Old sequence sort code was showing up on build profiles. Convert it to constexpr functions for much more efficient build-time execution. The sorting is still O(N^2), but our sequences are small enough it executes quickly. This reduced compilation time of a small convolution by more than 10% and time overall time spent in the compiler on a narrow build by %6.
[ROCm/composable_kernel commit: 15ed65db35]
* WIP: preparing to add transpose bq support
* WIP: handle both row/col layout for BQ windows/tile dstr
* Fix build
* WIP: adding some test, debugging numerical errors
* Fix all but pkint4 tests
* Remove test_gemm_quant_typed.cpp again
* update disabled tests
* add conversion from pkint4 for b matrix
* fix formatting
* fix formatting
* Fix tr_load and use override b datatype for clarity
* fix formatting
* make bquant preshuffle tests bqlayout column-major
[ROCm/composable_kernel commit: c363a98d41]
* wip: add aquant to grouped gemm quant example
* fix: properly handle hot loop count in aquant pipeline
* fix: add separate GemmConfig structs for AQuant, automatically select the correct one
* feat: finish support for a non-persistent kernel invocation for grouped gemm quant, and add support code to example
* refactor: cleaned up grouped gemm quant example a bit by reusing pipeline selection logic
* chore: add warp gemm dispatchers for a couple of TransposeC K=32 variants
* feat: add quant grouped gemm tests cases for aquant (regular and transpose C) and non-persistent kernel
* fix: update base pipeline classes according to changes in develop branch
* Revert "chore: add warp gemm dispatchers for a couple of TransposeC K=32 variants"
This reverts commit b3fd4d326d.
* feat: remove aquant config from grouped gemm quant example, update to add persistency as runtime parameter
* chore: removed work-around for aquant bug that has been fixed
* chore: fix typo in command-line parameters
* fix: correct K warp tile size for gfx950
* chore: incorrect warp tile configuration on gfx942
[ROCm/composable_kernel commit: fe07b5a1bf]
Sometimes there are not enough wait-states between v_mfma_f32... and v_accvgpr_read_b32 instructions if they are separated by s_cbranch.
The workaround is to read accvgprs to vgpr before branching.
[ROCm/composable_kernel commit: ca6143f0b2]
We don't want to add cmake formatting until we are in the super repo, but its handy if developers want to experiment with formatting. For now we should ignore .cmake-format.yaml.
[ROCm/composable_kernel commit: 7541d9b5b0]
We only want to import enums and types into the builder reflection code. But, some of the enums are included in much larger files or even big trees of include files. This leads to unintended mixing of code and very confusing interactions and symbol conflicts. We organize the includes and extract two new enum-only headers to help with decoupling in CK. This refactoring is critical if we want to include reflection in a device-operator "describe" method.
* Remove a few unnecessary includes from headers in builder/reflect/.
* Extract enums scheduler and pipeline to their own headers so they can be used without importing other code.
* Order includes alphabetically for better organization.
The immediate goal is to unblock reflection integration, and this type of cleanup helps the flexibility and robustness of the CK header library.
[ROCm/composable_kernel commit: f5b0af2272]
When I asked for a description of operators that didn't have ConvTraits, I was getting very long confusing errors about ConvTraits not being defined. Now we get specific errors explaining which concepts are violated, making it easier to know which code to generalize or update.
* Add concepts to conv_traits.hpp to get better error message.
* Put the correct requires clauses in the right places to get descriptive error messages.
* General cleanup of functions in conv_traits.hpp to make functions easier to read.
[ROCm/composable_kernel commit: 13f6d63565]
* initial poc
* factor out common parts in operator()
* cv4
* rest of the universal gemm pipelines
* fix test
* remove boilerplate from tile engine
* fix example
* fix example
* format
* fix tests build for gemm
* remove base pipeline codegen from gemm instance builder
* unify v3 logic with the rest of universal gemm pipelines
* fix build for multi abd test
* fix test gemm multi d
* fix build for weight preshuffle
* fix grouped gemm test
* fix grouped gemm multi d test
* fix grouped gemm preshuffle
* fix grouped gemm example except for quant
* fix gemm preshuffle
* fix splitk 2 stage example
* fix batched gemm example
* fix multid example
* fix multiabd example
* fix batched gemm test
* fixup
* fix examples build
* fix grouped gemm test build
* fix smoke builder
[ROCm/composable_kernel commit: d184eed823]
* Separate layouts into separate entities for input, weight, and output tensors.
* Add test for handling bias tensor layouts.
* Use instance string in builder tests.
* Add handling of output bias data types and layouts.
* Generalize handling of the elementwise ops.
* Test fix.
* Create builder for layouts.
* Layout builder improvements.
* Improve layout builder.
* Simplify bias layout handling.
* Code clean-up.
* Move layout utils into separate file.
* Remove hard-coded layout combinations.
* Small code clean-up.
* Move data type utils into a separate file.
* Add data types, layouts, and elementwise ops per conv tensor.
* Builder bug fixes after refactoring.
* Working baseline.
* Make signature definition look nice in the test code.
* Move TensorConfig into test implementations.
* Fix all fwd conv builder tests.
* Fix conv traits and descriptors tests.
* More factory assets under a separate directory.
* Fix building conv traits.
* Fix clang-format.
* Add Readme doc to describe the design.
* Add link to main Readme. Fix links in the builder design doc.
* Clean-up data type/layout/elementwise op conversions.
* Switch from dimension and tensor type specific layouts to a flat list of tensor layouts.
* Fix clang-formatting.
* Fix clang-format for test code.
* Simplify fwd conv signature definitions in the test code.
* Remove accidental edits.
* Fix comment string.
* Fix instance factory after rebase.
* Fix tests after rebase.
* Unify layout handling.
* Add more conv layout unit tests.
* Clang-format.
* Fix merge conflicts.
* Improve elementwise op handling.
---------
Co-authored-by: Ville Pietilä <>
[ROCm/composable_kernel commit: 9cb1f421bc]
This PR fixes a mismatch caused when PR #3146 was merged out of sync with develop, which made its intended changes ineffective. This PR reapplies those changes to move DataTypeTraits into a common file to mitigate code duplication.
Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
[ROCm/composable_kernel commit: 583fafc803]
* replace qr with async pipeline
* Add fp8fp32 to DTYPE_BITS
* Add kAlignmentRandVal to avoid compile fail
* format
---------
Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
[ROCm/composable_kernel commit: eb7f617713]
* LWPCK-4043: Add GPU reference implementations for CK Tile convolution
This commit implements GPU-based reference kernels for CK Tile convolution
operations to enable faster verification of optimized kernels, especially
for large tensors (>2GB).
Changes:
- Add naive_grouped_conv_fwd.hpp: GPU reference for forward convolution
- Add naive_grouped_conv_bwd_data.hpp: GPU reference for backward data
- Add naive_grouped_conv_bwd_weight.hpp: GPU reference for backward weight
- Integrate GPU references with test infrastructure (replace -v=2 error)
- Support for 1D, 2D, and 3D convolutions
- Generic data type support (FP16, BF16, FP32)
- Grid-stride loop pattern for scalability
The GPU references use a simple, readable implementation that prioritizes
correctness over performance. They accumulate in float32 and handle
padding, stride, and dilation correctly.
* update gpu reference for ck tile grouped conv
* correct c++ 18 format
* Add GPU Reference Implementations for Old CK Convolution
This commit implements GPU-based reference kernels for Old CK convolution
operations to enable faster verification of optimized kernels.
Changes:
- Fixed old CK forward GPU reference (naive_conv_fwd.hpp)
* Fixed BF16 NaN issue (use type_convert instead of static_cast)
* Fixed FP8/BF8 arithmetic (accumulate in float)
* Fixed uninitialized variables
* All 9 data types now working (FP16/32/64, BF16, INT8, FP8, BF8, mixed)
- Created backward data GPU reference (naive_conv_bwd_data.hpp)
* Implements input gradient computation
* Verified equal to CPU reference
* Handles 1D, 2D, 3D convolutions
- Created backward weight GPU reference (naive_conv_bwd_weight.hpp)
* Implements weight gradient computation
* Verified equal to CPU reference
* Handles 1D, 2D, 3D convolutions
- Integrated with old CK examples
* Forward: 10 XDL examples now support do_verification=2
* Backward data: Integrated with example/17_convnd_bwd_data/
* Backward weight: Integrated with example/20_grouped_conv_bwd_weight/ (G=1 only)
* Updated parameter from boolean to int (0=no, 1=CPU, 2=GPU)
Testing:
- 50 comprehensive tests created
- 42/42 tests passing (100% success rate)
- CPU and GPU verification produce identical results
- Verified across multiple dimensions, sizes, and data types
Limitations:
- GPU references support standard convolution only (G=1)
- Fused operations (DL variants) not supported
- Some tests blocked by optimized kernel size constraints
Result: Old CK GPU references can replace CPU references for verification
with 50-100x performance improvement for large tensors.
* Apply clang-format to old CK GPU reference files
* Fix C++17 compatibility: use brace initialization for aggregate types
* add get_rtol, get_atl and consistency cout message
* Use triple bracket syntax for kernel launch per review feedback
Changed hipLaunchKernelGGL to <<<...>>> syntax as suggested by @aosewski.
This is more idiomatic HIP/CUDA style and equally correct.
All tests still passing after this change.
* Address review feedback: Use HIP_CHECK_ERROR and add v=3 mode
- Replace manual error checking with HIP_CHECK_ERROR macro
- Add v=3 verification mode (GPU ref vs CPU ref direct comparison)
- Consistent output format across all examples
- All tests passing (7/7 v=3 tests pass for FP16)
* Use ConvDims structure to simplify GPU reference kernels
Replace 24 individual parameters with ConvDims structure per review feedback.
- Add conv_common.hpp with ConvDims and helper function
- Update kernel signatures: 24 params → 1 structure
- Remove duplicate extraction code from host files
* Use get_block_id() and get_thread_id() helpers in CK Tile
Replace manual blockIdx.x/threadIdx.x arithmetic with helper functions.
Updated 3 CK Tile GPU reference kernels per review feedback.
* Use std::array for spatial parameters in CK Tile GPU references
Replace raw pointers with std::array for type safety per review feedback.
- Add conv_common.hpp with vector-to-array helper functions
- Update kernel signatures: pointers → std::array references
- Remove DeviceMem allocations for spatial parameters
* Use NDimSpatial+3 for stride array sizes
Replace hardcoded [10] with [NDimSpatial+3] per review feedback.
Array sizes now correctly reflect actual dimensions needed.
* Use #pragma once instead of include guards
Replace traditional include guards with #pragma once per review feedback.
Updated 3 Old CK GPU reference headers.
* Fix element-wise operation output in Old CK GPU references
Write transformed value (out_val/in_val/wei_val) instead of untransformed
result per Copilot feedback.
This ensures element-wise operations are correctly applied to output.
* Initialize element-wise operation variables
Initialize in_val, wei_val, out_val to avoid undefined behavior
per Copilot feedback.
Updated backward data and backward weight kernels.
* Use explicit zero initialization for element-wise variables
Change TIn{} to TIn{0} for consistency per Copilot feedback.
All 3 kernels now use consistent zero initialization.
* Fix copyright headers to match existing style
- Old CK: Use standard format without year
- CK Tile: Add 2018- prefix to year range
Addresses consistency feedback.
* Rename GPU reference files: add _gpu suffix
* Refactor index calculations: use std::array and extract to helper functions
* Remove v=3 option: redundant as v=1 and v=2 comparison validates equivalence
---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
[ROCm/composable_kernel commit: 4baa4c9fae]
* Create Description::instance_string() function
To expose more reflection capabilities in MIOpen, we add the instance_string functionality to the ckr::Description class. This PR introduces a base class, adds the instance_string method, and implements the method by injecting the Traits::instance_string method through the ConvDescription constructor.
This will enable us to replace the specialized get_instance_string() method on device operations with a describe() method in a subsequent PR.
* Test describe().instance_string()
Update the instance string tests to also call `ckr::describe<Instance>().instance_string()`. This documents that the xld kernels are supported with describe(), but WMMA and DL kernels are not yet supported. Also update namespace and add a HasConvTraits concept.
[ROCm/composable_kernel commit: f29b67cf9b]
* [CK TILE GEMM] Refactor block_scale_gemm examples
- Split cpp file to reduce building time
- Support multiple GemmConfig
* [CK TILE GEMM] Refactor block_scale_gemm examples
- Update Readme
* feat(gemm_quant): add RRR and CRR layout support for aquant gemm
* test(gemm_quant): add unit tests for RRR and CRR layout support for aquant gemm
* fix: compilation error on gfx950 by omitting support for the gpu in example and unit tests
* fix: test cases compilation failure due to PR# 2095
* fix: make condition to filter out tests for gfx950 more explicit
* need to support the gfx950
* fix: add layout suppot for gfx950
* Extend pk_int4_t support for block_scale_gemm aquant CR and RR layout (#3277)
* WIP: add support for pk_int4_t for aquant mode layouts RR and CR
* test(block_scale_gemm): add unit tests for CRR and RRR layout when data type is int4 && aquant
* fix: compile time error for gfx950
* fix: minor bug where is_a_load_tr_v() was mising
* feat(block_scale_gemm): Add layout Col-Col-Row-Col (ABC-Aquant) for tensors in aquant (#3318)
* feat(block_scale_gemm): Add layout Col-Col-Row-Col (ABC-Aquant) for tensors in aquant
* test: add unit tests for new layout support CCRC for aquant block scale gemm
* docs: update changelog with new layout support info
* Update CHANGELOG.md
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* refactor: break test instances into multiple cpp files to reduce build time (#3319)
* feat(block_scale_gemm): Add layout Col-Col-Row-Col (ABC-Aquant) for tensors in aquant
* test: add unit tests for new layout support CCRC for aquant block scale gemm
* refactor: break test instances into multiple cpp files to reduce build time
* chore: rename file for better code readability
* fix: merge conflict resolution
* fix: remove memory pipeline because new layout is not compatible
* build: resolve build errors for gfx950 by modifying is_a_load_tr() & is_b_load_tr()
* refactor: address review comments
* solve the conflict
---------
Co-authored-by: Cong Ma <congma13@amd.com>
Co-authored-by: ThomasNing <thomas.ning@amd.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
[ROCm/composable_kernel commit: 6cb0bc2d11]
Refactor the builder factory code into multiple files and subdirectories and a ck_tile::builder::factory namespace.
The factory implements compile-time dispatch from high-level signature and algorithm descriptors to our existing specialized convolution kernel implementations.
Major changes in this PR:
Dispatch logic is explicit in the function make_conv_instance instead of implicit in template specialization selection.
Helper code is moved to a subdirectory builder/factory/helpers.
Helpers now have unit tests.
Factories are moved to their own files.
Code moved to namespaces ck_tile::builder::factory and ck_tile::builder::factory::internal.
This does not yet fix the problem of bad error messages, but the make_conv_instance function makes the poor error messages clear. The choice of algorithm must be much more robust (perhaps with explicit enumeration in the algorithm descriptor), so that the dispatch doesn't fail.
Quality changes:
Making dispatch explicit rather than implicit will improve robustness, readability, maintainability, testability, and extensibility.
Separating code into separate files and subdirectories helps readability and extensibility.
Adding unit tests for helpers documents behavior and will enable more complex logic and functionality.
Separating files (especially unit tests) helps clarify includes and dependencies and makes code easier to refactor.
[ROCm/composable_kernel commit: 280bc42191]
* Merge fwd conv groups in CK Tile.
* Fix building CK fwd convs.
* Add number of merged groups to conv fwd kernel name.
* Get number of merged groups from conv config.
* Rename GemmConfig to ConvConfig.
* Clean-up TODOs.
* Check that number of conv groups must be divisible by the number of merged groups.
* Improve error handling in the conv fwd example.
* Fix clang-format.
* Fix group offsets.
* Fix merge problem.
* Address feedback from code review.
* Fix clang-formatting.
[ROCm/composable_kernel commit: 66832861ad]
* Add help for example
* Refactore the compute reference batched contraction to manage stride-aware calculation and some code cleanings
* Add stride-aware reference for batched contraction with independent D tensor layouts
* Add -num_d argument for runtime D tensor count selection in batched contraction
* Add stride vector arguments in example code for testing non-contiguous batched contraction inputs
* Add descriptor-based architecture for batched contraction multi-dimensional stride support
* Add multi-dimensional non-contiguous stride support to batched contraction, num_d = 0
* Add complete multi-dimensional stride support via descriptors
* Enable vectorization in descriptor-based batched contraction. Add pad_tensor_view to local RunGemm
* Clean up batched contraction: remove old UniversalGemmKernel path
* Clean up batched contraction: remove legacy paths and finalize docs
* Optimize batched contraction example: pass dimension sizes not vectors
* correct the reference calculation, unsigned int to int
* Fix batched_contraction C++17 build errors for gfx90a CI
[ROCm/composable_kernel commit: 2d3020e5b0]
