mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-03 13:11:25 +00:00
1a4aa7fd89
ck_tile: add gtest unit tests for MX flatmm (gfx950)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
## Summary
- Add correctness unit tests for the MX-format flatmm kernel
(`example/ck_tile/18_flatmm/mxgemm`) under `test/ck_tile/flatmm/`
- Tests cover all five dtype combinations: FP4×FP4, FP8×FP8, FP6×FP6,
FP8×FP4, FP4×FP8
- Tests cover all four kernel dispatch paths (the `has_hot_loop` ×
`tail_num` product):
- `has_hot_loop=false, tail=ODD` (K=256, num_loop=1)
- `has_hot_loop=false, tail=EVEN` (K=512, num_loop=2)
- `has_hot_loop=true, tail=ODD` (K=768, num_loop=3)
- `has_hot_loop=true, tail=EVEN` (K=1024, num_loop=4)
- Remove unsupported `-split_k` CLI option from
`tile_example_mx_flatmm`; the pre-shuffled B layout is incompatible with
K-splitting and the option silently produced wrong results
## Changes
**New files (`test/ck_tile/flatmm/`):**
- `CMakeLists.txt` — builds 40 kernel instances as a shared OBJECT
library, links into 5 per-dtype test executables; forwards
`-DCK_TILE_USE_OCP_FP8` when `CK_USE_OCP_FP8` is ON
- `test_mx_flatmm_base.hpp` — base test fixture with
`run_test_with_validation(M, N, K, kbatch=1)`
- `test_mx_flatmm_fixtures.hpp` — concrete `TestMXFlatmm` typed test
class and type aliases
- `test_mx_flatmm_fp{4fp4,8fp8,6fp6,8fp4,4fp8}.cpp` — per-dtype
`TYPED_TEST_SUITE` files
**Modified files:**
- `example/ck_tile/18_flatmm/mxgemm/mx_flatmm_arch_traits.hpp` — moved
`preShuffleWeight` here (was in `mx_flatmm.cpp`) so it is includeable by
both the example and the tests
- `example/ck_tile/18_flatmm/mxgemm/mx_flatmm.cpp` / `run_mx_flatmm.inc`
— removed `-split_k` CLI arg, hardcoded `k_batch=1`, fixed `k_split`
formula, updated call sites after `preShuffleWeight` move
- `test/ck_tile/CMakeLists.txt` — added `add_subdirectory(flatmm)`
73 lines
2.9 KiB
CMake
73 lines
2.9 KiB
CMake
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
|
# SPDX-License-Identifier: MIT
|
|
|
|
################################################################################
|
|
# CK Tile Test Organization
|
|
################################################################################
|
|
# CK Tile tests can be run using several methods:
|
|
#
|
|
# 1. Global test labels (run tests across all operations):
|
|
# - ninja smoke - Fast tests (~30s on gfx90a)
|
|
# - ninja regression - Slower comprehensive tests
|
|
# - ninja check - All available tests
|
|
#
|
|
# 2. Operation-specific umbrella targets (run all tests for a specific operation):
|
|
# - ninja ck_tile_gemm_tests - All basic GEMM tests
|
|
# - ninja ck_tile_gemm_block_scale_tests - All GEMM with block-scale quantization tests
|
|
# - ninja ck_tile_gemm_streamk_tests - All GEMM StreamK tests
|
|
# - ninja ck_tile_grouped_gemm_quant_tests - All grouped GEMM quantization tests
|
|
# - ninja ck_tile_reduce_tests - All reduce operation tests
|
|
# - ninja ck_tile_fmha_tests - All FMHA (Flash Attention) tests
|
|
#
|
|
# 3. Individual test executables:
|
|
# - ninja test_<test_name> - Build specific test executable
|
|
# - ./build/bin/test_<test_name> - Run specific test directly
|
|
#
|
|
# These umbrella targets are useful when working on specific operations to quickly
|
|
# validate all related tests without running the entire test suite.
|
|
################################################################################
|
|
|
|
add_subdirectory(image_to_column)
|
|
add_subdirectory(gemm)
|
|
add_subdirectory(gemm_persistent_async_input)
|
|
add_subdirectory(gemm_weight_preshuffle)
|
|
add_subdirectory(batched_gemm)
|
|
add_subdirectory(grouped_gemm)
|
|
add_subdirectory(grouped_gemm_preshuffle)
|
|
add_subdirectory(grouped_gemm_multi_d)
|
|
add_subdirectory(grouped_gemm_quant)
|
|
add_subdirectory(grouped_gemm_abquant)
|
|
add_subdirectory(gemm_multi_d)
|
|
add_subdirectory(gemm_multi_abd)
|
|
add_subdirectory(gemm_streamk)
|
|
add_subdirectory(data_type)
|
|
add_subdirectory(container)
|
|
add_subdirectory(elementwise)
|
|
# Not including these tests as there is a bug on gfx90a and gfx942
|
|
# resulting in "GPU core dump"
|
|
#add_subdirectory(moe_smoothquant)
|
|
add_subdirectory(permute)
|
|
add_subdirectory(moe_sorting)
|
|
add_subdirectory(slice_tile)
|
|
add_subdirectory(memory_copy)
|
|
add_subdirectory(batched_transpose)
|
|
add_subdirectory(smoothquant)
|
|
add_subdirectory(topk_softmax)
|
|
add_subdirectory(add_rmsnorm2d_rdquant)
|
|
# add_subdirectory(layernorm2d)
|
|
# add_subdirectory(rmsnorm2d)
|
|
add_subdirectory(gemm_block_scale)
|
|
add_subdirectory(flatmm)
|
|
add_subdirectory(gemm_mx)
|
|
add_subdirectory(utility)
|
|
add_subdirectory(warp_gemm)
|
|
add_subdirectory(reduce)
|
|
add_subdirectory(core)
|
|
add_subdirectory(epilogue)
|
|
add_subdirectory(atomic_add_op)
|
|
add_subdirectory(fmha)
|
|
add_subdirectory(gemm_tile_engine)
|
|
add_subdirectory(pooling)
|
|
add_subdirectory(grouped_conv)
|
|
add_subdirectory(gemm_streamk_tile_engine)
|