The goal is support tail and mask policy in RVV builtins.
We focus on IR part first.
If the passthru operand is undef, we use tail agnostic, otherwise
use tail undisturbed.
Co-Authored-by: Hsiangkai Wang <Hsiangkai@gmail.com>
Reviewers: craig.topper, frasercrmck
Differential Revision: https://reviews.llvm.org/D117647
This patch introduces new intrinsics that enable the use of vsetvli in
contexts where only the returned vector length is of interest. The
pre-existing intrinsics are marked with side-effects, which prevents
even trivial optimizations on/across them.
These intrinsics are intended to be used in situations where the vector
length is fed in turn to RVV intrinsics or to vector-predication
intrinsics during loop vectorization, for example. Those codegen paths
ensure that instructions are generated with their own implicit vsetvli,
so the vector length and vtype can be relied upon to be correct.
No corresponding C builtins are planned at this stage, though that is a
possibility for the future if the need arises.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D117910
For vmsgeu.vi with 0, we know this is always true. So we can replace
it with vmset.m (unmasked) or vmset.m+vmand.mm (masked).
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D116584
Now the backend will select ~0 vl to a register and load instruction, we could use X0 to replace it.
Differential Revision: https://reviews.llvm.org/D116798
When we want to create an splat vector that only the first element is initialized, we could use vmv.s.x or vfmv.s.f to build it.
Differential Revision: https://reviews.llvm.org/D116277
This can be generalized to (srl (and X, C2), C) ->
(srli (slli X, (XLen-C3), (XLen-C3) + C). Where C2 is a mask with
C3 trailing ones.
This can avoid constant materialization for C2. This is beneficial
even when C2 can be selected to ANDI because the SLLI can become
C.SLLI, but C.ANDI cannot cover all the immediates of ANDI.
This also enables CSE in some cases of i8 sdiv by constant codegen.
Similar for (sra (sext_inreg X, i8), C).
With Zbb, sext_inreg of i8 and i16 are legal for sext.b and sext.h.
This transform makes the Zbb codegen the same as without Zbb. The
shifts are more compressible. This also exposes an opportunity for
CSE with another slli in the i16 sdiv by constant codegen.
These 3 switches map LMUL enum to instruction names. These follow
a regular pattern. Use a macro to reduce the number of source code
lines.
Reviewed By: arcbbb
Differential Revision: https://reviews.llvm.org/D116631
For large integers (for example, magic numbers generated by
TargetLowering::BuildSDIV when dividing by constant), we may
need about 4~8 instructions to build them.
In the same time, it just takes two instructions to load
constants (with extra cycles to access memory), so it may be
profitable to put these integers into constant pool.
Reviewed By: asb, craig.topper
Differential Revision: https://reviews.llvm.org/D114950
We already do this for splat nodes that carry a VL, but not for
splats that use VLMAX.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D115483
D113805 improved handling of i32 divu/remu on RV64. The basic idea
from that can be extended to (mul (and X, C2), C1) where C2 is any
mask constant.
We can replace the and with an SLLI by shifting by the number of
leading zeros in C2 if we also shift C1 left by XLen - lzcnt(C1)
bits. This will give the full product XLen additional trailing zeros,
putting the result in the output of MULHU. If we can't use ANDI,
ZEXT.H, or ZEXT.W, this will avoid materializing C2 in a register.
The downside is it make take 1 additional instruction to create C1.
But since that's not on the critical path, it can hopefully be
interleaved with other operations.
The previous tablegen pattern is replaced by custom isel code.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D115310
Add new hasVInstructions() which is currently equivalent.
Replace vector uses of hasStdExtZfh/F/D with new vector specific
versions. The vector spec no longer requires that the vectors implement the
same types as scalar. It only requires that the scalar type is
the maximum size the vectors can support. This is currently
implemented using the scalar rule we were using before.
Add new hasVInstructionsI64() begin using to qualify code that
requires i64 vector elements.
This is all NFC for now, but we can start using this to better
implement D112408 which introduces the Zve extensions.
Reviewed By: frasercrmck, eopXD
Differential Revision: https://reviews.llvm.org/D112496
Use SH1ADD/SH2ADD/SH3ADD along with LUI+ADDI to compose int32*3,
int32*5 and int32*9.
Reviewed By: craig.topper, luismarques
Differential Revision: https://reviews.llvm.org/D111484
Add the tail policy argument to LLVM IR intrinsics. There are two policies for tail elements. Tail agnostic means users do not care about the values in the tail elements and tail undisturbed means the values in the tail elements need to be kept after the operation. In order to let users control the tail policy, we add an additional argument at the end of the argument list.
For unmasked operations, we have no maskedoff and the tail policy is always tail agnostic. If users want to keep tail elements under unmasked operations, they could use all one mask in the masked operations to do it. So, we only add the additional argument for masked operations for most cases. There are exceptions listed below.
In this patch, we do not handle the following cases to reduce the complexity of the patch. There could be two separate patches for them.
* Use dest argument to control tail policy
vmerge.vvm/vmerge.vxm/vmerge.vim (add _t builtins with additional dest argument)
vfmerge.vfm (add _t builtins with additional dest argument)
vmv.v.v (add _t builtins with additional dest argument)
vmv.v.x (add _t builtins with additional dest argument)
vmv.v.i (add _t builtins with additional dest argument)
vfmv.v.f (add _t builtins with additional dest argument)
vadc.vvm/vadc.vxm/vadc.vim (add _t builtins with additional dest argument)
vsbc.vvm/vsbc.vxm (add _t builtins with additional dest argument)
* Always has tail argument for masked/unmasked intrinsics
Vector Single-Width Integer Multiply-Add Instructions (add _t and _mt builtins)
Vector Widening Integer Multiply-Add Instructions (add _t and _mt builtins)
Vector Single-Width Floating-Point Fused Multiply-Add Instructions (add _t and _mt builtins)
Vector Widening Floating-Point Fused Multiply-Add Instructions (add _t and _mt builtins)
Vector Reduction Operations (add _t and _mt builtins)
Vector Slideup Instructions (add _t and _mt builtins)
Vector Slidedown Instructions (add _t and _mt builtins)
Discussion: https://github.com/riscv/rvv-intrinsic-doc/pull/101
Differential Revision: https://reviews.llvm.org/D105092
Turn (and (shl x, c2), c1) -> (slli (srli x, c3-c2), c3) if c1 is a
shifted mask with no leading zeros and c3 trailing zeros where c3
is greater than c2.
Turn (and (shr x, c2), c1) -> (slli (srli x, c2+c3), c3) if c1 is a
shifted mask with c2 leading zeros and c3 trailing zeros.
When the leading zeros is C2+32 we can use SRLIW in place of SRLI.
SimplifyDemandedBits can turn srl into sra if the bits being shifted
in aren't demanded. This patch can recover the original sra in some cases.
I've renamed the tablegen class for detecting W users since the "overflowing operator"
term I originally borrowed from Operator.h does not include srl.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D109162
Soft deprecrate isNullValue/isAllOnesValue and update in tree
callers. This matches the changes to the APInt interface from
D109483.
Reviewed By: lattner
Differential Revision: https://reviews.llvm.org/D109535
If a sext_inreg is up for isel, and all its users are W instructions,
we can skip emitting the sext_inreg. This helpful if the producing
instruction can't become a W instruction.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D108966
X0 has special meaning for vsetvli, we need to make sure we never
create it a vsetvli that uses it by accident. This could happen
if the register coalescer coalesces a copy from X0 into this
instruction.
This patch splits the instruction so that we can have GPRNoX0
register class to use for the cases where we don't want the source
to be X0. The verifier won't let us explicitly use X0 on a GPRNoX0
operand so we need a separate pseudo for those cases.
I don't currently have a failing example for this. There was a
failure in D107957, but the coalescable copy from that example
should have been optimized away much earlier so I've fixed that.
This is not a complete fix. We still need to prevent the same
possible issue on the AVL operand of all of the vector instruction
pseudos. I don't want to make two versions of all of those so we
need to find a different solution for those. I have an idea I'm
going to try.
Differential Revision: https://reviews.llvm.org/D109110
Let the sext_inreg be selected to sext.w. Remove unneeded sext.w
during PostProcessISelDAG.
This gives opportunities for some other isel patterns to match
like the ADDIPair or matching mul with immediate to shXadd.
This becomes possible after D107658 started selecting W instructions
based on users. The sext.w will be considered a W user so isel
will often select a W instruction for the sext.w input and we can
just remove the sext.w. Otherwise we can combine the sext.w with
a ADD/SUB/MUL/SLLI to create a new W instruction in parallel
to the the original instruction.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D107708
DAGCombiner::visitStore can clear the upper bits of constants
used by stores. This leads prevents them from being recognized as
sign extended negative values making them more expensive to
materialize.
This patch uses the hasAllNBitUsers method from D107658 to make
a negative constant if none of the users care about the upper bits.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D108052
We normally select these when the root node is a sext_inreg, but
SimplifyDemandedBits can sometimes bypass the sext_inreg for some
users. This can create situation where sext_inreg+add/sub/mul/shl
is selected to a W instruction, and then the add/sub/mul/shl is
separately selected to a non-W instruction with the same inputs.
This patch tries to detect when it would still be ok to use a W
instruction without the sext_inreg by checking the direct users.
This can allow the W instruction to CSE with one created for a
sext_inreg+add/sub/mul/shl. To minimize complexity and cost of
checking, we make no attempt to determine if the CSE will happen
and just always use a W instruction when we can.
Differential Revision: https://reviews.llvm.org/D107658
Replace some existing isel patterns that are covered by the new
code. SLLIUWPat has been removed in favor of folding its root case
into the new code. The other uses in isel patterns for shXadd.uw
have been switched to using hardcoded AND masks.
This is based on the original version of D49585 from ARM. The final
version of that was made a DAG combine, but I've chosen to keep it
as custom isel. I'm not convinced DAG combine is as good with
shift pairs as it is with and+shift. I saw some issues optimizing
the shifts created by vscale lowering if an and isn't created for
from a shift pair.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D106230
If the upper 32 bits are zero and bit 31 is set, we might be able to
use zext.w to fill in the zeros after using an lui and/or addi.
Most of this patch is plumbing the subtarget features into the constant
materialization.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D105509
This helps us select W instructions in more cases. Most of the
affected tests have had the sign_extend_inreg or AND folded into
sextload/zextload.
Differential Revision: https://reviews.llvm.org/D104079
This uses 3 bits of data instead of 7. I'm wondering if we can use
bitfields for the lookup table key where this would matter.
I also name the shift_amount template to log2 since it is used
with more than just an srl now.
All that really matters is that the VLMAX of the preceding
instructions is the same as the VLMAX required by the mask
operation.
Also update the vmsge(u) handling to use the SEW/LMUL we use for
other mask register operations. We were matching it to the compare
before. Some cases will be improve if we fix masked compares to
use tail agnostic policy. I think they ignore the tail policy
anyway.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D103299
SEW=64 shifts only uses the log2(64) bits of shift amount. If we're
splatting a 64 bit value in 2 parts, we can avoid splatting the
upper bits and just let the low bits be sign extended. They won't
be read anyway.
For the purposes of SelectionDAG semantics of the generic ISD opcodes,
if hi was non-zero or bit 31 of the low is 1, the shift was already
undefined so it should be ok to replace high with sign extend of low.
In order do be able to find the split i64 value before it becomes
a stack operation, I added a new ISD opcode that will be expanded
to the stack spill in PreprocessISelDAG. This new node is conceptually
similar to BuildPairF64, but I expanded earlier so that we could
go through regular isel to get the right VLSE opcode for the LMUL.
BuildPairF64 is expanded in a CustomInserter.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D102521
The ComplexPattern is looking for an immediate in a certain range
that has a single use. This can be handled with a PatLeaf since
we aren't matching multiple patterns or checking any complicated
relationships between nodes.
This shrinks the isel table a little bit since tablegen no longer
has to generate patterns with commuted operands. With the PatLeaf,
tablegen can see we're matching an immediate which should always
be on the right hand side of add.
Reviewed By: benshi001
Differential Revision: https://reviews.llvm.org/D102510
The VSEW encoding isn't a useful value to pass around. It's better
to use SEW or log2(SEW) directly. The only real ugliness is that
the vsetvli IR intrinsics use the VSEW encoding, but it's easy
enough to decode that when the intrinsic is processed.
This shrinks the immediate that isel table needs to emit for these
instructions. Hoping this allows me to change OPC_EmitInteger to
use a better variable length encoding for representing negative
numbers. Similar to what was done a few months ago for OPC_CheckInteger.
The alternative encoding uses less bytes for negative numbers, but
increases the number of bytes need to encode 64 which was a very
common number in the RISCV table due to SEW=64. By using Log2 this
becomes 6 and is no longer a problem.
This adds a special operand type that is allowed to be either
an immediate or register. By giving it a unique operand type the
machine verifier will ignore it.
This perturbs a lot of tests but mostly it is just slightly different
instruction orders. Something bad did happen to some min/max reduction
tests. We're spilling vector registers when we weren't before.
Reviewed By: khchen
Differential Revision: https://reviews.llvm.org/D101246
This modifies my previous patch to push the strided load formation
to isel. This gives us opportunity to fold the splat into a .vx
operation first. Using a scalar register and a .vx operation reduces
vector register pressure which can be important for larger LMULs.
If we can't fold the splat into a .vx operation, then it can make
sense to use a strided load to free up the vector arithmetic
ALU to do actual arithmetic rather than tying it up with vmv.v.x.
Reviewed By: khchen
Differential Revision: https://reviews.llvm.org/D101138
We can have RISCVISelDAGToDAG.cpp call the VT only version by
finding the RISCVTargetLowering object via the Subtarget.
Make the static versions just global static functions in
RISCVISelLowering that can be called by static functions in that
file.
These instructions don't really exist, but we have ways we can
emulate them.
.vv will swap operands and use vmsle().vv. .vi will adjust the
immediate and use .vmsgt(u).vi when possible. For .vx we need to
use some of the multiple instruction sequences from the V extension
spec.
For unmasked vmsge(u).vx we use:
vmslt{u}.vx vd, va, x; vmnand.mm vd, vd, vd
For cases where mask and maskedoff are the same value then we have
vmsge{u}.vx v0, va, x, v0.t which is the vd==v0 case that
requires a temporary so we use:
vmslt{u}.vx vt, va, x; vmandnot.mm vd, vd, vt
For other masked cases we use this sequence:
vmslt{u}.vx vd, va, x, v0.t; vmxor.mm vd, vd, v0
We trust that register allocation will prevent vd in vmslt{u}.vx
from being v0 since v0 is still needed by the vmxor.
Differential Revision: https://reviews.llvm.org/D100925
This patch adds more optimized codegen for the above SETCC forms,
by matching the '.vi' vector forms when the immediate is a 5-bit signed
immediate plus 1. The immediate can be decremented and the corresponding
SET[U]LE or SET[U]GT forms can be matched.
This work was left as a TODO from D94168.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D100096
Many of the operands are handled the same or in the same order
for all these intrinsics. Factor out the code for selecting and
pushing them into the Operands vector.
Differential Revision: https://reviews.llvm.org/D99923
I missed a few intrinsics in 3dd4aa7d09
when I did this for masked loads and masked segment loads/stores.
Found while trying to share more code between these custom isel
functions.
This adds a new integer materialization strategy mainly targeted
at 64-bit constants like 0xffffffff where there are 32 or more trailing
ones with leading zeros. We can materialize these by using an addi -1
and srli to restore the leading zeros. This matches what gcc does.
I haven't limited to just these cases though. The implementation
here takes the constant, shifts out all the leading zeros and
shifts ones into the LSBs, creates the new sequence, adds an srli,
and checks if this is shorter than our original strategy.
I've separated the recursive portion into a standalone function
so I could append the new strategy outside of the recursion. Since
external users are no longer using the recursive function, I've
cleaned up the external interface to return the sequence instead of
taking a vector by reference.
Reviewed By: asb
Differential Revision: https://reviews.llvm.org/D98821
This matches what we do in our isel patterns. In our internal
testing we've found this is needed to make the fast register
allocator happy at -O0. Otherwise it may assign V0 to an earlier
operand and find itself with no registers left when it reaches
the mask operand. By using V0 explicitly, the fast register allocator
will see it when it checks for phys register usages before it
starts allocating vregs. I'll try to update this with a test case.
Unfortunately, this does appear to prevent some instruction reordering
by the pre-RA scheduler which leads to the increased spills seen in
some tests. I suspect that problem could already occur for other
instructions that already used V0 directly.
There's a lot of repeated code here that could do with some
wrapper functions. Not sure if that should be at the level of the
new code that deals with V0. That would require multiple output
parameters to pass the glue, chain and register back. Maybe it
should be at a higher level over the entire set of push_backs.
Reviewed By: frasercrmck, HsiangKai
Differential Revision: https://reviews.llvm.org/D99367
We look for this pattern frequently in isel patterns so its a
good idea to try to preserve it.
This also let's us remove our special isel handling for srliw
and use a direct pattern match of (srl (and X, 0xffffffff), C)
since no bits will be removed from the and mask.
Differential Revision: https://reviews.llvm.org/D99042
Previously we used selectImm for RV64 and isel patterns for
RV32. This should be NFC, but will allow RV32 and RV64 to share
improvements in the future. For example, it might be useful to
use BSETI from Zbs to make single bit constants.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D98877
This patch addresses a compiler crash resulting from passing a
fixed-length type to one that expects scalable vector types. An
assertion was added to prevent this regressing in the future.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D97868
This patch unifies the two disparate paths for lowering INSERT_SUBVECTOR
operations under one roof. Consequently, with this patch it is possible to
support any fixed-length subvector insertion, not just "cast-like" ones.
As before, support for the insertion of mask vectors will come in a
separate patch.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D97543
This will allow FrameIndex as the base address instead of
emitting a separate ADDI from isel. eliminateFrameIndex will likely turn
it back into an ADDI, but this makes things consistent with the
SDPatterns and VLPatterns.
I only tested one case for simplicity. I can test more if reviewers
want.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D97221
This patch unifies the two disparate paths for lowering
EXTRACT_SUBVECTOR operations under one roof. Consequently, with this
patch it is possible to support any fixed-length subvector extraction,
not just "cast-like" ones.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D97192
We just started using a ComplexPattern for sexti32. This updates
zexti32 to match.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D97231
This patch extends the support for RVV INSERT_SUBVECTOR to cover those
which don't align to a vector register boundary. Like the support for
EXTRACT_SUBVECTOR in D96959, it accomplishes this by extracting the
nearest register-sized subvector (a subregister operation), then sliding
the vector down with VSLIDEDOWN, inserting the subvector to the first
position, and sliding the vector back up again afterwards.
Unlike subvector extraction, for vectors that occupy less than a full
vector register we must preserve the untouched elements. We do this by
lowering to an LMUL=1 INSERT_SUBVECTOR using the above method and
lowering that to a VSLIDEUP with a zero offset. This uses a
tail-undisturbed policy and so has the effect of "sliding in" the
subvector elements while preserving the surrounding ones.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D96972
An i64 AssertZExt from a type smaller than i32 has at least 33
leading zeros which mean it has at least 33 sign bits.
Since we have a couple patterns that use two sexti32, I've
switched to a ComplexPattern so tablegen didn't have to generate
9 different permutations.
As noted in the FIXME, maybe we should just call computeNumSignBits,
but we don't have tests that benefit from that yet.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D97130
This should fix the issue reported in D96972.
I don't have a good test case for this without those changes.
Differential Revision: https://reviews.llvm.org/D97082
A previous patch moved the index versions. This moves the rest.
I also removed the custom lowering for VLEFF since we can now
do everything directly in the isel handling.
I had to update getLMUL to handle mask registers to index the
pseudo table correctly for VLE1/VSE1.
This is good for another 15K reduction in llc size.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D97097
This patch extends the support for RVV EXTRACT_SUBVECTOR to cover those
which don't align to a vector register boundary. It accomplishes this by
extracting the nearest register-sized subvector (a subregister
operation), then sliding the vector down with VSLIDEDOWN and extracting
the subvector from the first position (a COPY operation).
Since this procedure involves the use of VSCALE and multiplication, the
handling of such operations is done during lowering to simplify the
implementation and make use of DAG combining. This necessitated moving
some helper functions from RISCVISelDAGToDAG to RISCVTargetLowering.
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D96959
We don't currently create memory operands for these intrinsics,
but there was a suggestion of using the indexed load/store
intrinsics to implement isel for scalable vector gather/scatter.
That may propagate the memory operand from the gather/scatter
ISD nodes.
We had more combinations of data and index lmuls than we needed.
Also add some asserts to verify that the IndexVT and data VT have
the same element count when we isel these pseudo instructions.
There are many legal combinations of index and data VTs supported
for these intrinsics. This results in a lot of isel patterns in
RISCVGenDAGISel.inc.
By adding a separate table similar to what we use for segment
load/stores, we can more efficiently manually select these
intrinsics. We should also be able to reuse this table scalable
vector gather/scatter.
This reduces the llc binary size by ~56K.
Reviewed By: khchen
Differential Revision: https://reviews.llvm.org/D97033
Just like we do for isel patterns, we need to call selectVLOp
to prevent 0 from being selected to X0 by the default isel.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D97021
Intrinsic ID is a 32-bit value which made each row of the table 4
byte aligned. The remaining fields used 5 bytes. This meant 3 bytes
of padding per row.
This patch breaks the table into 4 separate tables and indexes them
by properties we know about the intrinsic. NF, masked,
strided, ordered, etc. The indexed load/store tables have no
padding in their rows now.
All together this reduces the size of llc binary by ~28K.
I'm considering adding similar tables for isel of non-segment
load/store as well to cut down the size of the isel table and
probably improve our isel performance. Those tables would need to
indexed from intrinsics, IR loads/stores, gathers/scatters, and
RISCVISD opcodes. So having a table that can be indexed without using
intrinsic ID is more flexible.
Reviewed By: HsiangKai
Differential Revision: https://reviews.llvm.org/D96894
These should be NOPs so we can just replace with the input. This
matches what SVE does with isel patterns for all permutations.
Custom isel saves us from having to list all permurations for
all LMULs.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D96921
This patch adds support for INSERT_SUBVECTOR and EXTRACT_SUBVECTOR
(nominally where both operands are scalable vector types) where the
vector, subvector, and index align sufficiently to allow decomposition
to subregister manipulation:
* For extracts, the extracted subvector must correctly align with the
lower elements of a vector register.
* For inserts, the inserted subvector must be at least one full vector
register, and correctly align as above.
This approach should work for fixed-length vector insertion/extraction
too, but that will come later.
Reviewed By: craig.topper, khchen, arcbbb
Differential Revision: https://reviews.llvm.org/D96873
A lot of the code for the masked and unmasked is the same. This
patch adds a boolean to handle the differences so we can share
the code.
Differential Revision: https://reviews.llvm.org/D96841
This is annoying because the condition code legalization belongs
to LegalizeDAG, but our custom handler runs in Legalize vector ops
which occurs earlier.
This adds some of the mask binary operations so that we can combine
multiple compares that we need for expansion.
I've also fixed up RISCVISelDAGToDAG.cpp to handle copies of masks.
This patch contains a subset of the integer setcc patch as well.
That patch is dependent on the integer binary ops patch. I'll rebase
based on what order the patches go in.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D96567
Unlike scalable vectors, I'm only using a ComplexPattern for
the immediate itself. The vmv_v_x is matched explicitly. We igore
the VL argument when matching a binary operator, but we do check
it when matching splat directly.
I left out tests for vXi64 as they fail on rv32 right now.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D96365
This patch handles cast-like insert_subvector & extract_subvector
in which case:
1. index starts from 0.
2. inserting a fixed-width vector into a scalable vector,
or extracting a fixed-width vector from a scalable vector.
Reviewed By: craig.topper, frasercrmck
Differential Revision: https://reviews.llvm.org/D96352
As of the current draft these are no longer being considered
for the bitmanip spec. It wasn't clear what sub extension they
belonged in in the 0.93 spec.
So remove them. They can always be added back if something changes.
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D96157
Rather than materializing the 0xffff immediate for the AND, use
a shift left to remove the upper bits and then shift in zeros
from the right.
This pattern occurs when type legalizing an i16 right shift.
I've implemented this with custom selection code for a number of
reasons. I've limited this to the AND having a single use. We need
to compensate for SimplifyDemandedBits altering the AND mask. I'm
using *W opcodes on RV64. We may want to generlize this in the
future. For all these reason it seemed easiest to do it this way.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D95774
In d2927f786e, I added patterns
to remove (and X, 31) from sllw/srlw/sraw shift amounts.
There is code in SelectionDAGISel.cpp that knows to use
computeKnownBits to fill in bits of the mask that were removed
by SimplifyDemandedBits based on bits being known zero.
The non-W shift patterns use immbottomxlenset which allows the
mask to have more than log2(xlen) trailing ones, but doesn't
have a call to computeKnownBits to fill in bits of the mask that may
have been cleared by SimplifyDemandedBits.
This patch copies code from X86 to handle more than log2(xlen)
bottom bits set and uses computeKnownBits to fill in missing bits
before counting.
Reviewed By: luismarques
Differential Revision: https://reviews.llvm.org/D95422
-Remove the ISD opcode for READ_VL. Just emit the MachineSDNode directly.
-Move segmented fault first only load intrinsic handling completely to
RISCVISelDAGToDAG.cpp and emit the ReadVL MachineSDNode there
instead of lowering to ISD opcodes first.
This avoids being dependent on SimplifyDemandedBits having cleared
those bits.
It could make sense to teach SimplifyDemandedBits to keep all
lower bits 1 in an AND mask when possible. This could be
implemented with slli+srli in the general case rather than
needing to materialize the constant.
For Zvlsseg, we need continuous vector registers for the values. We need
to define new register classes for the different combinations of (number
of fields and LMUL). For example,
when the number of fields(NF) = 3, LMUL = 2, the values will be assigned
to (V0M2, V2M2, V4M2), (V2M2, V4M2, V6M2), (V4M2, V6M2, V8M2), ...
We define the vlseg intrinsics with multiple outputs. There is no way to
describe the codegen patterns with multiple outputs in the tablegen
files. We do the codegen in RISCVISelDAGToDAG and use EXTRACT_SUBREG to
extract the values of output.
The multiple scalable vector values will be put into a struct. This
patch is depended on the support for scalable vector struct.
Differential Revision: https://reviews.llvm.org/D94229
MCTargetDesc includes headers from Utils and Utils includes headers
from MCTargetDesc. So from a library layering perspective it makes sense
for them to be in the same library. I guess the other option might be to
move the tablegen includes from RISCVMCTargetDesc.h to RISCVBaseInfo.h
so that RISCVBaseInfo.h didn't need to include RISCVMCTargetDesc.h.
Everything else that depends on Utils also depends on MCTargetDesc so
having one library seemed simpler.
Differential Revision: https://reviews.llvm.org/D93168
ComplexPatterns are kind of weird, they don't call any of the predicates on their operands. And their "complexity" used for tablegen ordering purposes in the matcher table is hand specified.
This started as an attempt to just use sext_inreg + SLOIPat to implement SLOIW just to have one less Select function. The matching for the or+shl is the same as long as you know the immediate is less than 32 for SLOIW. But that didn't work out because using uimm5 with SLOIPat didn't do anything if it was a ComplexPattern.
I realized I could just use a PatFrag with the opcodes I wanted to match and an immediate predicate would then evaluate correctly. This also computes the complexity just like any other pattern does. Then I just needed to check the constraints on the immediates in the predicate. Conveniently the predicate is evaluated after the fragment has been matched. So the structure has already been checked, we just need to find the constants.
I'll note that this is unusual, I didn't find any other targets looking through operands in PatFrag predicate. There is a PredicateCodeUsesOperands feature that can be used to collect the operands into an array that is used by AMDGPU/VOP3Instructions.td. I believe that feature exists to handle commuted matching, but since the nodes here use constants, they aren't ever commuted
Differential Revision: https://reviews.llvm.org/D91901
This patch extends the pattern-matching capability of vector-splatted
constants. When illegally-typed constants are legalized they are
canonically sign-extended to XLenVT. This preserves the sign and allows
us to match simm5. If they were zero-extended for whatever reason we'd
lose that ability: e.g. `(i8 -1) -> (XLenVT 255)` would not be matched
under the current logic.
To address this we first manually sign-extend the splatted constant from
the vector element type to int64_t. This preserves the semantics while
removing any implicitly-truncated bits.
The corresponding logic for uimm5 was not updated, the rationale being
that neither sign- nor zero-extending a legal uimm5 immediate should
change that (unless we expect actual "garbage" upper bits).
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D93837
This patch extends the SDNode ISel support for RVV from only the
vector/vector instructions to include the vector/scalar and
vector/immediate forms.
It uses splat_vector to carry the scalar in each case, except when
XLEN<SEW (RV32 SEW=64) when a custom node `SPLAT_VECTOR_I64` is used for
type-legalization and to encode the fact that the value is sign-extended
to SEW. When the scalar is a full 64-bit value we use a sequence to
materialize the constant into the vector register.
The non-intrinsic ISel patterns have also been split into their own
file.
Authored-by: Roger Ferrer Ibanez <rofirrim@gmail.com>
Co-Authored-by: Fraser Cormack <fraser@codeplay.com>
Reviewed By: craig.topper
Differential Revision: https://reviews.llvm.org/D93312
This patch adds two IR intrinsics for vsetvli instruction. One to set the vector length to a user specified value and one to set it to vlmax. The vlmax uses the X0 source register encoding.
Clang builtins will follow in a separate patch
Differential Revision: https://reviews.llvm.org/D92973
This node returns 2 results and uses a chain. As long as we use a DAG as part of the pseudo instruction definition where we can use the "set" operator, it looks like tablegen can handle use a pattern for this without a problem. I believe the original implementation was copied from PowerPC.
This also fixes the pseudo instruction so that it is marked as having side effects to match the definition of CSRRS and the RV64 instruction. And we don't need to explicitly clear mayLoad/mayStore since those can be inferred now.
Differential Revision: https://reviews.llvm.org/D92786
This should result in better utilization of RORIW since we
don't need to look for a SIGN_EXTEND_INREG that may not exist.
Also remove rotl/rotr isel matching to GREVI and just prefer RORI.
This is to keep consistency so we don't have to match ROLW/RORW
to GREVIW as well. I imagine RORI/RORIW performance will be the
same or better than GREVI.
Differential Revision: https://reviews.llvm.org/D91449
We need to make sure the upper 32 bits are all ones to ensure the result is properly sign extended. Previously we only checked the lower 32 bits of the mask. I've also added a check that the shift amount is less than 32. Without that the original code asserts inside maskLeadingOnes if the SROI check is removed or the SROIW pattern is checked first. I've refactored the code to use early outs to reduce nesting.
I've also updated SLOIW matching with the same changes, but I couldn't find a broken test case with the existing code.
Differential Revision: https://reviews.llvm.org/D90961
We need to ensure the upper 32 bits of the mask are zero.
So that the srl shifts zeroes into the lower 32 bits.
Differential Revision: https://reviews.llvm.org/D90585
We don't need custom matching, we just a need a predicate to check
the immediate is greater than 32. We can use the existing ImmSub32
to adjust the immediate.
I've also used the new predicate in the other location that used
ImmSub32. I tried to create a test case where we would break without
the greater than 32 check on that pattern, but DAG combine defeated me.
Still seemed safer to have it.
Differential Revision: https://reviews.llvm.org/D90546
DAGCombine doesn't canonicalize rotl/rotr with immediate so we
need patterns for both.
Remove the custom matcher for rotl to RORI and just use a SDNodeXForm
to convert the immediate instead. Doing this gives priority to the
rev32/rev16 versions of grevi over rori since an explicit immediate
is more precise than any immediate. I also added rotr patterns for
rev32/rev16. And removed the (or (shl), (shr)) patterns that should be
combined to rotl by DAG combine.
There is at least one other grev pattern that probably needs a
another rotr pattern, but we need more test coverage first.
Differential Revision: https://reviews.llvm.org/D90575
The code is looking for (sext_inreg (or (shl X, C2), (shr (and Y, C3), C1))).
We need to ensure X and Y are the same.
Differential Revision: https://reviews.llvm.org/D90580
In SelectionDAGBuilder always translate the fshl and fshr intrinsics to
FSHL and FSHR (or ROTL and ROTR) instead of lowering them to shifts and
ORs. Improve the legalization of FSHL and FSHR to avoid code quality
regressions.
Differential Revision: https://reviews.llvm.org/D77152
This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the ternary subset (zbt subextension) of the
experimental B extension of RISC-V.
It adds also the correspondent codegen tests.
This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf
Differential Revision: https://reviews.llvm.org/D79875
This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions belonging to both the permutation and the base
subsets of the experimental B extension of RISC-V.
It adds also the correspondent codegen tests.
This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf
Differential Revision: https://reviews.llvm.org/D79873
This patch provides optimization of bit manipulation operations by
enabling the +experimental-b target feature.
It adds matching of single block patterns of instructions to specific
bit-manip instructions from the base subset (zbb subextension) of the
experimental B extension of RISC-V.
It adds also the correspondent codegen tests.
This patch is based on Claire Wolf's proposal for the bit manipulation
extension of RISCV:
https://github.com/riscv/riscv-bitmanip/blob/master/bitmanip-0.92.pdf
Differential Revision: https://reviews.llvm.org/D79870
For an addition with an immediate in specific ranges, a pair of
addi-addi can be generated instead of the ordinary lui-addi-add serial.
Reviewed By: MaskRay, luismarques
Differential Revision: https://reviews.llvm.org/D82262
We can often fold an ADDI into the offset of load/store instructions:
(load (addi base, off1), off2) -> (load base, off1+off2)
(store val, (addi base, off1), off2) -> (store val, base, off1+off2)
This is possible when the off1+off2 continues to fit the 12-bit immediate.
We remove the previous restriction where we would never fold the ADDIs if
the load/stores had nonzero offsets. We now do the fold the the resulting
constant still fits a 12-bit immediate, or if off1 is a variable's address
and we know based on that variable's alignment that off1+offs2 won't overflow.
Differential Revision: https://reviews.llvm.org/D79690
Summary:
RISC-V uses a post-select peephole pass to optimise
`(load/store (ADDI $reg, %lo(addr)), 0)` into `(load/store $reg, %lo(addr))`.
This peephole wasn't firing for accesses to constant pools, which is how we
materialise most floating point constants.
This adds support for the constantpool case, which improves code generation for
lots of small FP loading examples. I have not added any tests because this
structure is well-covered by the `fp-imm.ll` testcases, as well as almost
all other uses of floating point constants in the RISC-V backend tests.
Reviewed By: luismarques, asb
Differential Revision: https://reviews.llvm.org/D79523
Summary:
RISC-V uses a post-select peephole pass to optimise
`(load/store (ADDI $reg, %lo(addr)), 0)` into `(load/store $reg, %lo(addr))`.
This peephole wasn't firing for accesses to constant pools, which is how we
materialise most floating point constants.
This adds support for the constantpool case, which improves code generation for
lots of small FP loading examples. I have not added any tests because this
structure is well-covered by the `fp-imm.ll` testcases, as well as almost
all other uses of floating point constants in the RISC-V backend tests.
Reviewed By: luismarques, asb
Differential Revision: https://reviews.llvm.org/D79523
For the downstream RISCV maintenance, it would be easier to inherent
RISCVISelDAGToDAG by including header and only override the method that needs
to be customized for the provider non-standard ISA extension without touching
RISCVISelDAGToDAG.cpp which may cause conflict when upgrading the downstream
LLVM version.
Differential Revision: https://reviews.llvm.org/D77117
This allows us to delete InlineAsm::Constraint_i workarounds in
SelectionDAGISel::SelectInlineAsmMemoryOperand overrides and
TargetLowering::getInlineAsmMemConstraint overrides.
They were introduced to X86 in r237517 to prevent crashes for
constraints like "=*imr". They were later copied to other targets.
This allows arguments with the constraint A to be lowered to input nodes
for RISC-V, which implies a memory address stored in a register.
This patch adds the minimal amount of code required to get operands with
the right constraints to compile.
https://reviews.llvm.org/D54296
llvm-svn: 369095
On RISC-V, the `cycle` CSR holds a 64-bit count of the number of clock
cycles executed by the core, from an arbitrary point in the past. This
matches the intended semantics of `@llvm.readcyclecounter()`, which we
currently leave to the default lowering (to the constant 0).
With this patch, we will now correctly lower this intrinsic to the
intended semantics, using the user-space instruction `rdcycle`. On
64-bit targets, we can directly lower to this instruction.
On 32-bit targets, we need to do more, as `rdcycle` only returns the low
32-bits of the `cycle` CSR. In this case, we perform a custom lowering,
based on the PowerPC lowering, using `rdcycleh` to obtain the high
32-bits of the `cycle` CSR. This custom lowering inserts a new basic
block which detects overflow in the high 32-bits of the `cycle` CSR
during reading (because multiple instructions are required to read). The
emitted assembly matches the suggested assembly in the RISC-V
specification.
Differential Revision: https://reviews.llvm.org/D64125
llvm-svn: 365201
The break isn't strictly needed yet as there is no subsequent entry in the
case. But adding to prevent mistakes further down the road.
llvm-svn: 351785
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
As discussed in the RFC
<http://lists.llvm.org/pipermail/llvm-dev/2018-October/126690.html>, 64-bit
RISC-V has i64 as the only legal integer type. This patch introduces patterns
to support codegen of the new instructions
introduced in RV64I: addiw, addiw, subw, sllw, slliw, srlw, srliw, sraw,
sraiw, ld, sd.
Custom selection code is needed for srliw as SimplifyDemandedBits will remove
lower bits from the mask, meaning the obvious pattern won't work:
def : Pat<(sext_inreg (srl (and GPR:$rs1, 0xffffffff), uimm5:$shamt), i32),
(SRLIW GPR:$rs1, uimm5:$shamt)>;
This is sufficient to compile and execute all of the GCC torture suite for
RV64I other than those files using frameaddr or returnaddr intrinsics
(LegalizeDAG doesn't know how to promote the operands - a future patch
addresses this).
When promoting i32 sltu/sltiu operands, it would be more efficient to use
sign-extension rather than zero-extension for RV64. A future patch adds a hook
to allow this.
Differential Revision: https://reviews.llvm.org/D52977
llvm-svn: 347973
This commit introduces support for materialising 64-bit constants for RV64I,
making use of the RISCVMatInt::generateInstSeq helper in order to share logic
for immediate materialisation with the MC layer (where it's used for the li
pseudoinstruction).
test/CodeGen/RISCV/imm.ll is updated to test RV64, and gains new 64-bit
constant tests. It would be preferable if anyext constant returns were sign
rather than zero extended (see PR39092). This patch simply adds an explicit
signext to the returns in imm.ll.
Further optimisations for constant materialisation are possible, most notably
for mask-like values which can be generated my loading -1 and shifting right.
A future patch will standardise on the C++ codepath for immediate selection on
RV32 as well as RV64, and then add further such optimisations to
RISCVMatInt::generateInstSeq in order to benefit both RV32 and RV64 for
codegen and li expansion.
Differential Revision: https://reviews.llvm.org/D52962
llvm-svn: 347042
r343712 performed this optimisation during instruction selection. As Eli
Friedman pointed out in post-commit review, implementing this as a DAGCombine
might allow opportunities for further optimisations.
llvm-svn: 343741
Although we can't write a tablegen pattern to remove redundant
splitf64+buildf64 pairs due to the multiple return values, we can handle it
with some C++ selection code. This is simpler than removing them after
instruction selection through RISCVDAGToDAGISel::PostprocessISelDAG, as was
done previously.
llvm-svn: 343712
Summary:
In r333455 we added a peephole to fix the corner cases that result
from separating base + offset lowering of global address.The
peephole didn't handle some of the cases because it only has a basic
block view instead of a function level view.
This patch replaces that logic with a machine function pass. In
addition to handling the original cases it handles uses of the global
address across blocks in function and folding an offset from LW\SW
instruction. This pass won't run for OptNone compilation, so there
will be a negative impact overall vs the old approach at O0.
Reviewers: asb, apazos, mgrang
Reviewed By: asb
Subscribers: MartinMosbeck, brucehoult, the_o, rogfer01, mgorny, rbar, johnrusso, simoncook, niosHD, kito-cheng, shiva0217, zzheng, llvm-commits, edward-jones
Differential Revision: https://reviews.llvm.org/D47857
llvm-svn: 335786
Summary:
Base and offset are always separated when a GlobalAddress node is lowered
(rL332641) as an optimization to reduce instruction count. However, this
optimization is not profitable if the Global Address ends up being used in only
instruction.
This patch adds peephole optimizations that merge an offset of
an address calculation into the LUI %%hi and ADD %lo of the lowering sequence.
The peephole handles three patterns:
1) ADDI (ADDI (LUI %hi(global)) %lo(global)), offset
--->
ADDI (LUI %hi(global + offset)) %lo(global + offset).
This generates:
lui a0, hi (global + offset)
add a0, a0, lo (global + offset)
Instead of
lui a0, hi (global)
addi a0, hi (global)
addi a0, offset
This pattern is for cases when the offset is small enough to fit in the
immediate filed of ADDI (less than 12 bits).
2) ADD ((ADDI (LUI %hi(global)) %lo(global)), (LUI hi_offset))
--->
offset = hi_offset << 12
ADDI (LUI %hi(global + offset)) %lo(global + offset)
Which generates the ASM:
lui a0, hi(global + offset)
addi a0, lo(global + offset)
Instead of:
lui a0, hi(global)
addi a0, lo(global)
lui a1, (offset)
add a0, a0, a1
This pattern is for cases when the offset doesn't fit in an immediate field
of ADDI but the lower 12 bits are all zeros.
3) ADD ((ADDI (LUI %hi(global)) %lo(global)), (ADDI lo_offset, (LUI hi_offset)))
--->
offset = global + offhi20<<12 + offlo12
ADDI (LUI %hi(global + offset)) %lo(global + offset)
Which generates the ASM:
lui a1, %hi(global + offset)
addi a1, %lo(global + offset)
Instead of:
lui a0, hi(global)
addi a0, lo(global)
lui a1, (offhi20)
addi a1, (offlo12)
add a0, a0, a1
This pattern is for cases when the offset doesn't fit in an immediate field
of ADDI and both the lower 1 bits and high 20 bits are non zero.
Reviewers: asb
Reviewed By: asb
Subscribers: rbar, johnrusso, simoncook, jordy.potman.lists, apazos,
niosHD, kito-cheng, shiva0217, zzheng, edward-jones, mgrang
llvm-svn: 333455
The DEBUG() macro is very generic so it might clash with other projects.
The renaming was done as follows:
- git grep -l 'DEBUG' | xargs sed -i 's/\bDEBUG\s\?(/LLVM_DEBUG(/g'
- git diff -U0 master | ../clang/tools/clang-format/clang-format-diff.py -i -p1 -style LLVM
- Manual change to APInt
- Manually chage DOCS as regex doesn't match it.
In the transition period the DEBUG() macro is still present and aliased
to the LLVM_DEBUG() one.
Differential Revision: https://reviews.llvm.org/D43624
llvm-svn: 332240
Inspired by r331508, I did a grep and found these.
Mostly just change from dyn_cast to cast. Some cases also showed a dyn_cast result being converted to bool, so those I changed to isa.
llvm-svn: 331577
Reverts rL330224, while issues with the C extension and missed common
subexpression elimination opportunities are addressed. Neither of these issues
are visible in current RISC-V backend unit tests, which clearly need
expanding.
llvm-svn: 330281
The implementation follows the MIPS backend and expands the
pseudo instruction directly during asm parsing. As the result, only
real MC instructions are emitted to the MCStreamer. Additionally,
PseudoLI instructions are emitted during codegen. The actual
expansion to real instructions is performed during MI to MC lowering
and is similar to the expansion performed by the GNU Assembler.
Differential Revision: https://reviews.llvm.org/D41949
Patch by Mario Werner.
llvm-svn: 330224
fadd.d is required in order to force floating point registers to be used in
test code, as parameters are passed in integer registers in the soft float
ABI.
Much of this patch is concerned with support for passing f64 on RV32D with a
soft-float ABI. Similar to Mips, introduce pseudoinstructions to build an f64
out of a pair of i32 and to split an f64 to a pair of i32. BUILD_PAIR and
EXTRACT_ELEMENT can't be used, as a BITCAST to i64 would be necessary, but i64
is not a legal type.
llvm-svn: 329871
Zakk Chen