are a
couple of issues:
1) It will use the VPR as a predicate operand, but doesn't have to be
instead a VPT block, which means we can assert while building up
the VPT block because we don't find another VPST to being a new
one.
2) VPSEL still requires a VPR operand even after tail predicating,
which means we can't remove it unless there is another
instruction, such as vcmp, that can provide the VPR def.
The first issue should be a relatively simple fix in the logic of the
LowOverheadLoops pass, whereas the second will require us to
represent the 'implicit' tail predication with an explicit value.
Differential Revision: https://reviews.llvm.org/D72629
Enables the masked gather pass to create a masked
gather loading from a base and vector of offsets.
This also enables v8i16 and v16i8 gather loads.
Differential Revision: https://reviews.llvm.org/D72330
ONE is currently softened to OGT | OLT. But the libcalls for OGT and OLT libcalls will trigger an exception for QNAN. At least for X86 with libgcc. UEQ on the other hand uses UO | OEQ. The UO and OEQ libcalls will not trigger an exception for QNAN.
This patch changes ONE to use the inverse of the UEQ lowering. So we now produce O & UNE. Technically the existing behavior was correct for a signalling ONE, but since I don't know how to generate one of those from clang that seemed like something we can deal with later as we would need to fix other predicates as well. Also removing spurious exceptions seemed better than missing an exception.
There are also problems with quiet OGT/OLT/OLE/OGE, but those are harder to fix.
Differential Revision: https://reviews.llvm.org/D72477
Summary:
This patch fixes pr23772 [ARM] r226200 can emit illegal thumb2 instruction: "sub sp, r12, #80".
The violation was that SUB and ADD (reg, immediate) instructions can only write to SP if the source register is also SP. So the above instructions was unpredictable.
To enforce that the instruction t2(ADD|SUB)ri does not write to SP we now enforce the destination register to be rGPR (That exclude PC and SP).
Different than the ARM specification, that defines one instruction that can read from SP, and one that can't, here we inserted one that can't write to SP, and other that can only write to SP as to reuse most of the hard-coded size optimizations.
When performing this change, it uncovered that emitting Thumb2 Reg plus Immediate could not emit all variants of ADD SP, SP #imm instructions before so it was refactored to be able to. (see test/CodeGen/Thumb2/mve-stacksplot.mir where we use a subw sp, sp, Imm12 variant )
It also uncovered a disassembly issue of adr.w instructions, that were only written as SUBW instructions (see llvm/test/MC/Disassembler/ARM/thumb2.txt).
Reviewers: eli.friedman, dmgreen, carwil, olista01, efriedma
Reviewed By: efriedma
Subscribers: john.brawn, efriedma, ostannard, kristof.beyls, hiraditya, dmgreen, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D70680
Architecturally, it's allowed to have MVE-I without an FPU, thus
-mfpu=none should not disable MVE-I, or moves to/from FP-registers.
This patch removes `+/-fpregs` from features unconditionally added to
target feature list, depending on FPU and moves the logic to Clang
driver, where the negative form (`-fpregs`) is conditionally added to
the target features list for the cases of `-mfloat-abi=soft`, or
`-mfpu=none` without either `+mve` or `+mve.fp`. Only the negative
form is added by the driver, the positive one is derived from other
features in the backend.
Differential Revision: https://reviews.llvm.org/D71843
After expanding the pseudo instructions, update the liveness info.
We do this in a post-order traversal of the loop, including its
exit blocks and preheader(s).
Differential Revision: https://reviews.llvm.org/D72131
This batch of intrinsics fills in all the shift instructions that take
a variable shift distance in a register, instead of an immediate. Some
of these instructions take a single shift distance in a scalar
register and apply it to all lanes; others take a vector of per-lane
distances.
These instructions are all basically one family, varying in whether
they saturate out-of-range values, and whether they round when bits
are shifted off the bottom. I've implemented them at the IR level by a
much smaller family of IR intrinsics, which take flag parameters to
indicate saturating and/or rounding (along with the usual one to
specify signed/unsigned integers).
An oddity is that all of them are //left// shift instructions – but if
you pass a negative shift count, they'll shift right. So the vector
shift distances are always vectors of //signed// integers, regardless
of whether you're considering the other input vector to be of signed
or unsigned. Also, even the simplest `vshlq` instruction in this
family (neither saturating nor rounding) has to be implemented as an
IR intrinsic, because the ordinary LLVM IR `shl` operation would
consider an out-of-range shift count to be undefined behavior.
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/D72329
This batch of intrinsics covers two sets of immediate shift
instructions, which have in common that they only overwrite part of
their output register and so they need an extra input giving its
previous value.
The VSLI and VSRI instructions shift each lane of the input vector
left or right just as if they were normal immediate VSHL/VSHR, but
then they only overwrite the output bits that correspond to actual
shifted bits of the input. So VSLI will leave the low n bits of each
output lane unchanged, and VSRI the same with the top n bits.
The V[Q][R]SHR[U]N family are all narrowing shifts: they take an input
vector of 2n-bit integers, shift each lane right by a constant, and
then narrowing the shifted result to only n bits. So they only
overwrite half of the n-bit lanes in the output register, and the B/T
suffix indicates whether it's the bottom or top half of each 2n-bit
lane.
I've implemented the whole of the latter family using a single IR
intrinsic `vshrn`, which takes a lot of i32 parameters indicating
which instruction it expands to (by specifying signedness of the input
and output types, whether it saturates and/or rounds, etc).
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/D72328
David Green