mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-07-12 02:05:50 +00:00
ad4e2e7624
[CK_TILE] [QuantGEMM] Fix SplitK tail handling and other improvements (#7199) This pull request introduces improved and more robust split-K support for quantized GEMM. The main changes add runtime validation, utility functions for split-K batch calculations, pointer offset handling for split-K in grouped kernels, and enhanced support for various tensor layouts. The changes also improve error handling and provide more flexibility for runtime tail handling in split-K pipelines. **Split-K Support and Validation Enhancements:** * Added runtime validation to ensure `k_batch` is a positive integer and that split-K configurations do not produce empty final batches or mismatched pipeline tails, with detailed error messages and logging for misconfiguration. [[1]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R1184-R1211) [[2]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L1161-R1250) * Introduced utility functions `get_splitk_batch_k_read` and `get_splitk_last_batch_k` to compute per-batch K read sizes and handle split rounding, ensuring correct and consistent split-K batch partitioning. [[1]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R206-R234) [[2]](diffhunk://#diff-635b89bdffa96b2b42f1632520cde36701d7d631e864185591f6b32f7645cf47L104-R107) [[3]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L388-R417) [[4]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L1161-R1250) * Changed the default value of `k_batch` in `QuantGemmHostArgs` to 1 (no split-K) for safer default behavior. **Pointer Offsets and Grouped Kernel Handling:** * Updated `QuantGroupedGemmKernel` to apply split-K per-batch offsets to all input pointers, mirroring the behavior of non-grouped kernels and ensuring correctness for split-K launches. * Modified AQ tensor view handling to correctly reflect the remaining K-groups from the split-K batch's offset position, improving accuracy for split-K in grouped kernels. **Pipeline and Layout Flexibility:** * Added support for runtime selection of split-K tail handling via a new template parameter `RuntimeSplitKTail_`, with new helper methods to dispatch GEMM pipelines accordingly. [[1]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R273) [[2]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R1496-R1567) [[3]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L1427) [[4]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L1447-R1629) [[5]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L1459-R1641) * Improved handling for tensor layout cases, including preshuffled B and both row-major and column-major AQ layouts, ensuring correct pointer arithmetic and compatibility checks. [[1]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R438-R454) [[2]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871L464-R516) [[3]](diffhunk://#diff-d000149a681cd42bfb9947872c603e556cea26cbd7fd4f8f60afc6595d975871R1184-R1211)
CK Tile Testing Guide
This document describes the test organization and available test targets for CK Tile operations.
Overview
CK Tile tests are organized with multiple levels of granularity to support different development workflows:
- Global test labels - Run tests across all operations
- Operation-specific umbrella targets - Run all tests for a specific operation
- Individual test executables - Run specific tests
Global Test Labels
These targets run tests across all CK operations (not just CK Tile):
ninja smoke
Run fast smoke tests (tests that complete within ~30 seconds on gfx90a).
ninja smoke
ninja regression
Run slower, more comprehensive regression tests.
ninja regression
ninja check
Run ALL available tests in the entire codebase.
ninja check
Operation-Specific Umbrella Targets
These targets allow you to run all tests for a specific CK Tile operation. This is useful when making changes to a particular operation and wanting to validate all related tests without running the entire test suite.
GEMM Operations
ck_tile_gemm_tests
Run all basic GEMM pipeline tests (memory, compute variants, persistent, etc.)
ninja ck_tile_gemm_tests
Test executables included:
test_ck_tile_gemm_pipeline_memtest_ck_tile_gemm_pipeline_compv3test_ck_tile_gemm_pipeline_compv4test_ck_tile_gemm_pipeline_persistenttest_ck_tile_gemm_pipeline_compv6test_ck_tile_gemm_pipeline_comp_async(gfx95 only)test_ck_tile_gemm_pipeline_*_wmmavariants (gfx11/gfx12 only)
ck_tile_gemm_block_scale_tests
Run all GEMM tests with block-scale quantization (AQuant, BQuant, ABQuant, etc.)
ninja ck_tile_gemm_block_scale_tests
Test executables included: 29 test executables covering:
- AQuant tests (memory pipelines, base layouts, prefill, preshuffle, transpose)
- ABQuant tests (base, padding, preshuffle)
- BQuant tests (1D/2D variants, transpose)
- BQuant with PreshuffleB (decode/prefill, 1D/2D)
- BQuant with PreshuffleQuant (decode/prefill, 1D/2D)
- RowColQuant and TensorQuant tests
ck_tile_gemm_streamk_tests
Run all GEMM StreamK tests (tile partitioner, reduction, smoke, extended)
ninja ck_tile_gemm_streamk_tests
Test executables included:
test_ck_tile_streamk_tile_partitionertest_ck_tile_streamk_reductiontest_ck_tile_streamk_smoketest_ck_tile_streamk_extended
ck_tile_grouped_gemm_quant_tests
Run all grouped GEMM quantization tests
ninja ck_tile_grouped_gemm_quant_tests
Test executables included:
test_ck_tile_grouped_gemm_quant_rowcoltest_ck_tile_grouped_gemm_quant_tensortest_ck_tile_grouped_gemm_quant_aquanttest_ck_tile_grouped_gemm_quant_bquanttest_ck_tile_grouped_gemm_quant_bquant_preshuffleb
Other Operations
ck_tile_fmha_tests
Run all FMHA (Flash Multi-Head Attention) tests
ninja ck_tile_fmha_tests
Test executables included: Forward and backward tests for fp16, bf16, fp8bf16, fp32
ck_tile_reduce_tests
Run all reduce operation tests
ninja ck_tile_reduce_tests
Test executables included:
test_ck_tile_reduce2dtest_ck_tile_multi_reduce2d_threadwisetest_ck_tile_multi_reduce2d_multiblock
Individual Test Executables
You can also build and run individual test executables:
Build a specific test
ninja test_ck_tile_gemm_pipeline_mem
Run a specific test directly
./build/bin/test_ck_tile_gemm_pipeline_mem
Run a specific test through ctest
ctest -R test_ck_tile_gemm_pipeline_mem --output-on-failure