[general]
version = "0.0.1"

[torch]
name = "moe"
src = [
  "core/scalar_type.hpp",
  "torch-ext/registration.h",
  "torch-ext/torch_binding.cpp",
  "torch-ext/torch_binding.h",
]
include = ["."]
pyroot = "torch-ext"
pyext = ["py", "json"]

[kernel.fp8]
capabilities = ["7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0"]
src = [
  "cuda_compat.h",
  "dispatch_utils.h",
  "fp8/amd/hip_float8.h",
  "fp8/amd/hip_float8_impl.h",
  "fp8/common.cu",
  "fp8/common.cuh",
  "fp8/vectorization.cuh",
]
include = ["."]
depends = ["torch"]


[kernel.moe]
capabilities = ["7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0"]
src = [
  "cuda_compat.h",
  "dispatch_utils.h",
  "moe/moe_align_sum_kernels.cu",
  "moe/topk_softmax_kernels.cu",
]
depends = ["torch"]

[kernel.moe-marlin]
capabilities = ["8.0", "8.6", "8.7", "8.9", "9.0"]
src = [
  "core/exception.hpp",
  "core/scalar_type.hpp",
  "marlin-moe/marlin_moe_ops.cu",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku4.cu",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku8b128.cu",
  "marlin-moe/marlin_kernels/marlin_moe_kernel.h",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku4.h",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku4b8.h",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku4b8.cu",
  "marlin-moe/marlin_kernels/marlin_moe_kernel_ku8b128.h",
]
include = ["."]
depends = ["torch"]

[kernel.activation]
capabilities = ["7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0"]
src = [
  "activation/activation_kernels.cu",
  "activation/cuda_compat.h",
  "activation/dispatch_utils.h",
]
depends = ["torch"]