* bool succeeded(Status)
- Return if the status corresponds to a success value.
* bool failed(Status)
- Return if the status corresponds to a failure value.
PiperOrigin-RevId: 237153884
Adds utility to convert slice bounds to a FlatAffineConstraints representation.
Adds utility to FlatAffineConstraints to promote loop IV symbol identifiers to dim identifiers.
PiperOrigin-RevId: 236973261
- change this for consistency - everything else similar takes/returns a
Function pointer - the FuncBuilder ctor,
Block/Value/Instruction::getFunction(), etc.
- saves a whole bunch of &s everywhere
PiperOrigin-RevId: 236928761
This fixes a bug: previously, during conversion function argument
attributes were neither beings passed through nor converted. This fix
extends DialectConversion to allow for simultaneous conversion of the
function type and the argument attributes.
This was important when lowering MLIR to LLVM where attribute
information (e.g. noalias) needs to be preserved in MLIR(LLVMDialect).
Longer run it seems reasonable that we want to convert both the
function attribute and its type and the argument attributes, but that
requires a small refactoring in Function.h to aggregate these three
fields in an inner struct, which will require some discussion.
PiperOrigin-RevId: 236709409
An analysis can be any class, but it must provide the following:
* A constructor for a given IR unit.
struct MyAnalysis {
// Compute this analysis with the provided module.
MyAnalysis(Module *module);
};
Analyses can be accessed from a Pass by calling either the 'getAnalysisResult<AnalysisT>' or 'getCachedAnalysisResult<AnalysisT>' methods. A FunctionPass may query for a cached analysis on the parent module with 'getCachedModuleAnalysisResult'. Similary, a ModulePass may query an analysis, it doesn't need to be cached, on a child function with 'getFunctionAnalysisResult'.
By default, when running a pass all cached analyses are set to be invalidated. If no transformation was performed, a pass can use the method 'markAllAnalysesPreserved' to preserve all analysis results. As noted above, preserving specific analyses is not yet supported.
PiperOrigin-RevId: 236505642
This CL changes dialect op source files (.h, .cpp, .td) to follow the following
convention:
<full-dialect-name>/<dialect-namespace>Ops.{h|cpp|td}
Builtin and standard dialects are specially treated, though. Both of them do
not have dialect namespace; the former is still named as BuiltinOps.* and the
latter is named as Ops.*.
Purely mechanical. NFC.
PiperOrigin-RevId: 236371358
*) Breaks fusion pass into multiple sub passes over nodes in data dependence graph:
- first pass fuses single-use producers into their unique consumer.
- second pass enables fusing for input-reuse by fusing sibling nodes which read from the same memref, but which do not share dependence edges.
- third pass fuses remaining producers into their consumers (Note that the sibling fusion pass may have transformed a producer with multiple uses into a single-use producer).
*) Fusion for input reuse is enabled by computing a sibling node slice using the load/load accesses to the same memref, and fusion safety is guaranteed by checking that the sibling node memref write region (to a different memref) is preserved.
*) Enables output vector and output matrix computations from KFAC patches-second-moment operation to fuse into a single loop nest and reuse input from the image patches operation.
*) Adds a generic loop utilitiy for finding all sequential loops in a loop nest.
*) Adds and updates unit tests.
PiperOrigin-RevId: 236350987
LoopFusion
- getConstDifference in LoopFusion is pending a refactoring to handle bounds
with min's and max's; it currently asserts on some useful test cases that we
want to experiment with. This CL changes getSliceBounds to be more
conservative so as to not trigger the assertion. Filed b/126426796 to track this.
PiperOrigin-RevId: 235826538
- clean up loop fusion CL options for promoting local buffers to fast memory
space
- add parameters to loop fusion pass instantiation
PiperOrigin-RevId: 235813419
This CL adds a primitive to perform stripmining of a loop by a given factor and
sinking it under multiple target loops.
In turn this is used to implement imperfectly nested loop tiling (with interchange) by repeatedly calling the stripmineSink primitive.
The API returns the point loops and allows repeated invocations of tiling to achieve declarative, multi-level, imperfectly-nested tiling.
Note that this CL is only concerned with the mechanical aspects and does not worry about analysis and legality.
The API is demonstrated in an example which creates an EDSC block, emits the corresponding MLIR and applies imperfectly-nested tiling:
```cpp
auto block = edsc::block({
For(ArrayRef<edsc::Expr>{i, j}, {zero, zero}, {M, N}, {one, one}, {
For(k1, zero, O, one, {
C({i, j, k1}) = A({i, j, k1}) + B({i, j, k1})
}),
For(k2, zero, O, one, {
C({i, j, k2}) = A({i, j, k2}) + B({i, j, k2})
}),
}),
});
// clang-format on
emitter.emitStmts(block.getBody());
auto l_i = emitter.getAffineForOp(i), l_j = emitter.getAffineForOp(j),
l_k1 = emitter.getAffineForOp(k1), l_k2 = emitter.getAffineForOp(k2);
auto indicesL1 = mlir::tile({l_i, l_j}, {512, 1024}, {l_k1, l_k2});
auto l_ii1 = indicesL1[0][0], l_jj1 = indicesL1[1][0];
mlir::tile({l_jj1, l_ii1}, {32, 16}, l_jj1);
```
The edsc::Expr for the induction variables (i, j, k_1, k_2) provide the programmatic hooks from which tiling can be applied declaratively.
PiperOrigin-RevId: 235548228
Analysis - NFC
- refactor AffineExprFlattener (-> SimpleAffineExprFlattener) so that it
doesn't depend on FlatAffineConstraints, and so that FlatAffineConstraints
could be moved out of IR/; the simplification that the IR needs for
AffineExpr's doesn't depend on FlatAffineConstraints
- have AffineExprFlattener derive from SimpleAffineExprFlattener to use for
all Analysis/Transforms purposes; override addLocalFloorDivId in the derived
class
- turn addAffineForOpDomain into a method on FlatAffineConstraints
- turn AffineForOp::getAsValueMap into an AffineValueMap ctor
PiperOrigin-RevId: 235283610
- compute slices precisely where the destination iteration depends on multiple source
iterations (instead of over-approximating to the whole source loop extent)
- update unionBoundingBox to deal with input with non-matching symbols
- reenable disabled backend test case
PiperOrigin-RevId: 234714069
- hoist DMAs past all loops immediately surrounding the region that the latter
is invariant on - do this at DMA generation time itself
PiperOrigin-RevId: 234628447
Expose the result types of edsc::Expr, which are now stored for all types of
Exprs and not only for the variadic ones. Require return types when an Expr is
constructed, if it will ever have some. An empty return type list is
interpreted as an Expr that does not create a value (e.g. `return` or `store`).
Conceptually, all edss::Exprs are now typed, with the type being a (potentially
empty) tuple of return types. Unbound expressions and Bindables must now be
constructed with a specific type they will take. This makes EDSC less
evidently type-polymorphic, but we can still write generic code such as
Expr sumOfSquares(Expr lhs, Expr rhs) { return lhs * lhs + rhs * rhs; }
and use it to construct different typed expressions as
sumOfSquares(Bindable(IndexType::get(ctx)), Bindable(IndexType::get(ctx)));
sumOfSquares(Bindable(FloatType::getF32(ctx)),
Bindable(FloatType::getF32(ctx)));
On the positive side, we get the following.
1. We can now perform type checking when constructing Exprs rather than during
MLIR emission. Nevertheless, this is still duplicates the Op::verify()
until we can factor out type checking from that.
2. MLIREmitter is significantly simplified.
3. ExprKind enum is only used for actual kinds of expressions. Data structures
are converging with AbstractOperation, and the users can now create a
VariadicExpr("canonical_op_name", {types}, {exprs}) for any operation, even
an unregistered one without having to extend the enum and make pervasive
changes to EDSCs.
On the negative side, we get the following.
1. Typed bindables are more verbose, even in Python.
2. We lose the ability to do print debugging for higher-level EDSC abstractions
that are implemented as multiple MLIR Ops, for example logical disjunction.
This is the step 2/n towards making EDSC extensible.
***
Move MLIR Op construction from MLIREmitter::emitExpr to Expr::build since Expr
now has sufficient information to build itself.
This is the step 3/n towards making EDSC extensible.
Both of these strive to minimize the amount of irrelevant changes. In
particular, this introduces more complex pretty-printing for affine and binary
expression to make sure tests continue to pass. It also relies on string
comparison to identify specific operations that an Expr produces.
PiperOrigin-RevId: 234609882
EDSC currently implement a block as a statement that is itself a list of
statements. This suffers from two modeling problems: (1) these blocks are not
addressable, i.e. one cannot create an instruction where thus constructed block
is a successor; (2) they support block nesting, which is not supported by MLIR
blocks. Furthermore, emitting such "compound statement" (misleadingly named
`Block` in Python bindings) does not actually produce a new Block in the IR.
Implement support for creating actual IR Blocks in EDSC. In particular, define
a new StmtBlock EDSC class that is neither an Expr nor a Stmt but contains a
list of Stmts. Additionally, StmtBlock may have (early-) typed arguments.
These arguments are Bindable expressions that can be used inside the block.
Provide two calls in the MLIREmitter, `emitBlock` that actually emits a new
block and `emitBlockBody` that only emits the instructions contained in the
block without creating a new block. In the latter case, the instructions must
not use block arguments.
Update Python bindings to make it clear when instruction emission happens
without creating a new block.
PiperOrigin-RevId: 234556474
generation pass to make it drop certain assumptions, complete TODOs.
- multiple fixes for getMemoryFootprintBytes
- pass loopDepth correctly from getMemoryFootprintBytes()
- use union while computing memory footprints
- bug fixes for addAffineForOpDomain
- take into account loop step
- add domains of other loop IVs in turn that might have been used in the bounds
- dma-generate: drop assumption of "non-unit stride loops being tile space loops
and skipping those and recursing to inner depths"; DMA generation is now purely
based on available fast mem capacity and memory footprint's calculated
- handle memory region compute failures/bailouts correctly from dma-generate
- loop tiling cleanup/NFC
- update some debug and error messages to use emitNote/emitError in
pipeline-data-transfer pass - NFC
PiperOrigin-RevId: 234245969
*) Adds utility to LoopUtils to perform loop interchange of two AffineForOps.
*) Adds utility to LoopUtils to sink a loop to a specified depth within a loop nest, using a series of loop interchanges.
*) Computes dependences between all loads and stores in the loop nest, and classifies each loop as parallel or sequential.
*) Computes loop interchange permutation required to sink sequential loops (and raise parallel loop nests) while preserving relative order among them.
*) Checks each dependence against the permutation to make sure that dependences would not be violated by the loop interchange transformation.
*) Calls loop interchange in LoopFusion pass on consumer loop nests before fusing in producers, sinking loops with loop carried dependences deeper into the consumer loop nest.
*) Adds and updates related unit tests.
PiperOrigin-RevId: 234158370
Function types are built-in in MLIR and affect the validity of the IR itself.
However, advanced target dialects such as the LLVM IR dialect may include
custom function types. Until now, dialect conversion was expecting function
types not to be converted to the custom type: although the signatures was
allowed to change, the outer type must have been an mlir::FunctionType. This
effectively prevented dialect conversion from creating instructions that
operate on values of the custom function type.
Dissociate function signature conversion from general type conversion.
Function signature conversion must still produce an mlir::FunctionType and is
used in places where built-in types are required to make IR valid. General
type conversion is used for SSA values, including function and block arguments
and function results.
Exercise this behavior in the LLVM IR dialect conversion by converting function
types to LLVM IR function pointer types. The pointer to a function is chosen
to provide consistent lowering of higher-order functions: while it is possible
to have a value of function type, it is not possible to create a function type
accepting a returning another function type.
PiperOrigin-RevId: 234124494
for dma-generate, loop-unroll.
- add -tile-sizes command line option for loop tiling to specify different tile
sizes for loops in a band
- clean up command line options for loop-unroll, dma-generate (remove
cl::hidden)
PiperOrigin-RevId: 234006232
- for the DMA transfers being pipelined through double buffering, generate
deallocs for the double buffers being alloc'ed
This change is along the lines of cl/233502632. We initially wanted to experiment with
scoped allocation - so the deallocation's were usually not necessary; however, they are
needed even with scoped allocations in some situations - for eg. when the enclosing loop
gets unrolled. The dealloc serves as an end of lifetime marker.
PiperOrigin-RevId: 233653463
In the current state, edsc::Expr and edsc::Stmt overload operators to construct
other Exprs and Stmts. This includes some unconventional overloads of the
`operator==` to create a comparison expression and of the `operator!` to create
a negation expression. This situation could lead to unpleasant surprises where
the code does not behave like expected. Make all Expr and Stmt construction
operators free functions and move them to the `edsc::op` namespace. Callers
willing to use these operators must explicitly include them with the `using`
declaration. This can be done in some local scope.
Additionally, we currently emit signed comparisons for order-comparison
operators. With namespaces, we can later introduce two sets of operators in
different namespace, e.g. `edsc::op::sign` and `edsc::op::unsign` to clearly
state which kind of comparison is implied.
PiperOrigin-RevId: 233578674
- for the DMA buffers being allocated (and their tags), generate corresponding deallocs
- minor related update to replaceAllMemRefUsesWith and PipelineDataTransfer pass
Code generation for DMA transfers was being done with the initial simplifying
assumption that the alloc's would map to scoped allocations, and so no
deallocations would be necessary. Drop this assumption to generalize. Note that
even with scoped allocations, unrolling loops that have scoped allocations
could create a series of allocations and exhaustion of fast memory. Having a
end of lifetime marker like a dealloc in fact allows creating new scopes if
necessary when lowering to a backend and still utilize scoped allocation.
DMA buffers created by -dma-generate are guaranteed to have either
non-overlapping lifetimes or nested lifetimes.
PiperOrigin-RevId: 233502632
*) Adds parameter to public API of MemRefRegion::compute for passing in the slice loop bounds to compute the memref region of the loop nest slice.
*) Exposes public method MemRefRegion::getRegionSize for computing the size of the memref region in bytes.
PiperOrigin-RevId: 232706165
* AffineStructures has moved to IR.
* simplifyAffineExpr/simplifyAffineMap/getFlattenedAffineExpr have moved to IR.
* makeComposedAffineApply/fullyComposeAffineMapAndOperands have moved to AffineOps.
* ComposeAffineMaps is replaced by AffineApplyOp::canonicalize and deleted.
PiperOrigin-RevId: 232586468
*) After a private memref buffer is created for a fused loop nest, dependences on the old memref are reduced, which can open up fusion opportunities. In these cases, users of the old memref are added back to the worklist to be reconsidered for fusion.
*) Fixed a bug in fusion insertion point dependence check where the memref being privatized was being skipped from the check.
PiperOrigin-RevId: 232477853
- use getAccessMap() instead of repeating it
- fold getMemRefRegion into MemRefRegion ctor (more natural, avoid heap
allocation and unique_ptr where possible)
- change extractForInductionVars - MutableArrayRef -> ArrayRef for the
arguments. Since the method is just returning copies of 'Value *', the client
can't mutate the pointers themselves; it's fine to mutate the 'Value''s
themselves, but that doesn't mutate the pointers to those.
- change the way extractForInductionVars returns (see b/123437690)
PiperOrigin-RevId: 232359277
loops), (2) take into account fast memory space capacity and lower 'dmaDepth'
to fit, (3) add location information for debug info / errors
- change dma-generate pass to work on blocks of instructions (start/end
iterators) instead of 'for' loops; complete TODOs - allows DMA generation for
straightline blocks of operation instructions interspersed b/w loops
- take into account fast memory capacity: check whether memory footprint fits
in fastMemoryCapacity parameter, and recurse/lower the depth at which DMA
generation is performed until it does fit in the provided memory
- add location information to MemRefRegion; any insufficient fast memory
capacity errors or debug info w.r.t dma generation shows location information
- allow DMA generation pass to be instantiated with a fast memory capacity
option (besides command line flag)
- change getMemRefRegion to return unique_ptr's
- change getMemRefFootprintBytes to work on a 'Block' instead of 'ForInst'
- other helper methods; add postDomInstFilter option for
replaceAllMemRefUsesWith; drop forInst->walkOps, add Block::walkOps methods
Eg. output
$ mlir-opt -dma-generate -dma-fast-mem-capacity=1 /tmp/single.mlir
/tmp/single.mlir:9:13: error: Total size of all DMA buffers' for this block exceeds fast memory capacity
for %i3 = (d0) -> (d0)(%i1) to (d0) -> (d0 + 32)(%i1) {
^
$ mlir-opt -debug-only=dma-generate -dma-generate -dma-fast-mem-capacity=400 /tmp/single.mlir
/tmp/single.mlir:9:13: note: 8 KiB of DMA buffers in fast memory space for this block
for %i3 = (d0) -> (d0)(%i1) to (d0) -> (d0 + 32)(%i1) {
PiperOrigin-RevId: 232297044
- fusion already includes the necessary analysis to create small/local buffers
post fusion; allocate these buffers in a higher memory space if the necessary
pass parameters are provided (threshold size, memory space id)
- although there will be a separate utility at some point to directly detect
and promote small local buffers to higher memory spaces, doing it while fusion
when possible is much less expensive, comes free with fusion analysis, and covers
a key common case.
PiperOrigin-RevId: 232063894
This CL applies the following simplifications to EDSCs:
1. Rename Block to StmtList because an MLIR Block is a different, not yet
supported, notion;
2. Rework Bindable to drop specific storage and just use it as a simple wrapper
around Expr. The only value of Bindable is to force a static cast when used by
the user to bind into the emitter. For all intended purposes, Bindable is just
a lightweight check that an Expr is Unbound. This simplifies usage and reduces
the API footprint. After playing with it for some time, it wasn't worth the API
cognition overhead;
3. Replace makeExprs and makeBindables by makeNewExprs and copyExprs which is
more explicit and less easy to misuse;
4. Add generally useful functionality to MLIREmitter:
a. expose zero and one for the ubiquitous common lower bounds and step;
b. add support to create already bound Exprs for all function arguments as
well as shapes and views for Exprs bound to memrefs.
5. Delete Stmt::operator= and replace by a `Stmt::set` method which is more
explicit.
6. Make Stmt::operator Expr() explicit.
7. Indexed.indices assertions are removed to pave the way for expressing slices
and views as well as to work with 0-D memrefs.
The CL plugs those simplifications with TableGen and allows emitting a full MLIR function for
pointwise add.
This "x.add" op is both type and rank-agnostic (by allowing ArrayRef of Expr
passed to For loops) and opens the door to spinning up a composable library of
existing and custom ops that should automate a lot of the tedious work in
TF/XLA -> MLIR.
Testing needs to be significantly improved but can be done in a separate CL.
PiperOrigin-RevId: 231982325
A performance issue was reported due to the usage of NestedMatcher in
ComposeAffineMaps. The main culprit was the ubiquitous copies that were
occuring when appending even a single element in `matchOne`.
This CL generally simplifies the implementation and removes one level of indirection by getting rid of
auxiliary storage as well as simplifying the API.
The users of the API are updated accordingly.
The implementation was tested on a heavily unrolled example with
ComposeAffineMaps and is now close in performance with an implementation based
on stateless InstWalker.
As a reminder, the whole ComposeAffineMaps pass is slated to disappear but the bug report was very useful as a stress test for NestedMatchers.
Lastly, the following cleanups reported by @aminim were addressed:
1. make NestedPatternContext scoped within runFunction rather than at the Pass level. This was caused by a previous misunderstanding of Pass lifetime;
2. use defensive assertions in the constructor of NestedPatternContext to make it clear a unique such locally scoped context is allowed to exist.
PiperOrigin-RevId: 231781279
a trivial inst walker :-) (reduces pass time from several minutes non-terminating to 120ms) - (fixes b/123541184)
- use a simple 7-line inst walker to collect affine_apply op's instead of the nested
matcher; -compose-affine-maps pass runs in 120ms now instead of 5 minutes + (non-
terminating / out of memory) - on a realistic test case that is 20,000 lines 12-d
loop nest
- this CL is also pushing for simple existing/standard patterns unless there
is a real efficiency issue (OTOH, fixing nested matcher to address this issue requires
cl/231400521)
- the improvement is from swapping out the nested walker as opposed to from a bug
or anything else that this CL changes
- update stale comment
PiperOrigin-RevId: 231623619
Cleanup a usage of functional::map that is deemed too obscure in
`reindexAffineIndices`. Also fix a stale comment in `reindexAffineIndices`.
PiperOrigin-RevId: 231211184
Addresses b/122486036
This CL addresses some leftover crumbs in AffineMap and IntegerSet by removing
the Null method and cleaning up the constructors.
As the ::Null uses were tracked down, opportunities appeared to untangle some
of the Parsing logic and make it explicit where AffineMap/IntegerSet have
ambiguous syntax. Previously, ambiguous cases were hidden behind the implicit
pointer values of AffineMap* and IntegerSet* that were passed as function
parameters. Depending the values of those pointers one of 3 behaviors could
occur.
This parsing logic convolution is one of the rare cases where I would advocate
for code duplication. The more proper fix would be to make the syntax
unambiguous or to allow some lookahead.
PiperOrigin-RevId: 231058512
This CL follows up on a memory leak issue related to SmallVector growth that
escapes the BumpPtrAllocator.
The fix is to properly use ArrayRef and placement new to define away the
issue.
The following renaming is also applied:
1. MLFunctionMatcher -> NestedPattern
2. MLFunctionMatches -> NestedMatch
As a consequence all allocations are now guaranteed to live on the BumpPtrAllocator.
PiperOrigin-RevId: 231047766
Example:
dma-generate options:
-dma-fast-mem-capacity - Set fast memory space ...
-dma-fast-mem-space=<uint> - Set fast memory space ...
loop-fusion options:
-fusion-compute-tolerance=<number> - Fractional increase in ...
-fusion-maximal - Enables maximal loop fusion
loop-tile options:
-tile-size=<uint> - Use this tile size for ...
loop-unroll options:
-unroll-factor=<uint> - Use this unroll factor ...
-unroll-full - Fully unroll loops
-unroll-full-threshold=<uint> - Unroll all loops with ...
-unroll-num-reps=<uint> - Unroll innermost loops ...
loop-unroll-jam options:
-unroll-jam-factor=<uint> - Use this unroll jam factor ...
PiperOrigin-RevId: 231019363
index remapping
- generate a sequence of single result affine_apply's for the index remapping
(instead of one multi result affine_apply)
- update dma-generate and loop-fusion test cases; while on this, change test cases
to use single result affine apply ops
- some fusion comment fix/cleanup
PiperOrigin-RevId: 230985830
- Update createAffineComputationSlice to generate a sequence of single result
affine apply ops instead of one multi-result affine apply
- update pipeline-data-transfer test case; while on this, also update the test
case to use only single result affine maps, and make it more robust to
change.
PiperOrigin-RevId: 230965478
This commit introduces a generic dialect conversion/lowering/legalization pass
and illustrates it on StandardOps->LLVMIR conversion.
It partially reuses the PatternRewriter infrastructure and adds the following
functionality:
- an actual pass;
- non-default pattern constructors;
- one-to-many rewrites;
- rewriting terminators with successors;
- not applying patterns iteratively (unlike the existing greedy rewrite driver);
- ability to change function signature;
- ability to change basic block argument types.
The latter two things required, given the existing API, to create new functions
in the same module. Eventually, this should converge with the rest of
PatternRewriter. However, we may want to keep two pass versions: "heavy" with
function/block argument conversion and "light" that only touches operations.
This pass creates new functions within a module as a means to change function
signature, then creates new blocks with converted argument types in the new
function. Then, it traverses the CFG in DFS-preorder to make sure defs are
converted before uses in the dominated blocks. The generic pass has a minimal
interface with two hooks: one to fill in the set of patterns, and another one
to convert types for functions and blocks. The patterns are defined as
separate classes that can be table-generated in the future.
The LLVM IR lowering pass partially inherits from the existing LLVM IR
translator, in particular for type conversion. It defines a conversion pattern
template, instantiated for different operations, and is a good candidate for
tablegen. The lowering does not yet support loads and stores and is not
connected to the translator as it would have broken the existing flows. Future
patches will add missing support before switching the translator in a single
patch.
PiperOrigin-RevId: 230951202
- introduce a way to compute union using symbolic rectangular bounding boxes
- handle multiple load/store op's to the same memref by taking a union of the regions
- command-line argument to provide capacity of the fast memory space
- minor change to replaceAllMemRefUsesWith to not generate affine_apply if the
supplied index remap was identity
PiperOrigin-RevId: 230848185
canonicalizations of operations. The ultimate important user of this is
going to be a funcBuilder->foldOrCreate<YourOp>(...) API, but for now it
is just a more convenient way to write certain classes of canonicalizations
(see the change in StandardOps.cpp).
NFC.
PiperOrigin-RevId: 230770021
- switch some debug info to emitError
- use a single constant op for zero index to make it easier to write/update
test cases; avoid creating new constant op's for common zero index cases
- test case cleanup
This is in preparation for an upcoming major update to this pass.
PiperOrigin-RevId: 230728379
- update fusion cost model to fuse while tolerating a certain amount of redundant
computation; add cl option -fusion-compute-tolerance
evaluate memory footprint and intermediate memory reduction
- emit debug info from -loop-fusion showing what was fused and why
- introduce function to compute memory footprint for a loop nest
- getMemRefRegion readability update - NFC
PiperOrigin-RevId: 230541857
- unrolling a single iteration loop by a factor of one should promote its body
into its parent; this makes it consistent with the behavior/expectation that
unrolling a loop by a factor equal to its trip count makes the loop go away.
PiperOrigin-RevId: 230426499
- ForInst::walkOps will also be used in an upcoming CL (cl/229438679); better to have
this instead of deriving from the InstWalker
PiperOrigin-RevId: 230413820
- the size of the private memref created for the slice should be based on
the memref region accessed at the depth at which the slice is being
materialized, i.e., symbolic in the outer IVs up until that depth, as opposed
to the region accessed based on the entire domain.
- leads to a significant contraction of the temporary / intermediate memref
whenever the memref isn't reduced to a single scalar (through store fwd'ing).
Other changes
- update to promoteIfSingleIteration - avoid introducing unnecessary identity
map affine_apply from IV; makes it much easier to write and read test cases
and pass output for all passes that use promoteIfSingleIteration; loop-fusion
test cases become much simpler
- fix replaceAllMemrefUsesWith bug that was exposed by the above update -
'domInstFilter' could be one of the ops erased due to a memref replacement in
it.
- fix getConstantBoundOnDimSize bug: a division by the coefficient of the identifier was
missing (the latter need not always be 1); add lbFloorDivisors output argument
- rename getBoundingConstantSizeAndShape -> getConstantBoundingSizeAndShape
PiperOrigin-RevId: 230405218
*) Do not remove loop nests which write to memrefs which escape the function.
*) Do not remove memrefs which escape the function (e.g. are used in the return instruction).
PiperOrigin-RevId: 230398630
This CL performs a bunch of cleanups related to EDSCs that are generally
useful in the context of using them with a simple wrapping C API (not in this
CL) and with simple language bindings to Python and Swift.
PiperOrigin-RevId: 230066505
*) Enables reduction of private memref size based on MemRef region accessed by fused slice.
*) Enables maximal fusion by creating a private memref to break a fusion-preventing dependence.
*) Adds maximal fusion flag to enable fusing as much as possible (though it still fuses the minimum cost computation slice).
PiperOrigin-RevId: 229936698
This CL fixes a misunderstanding in how to build DimOp which triggered
execution issues in the CPU path.
The problem is that, given a `memref<?x4x?x8x?xf32>`, the expressions to
construct the dynamic dimensions should be:
`dim %arg, 0 : memref<?x4x?x8x?xf32>`
`dim %arg, 2 : memref<?x4x?x8x?xf32>`
and
`dim %arg, 4 : memref<?x4x?x8x?xf32>`
Before this CL, we wold construct:
`dim %arg, 0 : memref<?x4x?x8x?xf32>`
`dim %arg, 1 : memref<?x4x?x8x?xf32>`
`dim %arg, 2 : memref<?x4x?x8x?xf32>`
and expect the other dimensions to be constants.
This assumption seems consistent at first glance with the syntax of alloc:
```
%tensor = alloc(%M, %N, %O) : memref<?x4x?x8x?xf32>
```
But this was actuallyincorrect.
This CL also makes the relevant functions available to EDSCs and removes
duplication of the incorrect function.
PiperOrigin-RevId: 229622766
*) Adds support for fusing into consumer loop nests with multiple loads from the same memref.
*) Adds support for reducing slice loop trip count by projecting out destination loop IVs greater than destination loop depth.
*) Removes dependence on src loop depth and simplifies cost model computation.
PiperOrigin-RevId: 229575126
This allows load, store and ForNest to be used with both Expr and Bindable.
This simplifies writing generic pieces of MLIR snippet.
For instance, a generic pointwise add can now be written:
```cpp
// Different Bindable ivs, one per loop in the loop nest.
auto ivs = makeBindables(shapeA.size());
Bindable zero, one;
// Same bindable, all equal to `zero`.
SmallVector<Bindable, 8> zeros(ivs.size(), zero);
// Same bindable, all equal to `one`.
SmallVector<Bindable, 8> ones(ivs.size(), one);
// clang-format off
Bindable A, B, C;
Stmt scalarA, scalarB, tmp;
Stmt block = edsc::Block({
ForNest(ivs, zeros, shapeA, ones, {
scalarA = load(A, ivs),
scalarB = load(B, ivs),
tmp = scalarA + scalarB,
store(tmp, C, ivs)
}),
});
// clang-format on
```
This CL also adds some extra support for pretty printing that will be used in
a future CL when we introduce standalone testing of EDSCs. At the momen twe
are lacking the basic infrastructure to write such tests.
PiperOrigin-RevId: 229375850
*) LoopFusion: Adds fusion cost function which compares the cost of the fused loop nest, with the cost of the two unfused loop nests to determine if it is profitable to fuse the candidate loop nests. The fusion cost function is run for various combinations for src/dst loop depths attempting find the minimum cost setting for src/dst loop depths which does not increase the computational cost when the loop nests are fused. Combinations of src/dst loop depth are evaluated attempting to maximize loop depth (i.e. take a bigger computation slice from the source loop nest, and insert it deeper in the destination loop nest for better locality).
*) LoopFusion: Adds utility to compute op instance count for loop nests, sliced loop nests, and to compute the cost of a loop nest fused with another sliced loop nest.
*) LoopFusion: canonicalizes slice bound AffineMaps (and updates related tests).
*) Analysis::Utils: Splits getBackwardComputationSlice into two functions: one which calculates and returns the slice loop bounds for analysis by LoopFusion, and the other for insertion of the computation slice (ones fusion has calculated the min-cost src/dst loop depths).
*) Test: Adds multiple unit tests to test the new functionality.
PiperOrigin-RevId: 229219757
This CL adds a short term remedy to an issue that was found during execution
tests.
Lowering of vector transfer ops uses the permutation map to determine which
ForInst have been super-vectorized. During materialization to HW vector sizes
however, some of those dimensions may be fully unrolled and do not appear in
the permutation map.
Such dimensions were then not clipped and may have accessed out of bounds.
This CL conservatively clips all dimensions to ensure no out of bounds access.
The longer term solution is still up for debate but will probably require
either passing more information between Materialization and lowering, or just
merging the 2 passes.
PiperOrigin-RevId: 228980787
This CL is the 6th and last on the path to simplifying AffineMap composition.
This removes `AffineValueMap::forwardSubstitutions` and replaces it by simple
calls to `fullyComposeAffineMapAndOperands`.
PiperOrigin-RevId: 228962580
This CL is the 5th on the path to simplifying AffineMap composition.
This removes the distinction between normalized single-result AffineMap and
more general composed multi-result map.
One nice byproduct of making the implementation driven by single-result is
that the multi-result extension is a trivial change: the implementation is
still single-result and we just use:
```
unsigned idx = getIndexOf(...);
map.getResult(idx);
```
This CL also fixes an AffineNormalizer implementation issue related to symbols.
Namely it stops performing substitutions on symbols in AffineNormalizer and
instead concatenates them all to be consistent with the call to
`AffineMap::compose(AffineMap)`. This latter call to `compose` cannot perform
simplifications of symbols coming from different maps based on positions only:
i.e. dims are applied and renumbered but symbols must be concatenated.
The only way to determine whether symbols from different AffineApply are the
same is to look at the concrete values. The canonicalizeMapAndOperands is thus
extended with behavior to support replacing operands that appear multiple
times.
Lastly, this CL demonstrates that the implementation is correct by rewriting
ComposeAffineMaps using only `makeComposedAffineApply`. The implementation
uses a matcher because AffineApplyOp are introduced as composed operations on
the fly instead of iteratively forwardSubstituting. For this purpose, a walker
would revisit freshly introduced AffineApplyOp. Regardless, ComposeAffineMaps
is scheduled to disappear, this CL replaces the implementation based on
iterative `forwardSubstitute` by a composed-by-construction
`makeComposedAffineApply`.
Remaining calls to `forwardSubstitute` will be removed in the next CL.
PiperOrigin-RevId: 228830443
This implements the lowering of `floordiv`, `ceildiv` and `mod` operators from
affine expressions to the arithmetic primitive operations. Integer division
rules in affine expressions explicitly require rounding towards either negative
or positive infinity unlike machine implementations that round towards zero.
In the general case, implementing `floordiv` and `ceildiv` using machine signed
division requires computing both the quotient and the remainder. When the
divisor is positive, this can be simplified by adjusting the dividend and the
quotient by one and switching signs.
In the current use cases, we are unlikely to encounter affine expressions with
negative divisors (affine divisions appear in loop transformations such as
tiling that guarantee that divisors are positive by construction). Therefore,
it is reasonable to use branch-free single-division implementation. In case of
affine maps, divisors can only be literals so we can check the sign and
implement the case for negative divisors when the need arises.
The affine lowering pass can still fail when applied to semi-affine maps
(division or modulo by a symbol).
PiperOrigin-RevId: 228668181
- the double buffer should be indexed (iv floordiv step) % 2 and NOT (iv % 2);
step wasn't being accounted for.
- fix test cases, enable failing test cases
PiperOrigin-RevId: 228635726
Supervectorization does not plan on handling multi-result AffineMaps and
non-canonical chains of > 1 AffineApplyOp.
This CL uses the simpler single-result unbounded AffineApplyOp in the
MaterializeVectors pass.
PiperOrigin-RevId: 228469085
This CL is the 2nd on the path to simplifying AffineMap composition.
This CL uses the now accepted `AffineExpr::compose(AffineMap)` to
implement `AffineMap::compose(AffineMap)`.
Implications of keeping the simplification function in
Analysis are documented where relevant.
PiperOrigin-RevId: 228276646
- refactor toAffineFromEq and the code surrounding it; refactor code into
FlatAffineConstraints::getSliceBounds
- add FlatAffineConstraints methods to detect identifiers as mod's and div's of other
identifiers
- add FlatAffineConstraints::getConstantLower/UpperBound
- Address b/122118218 (don't assert on invalid fusion depths cmdline flags -
instead, don't do anything; change cmdline flags
src-loop-depth -> fusion-src-loop-depth
- AffineExpr/Map print method update: don't fail on null instances (since we have
a wrapper around a pointer, it's avoidable); rationale: dump/print methods should
never fail if possible.
- Update memref-dataflow-opt to add an optimization to avoid a unnecessary call to
IsRangeOneToOne when it's trivially going to be true.
- Add additional test cases to exercise the new support
- update a few existing test cases since the maps are now generated uniformly with
all destination loop operands appearing for the backward slice
- Fix projectOut - fix wrong range for getBestElimCandidate.
- Fix for getConstantBoundOnDimSize() - didn't show up in any test cases since
we didn't have any non-hyperrectangular ones.
PiperOrigin-RevId: 228265152
- Detect 'mod' to replace the combination of floordiv, mul, and subtract when
possible at construction time; when 'c' is a power of two, this reduces the number of
operations; also more compact and readable. Update simplifyAdd for this.
On a side note:
- with the affine expr flattening we have, a mod expression like d0 mod c
would be flattened into d0 - c * q, c * q <= d0 <= c*q + c - 1, with 'q'
being added as the local variable (q = d0 floordiv c); as a result, a mod
was turned into a floordiv whenever the expression was reconstructed back,
i.e., as d0 - c * (d0 floordiv c); as a result of this change, we recover
the mod back.
- rename SimplifyAffineExpr -> SimplifyAffineStructures (pass had been renamed but
the file hadn't been).
PiperOrigin-RevId: 228258120
- when SSAValue/MLValue existed, code at several places was forced to create additional
aggregate temporaries of SmallVector<SSAValue/MLValue> to handle the conversion; get
rid of such redundant code
- use filling ctors instead of explicit loops
- for smallvectors, change insert(list.end(), ...) -> append(...
- improve comments at various places
- turn getMemRefAccess into MemRefAccess ctor and drop duplicated
getMemRefAccess. In the next CL, provide getAccess() accessors for load,
store, DMA op's to return a MemRefAccess.
PiperOrigin-RevId: 228243638
Supervectorization does not plan on handling multi-result AffineMaps and
non-canonical chains of > 1 AffineApplyOp.
This CL introduces a simpler abstraction and composition of single-result
unbounded AffineApplyOp by using the existing unbound AffineMap composition.
This CL adds a simple API call and relevant tests:
```c++
OpPointer<AffineApplyOp> makeNormalizedAffineApply(
FuncBuilder *b, Location loc, AffineMap map, ArrayRef<Value*> operands);
```
which creates a single-result unbounded AffineApplyOp.
The operands of AffineApplyOp are not themselves results of AffineApplyOp by
consrtuction.
This represent the simplest possible interface to complement the composition
of (mathematical) AffineMap, for the cases when we are interested in applying
it to Value*.
In this CL the composed AffineMap is not compressed (i.e. there exist operands
that are not part of the result). A followup commit will compress to normal
form.
The single-result unbounded AffineApplyOp abstraction will be used in a
followup CL to support the MaterializeVectors pass.
PiperOrigin-RevId: 227879021
Even though it is unexpected except in pathological cases, a nullptr clone may
be returned. This CL handles the nullptr return gracefuly.
PiperOrigin-RevId: 227764615
The strict requirement (i.e. at least 2 HW vectors in a super-vector) was a
premature optimization to avoid interfering with other vector code potentially
introduced via other means.
This CL avoids this premature optimization and the spurious errors it causes
when super-vector size == HW vector size (which is a possible corner case).
This may be revisited in the future.
PiperOrigin-RevId: 227763966
This corner was found when stress testing with a functional end-to-end CPU
path. In the case where the hardware vector size is 1x...x1 the `keep` vector
is empty and would result a crash.
While there is no reason to expect a 1x...x1 HW vector in practice, this case
can just gracefully degrade to scalar, which is what this CL allows.
PiperOrigin-RevId: 227761097
This change is mechanical and merges the LowerAffineApplyPass and
LowerIfAndForPass into a single LowerAffinePass. It makes a step towards
defining an "affine dialect" that would contain all polyhedral-related
constructs. The motivation for merging these two passes is based on retiring
MLFunctions and, eventually, transforming If and For statements into regular
operations. After that happens, LowerAffinePass becomes yet another
legalization.
PiperOrigin-RevId: 227566113
Existing implementation was created before ML/CFG unification refactoring and
did not concern itself with further lowering to separate concerns. As a
result, it emitted `affine_apply` instructions to implement `for` loop bounds
and `if` conditions and required a follow-up function pass to lower those
`affine_apply` to arithmetic primitives. In the unified function world,
LowerForAndIf is mostly a lowering pass with low complexity. As we move
towards a dialect for affine operations (including `for` and `if`), it makes
sense to lower `for` and `if` conditions directly to arithmetic primitives
instead of relying on `affine_apply`.
Expose `expandAffineExpr` function in LoweringUtils. Use this function
together with `expandAffineMaps` to emit primitives that implement loop and
branch conditions directly.
Also remove tests that become unnecessary after transforming LowerForAndIf into
a function pass.
PiperOrigin-RevId: 227563608
In LoweringUtils, extract out `expandAffineMap`. This function takes an affine
map and a list of values the map should be applied to and emits a sequence of
arithmetic instructions that implement the affine map. It is independent of
the AffineApplyOp and can be used in places where we need to insert an
evaluation of an affine map without relying on a (temporary) `affine_apply`
instruction. This prepares for a merge between LowerAffineApply and
LowerForAndIf passes.
Move the `expandAffineApply` function to the LowerAffineApply pass since it is
the only place that must be aware of the `affine_apply` instructions.
PiperOrigin-RevId: 227563439
The entire compiler now looks at structural properties of the function (e.g.
does it have one block, does it contain an if/for stmt, etc) so the only thing
holding up this difference is round tripping through the parser/printer syntax.
Removing this shrinks the compile by ~140LOC.
This is step 31/n towards merging instructions and statements. The last step
is updating the docs, which I will do as a separate patch in order to split it
from this mostly mechanical patch.
PiperOrigin-RevId: 227540453
This CL introduces a simple set of Embedded Domain-Specific Components (EDSCs)
in MLIR components:
1. a `Type` system of shell classes that closely matches the MLIR type system. These
types are subdivided into `Bindable` leaf expressions and non-bindable `Expr`
expressions;
2. an `MLIREmitter` class whose purpose is to:
a. maintain a map of `Bindable` leaf expressions to concrete SSAValue*;
b. provide helper functionality to specify bindings of `Bindable` classes to
SSAValue* while verifying comformable types;
c. traverse the `Expr` and emit the MLIR.
This is used on a concrete example to implement MemRef load/store with clipping in the
LowerVectorTransfer pass. More specifically, the following pseudo-C++ code:
```c++
MLFuncBuilder *b = ...;
Location location = ...;
Bindable zero, one, expr, size;
// EDSL expression
auto access = select(expr < zero, zero, select(expr < size, expr, size - one));
auto ssaValue = MLIREmitter(b)
.bind(zero, ...)
.bind(one, ...)
.bind(expr, ...)
.bind(size, ...)
.emit(location, access);
```
is used to emit all the MLIR for a clipped MemRef access.
This simple EDSL can easily be extended to more powerful patterns and should
serve as the counterpart to pattern matchers (and could potentially be unified
once we get enough experience).
In the future, most of this code should be TableGen'd but for now it has
concrete valuable uses: make MLIR programmable in a declarative fashion.
This CL also adds Stmt, proper supporting free functions and rewrites
VectorTransferLowering fully using EDSCs.
The code for creating the EDSCs emitting a VectorTransferReadOp as loops
with clipped loads is:
```c++
Stmt block = Block({
tmpAlloc = alloc(tmpMemRefType),
vectorView = vector_type_cast(tmpAlloc, vectorMemRefType),
ForNest(ivs, lbs, ubs, steps, {
scalarValue = load(scalarMemRef, accessInfo.clippedScalarAccessExprs),
store(scalarValue, tmpAlloc, accessInfo.tmpAccessExprs),
}),
vectorValue = load(vectorView, zero),
tmpDealloc = dealloc(tmpAlloc.getLHS())});
emitter.emitStmt(block);
```
where `accessInfo.clippedScalarAccessExprs)` is created with:
```c++
select(i + ii < zero, zero, select(i + ii < N, i + ii, N - one));
```
The generated MLIR resembles:
```mlir
%1 = dim %0, 0 : memref<?x?x?x?xf32>
%2 = dim %0, 1 : memref<?x?x?x?xf32>
%3 = dim %0, 2 : memref<?x?x?x?xf32>
%4 = dim %0, 3 : memref<?x?x?x?xf32>
%5 = alloc() : memref<5x4x3xf32>
%6 = vector_type_cast %5 : memref<5x4x3xf32>, memref<1xvector<5x4x3xf32>>
for %i4 = 0 to 3 {
for %i5 = 0 to 4 {
for %i6 = 0 to 5 {
%7 = affine_apply #map0(%i0, %i4)
%8 = cmpi "slt", %7, %c0 : index
%9 = affine_apply #map0(%i0, %i4)
%10 = cmpi "slt", %9, %1 : index
%11 = affine_apply #map0(%i0, %i4)
%12 = affine_apply #map1(%1, %c1)
%13 = select %10, %11, %12 : index
%14 = select %8, %c0, %13 : index
%15 = affine_apply #map0(%i3, %i6)
%16 = cmpi "slt", %15, %c0 : index
%17 = affine_apply #map0(%i3, %i6)
%18 = cmpi "slt", %17, %4 : index
%19 = affine_apply #map0(%i3, %i6)
%20 = affine_apply #map1(%4, %c1)
%21 = select %18, %19, %20 : index
%22 = select %16, %c0, %21 : index
%23 = load %0[%14, %i1, %i2, %22] : memref<?x?x?x?xf32>
store %23, %5[%i6, %i5, %i4] : memref<5x4x3xf32>
}
}
}
%24 = load %6[%c0] : memref<1xvector<5x4x3xf32>>
dealloc %5 : memref<5x4x3xf32>
```
In particular notice that only 3 out of the 4-d accesses are clipped: this
corresponds indeed to the number of dimensions in the super-vector.
This CL also addresses the cleanups resulting from the review of the prevous
CL and performs some refactoring to simplify the abstraction.
PiperOrigin-RevId: 227367414
on this to merge together the classes, but there may be other simplification
possible. I'll leave that to riverriddle@ as future work.
This is step 29/n towards merging instructions and statements.
PiperOrigin-RevId: 227328680
simplifying them in minor ways. The only significant cleanup here
is the constant folding pass. All the other changes are simple and easy,
but this is still enough to shrink the compiler by 45LOC.
The one pass left to merge is the CSE pass, which will be move involved, so I'm
splitting it out to its own patch (which I'll tackle right after this).
This is step 28/n towards merging instructions and statements.
PiperOrigin-RevId: 227328115
Remove an unnecessary restriction in forward substitution. Slightly
simplify LLVM IR lowering, which previously would crash if given an ML
function, it should now produce a clean error if given a function with an
if/for instruction in it, just like it does any other unsupported op.
This is step 27/n towards merging instructions and statements.
PiperOrigin-RevId: 227324542
representation, shrinking by 70LOC. The PatternRewriter class can probably
also be simplified as well, but one step at a time.
This is step 26/n towards merging instructions and statements. NFC.
PiperOrigin-RevId: 227324218
- introduce PostDominanceInfo in the right/complete way and use that for post
dominance check in store-load forwarding
- replace all uses of Analysis/Utils::dominates/properlyDominates with
DominanceInfo::dominates/properlyDominates
- drop all redundant copies of dominance methods in Analysis/Utils/
- in pipeline-data-transfer, replace dominates call with a much less expensive
check; similarly, substitute dominates() in checkMemRefAccessDependence with
a simpler check suitable for that context
- fix a bug in properlyDominates
- improve doc for 'for' instruction 'body'
PiperOrigin-RevId: 227320507
function pass, and eliminating the need to copy over code and do
interprocedural updates. While here, also improve it to make fewer empty
blocks, and rename it to "LowerIfAndFor" since that is what it does. This is
a net reduction of ~170 lines of code.
As drive-bys, change the splitBlock method to *not* insert an unconditional
branch, since that behavior is annoying for all clients. Also improve the
AsmPrinter to not crash when a block is referenced that isn't linked into a
function.
PiperOrigin-RevId: 227308856
- the load/store forwarding relies on memref dependence routines as well as
SSA/dominance to identify the memref store instance uniquely supplying a value
to a memref load, and replaces the result of that load with the value being
stored. The memref is also deleted when possible if only stores remain.
- add methods for post dominance for MLFunction blocks.
- remove duplicated getLoopDepth/getNestingDepth - move getNestingDepth,
getMemRefAccess, getNumCommonSurroundingLoops into Analysis/Utils (were
earlier static)
- add a helper method in FlatAffineConstraints - isRangeOneToOne.
PiperOrigin-RevId: 227252907
Function::walk functionality into f->walkInsts/Ops which allows visiting all
instructions, not just ops. Eliminate Function::getBody() and
Function::getReturn() helpers which crash in CFG functions, and were only kept
around as a bridge.
This is step 25/n towards merging instructions and statements.
PiperOrigin-RevId: 227243966
consistent and moving the using declarations over. Hopefully this is the last
truly massive patch in this refactoring.
This is step 21/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227178245
The last major renaming is Statement -> Instruction, which is why Statement and
Stmt still appears in various places.
This is step 19/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227163082
StmtResult -> InstResult, StmtOperand -> InstOperand, and remove the old names.
This is step 17/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227121537
OperationInst derives from it. This allows eliminating some forwarding
functions, other complex code handling multiple paths, and the 'isStatement'
bit tracked by Operation.
This is the last patch I think I can make before the big mechanical change
merging Operation into OperationInst, coming next.
This is step 15/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227077411
StmtSuccessorIterator/StmtSuccessorIterator, and rename and move the
CFGFunctionViewGraph pass to ViewFunctionGraph.
This is step 13/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227069438
FuncBuilder class. Also rename SSAValue.cpp to Value.cpp
This is step 12/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227067644
is the new base of the SSA value hierarchy. This CL also standardizes all the
nomenclature and comments to use 'Value' where appropriate. This also eliminates a large number of cast<MLValue>(x)'s, which is very soothing.
This is step 11/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227064624
This *only* changes the internal data structures, it does not affect the user visible syntax or structure of MLIR code. Function gets new "isCFG()" sorts of predicates as a transitional measure.
This patch is gross in a number of ways, largely in an effort to reduce the amount of mechanical churn in one go. It introduces a bunch of using decls to keep the old names alive for now, and a bunch of stuff needs to be renamed.
This is step 10/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 227044402
making it more similar to the CFG side of things. It is true that in a deeply
nested case that this is not a guaranteed O(1) time operation, and that 'get'
could lead compiler hackers to think this is cheap, but we need to merge these
and we can look into solutions for this in the future if it becomes a problem
in practice.
This is step 9/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 226983931
graph specializations for doing CFG traversals of ML Functions, making the two
sorts of functions have the same capabilities.
This is step 8/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 226968502
Supervectorization uses null pointers to SSA values as a means of communicating
the failure to vectorize. In operation vectorization, all operations producing
the values of operation arguments must be vectorized for the given operation to
be vectorized. The existing check verified if any of the value "def"
statements was vectorized instead, sometimes leading to assertions inside `isa`
called on a null pointer. Fix this to check that all "def" statements were
vectorized.
PiperOrigin-RevId: 226941552
from it. This is necessary progress to squaring away the parent relationship
that a StmtBlock has with its enclosing if/for/fn, and makes room for functions
to have more than one block in the future. This also removes IfClause and ForStmtBody.
This is step 5/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 226936541
for SSA values in terminators, but easily worked around. At the same time,
move the StmtOperand list in a OperationStmt to the end of its trailing
objects list so we can *reduce* the number of operands, without affecting
offsets to the other stuff in the allocation.
This is important because we want OperationStmts to be consequtive, including
their operands - we don't want to use an std::vector of operands like
Instructions have.
This is patch 4/n towards merging instructions and statements, NFC.
PiperOrigin-RevId: 226865727
clients to use OperationState instead. This makes MLFuncBuilder more similiar
to CFGFuncBuilder. This whole area will get tidied up more when cfg and ml
worlds get unified. This patch is just gardening, NFC.
PiperOrigin-RevId: 226701959
StmtBlock. This is more consistent with IfStmt and also conceptually makes
more sense - a forstmt "isn't" its body, it contains its body.
This is step 1/N towards merging BasicBlock and StmtBlock. This is required
because in the new regime StmtBlock will have a use list (just like BasicBlock
does) of operands, and ForStmt already has a use list for its induction
variable.
This is a mechanical patch, NFC.
PiperOrigin-RevId: 226684158
reuse existing ones.
- drop IterationDomainContext, redundant since FlatAffineConstraints has
MLValue information associated with its dimensions.
- refactor to use existing support
- leads to a reduction in LOC
- as a result of these changes, non-constant loop bounds get naturally
supported for dep analysis.
- update test cases to include a couple with non-constant loop bounds
- rename addBoundsFromForStmt -> addForStmtDomain
- complete TODO for getLoopIVs (handle 'if' statements)
PiperOrigin-RevId: 226082008
This introduces a generic lowering pass for ML functions. The pass is
parameterized by template arguments defining individual pattern rewriters.
Concrete lowering passes define individual pattern rewriters and inherit from
the generic class that takes care of allocating rewriters, traversing ML
functions and performing the actual rewrite.
While this is similar to the greedy pattern rewriter available in
Transform/Utils, it requires adjustments due to the ML/CFG duality. In
particular, ML function rewriters must be able to create statements, not only
operations, and need access to an MLFuncBuilder. When we move to using the
unified function type, the ML-specific rewriting will become unnecessary.
Use LowerVectorTransfers as a testbed for the generic pass.
PiperOrigin-RevId: 225887424
This operation is produced and used by the super-vectorization passes and has
been emitted as an abstract unregistered operation until now. For end-to-end
testing purposes, it has to be eventually lowered to LLVM IR. Matching
abstract operation by name goes into the opposite direction of the generic
lowering approach that is expected to be used for LLVM IR lowering in the
future. Register vector_type_cast operation as a part of the SuperVector
dialect.
Arguably, this operation is a special case of the `view` operation from the
Standard dialect. The semantics of `view` is not fully specified at this point
so it is safer to rely on a custom operation. Additionally, using a custom
operation may help to achieve clear dialect separation.
PiperOrigin-RevId: 225887305
provide unroll factors, and a cmd line argument to specify number of
innermost loop unroll repetitions.
- add function callback parameter for outside targets to provide unroll factors
- add a cmd line parameter to repeatedly apply innermost loop unroll a certain
number of times (to avoid using -loop-unroll -loop-unroll ...; instead
-unroll-num-reps=2).
- implement the callback for a target
- update test cases / usage
PiperOrigin-RevId: 225843191
*) Adds simple greedy fusion algorithm to drive experimentation. This algorithm greedily fuses loop nests with single-writer/single-reader memref dependences to improve locality.
*) Adds support for fusing slices of a loop nest computation: fusing one loop nest into another by adjusting the source loop nest's iteration bounds (after it is fused into the destination loop nest). This is accomplished by solving for the source loop nest's IVs in terms of the destination loop nests IVs and symbols using the dependece polyhedron, then creating AffineMaps of these functions for the loop bounds of the fused source loop.
*) Adds utility function 'insertMemRefComputationSlice' which computes and inserts computation slice from loop nest surrounding a source memref access into the loop nest surrounding the destingation memref access.
*) Adds FlatAffineConstraints::toAffineMap function which returns and AffineMap which represents an equality contraint where one dimension identifier is represented as a function of all others in the equality constraint.
*) Adds multiple fusion unit tests.
PiperOrigin-RevId: 225842944
- use addBoundsForForStmt
- getLoopIVs can return a vector of ForStmt * instead of const ForStmt *; the
returned things aren't owned / part of the stmt on which it's being called.
- other minor API cleanup
PiperOrigin-RevId: 225774301
From the beginning, vector_transfer_read and vector_transfer_write opreations
were intended as a mid-level vectorization abstraction. In particular, they
are lowered to the StandardOps dialect before further processing. As such, it
does not make sense to keep them at the same level as StandardOps. Introduce
the new SuperVectorOps dialect and move vector_transfer_* operations there.
This will be used as a testbed for the generic lowering/legalization pass.
PiperOrigin-RevId: 225554492
Originally, loop steps were implemented using `addi` and `constant` operations
because `affine_apply` was not handled in the first implementation. The
support for `affine_apply` has been added, use it to implement the update of
the loop induction variable. This is more consistent with the lower and upper
bounds of the loop that are also implemented as `affine_apply`, removes the
dependence of the converted function on the StandardOps dialect and makes it
clear from the CFG function that all operations on the loop induction variable
are purely affine.
PiperOrigin-RevId: 225165337
- loop step wasn't handled and there wasn't a TODO or an assertion; fix this.
- rename 'delay' to shift for consistency/readability.
- other readability changes.
- remove duplicate attribute print for DmaStartOp; fix misplaced attribute
print for DmaWaitOp
- add build method for AddFOp (unrelated to this CL, but add it anyway)
PiperOrigin-RevId: 224892958
- adding a conservative check for now (TODO: use the dependence analysis pass
once the latter is extended to deal with DMA ops). resolve an existing bug on
a test case.
- update test cases
PiperOrigin-RevId: 224869526
- add method normalizeConstraintsByGCD
- call normalizeConstraintsByGCD() and GCDTightenInequalities() at the end of
projectOut.
- remove call to GCDTightenInequalities() from getMemRefRegion
- change isEmpty() to check isEmptyByGCDTest() / hasInvalidConstraint() each
time an identifier is eliminated (to detect emptiness early).
- make FourierMotzkinEliminate, gaussianEliminateId(s),
GCDTightenInequalities() private
- improve / update stale comments
PiperOrigin-RevId: 224866741
- fix replaceAllMemRefUsesWith call to replace only inside loop body.
- handle the case where DMA buffers are dynamic; extend doubleBuffer() method
to handle dynamically shaped DMA buffers (pass the right operands to AllocOp)
- place alloc's for DMA buffers at the depth at which pipelining is being done
(instead of at top-level)
- add more test cases
PiperOrigin-RevId: 224852231
This was missing from the original commit. The implementation of
createLowerAffineApply was defined in the default namespace but declared in the
`mlir` namespace, which could lead to linking errors when it was used. Put the
definition in `mlir` namespace.
PiperOrigin-RevId: 224830894
are a max/min of several expressions.
- Extend loop tiling to handle non-constant loop bounds and bounds that
are a max/min of several expressions, i.e., bounds using multi-result affine
maps
- also fix b/120630124 as a result (the IR was in an invalid state when tiled
loop generation failed; SSA uses were created that weren't plugged into the IR).
PiperOrigin-RevId: 224604460
- generate DMAs correctly now using strided DMAs where needed
- add support for multi-level/nested strides; op still supports one level of
stride for now.
Other things
- add test case for symbolic lower/upper bound; cases where the DMA buffer
size can't be bounded by a known constant
- add test case for dynamic shapes where the DMA buffers are however bounded by
constants
- refactor some of the '-dma-generate' code
PiperOrigin-RevId: 224584529
This CL adds a pass that lowers VectorTransferReadOp and VectorTransferWriteOp
to a simple loop nest via local buffer allocations.
This is an MLIR->MLIR lowering based on builders.
A few TODOs are left to address in particular:
1. invert the permutation map so the accesses to the remote memref are coalesced;
2. pad the alloc for bank conflicts in local memory (e.g. GPUs shared_memory);
3. support broadcast / avoid copies when permutation_map is not of full column rank
4. add a proper "element_cast" op
One notable limitation is this does not plan on supporting boundary conditions.
It should be significantly easier to use pre-baked MLIR functions to handle such paddings.
This is left for future consideration.
Therefore the current CL only works properly for full-tile cases atm.
This CL also adds 2 simple tests:
```mlir
for %i0 = 0 to %M step 3 {
for %i1 = 0 to %N step 4 {
for %i2 = 0 to %O {
for %i3 = 0 to %P step 5 {
vector_transfer_write %f1, %A, %i0, %i1, %i2, %i3 {permutation_map: (d0, d1, d2, d3) -> (d3, d1, d0)} : vector<5x4x3xf32>, memref<?x?x?x?xf32, 0>, index, index, index, index
```
lowers into:
```mlir
for %i0 = 0 to %arg0 step 3 {
for %i1 = 0 to %arg1 step 4 {
for %i2 = 0 to %arg2 {
for %i3 = 0 to %arg3 step 5 {
%1 = alloc() : memref<5x4x3xf32>
%2 = "element_type_cast"(%1) : (memref<5x4x3xf32>) -> memref<1xvector<5x4x3xf32>>
store %cst, %2[%c0] : memref<1xvector<5x4x3xf32>>
for %i4 = 0 to 5 {
%3 = affine_apply (d0, d1) -> (d0 + d1) (%i3, %i4)
for %i5 = 0 to 4 {
%4 = affine_apply (d0, d1) -> (d0 + d1) (%i1, %i5)
for %i6 = 0 to 3 {
%5 = affine_apply (d0, d1) -> (d0 + d1) (%i0, %i6)
%6 = load %1[%i4, %i5, %i6] : memref<5x4x3xf32>
store %6, %0[%5, %4, %i2, %3] : memref<?x?x?x?xf32>
dealloc %1 : memref<5x4x3xf32>
```
and
```mlir
for %i0 = 0 to %M step 3 {
for %i1 = 0 to %N {
for %i2 = 0 to %O {
for %i3 = 0 to %P step 5 {
%f = vector_transfer_read %A, %i0, %i1, %i2, %i3 {permutation_map: (d0, d1, d2, d3) -> (d3, 0, d0)} : (memref<?x?x?x?xf32, 0>, index, index, index, index) -> vector<5x4x3xf32>
```
lowers into:
```mlir
for %i0 = 0 to %arg0 step 3 {
for %i1 = 0 to %arg1 {
for %i2 = 0 to %arg2 {
for %i3 = 0 to %arg3 step 5 {
%1 = alloc() : memref<5x4x3xf32>
%2 = "element_type_cast"(%1) : (memref<5x4x3xf32>) -> memref<1xvector<5x4x3xf32>>
for %i4 = 0 to 5 {
%3 = affine_apply (d0, d1) -> (d0 + d1) (%i3, %i4)
for %i5 = 0 to 4 {
for %i6 = 0 to 3 {
%4 = affine_apply (d0, d1) -> (d0 + d1) (%i0, %i6)
%5 = load %0[%4, %i1, %i2, %3] : memref<?x?x?x?xf32>
store %5, %1[%i4, %i5, %i6] : memref<5x4x3xf32>
%6 = load %2[%c0] : memref<1xvector<5x4x3xf32>>
dealloc %1 : memref<5x4x3xf32>
```
PiperOrigin-RevId: 224552717
This simplifies call-sites returning true after emitting an error. After the
conversion, dropped braces around single statement blocks as that seems more
common.
Also, switched to emitError method instead of emitting Error kind using the
emitDiagnostic method.
TESTED with existing unit tests
PiperOrigin-RevId: 224527868
This CLs adds proper error emission, removes NYI assertions and documents
assumptions that are required in the relevant functions.
PiperOrigin-RevId: 224377207
This CL adds the following free functions:
```
/// Returns the AffineExpr e o m.
AffineExpr compose(AffineExpr e, AffineMap m);
/// Returns the AffineExpr f o g.
AffineMap compose(AffineMap f, AffineMap g);
```
This addresses the issue that AffineMap composition is only available at a
distance via AffineValueMap and is thus unusable on Attributes.
This CL thus implements AffineMap composition in a more modular and composable
way.
This CL does not claim that it can be a good replacement for the
implementation in AffineValueMap, in particular it does not support bounded
maps atm.
Standalone tests are added that replicate some of the logic of the AffineMap
composition pass.
Lastly, affine map composition is used properly inside MaterializeVectors and
a standalone test is added that requires permutation_map composition with a
projection map.
PiperOrigin-RevId: 224376870
This CL hooks up and uses permutation_map in vector_transfer ops.
In particular, when going into the nuts and bolts of the implementation, it
became clear that cases arose that required supporting broadcast semantics.
Broadcast semantics are thus added to the general permutation_map.
The verify methods and tests are updated accordingly.
Examples of interest include.
Example 1:
The following MLIR snippet:
```mlir
for %i3 = 0 to %M {
for %i4 = 0 to %N {
for %i5 = 0 to %P {
%a5 = load %A[%i4, %i5, %i3] : memref<?x?x?xf32>
}}}
```
may vectorize with {permutation_map: (d0, d1, d2) -> (d2, d1)} into:
```mlir
for %i3 = 0 to %0 step 32 {
for %i4 = 0 to %1 {
for %i5 = 0 to %2 step 256 {
%4 = vector_transfer_read %arg0, %i4, %i5, %i3
{permutation_map: (d0, d1, d2) -> (d2, d1)} :
(memref<?x?x?xf32>, index, index) -> vector<32x256xf32>
}}}
````
Meaning that vector_transfer_read will be responsible for reading the 2-D slice:
`%arg0[%i4, %i5:%15+256, %i3:%i3+32]` into vector<32x256xf32>. This will
require a transposition when vector_transfer_read is further lowered.
Example 2:
The following MLIR snippet:
```mlir
%cst0 = constant 0 : index
for %i0 = 0 to %M {
%a0 = load %A[%cst0, %cst0] : memref<?x?xf32>
}
```
may vectorize with {permutation_map: (d0) -> (0)} into:
```mlir
for %i0 = 0 to %0 step 128 {
%3 = vector_transfer_read %arg0, %c0_0, %c0_0
{permutation_map: (d0, d1) -> (0)} :
(memref<?x?xf32>, index, index) -> vector<128xf32>
}
````
Meaning that vector_transfer_read will be responsible of reading the 0-D slice
`%arg0[%c0, %c0]` into vector<128xf32>. This will require a 1-D vector
broadcast when vector_transfer_read is further lowered.
Additionally, some minor cleanups and refactorings are performed.
One notable thing missing here is the composition with a projection map during
materialization. This is because I could not find an AffineMap composition
that operates on AffineMap directly: everything related to composition seems
to require going through SSAValue and only operates on AffinMap at a distance
via AffineValueMap. I have raised this concern a bunch of times already, the
followup CL will actually do something about it.
In the meantime, the projection is hacked at a minimum to pass verification
and materialiation tests are temporarily incorrect.
PiperOrigin-RevId: 224376828
The recently introduced `select` operation enables ConvertToCFG to support
min(max) in loop bounds. Individual min(max) is implemented as
`cmpi "lt"`(`cmpi "gt"`) followed by a `select` between the compared values.
Multiple results of an `affine_apply` operation extracted from the loop bounds
are reduced using min(max) in a sequential manner. While this may decrease the
potential for instruction-level parallelism, it is easier to recognize for the
following passes, in particular for the vectorizer.
PiperOrigin-RevId: 224376233
The implementation of OpPointer<OpType> provides an implicit conversion to
Operation *, but not to the underlying OpType *. This has led to
awkward-looking code when an OpPointer needs to be passed to a function
accepting an OpType *. For example,
if (auto someOp = genericOp.dyn_cast<OpType>())
someFunction(&*someOp);
where "&*" makes it harder to read. Arguably, one does not want to spell out
OpPointer<OpType> in the line with dyn_cast. More generally, OpPointer is now
being used as an owning pointer to OpType rather than to operation.
Replace the implicit conversion to Operation* with the conversion to OpType*
taking into account const-ness of the type. An Operation* can be obtained from
an OpType with a simple call. Since an instance of OpPointer owns the OpType
value, the pointer to it is never null. However, the OpType value may not be
associated with any Operation*. In this case, return nullptr when conversion
is attempted to maintain consistency with the existing null checks.
PiperOrigin-RevId: 224368103
cl/224246657); eliminate repeated evaluation of exprs in loop upper bounds.
- while on this, sweep through and fix potential repeated evaluation of
expressions in loop upper bounds
PiperOrigin-RevId: 224268918
update/improve/clean up API.
- update FlatAffineConstraints::getConstBoundDifference; return constant
differences between symbolic affine expressions, look at equalities as well.
- fix buffer size computation when generating DMAs symbolic in outer loops,
correctly handle symbols at various places (affine access maps, loop bounds,
loop IVs outer to the depth at which DMA generation is being done)
- bug fixes / complete some TODOs for getMemRefRegion
- refactor common code b/w memref dependence check and getMemRefRegion
- FlatAffineConstraints API update; added methods employ trivial checks /
detection - sufficient to handle hyper-rectangular cases in a precise way
while being fast / low complexity. Hyper-rectangular cases fall out as
trivial cases for these methods while other cases still do not cause failure
(either return conservative or return failure that is handled by the caller).
PiperOrigin-RevId: 224229879
The condition of the "if" statement is an integer set, defined as a conjunction
of affine constraints. An affine constraints consists of an affine expression
and a flag indicating whether the expression is strictly equal to zero or is
also allowed to be greater than zero. Affine maps, accepted by `affine_apply`
are also formed from affine expressions. Leverage this fact to implement the
checking of "if" conditions. Each affine expression from the integer set is
converted into an affine map. This map is applied to the arguments of the "if"
statement. The result of the application is compared with zero given the
equality flag to obtain the final boolean value. The conjunction of conditions
is tested sequentially with short-circuit branching to the "else" branch if any
of the condition evaluates to false.
Create an SESE region for the if statement (including its "then" and optional
"else" statement blocks) and append it to the end of the current region. The
conditional region consists of a sequence of condition-checking blocks that
implement the short-circuit scheme, followed by a "then" SESE region and an
"else" SESE region, and the continuation block that post-dominates all blocks
of the "if" statement. The flow of blocks that correspond to the "then" and
"else" clauses are constructed recursively, enabling easy nesting of "if"
statements and if-then-else-if chains.
Note that MLIR semantics does not require nor prohibit short-circuit
evaluation. Since affine expressions do not have side effects, there is no
observable difference in the program behavior. We may trade off extra
operations for operation-level parallelism opportunity by first performing all
`affine_apply` and comparison operations independently, and then performing a
tree pattern reduction of the resulting boolean values with the `muli i1`
operations (in absence of the dedicated bit operations). The pros and cons are
not clear, and since MLIR does not include parallel semantics, we prefer to
minimize the number of sequentially executed operations.
PiperOrigin-RevId: 223970248
This CL implements and uses VectorTransferOps in lieu of the former custom
call op. Tests are updated accordingly.
VectorTransferOps come in 2 flavors: VectorTransferReadOp and
VectorTransferWriteOp.
VectorTransferOps can be thought of as a backend-independent
pseudo op/library call that needs to be legalized to MLIR (whiteboxed) before
it can be lowered to backend-dependent IR.
Note that the current implementation does not yet support a real permutation
map. Proper support will come in a followup CL.
VectorTransferReadOp
====================
VectorTransferReadOp performs a blocking read from a scalar memref
location into a super-vector of the same elemental type. This operation is
called 'read' by opposition to 'load' because the super-vector granularity
is generally not representable with a single hardware register. As a
consequence, memory transfers will generally be required when lowering
VectorTransferReadOp. A VectorTransferReadOp is thus a mid-level abstraction
that supports super-vectorization with non-effecting padding for full-tile
only code.
A vector transfer read has semantics similar to a vector load, with additional
support for:
1. an optional value of the elemental type of the MemRef. This value
supports non-effecting padding and is inserted in places where the
vector read exceeds the MemRef bounds. If the value is not specified,
the access is statically guaranteed to be within bounds;
2. an attribute of type AffineMap to specify a slice of the original
MemRef access and its transposition into the super-vector shape. The
permutation_map is an unbounded AffineMap that must represent a
permutation from the MemRef dim space projected onto the vector dim
space.
Example:
```mlir
%A = alloc(%size1, %size2, %size3, %size4) : memref<?x?x?x?xf32>
...
%val = `ssa-value` : f32
// let %i, %j, %k, %l be ssa-values of type index
%v0 = vector_transfer_read %src, %i, %j, %k, %l
{permutation_map: (d0, d1, d2, d3) -> (d3, d1, d2)} :
(memref<?x?x?x?xf32>, index, index, index, index) ->
vector<16x32x64xf32>
%v1 = vector_transfer_read %src, %i, %j, %k, %l, %val
{permutation_map: (d0, d1, d2, d3) -> (d3, d1, d2)} :
(memref<?x?x?x?xf32>, index, index, index, index, f32) ->
vector<16x32x64xf32>
```
VectorTransferWriteOp
=====================
VectorTransferWriteOp performs a blocking write from a super-vector to
a scalar memref of the same elemental type. This operation is
called 'write' by opposition to 'store' because the super-vector
granularity is generally not representable with a single hardware register. As
a consequence, memory transfers will generally be required when lowering
VectorTransferWriteOp. A VectorTransferWriteOp is thus a mid-level
abstraction that supports super-vectorization with non-effecting padding
for full-tile only code.
A vector transfer write has semantics similar to a vector store, with
additional support for handling out-of-bounds situations.
Example:
```mlir
%A = alloc(%size1, %size2, %size3, %size4) : memref<?x?x?x?xf32>.
%val = `ssa-value` : vector<16x32x64xf32>
// let %i, %j, %k, %l be ssa-values of type index
vector_transfer_write %val, %src, %i, %j, %k, %l
{permutation_map: (d0, d1, d2, d3) -> (d3, d1, d2)} :
(vector<16x32x64xf32>, memref<?x?x?x?xf32>, index, index, index, index)
```
PiperOrigin-RevId: 223873234
The check for whether the memref was used in a non-derefencing context had to
be done inside, i.e., only for the op stmt's that the replacement was specified
to be performed on (by the domStmtFilter arg if provided). As such, it is
completely fine for example for a function to return a memref while the replacement
is being performed only a specific loop's body (as in the case of DMA
generation).
PiperOrigin-RevId: 223827753
The algorithm collects defining operations within a scoped hash table. The scopes within the hash table correspond to nodes within the dominance tree for a function. This cl only adds support for simple operations, i.e non side-effecting. Such operations, e.g. load/store/call, will be handled in later patches.
PiperOrigin-RevId: 223811328
class. This change is NFC, but allows for new kinds of patterns, specifically
LegalizationPatterns which will be allowed to change the types of things they
rewrite.
PiperOrigin-RevId: 223243783
Several things were suggested in post-submission reviews. In particular, use
pointers in function interfaces instead of references (still use references
internally). Clarify the behavior of the pass in presence of MLFunctions.
PiperOrigin-RevId: 222556851
This CL adds tooling for computing slices as an independent CL.
The first consumer of this analysis will be super-vector materialization in a
followup CL.
In particular, this adds:
1. a getForwardStaticSlice function with documentation, example and a
standalone unit test;
2. a getBackwardStaticSlice function with documentation, example and a
standalone unit test;
3. a getStaticSlice function with documentation, example and a standalone unit
test;
4. a topologicalSort function that is exercised through the getStaticSlice
unit test.
The getXXXStaticSlice functions take an additional root (resp. terminators)
parameter which acts as a boundary that the transitive propagation algorithm
is not allowed to cross.
PiperOrigin-RevId: 222446208
cases.
- fix bug in calculating index expressions for DMA buffers in certain cases
(affected tiled loop nests); add more test cases for better coverage.
- introduce an additional optional argument to replaceAllMemRefUsesWith;
additional operands to the index remap AffineMap can now be supplied by the
client.
- FlatAffineConstraints::addBoundsForStmt - fix off by one upper bound,
::composeMap - fix position bug.
- Some clean up and more comments
PiperOrigin-RevId: 222434628
This function pass replaces affine_apply operations in CFG functions with
sequences of primitive arithmetic instructions that form the affine map.
The actual replacement functionality is located in LoweringUtils as a
standalone function operating on an individual affine_apply operation and
inserting the result at the location of the original operation. It is expected
to be useful for other, target-specific lowering passes that may start at
MLFunction level that Deaffinator does not support.
PiperOrigin-RevId: 222406692
This CL refactors a few things in Vectorize.cpp:
1. a clear distinction is made between:
a. the LoadOp are the roots of vectorization and must be vectorized
eagerly and propagate their value; and
b. the StoreOp which are the terminals of vectorization and must be
vectorized late (i.e. they do not produce values that need to be
propagated).
2. the StoreOp must be vectorized late because in general it can store a value
that is not reachable from the subset of loads defined in the
current pattern. One trivial such case is storing a constant defined at the
top-level of the MLFunction and that needs to be turned into a splat.
3. a description of the algorithm is given;
4. the implementation matches the algorithm;
5. the last example is made parametric, in practice it will fully rely on the
implementation of vector_transfer_read/write which will handle boundary
conditions and padding. This will happen by lowering to a lower-level
abstraction either:
a. directly in MLIR (whether DMA or just loops or any async tasks in the
future) (whiteboxing);
b. in LLO/LLVM-IR/whatever blackbox library call/ search + swizzle inventor
one may want to use;
c. a partial mix of a. and b. (grey-boxing)
5. minor cleanups are applied;
6. mistakenly disabled unit tests are re-enabled (oopsie).
With this CL, this MLIR snippet:
```
mlfunc @vector_add_2d(%M : index, %N : index) -> memref<?x?xf32> {
%A = alloc (%M, %N) : memref<?x?xf32>
%B = alloc (%M, %N) : memref<?x?xf32>
%C = alloc (%M, %N) : memref<?x?xf32>
%f1 = constant 1.0 : f32
%f2 = constant 2.0 : f32
for %i0 = 0 to %M {
for %i1 = 0 to %N {
// non-scoped %f1
store %f1, %A[%i0, %i1] : memref<?x?xf32>
}
}
for %i4 = 0 to %M {
for %i5 = 0 to %N {
%a5 = load %A[%i4, %i5] : memref<?x?xf32>
%b5 = load %B[%i4, %i5] : memref<?x?xf32>
%s5 = addf %a5, %b5 : f32
// non-scoped %f1
%s6 = addf %s5, %f1 : f32
store %s6, %C[%i4, %i5] : memref<?x?xf32>
}
}
return %C : memref<?x?xf32>
}
```
vectorized with these arguments:
```
-vectorize -virtual-vector-size 256 --test-fastest-varying=0
```
vectorization produces this standard innermost-loop vectorized code:
```
mlfunc @vector_add_2d(%arg0 : index, %arg1 : index) -> memref<?x?xf32> {
%0 = alloc(%arg0, %arg1) : memref<?x?xf32>
%1 = alloc(%arg0, %arg1) : memref<?x?xf32>
%2 = alloc(%arg0, %arg1) : memref<?x?xf32>
%cst = constant 1.000000e+00 : f32
%cst_0 = constant 2.000000e+00 : f32
for %i0 = 0 to %arg0 {
for %i1 = 0 to %arg1 step 256 {
%cst_1 = constant splat<vector<256xf32>, 1.000000e+00> : vector<256xf32>
"vector_transfer_write"(%cst_1, %0, %i0, %i1) : (vector<256xf32>, memref<?x?xf32>, index, index) -> ()
}
}
for %i2 = 0 to %arg0 {
for %i3 = 0 to %arg1 step 256 {
%3 = "vector_transfer_read"(%0, %i2, %i3) : (memref<?x?xf32>, index, index) -> vector<256xf32>
%4 = "vector_transfer_read"(%1, %i2, %i3) : (memref<?x?xf32>, index, index) -> vector<256xf32>
%5 = addf %3, %4 : vector<256xf32>
%cst_2 = constant splat<vector<256xf32>, 1.000000e+00> : vector<256xf32>
%6 = addf %5, %cst_2 : vector<256xf32>
"vector_transfer_write"(%6, %2, %i2, %i3) : (vector<256xf32>, memref<?x?xf32>, index, index) -> ()
}
}
return %2 : memref<?x?xf32>
}
```
Of course, much more intricate n-D imperfectly-nested patterns can be emitted too in a fully declarative fashion, but this is enough for now.
PiperOrigin-RevId: 222280209
In the general case, loop bounds can be expressed as affine maps of the outer
loop iterators and function arguments. Relax the check for loop bounds to be
known integer constants and also accept one-dimensional affine bounds in
ConvertToCFG ForStmt lowering. Emit affine_apply operations for both the upper
and the lower bound. The semantics of MLFunctions guarantees that both bounds
can be computed before the loop starts iterating. Constant bounds are merely a
short-hand notation for zero-dimensional affine maps and get supported
transparently.
Multidimensional affine bounds are not yet supported because the target IR
dialect lacks min/max operations necessary to implement the corresponding
semantics.
PiperOrigin-RevId: 222275801
op-stats pass currently returns the number of occurrences of different operations in a Module. Useful for verifying transformation properties (e.g., 3 ops of specific dialect, 0 of another), but probably not useful outside of that so keeping it local to mlir-opt. This does not consider op attributes when counting.
PiperOrigin-RevId: 222259727
This CL adds some vector support in prevision of the upcoming vector
materialization pass. In particular this CL adds 2 functions to:
1. compute the multiplicity of a subvector shape in a supervector shape;
2. help match operations on strict super-vectors. This is defined for a given
subvector shape as an operation that manipulates a vector type that is an
integral multiple of the subtype, with multiplicity at least 2.
This CL also adds a TestUtil pass where we can dump arbitrary testing of
functions and analysis that operate at a much smaller granularity than a pass
(e.g. an analysis for which it is convenient to write a bit of artificial MLIR
and write some custom test). This is in order to keep using Filecheck for
things that essentially look and feel like C++ unit tests.
PiperOrigin-RevId: 222250910
and getMemRefRegion() to work with specified loop depths; add support for
outgoing DMAs, store op's.
- add support for getMemRefRegion symbolic in outer loops - hence support for
DMAs symbolic in outer surrounding loops.
- add DMA generation support for outgoing DMAs (store op's to lower memory
space); extend getMemoryRegion to store op's. -memref-bound-check now works
with store op's as well.
- fix dma-generate (references to the old memref in the dma_start op were also
being replaced with the new buffer); we need replace all memref uses to work
only on a subset of the uses - add a new optional argument for
replaceAllMemRefUsesWith. update replaceAllMemRefUsesWith to take an optional
'operation' argument to serve as a filter - if provided, only those uses that
are dominated by the filter are replaced.
- Add missing print for attributes for dma_start, dma_wait op's.
- update the FlatAffineConstraints API
PiperOrigin-RevId: 221889223
Array attributes can nested and function attributes can appear anywhere at that
level. They should be remapped to point to the generated CFGFunction after
ML-to-CFG conversion, similarly to plain function attributes. Extract the
nested attribute remapping functionality from the Parser to Utils. Extract out
the remapping function for individual Functions from the module remapping
function. Use these new functions in the ML-to-CFG conversion pass and in the
parser.
PiperOrigin-RevId: 221510997
These functions are declared in Transforms/LoopUtils.h (included to the
Transforms/Utils library) but were defined in the loop unrolling pass in
Transforms/LoopUnroll.cpp. As a result, targets depending only on
TransformUtils library but not on Transforms could get link errors. Move the
definitions to Transforms/Utils/LoopUtils.cpp where they should actually live.
This does not modify any code.
PiperOrigin-RevId: 221508882
This CL adds support for and a vectorization test to perform scalar 2-D addf.
The support extension notably comprises:
1. extend vectorizable test to exclude vector_transfer operations and
expose them to LoopAnalysis where they are needed. This is a temporary
solution a concrete MLIR Op exists;
2. add some more functional sugar mapKeys, apply and ScopeGuard (which became
relevant again);
3. fix improper shifting during coarsening;
4. rename unaligned load/store to vector_transfer_read/write and simplify the
design removing the unnecessary AllocOp that were introduced prematurely:
vector_transfer_read currently has the form:
(memref<?x?x?xf32>, index, index, index) -> vector<32x64x256xf32>
vector_transfer_write currently has the form:
(vector<32x64x256xf32>, memref<?x?x?xf32>, index, index, index) -> ()
5. adds vectorizeOperations which traverses the operations in a ForStmt and
rewrites them to their vector form;
6. add support for vector splat from a constant.
The relevant tests are also updated.
PiperOrigin-RevId: 221421426
Implement a pass converting a subset of MLFunctions to CFGFunctions. Currently
supports arbitrarily complex imperfect loop nests with statically constant
(i.e., not affine map) bounds filled with operations. Does NOT support
branches and non-constant loop bounds.
Conversion is performed per-function and the function names are preserved to
avoid breaking any external references to the current module. In-memory IR is
updated to point to the right functions in direct calls and constant loads.
This behavior is tested via a really hidden flag that enables function
renaming.
Inside each function, the control flow conversion is based on single-entry
single-exit regions, i.e. subgraphs of the CFG that have exactly one incoming
and exactly one outgoing edge. Since an MLFunction must have a single "return"
statement as per MLIR spec, it constitutes an SESE region. Individual
operations are appended to this region. Control flow statements are
recursively converted into such regions that are concatenated with the current
region. Bodies of the compound statement also form SESE regions, which allows
to nest control flow statements easily. Note that SESE regions are not
materialized in the code. It is sufficent to keep track of the end of the
region as the current instruction insertion point as long as all recursive
calls update the insertion point in the end.
The converter maintains a mapping between SSA values in ML functions and their
CFG counterparts. The mapping is used to find the operands for each operation
and is updated to contain the results of each operation as the conversion
continues.
PiperOrigin-RevId: 221162602
Change the storage type to APInt from int64_t for IntegerAttr (following the change to APFloat storage in FloatAttr). Effectively a direct change from int64_t to 64-bit APInt throughout (the bitwidth hardcoded). This change also adds a getInt convenience method to IntegerAttr and replaces previous getValue calls with getInt calls.
While this changes updates the storage type, it does not update all constant folding calls.
PiperOrigin-RevId: 221082788
Updates MemRefDependenceCheck to check and report on all memref access pairs at all loop nest depths.
Updates old and adds new memref dependence check tests.
Resolves multiple TODOs.
PiperOrigin-RevId: 220816515
- constant bounded memory regions, static shapes, no handling of
overlapping/duplicate regions (through union) for now; also only, load memory
op's.
- add build methods for DmaStartOp, DmaWaitOp.
- move getMemoryRegion() into Analysis/Utils and expose it.
- fix addIndexSet, getMemoryRegion() post switch to exclusive upper bounds;
update test cases for memref-bound-check and memref-dependence-check for
exclusive bounds (missed in a previous CL)
PiperOrigin-RevId: 220729810
Value type abstraction for locations differ from others in that a Location can NOT be null. NOTE: dyn_cast returns an Optional<T>.
PiperOrigin-RevId: 220682078
The passID is not currently stored in Pass but this avoids the unused variable warning. The passID is used to uniquely identify passes, currently this is only stored/used in PassInfo.
PiperOrigin-RevId: 220485662
This CL implement exclusive upper bound behavior as per b/116854378.
A followup CL will update the semantics of the for loop.
PiperOrigin-RevId: 220448963
Add static pass registration and change mlir-opt to use it. Future work is needed to refactor the registration for PassManager usage.
Change build targets to alwayslink to enforce registration.
PiperOrigin-RevId: 220390178
- simple perfectly nested band tiling with fixed tile sizes.
- only the hyper-rectangular case is handled, with other limitations of
getIndexSet applying (constant loop bounds, etc.); once
the latter utility is extended, tiled code generation should become more
general.
- Add FlatAffineConstraints::isHyperRectangular()
PiperOrigin-RevId: 220324933
- Builds access functions and iterations domains for each access.
- Builds dependence polyhedron constraint system which has equality constraints for equated access functions and inequality constraints for iteration domain loop bounds.
- Runs elimination on the dependence polyhedron to test if no dependence exists between the accesses.
- Adds a trivial LoopFusion transformation pass with a simple test policy to test dependence between accesses to the same memref in adjacent loops.
- The LoopFusion pass will be extended in subsequent CLs.
PiperOrigin-RevId: 219630898
This CL adds support for vectorization using more interesting 2-D and 3-D
patterns. Note in particular the fact that we match some pretty complex
imperfectly nested 2-D patterns with a quite minimal change to the
implementation: we just add a bit of recursion to traverse the matched
patterns and actually vectorize the loops.
For instance, vectorizing the following loop by 128:
```
for %i3 = 0 to %0 {
%7 = affine_apply (d0) -> (d0)(%i3)
%8 = load %arg0[%c0_0, %7] : memref<?x?xf32>
}
```
Currently generates:
```
#map0 = ()[s0] -> (s0 + 127)
#map1 = (d0) -> (d0)
for %i3 = 0 to #map0()[%0] step 128 {
%9 = affine_apply #map1(%i3)
%10 = alloc() : memref<1xvector<128xf32>>
%11 = "n_d_unaligned_load"(%arg0, %c0_0, %9, %10, %c0) :
(memref<?x?xf32>, index, index, memref<1xvector<128xf32>>, index) ->
(memref<?x?xf32>, index, index, memref<1xvector<128xf32>>, index)
%12 = load %10[%c0] : memref<1xvector<128xf32>>
}
```
The above is subject to evolution.
PiperOrigin-RevId: 219629745
FuncBuilder is useful to build a operation to replace an existing operation, so change the constructor to allow constructing it with an existing operation. Change FuncBuilder to contain (effectively) a tagged union of CFGFuncBuilder and MLFuncBuilder (as these should be cheap to copy and avoid allocating/deletion when created via a operation).
PiperOrigin-RevId: 219532952
Introduce analysis to check memref accesses (in MLFunctions) for out of bound
ones. It works as follows:
$ mlir-opt -memref-bound-check test/Transforms/memref-bound-check.mlir
/tmp/single.mlir:10:12: error: 'load' op memref out of upper bound access along dimension tensorflow/mlir#1
%x = load %A[%idxtensorflow/mlir#0, %idxtensorflow/mlir#1] : memref<9 x 9 x i32>
^
/tmp/single.mlir:10:12: error: 'load' op memref out of lower bound access along dimension tensorflow/mlir#1
%x = load %A[%idxtensorflow/mlir#0, %idxtensorflow/mlir#1] : memref<9 x 9 x i32>
^
/tmp/single.mlir:10:12: error: 'load' op memref out of upper bound access along dimension tensorflow/mlir#2
%x = load %A[%idxtensorflow/mlir#0, %idxtensorflow/mlir#1] : memref<9 x 9 x i32>
^
/tmp/single.mlir:10:12: error: 'load' op memref out of lower bound access along dimension tensorflow/mlir#2
%x = load %A[%idxtensorflow/mlir#0, %idxtensorflow/mlir#1] : memref<9 x 9 x i32>
^
/tmp/single.mlir:12:12: error: 'load' op memref out of upper bound access along dimension tensorflow/mlir#1
%y = load %B[%idy] : memref<128 x i32>
^
/tmp/single.mlir:12:12: error: 'load' op memref out of lower bound access along dimension tensorflow/mlir#1
%y = load %B[%idy] : memref<128 x i32>
^
#map0 = (d0, d1) -> (d0, d1)
#map1 = (d0, d1) -> (d0 * 128 - d1)
mlfunc @test() {
%0 = alloc() : memref<9x9xi32>
%1 = alloc() : memref<128xi32>
for %i0 = -1 to 9 {
for %i1 = -1 to 9 {
%2 = affine_apply #map0(%i0, %i1)
%3 = load %0[%2tensorflow/mlir#0, %2tensorflow/mlir#1] : memref<9x9xi32>
%4 = affine_apply #map1(%i0, %i1)
%5 = load %1[%4] : memref<128xi32>
}
}
return
}
- Improves productivity while manually / semi-automatically developing MLIR for
testing / prototyping; also provides an indirect way to catch errors in
transformations.
- This pass is an easy way to test the underlying affine analysis
machinery including low level routines.
Some code (in getMemoryRegion()) borrowed from @andydavis cl/218263256.
While on this:
- create mlir/Analysis/Passes.h; move Pass.h up from mlir/Transforms/ to mlir/
- fix a bug in AffineAnalysis.cpp::toAffineExpr
TODO: extend to non-constant loop bounds (straightforward). Will transparently
work for all accesses once floordiv, mod, ceildiv are supported in the
AffineMap -> FlatAffineConstraints conversion.
PiperOrigin-RevId: 219397961
This is done by changing Type to be a POD interface around an underlying pointer storage and adding in-class support for isa/dyn_cast/cast.
PiperOrigin-RevId: 219372163
This CL is a first in a series that implements early vectorization of
increasingly complex patterns. In particular, early vectorization will support
arbitrary loop nesting patterns (both perfectly and imperfectly nested), at
arbitrary depths in the loop tree.
This first CL builds the minimal support for applying 1-D patterns.
It relies on an unaligned load/store op abstraction that can be inplemented
differently on different HW.
Future CLs will support higher dimensional patterns, but 1-D patterns already
exhibit interesting properties.
In particular, we want to separate pattern matching (i.e. legality both
structural and dependency analysis based), from profitability analysis, from
application of the transformation.
As a consequence patterns may intersect and we need to verify that a pattern
can still apply by the time we get to applying it.
A non-greedy analysis on profitability that takes into account pattern
intersection is left for future work.
Additionally the CL makes the following cleanups:
1. the matches method now returns a value, not a reference;
2. added comments about the MLFunctionMatcher and MLFunctionMatches usage by
value;
3. added size and empty methods to matches;
4. added a negative vectorization test with a conditional, this exhibited a
but in the iterators. Iterators now return nullptr if the underlying storage
is nullpt.
PiperOrigin-RevId: 219299489
1) We incorrectly reassociated non-reassociative operations like subi, causing
miscompilations.
2) When constant folding, we didn't add users of the new constant back to the
worklist for reprocessing, causing us to miss some cases (pointed out by
Uday).
The code for tensorflow/mlir#2 is gross, but I'll add the new APIs in a followup patch.
PiperOrigin-RevId: 218803984
distinction. FunctionPasses can now choose to get called on all functions, or
have the driver split CFG/ML Functions up for them. NFC.
PiperOrigin-RevId: 218775885
make operations provide a list of canonicalizations that can be applied to
them. This allows canonicalization to be general to any IR definition.
As part of this, sink PatternMatch.h/cpp down to the IR library to fix a
layering problem.
PiperOrigin-RevId: 218773981
This is done by changing Attribute to be a POD interface around an underlying pointer storage and adding in-class support for isa/dyn_cast/cast.
PiperOrigin-RevId: 218764173
just having the pattern matcher in its own library. At this point,
lib/Transforms/*.cpp are all actually passes themselves (and will probably
eventually be themselves move to a new subdirectory as we accrete more).
PiperOrigin-RevId: 218745193
helper function, in preparation for it being used by other passes.
There is still a lot of room for improvement in its design, this patch is
intended as an NFC refactoring, and the improvements will continue after this
lands.
PiperOrigin-RevId: 218737116
- Introduce Fourier-Motzkin variable elimination to eliminate a dimension from
a system of linear equalities/inequalities. Update isEmpty to use this.
Since FM is only exact on rational/real spaces, an emptiness check based on
this is guaranteed to be exact whenever it says the underlying set is empty;
if it says, it's not empty, there may still be no integer points in it.
Also, supports a version that computes "dark shadows".
- Test this by checking for "always false" conditionals in if statements.
- Unique IntegerSet's that are small (few constraints, few variables). This
basically means the canonical empty set and other small sets that are
likely commonly used get uniqued; allows checking for the canonical empty set
by pointer. IntegerSet::kUniquingThreshold gives the threshold constraint size
for uniqui'ing.
- rename simplify-affine-expr -> simplify-affine-structures
Other cleanup
- IntegerSet::numConstraints, AffineMap::numResults are no longer needed;
remove them.
- add copy assignment operators for AffineMap, IntegerSet.
- rename Invalid() -> Null() on AffineExpr, AffineMap, IntegerSet
- Misc cleanup for FlatAffineConstraints API
PiperOrigin-RevId: 218690456
- Adds FlatAffineConstraints::isEmpty method to test if there are no solutions to the system.
- Adds GCD test check if equality constraints have no solution.
- Adds unit test cases.
PiperOrigin-RevId: 218546319
is a straight-forward change, but required adding missing moveBefore() methods
on operations (requiring moving some traits around to make C++ happy). This
also fixes a constness issue with the getBlock/getFunction() methods on
Instruction, and adds a missing getFunction() method on MLFuncBuilder.
PiperOrigin-RevId: 218523905
- Add a few canonicalization patterns to fold memref_cast into
load/store/dealloc.
- Canonicalize alloc(constant) into an alloc with a constant shape followed by
a cast.
- Add a new PatternRewriter::updatedRootInPlace API to make this more convenient.
SimplifyAllocConst and the testcase is heavily based on Uday's implementation work, just
in a different framework.
PiperOrigin-RevId: 218361237