This commit makes reductions part of the terminator. Instead of
`scf.yield`, `scf.reduce` now terminates the body of `scf.parallel` ops.
`scf.reduce` may contain an arbitrary number of reductions, with one
region per reduction.
Example:
```mlir
%init = arith.constant 0.0 : f32
%r:2 = scf.parallel (%iv) = (%lb) to (%ub) step (%step) init (%init, %init)
-> f32, f32 {
%elem_to_reduce1 = load %buffer1[%iv] : memref<100xf32>
%elem_to_reduce2 = load %buffer2[%iv] : memref<100xf32>
scf.reduce(%elem_to_reduce1, %elem_to_reduce2 : f32, f32) {
^bb0(%lhs : f32, %rhs: f32):
%res = arith.addf %lhs, %rhs : f32
scf.reduce.return %res : f32
}, {
^bb0(%lhs : f32, %rhs: f32):
%res = arith.mulf %lhs, %rhs : f32
scf.reduce.return %res : f32
}
}
```
`scf.reduce` operations can no longer be interleaved with other ops in
the body of `scf.parallel`. This simplifies the op and makes it possible
to assign the `RecursiveMemoryEffects` trait to `scf.reduce`. (This was
not possible before because the op was not a terminator, causing the op
to be DCE'd.)
Abort fusion if memref load may alias write, but not the exact alias.
Add alias check hook to `naivelyFuseParallelOps`, so user can customize
alias checking.
Use builtin alias analysis in `ParallelLoopFusion` pass.
There is a use case that we need to peel the first iteration out of the
for loop so that the peeled forOp can be canonicalized away and the
fillOp can be fused into the inner forall loop. For example, we have
nested loops as below
```
linalg.fill ins(...) outs(...)
scf.for %arg = %lb to %ub step %step
scf.forall ...
```
After the peeling transform, it is expected to be
```
scf.forall ...
linalg.fill ins(...) outs(...)
scf.for %arg = %(lb + step) to %ub step %step
scf.forall ...
```
This patch makes the most use of the existing peeling functions and adds
support for peeling the first iteration out of the loop.
Support loops without static boundaries. Since the number of iteration
is not known we need to predicate prologue and epilogue in case the
number of iterations is smaller than the number of stages.
This patch includes work from @chengjunlu
-Fix case where an op is scheduled in stage 0 and used with a distance
of 1
-Fix case where we don't peel the epilogue and a value not part of the
last stage is used outside the loop.
/llvm-project/mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp:703:5: error: non-void lambda does not return a value in all
control paths [-Werror,-Wreturn-type]
};
^
1 error generated.
The current implementation of tiling using `scf.for` is convoluted to
make sure that the destination passing style of the untiled program is
preserved. The addition of support to tile using `scf.forall` (adapted
from the transform operation in Linalg) in
https://github.com/llvm/llvm-project/pull/67083 used cloning of the
tiled operations to better streamline the implementation. This PR adapts
the other tiling methods to use a similar approach, making the
transformations (and handling destination passing style semantics) more
systematic.
---------
Co-authored-by: Abhishek-Varma <avarma094@gmail.com>
Loop peeling is not beneficial if the step size already divides "ub -
lb". There are currently some simple checks to prevent peeling in such
cases when lb, ub, step are constants. This commit adds support for IR
that is the result of loop peeling in the general case; i.e., lb, ub,
step do not necessarily have to be constants.
This change adds a new affine_map simplification rule for semi-affine
maps that appear during loop peeling and are guaranteed to evaluate to a
constant zero. Affine maps such as:
```
(1) affine_map<()[ub, step] -> ((ub - ub mod step) mod step)
(2) affine_map<()[ub, lb, step] -> ((ub - (ub - lb) mod step - lb) mod step)
(3) ^ may contain additional summands
```
Other affine maps with modulo expressions are not supported by the new
simplification rule.
This fixes#71469.
Expose loop results, which correspond to the region iter_arg values that
are returned from the loop when there are no more iterations. Exposing
loop results is optional because some loops (e.g., `scf.while`) do not
have a 1-to-1 mapping between region iter_args and op results.
Also add additional helper functions to query tied
results/iter_args/inits.
The `LoopLikeOpInterface` already has interface methods to query inits
and iter_args. This commit adds helper functions to query tied
init/iter_arg pairs and removes the corresponding functions for
`scf::ForOp`.
Fix a crash reported in #64331. The crash is described in the following
comment:
> It looks like the bug is being caused by the command line argument
--scf-parallel-loop-tiling=parallel-loop-tile-sizes=0. More
specifically, --scf-parallel-loop-tiling=parallel-loop-tile-sizes sets
the tileSize variable to 0 on [this
line](7cc1bfaf37/mlir/lib/Dialect/SCF/Transforms/ParallelLoopTiling.cpp (L67)).
tileSize is then used on [this
line](7cc1bfaf37/mlir/lib/Dialect/SCF/Transforms/ParallelLoopTiling.cpp (L117))
causing a divide by zero exception.
This PR will:
1. Call `signalPassFail()` when 0 is passed as a tile size.
2. Avoid the divide by zero that causes the crash.
Note: This is my first PR for MLIR, so please liberally critique it.
A recent change modified the parameter tileSize from Value to
OpFoldResult. Therefore we should call getAsOpFoldResult before passing
on the tileSize.
Adjust a test regarding this new behavior.
Similar to `scf::tileUsingSCFForOp` that is a method that tiles
operations that implement the `TilingInterface`, using `scf.for`
operations, this method introduces tiling of operations using
`scf.forall`. Most of this implementation is derived from
`linalg::tileToForallOp` method. Eventually that method will either be
deprecated or moved to use the method introduced here.
Add a new interface method that returns the yielded values.
Also add a verifier that checks the number of inits/iter_args/yielded
values. Most of the checked invariants (but not all of them) are already
covered by the `RegionBranchOpInterface`, but the `LoopLikeOpInterface`
now provides (additional) error messages that are easier to read.
The TableGen code generator now generates C++ code that returns a single
`OpOperand &` for `get...Mutable` of operands that are not variadic and
not optional. `OpOperand::set`/`assign` can be used to set a value (same
as `MutableOperandRange::assign`). This is safer than
`MutableOperandRange` because only single values (and no longer
`ValueRange`) can be assigned.
E.g.:
```
// Assignment of multiple values to non-variadic operand.
// Before: Compiles, but produces invalid op.
// After: Compilation error.
extractSliceOp.getSourceMutable().assign({v1, v2});
```
The `BufferizableOpInterface` implementation of `scf.for` currently
assumes that an OpResult does not alias with any tensor apart from the
corresponding init OpOperand. Newly allocated buffers (inside of the
loop) are also allowed. The current implementation checks whether the
respective init_arg and yielded value are equivalent. This is overly
strict and causes extra buffer allocations/copies when yielding a new
buffer allocation from a loop.
`affine::replaceForOpWithNewYields` and `replaceLoopWithNewYields` (for
"scf.for") are now interface methods and additional loop-carried
variables can now be added to "scf.for"/"affine.for" uniformly. (No more
`TypeSwitch` needed.)
Note: `scf.while` and other loops with loop-carried variables can
implement `replaceWithAdditionalYields`, but to keep this commit small,
that is not done in this commit.
* "init" operands are specified with `MutableOperandRange` (which gives
access to the underlying `OpOperand *`). No more magic numbers.
* Remove most interface methods and make them helper functions. Only
`getInitsMutable` should be implemented.
* Provide separate helper functions for accessing mutable/immutable
operands (`OpOperand`/`Value`, in line with #66515): `getInitsMutable`
and `getInits` (same naming convention as auto-generated op accessors).
`getInputOperands` was not renamed because this function cannot return a
`MutableOperandRange` (because the operands are not necessarily
consecutive). `OpOperandVector` is no longer needed.
* The new `getDpsInits`/`getDpsInitsMutable` is more efficient than the
old `getDpsInitOperands` because no `SmallVector` is created. The new
functions return a range of operands.
* Fix a bug in `getDpsInputOperands`: out-of-bounds operands were
potentially returned.
This is necessary to support deallocation of IR with gpu.launch
operations because it does not implement the RegionBranchOpInterface.
Implementing the interface would require it to support regions with
unstructured control flow and produced arguments/results.
This commit implements `LoopLikeOpInterface` on `scf.while`. This
enables LICM (and potentially other transforms) on `scf.while`.
`LoopLikeOpInterface::getLoopBody()` is renamed to `getLoopRegions` and
can now return multiple regions.
Also fix a bug in the default implementation of
`LoopLikeOpInterface::isDefinedOutsideOfLoop()`, which returned "false"
for some values that are defined outside of the loop (in a nested op, in
such a way that the value does not dominate the loop). This interface is
currently only used for LICM and there is no way to trigger this bug, so
no test is added.
This function was inconsistent with the remaining API because it
accepted `OpOperand &` that do not belong to the op. All the other
functions assert. This helper function is also not really necessary, as
the iter_arg number is identical to the result number.
In line with #66515, change `MutableArrayRange::begin`/`end` to
enumerate `OpOperand &` instead of `Value`. Also remove
`ForOp::getIterOpOperands`/`setIterArg`, which are now redundant.
Note: `MutableOperandRange` cannot be made a derived class of
`indexed_accessor_range_base` (like `OperandRange`), because
`MutableOperandRange::assign` can change the number of operands in the
range.
This commit removes the deallocation capabilities of
one-shot-bufferization. One-shot-bufferization should never deallocate
any memrefs as this should be entirely handled by the
ownership-based-buffer-deallocation pass going forward. This means the
`allow-return-allocs` pass option will default to true now,
`create-deallocs` defaults to false and they, as well as the escape
attribute indicating whether a memref escapes the current region, will
be removed. A new `allow-return-allocs-from-loops` option is added as a
temporary workaround for some bufferization limitations.
`operator[]` returns `OpOperand &` instead of `Value`.
* This allows users to get OpOperands by name instead of "magic" number.
E.g., `extractSliceOp->getOpOperand(0)` can be written as
`extractSliceOp.getSourceMutable()[0]`.
* `OperandRange` provides a read-only API to operands: `operator[]`
returns `Value`. `MutableOperandRange` now provides a mutable API:
`operator[]` returns `OpOperand &`, which can be used to set operands.
Note: The TableGen code generator could be changed to return `OpOperand
&` (instead of `MutableOperandRange`) for non-variadic and non-optional
arguments in a subsequent change. Then the `[0]` part in the above
example would no longer be necessary.
* Always use the auto-generated `getInitArgs` function. Remove the
hand-written `getInitOperands` duplicate.
* Remove `hasIterOperands` and `getNumIterOperands`. The names were
inconsistent because the "arg" is called `initArgs` in TableGen. Use
`getInitArgs().size()` instead.
* Fix verification around ops with no results.
This change adds a method to modify the ConversionTarget used during
`transform.apply_conversion_patterns` to the
`ConversionPatternDescriptorOpInterface`. This is needed when the TypeConverter
is used to dictate the dynamic legality of operations, as in "structural"
conversion patterns present in, for example, the SCF and func dialects.
As a first use case/test, this change also adds a
`transform.apply_patterns.scf.structural_conversions` operation to the SCF
dialect.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D158672
The scf.forall.in_parallel terminator operation has a nested graph region with the NoTerminator trait. Such regions are not supported by the default implementations. Therefore, this commit adds a specialized implementation for
this operation which only covers the case where the nested region is empty.
This is because after bufferization, ops like tensor.parallel_insert_slice were already converted to memref operations residing int the scf.forall only and the nested region of scf.forall.in_parallel ends up empty.
The scf.forall.in_parallel terminator operation has a nested graph region with
the NoTerminator trait. Such regions are not supported by the default
implementations. Therefore, this commit adds a specialized implementation for
this operation which only covers the case where the nested region is empty.
This is because after bufferization, ops like tensor.parallel_insert_slice were
already converted to memref operations residing int the scf.forall only and the
nested region of scf.forall.in_parallel ends up empty.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D158979
This is the first commit in a series with the goal to rework the
BufferDeallocation pass. Currently, this pass heavily relies on copies
to perform correct deallocations, which leads to very slow code and
potentially high memory usage. Additionally, there are unsupported cases
such as returning memrefs which this series of commits aims to add
support for as well.
This first commit removes the deallocation capabilities of
one-shot-bufferization.One-shot-bufferization should never deallocate any
memrefs as this should be entirely handled by the buffer-deallocation pass
going forward. This means the allow-return-allocs pass option will
default to true now, create-deallocs defaults to false and they, as well
as the escape attribute indicating whether a memref escapes the current region,
will be removed.
The documentation should w.r.t. these pass option changes should also be
updated in this commit.
Reviewed By: springerm
Differential Revision: https://reviews.llvm.org/D156662
One-Shot Bufferize correctly handles RaW conflicts around repetitive regions (loops). Specical handling is needed for parallel regions. These are a special kind of repetitive regions that can have additional RaW conflicts that would not be present if the regions would be executed sequentially.
Example:
```
%0 = bufferization.alloc_tensor()
scf.forall ... {
%1 = linalg.fill ins(...) outs(%0)
...
scf.forall.in_parallel {
tensor.parallel_insert_slice %1 into ...
}
}
```
A separate (private) buffer must be allocated for each iteration of the `scf.forall` loop.
This change adds a new interface method to `BufferizableOpInterface` to detect parallel regions. By default, regions are assumed to be sequential.
A buffer is privatized if an OpOperand bufferizes to a memory read inside a parallel region that is different from the parallel region where operand's value is defined.
Differential Revision: https://reviews.llvm.org/D159286
This revision adds support for unstructured control flow to the bufferization infrastructure. In particular: regions with multiple blocks, `cf.br`, `cf.cond_br`.
Two helper templates are added to `BufferizableOpInterface.h`, which can be implemented by ops that supported unstructured control flow in their regions (e.g., `func.func`) and ops that branch to another block (e.g., `cf.br`).
A block signature is always bufferized together with the op that owns the block.
Differential Revision: https://reviews.llvm.org/D158094
This patch improves the reduction tiling for linalg to support multiple
reduction dimensions.
Reviewed By: mravishankar
Differential Revision: https://reviews.llvm.org/D158005
`getBufferType` computes the bufferized type of an SSA value without bufferizing any IR. This is useful for predicting the bufferized type of iter_args of a loop.
To avoid endless recursion (e.g., in the case of "scf.for", the type of the iter_arg depends on the type of init_arg and the type of the yielded value; the type of the yielded value depends on the type of the iter_arg again), `fixedTypes` was used to fall back to "fixed" type. A simpler way is to maintain an "invocation stack". `getBufferType` implementations can then inspect the invocation stack to detect repetitive computations (typically when computing the bufferized type of a block argument).
Also improve error messages in case of inconsistent memory spaces inside of a loop.
Differential Revision: https://reviews.llvm.org/D158060
This revision is needed to support bufferization of `cf.br`/`cf.cond_br`. It will also be useful for better analysis of loop ops.
This revision generalizes `getAliasingOpResults` to `getAliasingValues`. An OpOperand can now not only alias with OpResults but also with BlockArguments. In the case of `cf.br` (will be added in a later revision): a `cf.br` operand will alias with the corresponding argument of the destination block.
If an op does not implement the `BufferizableOpInterface`, the analysis in conservative. It previously assumed that an OpOperand may alias with each OpResult. It now assumes that an OpOperand may alias with each OpResult and each BlockArgument of the entry block.
Differential Revision: https://reviews.llvm.org/D157957
Add two new helper functions `getBeforeBody` and `getAfterBody` to be consistent with "scf.for" (`getBody`) and to show in the API that both regions have exactly one block. Also simplify some code that assumed that there can be more than one block in a region.
Differential Revision: https://reviews.llvm.org/D157860
These two headers both contained a strange mix of definitions related to
both patterns and non-pattern transforms. Put patterns and "populate"
functions into Patterns.h and standalone transforms into Transforms.h.
Depends On: D155223
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D155454
Add a simple transform operation to the NVGPU extension that performs
software pipelining of copies to shared memory. The functionality is
extremely minimalistic in this version and only supports copies from
global to shared memory inside an `scf.for` loop with either
`vector.transfer` or `nvgpu.device_async_copy` operations when
pipelining preconditions are already satisfied in the IR. This is the
minimally useful version that uses the more general loop pipeliner in an
NVGPU-specific way. Further extensions and orthogonalizations will be
necessary.
This required a change to the loop pipeliner itself to properly
propagate errors should the predicate generator fail.
This is loosely inspired from the vesion in IREE, but has less unsafe
assumptions and more principled way of communicating decisions.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D155223
Add the missing one-to-n structural type conversion pattern for the
scf.for operation.
Reviewed By: ingomueller-net
Differential Revision: https://reviews.llvm.org/D154299
In two places, a ResultRange was copied into a SmallVector just to be
passed as a ValueRange argument. With this patch, the ResultRanges are
passed directly, avoiding a copy.
Reviewed By: ingomueller-net
Differential Revision: https://reviews.llvm.org/D154685
