We have some long-standing missing shuffle optimizations that could
use this transform via VectorCombine now:
https://bugs.llvm.org/show_bug.cgi?id=35454
(and we still don't get that case in the backend either)
This function is apparently templated because there's existing code
in IR that treats mask values as unsigned and backend code that
treats masks values as signed.
The mask values are not endian-dependent (as shown by the existing
bitcast transform from DAGCombiner).
Differential Revision: https://reviews.llvm.org/D76508
Summary:
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: dylanmckay, sdardis, nemanjai, hiraditya, kbarton, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76551
createPSADBW uses SplitsOpsAndApply so should be able to handle
any size.
Restrict the extract result type to i32 or i64 since that's what
we have coverage for today and probably matches what the
isSimple() check gave us before.
Differential Revision: https://reviews.llvm.org/D76560
This patch attempts to more accurately model the reduction of
power of 2 vectors of types we natively support. This takes into
account the narrowing of vectors that occur as we go from 512
bits to 256 bits, to 128 bits. It also takes into account the use
of wider elements in the shuffles for the first 2 steps of a
reduction from 128 bits. And uses a v8i16 shift for the final step
of vXi8 reduction.
The default implementation uses the legalized type for the arithmetic
for all levels. And uses the single source permute cost of the
legalized type for all levels. This penalizes things like
lack of v16i8 pshufb on pre-sse3 targets and the splitting and
joining that needs to be done for integer types on AVX1. We never
need v16i8 shuffle for a reduction and we only need split AVX1 ops
when type the type wide and needs to be split. I think we're still
over costing splits and joins for AVX1, but we're closer now.
I've also removed all pairwise special casing because I don't
think we ever want to generate that on X86. I've also adjusted
the add handling to more accurately account for any type splitting
that occurs before we reach a legal type.
Differential Revision: https://reviews.llvm.org/D76478
SplitsOpsAndApply will take care of any needed splitting correctly.
All that we need to check is that the vector element count is a
power of 2.
Differential Revision: https://reviews.llvm.org/D76558
We often widen xmm/ymm vectors to ymm/zmm by insertion into an undef base vector. By letting getTargetShuffleAndZeroables track the undef elts we can help avoid a lot of unnecessary cross-lane shuffles.
Fixes PR44694
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.
-fuse-init-array is now the CC1 default but TargetLoweringObjectFileELF::UseInitArray still defaults to false.
The following two unknown OS target triples continue using .ctors/.dtors because InitializeELF is not called.
clang -target i386 -c a.c
clang -target x86_64 -c a.c
This cleanup fixes this as a bonus.
X86SpeculativeLoadHardeningPass::tracePredStateThroughCall can call
MCContext::createTempSymbol before TargetLoweringObjectFileELF::Initialize().
We need to call TargetLoweringObjectFileELF::Initialize() ealier.
test/CodeGen/X86/speculative-load-hardening-indirect.ll
Differential Revision: https://reviews.llvm.org/D71360
UseInitArray is now the CC1 default but TargetLoweringObjectFileELF::UseInitArray still defaults to false.
The following two unknown OS target triples continue using .ctors/.dtors because InitializeELF is not called.
clang -target i386 -c a.c
clang -target x86_64 -c a.c
This cleanup fixes this as a bonus.
Differential Revision: https://reviews.llvm.org/D71360
The combine tries to put the broadcast in either the integer or
fp domain to match the bitcast domain. But we can only do this
if the broadcast size is 32 or larger.
Previously we multiplied the cost for the table entries by the number of splits needed. But that implies that each split goes through a reduction to scalar independently. I think what really happens is that the we AND/OR the split pieces until we're down to a single value with a legal type and then do special reduction sequence on that.
So to model that this patch takes the number of splits minus one multiplied by the cost of a AND/OR at the legal element count and adds that on top of the table lookup.
Differential Revision: https://reviews.llvm.org/D76400
Summary:
This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790
Reviewers: courbet
Subscribers: jholewinski, arsenm, dschuff, jyknight, sdardis, nemanjai, jvesely, nhaehnle, sbc100, jgravelle-google, hiraditya, aheejin, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D76348
Under certain circumstances we'll end up in the position where the negated shift amount will get truncated to the type specified getScalarShiftAmountTy(), so we need to test for a truncated version of the shift amount as well.
This allows us to remove half of the remaining patterns tested for by X86ISelLowering's combineOrShiftToFunnelShift.
This fails an assert because the type is marked in the calling
convention td file as needing promotion, but the code doesn't know
how to do it.
It also much more complicated because we try to pass these in
xmm/ymm/zmm registers. As of a few weeks ago we do this promotion
from getRegisterTypeForCallingConv before the td file generated
code gets involved.
Add optional support for opt-in partial reduction cases by providing an optional partial mask to indicate which elements have been extracted for the scalar reduction.
We currently only ever use this for lowering constant uniform values (shift/rotate by immediate) so we can safely enable it by default (it treats the undef bits as zero when extracting constants).
This is necessary for an upcoming patch that will use SimplifyDemandedBits more aggressively on funnel shift amounts and causes regressions in vXi64 constant without it.
alignBranches is X86 specific, change the name in a
more general one since other target can do some state
chang before and after emitting the instruction.
Now that D75203 has landed and baked for a few days, extend the basic approach to prefix padding as well. The patch itself is fairly straight forward.
For the moment, this patch adds the functional support and some basic testing there of, but defaults to not enabling prefix padding. I want to be able to phrase a separate patch which adds the target specific reasoning and test it cleanly. I haven't decided whether I want to common it with the nop logic or not.
Differential Revision: https://reviews.llvm.org/D75300
Merge the INSERT_VECTOR_ELT/SCALAR_TO_VECTOR and PINSRW/PINSRB shuffle mask paths - they both do the same thing (find source vector + handle implicit zero extension). The PINSRW/PINSRB path also handled in the insertion of zero case which needed to be added to the general case as well.
This is a little more complicated than I'd like it to be. We have
to manually match a trunc+srl+load pattern that generic DAG
combine won't do for us due to isTypeDesirableForOp.
If we're extracting the 0'th index of a v16i8 vector we're better off using MOVD than PEXTRB, unless we're storing the value or we require the implicit zero extension of PEXTRB.
The biggest perf diff is on SLM targets where MOVD (uops=1, lat=3 tp=1) is notably faster than PEXTRB (uops=2, lat=5, tp=4).
This matches what we already do for PEXTRW.
Differential Revision: https://reviews.llvm.org/D76138
Back in D42616, we switched our default nop length from 15 to 10 bytes because some platforms have painful decode stalls when encountering multiple instruction prefixes. (10 byte long nops come from the fact that prefixes are used to pad after 8 bytes, and some platforms have issues w/more than two prefixes.)
Based on Agner's guides, it appears to be the case that modern Intel (SandyBridge and later) can decode an arbitrary number of prefixes without issue. Intel's guide only provides up to 9 bytes; I read that as providing a safe default for all their chips. Older chips and Atom series have serious decode stalls. I can't find a conclusive reference beyond those two.
Differential Revision: https://reviews.llvm.org/D75945
For context, the proposed RISC-V bit manipulation extension has a subset
of instructions which require one of two SubtargetFeatures to be
enabled, 'zbb' or 'zbp', and there is no defined feature which both of
these can imply to use as a constraint either (see comments in D65649).
AssemblerPredicates allow multiple SubtargetFeatures to be declared in
the "AssemblerCondString" field, separated by commas, and this means
that the two features must both be enabled. There is no equivalent to
say that _either_ feature X or feature Y must be enabled, short of
creating a dummy SubtargetFeature for this purpose and having features X
and Y imply the new feature.
To solve the case where X or Y is needed without adding a new feature,
and to better match a typical TableGen style, this replaces the existing
"AssemblerCondString" with a dag "AssemblerCondDag" which represents the
same information. Two operators are defined for use with
AssemblerCondDag, "all_of", which matches the current behaviour, and
"any_of", which adds the new proposed ORing features functionality.
This was originally proposed in the RFC at
http://lists.llvm.org/pipermail/llvm-dev/2020-February/139138.html
Changes to all current backends are mechanical to support the replaced
functionality, and are NFCI.
At this stage, it is illegal to combine features with ands and ors in a
single AssemblerCondDag. I suspect this case is sufficiently rare that
adding more complex changes to support it are unnecessary.
Differential Revision: https://reviews.llvm.org/D74338
We can form vpbroadcastw with a folded load.
We had patterns for i16->i32 zextload/extload, but nothing prevents
i64 from occuring.
I'd like to move this all to DAG combine to fix more cases, but
this is trivial fix to minimize test diffs when moving to a combine.
Lets us remove another SLM proc family flag usage.
This is NFC, but we should probably check whether atom/glm/knl? should be using this flag as well...
Summary:
Currently, a BoundaryAlign fragment may be inserted after the branch
that needs to be aligned to truncate the current fragment, this fragment is
unused at most of time. To avoid that, we can insert a new empty Data
fragment instead. Non-relaxable instruction is usually emitted into Data
fragment, so the inserted empty Data fragment will be reused at a high
possibility.
Reviewers: annita.zhang, reames, MaskRay, craig.topper, LuoYuanke, jyknight
Reviewed By: reames, LuoYuanke
Subscribers: llvm-commits, dexonsmith, hiraditya
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75438
For i32 and i64 cases, X86ISD::SHLD/SHRD are close enough to ISD::FSHL/FSHR that we can use them directly, we just need to account for the operand commutation for SHRD.
The i16 SHLD/SHRD case is annoying as the shift amount is modulo-32 (vs funnel shift modulo-16), so I've added X86ISD::FSHL/FSHR equivalents, which matches the generic implementation in all other terms.
Something I'm slightly concerned with is that ISD::FSHL/FSHR legality is controlled by the Subtarget.isSHLDSlow() feature flag - we don't normally use non-ISA features for this but it allows the DAG combines to continue to operate after legalization in a lot more cases.
The X86 *bits.ll changes are all affected by the same issue - we now have a "FSHR(-1,-1,amt) -> ROTR(-1,amt) -> (-1)" simplification that reduces the dependencies enough for the branch fall through code to mess up.
Differential Revision: https://reviews.llvm.org/D75748
Refines the gather/scatter cost model, but also changes the TTI
function getIntrinsicInstrCost to accept an additional parameter
which is needed for the gather/scatter cost evaluation.
This did require trivial changes in some non-ARM backends to
adopt the new parameter.
Extending gathers and truncating scatters are now priced cheaper.
Differential Revision: https://reviews.llvm.org/D75525
For pre-AVX512 targets, combine binary shuffles to X86ISD::VPERM2X128 if possible. This mainly helps optimize the blend(extract_subvector(x,1),y) pattern.
At some point soon we're going to have make a decision about when to combine AVX512 shuffles more aggressively - we bail out if there is any change in element size (to protect predicate mask merging) which means we miss out on a lot of optimizations.
isTypeDesirableForOp prevents loads from being shrunk to i16 by DAG
combine. Because of this we can't just match the broadcast and a
scalar load. So look for broadcast+truncate+load and form a
vbroadcast_load during DAG combine. This replaces what was
previously done as an isel pattern and I think fixes it so we
won't change the size of a volatile load. But my main motivation
is just to clean up our isel patterns.
Patch by Zola Bridges!
From the review:
"""
I moved these functions to X86InstrInfo.cpp, so they are available from
another pass. In addition, this is a step toward resolving the FIXME to
move this metadata to the instruction tables.
This is the final step to make these two data invariance checks
available for non-SLH passes.
The other two steps were here:
- https://reviews.llvm.org/D70283
- https://reviews.llvm.org/D75650
Tested via llvm-lit llvm/test/CodeGen/X86/speculative-load-hardening*
"""
Differential Revision: https://reviews.llvm.org/D75654
Patch by Zola Bridges!
From the review:
"""
In this changeset (https://reviews.llvm.org/D70283), I added a liveness
check everywhere the isDataInvariant* functions were used, so that I
could safely delete the checks within the function. I mistakenly left
that deletion out of the patch. The result is that the same condition is
checked twice for some instructions which is functionally fine, but not
good. This change deletes the redundant check that I intended to delete
in the last change.
This is the second of three patches that will make the data invariance
checks available for non-SLH passes and enable the FIXMEs related to
moving this metadata to the instruction tables to be resolved.
Tested via llvm-lit llvm/test/CodeGen/X86/speculative-load-hardening*
"""
Differential Revision: https://reviews.llvm.org/D75650
If we don't need the upper subvector elements of the BLENDI node then use a smaller vector size.
This causes a couple of minor regressions in insertelement-ones.ll which are more examples of PR26018; given how cheap allones generation is I don't consider that a showstopper, just an annoyance (and there's plenty of other poor codegen cases in that file).
If we're inserting a scalar that is smaller than the element
size of the final VT, the value of the extra bits doesn't matter.
Previously we any_extended in the scalar domain before inserting.
This patch changes this to use a broadcast of the original
scalar type and then a bitcast to the final type. This might
enable the use of a broadcast load.
This recovers regressions from 07d68c24aa
and 9fcd212e2f without relying on
alignment of the load.
Differential Revision: https://reviews.llvm.org/D75835
This was selecting VBROADCASTW which turned the 8-bit load into
a 16-bit load if it happened to be 2 byte aligned.
I have a plan to fix the regression with a follow up patch
which I'll post shortly.
In 172eee9c, we tried to avoid these by modelling the callee as
internally resetting the stack pointer.
However, for the majority of functions with reserved stack frames, this
would lead LLVM to emit extra SP adjustments to undo the callee's
internal adjustment. This lead us to fix the problem further on down the
pipeline in eliminateCallFramePseudoInstr. In 5b79e603d3, I added
use a heuristic to try to detect when the adjustment would be
unreachable.
This heuristic is imperfect, and when exception handling is involved, it
fails to fire. The new test is an example of this. Simply throwing an
exception with an active cleanup emits dead SP adjustments after the
throw. Not only are they dead, but if they were executed, they would be
incorrect, so they are confusing.
This change essentially reverts 172eee9c and makes the 5b79e603d3
heuristic responsible for preventing unreachable stack adjustments. This
means we may emit unreachable stack adjustments for functions using EH
with unreserved call frames, but that is not very many these days. Back
in 2016 when this change was added, we were focused on 32-bit, which we
observed to have fewer reserved frames.
Fixes PR45064
Reviewed By: hans
Differential Revision: https://reviews.llvm.org/D75712
There was a noop bitconvert in the load pattern. While there
also make all the sources refer to src_v.RC even though its the
same as _.RC, but its consistent.
The original code could create a bitcast from f64 to i64 and back
on 32-bit targets. This was only working because getBitcast was
able to fold the casts away to avoid leaving the illegal i64 type.
Now we handle the scalar case directly by broadcasting using the
scalar type as the element type. Then bitcasting to the final VT.
This works since we ensure the scalar type is the same size as
the final VT element type. No more casts to i64.
For the vector case, we cast to VT or subvector of VT. And then
do the broadcast.
I think this all matches what we generated before, just in a more
readable way.
Summary: X86 can reduce the bytes of NOP by padding instructions with prefixes to get a better peformance in some cases. So a private member function `determinePaddingPrefix` is added to determine which prefix is the most suitable.
Reviewers: annita.zhang, reames, MaskRay, craig.topper, LuoYuanke, jyknight
Reviewed By: reames
Subscribers: llvm-commits, dexonsmith, hiraditya
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75357
If we have an explicit align directive, we currently default to emitting nops to fill the space. As discussed in the context of the prefix padding work for branch alignment (D72225), we're allowed to play other tricks such as extending the size of previous instructions instead.
This patch will convert near jumps to far jumps if doing so decreases the number of bytes of nops needed for a following align. It does so as a post-pass after relaxation is complete. It intentionally works without moving any labels or doing anything which might require another round of relaxation.
The point of this patch is mainly to mock out the approach. The optimization implemented is real, and possibly useful, but the main point is to demonstrate an approach for implementing such "pad previous instruction" approaches. The key notion in this patch is to treat padding previous instructions as an optional optimization, not as a core part of relaxation. The benefit to this is that we avoid the potential concern about increasing the distance between two labels and thus causing further potentially non-local code grown due to relaxation. The downside is that we may miss some opportunities to avoid nops.
For the moment, this patch only implements a small set of existing relaxations.. Assuming the approach is satisfactory, I plan to extend this to a broader set of instructions where there are obvious "relaxations" which are roughly performance equivalent.
Note that this patch *doesn't* change which instructions are relaxable. We may wish to explore that separately to increase optimization opportunity, but I figured that deserved it's own separate discussion.
There are possible downsides to this optimization (and all "pad previous instruction" variants). The major two are potentially increasing instruction fetch and perturbing uop caching. (i.e. the usual alignment risks) Specifically:
* If we pad an instruction such that it crosses a fetch window (16 bytes on modern X86-64), we may cause the decoder to have to trigger a fetch it wouldn't have otherwise. This can effect both decode speed, and icache pressure.
* Intel's uop caching have particular restrictions on instruction combinations which can fit in a particular way. By moving around instructions, we can both cause misses an change misses into hits. Many of the most painful cases are around branch density, so I don't expect this to be too bad on the whole.
On the whole, I expect to see small swings (i.e. the typical alignment change problem), but nothing major or systematic in either direction.
Differential Revision: https://reviews.llvm.org/D75203
Previously we tried to promote these to xmm/ymm/zmm by promoting
in the X86CallingConv.td file. But this breaks when we run out
of xmm/ymm/zmm registers and need to fall back to memory. We end
up trying to create a non-sensical scalar to vector. This lead
to an assertion. The new tests in avx512-calling-conv.ll all
trigger this assertion.
Since we really want to treat these types like we do on avx2,
it seems better to promote them before the calling convention
code gets involved. Except when the calling convention is one
that passes the vXi1 type in a k register.
The changes in avx512-regcall-Mask.ll are because we indicated
that xmm/ymm/zmm types should be passed indirectly for the
Win64 ABI before we go to the common lines that promoted the
vXi1 types. This caused the promoted types to be picked up by
the default calling convention code. Now we promote them earlier
so they get passed indirectly as though they were xmm/ymm/zmm.
Differential Revision: https://reviews.llvm.org/D75154
I believe this is the correct fix for D75506 rather than disabling all commuting. We can still commute the remaining two sources.
Differential Revision:m https://reviews.llvm.org/D75526
SLH had two functions named isDataInvariant and isDataInvariantLoad that
checked whether the passed instruction was data invariant. For some instructions,
if the EFLAGS were dead then they were considered data invariant, otherwise
they were not considered data invariant.
In this patch, I extracted that EFLAGS liveness check and made it
explicit at every call to isDataInvariant and isDataInvariantLoad.
This makes the isDataInvariant function behave more generally
and preserves the liveness check behavior that SLH would like to have.
Tested via llvm-lit llvm/test/CodeGen/X86/speculative-load-hardening*
This is the first step in making these two data invariance checks
available for non-SLH passes. The second step is to move the passes from
SLH to X86InstrInfo.cpp. I'll follow up with a patch that does that.
Differential Revision: https://reviews.llvm.org/D70283
https://reviews.llvm.org/D42848 only handled CFA related cfi directives but
didn't handle CSR related cfi. The patch adds the CSR part. Basically it reuses
the framework created in D42848. For each basicblock, the patch tracks which
CSR set have been saved at its CFG predecessors's exits, and compare the CSR
set with the set at its previous basicblock's exit (The previous block is the
block laid before the current block). If the saved CSR set at its previous
basicblock's exit is larger, .cfi_restore will be inserted.
The patch also generates proper .cfi_restore in epilogue to make sure the
saved CSR set is consistent for the incoming edges of each block.
Differential Revision: https://reviews.llvm.org/D74303
If we would emit a VBROADCAST node, we can instead directly emit
a VBROADCAST_LOAD. This allows us to get rid of the special case
to use an f64 load on 32-bit targets for vXi64.
I believe there is more cleanup we can do later in this function,
but I'll do that in follow ups.
uops.info says these should be 15 cycle instructions. Uops.info also shows the 512-bit form uses port 0 and 5 for both register and memory. We had memory using 0 and 1.
Differential Revision: https://reviews.llvm.org/D75549
If we go with D75412, we no longer depend on the scalar type directly. So we don't need to avoid using i64. We already have AVX1 fallback patterns with i32 and i64 scalar types so we don't need to avoid using integer types on AVX1.
Differential Revision: https://reviews.llvm.org/D75413
Also add a DAG combine to combine different sized broadcasts from
constant pool to avoid a regression.
Differential Revision: https://reviews.llvm.org/D75412
The build_vector needs to be the only user of the data, but the
chain will likely have another use. So we can't make sure the
build_vector is the only user of the node.
X86 has several instructions which are documented as enabling interrupts exactly one instruction *after* the one which changes the SS segment register. Inserting a nop between these two instructions allows an interrupt to arrive before the execution of the following instruction which changes semantic behaviour.
The list of instructions is documented in "Table 24-3. Format of Interruptibility State" in Volume 3c of the Intel manual. They basically all come down to different ways to write to the SS register.
Differential Revision: https://reviews.llvm.org/D75359
- Remove unnecessary includes from the headers
- Fix cppcheck definition/declaration arg mismatch warnings
- Tidyup old comments (MVT usage was removed a long time ago)
- Use SmallVector::append for repeated mask entries
Summary:
Currently the boundaryalign fragment caches its size during the process
of layout and then it is relaxed and update the size in each iteration. This
behaviour is unnecessary and ugly.
Reviewers: annita.zhang, reames, MaskRay, craig.topper, LuoYuanke, jyknight
Reviewed By: MaskRay
Subscribers: hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75404
These AddToWorklist calls were added in 84cd968f75.
It's possible the SimplifyDemandedBits/SimplifyDemandedVectorElts
triggered CSE that deleted N. Detect that and avoid adding N
to the worklist.
Fixes PR45067.
This removes everything but int_x86_avx512_mask_vcvtph2ps_512 which provides the SAE variant, but even this can use the fpext generic if the rounding control is the default.
Differential Revision: https://reviews.llvm.org/D75162
MCObjectStreamer is more suitable to create fragments than
X86AsmBackend, for example, the function getOrCreateDataFragment is
defined in MCObjectStreamer.
Differential Revision: https://reviews.llvm.org/D75351
When bundle is enabled, data fragment itself has a space to emit NOP
to bundle-align instructions. The behaviour makes it impossible for
us to determine whether the macro fusion really happen when emitting
instructions. In addition, boundary-align fragment is also used to
emit NOPs to align instructions, currently using them together sometimes
makes code crazy.
Differential Revision: https://reviews.llvm.org/D75346
We already combine non extending loads with broadcasts in DAG
combine. All these patterns are picking up is the aligned extload
special case. But the only lit test we have that exercsises it is
using v8i1 load that datalayout is reporting align 8 for. That
seems generous. So without a realistic test case I don't think
there is much value in these patterns.
The code changes here are hopefully straightforward:
1. Use MachineInstruction flags to decide if FP ops can be reassociated
(use both "reassoc" and "nsz" to be consistent with IR transforms;
we probably don't need "nsz", but that's a safer interpretation of
the FMF).
2. Check that both nodes allow reassociation to change instructions.
This is a stronger requirement than we've usually implemented in
IR/DAG, but this is needed to solve the motivating bug (see below),
and it seems unlikely to impede optimization at this late stage.
3. Intersect/propagate MachineIR flags to enable further reassociation
in MachineCombiner.
We managed to make MachineCombiner flexible enough that no changes are
needed to that pass itself. So this patch should only affect x86
(assuming no other targets have implemented the hooks using MachineIR
flags yet).
The motivating example in PR43609 is another case of fast-math transforms
interacting badly with special FP ops created during lowering:
https://bugs.llvm.org/show_bug.cgi?id=43609
The special fadd ops used for converting int to FP assume that they will
not be altered, so those are created without FMF.
However, the MachineCombiner pass was being enabled for FP ops using the
global/function-level TargetOption for "UnsafeFPMath". We managed to run
instruction/node-level FMF all the way down to MachineIR sometime in the
last 1-2 years though, so we can do better now.
The test diffs require some explanation:
1. llvm/test/CodeGen/X86/fmf-flags.ll - no target option for unsafe math was
specified here, so MachineCombiner kicks in where it did not previously;
to make it behave consistently, we need to specify a CPU schedule model,
so use the default model, and there are no code diffs.
2. llvm/test/CodeGen/X86/machine-combiner.ll - replace the target option for
unsafe math with the equivalent IR-level flags, and there are no code diffs;
we can't remove the NaN/nsz options because those are still used to drive
x86 fmin/fmax codegen (special SDAG opcodes).
3. llvm/test/CodeGen/X86/pow.ll - similar to #1
4. llvm/test/CodeGen/X86/sqrt-fastmath.ll - similar to #1, but MachineCombiner
does some reassociation of the estimate sequence ops; presumably these are
perf wins based on latency/throughput (and we get some reduction of move
instructions too); I'm not sure how it affects numerical accuracy, but the
test reflects reality better now because we would expect MachineCombiner to
be enabled if the IR was generated via something like "-ffast-math" with clang.
5. llvm/test/CodeGen/X86/vec_int_to_fp.ll - this is the test added to model PR43609;
the fadds are not reassociated now, so we should get the expected results.
6. llvm/test/CodeGen/X86/vector-reduce-fadd-fast.ll - similar to #1
7. llvm/test/CodeGen/X86/vector-reduce-fmul-fast.ll - similar to #1
Differential Revision: https://reviews.llvm.org/D74851
This tries to improve the accuracy of extract/insert element costs by accounting for subvector extraction/insertion for >128-bit vectors and the shuffling of elements to/from the 0'th index.
It also adds INSERTPS for f32 types and PINSR/PEXTR costs for integer types (at the moment we assume the same cost as MOVD/MOVQ - which isn't always true).
Differential Revision: https://reviews.llvm.org/D74976
Previously this code was called into two ways, either a FrameIndexSDNode
was passed in StackSlot. Or a load node was passed in the argument
called StackSlot. This was determined by a dyn_cast to FrameIndexSDNode.
In the case of a load, we had to go find the real pointer from
operand 0 and cast the node to MemSDNode to find the pointer info.
For the stack slot case, the code assumed that the stack slot
was perfectly aligned despite not being the creator of the slot.
This commit modifies the interface to make the caller responsible
for passing all of the required information to avoid all the
guess work and reverse engineering.
I'm not aware of any issues with the original code after an
earlier commit to fix the alignment of one of the stack objects.
This is just clean up to make the code less surprising.
Summary: ParsingInlineAsm was a misleading name. These values are only set for MS-style inline assembly.
Reviewed By: rnk
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75198
This node reads the rounding control which means it needs to be ordered properly with operations that change the rounding control. So it needs to be chained to maintain order.
This patch adds a chain input and output to the node and connects it to the chain in SelectionDAGBuilder. I've update all in-tree targets to connect their chain through their lowering code.
Differential Revision: https://reviews.llvm.org/D75132
This reverts commit eee22ec3c3.
This is not the correct fix, the root cause seems to be a bug in the
stage1 host clang compiler. See https://reviews.llvm.org/D75091 for more
discussion.
Instead add it when we make the machine nodes during instruction
selections.
This makes this ISD node closer to ISD::MGATHER. Trying to see
if we remove the X86 specific ones.
Previously we emitted an fmadd and a fmadd+fneg and combined them with a shufflevector. But this doesn't follow the correct exception behavior for unselected elements so the backend can't merge them into the fmaddsub/fmsubadd instructions.
This patch restores the the fmaddsub intrinsics so we don't have two arithmetic operations. We lose out on optimization opportunity in the non-strict FP case, but I don't think this is a big loss. If someone gives us a test case we can look into adding instcombine/dagcombine improvements. I'd rather not have the frontend do completely different things for strict and non-strict.
This still has problems because target specific intrinsics don't support strict semantics yet. We also still have all of the problems with masking. But we at least generate the right instruction in constrained mode now.
Differential Revision: https://reviews.llvm.org/D74268
I'm hoping to begin improving shuffle combining across different vector sizes, but before that we must ensure that all existing getTargetShuffleInputs calls must bail if the inputs aren't the same size.
D74976 will handle larger vector types, but since SLM doesn't support AVX+ then we will always be extracting from 128-bit vectors so don't need to scale the cost.
The gather intrinsics use a floating point mask when the result
type is FP. But we call DemandedBits on the mask assuming its an
integer type. We also use integer types when we create it from
generic IR. So add a bitcast to the intrinsic path to guarantee
the integer type.
The type profile we use for the isel patterns lied about how
many operands the gather/scatter node has to skip the index
and scale operands. This allowed us to expand the baseptr
operand into base, displacement, and segment and then merge
the index and scale with them in the final instruction during
isel. This is kind of a hack that relies on isel not checking the
number of operands at all.
This commit switches to custom isel where we can manage this
directly without relying on holes in the isel checking.
Leave the gather/scatter subclasses, but make them inherit from
MemIntrinsicSDNode and delete their constructor and destructor.
This way we can still have the getIndex, getMask, etc. convenience
functions.
If a simplication occurs the operand will be added to the worklist.
But since the demanded mask was based on N, we need to make sure
we revisit N in case there are more simplifications to be done.
Returning SDValue(N, 0) as we do, only tells DAG combine that
something changed, but that won't make it add anything to the
worklist.
Found while playing around with using VEXTRACT_STORE in more cases.
But I guess this doesn't affect any of our existing tests.
We can use MOVLPS which will load 64 bits, but we need a v4f32
result type. We already have isel patterns for this.
The code here is a little hacky. We can probably improve it with
more isel patterns.
This is similar to using movd which we do for sse2 targets.
I've added a DAG combine for VEXTRACT_STORE to use SimplifyDemandedVectorElts
to clean up some artifacts from type legalization.
Similar to what do for other operations that use a subset of bits.
Allows us to remove a pattern that shrinks a load. Which was
incorrect if the load was volatile.
This moves all the logic of converting LLVM Triples to
MachO::CPU_(SUB_)TYPE from the specific target (Target)AsmBackend to
more convenient functions in lib/BinaryFormat.
This also gets rid of the separate two X86AsmBackend classes.
The previous attempt was to add it to libObject, but that adds an
unnecessary dependency to libObject from all the targets.
Differential Revision: https://reviews.llvm.org/D74808
isPrefix was added to support the patches to align branches.
it relies on a switch over instruction names.
This moves those opcodes to a new format so the information is
tablegen and we can just check for a specific value in some bits
in TSFlags instead.
I've left the other function in place for now so that the
existing patches in phabricator will still work. I'll work with
the owner to get them migrated.
At this point in the code we know that Op1 or Op2 is
all ones. Y points to the other operand. In the case that
Op2 is zero, Op1 must be all ones and Y is Op2. The OR
ORs Y into Res. But if Y is 0 the OR will be folded away by
getNode so we don't need to check for it.
The combineSelect code was casting to i64 without any check that
i64 was legal. This can break after type legalization.
It also required splitting the mmx register on 32-bit targets.
It's not clear that this makes sense. Instead switch to using
a cmov pseudo like we do for XMM/YMM/ZMM.
VK1 was being used as the output of the copy to regclass, but it
should be VK2/VK4. Shouldn't matter in practice though since
VK1/VK2/VK4/VK8/VK16 are all identicaly and just have different VTs.
The motivating case is seen in "splat4_v8f32_load_store" and based on code in PR42024:
https://bugs.llvm.org/show_bug.cgi?id=42024
(I haven't stepped through the v8i32 sibling test yet to see why that diverged.)
There are other potential improvements visible like allowing scalarization or vector
narrowing.
Differential Revision: https://reviews.llvm.org/D74909
This moves all the logic of converting LLVM Triples to
MachO::CPU_(SUB_)TYPE from the specific target (Target)AsmBackend to
more convenient functions in libObject.
This also gets rid of the separate two X86AsmBackend classes.
Differential Revision: https://reviews.llvm.org/D74808
There's a lot of old leftover code in LowerBRCOND. Especially
the detecting or AND or OR of X86ISD::SETCC nodes. Those were
needed before LegalizeDAG was changed to visit nodes before
their operands.
It also relied on reversing the output of LowerSETCC to find the
flags producing node to use for the X86ISD::BRCOND node.
Rather than using LowerSETCC this patch uses emitFlagsForSetcc to
handle the integer ISD::SETCC case. This gives the flag producer
and the comparison code to use directly. I've removed the addTest
flag and just produce a X86ISD::BRCOND and return immediately.
Floating point ISD::SETCC case is just an X86ISD::FCMP with special
care for OEQ and UNE derived from the previous code. I've left
f128 out so it will emit a test. And LowerSETCC will be called
later to produce a libcall and X86ISD::SETCC. We have combines
that can merge the test and X86ISD::SETCC.
We need to handle two cases for overflow ops. Either they are used
directly or they have a seteq 0 or setne 1 to invert the overflow.
The old code did not handle the setne 1 case, but I think some
other combines were making up for it.
If we fail to find a condition, we'll wrap an AND with 1 on the
original condition and tell emitFlagsForSetcc to emit a compare
with 0. This will pickup the LowerAndToBT and or the EmitTest case.
I kept the isTruncWithZeroHighBitsInput call, but we might be able
to fold that in to emitFlagsForSetcc.
Differential Revision: https://reviews.llvm.org/D74750
Only handle power of 2 element count for simplicity. Not sure what to do with vXf64->vXf16 fp_round to avoid double rounding
Differential Revision: https://reviews.llvm.org/D74886
We only need to split after type legalization. If we're before
we can just use a wide store and type legalization will split it.
Add a v128i1 test to exercise it post type legalization.
On some targets, like SPARC, forming overflow ops is only profitable if
the math result is used: https://godbolt.org/z/DxSmdB
This patch adds a new MathUsed parameter to allow the targets
to make the decision and defaults to only allowing it
if the math result is used. That is the conservative choice.
This patch also updates AArch64ISelLowering, X86ISelLowering,
ARMISelLowering.h, SystemZISelLowering.h to allow forming overflow
ops if the math result is not used. On those targets using the
overflow intrinsic for the overflow check only generates better code.
Reviewers: nikic, RKSimon, lebedev.ri, spatel
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D74722
LoweSELECT will detect the constant inputs and convert to scalar
selects, but we can do it directly here.
I might remove some of the code from LowerSELECT and move it to
DAG combine so doing this explicitly will make us less dependent
on it happening in lowering.
Summary:
Making `Scale` a `TypeSize` in AArch64InstrInfo::getMemOpInfo,
has the effect that all places where this information is used
(notably, TargetInstrInfo::getMemOperandWithOffset) will need
to consider Scale - and derived, Offset - possibly being scalable.
This patch adds a new operand `bool &OffsetIsScalable` to
TargetInstrInfo::getMemOperandWithOffset and fixes up all
the places where this function is used, to consider the
offset possibly being scalable.
In most cases, this means bailing out because the algorithm does not
(or cannot) support scalable offsets in places where it does some
form of alias checking for example.
Reviewers: rovka, efriedma, kristof.beyls
Reviewed By: efriedma
Subscribers: wuzish, kerbowa, MatzeB, arsenm, nemanjai, jvesely, nhaehnle, hiraditya, kbarton, javed.absar, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Jim, lenary, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D72758
This patch enables the debug entry values feature.
- Remove the (CC1) experimental -femit-debug-entry-values option
- Enable it for x86, arm and aarch64 targets
- Resolve the test failures
- Leave the llc experimental option for targets that do not
support the CallSiteInfo yet
Differential Revision: https://reviews.llvm.org/D73534
This helps this transform occur earlier so we can fold the not
with setcc. If we delay it until after type legalization we might
have introduced instructions to widen the mask if the vselect was
widened. This can prevent the not from making it to the setcc.
We could of course add more DAG combines to handle that, but
moving this earlier is easier.
D73835 will make IRBuilder no longer trivially copyable. This patch
deletes the copy constructor in advance, to separate out the breakage.
Currently, the IRBuilder copy constructor is usually used by accident,
not by intention. In rG7c362b25d7a9 I've fixed a number of cases where
functions accepted IRBuilder rather than IRBuilder &, thus performing
an unnecessary copy. In rG5f7b92b1b4d6 I've fixed cases where an
IRBuilder was copied, while an InsertPointGuard should have been used
instead.
The only non-trivial use of the copy constructor is the
getIRBForDbgInsertion() helper, for which I separated construction and
setting of the insertion point in this patch.
Differential Revision: https://reviews.llvm.org/D74693
Both of those functions only have a single caller starting
at LowerSETCC. Just handle floating point directly in LowerSETCC.
This removes the need to pass Chain and IsSignaling all the way
down.
Summary:
Many directives are unavailable, and support for others may be limited.
This first draft has preliminary support for:
- conditional directives (including errors),
- data allocation (unsigned types up to 8 bytes, and ALIGN),
- equates/variables (numeric and text),
- and procedure directives (without parameters),
as well as COMMENT, ECHO, INCLUDE, INCLUDELIB, PUBLIC, and EXTERN. Text variables (aka text macros) are expanded in-place wherever the identifier occurs.
We deliberately ignore all ml.exe processor directives.
Prominent features not yet supported:
- structs
- macros (both procedures and functions)
- procedures (with specified parameters)
- substitution & expansion operators
Conditional directives are complicated by the fact that "ifdef rax" is a valid way to check if a file is being assembled for a 64-bit x86 processor; we add support for "ifdef <register>" in general, which requires adding a tryParseRegister method to all MCTargetAsmParsers. (Some targets require backtracking in the non-register case.)
Reviewers: rnk, thakis
Reviewed By: thakis
Subscribers: kerbowa, merge_guards_bot, wuzish, arsenm, dschuff, jyknight, dylanmckay, sdardis, nemanjai, jvesely, mgorny, sbc100, jgravelle-google, hiraditya, aheejin, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, sabuasal, niosHD, jrtc27, MaskRay, zzheng, edward-jones, atanasyan, rogfer01, MartinMosbeck, brucehoult, the_o, PkmX, jocewei, jsji, Jim, s.egerton, pzheng, sameer.abuasal, apazos, luismarques, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D72680
The unseen logic diff occurs because MayFoldLoad() is defined like this:
static bool MayFoldLoad(SDValue Op) {
return Op.hasOneUse() && ISD::isNormalLoad(Op.getNode());
}
The test diffs here all seem ok to me on screen/paper, but it's hard to know
if that will lead to universally better perf for all targets. For example,
if a target implements broadcast from mem as multiple uops, we would have to
weigh the potential reduction of instructions and register pressure vs.
possible increase in number of uops. I don't know if we can make a truly
informed decision on this at compile-time.
The motivating case that I'm looking at in PR42024:
https://bugs.llvm.org/show_bug.cgi?id=42024
...resembles the diff in extract-concat.ll, but we're not going to change the
larger example there without at least 1 other fix.
Differential Revision: https://reviews.llvm.org/D74088
This allows it to work properly with masked inc/dec for avx512. Those
would have a vselect as the root node so didn't get a chance to call
combineIncDecVector.
This also simplifies the logic because we don't have to manage
the topological ordering.
Sanjay Patel