fused_lamb.py 968 Bytes
Newer Older
Samyam Rajbhandari's avatar
Samyam Rajbhandari committed
1
2
3
"""
Copyright 2020 The Microsoft DeepSpeed Team
"""
4
5
6
7
8
9
10
11
12
13
14
15
16
17
import torch
from .builder import CUDAOpBuilder


class FusedLambBuilder(CUDAOpBuilder):
    BUILD_VAR = 'DS_BUILD_FUSED_LAMB'
    NAME = "fused_lamb"

    def __init__(self):
        super().__init__(name=self.NAME)

    def absolute_name(self):
        return f'deepspeed.ops.lamb.{self.NAME}_op'

401qingkong's avatar
401qingkong committed
18
19
20
21
22
    def sources(self, is_rocm_pytorch):
        if is_rocm_pytorch:
            return ['csrc/lamb/hip/fused_lamb_hip.cpp', 'csrc/lamb/hip/fused_lamb_hip_kernel.hip']
        else:
            return ['csrc/lamb/fused_lamb_cuda.cpp', 'csrc/lamb/fused_lamb_cuda_kernel.cu']
23
24
25
26
27
28
29
    def include_paths(self):
        return ['csrc/includes']

    def cxx_args(self):
        return ['-O3'] + self.version_dependent_macros()

    def nvcc_args(self):
30
31
        return ['-lineinfo',
                '-O3',
401qingkong's avatar
401qingkong committed
32
33
                #'--use_fast_math'
                ]# + self.version_dependent_macros() + self.compute_capability_args()