|
This change standardizes the order of the parameters for `Constant[XXX]
Ops` to match with all other `Op` `build()` constructors.
In all instances of generated code for the MLIR dialects's Ops (that is
the TableGen using the .td files to create the .h.inc/.cpp.inc files),
the desired result type is always specified before the value.
Examples:
```
// ArithOps.h.inc
class ConstantOp : public ::mlir::Op<ConstantOp, ::mlir::OpTrait::ZeroRegions, ::mlir::OpTrait::OneResult, ::mlir::OpTrait::OneTypedResult<::mlir::Type>::Impl, ::mlir::OpTrait::ZeroSuccessors, ::mlir::OpTrait::ZeroOperands, ::mlir::OpTrait::OpInvariants, ::mlir::BytecodeOpInterface::Trait, ::mlir::OpTrait::ConstantLike, ::mlir::ConditionallySpeculatable::Trait, ::mlir::OpTrait::AlwaysSpeculatableImplTrait, ::mlir::MemoryEffectOpInterface::Trait, ::mlir::OpAsmOpInterface::Trait, ::mlir::InferIntRangeInterface::Trait, ::mlir::InferTypeOpInterface::Trait> {
public:
....
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::Type result, ::mlir::TypedAttr value);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::TypedAttr value);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::TypeRange resultTypes, ::mlir::TypedAttr value);
static void build(::mlir::OpBuilder &, ::mlir::OperationState &odsState, ::mlir::TypeRange resultTypes, ::mlir::ValueRange operands, ::llvm::ArrayRef<::mlir::NamedAttribute> attributes = {});
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::ValueRange operands, ::llvm::ArrayRef<::mlir::NamedAttribute> attributes = {});
...
```
```
// ArithOps.h.inc
class SubIOp : public ::mlir::Op<SubIOp, ::mlir::OpTrait::ZeroRegions, ::mlir::OpTrait::OneResult, ::mlir::OpTrait::OneTypedResult<::mlir::Type>::Impl, ::mlir::OpTrait::ZeroSuccessors, ::mlir::OpTrait::NOperands<2>::Impl, ::mlir::OpTrait::OpInvariants, ::mlir::BytecodeOpInterface::Trait, ::mlir::ConditionallySpeculatable::Trait, ::mlir::OpTrait::AlwaysSpeculatableImplTrait, ::mlir::MemoryEffectOpInterface::Trait, ::mlir::InferIntRangeInterface::Trait, ::mlir::arith::ArithIntegerOverflowFlagsInterface::Trait, ::mlir::OpTrait::SameOperandsAndResultType, ::mlir::VectorUnrollOpInterface::Trait, ::mlir::OpTrait::Elementwise, ::mlir::OpTrait::Scalarizable, ::mlir::OpTrait::Vectorizable, ::mlir::OpTrait::Tensorizable, ::mlir::InferTypeOpInterface::Trait> {
public:
...
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::Type result, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlagsAttr overflowFlags);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlagsAttr overflowFlags);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::TypeRange resultTypes, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlagsAttr overflowFlags);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::Type result, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlags overflowFlags = ::mlir::arith::IntegerOverflowFlags::none);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlags overflowFlags = ::mlir::arith::IntegerOverflowFlags::none);
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::TypeRange resultTypes, ::mlir::Value lhs, ::mlir::Value rhs, ::mlir::arith::IntegerOverflowFlags overflowFlags = ::mlir::arith::IntegerOverflowFlags::none);
static void build(::mlir::OpBuilder &, ::mlir::OperationState &odsState, ::mlir::TypeRange resultTypes, ::mlir::ValueRange operands, ::llvm::ArrayRef<::mlir::NamedAttribute> attributes = {});
static void build(::mlir::OpBuilder &odsBuilder, ::mlir::OperationState &odsState, ::mlir::ValueRange operands, ::llvm::ArrayRef<::mlir::NamedAttribute> attributes = {});
...
```
In comparison, in the distinct case of `ConstantIntOp` and
`ConstantFloatOp`, the ordering of the result type and the value is
switched.
Thus, this PR corrects the ordering of the aforementioned
`Constant[XXX]Ops` to match with other constructors.
|
|
This is an implementation for [RFC: Supporting Sub-Channel Quantization
in
MLIR](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694).
In order to make the review process easier, the PR has been divided into
the following commit labels:
1. **Add implementation for sub-channel type:** Includes the class
design for `UniformQuantizedSubChannelType`, printer/parser and bytecode
read/write support. The existing types (per-tensor and per-axis) are
unaltered.
2. **Add implementation for sub-channel type:** Lowering of
`quant.qcast` and `quant.dcast` operations to Linalg operations.
3. **Adding C/Python Apis:** We first define he C-APIs and build the
Python-APIs on top of those.
4. **Add pass to normalize generic ....:** This pass normalizes
sub-channel quantized types to per-tensor per-axis types, if possible.
A design note:
- **Explicitly storing the `quantized_dimensions`, even when they can be
derived for ranked tensor.**
While it's possible to infer quantized dimensions from the static shape
of the scales (or zero-points) tensor for ranked
data tensors
([ref](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694/3)
for background), there are cases where this can lead to ambiguity and
issues with round-tripping.
```
Consider the example: tensor<2x4x!quant.uniform<i8:f32:{0:2, 0:2}, {{s00:z00, s01:z01}}>>
```
The shape of the scales tensor is [1, 2], which might suggest that only
axis 1 is quantized. While this inference is technically correct, as the
block size for axis 0 is a degenerate case (equal to the dimension
size), it can cause problems with round-tripping. Therefore, even for
ranked tensors, we are explicitly storing the quantized dimensions.
Suggestions welcome!
PS: I understand that the upcoming holidays may impact your schedule, so
please take your time with the review. There's no rush.
|
