Add patterns that use a normal, non-wrapping, add and sub nodes along
with an arm vshr imm node.
Differential Revision: https://reviews.llvm.org/D77065
Fix for the address optimization for gathers and scatters which would in
some complex cases push out instructions not to the vector loop preheader,
but to other locations as well which lead to a scrambled order and the
compilation failing.
This patch ensures that said instructions are always pushed to the end
of the vector loop preheader.
Differential Revision: https://reviews.llvm.org/D78293
Summary:
The INLINEASM MIR instructions use immediate operands to encode the values of some operands.
The MachineInstr pretty printer function already handles those operands and prints human readable annotations instead of the immediates. This patch adds similar annotations to the output of the MIRPrinter, however uses the new MIROperandComment feature.
Reviewers: SjoerdMeijer, arsenm, efriedma
Reviewed By: arsenm
Subscribers: qcolombet, sdardis, jvesely, wdng, nhaehnle, hiraditya, jrtc27, atanasyan, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D78088
The pass was incorrectly reverting back to a "T" when something wrote
to VPR inside a "E" block. This is not the correct behaviour, the
predicate should stay the same.
Differential Revision: https://reviews.llvm.org/D77798
This patch adds an analysis of the offset addresses used by gathers
and scatters to the MVEGatherScatterLowering pass to find
multiplications and additions that are loop invariant and thus can
be moved into the loop preheader, avoiding to execute them each time.
Differential Revision: https://reviews.llvm.org/D76681
If the stack pointer is altered for local variables and we are generating
Thumb2 execute-only code the .pad directive is missing.
Usually the size of the adjustment is stored in a PC-relative location
and loaded into a register which is then added to the stack pointer.
However when we are generating execute-only code code the size of the
adjustment is instead generated using the MOVW/MOVT instruction pair.
As a by product of handling the execute-only case this also fixes an
existing issue that in the none execute-only case the .pad directive was
generated against the load of the constant to a register instruction,
instead of the instruction which adds the register to the stack pointer.
Differential Revision: https://reviews.llvm.org/D76849
Currently when the target is big-endian vmov.i64 reverses the order of the two
words of the vector. This is correct only when the underlying element type is
32-bit, as actually what it should be doing is considering it a vector of the
underlying type and reversing the elements of that.
Differential Revision: https://reviews.llvm.org/D76515
If we have an element-wise vmov immediate instruction then a subsequent vrev
with width greater or equal to the vmov element width, then that vrev won't do
anything. Add a DAG combine to convert bitcasts that would become such vrevs
into vector_reg_casts instead.
Differential Revision: https://reviews.llvm.org/D76514
This adds MVE vmull patterns, which are conceptually the same as
mul(vmovl, vmovl), and so the tablegen patterns follow the same
structure.
For i8 and i16 this is simple enough, but in the i32 version the
multiply (in 64bits) is illegal, meaning we need to catch the pattern
earlier in a dag fold. Because bitcasts are involved in the zext
versions and the patterns are a little different in little and big
endian. I have only added little endian support in this patch.
Differential Revision: https://reviews.llvm.org/D76740
The unpredictable/hasSideEffects flag is usually inferred by tablegen
from whether the instruction has a tablegen pattern (and that pattern
only has a single output instruction). Now that the MVE intrinsics are
all committed and producing code, the remaining instructions still
marked as unpredictable need to be specially handled. This adds the flag
directly to instructions that need it, notably the V*MLAL instructions
and some of the MOV's.
Differential Revision: https://reviews.llvm.org/D76910
This allows the MVE VPT Block insertion pass to remove VPNOTs in
order to create more complex VPT blocks such as TE, TEET, TETE, etc.
Differential Revision: https://reviews.llvm.org/D75993
Add a bit more logic into the 'FalseLaneZeros' tracking to enable
horizontal reductions and also make the VADDV variants
validForTailPredication.
Differential Revision: https://reviews.llvm.org/D76708
In the original batch of MVE VMOVimm code generation VMOV.i64 was left
out due to the way it was done downstream. It turns out that it's fairly
simple though. This adds the codegen for it, similar to NEON.
Bigendian is technically incorrect in this version, which John is fixing
in a Neon patch.
Given that some instructions generate wider result elements than
their inputs, flag them as being able to generate non zeros in the
false lanes.
Differential Revision: https://reviews.llvm.org/D76766
Some MVE floating point instructions have gpr register variants that take
the scalar gpr value and splat them to all lanes. In order to accept
them in loops, the shuffle_vector and insert need to be sunk down into
the loop, next to the instruction so that ISel can see the whole
pattern.
This does that sinking for FAdd, FSub, FMul and FCmp. The patterns for
mul are slightly more constrained as there are no fms variants taking
register arguments.
Differential Revision: https://reviews.llvm.org/D76023
Summary:
This patch implements the following CDE intrinsics:
T __arm_vcx1q_m(int coproc, T inactive, uint32_t imm, mve_pred_t p);
T __arm_vcx2q_m(int coproc, T inactive, U n, uint32_t imm, mve_pred_t p);
T __arm_vcx3q_m(int coproc, T inactive, U n, V m, uint32_t imm, mve_pred_t p);
T __arm_vcx1qa_m(int coproc, T acc, uint32_t imm, mve_pred_t p);
T __arm_vcx2qa_m(int coproc, T acc, U n, uint32_t imm, mve_pred_t p);
T __arm_vcx3qa_m(int coproc, T acc, U n, V m, uint32_t imm, mve_pred_t p);
The intrinsics are not part of the released ACLE spec, but internally at
Arm we have reached consensus to add them to the next ACLE release.
Reviewers: simon_tatham, MarkMurrayARM, ostannard, dmgreen
Reviewed By: simon_tatham
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76610
Summary:
These were accidentally left out of D76123. I added tests for the
other three instructions in this small cross-product family (vqdmlah,
vqrdmlah, vqrdmlash) but missed this one.
Reviewers: miyuki
Reviewed By: miyuki
Subscribers: kristof.beyls, dmgreen, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76714
This adds a simple fold to combine VMOVrh load to a integer load.
Similar to what is already performed for BITCAST, but needs to account
for the types being of different sizes, creating an zero extending load.
Differential Revision: https://reviews.llvm.org/D76485
We deliberately split stores of the form
store(truncate(larger-than-legal-type)) into two stores, allowing each
store to perform part of the truncate for free.
There are times however where it makes more sense to use VMOVN to
de-interlace the results back into a single vector, and store that in
one go. This adds a check for that situation, not splitting the store if
it looks like a VMOVN can be more useful.
Differential Revision: https://reviews.llvm.org/D76511
When decided whether to generate a post-inc load/store, look at the
other memory nodes that use the same base address and, if any proceed
the current node, then don't do the combine.
The change only seems to be affecting the Arm backend, which I was
surprised at, but it appears to fix a lot of our issues around MVE
masked load/stores having to store a temporary address after an early
post-increment on a shared base address.
Differential Revision: https://reviews.llvm.org/D75847
Summary:
I've implemented them as target-specific IR intrinsics rather than
using `@llvm.experimental.vector.reduce.add`, on the grounds that the
'experimental' intrinsic doesn't currently have much code generation
benefit, and my replacements encapsulate the sign- or zero-extension
so that you don't expose the illegal MVE vector type (`<4 x i64>`) in
IR.
The machine instructions come in two versions: with and without an
input accumulator. My new IR intrinsics, like the 'experimental' one,
don't take an accumulator parameter: we represent that by just adding
on the input value using an ordinary i32 or i64 add. So if you write
the `vaddvaq` C-language intrinsic with an input accumulator of zero,
it can be optimised to VADDV, and conversely, if you write something
like `x += vaddvq(y)` then that can be combined into VADDVA.
Most of this is achieved in isel lowering, by converting these IR
intrinsics into the existing `ARMISD::VADDV` family of custom SDNode
types. For the difficult case (64-bit accumulators), isel lowering
already implements the optimization of folding an addition into a
VADDLV to make a VADDLVA; so once we've made a VADDLV, our job is
already done, except that I had to introduce a parallel set of ARMISD
nodes for the //predicated// forms of VADDLV.
For the simpler VADDV, we handle the predicated form by just leaving
the IR intrinsic alone and matching it in an ordinary dag pattern.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76491
Summary:
I've implemented these as target-specific IR intrinsics, because
they're not //quite// enough like @llvm.experimental.vector.reduce.min
(which doesn't take the extra scalar parameter). Also this keeps the
predicated and unpredicated versions looking similar, and the
floating-point minnm/maxnm versions fold into the same schema.
We had a couple of min/max reductions already implemented, from the
initial pathfinding exercise in D67158. Those were done by having
separate IR intrinsic names for the signed and unsigned integer
versions; as part of this commit, I've changed them to use a flag
parameter indicating signedness, which is how we ended up deciding
that the rest of the MVE intrinsics family ought to work. So now
hopefully the ewhole lot is consistent.
In the new llc test, the output code from the `v8f16` test functions
looks quite unpleasant, but most of it is PCS lowering (you can't pass
a `half` directly in or out of a function). In other circumstances,
where you do something else with your `half` in the same function, it
doesn't look nearly as nasty.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76490
Summary:
This patch implements the following intrinsics:
uint8x16_t __arm_vcx1q_u8 (int coproc, uint32_t imm);
T __arm_vcx1qa(int coproc, T acc, uint32_t imm);
T __arm_vcx2q(int coproc, T n, uint32_t imm);
uint8x16_t __arm_vcx2q_u8(int coproc, T n, uint32_t imm);
T __arm_vcx2qa(int coproc, T acc, U n, uint32_t imm);
T __arm_vcx3q(int coproc, T n, U m, uint32_t imm);
uint8x16_t __arm_vcx3q_u8(int coproc, T n, U m, uint32_t imm);
T __arm_vcx3qa(int coproc, T acc, U n, V m, uint32_t imm);
Most of them are polymorphic. Furthermore, some intrinsics are
polymorphic by 2 or 3 parameter types, such polymorphism is not
supported by the existing MVE/CDE tablegen backends, also we don't
really want to have a combinatorial explosion caused by 1000 different
combinations of 3 vector types. Because of this some intrinsics are
implemented as macros involving a cast of the polymorphic arguments to
uint8x16_t.
The IR intrinsics are even more restricted in terms of types: all MVE
vectors are cast to v16i8.
Reviewers: simon_tatham, MarkMurrayARM, dmgreen, ostannard
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76299
Summary:
This change implements ACLE CDE intrinsics that translate to
instructions working with general-purpose registers.
The specification is available at
https://static.docs.arm.com/101028/0010/ACLE_2019Q4_release-0010.pdf
Each ACLE intrinsic gets a corresponding LLVM IR intrinsic (because
they have distinct function prototypes). Dual-register operands are
represented as pairs of i32 values. Because of this the instruction
selection for these intrinsics cannot be represented as TableGen
patterns and requires custom C++ code.
Reviewers: simon_tatham, MarkMurrayARM, dmgreen, ostannard
Reviewed By: MarkMurrayARM
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76296
The MVE VDUP instruction take a GPR and splats into every lane of a
vector register. Unlike NEON we do not have a VDUPLANE equivalent
instruction, doing the same splat from a fp register. Previously a VDUP
to a v4f32/v8f16 would be represented as a (v4f32 VDUP f32), which
would mean the instruction pattern needs to add a COPY_TO_REGCLASS to
the GPR.
Instead this now converts that earlier during an ISel DAG combine,
converting (VDUP x) to (VDUP (bitcast x)). This can allow instruction
selection to tell that the input needs to be an i32, which in one of the
testcases allows it to use ldr (or specifically ldm) over (vldr;vmov).
Whilst being simple enough for floats, as the types sizes are the same,
these is no BITCAST equivalent for getting a half into a i32. This uses
a VMOVrh ARMISD node, which doesn't know the same tricks yet.
Differential Revision: https://reviews.llvm.org/D76292
Summary:
This is another set of instructions too complicated to be sensibly
expressed in IR by anything short of a target-specific intrinsic.
Given input vectors a,b, the instruction generates intermediate values
2*(a[0]*b[0]+a[1]+b[1]), 2*(a[2]*b[2]+a[3]+b[3]), etc; takes the high
half of each double-width values, and overwrites half the lanes in the
output vector c, which you therefore have to provide the input value
of. Optionally you can swap the elements of b so that the are things
like a[0]*b[1]+a[1]*b[0]; optionally you can round to nearest when
taking the high half; and optionally you can take the difference
rather than sum of the two products. Finally, saturation is applied
when converting back to a single-width vector lane.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76359
When optimising for code size at the expense of performance, it is often
worth saving and restoring some of r0-r3, if IPRA will be able to take
advantage of them. This doesn't cost any extra code size if we already
have a PUSH/POP pair, and increases the number of available registers
across any calls to the function.
We already have an optimisation which tries fold the subtract/add of the
SP into the PUSH/POP by using extra registers, which somewhat conflicts
with this. I've made the new optimisation less aggressive in cases where
the existing one is likely to trigger, which gives better results than
either of these optimisations by themselves.
Differential revision: https://reviews.llvm.org/D69936
Summary:
These are complicated integer multiply+add instructions with extra
saturation, taking the high half of a double-width product, and
optional rounding. There's no sensible way to represent that in
standard IR, so I've converted the clang builtins directly to
target-specific intrinsics.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76123
Summary:
These instructions compute multiply+add in integers, with one of the
operands being a splat of a scalar. (VMLA and VMLAS differ in whether
the splat operand is a multiplier or the addend.)
I've represented these in IR using existing standard IR operations for
the unpredicated forms. The predicated forms are done with target-
specific intrinsics, as usual.
When operating on n-bit vector lanes, only the bottom n bits of the
i32 scalar operand are used. So we have to tell that to isel lowering,
to allow it to remove a pointless sign- or zero-extension instruction
on that input register. That's done in `PerformIntrinsicCombine`, but
first I had to enable `PerformIntrinsicCombine` for MVE targets
(previously all the intrinsics it handled were for NEON), and make it
a method of `ARMTargetLowering` so that it can get at
`SimplifyDemandedBits`.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76122
ISD::ROTL/ROTR rotation values are guaranteed to act as a modulo amount, so for power-of-2 bitwidths we only need the lowest bits.
Differential Revision: https://reviews.llvm.org/D76201
The ASRL/LSRL long shifts are generated from 64bit shifts. Once we have
them, it might turn out that enough of the 64bit result was not required
that we can use a smaller shift to perform the same result. As the
smaller shift can in general be folded in more way, such as into add
instructions in one of the test cases here, we can use the demand bit
analysis to prefer the smaller shifts where we can.
Differential Revision: https://reviews.llvm.org/D75371
This changes the way that asrl and lsrl intrinsics are lowered, going
via a the ISEL ASRL and LSLL nodes instead of straight to machine nodes.
On top of that, it adds some constant folds for long shifts, in case it
turns out that the shift amount was either constant or 0.
Differential Revision: https://reviews.llvm.org/D75553
Summary:
This adds the ACLE intrinsic family for the VFMA and VFMS
instructions, which perform fused multiply-add on vectors of floats.
I've represented the unpredicated versions in IR using the cross-
platform `@llvm.fma` IR intrinsic. We already had isel rules to
convert one of those into a vector VFMA in the simplest possible way;
but we didn't have rules to detect a negated argument and turn it into
VFMS, or rules to detect a splat argument and turn it into one of the
two vector/scalar forms of the instruction. Now we have all of those.
The predicated form uses a target-specific intrinsic as usual, but
I've stuck to just one, for a predicated FMA. The subtraction and
splat versions are code-generated by passing an fneg or a splat as one
of its operands, the same way as the unpredicated version.
In arm_mve_defs.h, I've had to introduce a tiny extra piece of
infrastructure: a record `id` for use in codegen dags which implements
the identity function. (Just because you can't declare a Tablegen
value of type dag which is //only// a `$varname`: you have to wrap it
in something. Now I can write `(id $varname)` to get the same effect.)
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75998
David Green