tilelang.carver.arch.cuda ========================= .. py:module:: tilelang.carver.arch.cuda Attributes ---------- .. autoapisummary:: tilelang.carver.arch.cuda.volta_tensorcore_supported tilelang.carver.arch.cuda.ampere_tensorcore_supported tilelang.carver.arch.cuda.ada_tensorcore_supported tilelang.carver.arch.cuda.hopper_tensorcore_supported Classes ------- .. autoapisummary:: tilelang.carver.arch.cuda.TensorInstruction tilelang.carver.arch.cuda.CUDA Functions --------- .. autoapisummary:: tilelang.carver.arch.cuda.check_sm_version tilelang.carver.arch.cuda.is_cuda_arch tilelang.carver.arch.cuda.is_volta_arch tilelang.carver.arch.cuda.is_ampere_arch tilelang.carver.arch.cuda.is_ada_arch tilelang.carver.arch.cuda.is_hopper_arch tilelang.carver.arch.cuda.has_mma_support tilelang.carver.arch.cuda.is_tensorcore_supported_precision Module Contents --------------- .. py:function:: check_sm_version(arch) .. py:function:: is_cuda_arch(arch) .. py:function:: is_volta_arch(arch) .. py:function:: is_ampere_arch(arch) .. py:function:: is_ada_arch(arch) .. py:function:: is_hopper_arch(arch) .. py:function:: has_mma_support(arch) .. py:data:: volta_tensorcore_supported :value: [('float16', 'float32'), ('float16', 'float16')] .. py:data:: ampere_tensorcore_supported :value: [('bfloat16', 'float32'), ('float16', 'float32'), ('float16', 'float16'), ('int8', 'int32'),... .. py:data:: ada_tensorcore_supported :value: [('bfloat16', 'float32'), ('float16', 'float32'), ('float16', 'float16'), ('int8', 'int32'),... .. py:data:: hopper_tensorcore_supported :value: [('bfloat16', 'float32'), ('float16', 'float32'), ('float16', 'float16'), ('int8', 'int32'),... .. py:function:: is_tensorcore_supported_precision(in_dtype, accum_dtype, arch) .. py:class:: TensorInstruction(name, shape) Bases: :py:obj:`object` .. py:attribute:: name :type: str .. py:attribute:: shape :type: List[int] .. py:class:: CUDA(target) Bases: :py:obj:`tilelang.carver.arch.arch_base.TileDevice` Represents the architecture of a computing device, capturing various hardware specifications. .. py:attribute:: target .. py:attribute:: sm_version .. py:attribute:: name .. py:attribute:: device :type: tvm.runtime.Device .. py:attribute:: platform :type: str :value: 'CUDA' .. py:attribute:: smem_cap .. py:attribute:: compute_max_core .. py:attribute:: warp_size .. py:attribute:: compute_capability .. py:attribute:: reg_cap :type: int :value: 65536 .. py:attribute:: max_smem_usage :type: int .. py:attribute:: sm_partition :type: int :value: 4 .. py:attribute:: l2_cache_size_bytes :type: int .. py:attribute:: transaction_size :type: List[int] :value: [32, 128] .. py:attribute:: bandwidth :type: List[int] :value: [750, 12080] .. py:attribute:: available_tensor_instructions :type: List[TensorInstruction] :value: None .. py:method:: get_avaliable_tensorintrin_shapes() .. py:method:: __repr__()