This pull request builds on #3267 by proving the "validation" infrastructure, the means to compare a set of `Outputs`.
The design of the validation infrastructure is relatively straight forward:
- Each SIGNATURE should come with a `validate()` implementation, which should be implemented in a similar way that the other functions/types from `testing.hpp` are implemented.
- `validate()` returns a `ValidationReport`, which is a structure that keeps all relevant information about comparing the tensors from two `Outputs`. Note that crucially, `validate()` should not do any reporting by itself. Rather, glue logic should be implemented by the user to turn `ValidationReport` into a relevant error message.
- You can see this clue code for CK-Builder itself in `testing_utils.hpp`, its `MatchesReference()`. This functionality is relatively barebones right now, it will be expanded upon in a different PR to keep the scope of this one down.
The comparison is done on the GPU (using an atomic for now), to keep tests relatively quick. Some notable items from this PR:
- To help compare the tensors and with writing tests, I've written a generic function `tensor_foreach` which invokes a callback on every element of a tensor.
- For that it was useful that the `TensorDescriptor` has a rank which is known at compile-time, so I've changed the implementation of `TensorDescriptor` for that. I felt like it was a better approach than keeping it dynamic, for multiple reasons:
- This is C++ and we should use static typing where possible and useful. This way, we don't have to implement runtime assertions about the tensor rank.
- We know already know the rank of tensors statically, as it can be derived from the SIGNATURE.
- It simpifies the implementation of `tensor_foreach` and other comparison code.
- There are a lot of new tests for validating the validation implementation, validating validation validation tests (Only 3 recursive levels though...). For a few of those functions, I felt like it would be useful to expose them to the user.
- Doc comments everywhere.
[ROCm/composable_kernel commit: e6e7dc2910]
Composable Kernel Builder Design Documentation
This directory contains the builder framework for Composable Kernel, which provides a compile-time, type-safe interface for constructing convolution operations with various configurations.
Table of Contents
Convolution Signature
Overview
The convolution signature system provides a compile-time description of grouped convolution operations. A signature is a collection of properties that fully characterize a convolution kernel's mathematical and operational behavior, enabling:
- Compile-time validation: Ensures type safety and correctness before kernel instantiation
- Kernel selection: Matches user requirements to optimized implementations
- Specialization: Enables optimized code paths for specific configurations
- Composability: Supports building complex operations from simpler components
The signature leverages modern C++20 features, particularly concepts, to provide expressive, self-documenting interfaces with compile-time guarantees.
Architecture
The signature system is organized into a hierarchical structure:
┌─────────────────────────────────────────────────────────┐
│ ConvSignature │
├─────────────────────────────────────────────────────────┤
│ Properties: │
│ • spatial_dim: int (1D, 2D, or 3D) │
│ • direction: ConvDirection (Fwd/BwdData/BwdWeight) │
│ • data_type: DataType (default data type) │
│ • accumulation_data_type: DataType │
│ • input: ConvTensor ──┐ │
│ • weight: ConvTensor ──│ │
│ • output: ConvTensor ──│ │
└──────────────────────────────────┼──────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ ConvTensor │
├─────────────────────────────────────────┤
│ ╔═════════════════════════════════════╗ │
│ ║ TensorConfig (required) ║ │
│ ╠═════════════════════════════════════╣ │
│ ║ • layout: ConvLayout ║ │
│ ║ • data_type: DataType (optional) ║ │
│ ║ • compute_type: DataType (optional)║ │
│ ╚═════════════════════════════════════╝ │
│ │
│ ┌─────────────────────────────────────┐ │
│ │ TensorOperation (optional) │ │
│ ├─────────────────────────────────────┤ │
│ │ • elementwise_operation │ │
│ │ • auxiliary_operand_configs[] │ │
│ │ (each is also ConvTensor) ◄───────┼─┐
│ └─────────────────────────────────────┘ │ │
└─────────────────────────────────────────┘ │
│
Recursive ───────────────┘
Key Design Points:
- ConvSignature contains three ConvTensor instances (input, weight, output)
- All tensors share the same ConvTensor structure
- Each ConvTensor has:
- TensorConfig (required): Defines layout as well as optional data and compute type overrides
- TensorOperation (optional): Defines fused elementwise operations
- Auxiliary operands (e.g., bias) in TensorOperation also use the ConvTensor type
Core Components
1. Signature Level
The top-level signature contains global properties that apply to the entire convolution operation:
template <typename T>
concept ConvSignatureDescriptor = requires(T t) {
{ t.spatial_dim } -> std::convertible_to<unsigned int>; // 1, 2, or 3
{ t.data_type } -> std::convertible_to<DataType>; // Default data type
{ t.input } -> ConvTensorDescriptor;
{ t.weight } -> ConvTensorDescriptor;
{ t.output } -> ConvTensorDescriptor;
requires ConvolutionDirectionWellDefinedIfProvided<T>; // Optional direction
};
Properties:
spatial_dim: Dimensionality of the convolution (1D, 2D, or 3D)direction: Operation type (optional, defaults to FORWARD)FORWARD: Standard forward convolutionBACKWARD_DATA: Gradient computation w.r.t. inputBACKWARD_WEIGHT: Gradient computation w.r.t. weights
data_type: Default data type for all tensors (FP32, FP16, BF16, FP8, I8, U8)accumulation_data_type: Type used for internal accumulation
2. Tensor Level
Each tensor (input, weight, output) has its own descriptor:
template <typename T>
concept ConvTensorDescriptor = requires(T t) {
{ t.config } -> TensorConfigDescriptor;
requires ElementwiseOpWellDefinedIfProvided<T>;
};
A tensor descriptor encapsulates:
- Configuration: Layout and data type information
- Operation (optional): Fused elementwise operations on this tensor
3. Tensor Configuration
Describes the memory layout and data types:
template <typename T>
concept TensorConfigDescriptor = requires(T t) {
{ t.layout } -> std::convertible_to<ConvLayout>;
{ t.data_type } -> std::convertible_to<DataType>; // Optional override
};
Layout Types (dimension-specific):
-
1D Convolution:
- Input:
GNCW,GNWC,NWGC,NGCW,G_NW_C_strided - Weight:
GKXC,GKCX,KXGC,G_K_X_C_strided - Output:
GNKW,GNWK,NWGK,NGKW,G_NW_K_strided
- Input:
-
2D Convolution:
- Input:
GNCHW,GNHWC,NHWGC,NGCHW,G_NHW_C_strided - Weight:
GKYXC,GKCYX,KYXGC,G_K_YX_C_strided - Output:
GNKHW,GNHWK,NHWGK,NGKHW,G_NHW_K_strided
- Input:
-
3D Convolution:
- Input:
GNCDHW,GNDHWC,NDHWGC,NGCDHW,G_NDHW_C_strided - Weight:
GKZYXC,GKCZYX,KZYXGC,G_K_ZYX_C_strided - Output:
GNKDHW,GNDHWK,NDHWGK,NGKDHW,G_NDHW_K_strided
- Input:
Where:
G= GroupsN= Batch sizeC= Input channelsK= Output channels (filters)W,H,D= Width, Height, Depth (spatial dimensions)X,Y,Z= Filter dimensions
4. Tensor Operations
Describes fused elementwise operations applied to a tensor:
template <typename T>
concept TensorOperatorDescriptor = requires(T t) {
{ t.elementwise_operation } -> std::convertible_to<ElementwiseOperation>;
requires AuxiliaryOperandConfigsWellDefinedIfProvided<T>;
};
Supported Operations:
PASS_THROUGH: No operation (identity)SCALE: Multiply by a scalarCLAMP: Clamp values to a rangeBIAS_BNORM_CLAMP: Bias addition + batch normalization + clampSCALEADD_SCALEADD_RELU: Fused scale-add operations + ReLU activation
Auxiliary Operands:
Some operations require additional tensor inputs (e.g., bias tensors, scaling factors). These are specified through auxiliary_operand_configs, which is an array of TensorConfigDescriptor objects describing the layout and data type of each auxiliary input.
Concepts and Validation
The signature system uses C++20 concepts for compile-time validation at multiple levels:
Constraint Concepts
// Spatial dimension must be 1, 2, or 3
template <auto N>
concept ConvSpatialDim = std::is_integral_v<decltype(N)> && (N == 1 || N == 2 || N == 3);
// Valid data types for convolution
template <DataType T>
concept ValidConvDataType =
(T == DataType::FP32) || (T == DataType::FP16) || (T == DataType::BF16) ||
(T == DataType::FP8) || (T == DataType::I8) || (T == DataType::U8);
Validation Concept
// Validates a complete signature
template <auto Sig>
concept ValidConvSignature = requires {
requires ConvSpatialDim<Sig.spatial_dim>;
requires ValidConvDataType<Sig.data_type>;
};
Tensor Descriptors
The layout/data type/elementwise operation are described per tensor. This multi-level hierarchy allows:
- Flexibility: Each tensor can have independent layout and data type
- Reusability: Common configurations can be shared across different signatures
- Extensibility: New properties can be added to specific levels without affecting others
- Clarity: Separates concerns (global properties vs. tensor-specific properties)
Optional Signature Fields
Several fields in the signature are optional:
direction: Defaults toFORWARDif not specified, reducing boilerplate for the common case- Tensor
data_type: Falls back to signature's default, allowing mixed-precision with minimal specification - Tensor
operation: Defaults toPASS_THROUGH, supporting both fused and non-fused operations with the same interface
This design follows the principle of "make the common case simple, the complex case possible."
Convolution Algorithm
Convolution Factory
Convolution factory builds the instance based on the convolution signature and convolution algorithm. The signature and the algorithm descriptions are dispatched to the relevant algorithm specific factory for instance creation. The convolution factory design is described in a separate Readme.