Commit Graph

9002 Commits

Author SHA1 Message Date
Matthias Springer 65ef43e288 [mlir][linalg][bufferize][NFC] Check return value of getResultBuffer
In a subsequent commit, getResultBuffer can return a "null" Value. This is the case when the returned buffer from an scf.if is not unique.

This commit is in preparation for scf.if support to keep the next commit smaller.

Differential Revision: https://reviews.llvm.org/D111927
2021-10-21 17:24:17 +09:00
Matthias Springer 9c55e718f5 [mlir][linalg][bufferize] Bufferize using PostOrder traversal
This is required for bufferization of scf::IfOp, which is added in a subsequent commit.

Some ops (scf::ForOp, TiledLoopOp) require PreOrder traversal to make sure that bbArgs are mapped before bufferizing the loop body.

Differential Revision: https://reviews.llvm.org/D111924
2021-10-21 17:21:52 +09:00
Mehdi Amini cb11ddb96c Revert "[MLIR][OpenMP] Add support for ordered construct"
This reverts commit dc2be87ecf.

Seems like this broke all the CI bots.
2021-10-21 04:53:45 +00:00
Peixin-Qiao dc2be87ecf [MLIR][OpenMP] Add support for ordered construct
This patch supports the ordered construct in OpenMP dialect following
Section 2.19.9 of the OpenMP 5.1 standard. Also lowering to LLVM IR
using OpenMP IRBduiler. Lowering to LLVM IR for ordered simd directive
is not supported yet since LLVM optimization passes do not support it
for now.

Reviewed By: kiranchandramohan, clementval, ftynse, shraiysh

Differential Revision: https://reviews.llvm.org/D110015
2021-10-21 09:16:04 +08:00
Aart Bik bd5494d127 [mlir][sparse] make index type explicit in public API of support library
The current implementation used explicit index->int64_t casts for some, but
not all instances of passing values of type "index" in and from the sparse
support library. This revision makes the situation more consistent by
using new "index_t" type at all such places  (which allows for less trivial
casting in the generated MLIR code).  Note that the current revision still
assumes that "index" is 64-bit wide. If we want to support targets with
alternative "index" bit widths, we need to build the support library different.
But the current revision is a step forward by making this requirement explicit
and more visible.

Reviewed By: wrengr

Differential Revision: https://reviews.llvm.org/D112122
2021-10-20 12:46:31 -07:00
Ahmed S. Taei a3dd4e7770 Drop transfer_read inner most unit dimensions
Add a pattern to take a rank-reducing subview and drop inner most
contiguous unit dim.
This is useful when lowering vector to backends with 1d vector types.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D111561
2021-10-20 19:27:04 +00:00
Alex Zinenko 310736e098 [mlir] fix region property generation in python bindings 2021-10-20 19:00:59 +02:00
Shraiysh Vaishay c4c7e06bd7 [MLIR][OpenMP] Shifted hint from CriticalOp to CriticalDeclareOp
According to the OpenMP 5.0 standard, names and hints of critical operation are
closely related. The following are the restrictions on them:
 - Unless the effect is as if `hint(omp_sync_hint_none)` was specified, the
   critical construct must specify a name.
 - If the hint clause is specified, each of the critical constructs with the
   same name must have a hint clause for which the hint-expression evaluates to
   the same value.

These restrictions will be enforced by design if the hint expression is a part
of the `omp.critical.declare` operation.
 - Any operation with no "name" will be considered to have
   `hint(omp_sync_hint_none)`.
 - All the operations with the same "name" will have the same hint value.

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D112134
2021-10-20 21:36:09 +05:30
Jacques Pienaar 6a99423390 [mlir] Expand prefixing to OpFormatGen
Follow up to also use the prefixed emitters in OpFormatGen (moved
getGetterName(s) and getSetterName(s) to Operator as that is most
convenient usage wise even though it just depends on Dialect). Prefix
accessors in Test dialect and follow up on missed changes in
OpDefinitionsGen.

Differential Revision: https://reviews.llvm.org/D112118
2021-10-20 07:08:37 -07:00
Nicolas Vasilache 6bb7d2474f [mlir][Linalg] Add a first vectorization pattern for conv1d in NWCxWCF format.
This revision uses the newly refactored StructuredGenerator to create a simple vectorization for conv1d_nwc_wcf.

Note that the pattern is not specific to the op and is technically not even specific to the ConvolutionOpInterface (modulo minor details related to dilations and strides).

The overall design follows the same ideas as the lowering of vector::ContractionOp -> vector::OuterProduct: it seeks to be minimally complex, composable and extensible while avoiding inference analysis. Instead, we metaprogram the maps/indexings we expect and we match against them.

This is just a first stab and still needs to be evaluated for performance.
Other tradeoffs are possible that should be explored.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D111894
2021-10-20 13:54:18 +00:00
Stella Laurenzo a897590f11 Add MLIR_INSTALL_AGGREGATE_OBJECTS and default it to ON.
* Package maintainers can opt to disable installation of these objects.
* Per discussion on https://reviews.llvm.org/D111504

Differential Revision: https://reviews.llvm.org/D112090
2021-10-19 16:14:04 -07:00
Kojo Acquah 9c62bb55f4 Implementation of `ReshapeNoopOptimization` canonicalizer.
This canonicalizer replaces reshapes of constant tensors that contain the updated shape (skipping the reshape operation).

Differential Revision: https://reviews.llvm.org/D112038
2021-10-19 16:07:34 -07:00
Mehdi Amini e2f16be599 Fix clang-tidy warnings in MLIR Python bindings (NFC) 2021-10-19 17:15:20 +00:00
Shraiysh Vaishay 10e08784ca [MLIR][OpenMP][NFC] Moved Synchronization Hint related functions
The functions are moved above the parseClauses function as they
will be used inside it to parse `hint` clause

Reviewed By: clementval

Differential Revision: https://reviews.llvm.org/D112071
2021-10-19 19:38:31 +05:30
bakhtiyar f97f946839 Canonicalize max/min operations on integers.
Reviewed By: ezhulenev

Differential Revision: https://reviews.llvm.org/D112051
2021-10-19 05:25:59 -07:00
Shraiysh Vaishay d576f45014 [MLIR][OpenMP] Added parseClauses
Code reorganized in OpenMPDialect.cpp to have all functions corresponding to an operation together.

Added parseClauses function to avoid code duplication while parsing clauses in OpenMP operations. Also added printers and verifiers for clauses, which are being used for multiple operations.

Reviewed By: kiranchandramohan, peixin

Differential Revision: https://reviews.llvm.org/D110903
2021-10-19 17:31:36 +05:30
Vladislav Vinogradov e41ebbecf9 [mlir][RFC] Refactor layout representation in MemRefType
The change is based on the proposal from the following discussion:
https://llvm.discourse.group/t/rfc-memreftype-affine-maps-list-vs-single-item/3968

* Introduce `MemRefLayoutAttr` interface to get `AffineMap` from an `Attribute`
  (`AffineMapAttr` implements this interface).
* Store layout as a single generic `MemRefLayoutAttr`.

This change removes the affine map composition feature and related API.
Actually, while the `MemRefType` itself supported it, almost none of the upstream
can work with more than 1 affine map in `MemRefType`.

The introduced `MemRefLayoutAttr` allows to re-implement this feature
in a more stable way - via separate attribute class.

Also the interface allows to use different layout representations rather than affine maps.
For example, the described "stride + offset" form, which is currently supported in ASM parser only,
can now be expressed as separate attribute.

Reviewed By: ftynse, bondhugula

Differential Revision: https://reviews.llvm.org/D111553
2021-10-19 12:31:15 +03:00
Mogball 21bb463e96 [mlir] fix bugs with NamedAttrList
- `assign` with ArrayRef was calling `append`
- `assign` with empty ArrayRef was not clearing storage

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D112043
2021-10-19 01:30:00 +00:00
Matthias Springer fd26ca4e75 [mlir][scf] Add insideMutuallyExclusiveBranches helper
This helper function checks if two given ops are in mutually exclusive branches of the same scf::IfOp.

Differential Revision: https://reviews.llvm.org/D111957
2021-10-19 09:09:07 +09:00
Matthias Springer 252386ac81 [mlir] Add enclosingOpOk parameter to properlyDominates
Differential Revision: https://reviews.llvm.org/D111959
2021-10-19 08:59:56 +09:00
not-jenni 4ada6c2aaf [mlir][tosa] Adds a canonicalization to the transpose op if the perms are a no op
Reviewed By: rsuderman

Differential Revision: https://reviews.llvm.org/D112037
2021-10-18 16:30:53 -07:00
wren romano bd0cae6d16 [mlir][sparse] Renaming variables for consistency/clarity
Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D112029
2021-10-18 15:12:03 -07:00
Aart Bik 9d1db3d4a1 [mlir][sparse] generalize sparse_tensor.convert on static/dynamic dimension sizes
This revison lifts the artificial restriction on having exact matches between
source and destination type shapes. A static size may become dynamic. We still
reject changing a dynamic size into a static size to avoid the need for a
runtime "assert" on the conversion. This revision also refactors some of the
conversion code to share same-content buffers.

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D111915
2021-10-18 13:54:03 -07:00
Andrew Young 44b22f6f40
[MLIR] Expose optional attribute parsing functions
The functionality already exists in AsmParser to parse optional ArrayAttrs and
StringAttrs, but only if they are added to a NamedAttrList.  This moves the
code to parse an optional attribute and add it to an list into a common
template, and exposes the simpler functionality of just parsing the optional
attributes.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111918
2021-10-18 11:45:43 -07:00
Eugene Zhulenev bf32bb7e05 [mlir] Update approximation range for Tanh operation
Use wider range for approximating Tanh to match results computed in Eigen with AVX.

Reviewed By: cota

Differential Revision: https://reviews.llvm.org/D112011
2021-10-18 10:57:31 -07:00
Caitlyn Cano 2ea5e7ba57 [mlir] SPIR-V: add sin, cos, log, sqrt OCL ops
Differential Revision: https://reviews.llvm.org/D111884
2021-10-18 20:48:59 +03:00
Jacques Pienaar 62bf850910 [mlir] Flipping Test dialect to prefixed form _Both
Starting with a mostly NFC change to be able to differentiate between
mechanical changes from ones that require more detailed review.

This will be used to flush out flow before flipping dialects used
outside local testing. As this dialect is not intended to be used
generally rather than in tests in core, I will not be following 2 week
staging approach here.
2021-10-18 10:00:37 -07:00
Mogball 4a5ff56b14 [mlir][docs] Fix name of get arith->LLVM patterns in docs 2021-10-18 16:04:17 +00:00
Mathieu Fehr 6d6ad76a60 [mlir][NFC] Provide accessor for TableGen record for constraints
Besides accessing the record, there is currently no way to access all possible
constraint informations, such as the base constraint of a variadic constraint
for example.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111719
2021-10-18 15:46:08 +00:00
Mathieu Fehr d78136121e [mlir] Add AnyAttrOf tablegen attribute constraint
AnyAttrOf, similar to AnyTypeOf, expects the attribute to be one of the
given attributes.
For instance, `AnyAttrOf<[I32Attr, StrAttr]>` expects either a `I32Attr`,
or a `StrAttr`.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111739
2021-10-18 15:45:25 +00:00
River Riddle 20211b3261 [mlir] Fix tsan failure in PassCrashRecovery
Don't set printOpOnDiagnostic, as this is not safe to call from a threaded context.

Differential Revision: https://reviews.llvm.org/D111752
2021-10-18 15:25:21 +00:00
River Riddle a77cd55dea [mlir] Add support for specifying printing flags when adding an op to a Diagnostic
This removes edge cases where the default flags we want to use
during printing (e.g. local scope, eliding attributes, etc.)
get missed/dropped.

Differential Revision: https://reviews.llvm.org/D111761
2021-10-18 15:07:55 +00:00
Ahmed Taei b0c4aaff24 Allow only valid vector.shape_cast transitive folding
When folding A->B->C => A->C only accept A->C that is valid shape cast

Reviewed By: ThomasRaoux, nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111473
2021-10-18 07:57:55 -07:00
rkayaith d5429a13da [mlir][python] Add 'loc' property to ops
Add a read-only `loc` property to Operation and OpView

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D111972
2021-10-18 16:01:12 +02:00
William S. Moses 40b9c39db1 [MLIR][LLVM] Add memset intrinsic
Add memset intrinsic into LLVM dialect

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D111906
2021-10-16 18:20:48 -04:00
Chris Lattner ecbee4804d [Builders.h] Silence a warning by adding a cast.
The no-result version of createOrFold calls 'tryFold' but
ignores the result since it doesn't matter what it produced.
Explicitly cast to void to silence this warning:

../llvm/mlir/include/mlir/IR/Builders.h:454:5: warning: ignoring return value of function declared with 'nodiscard' attribute [-Wunused-result]
    tryFold(op.getOperation(), unused);
    ^~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~

Differential Revision: https://reviews.llvm.org/D111951
2021-10-16 13:30:28 -07:00
Mehdi Amini d0d991cd23 Improve fatal error message when an Attribute or Type wasn't initialized by a dialect (NFC)
The existing message hints that the dialect may not be loaded, but there
is also the possibility that the dialect was loaded and the initialize()
method didn't include the Type/Attribute.
2021-10-16 20:19:35 +00:00
Kazu Hirata 939a808670 Use llvm::is_contained (NFC) 2021-10-16 07:52:21 -07:00
Matthias Springer e7bb8dd929 [mlir][linalg][bufferize] Relax rules for extract_slice/insert_slice matching
The rules were too restrictive, causing out-of-place bufferization when the result of two ExtractSliceOp is fed into an InsertSliceOp.

Differential Revision: https://reviews.llvm.org/D111861
2021-10-16 17:08:47 +09:00
Groverkss 52d6c5df85 [MLIR] Generalize Affine dependence analysis using Affine Relations
This patch removes code very specific to affine dependence analysis and
refactors it as a FlatAfffineRelation.

A FlatAffineRelation represents a set of ordered pairs (domain -> range) where
"domain" and "range" are tuples of identifiers. These relations are used to
represent an "access relation" for memory access on a memref.  An access
relation maps elements of an iteration domain to the element(s) of an array
domain accessed by that iteration of the associated statement through some
array reference.  The dependence relation representing the dependence
constraints between two memory accesses can be built by composing the access
relation of the destination access by the inverse of the access relation of
source access.

This patch does not change the functionality of the existing dependence
analysis in checkMemrefAccessDependence, but refactors it to use
FlatAffineRelations to deduplicate code and enable code reuse for future
development of features like scheduling, value-based dependence analysis, etc.

Reviewed By: bondhugula

Differential Revision: https://reviews.llvm.org/D110563
2021-10-16 12:16:42 +05:30
Jacques Pienaar 965ec6dbe7 [mlir] Add folder for shape.add 2021-10-15 17:30:17 -07:00
Aart Bik e9b1c974be [mlir][sparse] run less combinations of SpMM in test (to reduce runtime)
This revision also adds a few passes to the sparse compiler part to unify the transformation sequence with all other paths we currently use.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111900
2021-10-15 16:04:01 -07:00
Geoffrey Martin-Noble efc6fe963c [MLIR][TOSA] Drop "OnTensors" suffix
This is the only lowering to Linalg Tosa has, so it's needlessly
verbose. Likely this was a carry over from IREE's usage where we
originally lowered to linalg on buffers (the only linalg that existed at
the time), so the everything on tensors needed the suffix. We're dropping
it in IREE also, having transitioned entirely to using Linalg on
tensors.

Reviewed By: sjarus

Differential Revision: https://reviews.llvm.org/D111911
2021-10-15 16:01:19 -07:00
Aart Bik b24788abd8 [mlir][sparse] implement sparse tensor init operation
Next step towards supporting sparse tensors outputs.
Also some minor refactoring of enum constants as well
as replacing tensor arguments with proper buffer arguments
(latter is required for more general sizes arguments for
the sparse_tensor.init operation, as well as more general
spares_tensor.convert operations later)

Reviewed By: wrengr

Differential Revision: https://reviews.llvm.org/D111771
2021-10-15 09:33:16 -07:00
Nicolas Vasilache 60802715d1 [mlir][vector] NFC - Refactor and extract a helper StructuredGenerator class
Differential Revision: https://reviews.llvm.org/D111893
2021-10-15 16:01:59 +00:00
Andrew Young 844706701e
[MLIR] Add `KeywordOrString` handling to AsmParser
This adds a new parser and printer for text which may be a keyword or a
string. When printing, it will attempt to print the text as a keyword,
but if it has any special or non-printable characters, it will be
printed as an escaped string.  When parsing, it will parse either a
valid keyword or a potentially escaped string. The printer allows for an
empty string, in which case it prints `""`.

This new function is used for printing the name in NamedAttributes, and
for printing the symbol name after the `@`. In CIRCT we are using this
to print module port names, which are conceptually similar to named
function arguments.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111683
2021-10-15 00:08:34 -07:00
Mogball 44610c01ae [MLIR][ODS] default-valued strings should be in quotes
`DefaultValuedAttr<StrAttr, "">` and `ConstantAttr<StrAttr, "">`
result in bugs in which TableGen will not recognize that the attribute
has a default value, because `""` is an empty TableGen string.

Strings no longer have special treatment. Instead, string values must be
wrapped in quotes: "\"foo\"". Two helpers, `DefaultValuedStrAttr` and
`ConstantStrAttr` have been added to keep code clean.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111855
2021-10-15 03:00:41 +00:00
Matthias Springer 7dd7078760 [mlir][linalg][bufferize] Handle scf::ForOp correctly in bufferizesToMemoryRead
From the perspective of analysis, scf::ForOp is treated as a black box. Basic block arguments do not alias with their respective OpOperands on the ForOp, so they do not participate in conflict analysis with ops defined outside of the loop.

However, bufferizesToMemoryRead and bufferizesToMemoryWrite on the scf::ForOp itself are used to determine how the scf::ForOp interacts with its surrounding ops.

Differential Revision: https://reviews.llvm.org/D111775
2021-10-15 11:24:21 +09:00
Matthias Springer d3cb6bf2d4 [mlir][linalg][bufferize] Rewrite conflict detection
For each memory read, follow SSA use-def chains to find the op that produces the data being read (i.e., the most recent write). A memory write to an alias is a conflict if it takes places after the "most recent write" but before the read.

This CL introduces two main changes:
* There is a concise definition of a conflict. Given a piece of IR with InPlaceSpec annotations and a computes alias set, it is easy to compute whether this program has a conflict. No need to consider multiple cases such as "read of operand after in-place write" etc.
* No need to check for clobbering.

Differential Revision: https://reviews.llvm.org/D111287
2021-10-15 10:31:02 +09:00
Jacques Pienaar 65c9907c80 [mlir][ods] Enable emitting getter/setter prefix
Allow emitting get & set prefix for accessors generated for ops. If
enabled, then the argument/return/region name gets converted from
snake_case to UpperCamel and prefix added. The attribute also allows
generating both the current "raw" method along with the prefix'd one to
make it easier to stage changes.

The option is added on the dialect and currently defaults to existing
raw behavior. The expectation is that the staging where both are
generated would be short lived and so optimized to keeping the changes
local/less invasive (it just generates two functions for each accessor
with the same body - most of these internally again call a helper
function). But generation can be optimized if needed.

I'm unsure about OpAdaptor classes as there it is all get methods (it is
a named view into raw data structures), so prefix doesn't add much.

This starts with emitting raw-only form (as current behavior) as
default, then one can opt-in to raw & prefixed, then just prefixed. The
default in OpBase will switch to prefixed-only to be consistent with
MLIR style guide. And the option potentially removed later (considered
enabling specifying prefix but current discussion more pro keeping it
limited and stuck with that).

Also add more explicit checking for pruned functions to avoid emitting
where no function was added (and so avoiding dereferencing nullptr)
during op def/decl generation.

See https://bugs.llvm.org/show_bug.cgi?id=51916 for further discussion.

Differential Revision: https://reviews.llvm.org/D111033
2021-10-14 15:58:44 -07:00
Stella Laurenzo 8e54f24f89 Disable add_mlir_aggregate() debug file generation.
* Leaves it as a commented out area with a note on how to debug.
2021-10-14 14:32:27 -07:00
Mogball cb3aa49ec0 [MLIR][arith] fix references to std.constant in comments
Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D111820
2021-10-14 20:38:47 +00:00
thomasraoux afad0cdf31 [mlir][vector] Refactor linalg vectorization for reductions
Emit reduction during op vectorization instead of doing it when creating the
transfer write. This allow us to not broadcast output arguments for reduction
initial value.

Differential Revision: https://reviews.llvm.org/D111825
2021-10-14 13:37:56 -07:00
Rob Suderman 59dd418e89 [mlir][tosa] Fix tosa.cast UiToFp32 for tosa-to-linalg
Part of the arith update broke UiToFp32. Fixed the lowering and included a new
test to detect a regression.

Differential Revision: https://reviews.llvm.org/D111772
2021-10-14 11:34:10 -07:00
Nicolas Vasilache 82dd977baf [mlir][Linalg] Tighten canonicalization of InsertSliceOp that triggers infinite loop
I am unclear this is reproducible with correct IR but atm the verifier for InsertSliceOp
is not powerful enough and this triggers an infinite loop that is worth fixing independently.

Differential Revision: https://reviews.llvm.org/D111812
2021-10-14 15:26:03 +00:00
Nicolas Vasilache 0eeaad3012 [mlir][Linalg] Fix insertion point in comprehensive bufferization 2021-10-14 15:24:09 +00:00
Tobias Gysi 3f335ffffe [mlir][linalg] Fix FusionOnTensors header and make local method static (NFC).
Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111798
2021-10-14 14:09:38 +00:00
Nicolas Vasilache 012c0cc7c3 [mlir] NFC - Avoid unused symbol in opt mode. 2021-10-14 11:26:33 +00:00
Alex Zinenko 18fbd5fe34 [mlir][python] Better support for variadic regions in Python bindings
Improve support for variadic regions in ODS-generated operation view classes.
In particular, make generated constructors take an extra argument that
specifies the number of variadic regions if the operation has them. Previously,
there was no mechanism to specify a non-zero number of variadic regions. Also
generate named accessors to regions.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D111783
2021-10-14 13:15:13 +02:00
Alex Zinenko a04c0b7ed2 [mlir][python] Fix MemRefType IsAFunction in Python bindings
MemRefType was using a wrong `isa` function in the bindings code, which
could lead to invalid IR being constructed. Also run the verifier in
memref dialect tests.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111784
2021-10-14 13:12:37 +02:00
Uday Bondhugula 05fb26062c [MLIR] Fix assert crash when an unregistered dialect op is encountered
Fix assert crash when an unregistered dialect op is encountered during
parsing and `-allow-unregistered-dialect' isn't on. Instead, emit an
error.

While on this, clean up "registered" vs "loaded" on `getDialect()` and
local clang-tidy warnings.

https://llvm.discourse.group/t/assert-behavior-on-unregistered-dialect-ops/4402

Differential Revision: https://reviews.llvm.org/D111628
2021-10-14 15:43:53 +05:30
Tobias Gysi a8f69be61f [mlir][linalg] Expose flag to control nofold attribute when padding.
Setting the nofold attribute enables packing an operand. At the moment, the attribute is set by default. The pack introduces a callback to control the flag.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111718
2021-10-14 10:07:07 +00:00
Tobias Gysi eaa52750ce [mlir][linalg] Verify every LinalgOp has a body.
After removing the last LinalgOps that have no region attached we can verify there is a region. The patch performs the following changes:
- Move the SingleBlockImplicitTerminator trait further up the the structured op base class.
- Adapt the LinalgOp verification since the trait only check if there is 0 or 1 block.
- Introduce a getBlock method on the LinalgOp interface.
- Access the LinalgOp body using either getBlock() or getBody() if the concrete operation type is known.

This patch is a follow up to https://reviews.llvm.org/D111233.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111393
2021-10-14 09:08:39 +00:00
Stella Laurenzo fe6d9937b3 [mlir] Ability to build CAPI dylibs from out of tree projects against installed LLVM.
* Incorporates a reworked version of D106419 (which I have closed but has comments on it).
* Extends the standalone example to include a minimal CAPI (for registering its dialect) and a test which, from out of tree, creates an aggregate dylib and links a little sample program against it. This will likely only work today in *static* MLIR builds (until the TypeID fiasco is finally put to bed). It should work on all platforms, though (including Windows - albeit I haven't tried this exact incarnation there).
* This is the biggest pre-requisite to being able to build out of tree MLIR Python-based projects from an installed MLIR/LLVM.
* I am rather nauseated by the CMake shenanigans I had to endure to get this working. The primary complexity, above and beyond the previous patch is because (with no reason given), it is impossible to export target properties that contain generator expressions... because, of course it isn't. In this case, the primary reason we use generator expressions on the individual embedded libraries is to support arbitrary ordering. Since that need doesn't apply to out of tree (which import everything via FindPackage at the outset), we fall back to a more imperative way of doing the same thing if we detect that the target was imported. Gross, but I don't expect it to need a lot of maintenance.
* There should be a relatively straight-forward path from here to rebase libMLIR.so on top of this facility and also make it include the CAPI.

Differential Revision: https://reviews.llvm.org/D111504
2021-10-13 18:45:55 -07:00
wren romano 5167c36ab4 [mlir][sparse] Misc code cleanup
Depends On D111763

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D111766
2021-10-13 16:39:29 -07:00
wren romano 63d4fc9483 [mlir][sparse] Factoring out helper functions for generating constants
Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D111763
2021-10-13 16:19:55 -07:00
Aart Bik a652e5b53a [mlir][sparse] emergency fix after constant -> arith.constant change
Reviewed By: Mogball

Differential Revision: https://reviews.llvm.org/D111743
2021-10-13 10:26:17 -07:00
Aart Bik 35517a251d [mlir][sparse] add init sparse tensor operation
This is the first step towards supporting general sparse tensors as output
of operations. The init sparse tensor is used to materialize an empty sparse
tensor of given shape and sparsity into a subsequent computation (similar to
the dense tensor init operation counterpart).

Example:
  %c = sparse_tensor.init %d1, %d2 : tensor<?x?xf32, #SparseMatrix>
  %0 = linalg.matmul
    ins(%a, %b: tensor<?x?xf32>, tensor<?x?xf32>)
    outs(%c: tensor<?x?xf32, #SparseMatrix>) -> tensor<?x?xf32, #SparseMatrix>

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D111684
2021-10-13 09:47:56 -07:00
xndcn 8c1553f0d7 [mlir][spirv] Add memory semantics verify for atomic operations
Differential Revision: https://reviews.llvm.org/D111510
2021-10-14 00:00:55 +08:00
Alex Zinenko 2b55e14384 [mlir] fix python bindings cmake 2021-10-13 17:29:45 +02:00
Alex Zinenko 7fd6f40dbd [mlir][python] Add custom constructor for memref load
The type can be inferred trivially, but it is currently done as string
stitching between ODS and C++ and is not easily exposed to Python.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111712
2021-10-13 17:11:02 +02:00
thomasraoux cc83c2444f [mlir][vector] Add canonicalization extract + splat
Make canonicalization working on broadcast also work on splat op.

Differential Revision: https://reviews.llvm.org/D111690
2021-10-13 08:08:46 -07:00
Alex Zinenko 90a6c3c2e4 [mlir] Fix typos in the Python bindings doc 2021-10-13 14:40:49 +02:00
Alex Zinenko 78f2dae00d [mlir][python] Provide some methods and properties for API completeness
When writing the user-facing documentation, I noticed several inconsistencies
and asymmetries in the Python API we provide. Fix them by adding:

- the `owner` property to regions, similarly to blocks;
- the `isinstance` method to any class derived from `PyConcreteAttr`,
  `PyConcreteValue` and `PyConreteAffineExpr`, similar to `PyConcreteType` to
  enable `isa`-like calls without having to handle exceptions;
- a mechanism to create the first block in the region as we could only create
  blocks relative to other blocks, with is impossible in an empty region.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D111556
2021-10-13 14:30:55 +02:00
Mehdi Amini 6f5c685839 Avoid one branch in Operation::getOpOperand()/getOperand() (NFC)
Skip the check on "hasOperandStorage" since the array will be indexed anyway.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111696
2021-10-13 09:07:06 +00:00
Jacques Pienaar e67cbbef03 [mlir][python] Expose CallSiteLoc Python side
This exposes creating a CallSiteLoc with a callee & list of frames for
callers. Follows the creation approach in C++ side where a list of
frames may be provided.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D111670
2021-10-13 10:25:40 +02:00
Matthias Springer c161084723 [mlir] Make OpResult usable with DenseSet etc.
`DenseSet<Value>` works but `DenseSet<OpResult>` currently results in a compile error.

Differential Revision: https://reviews.llvm.org/D111694
2021-10-13 15:58:08 +09:00
Stella Laurenzo 3d92722f74 [mlir][python] Add a warning for cmake version < 3.15.
As discussed on discord, we have never actually been able to build with the project-wide published min version of 3.14.3. The buildbot that tests the Python configuration is currently pinned to 3.19.1, and there are a number of non-version/policy controlled features that Python building relies on that makes it unreliable with older versions. Some of the issues are pretty fundamental and I don't know how to do them on the older version. I think that, as an optional feature, at least advertising the PSA as in this patch is a good middle ground until the next project-wide CMake version bump.

Also moves setup logic to a macro so that everyone can use it.
2021-10-12 21:13:54 -07:00
Mogball a54f4eae0e [MLIR] Replace std ops with arith dialect ops
Precursor: https://reviews.llvm.org/D110200

Removed redundant ops from the standard dialect that were moved to the
`arith` or `math` dialects.

Renamed all instances of operations in the codebase and in tests.

Reviewed By: rriddle, jpienaar

Differential Revision: https://reviews.llvm.org/D110797
2021-10-13 03:07:03 +00:00
Weiwei Li c0a6381e49 [mlir][SPIRVToLLVM] Solve ExecutionModeOp redefinition and add OpTypeSampledImage into SPV_Type
1. To avoid two ExecutionModeOp using the same name, adding the value of execution mode in name when converting to LLVM dialect.
2. To avoid syntax error in spv.OpLoad, add OpTypeSampledImage into SPV_Type.

Reviewed by:antiagainst

Differential revision:https://reviews.llvm.org/D111193
2021-10-13 10:03:25 +08:00
Matthias Springer 441485cc3d [mlir][linalg][bufferize][NFC] Remove InSpaceSpec from bufferizesToMemoryWrite
Move functionality into a separate function `isInplaceMemoryWrite`.

Differential Revision: https://reviews.llvm.org/D111040
2021-10-13 09:51:22 +09:00
Matthias Springer 0ee5323189 [mlir][linalg][bufferize][NFC] Simplify getAliasingOpOperand signature
Differential Revision: https://reviews.llvm.org/D110982
2021-10-13 09:48:28 +09:00
Matthias Springer 0292236d0c [mlir][linalg][bufferize][NFC] Simplify allocateBuffersForResults
Reuse `getResultBuffer`.

Differential Revision: https://reviews.llvm.org/D110941
2021-10-13 09:32:05 +09:00
Matthias Springer 7c3a8108b3 [mlir][linalg][bufferize] Put buffer copying in separate function
This is to avoid code duplication.

Differential Revision: https://reviews.llvm.org/D110940
2021-10-13 09:24:56 +09:00
Matthias Springer fd31620420 [mlir][linalg][bufferize][NFC] Conflict detection funcs take OpOperand and OpResult
By doing so, it is not necessary to get the OpOperand a second time via
getAliasingOpOperand. Also, code slightly more readable because we do
not have to deal with Optional<> return value.

Differential Revision: https://reviews.llvm.org/D110918
2021-10-13 09:17:31 +09:00
Matthias Springer a6ddb7cec5 [mlir][linalg][bufferize][NFC] Reduce code duplication around bufferizableInPlaceAnalysis
Differential Revision: https://reviews.llvm.org/D111380
2021-10-13 09:08:58 +09:00
thomasraoux aa71f487f3 [mlir] update new linalg vectorization tests after vectorization fix 2021-10-12 16:10:30 -07:00
thomasraoux 7c97e328b3 [mlir][linalg] Fix generic reduction vectorization
We shouldn't broadcast the original value when doing reduction. Instead
we compute the reduction and then combine it with the original value.

Differential Revision: https://reviews.llvm.org/D111666
2021-10-12 15:46:04 -07:00
Diego Caballero eeb09fd646 [mlir][Linalg] Enable vectorization of 'mul', 'and', 'or' and 'xor' reductions
This patch adds support for vectorizing 'mul', 'and', 'or' anx 'xor' reductions
to Linalg.

Reviewed By: pifon2a, ThomasRaoux, aartbik

Differential Revision: https://reviews.llvm.org/D111565
2021-10-12 21:08:23 +00:00
Diego Caballero 5c1d356c18 [mlir][Linalg] Enable vectorization of explicit broadcasts
This patch teaches `isProjectedPermutation` and `inverseAndBroadcastProjectedPermutation`
utilities to deal with maps representing an explicit broadcast, e.g., (d0, d1) -> (d0, 0).
This extension is needed to enable vectorization of such explicit broadcast in Linalg.

Reviewed By: pifon2a, nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111563
2021-10-12 21:08:22 +00:00
Rob Suderman 95e4b71519 [mlir][tosa] Fix tosa average_pool2d to linalg type issue
Average pool assumed the same input/output type. Result type for integers
is always an i32, should be updated appropriately.

Reviewed By: GMNGeoffrey

Differential Revision: https://reviews.llvm.org/D111590
2021-10-12 13:09:21 -07:00
Jacques Pienaar 04d76d3694 [mlir][python] Add nameloc getter
Expose the nameloc getter to Python API.

Differential Revision: https://reviews.llvm.org/D111663
2021-10-12 12:45:57 -07:00
Tobias Gysi 63638ed181 [mlir][linalg] Enable the vector transfer lowering by default.
Adapt CodegenStartegy to used the vector transfer lowering patterns by default.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111649
2021-10-12 18:11:05 +00:00
Benjamin Kramer f67d57c95f [mlir][Shape] Add a pattern to turn extract from shape_of into tensor.dim
If I remember correctly this wasn't done previously because dim used to
be in the memref dialect.

Differential Revision: https://reviews.llvm.org/D111651
2021-10-12 19:09:21 +02:00
Aart Bik 9fc35c306c [mlir][sparse] add action enum, group enums together, few asserts
Some random changes that were hanging around in my workspace. Also,
a tiny step towards creating a header file for the sparse utils lib.

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D111589
2021-10-12 08:50:51 -07:00
Lei Zhang 519b350de0 [mlir][vector] Add folder for no-op InsertStridedSliceOp
Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111636
2021-10-12 11:41:35 -04:00
Mike Urbach 55e76c70a4 [mlir] Limit Python dependency to Development.Module when possible.
After CMake 3.18, we are able to limit the scope of the search to just
Development.Module. Searching for Development will fail in situations
where the Python libraries are not available. When possible, limit to
just Development.Module. See:
https://pybind11.readthedocs.io/en/stable/compiling.html#findpython-mode

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D111585
2021-10-12 08:31:06 -07:00
Tobias Gysi c8faeb1edd [mlir][linalg] Add switch to disable/enable vector transfer lowering.
Add a switch to code gen strategy to disable/enable the vector transfer lowering and disable it by default.

Differential Revision: https://reviews.llvm.org/D111647
2021-10-12 15:06:07 +00:00
Tobias Gysi 97d67f7438 [mlir][linalg] Add vector transfer lowering patters to codegen strategy.
Add the vector transfer patterns and introduce the max transfer rank option on the codegen strategy.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111635
2021-10-12 14:19:45 +00:00
Nicolas Vasilache b24c91fffc [mlir][Vector][Bigfix] Fix vector transfer to store lowering to insert a proper ExtractOp
Differential Revision: https://reviews.llvm.org/D111641
2021-10-12 13:28:12 +00:00
Nicolas Vasilache 0a7f81a451 mlir][Vector] Fix spuriously disabled test. 2021-10-12 12:56:40 +00:00
Nicolas Vasilache 753a67b5c9 [mlir][Linalg] Refactor and improve vectorization to add support for reduction into 0-d tensors.
This revision takes advantage of the recently added support for 0-d transfers and vector.multi_reduction that return a scalar.

Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D111626
2021-10-12 12:47:36 +00:00
Lei Zhang bdd37c9f49 [mlir][tensor] Add some folders for insert/extract slice ops
* Fold extract_slice immediately after insert_slice.
* Fold overlapping insert_slice.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D111439
2021-10-12 08:40:54 -04:00
Nicolas Vasilache 0c74b12a2e [mlir][Vector] NFC - Add test to exercise lowering of vector.transfer to scf
This revision also renames and moves some tests around.

Differential Revision: https://reviews.llvm.org/D111606
2021-10-12 12:38:33 +00:00
Nicolas Vasilache 47f7938a94 [mlir][Vector] Add support for lowering 0-d transfers to load/store.
Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D111603
2021-10-12 12:35:19 +00:00
Nicolas Vasilache 67b10532c6 [mlir][Vector] Allow a 0-d for for vector transfer ops.
This revision updates the op semantics, printer, parser and verifier to allow 0-d transfers.
Until 0-d vectors are available, such transfers have a special form that transits through vector<1xt>.
This is a stepping stone towards the longer term work of adding 0-d vectors and will help significantly reduce corner cases in vectorization.

Transformations and lowerings do not yet support this form, extensions will follow.

Differential Revision: https://reviews.llvm.org/D111559
2021-10-12 11:48:42 +00:00
Nicolas Vasilache 8f1650cb65 [mlir][Linalg] NFC - Refactor vector.broadcast op verification logic and make it available as a precondition in Linalg vectorization.
Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D111558
2021-10-12 11:35:34 +00:00
Nicolas Vasilache 31270eb165 [mlir][Vector] Let vector.multi_reduction reduce down to a scalar.
vector.multi_reduction currently does not allow reducing down to a scalar.
This creates corner cases that are hard to handle during vectorization.
This revision extends the semantics and adds the proper transforms, lowerings and canonicalizations to allow lowering out of vector.multi_reduction to other abstractions all the way to LLVM.

In a future, where we will also allow 0-d vectors, scalars will still be relevant: 0-d vector and scalars are not equivalent on all hardware.

In the process, splice out the implementation patterns related to vector.multi_reduce into a new file.

Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D111442
2021-10-12 11:03:54 +00:00
Shraiysh Vaishay 7a79c6afea [mlir][OpenMP] OpenMP Synchronization Hints stored as IntegerAttr
`hint-expression` is an IntegerAttr, because it can be a combination of multiple values from the enum `omp_sync_hint_t` (Section 2.17.12 of OpenMP 5.0)

Reviewed By: ftynse, kiranchandramohan

Differential Revision: https://reviews.llvm.org/D111360
2021-10-12 11:01:19 +00:00
Vladislav Vinogradov d6296c3b00 [mlir] Allow to use vformat utility with MLIR classes
Make `raw_ostream operator<<` follow const correctness semantic,
since it is a requirement of FormatVariadic implementation.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111547
2021-10-12 13:28:32 +03:00
Caitlyn Cano c6828e0cea [mlir] Make ConversionTarget dynamic legality callbacks composable
* Change callback signature `bool(Operation *)` -> `Optional<bool>(Operation *)`
* addDynamicallyLegalOp add callback to the chain
* If callback returned empty `Optional` next callback in chain will be called

Differential Revision: https://reviews.llvm.org/D110487
2021-10-12 13:05:54 +03:00
Vladislav Vinogradov c6390f19f2 [mlir] Fix AsmPrinter for types with sub elements
Call `printType(subElemType)` instead of `os << subElemType` for them.
It allows to handle type aliases inside complex types.

As a side effect, fixed `test.int` parsing.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111536
2021-10-12 12:08:16 +03:00
Vladislav Vinogradov 505afd1e64 [mlir] Clean up boolean flags usage in LIT tests
* Call `llvm_canonicalize_cmake_booleans` for all CMake options,
  which are propagated to `lit.local.cfg` files.
* Use Python native boolean values instead of strings for such options.

This fixes the cases, when CMake variables have values other than `ON` (like `TRUE`).
This might happen due to IDE integration or due to CMake preset usage.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D110073
2021-10-12 11:44:48 +03:00
Jacques Pienaar c3dcf39554 [mlir] Restrict to requiring traits when using InferTensorType trait.
Avoids running into segfaults accidentally.

Differential Revision: https://reviews.llvm.org/D110297
2021-10-11 14:56:28 -07:00
Daniel Resnick 1760d8b36b [mlir][ODS] Support result type inference in custom assembly format
Operations that have the InferTypeOpInterface trait can now omit the return
types in their custom assembly formats.

Differential Revision: https://reviews.llvm.org/D111326
2021-10-11 14:07:56 -06:00
Aart Bik 849f016ce8 [mlir][sparse] accept affine subscripts in outer dimensions of dense memrefs
This relaxes vectorization of dense memrefs a bit so that affine expressions
are allowed in more outer dimensions. Vectorization of non unit stride
references is disabled though, since this seems ineffective anyway.

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D111469
2021-10-11 11:45:14 -07:00
Alex Zinenko bacb0cac15 [mlir] add user-level documentation for Python bindings
Until now, we only had documentation oriented towards developers of the
bindings. Provide some documentation for users of the bindings that don't want
or need to understand the inner workings.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111540
2021-10-11 17:23:00 +02:00
Uday Bondhugula b2217b36fe [MLIR] Fix affine loop unroll corner case for full unroll
Fix affine loop unroll for zero trip count loops. Add missing check.

Differential Revision: https://reviews.llvm.org/D111375
2021-10-11 10:22:24 +05:30
Amy Zhuang 5ce368cfe2 [mlir] Vectorize induction variables
1. Add support to vectorize induction variables of loops that are
   not mapped to any vector dimension in SuperVectorize pass.
2. Fix a bug in getForInductionVarOwner.

Reviewed By: dcaballe

Differential Revision: https://reviews.llvm.org/D111370
2021-10-09 12:40:24 -07:00
Mehdi Amini dda810c332 Fix a comment at call-site to match the declared parameter (NFC)
(clang-tidy warning)
2021-10-09 17:57:53 +00:00
Mehdi Amini 8c9f506d8c Disable mlir/test/mlir-cpu-runner/async-group.mlir with ASAN
This test is crashing 9 out of 10 runs in CI, but I can't reproduce
locally right now. Disabling to get the CI back to green and avoid
backsliding with more ASAN issues that would go unnoticed.
2021-10-09 03:02:53 +00:00
Reid Kleckner 89b57061f7 Move TargetRegistry.(h|cpp) from Support to MC
This moves the registry higher in the LLVM library dependency stack.
Every client of the target registry needs to link against MC anyway to
actually use the target, so we might as well move this out of Support.

This allows us to ensure that Support doesn't have includes from MC/*.

Differential Revision: https://reviews.llvm.org/D111454
2021-10-08 14:51:48 -07:00
Emilio Cota 57c56cf20c X86Vector: relax checks in rsqrt's integration test
Instead of hard-coding results for both Intel and AMD, let's relax
the checks to simplify the test while supporting both implementations.
Note that:
- If a new hardware implementation comes up in the future, it is likely
  to pass the relaxed tests, i.e. no future maintenance burden for us.
- If something terribly wrong happens (e.g. instead of rsqrt we
  execute 1/sqrt), the tests will probably catch it, since the relaxed
  tests expect low precision (e.g. rsqrt(1) != 1.0).

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D111461
2021-10-08 13:59:18 -07:00
Nikita Popov cfb53d8e6d [NFC] Make some includes explicit
Avoid relying on a number of indirect includes that currently
happen through the Hashing.h header in DenseMapInfo.h.
2021-10-08 20:34:48 +02:00
Stella Laurenzo a201829a20 Fix parsing of hex-format index dense tensor attributes.
TensorLiteralParser::getHexAttr does a isIntOrIndexOrFloat check and properly handles index elements, but TensorLiteralParser::getAttr that calls into it has a mismatched check. This just makes the checks match so that index element attrs can parse when of type tensor.

Reviewed By: rriddle

Differential Revision: https://reviews.llvm.org/D111374
2021-10-08 15:44:02 +00:00
Matthias Springer f8453ea75f [mlir][linalg][bufferize] Rewrite "write into non-writable memory" detection
The purpose of this revision is to make "write into non-writable memory" conflict detection easier to understand.

The main idea is that there is a conflict in the case of inplace bufferization if:

1. Someone writes to (an alias of) opOperand, opResult or the to-be-bufferized op writes itself.
2. And, opOperand or opResult aliases a non-writable buffer.

Differential Revision: https://reviews.llvm.org/D111379
2021-10-08 21:27:49 +09:00
Lei Zhang 4cd7ff6728 [mlir][linalg] Constant fold linalg.generic that are transposes
This commit adds a pattern to perform constant folding on linalg
generic ops which are essentially transposes. We see real cases
where model importers may generate such patterns.

Reviewed By: mravishankar

Differential Revision: https://reviews.llvm.org/D110597
2021-10-08 08:09:13 -04:00
Eugene Zhulenev e2a37bb540 [mlir] Add alignment option to constant tensor bufferization pass
Reviewed By: bkramer

Differential Revision: https://reviews.llvm.org/D111364
2021-10-08 03:17:20 -07:00
Alex Zinenko b164f23c29 [mlir][python] support taking ops instead of values in op constructors
Introduce support for accepting ops instead of values when constructing ops. A
single-result op can be used instead of a value, including in lists of values,
and any op can be used instead of a list of values. This is similar to, but
more powerful, than the C++ API that allows for implicitly casting an OpType to
Value if it is statically known to have a single result - the cast in Python is
based on the op dynamically having a single result, and also handles the
multi-result case. This allows to build IR in a more concise way:

    op = dialect.produce_multiple_results()
    other = dialect.produce_single_result()
    dialect.consume_multiple_results(other, op)

instead of having to access the results manually

    op = dialect.produce.multiple_results()
    other = dialect.produce_single_result()
    dialect.consume_multiple_results(other.result, op.operation.results)

The dispatch is implemented directly in Python and is triggered automatically
for autogenerated OpView subclasses. Extension OpView classes should use the
functions provided in ods_common.py if they want to implement this behavior.
An alternative could be to implement the dispatch in the C++ bindings code, but
it would require to forward opaque types through all Python functions down to a
binding call, which makes it hard to inspect them in Python, e.g., to obtain
the types of values.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D111306
2021-10-08 09:49:48 +02:00
Tobias Gysi 8ed2e8e04f [mlir][linalg] Retire Linalg ConvOp.
The convolution op is one of the remaining hard coded Linalg operations that have no region attached. It got obsolete due to the OpDSL convolution operations. Removing it allows us to delete specialized code and tests that are not needed for the OpDSL counterparts that rely on the standard code paths.

Test needed due to specialized implementations are removed. Tiling and fusion tests are replaced by variants using linalg.conv_2d.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D111233
2021-10-08 06:56:37 +00:00
Tobias Gysi 23800b05be [mlir][linalg] Add loop interchange to CodegenStrategy.
Add a loop interchange pass and integrate it with CodegenStrategy.

This patch depends on https://reviews.llvm.org/D110728 and https://reviews.llvm.org/D110746.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D110748
2021-10-08 06:39:22 +00:00
Tobias Gysi 1ebd197bc5 [mlir][linalg] Add generalization to CodegenStrategy.
Add a generalization pass and integrate it with CodegenStrategy.

This patch depends on https://reviews.llvm.org/D110728.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D110746
2021-10-08 06:31:19 +00:00
Matthias Springer 7dfd3bb034 [mlir][linalg][bufferize][NFC] API change of aliasesNonWritableBuffer
The function now takes a Value instead of an OpOperand.

Differential Revision: https://reviews.llvm.org/D111378
2021-10-08 14:47:29 +09:00
Matthias Springer 89b2f29d62 [mlir][linalg][bufferize] Fix/add missing case to getAliasingOpOperand
Differential Revision: https://reviews.llvm.org/D111377
2021-10-08 14:38:47 +09:00
Matthias Springer a046154057 [mlir][linalg][bufferize] Add bufferRelation to op interface
Currently supported are: BufferRelation::None, BufferRelation::Equivalent.

Differential Revision: https://reviews.llvm.org/D111376
2021-10-08 14:28:24 +09:00
Mehdi Amini 70a8d61ca1 evert "[mlir] Limit Python dependency to Development.Module when possible."
This reverts commit 7aebdfc4fc.

The build is broken with errors like:

GPUPasses.cpp:(.text.pybind11_object_init[pybind11_object_init]+0x118): undefined reference to `PyExc_TypeError'
2021-10-08 05:18:52 +00:00
Mike Urbach 7aebdfc4fc [mlir] Limit Python dependency to Development.Module when possible.
After CMake 3.18, we are able to limit the scope of the
find_package(Python3 ...) search to just Development.Module. Searching
for Development will fail in manylinux builds, and isn't necessary
since we are not embedding the Python interpreter. For more information, see:
https://pybind11.readthedocs.io/en/stable/compiling.html#findpython-mode

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D111383
2021-10-07 23:05:43 -06:00
Mehdi Amini 82cd8b81aa Fix test-rsqrt.mlir to accept AMD's approximation of rsqrt as well
These kind of function can behave differently on these X86 chips, there
isn't really "one true answer" so we'll accept both.

Also remove spurious passes and use mattr="avx" to match the instruction
used here.

Differential Revision: https://reviews.llvm.org/D111373
2021-10-08 04:24:24 +00:00
MaheshRavishankar 4281946390 [mlir][Tensor] Add ReifyRankedShapedTypeOpInterface to tensor.extract_slice.
Differential Revision: https://reviews.llvm.org/D111263
2021-10-07 17:10:35 -07:00
Amy Zhuang 5d001f58f2 [mlir] Fix a bug in Affine LICM.
Currently Affine LICM checks iterOperands and does not hoist out any
instruction containing iterOperands. We should check iterArgs instead.

Reviewed By: bondhugula

Differential Revision: https://reviews.llvm.org/D111090
2021-10-07 15:46:43 -07:00
Kiran Chandramohan 3b01cf9286
[mlir][openmp] Add an interface for Outlineable OpenMP ops
Add an interface for outlineable OpenMP operations.
This patch was initially done in fir-dev and is now needed
for the upstreaming.

Reviewed By: schweitz

Differential Revision: https://reviews.llvm.org/D111310
2021-10-07 20:53:48 +02:00
Stella Laurenzo c5f445d143 [mlir][python] Temporarily disable test for converting unsupported DenseElementsAttr types to a buffer.
* Need to investigate the proper solution to https://github.com/pybind/pybind11/issues/3336 or engineer something different.
* The attempt to produce an empty buffer_info as a workaround triggers asan/ubsan.
* Usage of this API does not arise naturally in practice yet, and it is more important to be asan/crash clean than have a solution right now.
* Switching back to raising an exception, even though that triggers terminate().
2021-10-07 11:50:57 -07:00
Stella Laurenzo 5d6d30edf8 [mlir] Extend C and Python API to support bulk loading of DenseElementsAttr.
* This already half existed in terms of reading the raw buffer backing a DenseElementsAttr.
* Documented the precise expectations of the buffer layout.
* Extended the Python API to support construction from bitcasted buffers, allowing construction of all primitive element types (even those that lack a compatible representation in Python).
* Specifically, the Python API can now load all integer types at all bit widths and all floating point types (f16, f32, f64, bf16).

Differential Revision: https://reviews.llvm.org/D111284
2021-10-07 08:42:12 -07:00
Kazu Hirata 80e39366ee [lldb, mlir] Migrate from getNumArgOperands and arg_operands (NFC)
Note that getNumArgOperands and arg_operands are considered legacy
names.  See llvm/include/llvm/IR/InstrTypes.h for details.
2021-10-07 08:29:42 -07:00
Matthias Springer 56bf688a09 [mlir][linalg][bufferize][NFC] Simplify getAliasingOpResult()
The signature of this function was confusing. Check for hasKnownBufferizationAliasingBehavior separately when needed.

Differential Revision: https://reviews.llvm.org/D110916
2021-10-07 22:41:21 +09:00
Lei Zhang 3964c1db91 [mlir][vector] Split populateVectorContractLoweringPatterns
It was bundling quite a lot of patterns that convert high-D
vector ops into low-D elementary ops. It might not be good
for all of the patterns to happen for a particular downstream
user. For example, `ShapeCastOpRewritePattern` rewrites
`vector.shape_cast` into data movement extract/insert ops.

Instead, split the entry point into multiple ones so users
can pull in patterns on demand.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D111225
2021-10-07 09:39:26 -04:00
Matthias Springer 1d24b8c603 [mlir][linalg][bufferize][NFC] Change bufferizableInPlaceAnalysis signature
Move getInplaceableOpResult() call into bufferizableInPlaceAnalysis.

Note: The only goal of this change is to make the signature of bufferizableInPlaceAnalysis smaller. (Fewer arguments.)

Differential Revision: https://reviews.llvm.org/D110915
2021-10-07 22:35:40 +09:00
Matthias Springer 6b1f653c94 [mlir][linalg][bufferize] tensor.cast may require a copy
Differential Revision: https://reviews.llvm.org/D110806
2021-10-07 22:24:05 +09:00
Eugene Zhulenev 8276ac13e9 [mlir] Add alignment attribute to memref.global
Revived https://reviews.llvm.org/D102435

Add alignment attribute to `memref.global` and propagate it to llvm global in memref->llvm lowering

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D111309
2021-10-07 06:21:57 -07:00
Adrian Kuegel 2bb208ddfd [mlir] Don't allow dynamic extent tensor types for ConstShapeOp.
ConstShapeOp has a constant shape, so its type can always be static.
We still allow it to have ShapeType though.

Differential Revision: https://reviews.llvm.org/D111139
2021-10-07 10:56:16 +02:00