[general]
version = "0.0.1"

[torch]
name = "quantization_eetq"
src = [
  "torch-ext/registration.h",
  "torch-ext/torch_binding.cpp",
  "torch-ext/torch_binding.h"
]
pyroot = "torch-ext"

[kernel.cutlass_kernels]
capabilities = [ "7.0", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0" ]
src = [
  "cutlass_extensions/include/cutlass_extensions/arch/mma.h",
  "cutlass_extensions/include/cutlass_extensions/compute_occupancy.h",
  "cutlass_extensions/include/cutlass_extensions/epilogue/epilogue_quant_helper.h",
  "cutlass_extensions/include/cutlass_extensions/epilogue/thread/ft_fused_activations.h",
  "cutlass_extensions/include/cutlass_extensions/epilogue/threadblock/epilogue_per_row_per_col_scale.h",
  "cutlass_extensions/include/cutlass_extensions/epilogue/threadblock/epilogue_tensor_op_int32.h",
  "cutlass_extensions/include/cutlass_extensions/epilogue_helpers.h",
  "cutlass_extensions/include/cutlass_extensions/ft_gemm_configs.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/kernel/default_fpA_intB_traits.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/kernel/fpA_intB_gemm.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/kernel/fpA_intB_gemm_with_broadcast.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/kernel/mixed_gemm_B_layout.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/default_dq_mma.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/default_dq_mma_multistage.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/default_dq_mma_pipelined.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/default_mma.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/default_mma_bf16.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/dq_mma_base.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/dq_mma_multistage.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/threadblock/dq_mma_pipelined.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/warp/default_mma_tensor_op.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/warp/mma_tensorop_compute_B_with_f16.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/warp/mma_tensorop_dequantizer.h",
  "cutlass_extensions/include/cutlass_extensions/interleaved_numeric_conversion.h",
  "cutlass_extensions/include/cutlass_extensions/tile_interleaved_layout.h",
  "cutlass_kernels/cutlass_heuristic.cu",
  "cutlass_kernels/cutlass_heuristic.h",
  "cutlass_kernels/cutlass_preprocessors.cc",
  "cutlass_kernels/cutlass_preprocessors.h",
  "cutlass_kernels/fpA_intB_gemm.cu",
  "cutlass_kernels/fpA_intB_gemm.h",
  "cutlass_kernels/fpA_intB_gemm/fpA_intB_gemm.h",
  "cutlass_kernels/fpA_intB_gemm/fpA_intB_gemm_template.h",
  "cutlass_kernels/fpA_intB_gemm_wrapper.cu",
  "cutlass_kernels/fpA_intB_gemm_wrapper.h",
  "weightOnlyBatchedGemv/common.h",
  "weightOnlyBatchedGemv/enabled.h",
  "utils/activation_types.h",
  "utils/cuda_utils.h",
  "utils/logger.cc",
  "utils/logger.h",
  "utils/string_utils.h",
  "utils/torch_utils.h",
]
depends = [ "cutlass_2_10", "torch" ]
include = [ ".", "utils", "cutlass_extensions/include" ]

[kernel.weight_only_batched_gemv]
capabilities = [ "7.0", "7.5", "8.0", "8.6", "8.7", "8.9", "9.0" ]
src = [
  "cutlass_extensions/include/cutlass_extensions/interleaved_numeric_conversion.h",
  "cutlass_extensions/include/cutlass_extensions/gemm/kernel/mixed_gemm_B_layout.h",
  "weightOnlyBatchedGemv/common.h",
  "weightOnlyBatchedGemv/enabled.h",
  "weightOnlyBatchedGemv/kernel.h",
  "weightOnlyBatchedGemv/kernelLauncher.cu",
  "weightOnlyBatchedGemv/kernelLauncher.h",
  "weightOnlyBatchedGemv/utility.h",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs1Int4b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs1Int8b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs2Int4b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs2Int8b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs3Int4b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs3Int8b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs4Int4b.cu",
  "weightOnlyBatchedGemv/weightOnlyBatchedGemvBs4Int8b.cu",
]
depends = [ "cutlass_2_10", "torch" ]
include = [ "cutlass_extensions/include" ]