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
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
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
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
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
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
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
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 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
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
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
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
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
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
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
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
