Migrate the X86 backend from X86ISD opcodes ADDS and SUBS to generic
ISD opcodes SADDSAT and SSUBSAT. This also improves scodegen for
@llvm.sadd.sat() and @llvm.ssub.sat() intrinsics.
This is a followup to D55787 and part of PR40056.
Differential Revision: https://reviews.llvm.org/D55833
llvm-svn: 349520
Replace the X86ISD opcodes ADDUS and SUBUS with generic ISD opcodes
UADDSAT and USUBSAT. As a side-effect, this also makes codegen for
the @llvm.uadd.sat and @llvm.usub.sat intrinsics reasonable.
This only replaces use in the X86 backend, and does not move any of
the ADDUS/SUBUS X86 specific combines into generic codegen.
Differential Revision: https://reviews.llvm.org/D55787
llvm-svn: 349481
Previously, the extend_vector_inreg opcode required their input register to be the same total width as their output. But this doesn't match up with how the X86 instructions are defined. For X86 the input just needs to be a legal type with at least enough elements to cover the output.
This patch weakens the check on these nodes and allows them to be used as long as they have more input elements than output elements. I haven't changed type legalization behavior so it will still create them with matching input and output sizes.
X86 will custom legalize these nodes by shrinking the input to be a 128 bit vector and once we've done that we treat them as legal operations. We still have one case during type legalization where we must custom handle v64i8 on avx512f targets without avx512bw where v64i8 isn't a legal type. In this case we will custom type legalize to a *extend_vector_inreg with a v16i8 input. After that the input is a legal type so type legalization should ignore the node and doesn't need to know about the relaxed restriction. We are no longer allowed to use the default expansion for these nodes during vector op legalization since the default expansion uses a shuffle which required the widths to match. Custom legalization for all types will prevent us from reaching the default expansion code.
I believe DAG combine works correctly with the released restriction because it doesn't check the number of input elements.
The rest of the patch is changing X86 to use either the vector_inreg nodes or the regular zero_extend/sign_extend nodes. I had to add additional isel patterns to handle any_extend during isel since simplifydemandedbits can create them at any time so we can't legalize to zero_extend before isel. We don't yet create any_extend_vector_inreg in simplifydemandedbits.
Differential Revision: https://reviews.llvm.org/D54346
llvm-svn: 346784
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
SimplifyDemandedBits can turn a sign_extend back into an any_extend and trigger an infinite loop. So instead legalize it the same way as a sign_extend, but preserve the opcode. Then just pattern match it the same as sign_extend during isel.
I don't have a reduced test case for such an infinite loop yet.
llvm-svn: 346170
These promotions add additional bitcasts to the SelectionDAG that can pessimize computeKnownBits/computeNumSignBits. It also seems to interfere with broadcast formation.
This patch removes the promotion and adds isel patterns instead.
The increased table size is more than I would like, but hopefully we can find some canonicalizations or other tricks to start pruning out patterns going forward.
Differential Revision: https://reviews.llvm.org/D53268
llvm-svn: 345408
This B/W VPTEST instructions are only available with AVX512BW. But lowering should prevent any byte or word elements from getting to isel so this can't be exposed.
llvm-svn: 345112
I've included a fix to DAGCombiner::ForwardStoreValueToDirectLoad that I believe will prevent the previous miscompile.
Original commit message:
Theoretically this was done to simplify the amount of isel patterns that were needed. But it also meant a substantial number of our isel patterns have to match an explicit bitcast. By making the vXi32/vXi16/vXi8 types legal for loads, DAG combiner should be able to change the load type to rem
I had to add some additional plain load instruction patterns and a few other special cases, but overall the isel table has reduced in size by ~12000 bytes. So it looks like this promotion was hurting us more than helping.
I still have one crash in vector-trunc.ll that I'm hoping @RKSimon can help with. It seems to relate to using getTargetConstantFromNode on a load that was shrunk due to an extract_subvector combine after the constant pool entry was created. So we end up decoding more mask elements than the lo
I'm hoping this patch will simplify the number of patterns needed to remove the and/or/xor promotion.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: llvm-commits, RKSimon
Differential Revision: https://reviews.llvm.org/D53306
llvm-svn: 344965
This makes fast isel treat all legal vector types the same way. Previously only vXi64 was in the fast-isel tables.
This unfortunately prevents matching of andn by fast-isel for these types since the requires SelectionDAG. But we already had this issue for vXi64. So at least we're consistent now.
Interestinly it looks like fast-isel can't handle instructions with constant vector arguments so the the not part of the andn patterns is selected with SelectionDAG. This explains why VPTERNLOG shows up in some of the tests.
This is a subset of D53268. As I make progress on that, I will try to reduce the number of lines in the tablegen files.
llvm-svn: 344884
Summary:
Theoretically this was done to simplify the amount of isel patterns that were needed. But it also meant a substantial number of our isel patterns have to match an explicit bitcast. By making the vXi32/vXi16/vXi8 types legal for loads, DAG combiner should be able to change the load type to remove the bitcast.
I had to add some additional plain load instruction patterns and a few other special cases, but overall the isel table has reduced in size by ~12000 bytes. So it looks like this promotion was hurting us more than helping.
I still have one crash in vector-trunc.ll that I'm hoping @RKSimon can help with. It seems to relate to using getTargetConstantFromNode on a load that was shrunk due to an extract_subvector combine after the constant pool entry was created. So we end up decoding more mask elements than the load size.
I'm hoping this patch will simplify the number of patterns needed to remove the and/or/xor promotion.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: llvm-commits, RKSimon
Differential Revision: https://reviews.llvm.org/D53306
llvm-svn: 344877
These included a bitcast of a load from v4f32 to v2f64, but DAG combine should have already changed the type of the load to remove the cast.
llvm-svn: 344573
Currently we hardcode instructions with ReadAfterLd if the register operands don't need to be available until the folded load has completed. This doesn't take into account the different load latencies of different memory operands (PR36957).
This patch adds a ReadAfterFold def into X86FoldableSchedWrite to replace ReadAfterLd, allowing us to specify the load latency at a scheduler class level.
I've added ReadAfterVec*Ld classes that match the XMM/Scl, XMM and YMM/ZMM WriteVecLoad classes that we currently use, we can tweak these values in future patches once this infrastructure is in place.
Differential Revision: https://reviews.llvm.org/D52886
llvm-svn: 343868
Our lowering that tries to avoid this sign extend can be defeated by the DAG combine folding it with a truncate.
The pattern needs to extend to an v8i32 then truncate back down to v8i16.
llvm-svn: 342830
Summary:
Previously the value being stored is the last operand in SDNode. This causes the type legalizer to visit the mask operand before the value operand. The type legalizer was more complicated because of this since we want the type of the value to drive the decisions.
This patch moves the value to be the first operand so we visit it first during type legalization. It also simplifies the type legalization code accordingly.
X86 is currently the only in tree target that uses this SDNode. Not sure if there are any users out of tree.
Reviewers: RKSimon, delena, hfinkel, eli.friedman
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D50402
llvm-svn: 340689
AVX512 added new versions of these intrinsics that take a rounding mode. If the rounding mode is 4 the new intrinsics are equivalent to the old intrinsics.
The AVX512 intrinsics were being lowered to ISD opcodes, but the legacy SSE intrinsics were left as intrinsics. This resulted in the AVX512 instructions needing separate patterns for the ISD opcodes and the legacy SSE intrinsics.
Now we convert SSE intrinsics and AVX512 intrinsics with rounding mode 4 to the same ISD opcode so we can share the isel patterns.
llvm-svn: 339749
Ideally our ISD node types going into the isel table would have types consistent with their instruction domain. This prevents us having to duplicate patterns with different types for the same instruction.
Unfortunately, it seems our shuffle combining is currently relying on this a little remove some bitcasts. This seems to enable some switching between shufps and shufd. Hopefully there's some way we can address this in the combining.
Differential Revision: https://reviews.llvm.org/D49280
llvm-svn: 337590
The X86ISD::MOVLHPS/MOVHLPS should now only be emitted in SSE1 only. This means that the v2i64/v2f64 types would be illegal thus we don't need these patterns.
llvm-svn: 337349
Previously we passed 'null_frag' into the instruction definition. The multiclass is shared with MOVHPD which doesn't use null_frag. It turns out by passing X86Movsd it produces patterns equivalent to some standalone patterns.
llvm-svn: 337299
This amounts to pretty ridiculous number of patterns. Ideally we'd canonicalize the X86ISD::VRNDSCALE earlier to reuse those patterns. I briefly looked into doing that, but some strict FP operations could still get converted to rint and nearbyint during isel. It's probably still worthwhile to look into. This patch is meant as a starting point to work from.
llvm-svn: 337234
This unfortunately requires a bunch of bitcasts to be added added to SUBREG_TO_REG, COPY_TO_REGCLASS, and instructions in output patterns. Otherwise tablegen seems to default to picking f128 and then we fail when something tries to get the register class for f128 which isn't always valid.
The test changes are because we were previously mixing fr128 and vr128 due to contrainRegClass finding FR128 first and passes like live range shrinking weren't handling that well.
llvm-svn: 337147
indices used by AVX2 and AVX-512 gather instructions.
The index vector is hardened by broadcasting the predicate state
into a vector register and then or-ing. We don't even have to worry
about EFLAGS here.
I've added a test for all of the gather intrinsics to make sure that we
don't miss one. A particularly interesting creation is the gather
prefetch, which needs to be marked as potentially "loading" to get the
correct behavior. It's a memory access in many ways, and is actually
relevant for SLH. Based on discussion with Craig in review, I've moved
it to be `mayLoad` and `mayStore` rather than generic side effects. This
matches how we model other prefetch instructions.
Many thanks to Craig for the review here.
Differential Revision: https://reviews.llvm.org/D49336
llvm-svn: 337144
AVX512 doesn't have an immediate controlled blend instruction. But blend throughput is still better than movss/sd on SKX.
This commit changes AVX512 to use the AVX blend instructions instead of MOVSS/MOVSD. This constrains the register allocation since it won't be able to use XMM16-31, but hopefully the increased throughput and reduced port 5 pressure makes up for that.
llvm-svn: 337083
This is not an optimization we should be doing in isel. This is more suitable for a DAG combine.
My main concern is a future time when we support more FPENV. Changing a packed op to a scalar op could cause us to miss some exceptions that should have occured if we had done a packed op. A DAG combine would be better able to manage this.
llvm-svn: 336971
We now use llvm.fma.f32/f64 or llvm.x86.fmadd.f32/f64 intrinsics that use scalar types rather than vector types. So we don't these special ISD nodes that operate on the lowest element of a vector.
llvm-svn: 336883
This converts them to what clang is now using for codegen. Unfortunately, there seem to be a few kinks to work out still. I'll try to address with follow up patches.
llvm-svn: 336871
Before revision 336728, the "mayLoad" flag for instruction (V)MOVLPSrm was
inferred directly from the "default" pattern associated with the instruction
definition.
r336728 removed special node X86Movlps, and all the patterns associated to it.
Now instruction (V)MOVLPSrm doesn't have a pattern associated to it, and the
'mayLoad/hasSideEffects' flags are left unset.
When the instruction info is emitted by tablegen, method
CodeGenDAGPatterns::InferInstructionFlags() sees that (V)MOVLPSrm doesn't have a
pattern, and flags are undefined. So, it conservatively sets the
"hasSideEffects" flag for it.
As a consequence, we were losing the 'mayLoad' flag, and we were gaining a
'hasSideEffect' flag in its place.
This patch fixes the issue (originally reported by Michael Holmen).
The mca tests show the differences in the instruction info flags. Instructions
that were affected by this problem were: MOVLPSrm/VMOVLPSrm/VMOVLPSZ128rm.
Differential Revision: https://reviews.llvm.org/D49182
llvm-svn: 336818
These patterns looked for a MOVSS/SD followed by a scalar_to_vector. Or a scalar_to_vector followed by a load.
In both cases we emitted a MOVSS/SD for the MOVSS/SD part, a REG_CLASS for the scalar_to_vector, and a MOVSS/SD for the load.
But we have patterns that do each of those 3 things individually so there's no reason to build large patterns.
Most of the test changes are just reorderings. The one test that had a meaningful change is pr30430.ll and it appears to be a regression. But its doing -O0 so I think it missed a lot of opportunities and was just getting lucky before.
llvm-svn: 336762
Some added 20 and some added 15. Its unclear when to use which value and whether they are required at all.
This patch removes them all. If we start finding real world issues we may need to add them back with proper tests.
llvm-svn: 336735
These ISD nodes try to select the MOVLPS and MOVLPD instructions which are special load only instructions. They load data and merge it into the lower 64-bits of an XMM register. They are logically equivalent to our MOVSD node plus a load.
There was only one place in X86ISelLowering that used MOVLPD and no places that selected MOVLPS. The one place that selected MOVLPD had to choose between it and MOVSD based on whether there was a load. But lowering is too early to tell if the load can really be folded. So in isel we have patterns that use MOVSD for MOVLPD if we can't find a load.
We also had patterns that select the MOVLPD instruction for a MOVSD if we can find a load, but didn't choose the MOVLPD ISD opcode for some reason.
So it seems better to just standardize on MOVSD ISD opcode and manage MOVSD vs MOVLPD instruction with isel patterns.
llvm-svn: 336728
These patterns mapped (v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))) to a MOVSD and an zeroing XOR. But the complexity of a pattern for (v2f64 (X86vzmovl (v2f64))) that selects MOVQ is artificially and hides this MOVSD pattern.
Weirder still, the SSE version of the pattern was explicitly blocked on SSE41, but yet we had copied it to AVX and AVX512.
llvm-svn: 336556
This allows us to handle masking in a very similar way to the default rounding version that uses llvm.fma.
I had to add new rounding mode CodeGenOnly instructions to support isel when we can't find a movss to grab the upper bits from to use the b_Int instruction.
Fast-isel tests have been updated to match new clang codegen.
We are currently having trouble folding fneg into the new intrinsic. I'm going to correct that in a follow up patch to keep the size of this one down.
A future patch will also remove the old intrinsics.
llvm-svn: 336506
We've removed the legacy FMA3 intrinsics and are now using llvm.fma and extractelement/insertelement. So we don't need patterns for the nodes that could only be created by the old intrinscis. Those ISD opcodes still exist because we haven't dropped the AVX512 intrinsics yet, but those should go to EVEX instructions.
llvm-svn: 336457
We have patterns for SELECTS that top at v1i1 and we have a pattern for (v1i1 (scalar_to_vector GR8)). The patterns being removed here do the same thing as the two other patterns combined so there is no need for them.
llvm-svn: 336305
I don't believe there is any real reason to have separate X86 specific opcodes for vector compares. Setcc has the same behavior just uses a different encoding for the condition code.
I had to change the CondCodeAction for SETLT and SETLE to prevent some transforms from changing SETGT lowering.
Differential Revision: https://reviews.llvm.org/D43608
llvm-svn: 335173
This patch handles back-end folding of generic patterns created by lowering the
X86 rounding intrinsics to native IR in cases where the instruction isn't a
straightforward packed values rounding operation, but a masked operation or a
scalar operation.
Differential Revision: https://reviews.llvm.org/D45203
llvm-svn: 335037
This adds an EVEX2VEXOverride string to the X86 instruction class in X86InstrFormats.td. If this field is set it will add manual entry in the EVEX->VEX tables that doesn't check the encoding information.
Then use this mechanism to map VMOVDU/A8/16, 128-bit VALIGN, and VPSHUFF/I instructions to VEX instructions.
Finally, remove the manual table from the emitter.
This has the bonus of fully sorting the autogenerated EVEX->VEX tables by their EVEX instruction enum value. We may be able to use this to do a binary search for the conversion and get rid of the need to create a DenseMap.
llvm-svn: 335018
EVEX makes heavy use of the VEX.W bit to indicate 64-bit element vs 32-bit elements. Many of the VEX instructions were split into 2 versions with different masking granularity.
The EVEX->VEX table generate can collapse the two versions if the VEX version uses is tagged as VEX_WIG. But if the VEX version is instead marked VEX.W==0 we can't combine them because we don't know if there is also a VEX version with VEX.W==1.
This patch adds a new VEX_W1X tag that indicates the EVEX instruction encodes with VEX.W==1, but is safe to convert to a VEX instruction with VEX.W==0.
This allows us to remove a bunch of manual EVEX->VEX table entries. We may want to look into splitting up the VEX_WPrefix field which would simplify the disassembler.
llvm-svn: 335017
The instructions that use this class don't have another source register. So I think this was just marking one of the address operands as ReadAfterLd?
llvm-svn: 334994
Rather than having an exclusion list in tablegen sources, add a flag to the X86 instruction records that can be used to suppress checking for convertibility.
llvm-svn: 334971
We already have these aliases for EVEX enocded instructions, but not for the GPR, MMX, SSE, and VEX versions.
Also remove the vpextrw.s EVEX alias. That's not something gas implements.
llvm-svn: 334922
The .s assembly strings allow the reversed forms to be targeted from assembly which matches gas behavior. But when printing the instructions we should print them without the .s to match other tooling like objdump. By using InstAliases we can use the normal string in the instruction and just hide it from the assembly parser.
Ideally we'd add the .s versions to the legacy SSE and VEX versions as well for full compatibility with gas. Not sure how we got to state where only EVEX was supported.
llvm-svn: 334920
VMOVPQIto64Zmr is not a 64-bit mode only instruction. But I don't know how to test this because VMOVPQIto64mr should always have priority over it in 32-bit mode since its only advantage is XMM16-XMM31 which aren't usable in 32-bit mode.
VMOVPQIto64Zrr is a 64-bit mode only instruction, but we don't need to explicitly mark it as such because it uses a GR64 register which won't parse in 32-bit mode.
llvm-svn: 334896
Summary:
The tests in:
https://bugs.llvm.org/show_bug.cgi?id=37751
...show miscompiles because we wrongly mapped and folded x86-specific intrinsics into generic DAG nodes.
This patch corrects the mappings in X86IntrinsicsInfo.h and adds isel matching corresponding to the new patterns. The complete tests for the failure cases should be in avx-cvttp2si.ll and sse-cvttp2si.ll and avx512-cvttp2i.ll
Reviewers: RKSimon, gbedwell, spatel
Reviewed By: spatel
Subscribers: mcrosier, llvm-commits
Differential Revision: https://reviews.llvm.org/D47993
llvm-svn: 334685
Previously we were whitelisting in instructions based on their SchedRW value. With the masked store instructions explicitly removed via NotMemoryFoldable, we don't seem to need this check anymore.
llvm-svn: 334563
We were missing packed isel folding patterns for all of sse41, avx, and avx512.
For some reason avx512 had scalar load folding patterns under optsize(due to partial/undef reg update), but we didn't have the equivalent sse41 and avx patterns.
Sometimes we would get load folding due to peephole pass anyway, but we're also missing avx512 instructions from the load folding table. I'll try to fix that in another patch.
Some of this was spotted in the review for D47993.
This patch adds all the folds to isel, adds a few spot tests, and disables the peephole pass on a few tests to ensure we're testing some of these patterns.
llvm-svn: 334460
Necessary for D46276 as even though btver2 doesn't use these instructions, its now flagged as complete so complains if ANY instruction isn't tagged.....
UnsupportedFeatures wouldn't help here as these instructions don't appear to have a feature predicate (like a lot of AVX512).
llvm-svn: 334423
Summary:
This fixes most of the scheduling info for SKX vector operations.
I had to split a lot of the YMM/ZMM classes into separate classes for YMM and ZMM.
The before/after llvm-exegesis analysis are in the phabricator diff.
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47721
llvm-svn: 334407
The index size is represented by the letter after the 'v'. The number represents the memory size. If an 'x' appears after the number its means the index register can be from VR128X/VR256X instead of VR128/VR256.
As vy512mem uses a VR256X index it should have an x.
And vz256mem uses a VR512 index so it shouldn't have an x.
I admit these names kind of suck and are confusing.
llvm-svn: 334120
Summary:
The new rules are straightforward. The main rules to keep in mind
are:
1. NAME is an implicit template argument of class and multiclass,
and will be substituted by the name of the instantiating def/defm.
2. The name of a def/defm in a multiclass must contain a reference
to NAME. If such a reference is not present, it is automatically
prepended.
And for some additional subtleties, consider these:
3. defm with no name generates a unique name but has no special
behavior otherwise.
4. def with no name generates an anonymous record, whose name is
unique but undefined. In particular, the name won't contain a
reference to NAME.
Keeping rules 1&2 in mind should allow a predictable behavior of
name resolution that is simple to follow.
The old "rules" were rather surprising: sometimes (but not always),
NAME would correspond to the name of the toplevel defm. They were
also plain bonkers when you pushed them to their limits, as the old
version of the TableGen test case shows.
Having NAME correspond to the name of the toplevel defm introduces
"spooky action at a distance" and breaks composability:
refactoring the upper layers of a hierarchy of nested multiclass
instantiations can cause unexpected breakage by changing the value
of NAME at a lower level of the hierarchy. The new rules don't
suffer from this problem.
Some existing .td files have to be adjusted because they ended up
depending on the details of the old implementation.
Change-Id: I694095231565b30f563e6fd0417b41ee01a12589
Reviewers: tra, simon_tatham, craig.topper, MartinO, arsenm, javed.absar
Subscribers: wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D47430
llvm-svn: 333900
These instructions are unusual in that they operate on 4 consecutive registers so supporting them in codegen will be more difficult than normal.
Includes an assembler check to warn if the source register is not the first register of a 4 register group.
llvm-svn: 333812
We only need the extractelt that corresponds to the register we're trying to insert back into. We can't guarantee the others haven't been optimized out depending on how those operands were produced.
So instead just look for an FR32/FR64 input and emit a COPY_TO_REGCLASS to VR128 in the output pattern. This matches what we do for ADD/SUB/MUL/DIV.
llvm-svn: 333473
1. Introduction of mask scalar TableGen patterns.
2. Introduction of new scalar move TableGen patterns
and refactoring of existing ones.
3. Folding of pattern created by introducing scalar
masking in Clang header files.
Patch by tkrupa
Differential Revision: https://reviews.llvm.org/D47012
llvm-svn: 333419
These do the same thing with the first and second sources swapped. They previously came from separate intrinsics that specified different masking behavior. But we can cover that with isel patterns and a single node.
This is a step towards reducing the number of intrinsics needed.
A bunch of tests change because we are now biased to choosing VPERMT over VPERMI when there is nothing to signal that commuting is beneficial.
llvm-svn: 333383
This basically reverts r280696 in favor of using extra patterns as mentioned as an alternative in that commit message. For now I've only added the cases we have test cases for, but it should be easy to add more in the future.
This will help to convert VPERMI2PS/VPERMT2PS intrinsics to use a single ISD node opcode. And hopefully allow some intrinsics to be removed.
llvm-svn: 333365
This property is needed in order to follow values movement between
registers. This property is used in TII to implement method that
returns true if simple copy like instruction is recognized, along
with source and destination machine operands.
Patch by Nikola Prica.
Differential Revision: https://reviews.llvm.org/D45204
llvm-svn: 333093
BtVer2 - fix NumMicroOp and account for the Lat+6cy GPR->XMM and Lat+1cy XMm->GPR delays (see rL332737)
The high number of MOVD/MOVQ equivalent instructions meant that there were a number of missed patterns in SNB/Znver1:
SNB - add missing GPR<->MMX costs (taken from Agner / Intel AOM)
Znver1 - add missing GPR<->XMM MOVQ costs (taken from Agner)
llvm-svn: 332745
Retag some instructions that were missed when we split off vector load/store/moves - MOVQ/MOVD etc.
Fixes BtVer2/SLM which have different behaviours for GPR stores.
llvm-svn: 332718
Retag some instructions that were missed when we split off vector load/store/moves - MOVSS/MOVSD/MOVHPD/MOVHPD/MOVLPD/MOVLPS etc.
Fixes BtVer2/SLM which have different behaviours for GPR stores.
llvm-svn: 332714
The FIXME comments were about preventing load folding to avoid a partial xmm update. But these instructions use GPR as input when the load isn't folded. This won't help prevent a partial xmm update.
llvm-svn: 332573
A lot of the models still have too many InstRW overrides for these new classes - this needs cleaning up but I wanted to get the classes in first
llvm-svn: 332451
BtVer2 - Fixes schedules for (V)CVTPS2PD instructions
A lot of the Intel models still have too many InstRW overrides for these new classes - this needs cleaning up but I wanted to get the classes in first
llvm-svn: 332376
Btver2 - VCVTPH2PSYrm needs to double pump the AGU
Broadwell - missing VCVTPS2PH*mr stores extra latency
Allows us to remove the WriteCvtF2FSt conversion store class
llvm-svn: 332357
Clang's codegen now uses 128-bit masked load/store intrinsics in IR. The backend will widen to 512-bits on AVX512F targets.
So this patch adds patterns to detect codegen's widening and patterns for AVX512VL that don't get widened.
We may be able to drop some of the old patterns, but I leave that for a future patch.
llvm-svn: 332049
Split to support single/double for scalar, XMM and YMM/ZMM instructions - removing InstrRW overrides for these instructions.
Fixes Atom ADDSUBPD instruction and reclassifies VFPCLASS as WriteFCmp which is closer in behaviour.
llvm-svn: 331672
WriteFRcp/WriteFRsqrt are split to support scalar, XMM and YMM/ZMM instructions.
WriteFSqrt is split into single/double/long-double sizes and scalar, XMM, YMM and ZMM instructions.
This removes all InstrRW overrides for these instructions.
NOTE: There were a couple of typos in the Znver1 model - notably a 1cy throughput for SQRT that is highly unlikely and doesn't tally with Agner.
NOTE: I had to add Agner's numbers for several targets for WriteFSqrt80.
llvm-svn: 331629
Split off from SchedWriteFAdd for fp rounding/bit-manipulation instructions.
Fixes an issue on btver2 which only had the ymm version using the JSTC pipe instead of JFPA.
llvm-svn: 331515
This took a bit of extra work as on Intel targets the old (V)PSLLDrr/(V)PSLLDrm style instructions act differently - I ended up creating WriteVecShiftImm classes for XMM/YMM/ZMM vector shift by immediate and retaining WriteVecShift as the default (used only by MMX) plus WriteVecShiftX/WriteVecShiftY. X86SchedWriteWidths hides most of this thank goodness.
llvm-svn: 331472
Also retagged VDBPSADBW instructions as SchedWritePSADBW instead of SchedWriteVecIMul which matches the behaviour on SkylakeServer (the only thing that supports it...)
llvm-svn: 331445
We need to split most of the scheduler classes by vector width to remove more of the InstRW overrides, this patch should make this easier/tidier by allowing us to pass the X86SchedWriteWidths wrapper to multi-width multiclasses and then split as required.
I've included fields for Scl (scalar float/double), MMX (MMX integer), XMM, YMM and ZMM widths. These fields mostly share the same classes but it should give us the flexibility that we may need in the future.
This patch has replaced a set of example SSE/AVX512 instruction cases but isn't exhaustive as it gets very noisy before we really need the functionality.
Differential Revision: https://reviews.llvm.org/D46266
llvm-svn: 331208
Many of these aliases exist to give one syntax or the other a slightly different mnemonic and the other variant gets a duplicate of its normal mnemonic
This patch restricts a lot of these to only one variant so we don't get the duplication.
This removes a lot of duplicate entries from the matcher table. It also reduces the number of warnings printed when you enable the ambiguous match warning in tablegen.
llvm-svn: 331117
Split off pinsr/pextr and extractps instructions.
(Mostly) fixes PR36887.
Note: It might be worth adding a WriteFInsertLd class as well in the future.
Differential Revision: https://reviews.llvm.org/D45929
llvm-svn: 330714
Split the fp and integer vector logical instruction scheduler classes - older CPUs especially often handled these on different pipes.
This unearthed a couple of things that are also handled in this patch:
(1) We were tagging avx512 fp logic ops as WriteFAdd, probably because of the lack of WriteFLogic
(2) SandyBridge had integer logic ops only using Port5, when afaict they can use Ports015.
(3) Cleaned up x86 FCHS/FABS scheduling as they are typically treated as fp logic ops.
Differential Revision: https://reviews.llvm.org/D45629
llvm-svn: 330480
Split VCMP/VMAX/VMIN instructions off to WriteFCmp and VCOMIS instructions off to WriteFCom instead of assuming they match WriteFAdd
Differential Revision: https://reviews.llvm.org/D45656
llvm-svn: 330179
This removes the last of the x86 schedule itineraries, I'm intending to cleanup the remaining uses of NoItinerary/OpndItins/etc. before resolving PR37093.
llvm-svn: 329967
It's failing on the bots and I'm not sure why.
This reverts:
[X86] Synchronize the SchedRW on some EVEX instructions with their VEX equivalents.
[X86] Use WriteFShuffle256 for VEXTRACTF128 to be consistent with VEXTRACTI128 which uses WriteShuffle256.
[X86] Remove some InstRWs for plain store instructions on Sandy Bridge.
[X86] Auto-generate complete checks. NFC
llvm-svn: 329256
These both use a 16-bit load, but one used loadi16_anyext and the other used extloadi32i16. The only difference between them is that loadi16_anyext checked that the load was at least 2 byte aligned and non-volatile. But the alignment doesn't matter here. Just use extloadi32i16 for both.
llvm-svn: 329154
Summary:
It seems many CPUs don't implement this instruction as well as the other vector multiplies. Often using a multi uop flow. Silvermont in particular has a 7 uop flow with 11 cycle throughput. Sandy Bridge implements it as a single uop with 5 cycle latency and 1 cycle throughput. But Haswell and later use 2 uops with 10 cycle latency and 2 cycle throughput.
This patch adds a new X86SchedWritePair we can use to tag this instruction separately. I've provided correct information for Silvermont, Btver2, and Sandy Bridge. I've removed the InstRWs for SandyBridge. I've left Haswell/Broadwell/Skylake InstRWs in place because I wasn't sure how to account for the different load latency between 128 and 256 bits. I also left Znver1 InstRWs in place because the existing values don't match Agner's spreadsheet.
I also left a FIXME in the SandyBridge model because it being used for the "generic" model is too optimistic for the 256/512-bit versions since those are multiple uops on all known CPUs.
Reviewers: RKSimon, GGanesh, courbet
Reviewed By: RKSimon
Subscribers: gchatelet, gbedwell, andreadb, llvm-commits
Differential Revision: https://reviews.llvm.org/D44972
llvm-svn: 328914
This better able to detect undef and zeros pieces in the concat. Or cases when only one subvector is non-zero. This allows us to avoid silly things like double inserts into progressively larger undefs.
This still builds 512 bit concats of 128 bits by building up through 256 bits first. But I don't know if that's best.
We probably want to merge this with the vXi1 concat code since they are very similar.
llvm-svn: 327454
This instruction can be thought of as reading either the even elements of a vXi32 input or the lower half of each element of a vXi64 input. We currently use the vXi32 interpretation, but vXi64 matches better with its broadcast behavior in EVEX.
I'm looking at moving MULDQ/MULUDQ creation to a DAG combine so we can do it when AVX512DQ is enabled without having to go through Custom lowering. But in some of the test cases we failed to use a broadcast load due to the size difference. This should help with that.
I'm also wondering if we can model these instructions in native IR and remove the intrinsics and I think using a vXi64 type will work better with that.
llvm-svn: 326991
We were previously doing this with isel patterns. Moving it to op legalization gives us chance to see the required bitcast earlier. And it lets us remove some isel patterns.
llvm-svn: 326669
While the description for the instruction does mention OR, its talking about how the individual classification test results are ORed together.
The incoming mask is used as a zeroing write mask. If the bit is 1 the classification is written to the output. The bit is 0 the output is 0. This equivalent to an AND.
Here is pseudocode from the intrinsics guide
FOR j := 0 to 1
i := j*64
IF k1[j]
k[j] := CheckFPClass_FP64(a[i+63:i], imm8[7:0])
ELSE
k[j] := 0
FI
ENDFOR
k[MAX:2] := 0
llvm-svn: 326306
This portion can be matched by other patterns. We don't need it to make the larger pattern valid. It's sufficient to have a v1i1 mask input without caring where it came from.
llvm-svn: 325999
These can be created by type legalization promoting the inputs to select to match scalar boolean contents.
We were trying to pattern match them away during isel, but its better to just remove them from the DAG.
I've cleaned up some patterns to not check for this 'and' anymore. But I suspect this has also opened up opportunities for pattern removal.
llvm-svn: 325949
Canonicalize EQ/NE PCMPM to have build vector all zeros on the RHS so we don't have to pattern match it in both locations. This significantly reduces the number of isel patterns needed since we also had to multiply it out with loads being in either operand of the 'and' input node and in the 'and' masking node.
This removes over 24000 bytes from the isel table.
llvm-svn: 325526
Previously we used the immediate encoding if the load was in operand 0 and the short encoding if the load was in operand 1.
This added an insane number of bytes to the size of the isel table. I'm wondering if we should always use the immediate form during isel and change to the short form during emission. This would remove the need to pattern match every combination for both the immediate form and the short form during isel. We could do the same with vpcmpgt
llvm-svn: 325456
Tag AVX512 variants to match SSE/AVX originals.
We only tagged it with the itinerary class, so completeness checks were erroneously passed (PR35639).
llvm-svn: 324901
ISD::ADD implies individual vector element addition with no carries between elements. But for a vXi1 type that would be the same as XOR. And we already turn ISD::ADD into ISD::XOR for all vXi1 types during lowering. So the ISD::ADD pattern would never be able to match anyway.
KADD is different, it adds the elements but also propagates a carry between them. This just a way of doing an add in k-register without bitcasting to the scalar domain. There's still no way to match the pattern, but at least its not obviously wrong.
llvm-svn: 324861
We always created X86ISD::SHUF128 with a 64-bit element type so we can use isel patterns to detect a bitconvert to 32-bit to handle masking.
The test changes are because we also match the bitconvert even if there is no masking. This leads to unnecessary isel pattern, but it requires more multiclass hackery in tablegen to get rid of it.
llvm-svn: 324205
Legalization is still biased to turn LT compares in to GT by swapping operands to avoid needing extra isel patterns to commute.
I'm hoping to remove TESTM/TESTNM next and this should simplify that by making EQ/NE more similar.
llvm-svn: 323604
Previously we had to materialize all 1s in a register using vpternlog or pcmpeq and then xor with that. By using vpternlog directly we can do it in one operation.
This is implemented using isel patterns, but we should maybe consider creating a generalized vpternlog combiner.
llvm-svn: 323572
There are a couple tricky things with this patch.
I had to add an override of isVectorLoadExtDesirable to stop DAG combine from combining sign_extend with loads after legalization since we legalize sextload using a load+sign_extend. Overriding this hook actually prevents a lot sextloads from being created in the first place.
I also had to add isel patterns because DAG combine blindly combines sign_extend+truncate to a smaller sign_extend which defeats what legalization was trying to do.
Differential Revision: https://reviews.llvm.org/D42407
llvm-svn: 323301
All other intrinsic instructions put the _Int on the end. This make these instructions consistent and gets the prefix instregexs in the scheduler models to pick them up.
llvm-svn: 323261
Add missing patterns for inserting v1i1 into a zero vector. Use insert_subvector to zero upper bits before inserting an element into a vXi1 vector. Replace kshift based isel pattern with insert_subvector based pattern now that code that caused the pattern has been fixed to emit insert_subvector.
llvm-svn: 323173
1. ReachingDefsAnalysis - Allows to identify for each instruction what is the “closest” reaching def of a certain register. Used by BreakFalseDeps (for clearance calculation) and ExecutionDomainFix (for arbitrating conflicting domains).
2. ExecutionDomainFix - Changes the variant of the instructions in order to minimize domain crossings.
3. BreakFalseDeps - Breaks false dependencies.
4. LoopTraversal - Creatws a traversal order of the basic blocks that is optimal for loops (introduced in revision L293571). Both ExecutionDomainFix and ReachingDefsAnalysis use this to determine the order they will traverse the basic blocks.
This also included the following changes to ExcecutionDepsFix original logic:
1. BreakFalseDeps and ReachingDefsAnalysis logic no longer restricted by a register class.
2. ReachingDefsAnalysis tracks liveness of reg units instead of reg indices into a given reg class.
Additional changes in affected files:
1. X86 and ARM targets now inherit from ExecutionDomainFix instead of ExecutionDepsFix. BreakFalseDeps also was added to the passes they activate.
2. Comments and references to ExecutionDepsFix replaced with ExecutionDomainFix and BreakFalseDeps, as appropriate.
Additional refactoring changes will follow.
This commit is (almost) NFC.
The only functional change is that now BreakFalseDeps will break dependency for all register classes.
Since no additional instructions were added to the list of instructions that have false dependencies, there is no actual change yet.
In a future commit several instructions (and tests) will be added.
This is the first of multiple patches that fix bugzilla https://bugs.llvm.org/show_bug.cgi?id=33869
Most of the patches are intended at refactoring the existent code.
Additional relevant reviews:
https://reviews.llvm.org/D40331https://reviews.llvm.org/D40332https://reviews.llvm.org/D40333https://reviews.llvm.org/D40334
Differential Revision: https://reviews.llvm.org/D40330
Change-Id: Icaeb75e014eff96a8f721377783f9a3e6c679275
llvm-svn: 323087
Summary:
For example, VSQRTSDZr and VSQRTSSZr were missing the predicate.
Also fix braces indentation and braces for consistency.
Reviewers: craig.topper, RKSimon
Suscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41983
llvm-svn: 322478
Extend vXi1 conditions of vXi8/vXi16 selects even before type legalization gets a chance to split wide vectors. Previously we would only extend 128 and 256 bit vectors. But if we start with a 512 bit vector or wider that needs to be split we wouldn't extend until after the split had taken place. By extending early we improve the results of type legalization.
Don't widen condition of 128/256 bit vXi16/vXi8 selects when we have BWI but not VLX. We can still use a mask register by widening the select to 512-bits instead. This is similar to what we do for compares already.
llvm-svn: 322450
If the index is v2i64 we can use the scatter instruction that has v4i32/v4f32 data register, v2i64 index, and v2i1 mask. Similar was already done for gather.
Implement custom widening for v2i32 data to remove the code that reverses type legalization during lowering.
llvm-svn: 322254
The pattern was this
def : Pat<(i32 (zext (i8 (bitconvert (v8i1 VK8:$src))))),
(MOVZX32rr8 (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS VK8:$src, GR32)), sub_8bit))>, Requires<[NoDQI]>;
but if you just let (i32 (zext X)) match byte itself you'll get MOVZX32rr8. And if you let (i8 (bitconvert (v8i1 VK8:$src))) match by itself you'll get (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS VK8:$src, GR32)), sub_8bit).
So we can just let isel do the two patterns naturally.
llvm-svn: 322049
Summary:
There are few oddities that occur due to v1i1, v8i1, v16i1 being legal without v2i1 and v4i1 being legal when we don't have VLX. Particularly during legalization of v2i32/v4i32/v2i64/v4i64 masked gather/scatter/load/store. We end up promoting the mask argument to these during type legalization and then have to widen the promoted type to v8iX/v16iX and truncate it to get the element size back down to v8i1/v16i1 to use a 512-bit operation. Since need to fill the upper bits of the mask we have to fill with 0s at the promoted type.
It would be better if we could just have the v2i1/v4i1 types as legal so they don't undergo any promotion. Then we can just widen with 0s directly in a k register. There are no real v4i1/v2i1 instructions anyway. Everything is done on a larger register anyway.
This also fixes an issue that we couldn't implement a masked vextractf32x4 from zmm to xmm properly.
We now have to support widening more compares to 512-bit to get a mask result out so new tablegen patterns got added.
I had to hack the legalizer for widening the operand of a setcc a bit so it didn't try create a setcc returning v4i32, extract from it, then try to promote it using a sign extend to v2i1. Now we create the setcc with v4i1 if the original setcc's result type is v2i1. Then extract that and don't sign extend it at all.
There's definitely room for improvement with some follow up patches.
Reviewers: RKSimon, zvi, guyblank
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41560
llvm-svn: 321967
Without this we allow "vmovd %rax, %xmm0", but not "vmovd %rax, %xmm16"
This exists due to continue a silly bug where really old versions of the GNU assembler required movd instead of movq on these instructions. This compatibility hack then crept forward to avx version too, but we didn't propagate it to avx512.
llvm-svn: 321903
We can use zmm move with zero masking for this. We already had patterns for using a masked move, but we didn't check for the zero masking case separately.
llvm-svn: 321612
We end up using an i8 load via an isel pattern from v8i1 anyway. This just makes it more explicit. This seems to improve codgen in some cases and I'd like to kill off some of the load patterns.
llvm-svn: 321598
Nikita Popov