WaveCut/orbitquant-packed-matmul

1[general]2name = "orbitquant-packed-matmul"3version = 14edition = 55license = "Apache-2.0"6backends = ["cpu", "cuda", "metal"]7upstream = "https://github.com/iamwavecut/OrbitQuant"8source = "https://huggingface.co/WaveCut/orbitquant-packed-matmul"9 10[general.hub]11repo-id = "WaveCut/orbitquant-packed-matmul"12 13[torch]14src = [15  "torch-ext/torch_binding.cpp",16  "torch-ext/torch_binding.h",17]18 19[torch.stable-abi]20cpu = "2.11"21 22[kernel.packed_matmul_cpu]23backend = "cpu"24depends = ["torch"]25include = ["orbitquant_packed_matmul_cpu"]26src = [27  "orbitquant_packed_matmul_cpu/cpu_isa.cpp",28  "orbitquant_packed_matmul_cpu/cpu_kernel_args.h",29  "orbitquant_packed_matmul_cpu/cpu_pool.cpp",30  "orbitquant_packed_matmul_cpu/cpu_pool.h",31  "orbitquant_packed_matmul_cpu/cpu_threads.cpp",32  "orbitquant_packed_matmul_cpu/cpu_threads.h",33  "orbitquant_packed_matmul_cpu/packed_adaln_cpu.cpp",34  "orbitquant_packed_matmul_cpu/packed_matmul_cpu.cpp",35  "orbitquant_packed_matmul_cpu/packed_matmul_cpu.h",36  "orbitquant_packed_matmul_cpu/packed_matmul_scalar.cpp",37  "orbitquant_packed_matmul_cpu/packed_matmul_neon.cpp",38  "orbitquant_packed_matmul_cpu/packed_matmul_x86_avx512.cpp",39  "orbitquant_packed_matmul_cpu/quantize_activations_cpu.cpp",40]41 42[kernel.packed_matmul_cpu_x86_avx2]43backend = "cpu"44depends = ["torch"]45include = ["orbitquant_packed_matmul_cpu"]46cxx-flags = ["$<$<CXX_COMPILER_ID:MSVC>:/arch:AVX2>"]47src = [48  "orbitquant_packed_matmul_cpu/cpu_msvc_avx2.cpp",49  "orbitquant_packed_matmul_cpu/packed_matmul_x86.cpp",50]51 52[kernel.packed_matmul_cuda]53backend = "cuda"54depends = ["torch"]55src = ["orbitquant_packed_matmul_cuda/packed_matmul.cu"]56 57[kernel.packed_matmul_metal]58backend = "metal"59depends = ["torch"]60src = [61  "orbitquant_packed_matmul_metal/packed_matmul.mm",62  "orbitquant_packed_matmul_metal/packed_matmul.metal",63]64