The approach for handling reductions in the outer most
dimension follows that for inner most dimensions, outlined
below
First, transpose to move reduction dims, if needed
Convert reduction from n-d to 2-d canonical form
Then, for outer reductions, we emit the appropriate op
(add/mul/min/max/or/and/xor) and combine the results.
Differential Revision: https://reviews.llvm.org/D107675
The existing vector transforms reduce the dimension of transfer_read
ops. However, beyond a certain point, the vector op actually has
to be reduced to a scalar load, since we can't load a zero-dimension
vector. This handles this case.
Note that in the longer term, it may be preferaby to support
zero-dimension vectors. see
https://llvm.discourse.group/t/should-we-have-0-d-vectors/3097.
Differential Revision: https://reviews.llvm.org/D103432
Historically the builtin dialect has had an empty namespace. This has unfortunately created a very awkward situation, where many utilities either have to special case the empty namespace, or just don't work at all right now. This revision adds a namespace to the builtin dialect, and starts to cleanup some of the utilities to no longer handle empty namespaces. For now, the assembly form of builtin operations does not require the `builtin.` prefix. (This should likely be re-evaluated though)
Differential Revision: https://reviews.llvm.org/D105149
This simplifies the vector to LLVM lowering. Previously, both vector.load/store and vector.transfer_read/write lowered directly to LLVM. With this commit, there is a single path to LLVM vector load/store instructions and vector.transfer_read/write ops must first be lowered to vector.load/store ops.
* Remove vector.transfer_read/write to LLVM lowering.
* Allow non-unit memref strides on all but the most minor dimension for vector.load/store ops.
* Add maxTransferRank option to populateVectorTransferLoweringPatterns.
* vector.transfer_reads with changing element type can no longer be lowered to LLVM. (This functionality is needed only for SPIRV.)
Differential Revision: https://reviews.llvm.org/D106118
Simplify vector unrolling pattern to be more aligned with rest of the
patterns and be closer to vector distribution.
The new implementation uses ExtractStridedSlice/InsertStridedSlice
instead of the Tuple ops. After this change the ops based on Tuple don't
have any more used so they can be removed.
This allows removing signifcant amount of dead code and will allow
extending the unrolling code going forward.
Differential Revision: https://reviews.llvm.org/D105381
Uses elementwise interface to generalize canonicalization pattern and add a new
pattern for vector.contract case.
Differential Revision: https://reviews.llvm.org/D104343
The current code does not preserve the order of the parallel
dimensions when doing multi-reductions and thus we can end
up in scenarios where the result shape does not match the
desired shape after reduction.
This patch fixes that by ensuring that the parallel indices
are in order and then concatenates them to the reduction dimensions
so that the reduction dimensions are innermost.
Differential Revision: https://reviews.llvm.org/D104884
The patch changes the pretty printed FillOp operand order from output, value to value, output. The change is a follow up to https://reviews.llvm.org/D104121 that passes the fill value using a scalar input instead of the former capture semantics.
Differential Revision: https://reviews.llvm.org/D104356
This provides a sizable compile time improvement by seeding
the worklist in an order that leads to less iterations of the
worklist.
This patch only changes the behavior of the Canonicalize pass
itself, it does not affect other passes that use the
GreedyPatternRewrite driver
Differential Revision: https://reviews.llvm.org/D103053
VectorTransferPermutationMapLoweringPatterns can be enabled via a pass option. These additional patterns lower permutation maps to minor identity maps with broadcasting, if possible, allowing for more efficient vector load/stores. The option is deactivated by default.
Differential Revision: https://reviews.llvm.org/D102593
Broadcast dimensions of vector transfer ops are always in-bounds. This is consistent with the fact that the starting position of a transfer is always in-bounds.
Differential Revision: https://reviews.llvm.org/D102566
Add TransferWritePermutationLowering, which replaces permutation maps of TransferWriteOps with vector.transpose.
Differential Revision: https://reviews.llvm.org/D102548
Broadcast dimensions of a vector transfer op have no corresponding dimension in the mask vector. E.g., a 2-D TransferReadOp, where one dimension is a broadcast, can have a 1-D `mask` attribute.
This commit also adds a few additional transfer op integration tests for various combinations of broadcasts, masking, dim transposes, etc.
Differential Revision: https://reviews.llvm.org/D101745
VectorTransfer split previously only split read xfer ops. This adds
the same logic to write ops. The resulting code involves 2
conditionals for write ops while read ops only needed 1, but the created
ops are built upon the same patterns, so pattern matching/expectations
are all consistent other than in regards to the if/else ops.
Differential Revision: https://reviews.llvm.org/D102157
Three patterns are added to convert into vector.multi_reduction into a
sequence of vector.reduction as the following:
- Transpose the inputs so inner most dimensions are always reduction.
- Reduce rank of vector.multi_reduction into 2d with inner most
reduction dim (get the 2d canical form)
- 2D canonical form is converted into a sequence of vector.reduction.
There are two things we might worth in a follow up diff:
- An scf.for (maybe optionally) around vector.reduction instead of unrolling it.
- Breakdown the vector.reduction into a sequence of vector.reduction
(e.g tree-based reduction) instead of relying on how downstream dialects
handle it.
Note: this will requires passing target-vector-length
Differential Revision: https://reviews.llvm.org/D101570
The current implementation allows for TransferWriteOps with broadcasts that do not make sense. E.g., a broadcast could write a vector into a single (scalar) memory location, which is effectively the same as writing only the last element of the vector.
Differential Revision: https://reviews.llvm.org/D100842
Move the existing optimization for transfer op on tensor to folder and
canonicalization. This handles the write after write case and read after write
and also add write after read case.
Differential Revision: https://reviews.llvm.org/D100597
The patch enables the use of index type in vectors. It is a prerequisite to support vectorization for indexed Linalg operations. This refactoring became possible due to the newly introduced data layout infrastructure. The data layout of a module defines the bitwidth of the index type needed to verify bitcasts and similar vector operations.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D99948
Also factors out out-of-bounds mask generation from vector.transfer_read/write into a new MaterializeTransferMask pattern.
Differential Revision: https://reviews.llvm.org/D100001
This is in preparation for adding a new "mask" operand. The existing "masked" attribute was used to specify dimensions that may be out-of-bounds. Such transfers can be lowered to masked load/stores. The new "in_bounds" attribute is used to specify dimensions that are guaranteed to be within bounds. (Semantics is inverted.)
Differential Revision: https://reviews.llvm.org/D99639
Convert transfer_read ops with permutation maps into simpler
transfer_read with minority map + vector.braodcast and vector.transpose.
And transfer_read with leading dimensions broacast into transfer_read of
lower rank.
Differential Revision: https://reviews.llvm.org/D99019
This reverts commit 361b7d125b by Chris
Lattner <clattner@nondot.org> dated Fri Mar 19 21:22:15 2021 -0700.
The change to the greedy rewriter driver picking a different order was
made without adequate analysis of the trade-offs and experimentation. A
change like this has far reaching consequences on transformation
pipelines, and a major impact upstream and downstream. For eg., one
can’t be sure that it doesn’t slow down a large number of cases by small
amounts or create other issues. More discussion here:
https://llvm.discourse.group/t/speeding-up-canonicalize/3015/25
Reverting this so that improvements to the traversal order can be made
on a clean slate, in bigger steps, and higher bar.
Differential Revision: https://reviews.llvm.org/D99329
This reapplies b5d9a3c / https://reviews.llvm.org/D98609 with a one line fix in
processExistingConstants to skip() when erasing a constant we've already seen.
Original commit message:
1) Change the canonicalizer to walk the function in top-down order instead of
bottom-up order. This composes well with the "top down" nature of constant
folding and simplification, reducing iterations and re-evaluation of ops in
simple cases.
2) Explicitly enter existing constants into the OperationFolder table before
canonicalizing. Previously we would "constant fold" them and rematerialize
them, wastefully recreating a bunch fo constants, which lead to pointless
memory traffic.
Both changes together provide a 33% speedup for canonicalize on some mid-size
CIRCT examples.
One artifact of this change is that the constants generated in normal pattern
application get inserted at the top of the function as the patterns are applied.
Because of this, we get "inverted" constants more often, which is an aethetic
change to the IR but does permute some testcases.
Differential Revision: https://reviews.llvm.org/D99006
This reverts commit b5d9a3c923.
The commit introduced a memory error in canonicalization/operation
walking that is exposed when compiled with ASAN. It leads to crashes in
some "release" configurations.
Two changes:
1) Change the canonicalizer to walk the function in top-down order instead of
bottom-up order. This composes well with the "top down" nature of constant
folding and simplification, reducing iterations and re-evaluation of ops in
simple cases.
2) Explicitly enter existing constants into the OperationFolder table before
canonicalizing. Previously we would "constant fold" them and rematerialize
them, wastefully recreating a bunch fo constants, which lead to pointless
memory traffic.
Both changes together provide a 33% speedup for canonicalize on some mid-size
CIRCT examples.
One artifact of this change is that the constants generated in normal pattern
application get inserted at the top of the function as the patterns are applied.
Because of this, we get "inverted" constants more often, which is an aethetic
change to the IR but does permute some testcases.
Differential Revision: https://reviews.llvm.org/D98609
This patch introduces progressive lowering patterns for rewriting
vector.transfer_read/write to vector.load/store and vector.broadcast
in certain supported cases.
Reviewed By: dcaballe, nicolasvasilache
Differential Revision: https://reviews.llvm.org/D97822
With the new vector.load/store operations, there is no need to go through
unmasked transfer operations (which will canonicalized to l/s anyway).
Reviewed By: dcaballe
Differential Revision: https://reviews.llvm.org/D98056
Add a folder to rewrite a sequence such as:
```
%t1 = ...
%v = vector.transfer_read %t0[%c0...], {masked = [false...]} :
tensor<static_sizesxf32>, vector<static_sizesxf32>
%t2 = vector.transfer_write %v, %t1[%c0...] {masked = [false...]} :
vector<static_sizesxf32>, tensor<static_sizesxf32>
```
into:
```
%t0
```
The producer of t1 may or may not be DCE'd depending on whether it is a
block argument or has side effects.
Differential revision: https://reviews.llvm.org/D97934
Similar to mask-load/store and compress/expand, the gather and
scatter operation now allow for higher dimension uses. Note that
to support the mixed-type index, the new syntax is:
vector.gather %base [%i,%j] [%kvector] ....
The first client of this generalization is the sparse compiler,
which needs to define scatter and gathers on dense operands
of higher dimensions too.
Reviewed By: bixia
Differential Revision: https://reviews.llvm.org/D97422
This commit introduced a cyclic dependency:
Memref dialect depends on Standard because it used ConstantIndexOp.
Std depends on the MemRef dialect in its EDSC/Intrinsics.h
Working on a fix.
This reverts commit 8aa6c3765b.
Create the memref dialect and move several dialect-specific ops without
dependencies to other ops from std dialect to this dialect.
Moved ops:
AllocOp -> MemRef_AllocOp
AllocaOp -> MemRef_AllocaOp
DeallocOp -> MemRef_DeallocOp
MemRefCastOp -> MemRef_CastOp
GetGlobalMemRefOp -> MemRef_GetGlobalOp
GlobalMemRefOp -> MemRef_GlobalOp
PrefetchOp -> MemRef_PrefetchOp
ReshapeOp -> MemRef_ReshapeOp
StoreOp -> MemRef_StoreOp
TransposeOp -> MemRef_TransposeOp
ViewOp -> MemRef_ViewOp
The roadmap to split the memref dialect from std is discussed here:
https://llvm.discourse.group/t/rfc-split-the-memref-dialect-from-std/2667
Differential Revision: https://reviews.llvm.org/D96425
Some of the lowering of vector.contract didn't support integer case. Since
reduction of integer cannot accumulate we always break up the reduction op, it
should be merged by a separate canonicalization if possible.
Differential Revision: https://reviews.llvm.org/D96461
Currently, vector.contract joins the intermediate result and the accumulator
argument (of ranks K) using summation. We desire more joining operations ---
such as max --- to help vector.contract express reductions. This change extends
Vector_ContractionOp to take an optional attribute (called "kind", of enum type
CombiningKind) specifying the joining operation to be add/mul/min/max for int/fp
, and and/or/xor for int only. By default this attribute has value "add".
To implement this we also need to extend vector.outerproduct, since
vector.contract gets transformed to vector.outerproduct (and that to
vector.fma). The extension for vector.outerproduct is also an optional kind
attribute that uses the same enum type and possible values. The default is
"add". In case of max/min we transform vector.outerproduct to a combination of
compare and select.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D93280
Align the vector gather/scatter/expand/compress API with
the vector load/store/maskedload/maskedstore API.
Reviewed By: aartbik
Differential Revision: https://reviews.llvm.org/D96396
