Files
composable_kernel/test/ck_tile/gemm_streamk/CMakeLists.txt
arai713 715395bc86 [CK_TILE] Stream-K Gemm Example for fp8 and bf8 (#3041)
* Addition of streamk fp8 example for CK Tile

* Adding in bf8 streamk example in CK Tile

* Refactoring fp8/bf8 unit tests

Refactored the unit tests for fp8/bf8 to utilize the test harness.
Implemented smoke tests with layouts: CCR, CRR, RCR, RRR for fp8/bf8.
The tests are using 128x128x32 for the tile configuration, as other
configurations revealed implementation gaps that are currently being
documented.
2025-10-27 19:29:03 -07:00

160 lines
16 KiB
CMake

set(EXAMPLE_GEMM_COMPILE_OPTIONS)
if(CK_USE_OCP_FP8)
list(APPEND EXAMPLE_GEMM_COMPILE_OPTIONS -DCK_TILE_USE_OCP_FP8)
endif()
set(EXAMPLE_GEMM_COMPILE_COMPUTE_V4_OPTIONS)
if(CK_USE_OCP_FP8)
list(APPEND EXAMPLE_GEMM_COMPILE_COMPUTE_V4_OPTIONS -DCK_TILE_USE_OCP_FP8)
endif()
list(APPEND EXAMPLE_GEMM_COMPILE_COMPUTE_V4_OPTIONS
-mllvm
-enable-noalias-to-md-conversion=0
)
set(EXAMPLE_GEMM_COMPILE_COMPUTE_ASYNC_OPTIONS ${EXAMPLE_GEMM_COMPILE_COMPUTE_V4_OPTIONS})
# Currently test_ck_tile_streamk is only built on gfx9
if(GPU_TARGETS MATCHES "gfx9")
include_directories(BEFORE ${CMAKE_CURRENT_SOURCE_DIR})
#TODO: support all arches
#TODO: current c-shuffle only supports C layout as R
add_gtest_executable(test_ck_tile_streamk_smoke
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_rrr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_rrc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_rcr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_rcc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_crr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_crc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_ccr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f16_ccc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_rrr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_rrc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_rcr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_rcc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_crr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_crc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_ccr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
#${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf16_ccc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f8_rrr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f8_rcr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f8_crr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/f8_ccr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf8_rrr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf8_rcr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf8_crr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/bf8_ccr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
)
# TODO: enable extended tests after tolerances for atomic reductions are addressed.
# add_gtest_executable(test_ck_tile_streamk_extended
# # compv3 pipeline
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rrr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rrc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rcr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rcc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_crr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_crc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_ccr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_ccc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rrr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rrc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rcr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_rcc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_crr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_crc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_ccr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/f16_ccc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rrr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rrc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rcr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rcc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_crr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_crc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_ccr_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_ccc_compv3_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rrr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rrc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rcr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_rcc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_crr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_crc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_ccr_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv3/bf16_ccc_compv3_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # TODO: add compv4 pipeline
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rrr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rrc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rcr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rcc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_crr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_crc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_ccr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_ccc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rrr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rrc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rcr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_rcc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_crr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_crc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_ccr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/f16_ccc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rrr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rrc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rcr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rcc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_crr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_crc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_ccr_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_ccc_compv4_256x256x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rrr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rrc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rcr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_rcc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_crr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_crc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_ccr_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/compv4/bf16_ccc_compv4_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# # mem pipeline
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_rrr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_rrc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_rcr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_rcc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_crr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_crc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_ccr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/f16_ccc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_rrr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_rrc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_rcr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_rcc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_crr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_crc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# ${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_ccr_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# #${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/mem/bf16_ccc_mem_128x128x32_2x2x1_32x32x16_NonPersistent.cpp
# )
add_gtest_executable(test_ck_tile_streamk_tile_partitioner test_streamk_tile_partitioner.cpp)
add_gtest_executable(test_ck_tile_streamk_reboot_smoke
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/test_gemm_streamk_reboot_fp16_persistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/test_gemm_streamk_reboot_bf16_persistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/test_gemm_streamk_reboot_fp16_nonpersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/smoke_tests/test_gemm_streamk_reboot_bf16_nonpersistent.cpp
test_gemm_streamk_reboot_util.cpp)
add_gtest_executable(test_ck_tile_streamk_reboot_extended
${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/test_gemm_streamk_reboot_fp16_persistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/test_gemm_streamk_reboot_bf16_persistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/test_gemm_streamk_reboot_fp16_nonpersistent.cpp
${CMAKE_CURRENT_SOURCE_DIR}/extended_tests/test_gemm_streamk_reboot_bf16_nonpersistent.cpp
test_gemm_streamk_reboot_util.cpp)
target_compile_options(test_ck_tile_streamk_smoke PRIVATE ${EXAMPLE_GEMM_COMPILE_OPTIONS})
else()
message(DEBUG "Skipping test_ck_tile_streamk tests for current target")
endif()