mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 12:54:11 +08:00
26a91a9f04
Summary: Modified files in `benchmarks/tensorexpr` to add support for NVIDIA's Fuser for the jit compiler. This support has some modifications besides adding an option to support the NVIDIA fuser: * Adds FP16 Datatype support * Fixes SOL/Algo calculations to generally use the data type instead of being fixed to 4 bytes * Adds IR printing and kernel printing knobs * Adds a knob `input_iter` to create ranges of inputs currently only for reductions * Adds further reduction support for Inner and Outer dimension reductions that are compatible with the `input_iter` knob. * Added `simple_element`, `reduce2d_inner`, and `reduce2d_outer` to isolate performance on elementwise and reduction operations in the most minimal fashion. Pull Request resolved: https://github.com/pytorch/pytorch/pull/44101 Reviewed By: ngimel Differential Revision: D23713658 Pulled By: bertmaher fbshipit-source-id: d6b83cfab559aefe107c23b3c0f2df9923b3adc1
66 lines
1.7 KiB
Python
66 lines
1.7 KiB
Python
from . import benchmark
|
|
|
|
|
|
class PoolingBench(benchmark.Benchmark):
|
|
def __init__(self, case, mode, device, dtype, kernel_size, N, C, H, W):
|
|
super().__init__(mode, device)
|
|
self.case = case
|
|
self.kernel_size = kernel_size
|
|
self.N = N
|
|
self.C = C
|
|
self.H = H
|
|
self.W = W
|
|
self.data = self.rand(
|
|
[N, C, H, W], device=device, dtype=dtype, requires_grad=self.requires_grad
|
|
)
|
|
|
|
def forward(self):
|
|
if self.case == "maxpool":
|
|
y = self.max_pool2d(self.data, self.kernel_size, stride=1)
|
|
elif self.case == "avgpool":
|
|
y = self.avg_pool2d(self.data, self.kernel_size, stride=1)
|
|
return y
|
|
|
|
def config(self):
|
|
return [self.kernel_size, self.N, self.C, self.H, self.W]
|
|
|
|
def memory_workload(self):
|
|
if self.mode == "fwd":
|
|
sol_count = 1 + 1
|
|
algorithmic_count = 1 + 1
|
|
else:
|
|
sol_count = (1 + 1) + (1 + 1)
|
|
algorithmic_count = (1 + 1) + (2 + 1)
|
|
|
|
buffer_size = self.N * self.C * self.H * self.W
|
|
return {
|
|
"sol": buffer_size * sol_count,
|
|
"algorithmic": buffer_size * algorithmic_count,
|
|
}
|
|
|
|
@staticmethod
|
|
def default_configs():
|
|
return [[3, 16, 32, 256, 256]]
|
|
|
|
|
|
class MaxPoolBench(PoolingBench):
|
|
def __init__(self, *args):
|
|
super().__init__("maxpool", *args)
|
|
|
|
@staticmethod
|
|
def module():
|
|
return "maxpool"
|
|
|
|
|
|
class AvgPoolBench(PoolingBench):
|
|
def __init__(self, *args):
|
|
super().__init__("avgpool", *args)
|
|
|
|
@staticmethod
|
|
def module():
|
|
return "avgpool"
|
|
|
|
|
|
benchmark.register_benchmark_class(MaxPoolBench)
|
|
benchmark.register_benchmark_class(AvgPoolBench)
|