Using PACK for truncations leaves us with intermediate shuffles that can be tricky to remove while the truncation tree is being formed.
This fold helps pull out the PERMQ case which is one of the most common, avoiding some costly lane-crossing shuffles.
A future patch will begin adding more general shuffle folding, which we should be able to use for HADD/HSUB as well.
Pulled out from the ongoing work on D66004, currently we don't do a good job of simplifying variable shuffle masks that have already lowered to constant pool entries.
This patch adds SimplifyDemandedVectorEltsForTargetShuffle (a custom x86 helper) to first try SimplifyDemandedVectorElts (which we already do) and then constant pool simplification to help mark undefined elements.
To prevent lowering/combines infinite loops, we only handle basic constant pool loads instead of creating new BUILD_VECTOR nodes for lowering - e.g. we don't try to convert them to broadcast/vzext_load - there might be some benefit to this but if so I'd rather we come up with some way to reuse existing code than reimplement a lot of BUILD_VECTOR code.
Differential Revision: https://reviews.llvm.org/D81791
Default legalization will create two v8i64 truncs to v8i32, concat
them to v16i32, and then truncate the rest of the way to v16i8.
Instead we can truncate directly from v8i64 to v8i8 in the lower
half of an xmm. Then concat the two halves to use vpunpcklqdq.
This is the same number of uops, but the dependency chain through
the uops is better since the halves are merged at the end.
I had to had SimplifyDemandedBits support for VTRUNC to prevent
a regression on vector-trunc-math.ll. combineTruncatedArithmetic
no longer gets a chance to shrink vXi64 mul so we were producing
the v8i64 multiply sequence using multiple PMULUDQs. With the
demanded bits fix we are able to prune out the extra ops leaving
just two PMULUDQs, one for each v8i64 half. This is twice the
width of the 2 v8i32 PMULLDs we had before, but PMULUDQ is 1
uop and PMULLD is 2. We also save some truncates. It's probably
worth using PMULUDQ even when PMULLQ is available since the latter
is 3 uops, but that will require a different change.
Differential Revision: https://reviews.llvm.org/D79231
We generate PACK instructions with an undef second source when we are truncating from a 128-bit vector to something narrower and we don't care about the upper bits of the vector register. The register allocation process will always assign untied undef uses to xmm0. This creates a false dependency on xmm0.
By adding these instructions to hasUndefRegUpdate, we can get the BreakFalseDeps pass to reassign the source to match the other input. Normally this interface is used for instructions that might need an xor inserted to break the dependency. But the pass also has a heuristic that tries to use the same register as other sources. That should always be possible for these instructions so we'll never trigger the xor dependency break.
Differential Revision: https://reviews.llvm.org/D79032
truncateVectorWithPACK has its own vector length controls, so we can rely on those directly. This helps some existing truncation to subvector tests, which were being combined later during shuffle lowering at which point the sign/zero bit detection had become obscured preventing lowerShuffleWithPACK working as well as it could.
If we're packing from 128-bits to 64-bits then we don't need the RHS argument. This helps with register allocation, especially as we avoid repeating a use of the input value.
Similar to the lowerV16I8Shuffle implementation, for binary compaction v8i16 shuffles we can avoid the PUNPCKLDQ(PSHUFB,PSHUFB) pattern on SSE41+ targets by using PACKUSDW and PBLENDW. Before SSE41 we would need to use PACKSSDW but that requires sign extension that seems to destroy any gains, even on targets without PSHUFB.
This is a bigger gain on AMD than Intel targets but should never be a regression, and avoiding the shuffle mask load(s) is always useful.
Noticed in codegen while dealing with PR31443.
Extend lowerShuffleWithPACK/matchShuffleWithPACK/createPackShuffleMask to handle compaction style shuffle masks that can be lowered to chains of PACKSS/PACKUS if their inputs are suitably sign/zero extended.
This helps avoid PSHUFB (and its mask load) for short shuffle chains, shuffle combining will still replace with a PSHUFB if we have enough shuffles as getFauxShuffleMask should recognise the PACKSS/PACKUS chains.
If canLowerByDroppingEvenElements indicates that the shuffle is a N:1 compaction pattern and the inputs are suitably sign/zero extended then we can use a chain of PACKSS/PACKUS to compact.
This helps avoid PSHUFB (and its mask load) for short shuffle chains, shuffle combining will still replace with a PSHUFB if we have enough shuffles as getFauxShuffleMask can recognise PACKSS/PACKUS chains.
Now that rG18c19441d105 has improved VPERM2X128 handling, we can perform this to improve x64->x32 truncation without poor cross-lane issues.
Someday combineX86ShufflesRecursively will handle this, but we're still really bad at dealing with different vector widths.
Summary:
The 2 source operands commutable instructions are encoded in the
VEX.VVVV field and the r/m field of the MODRM byte plus the VEX.B
field.
The VEX.B field is missing from the 2-byte VEX encoding. If the
VEX.VVVV source is 0-7 and the other register is 8-15 we can
swap them to avoid needing the VEX.B field. This works as long as
the VEX.W, VEX.mmmmm, and VEX.X fields are also not needed.
Fixes PR36706.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68550
This enables use of the saturating truncate instructions when the
result type is less than 128 bits. It also enables the use of
saturating truncate instructions on KNL when the input is less
than 512 bits. We can do this by widening the input and then
extracting the result.
llvm-svn: 374731
Since the input type is larger than 256-bits we'll need to some
concatenating to reassemble the results. The pack instructions
ability to concatenate while packing make this a shorter/faster
sequence.
llvm-svn: 374643
This adds "min-legal-vector-width"="256" function attributes to
all the tests for a larger than 256-bit input. Also switch any
larger than 512-bit inputs to use a load. This makes the
arguments consistent with min-legal-vector-width attribute which
should usually be at least as large as the arguments.
The SKX configuration will avoid using zmm registers on the
modified test cases. For many of them we should use something
closer to the AVX2 codegen with pack instructions instead of
the avx512 saturating truncates.
llvm-svn: 374642
If we don't have VLX we won't end up selecting a saturating
truncate for 256-bit or smaller vectors so we should just use
the pack lowering.
llvm-svn: 374487
Summary:
The default legalization for v16i64->v16i8 tries to create a multiple stage truncate concatenating after each stage and truncating again. But avx512 implements truncates with multiple uops. So it should be better to truncate all the way to the desired element size and then concatenate the pieces using unpckl instructions. This minimizes the number of 2 uop truncates. The unpcks are all single uop instructions.
I tried to handle this by just custom splitting the v16i64->v16i8 shuffle. And hoped that the DAG combiner would leave the two halves in the state needed to make D68374 do the job for each half. This worked for the first half, but the second half got messed up. So I've implemented custom handling for v8i64->v8i8 when v8i64 needs to be split to produce the VTRUNCs directly.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68428
llvm-svn: 373864
We already do this for ISD::TRUNCATE, but we can do the same for X86ISD::VTRUNC
Differential Revision: https://reviews.llvm.org/D68432
llvm-svn: 373765
gcc and icc pass these types in zmm registers in zmm registers.
This patch implements a quick hack to override the register
type before calling convention handling to one that is legal.
Longer term we might want to do something similar to 256-bit
integer registers on AVX1 where we just split all the operations.
Fixes PR42957
Differential Revision: https://reviews.llvm.org/D66708
llvm-svn: 370495
We need AVX512BW to be able to truncate an i16 vector. If we don't
have that we have to extend i16->i32, then trunc, i32->i8. But we
won't be able to remove the min/max if we do that. At least not
without more special handling.
llvm-svn: 368623
The assert that caused this to be reverted should be fixed now.
Original commit message:
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 368183
This reverts commit 3de33245d2c992c9e0af60372043540b60f3a810.
This commit broke the MSan buildbots. See
https://reviews.llvm.org/rL367901 for more information.
llvm-svn: 368107
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 367901
If anything called the recursive isKnownNeverNaN/computeKnownBits/ComputeNumSignBits/SimplifyDemandedBits/SimplifyMultipleUseDemandedBits with an incorrect depth then we could continue to recurse if we'd already exceeded the depth limit.
This replaces the limit check (Depth == 6) with a (Depth >= 6) to make sure that we don't circumvent it.
This causes a couple of regressions as a mixture of calls (SimplifyMultipleUseDemandedBits + combineX86ShufflesRecursively) were calling with depths that were already over the limit. I've fixed SimplifyMultipleUseDemandedBits to not do this. combineX86ShufflesRecursively is trickier as we get a lot of regressions if we reduce its own limit from 8 to 6 (it also starts at Depth == 1 instead of Depth == 0 like the others....) - I'll see what I can do in future patches.
llvm-svn: 367171
If all the demanded elts are from one operand and are inline, then we can use the operand directly.
The changes are mainly from SSE41 targets which has blendvpd but not cmpgtq, allowing the v2i64 comparison to be simplified as we only need the signbit from alternate v4i32 elements.
llvm-svn: 366817
This is a potentially large perf win for AVX1 targets because of the way we
auto-vectorize to 256-bit but then expect the backend to legalize/optimize
for the half-implemented AVX1 ISA.
On the motivating example from PR37428 (even though this patch doesn't solve
the vector shift issue):
https://bugs.llvm.org/show_bug.cgi?id=37428
...there's a 16% speedup when compiling with "-mavx" (perf tested on Haswell)
because we eliminate the remaining 256-bit vblendv ops.
I added comments on a couple of tests that require further work. If we have
256-bit logic ops separating the vselect and extract, we should probably narrow
everything to 128-bit, but that requires a larger pattern match.
Differential Revision: https://reviews.llvm.org/D62969
llvm-svn: 362797
combineVectorTruncationWithPACKUS is currently splitting the upper bit bit masking into 128-bit subregs and then concatenating them back together.
This was originally done to avoid regressions that caused existing subregs to be concatenated to the larger type just for the AND masking before being extracted again. This was fixed by @spatel (notably rL303997 and rL347356).
This also lets SimplifyDemandedBits do some further improvements before it hits the recursive depth limit.
My only annoyance with this is that we were broadcasting some xmm masks but we seem to have lost them by moving to ymm - but that's a known issue as the logic in lowerBuildVectorAsBroadcast isn't great.
Differential Revision: https://reviews.llvm.org/D60375#inline-539623
llvm-svn: 358692
Various SelectionDAG non-combine operations (e.g. the getNode smart
constructor and legalization) may leave dangling nodes by applying
optimizations or not fully pruning unused result values. This can
result in nodes that are never added to the worklist and therefore can
not be pruned.
Add a node inserter as the current node deleter to make sure such
nodes have the chance of being pruned.
Many minor changes, mostly positive.
llvm-svn: 356996
Similar to D57867 - this is a small patch with lots of test diffs.
With half-vector-width narrowing potential, using an extract + 128-bit vshufps
is a win because it replaces a 256-bit shuffle with a 128-bit shufle.
This seems like it should be a win even for targets with 'fast-variable-shuffle',
but we are intentionally deferring that to an independent change to make sure
that is true.
Differential Revision: https://reviews.llvm.org/D58181
llvm-svn: 354279
With avx512f but not avx512bw we need to extend to v16i32 then truncate that to to v16i8. Previously we emitted both nodes during lowering, but I'm trying to switch to using target independent nodes and with that switched the extend+truncate wou
This patch changes the implementation to what will be necessary with that patch which helps minimize test diffs.
llvm-svn: 346552
Summary:
I've noticed that the bitcasts we introduce for these make computeKnownBits and computeNumSignBits not work well in LegalizeVectorOps. LegalizeVectorOps legalizes bottom up while LegalizeDAG legalizes top down. The bottom up strategy for LegalizeVectorOps means operands are legalized before their uses. So we promote and/or/xor before we legalize the operands that use them making computeKnownBits/computeNumSignBits in places like LowerTruncate suboptimal. I looked at changing LegalizeVectorOps to be top down as well, but that was more disruptive and caused some regressions. I also looked at just moving promotion of binops to LegalizeDAG, but that had a few issues one around matching AND,ANDN,OR into VSELECT because I had to create ANDN as vXi64, but the other nodes hadn't legalized yet, I didn't look too hard at fixing that.
This patch seems to produce better results overall than my other attempts. We now form broadcasts of constants better in some cases. For at least some of them the AND was being introduced in LegalizeDAG, promoted to vXi64, and the BUILD_VECTOR was also legalized there. I think we got bad ordering of that. Now the promotion is out of the legalizer so we handle this better.
In the longer term I think we really should evaluate whether we should be doing this promotion at all. It's really there to reduce isel pattern count, but I'm wondering if we'd be better served just eating the pattern cost or doing C++ based isel for vector and/or/xor in X86ISelDAGToDAG. The masked and/or/xor will definitely be difficult in patterns if a bitcast gets between the vselect and the and/or/xor node. That becomes a lot of permutations to cover.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D53107
llvm-svn: 344487
This may give slightly better opportunities for DAG combine to simplify with the operations before the setcc. It also matches the type the xors will eventually be promoted to anyway so it saves a legalization step.
Almost all of the test changes are because our constant pool entry is now v2i64 instead of v4i32 on 64-bit targets. On 32-bit targets getConstant should be emitting a v4i32 build_vector and a v4i32->v2i64 bitcast.
There are a couple test cases where it appears we now combine a bitwise not with one of these xors which caused a new constant vector to be generated. This prevented a constant pool entry from being shared. But if that's an issue we're concerned about, it seems we need to address it another way that just relying a bitcast to hide it.
This came about from experiments I've been trying with pushing the promotion of and/or/xor to vXi64 later than LegalizeVectorOps where it is today. We run LegalizeVectorOps in a bottom up order. So the and/or/xor are promoted before their users are legalized. The bitcasts added for the promotion act as a barrier to computeKnownBits if we try to use it during vector legalization of a later operation. So by moving the promotion out we can hopefully get better results from computeKnownBits/computeNumSignBits like in LowerTruncate on AVX512. I've also looked at running LegalizeVectorOps in a top down order like LegalizeDAG, but thats showing some other issues.
llvm-svn: 344071
Simplify combineVectorTruncationWithPACKUS to mask the upper bits followed by calling truncateVectorWithPACK instead of duplicating with similar code.
This results in the codegen using (V)PACKUSDW on SSE41+ targets for vXi64/vXi32 inputs where before it always used PACKUSWB (along with a lot more bitcasting).
I've raised PR37749 as until we avoid unnecessary concats back to 256-bit for bitwise ops, we can't avoid splitting the input value into 128-bit subvectors for masking.
llvm-svn: 334289
We have some combines/lowerings that attempt to use PACKSS-then-PACKUS and others that use PACKUS-then-PACKSS.
PACKUS is much easier to combine with if we know the upper bits are zero as ComputeKnownBits can easily see through BITCASTs etc. especially now that rL333995 and rL334007 have landed. It also effectively works at byte level which further simplifies shuffle combines.
The only (minor) annoyances are that ComputeKnownBits can sometimes take longer as it doesn't fail as quickly as ComputeNumSignBits (but I'm not seeing any actual regressions in tests) and PACKUSDW only became available after SSE41 so we have more codegen diffs between targets.
llvm-svn: 334276
Summary:
New unsigned saturation downconvert patterns detection was implemented in
X86 Codegen:
(truncate (smin (smax (x, C1), C2)) to dest_type),
where C1 >= 0 and C2 is unsigned max of destination type.
(truncate (smax (smin (x, C2), C1)) to dest_type)
where C1 >= 0, C2 is unsigned max of destination type and C1 <= C2.
These two patterns are equivalent to:
(truncate (umin (smax(x, C1), unsigned_max_of_dest_type)) to dest_type)
Reviewers: RKSimon
Subscribers: llvm-commits, a.elovikov
Differential Revision: https://reviews.llvm.org/D45315
llvm-svn: 332336
If we have a clamp pattern, SMIN(SMAX(X, LO),HI) or SMAX(SMIN(X, HI),LO) then we can deduce that the number of signbits (zeros/ones) will be at least the minimum of the LO and HI constants.
ComputeKnownBits equivalent of D43338.
Differential Revision: https://reviews.llvm.org/D43463
llvm-svn: 325521
If we have a clamp pattern, SMIN(SMAX(X, LO),HI) or SMAX(SMIN(X, HI),LO) then we can deduce that the number of signbits will be at least the minimum of the LO and HI constants.
I haven't bothered with the UMIN/UMAX equivalent as (1) we don't have any current use cases and (2) I wonder if we'd be better off immediately falling back for ComputeKnownBits for UMIN/UMAX which already has optimization patterns useful for unsigned cases.
Differential Revision: https://reviews.llvm.org/D43338
llvm-svn: 325450
This reverts commit r323991.
This commit breaks target that don't model all the register constraints
in TableGen. So far the workaround was to set the
hasExtraXXXRegAllocReq, but it proves that it doesn't cover all the
cases.
For instance, when mutating an instruction (like in the lowering of
COPYs) the isRenamable flag is not properly updated. The same problem
will happen when attaching machine operand from one instruction to
another.
Geoff Berry is working on a fix in https://reviews.llvm.org/D43042.
llvm-svn: 325421
While the AVX512 VTRUNCS/VTRUNCUS instructions require legal types, truncateVectorWithPACK handles cases with multiples of legal types through splitting/concatenation. So we just need to ensure that the src/dst scalar types are correct and leave truncateVectorWithPACK to handle the rest of it.
llvm-svn: 325127
