Currently, MlirTranslateMain only executes one of the requested translations, and does not error if multiple are specified. This commit enables translations to be chained in the specified order.
This makes round-trip tests easier, since existing import/export passes can be reused and no combined round-trip passes have to be registered (example: mlir-translate -serialize-spirv -deserialize-spirv).
Additionally, by leveraging TranslateRegistration with file-to-file TranslateFunctions, generic pre- and post-processing can be added before/after conversion to/from MLIR.
Reviewed By: lattner, Mogball
Differential Revision: https://reviews.llvm.org/D143719
Do not create an Affine ops for expanded size because the affine op is
too complicated which would hit an assertion in affine ops
simplification.
Reviewed By: mravishankar
Differential Revision: https://reviews.llvm.org/D144151
This patch reintroduces an API to create operations with a pre-existing
DictionaryAttr. This API does not populate the attributes with any
default attributes the operation may have, like the API that takes a
NamedAttrList does. NamedAttrList is effective at not re-hashing the
attributes if no default attributes were added, but this new API speeds
up clone-heavy workloads slightly (~5%).
Reviewed By: rriddle
Differential Revision: https://reviews.llvm.org/D144204
Allow user to apply multi-buffering transformation for cases where proving
that there is no loop carried dependency is not trivial. In this case user needs
to ensure that the data are written and read in the same iteration otherwise the
result is incorrect.
Differential Revision: https://reviews.llvm.org/D144227
This commit moves the access group translation into the
LoopAnnotationTranslation class as these two metadata kinds only appear
together.
Drops the access group cleanup from `ModuleTranslation::forgetMapping`
as this is only used on function regions. Access groups only appear in the
region of a global metadata operation and will thus not be cleaned here.
Analogous to https://reviews.llvm.org/D143577
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144253
This commit introduces the DINamespaceAttr to model LLVM's DINamespace metadata.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144243
This commit ensures that the importing of LLVM modules first verifies
that the module is even valid. As many tests did not work with valid
LLVM IR, they were fixed as part of this commit.
Some error messages were only reachable with invalid input IR, thus they
were replaced with a failures.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144186
This revision adds atomic support to the LoadOp. It chooses
to print the atomic keywords together with the syncscope and
ordering arguments, which simplifies parsing and printing compared
to the LLVM IR printer that puts the atomic keyword at the beginning.
It uses the ordering attribute to check if the load is atomic.
The revision also implements verifiers to ensure the constraints
that apply to atomic load operations are checked.
Reviewed By: Dinistro
Differential Revision: https://reviews.llvm.org/D144112
This loosens the requirement of no passthrough function attribute being
present to checking for specific attributes that prevent inlining. Since
these attributes are no longer strictly passthrough, they should
eventually be upgraded to some form of addressable attributes.
Drops the expensive StringSwitches over call and function attributes in
favor of selectively disallowing attributes that prevent inlining
(similiar to the LLVM inliner), thereby moving to a less conservative
approach.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144104
No need for a temp COO and sort even when converting dense -> CSC, we can instead rotate the loop to yield a ordered coordinates at beginning.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D144213
We will support symmetric MTX without expanding the data in the sparse tensor
storage.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D144059
This commit adds special handling for the debug intrinsic value
handling. LLVM allows to relax the def before use property for debug
intrinsics, so this property cannot be assumed for metadata values.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144177
A new transform dialect op is introduced to perform the rewrite.
The test pass option is now obsolete and is removed in favor of the transform.
In the process I realized the tensor.pad nofold attribute was not taken into account
and added support to emit a bufferization.alloc_tensor + linalg.copy.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D143943
This is permitted by the op, but the current lowering generates invalid IR.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D144090
This commit renames the "di_void_result_type" to "di_null_type" as LLVM
does use null not exclusively for void types. An added test demonstrates
this for variadic function declarations, whose DISubroutine indicates the
start of variadic types with `null`.
Reviewed By: gysit
Differential Revision: https://reviews.llvm.org/D144109
Although specifying an index that is out of bounds for both `memref.dim`
and `tensor.dim` produces an undefined behavior, this is still valid IR.
In particular, we could expose an out of bound index because of some
optimizations, for instance as demonstrated with
https://github.com/llvm/llvm-project/issues/60295, and this shouldn't
cause the compiler to abort.
This patch removes the overzealous verifier checks and properly handles
out of bound indices (as in it doesn't crash the compiler, but still
produces UB).
This fixes https://github.com/llvm/llvm-project/issues/60295.
Note: That `shape.dim` has a similar problem but we're not supposed to
produce UB in this case. Instead we're supposed to propagate an error in
the resulting value and I don't know how to do that at the moment. Hence I
left this part out of the patch.
Differential Revision: https://reviews.llvm.org/D143999
The revision adds support for importing alias.scope and noalias metadata
from LLVM IR into LLVM dialect. It also adds a verifier to the
AliasScopeMetadataOp to check that the associated domain exists
and is of type AliasScopeDomainMetadataOp.
Reviewed By: Dinistro
Differential Revision: https://reviews.llvm.org/D143923
Plain `getVectorType()` can be quite confusing and error-prone
given that, well, vector ops always work on vector types, and
it can commonly involve both source and result vectors. So this
commit makes various such accessor methods to be explicit w.r.t.
source or result vectors.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D144159
* Flattening/simplifying some nested conditionals
* const-ifying some local variables
Depends On D143800
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D143949
The inner tiling sizes could be dynamic (which are Values). In this
context, they should be added to tiledOperand when cloning the op.
Reviewed By: chelini
Differential Revision: https://reviews.llvm.org/D143978
Upcoming AMD hardware will include functions that accept 8-bit floats.
Specifically, there are MFMA instructions that accept 8-bit floats,
either using the same or mixed formats. This patch adds MLIR wrappers
for these intrinsics and explicitly adds support for 8-bit floats in
the gpu-to-rocdl conversion by way of amdgpu-to-rocdl.
Since LLVM does not have f8 types, when targeting LLVM for compilation
on an AMD GPU, both f8 types used on AMD hardware (f8E5M2FNUZ and
f8E4M3FNUZ) are rewritten to i8.
This patch also relaxes the restriction that the types of both source
operands to a amdgpu.mfma instructions match exactly, as this is not
necessarily required for the bf8 (f8E5M2FNUZ) and fp8 (f8E4M3FNUZ)
instructions. In addition, since the buffer_{load,store} operations
maintain a whitelist of permitted types, we add the relevant f8 types
to that list.
This patch does not add any implementations of arithmetic operations
for f8 types.
Reviewed By: jakeh-gc
Differential Revision: https://reviews.llvm.org/D143956
This transform materializes a buffer allocation for a given tensor value. All uses of the original value are replaced with the allocation.
Certain non-DPS ops may have an optimized lowering path that bufferizes the entire defining op. Such optimization is added for `tensor.pad` as part of this change.
The resulting IR can be further bufferized with One-Shot Bufferize.
Differential Revision: https://reviews.llvm.org/D144022
This transform op returns a value handle pointing to the specified OpResult of the targeted op.
Differential Revision: https://reviews.llvm.org/D144087
`restrict` is similar to the C++ restrict keyword. Results of `to_tensor` that have the `restrict` attribute are guaranteed to not alias any other `to_tensor` result (after bufferization).
Note: Since `to_memref` ops are not supported by One-Shot Bufferize and all bufferizable ops follow DPS rules (i.e., the buffer of the result is the buffer of an operand or an alias thereof), the buffer of a `to_tensor` op that has the `restrict` attribute is always an entirely "new" buffer that is not aliasing with the future buffer of any tensor value in the entire program. This makes such `to_tensor` ops "safe" from a bufferization perspective; they cannot cause RaW conflicts.
Differential Revision: https://reviews.llvm.org/D144021
This patch simply adds `vectorize_nd_extract` (that's currently only
used for the `vectorize` Op) to `masked_vectorize`. A test is added to
verify that it works as expected - it prevents the masked vectorisation
of `tensor.extract`, which is currently not supported.
Differential Revision: https://reviews.llvm.org/D142634
Introduce a pattern to 'push down' a `tensor.unpack` through a
`tensor.pad`. The propagation happens if the unpack does not touch the
padded dimensions.
Reviewed By: hanchung
Differential Revision: https://reviews.llvm.org/D143907
When changing IR in a RewriterPattern, all changes must go through the
rewriter. There are several convenience functions in RewriterBase that
help with high-level modifications, such as replaceAllUsesWith for
Values, but there is currently none to do the same task for Blocks.
Reviewed By: mehdi_amini, ingomueller-net
Differential Revision: https://reviews.llvm.org/D142525
This may be seen as a hack, but it allows for any piece of MLIR to be able
to end up with DWARF debug info through LLVM.
Assuming the operations in the function have location such as FileLineCol,
this provides backtraces with line tables and allows to step in a debugger.
That makes this pass a perfect companion to -snapshot-op-locations
It was also the default behavior of MLIR to LLVM IR translation until MLIR
got support for proper debug info attributes.
Differential Revision: https://reviews.llvm.org/D144069
Support for masking static shapes was already implemented in the past
but not enabled so this patch is just removing a pre-condition check and
adding some tests with static shapes.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D143937
This patch adds support for masked vector.gather ops using the
vector.mask representation. It includes the implementation of the
MaskableOpInterface, Linalg vectorizer support and lowering to LLVM.
Reviewed By: ThomasRaoux
Differential Revision: https://reviews.llvm.org/D143939
