TosaOp defintion had an artificial constraint that the input/output types
needed to be ranked to invoke the quantization builder. This is correct as an
unranked tensor could still be quantized.
Reviewed By: NatashaKnk
Differential Revision: https://reviews.llvm.org/D109863
We are having issues running the integration test of the sparse compiler
on AArch64 (crashing in the lib). This revision adds more assertions.
Reviewed By: jsetoain
Differential Revision: https://reviews.llvm.org/D109861
Express the input shape definitions of convolution and pooling operations in terms of the output shapes, filter shapes, strides, and dilations.
Reviewed By: shabalin, rsuderman, stellaraccident
Differential Revision: https://reviews.llvm.org/D109815
This enables the sparsification of more kernels, such as convolutions
where there is a x(i+j) subscript. It also enables more tensor invariants
such as x(1) or other affine subscripts such as x(i+1). Currently, we
reject sparsity altogether for such tensors. Despite this restriction,
however, we can already handle a lot more kernels with compound subscripts
for dense access (viz. convolution with dense input and sparse filter).
Some unit tests and an integration test demonstrate new capability.
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109783
Adds a new rewrite directive returnType that can be added at the end of an op's
argument list to explicitly specify return types.
```
(OpX $v0, $v1, (returnType "$_builder.getI32Type()"))
```
Pass in a bound value to copy its return type, or pass a native code call to
dynamically create new types.
```
(OpX $v0, $v1, (returnType $v0, (NativeCodeCall<"..."> $v1)))
```
Reviewed By: jpienaar
Differential Revision: https://reviews.llvm.org/D109472
There are two main versions of depthwise conv depending whether the multiplier
is 1 or not. In cases where m == 1 we should use the version without the
multiplier channel as it can perform greater optimization.
Add lowering for the quantized/float versions to have a multiplier of one.
Reviewed By: antiagainst
Differential Revision: https://reviews.llvm.org/D108959
This revision fixes a corner case that could appear due to incorrect insertion point behavior in comprehensive bufferization.
Differential Revision: https://reviews.llvm.org/D109830
Summary: Making the late transformations opt-in results in less surprising behavior when composing multiple calls to the codegen strategy.
Reviewers:
Subscribers:
Differential revision: https://reviews.llvm.org/D109820
AliasInfo can now use union-find for a much more efficient implementation.
This brings no functional changes but large performance gains on more complex examples.
Differential Revision: https://reviews.llvm.org/D109819
Create a new document that explain both stages of the process in a single
place, merge and deduplicate the content from the two previous documents. Also
extend the documentation to account for the recent changes in pass structure
due to standard dialect splitting and translation being more flexible.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D109605
This seems in-line with the intent and how we build tools around it.
Update the description for the flag accordingly.
Also use an injected thread pool in MLIROptMain, now we will create
threads up-front and reuse them across split buffers.
Differential Revision: https://reviews.llvm.org/D109802
Both copy/alloc ops are using memref dialect after this change.
Reviewed By: silvas, mehdi_amini
Differential Revision: https://reviews.llvm.org/D109480
Add the makeComposedExtractSliceOp method that creates an ExtractSliceOp and folds chains of ExtractSliceOps by computing the sum of their offsets and by multiplying their strides.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D109601
This tiling option scalarizes all dynamic dimensions, i.e., it tiles all dynamic dimensions by 1.
This option is useful for linalg ops with partly dynamic tensor dimensions. E.g., such ops can appear in the partial iteration after loop peeling. After scalarizing dynamic dims, those ops can be vectorized.
Differential Revision: https://reviews.llvm.org/D109268
Only scf.for loops are supported at the moment. linalg.tiled_loop support will be added in a subsequent commit.
Only static tensor sizes are supported. Loops for dynamic tensor sizes can be peeled, but the generated code is not optimal due to a missing canonicalization pattern.
Differential Revision: https://reviews.llvm.org/D109043
* Revert https://reviews.llvm.org/D107307 so that both LHS and RHS have
the same layout with K0 as the innermost dimension.
* Continuing from https://reviews.llvm.org/D107003, move also 'K'
to the outer side, so that now the inter-tile dimensions as all outer,
and the intra-tile dimensions are all inner.
Reviewed By: asaadaldien
Differential Revision: https://reviews.llvm.org/D109692
This revision allows hoisting static alloc/dealloc pairs as high as possible during ComprehensiveBufferization.
This also aligns such allocated buffers to 128B by default.
This change exhibited some issues wrt insertion points and a missing copy that are also fixed in this revision; tests are updated accordingly.
Differential Revision: https://reviews.llvm.org/D109684
mlir-cpu-runner/math_polynomial_approx.mlir
This test case is currently failing on SystemZ, but it does not appear to
necessarily be a target specific problem. See discussion at
https://bugs.llvm.org/show_bug.cgi?id=51204.
Previously, we would insert a DimOp and rely on later canonicalizations.
Unfortunately, reifyShape kind of rewrites are not canonicalizations anymore.
This introduces undesirable pass dependencies.
Instead, immediately reify the result shape and avoid the DimOp altogether.
This is akin to a local folding, which avoids introducing more reliance on `-resolve-shaped-type-result-dims` (similar to compositions of `affine.apply` by construction to avoid chains of size > 1).
It does not completely get rid of the reliance on the pass as the process is merely local: calling the pass may still be necessary for global effects. Indeed, one of the tests still requires the pass.
Differential Revision: https://reviews.llvm.org/D109571
Types and attributes now have a `hasTrait` function that allow users to check
if a type defines a trait.
Also, AbstractType and AbstractAttribute has now a `hasTraitFn` field to carry
the implementation of the `hasTrait` function of the concrete type or attribute.
This patch also adds the remaining functions to access type and attribute traits
in TableGen.
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D105202
The conversion pattern is particularly useful for conversion of
block arguments in the master op.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D109610
Tosa.while shape inference requires repeatedly running shape inference across
the body of the loop until the types become static as we do not know the number
of iterations required by the loop body. Once the least specific arguments are
known they are propagated to both regions.
To determine the final end type, the least restrictive types are determined
from all yields.
Differential Revision: https://reviews.llvm.org/D108801
The original version of the bufferization pattern for linalg.generic would
manually clone operations within the region to the bufferized clone of the
operation. This triggers legality requirements on those operations in the
conversion infra. Instead, this now uses the rewriter to inline the region
instead, avoiding those legality requirements.
Differential Revision: https://reviews.llvm.org/D109581
Generate an scf.for instead of an scf.if for the partial iteration. This is for consistency reasons: The peeling of linalg.tiled_loop also uses another loop for the partial iteration.
Note: Canonicalizations patterns may rewrite partial iterations to scf.if afterwards.
Differential Revision: https://reviews.llvm.org/D109568
Extend the signature of the tile loop nest region builder to take all operand values to use and not just the scf::For iterArgs. This change allows us to pass in all block arguments of TiledLoop and use them directly instead of replacing them after the loop generation.
Reviewed By: pifon2a
Differential Revision: https://reviews.llvm.org/D109569
This revision fixes the traversal order of extract_slice during the inplace analysis.
It was previously thought that such ops could be analyzed at the very end.
This is unfortunately not true as the AliasInfo for dependents of these ops need to be updated.
This change allows the aliases introduced by the bufferization of extract_slice to be properly propagated.
Differential Revision: https://reviews.llvm.org/D109519
Switches to adding target specific, private includes instead of adding
global includes.
Reviewed By: ftynse
Differential Revision: https://reviews.llvm.org/D109494
This PR adds missing AtomicRMWKind::min/max cases which we would like to use for min/max reduction loop vectorizations.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D104881
This renames the primary methods for creating a zero value to `getZero`
instead of `getNullValue` and renames predicates like `isAllOnesValue`
to simply `isAllOnes`. This achieves two things:
1) This starts standardizing predicates across the LLVM codebase,
following (in this case) ConstantInt. The word "Value" doesn't
convey anything of merit, and is missing in some of the other things.
2) Calling an integer "null" doesn't make any sense. The original sin
here is mine and I've regretted it for years. This moves us to calling
it "zero" instead, which is correct!
APInt is widely used and I don't think anyone is keen to take massive source
breakage on anything so core, at least not all in one go. As such, this
doesn't actually delete any entrypoints, it "soft deprecates" them with a
comment.
Included in this patch are changes to a bunch of the codebase, but there are
more. We should normalize SelectionDAG and other APIs as well, which would
make the API change more mechanical.
Differential Revision: https://reviews.llvm.org/D109483
Further enhance the set of operations that can be handled by the sparse compiler
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109413
Conversion to the LLVM dialect is being refactored to be more progressive and
is now performed as a series of independent passes converting different
dialects. These passes may produce `unrealized_conversion_cast` operations that
represent pending conversions between built-in and LLVM dialect types.
Historically, a more monolithic Standard-to-LLVM conversion pass did not need
these casts as all operations were converted in one shot. Previous refactorings
have led to the requirement of running the Standard-to-LLVM conversion pass to
clean up `unrealized_conversion_cast`s even though the IR had no standard
operations in it. The pass must have been also run the last among all to-LLVM
passes, in contradiction with the partial conversion logic. Additionally, the
way it was set up could produce invalid operations by removing casts between
LLVM and built-in types even when the consumer did not accept the uncasted
type, or could lead to cryptic conversion errors (recursive application of the
rewrite pattern on `unrealized_conversion_cast` as a means to indicate failure
to eliminate casts).
In fact, the need to eliminate A->B->A `unrealized_conversion_cast`s is not
specific to to-LLVM conversions and can be factored out into a separate type
reconciliation pass, which is achieved in this commit. While the cast operation
itself has a folder pattern, it is insufficient in most conversion passes as
the folder only applies to the second cast. Without complex legality setup in
the conversion target, the conversion infra will either consider the cast
operations valid and not fold them (a separate canonicalization would be
necessary to trigger the folding), or consider the first cast invalid upon
generation and stop with error. The pattern provided by the reconciliation pass
applies to the first cast operation instead. Furthermore, having a separate
pass makes it clear when `unrealized_conversion_cast`s could not have been
eliminated since it is the only reason why this pass can fail.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D109507
Fix extra space print for llvm global op when the 'unamed_addr'
attribute was empty. This led to two spaces being printed in the custom
form between non-whitespace chars. A round trip would add an extra space
to a typical spaced form. NFC.
Differential Revision: https://reviews.llvm.org/D109502
OpenMP reductions need a neutral element, so we match some known reduction
kinds (integer add/mul/or/and/xor, float add/mul, integer and float min/max) to
define the neutral element and the atomic version when possible to express
using atomicrmw (everything except float mul). The SCF-to-OpenMP pass becomes a
module pass because it now needs to introduce new symbols for reduction
declarations in the module.
Reviewed By: chelini
Differential Revision: https://reviews.llvm.org/D107549
Fold dim ops of scf.for results to dim ops of the respective iter args if the loop is shape preserving.
Differential Revision: https://reviews.llvm.org/D109430
Fold dim ops of linalg.tiled_loop results to dim ops of the respective iter args if the loop is shape preserving.
Differential Revision: https://reviews.llvm.org/D109431
Run a small analysis to see if the runtime type of the iter_arg is changing. Fold only if the runtime type stays the same. (Same as `DimOfIterArgFolder` in SCF.)
Differential Revision: https://reviews.llvm.org/D109299
When tiling a LinalgOp, extract_slice/insert_slice pairs are inserted. To avoid going out-of-bounds when the tile size does not divide the shape size evenly (at the boundary), AffineMin ops are inserted. Some ops have assumptions regarding the dimensions of inputs/outputs. E.g., in a `A * B` matmul, `dim(A, 1) == dim(B, 0)`. However, loop bounds use either `dim(A, 1)` or `dim(B, 0)`.
With this change, AffineMin ops are expressed in terms of loop bounds instead of tensor sizes. (Both have the same runtime value.) This simplifies canonicalizations.
Differential Revision: https://reviews.llvm.org/D109267
This patch (e4635e6328) fixed a bug where a newly generated/reused
constant wouldn't dominate a folded operation. It did so by calling
isBeforeInBlock to move the constant around on demand. This introduced
a significant compile time regression, because "isBeforeInBlock" is
O(n) in the size of a block the first time it is called, and the cache
is invalidated any time canonicalize changes something big in the block.
This fixes LLVM PR51738 and this CIRCT issue:
https://github.com/llvm/circt/issues/1700
This does affect the order of constants left in the top of a block,
I staged in the testsuite changes in rG42431b8207a5.
Differential Revision: https://reviews.llvm.org/D109454
This patch refactors the existing implementation of computing an explicit
representation of an identifier as a floordiv in terms of other identifiers and
exposes this computation as a public function.
The computation of this representation is required to support local identifiers
in PresburgerSet subtract, complement and isEqual.
Reviewed By: bondhugula, arjunp
Differential Revision: https://reviews.llvm.org/D106662
Add loop coalesce utility for affine.for. This expects loops to have
been normalized a-priori. This works for both constant as well non
constant upper bounds having single/multiple result upper bound affine
map.
With contributions from Arnab Dutta and Uday Bondhugula.
Reviewed By: bondhugula, ayzhuang
Differential Revision: https://reviews.llvm.org/D108126
Right now all but the last bullet are relying on applied "must not" that
isn't there and the last bullet is a "must".
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D109389
The lowering has been incorrectly using the operands of the original op instead
of rewritten operands provided to matchAndRewrite call. This may lead to
spurious materializations and generally invalid IR.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D109355
The conversion has been incorrectly using the operands of the original
operation instead of the converted operands provided to the matchAndRewrite
call. This may lead to spurious materializations and generally invalid IR if
the producer of the original operands is deleted in the process of conversion.
Reviewed By: csigg
Differential Revision: https://reviews.llvm.org/D109356
It looks like it was a typo. Instead of `*maybeConstantIndex`,
`initTensorOp.getStaticSize(*maybeConstantIndex)` should be used to access the
dim size of the tensor. There is a test for that in `canonicalize.mlir`, but it
was working correctly because `ReplaceStaticShapeDims` was canonicalizing DimOp
before `FoldInitTensorWithDimOp`. So, to make the patterns more "orthogonal",
this case is disabled.
Differential Revision: https://reviews.llvm.org/D109247
Previously only await inside the async function (coroutine after lowering to async runtime) would check the error state
Reviewed By: mehdi_amini
Differential Revision: https://reviews.llvm.org/D109229
Create a gpu memset op and corresponding CUDA and ROCm wrappers.
Reviewed By: herhut, lorenrose1013
Differential Revision: https://reviews.llvm.org/D107548
FuncOp always lowers to an LLVM external linkage presently. This makes it impossible to define functions in mlir which are local to the current module. Until MLIR FuncOps have a more formal linkage specification, this commit allows funcop's to have an optionally specified llvm.linkage attribute, whose value will be used as the linkage of the llvm funcop when lowered.
Differential Revision: https://reviews.llvm.org/D108524
Support LLVM linkage
This makes the IR more readable, in particular when this will be used on
the builtin func outside of the LLVM dialect.
Reviewed By: wsmoses
Differential Revision: https://reviews.llvm.org/D109209
This simplifies setting up sparse tensors through C-style data structures.
Useful for runtimes that want to interact with MLIR-generated code
without knowning about all bufferization details (viz. memrefs).
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109251
The sparse index order must always be satisfied, but this
may give a choice in topsorts for several cases. We broke
ties in favor of any dense index order, since this gives
good locality. However, breaking ties in favor of pushing
unrelated indices into sparse iteration spaces gives better
asymptotic complexity. This revision improves the heuristic.
Note that in the long run, we are really interested in using
ML for ML to find the best loop ordering as a replacement for
such heuristics.
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109100
DialectAsmParser::parseKeyword is rejecting `'i' digit+` while it is
a valid identifier according to mlir/docs/LangRef.md.
Integer types actually used to be TOK_KEYWORD a while back before the
change: 6af866c58d.
This patch Modifies `isCurrentTokenAKeyword` to return true for tokens that
match integer types too.
The motivation for this change is the parsing of `!fir.type<{` `component-name: component-type,`+ `}>`
type in FIR that represent Fortran derived types. The component-names are
parsed as keywords, and can very well be i32 or any ixxx (which are
valid Fortran derived type component names).
The Quant dialect type parser had to be modified since it relied on `iw` not
being parsed as keywords.
Differential Revision: https://reviews.llvm.org/D108913
The limitation on iter_args introduced with D108806 is too restricting. Changes of the runtime type should be allowed.
Extends the dim op canonicalization with a simple analysis to determine when it is safe to canonicalize.
Differential Revision: https://reviews.llvm.org/D109125
* Now that packaging has stabilized, removes old mechanisms for loading extensions, preferring direct importing.
* Removes _cext_loader.py, _dlloader.py as unnecessary.
* Fixes the path where the CAPI dll is written on Windows. This enables that path of least resistance loading behavior to work with no further drama (see: https://bugs.python.org/issue36085).
* With this patch, `ninja check-mlir` on Windows with Python bindings works for me, modulo some failures that are actually due to a couple of pre-existing Windows bugs. I think this is the first time the Windows Python bindings have worked upstream.
* Downstream changes needed:
* If downstreams are using the now removed `load_extension`, `reexport_cext`, etc, then those should be replaced with normal import statements as done in this patch.
Reviewed By: jdd, aartbik
Differential Revision: https://reviews.llvm.org/D108489
The translation to LLVM IR used to construct sequential constants by recurring
down to individual elements, creating constant values for them, and wrapping
them into aggregate constants in post-order. This is highly inefficient for
large constants with known data such as DenseElementsAttr. Use LLVM's
ConstantData for the innermost dimension instead. LLVM does seem to support
data constants for nested sequential constants so the outer dimensions are
still handled recursively. Nevertheless, this speeds up the translation of
large constants with equal dimensions by up to 30x.
Users are advised to rewrite large constants to use flat types before
translating to LLVM IR if more efficiency in translation is necessary. This is
not done automatically as the translation is not aware of the expectations of
the overall compilation flow about type changes and indexing, in particular for
global constants with external linkage.
Reviewed By: silvas
Differential Revision: https://reviews.llvm.org/D109152
Add an operation omp.critical.declare to declare names/symbols of
critical sections. Named omp.critical operations should use symbols
declared by omp.critical.declare. Having a declare operation ensures
that the names of critical sections are global and unique. In the
lowering flow to LLVM IR, the OpenMP IRBuilder creates unique names
for critical sections.
Reviewed By: ftynse, jeanPerier
Differential Revision: https://reviews.llvm.org/D108713
This upstreams the Cpp emitter, initially presented with [1], from [2]
to MLIR core. Together with the previously upstreamed EmitC dialect [3],
the target allows to translate MLIR to C/C++.
[1] https://reviews.llvm.org/D76571
[2] https://github.com/iml130/mlir-emitc
[3] https://reviews.llvm.org/D103969
Co-authored-by: Jacques Pienaar <jpienaar@google.com>
Co-authored-by: Simon Camphausen <simon.camphausen@iml.fraunhofer.de>
Co-authored-by: Oliver Scherf <oliver.scherf@iml.fraunhofer.de>
Reviewed By: jpienaar
Differential Revision: https://reviews.llvm.org/D104632
Use the recently introduced OpenMPIRBuilder facility to transate OpenMP
workshare loops with reductions to LLVM IR calling OpenMP runtime. Most of the
heavy lifting is done at the OpenMPIRBuilder. When other OpenMP dialect
constructs grow support for reductions, the translation can be updated to
operate on, e.g., an operation interface for all reduction containers instead
of workshare loops specifically. Designing such a generic translation for the
single operation that currently supports reductions is premature since we don't
know how the reduction modeling itself will be generalized.
Reviewed By: kiranchandramohan
Differential Revision: https://reviews.llvm.org/D107343
(1) renamed SparseTensor to SparseTensorCOO, the other one remains SparseTensorStorage to focus on contrast
(2) documents difference between public API exclusively for compiler-generated code and methods that could be used by other runtimes (TBD) that want to interact with MLIR
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D109039
Add method to get NameLoc. Treat null child location as unknown to avoid
needing to create UnknownLoc in C API where child loc is not needed.
Differential Revision: https://reviews.llvm.org/D108678
This patch is to add Image Operands in SPIR-V Dialect and also let ImageDrefGather to use Image Operands.
Image Operands are used in many image instructions. "Image Operands encodes what oprands follow, as per Image Operands". And ususally, they are optional to image instructions.
The format of image operands looks like:
%0 = spv.ImageXXXX %1, ... %3 : f32 ["Bias|Lod"](%4, %5 : f32, f32) -> ...
This patch doesn’t implement all operands (see Section 3.14 in SPIR-V Spec) but provides a skeleton of it. There is TODO in verifyImageOperands function.
Co-authored: Alan Liu <alanliu.yf@gmail.com>
Reviewed by: antiagainst
Differential Revision: https://reviews.llvm.org/D108501
In D104421, we changed the API for pass registration.
Before you would write:
void registerPass("my-pass", "My Pass Description.",
[] { return createMyPass(); });
while now you’d only write:
void registerPass([] { return createMyPass(); });
If you’re using TableGen to define your pass registration, you shouldn’t have anything to do. If you’re using directly the C++ API here are some changes.
Your project may also be broken even if you use TableGen and you call the
generated registration API in case your pass implementation didn’t inherit from
the MyPassBase class generated by TableGen.
If you don't use TableGen, the "my-pass" and "My Pass Description." fields must
be provided by overriding methods on the pass itself:
llvm::StringRef getArgument() const final { return "my-pass"; }
llvm::StringRef getDescription() const final {
return "My Pass Description.";
}
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D104429
Trying to reduce confusion by having the name of the public method match that of the private method for handling the recursion. Also adding some comments to SparseTensorStorage::fromCOO to help clarify what the recursive calls are doing in the dense case.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D108954
Aart Bik