cmake_minimum_required(VERSION 3.15) project(ck_app LANGUAGES CXX) if (DTYPES) add_definitions(-DDTYPES) if (DTYPES MATCHES "int8") add_definitions(-DCK_ENABLE_INT8) if(NOT DEFINED ${CK_ENABLE_INT8}) set(CK_ENABLE_INT8 "ON") endif() endif() if (DTYPES MATCHES "fp8") add_definitions(-DCK_ENABLE_FP8) if(NOT DEFINED ${CK_ENABLE_FP8}) set(CK_ENABLE_FP8 "ON") endif() endif() if (DTYPES MATCHES "fp16") add_definitions(-DCK_ENABLE_FP16) if(NOT DEFINED ${CK_ENABLE_FP16}) set(CK_ENABLE_FP16 "ON") endif() endif() if (DTYPES MATCHES "fp32") add_definitions(-DCK_ENABLE_FP32) if(NOT DEFINED ${CK_ENABLE_FP32}) set(CK_ENABLE_FP32 "ON") endif() endif() if (DTYPES MATCHES "fp64") add_definitions(-DCK_ENABLE_FP64) if(NOT DEFINED ${CK_ENABLE_FP64}) set(CK_ENABLE_FP64 "ON") endif() endif() if (DTYPES MATCHES "bf16") add_definitions(-DCK_ENABLE_BF16) if(NOT DEFINED ${CK_ENABLE_BF16}) set(CK_ENABLE_BF16 "ON") endif() endif() message("DTYPES macro set to ${DTYPES}") else() add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16) if(NOT DEFINED ${CK_ENABLE_ALL_DTYPES}) set(CK_ENABLE_ALL_DTYPES "ON") endif() endif() find_package(composable_kernel COMPONENTS device_other_operations device_gemm_operations device_conv_operations device_contraction_operations device_reduction_operations) find_package(hip REQUIRED PATHS /opt/rocm /opt/rocm/llvm $ENV{HIP_PATH} $ENV{ROCM_PATH}) message(STATUS "Build with HIP ${hip_VERSION}") add_subdirectory(01_gemm) add_subdirectory(02_gemm_add_add_fastgelu) add_subdirectory(03_gemm_layernorm) add_subdirectory(04_contraction) add_subdirectory(05_layernorm) add_subdirectory(06_softmax) add_subdirectory(07_grouped_convnd_fwd) add_subdirectory(08_fused_attention) add_subdirectory(09_quantization) add_subdirectory(10_grouped_convnd_bwd_data) add_subdirectory(11_grouped_conv_bwd_weight) add_subdirectory(12_elementwise_normalization) add_subdirectory(13_batchnorm) add_subdirectory(14_instance_id) add_subdirectory(15_convnd_bwd_data) add_subdirectory(15_gemm_add_multiply) add_subdirectory(15_reduce) add_subdirectory(16_convnd_fwd) add_subdirectory(17_grouped_gemm_fastgelu) add_subdirectory(18_groupnorm) add_subdirectory(19_pool) add_subdirectory(20_splitk_gemm) add_subdirectory(21_grouped_gemm_bias) add_subdirectory(22_grouped_gemm) add_subdirectory(22_im2col_col2im)