NVIDIA GPU自动调度神经网络

对特定设备和工作负载进行自动调整对于获得最佳性能至关重要。这是有关如何使用自动调度器为NVIDIA GPU调整整个神经网络。

为了自动调整神经网络，将网络划分为小的子图，并对其进行独立调整。每个子图被视为一个搜索任务。任务调度程序可以对时间进行分片，并为这些任务动态分配时间资源。任务调度程序可以预测每个任务对端到端执行时间的影响，确定可以最大程度地减少执行时间的任务的优先级。

对于每个子图，使用compute声明tvm/python/topi获取张量表达式形式的计算DAG。使用自动调度器来构造此DAG的搜索空间，并搜索良好的调度（低级优化）。

与依靠手动模板定义搜索空间的基于模板的autotvm不同，自动调度程序不需要任何调度模板。换句话说，自动调度程序仅在其中使用tvm/python/topi计算声明，而不使用现有的调度模板。

本文无法在Windows或最新版本的macOS上运行。要使其运行，需要将本文的内容包装在一个if __name__ == "__main__":块中。

import numpy as np

import tvm

from tvm import relay, auto_scheduler

import tvm.relay.testing

from tvm.contrib import graph_runtime

定义网络

需要使用中继前端API定义网络。可以加载一些预定义的网络tvm.relay.testing。从MXNet，ONNX，PyTorch和TensorFlow加载模型。

对于卷积神经网络，尽管自动调度程序可以在任何布局下正常工作，但发现使用NHWC布局通常可以实现最佳性能。还使用自动调度程序对NHWC布局实施了更多优化。建议将模型转换为NHWC布局以使用自动调度程序。可以使用ConvertLayout传递在TVM中进行布局转换。

def get_network(name, batch_size, layout="NHWC", dtype="float32"):

"""Get the symbol definition and random weight of a network"""

# auto-scheduler prefers NHWC layout

if layout == "NHWC":

image_shape = (224, 224, 3)

elif layout == "NCHW":

image_shape = (3, 224, 224)

else:

raise ValueError("Invalid layout: " + layout)

input_shape = (batch_size,) + image_shape

output_shape = (batch_size, 1000)

if name.startswith("resnet-"):

n_layer = int(name.split("-")[1])

mod, params = relay.testing.resnet.get_workload(

num_layers=n_layer,

batch_size=batch_size,

layout=layout,

dtype=dtype,

image_shape=image_shape,

)

elif name.startswith("resnet3d-"):

n_layer = int(name.split("-")[1])

mod, params = relay.testing.resnet.get_workload(

num_layers=n_layer,

batch_size=batch_size,

layout=layout,

dtype=dtype,

image_shape=image_shape,

)

elif name == "mobilenet":

mod, params = relay.testing.mobilenet.get_workload(

batch_size=batch_size, layout=layout, dtype=dtype, image_shape=image_shape

)

elif name == "squeezenet_v1.1":

assert layout == "NCHW", "squeezenet_v1.1 only supports NCHW layout"

mod, params = relay.testing.squeezenet.get_workload(

version="1.1",

batch_size=batch_size,

dtype=dtype,

image_shape=image_shape,

)

elif name == "inception_v3":

input_shape = (batch_size, 3, 299, 299) if layout == "NCHW" else (batch_size, 299, 299, 3)

mod, params = relay.testing.inception_v3.get_workload(batch_size=batch_size, dtype=dtype)

elif name == "mxnet":

# an example for mxnet model

from mxnet.gluon.model_zoo.vision import get_model

assert layout == "NCHW"

block = get_model("resnet18_v1", pretrained=True)

mod, params = relay.frontend.from_mxnet(block, shape={"data": input_shape}, dtype=dtype)

net = mod["main"]

net = relay.Function(

net.params, relay.nn.softmax(net.body), None, net.type_params, net.attrs

)

mod = tvm.IRModule.from_expr(net)

return mod, params, input_shape, output_shape

# Define the neural network and compilation target

network = "resnet-18"

batch_size = 1

layout = "NHWC"

target = tvm.target.Target("cuda")

dtype = "float32"

log_file = "%s-%s-B%d-%s.json" % (network, layout, batch_size, target.kind.name)

提取搜索任务

接下来，从网络中提取搜索任务及其权重。任务的权重是该任务的子图在整个网络中的出现次数。通过使用权重，可以将网络的端到端延迟近似为，其中sum(latency[t] * weight[t])latency[t]是任务的延迟，weight[t]是任务的权重。任务调度程序将仅优化此目标。

# Extract tasks from the network

print("Extract tasks...")

mod, params, input_shape, output_shape = get_network(network, batch_size, layout, dtype=dtype)

tasks, task_weights = auto_scheduler.extract_tasks(mod["main"], params, target)

for idx, task in enumerate(tasks):

print("========== Task %d (workload key: %s) ==========" % (idx, task.workload_key))

print(task.compute_dag)

输出：

Extract tasks...

========== Task 0 (workload key: ["d7b65649a4dd54becea0a52aabbc5af5", 1, 1000, 1, 1000]) ==========

placeholder = PLACEHOLDER [1, 1000]

T_softmax_maxelem(i0) max= placeholder[i0, k]

T_softmax_exp(i0, i1) = tir.exp((placeholder[i0, i1] - T_softmax_maxelem[i0]))

T_softmax_expsum(i0) += T_softmax_exp[i0, k]

T_softmax_norm(i0, i1) = (T_softmax_exp[i0, i1]/T_softmax_expsum[i0])

========== Task 1 (workload key: ["9847f8cc0b305137f49f2c5c0c8ab25d", 1, 512, 1000, 512, 1000, 1, 1000]) ==========

placeholder = PLACEHOLDER [1, 512]

placeholder = PLACEHOLDER [1000, 512]

T_dense(i, j) += (placeholder[i, k]*placeholder[j, k])

placeholder = PLACEHOLDER [1000]

T_add(ax0, ax1) = (T_dense[ax0, ax1] + placeholder[ax1])

========== Task 2 (workload key: ["69115f188984ae34ede37c3b8ca40b43", 1, 7, 7, 512, 1, 1, 1, 512]) ==========

placeholder = PLACEHOLDER [1, 7, 7, 512]

tensor(ax0, ax1, ax2, ax3) += placeholder[ax0, ((ax1*7) + rv0), ((ax2*7) + rv1), ax3]

tensor(ax0, ax1, ax2, ax3) = (tensor[ax0, ax1, ax2, ax3]/(float32((select((bool)1, ((ax1 + 1)*7), (((ax1 + 1)*7) + 1)) - (ax1*7)))*float32((select((bool)1, ((ax2 + 1)*7), (((ax2 + 1)*7) + 1)) - (ax2*7)))))

========== Task 3 (workload key: ["ad6cecbf5d85cb1cda3c2bb7af170211", 1, 7, 7, 512, 4, 4, 512, 512, 1, 7, 7, 512, 1, 1, 1, 512, 1, 1, 1, 512, 1, 7, 7, 512]) ==========

placeholder = PLACEHOLDER [1, 7, 7, 512]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 8)) && (i2 >= 1)) && (i2 < 8)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 16), ((floormod(floordiv(p, 4), 4)*2) + eps), ((floormod(p, 4)*2) + nu), ci]

B(i, j) = select(((floormod(i, 4) == 3) && (floormod(j, 4) == 3)), 1f, select(((floormod(i, 4) == 3) && (floormod(j, 4) == 2)), ..(OMITTED).. ormod(i, 4) == 0) && (floormod(j, 4) == 1)), 0f, select(((floormod(i, 4) == 0) && (floormod(j, 4) == 0)), 1f, 0f))))))))))))))))

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 512, 512]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

A(i, j) = select(((floormod(i, 4) == 3) && (floormod(j, 2) == 1)), 1f, select(((floormod(i, 4) == 3) && (floormod(j, 2) == 0)), ..(OMITTED).. ct(((floormod(i, 4) == 0) && (floormod(j, 2) == 1)), 0f, select(((floormod(i, 4) == 0) && (floormod(j, 2) == 0)), 1f, 0f))))))))

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*4)*4) + (floordiv(h, 2)*4)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 7, 7, 512]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

placeholder = PLACEHOLDER [1, 1, 1, 512]

T_multiply(ax0, ax1, ax2, ax3) = (T_add[ax0, ax1, ax2, ax3]*placeholder[ax0, 0, 0, ax3])

placeholder = PLACEHOLDER [1, 1, 1, 512]

T_add(ax0, ax1, ax2, ax3) = (T_multiply[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 4 (workload key: ["3a69f9fbc63760d99e36b4c17b3bfc57", 1, 7, 7, 512, 4, 4, 512, 512, 1, 1, 1, 512, 1, 7, 7, 512]) ==========

placeholder = PLACEHOLDER [1, 7, 7, 512]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 8)) && (i2 >= 1)) && (i2 < 8)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 16), ((floormod(floordiv(p, 4), 4)*2) + eps), ((floormod(p, 4)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 512, 512]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*4)*4) + (floordiv(h, 2)*4)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 1, 1, 512]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 5 (workload key: ["d730bcd28f0920f6b97245e2a11bd8d6", 1, 7, 7, 512, 4, 4, 512, 512, 1, 7, 7, 512, 1, 7, 7, 512]) ==========

placeholder = PLACEHOLDER [1, 7, 7, 512]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 8)) && (i2 >= 1)) && (i2 < 8)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 16), ((floormod(floordiv(p, 4), 4)*2) + eps), ((floormod(p, 4)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 512, 512]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*4)*4) + (floordiv(h, 2)*4)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 7, 7, 512]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

========== Task 6 (workload key: ["12b88bedece6984af589a28b43e0f3c4", 1, 14, 14, 256, 3, 3, 256, 512, 1, 1, 1, 512, 1, 7, 7, 512]) ==========

placeholder = PLACEHOLDER [1, 14, 14, 256]

PaddedInput(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 15)) && (i2 >= 1)) && (i2 < 15)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

placeholder = PLACEHOLDER [3, 3, 256, 512]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

placeholder = PLACEHOLDER [1, 1, 1, 512]

T_add(ax0, ax1, ax2, ax3) = (Conv2dOutput[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 7 (workload key: ["f3b6c10fcc6ce01ff01add933e4d21e9", 1, 14, 14, 256, 4, 4, 256, 256, 1, 14, 14, 256, 1, 1, 1, 256, 1, 14, 14, 256]) ==========

placeholder = PLACEHOLDER [1, 14, 14, 256]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 15)) && (i2 >= 1)) && (i2 < 15)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 49), ((floormod(floordiv(p, 7), 7)*2) + eps), ((floormod(p, 7)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 256, 256]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*7)*7) + (floordiv(h, 2)*7)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 14, 14, 256]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

placeholder = PLACEHOLDER [1, 1, 1, 256]

T_add(ax0, ax1, ax2, ax3) = (T_add[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 8 (workload key: ["b8b52b9be9df6102466a22a014c44c1f", 1, 14, 14, 256, 4, 4, 256, 256, 1, 1, 1, 256, 1, 14, 14, 256]) ==========

placeholder = PLACEHOLDER [1, 14, 14, 256]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 15)) && (i2 >= 1)) && (i2 < 15)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 49), ((floormod(floordiv(p, 7), 7)*2) + eps), ((floormod(p, 7)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 256, 256]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*7)*7) + (floordiv(h, 2)*7)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 1, 1, 256]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 9 (workload key: ["d374e472bd9d8164892b9e28a0a8cb59", 1, 14, 14, 256, 4, 4, 256, 256, 1, 14, 14, 256, 1, 14, 14, 256]) ==========

placeholder = PLACEHOLDER [1, 14, 14, 256]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 15)) && (i2 >= 1)) && (i2 < 15)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 49), ((floormod(floordiv(p, 7), 7)*2) + eps), ((floormod(p, 7)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 256, 256]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*7)*7) + (floordiv(h, 2)*7)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 14, 14, 256]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

========== Task 10 (workload key: ["12b88bedece6984af589a28b43e0f3c4", 1, 28, 28, 128, 3, 3, 128, 256, 1, 1, 1, 256, 1, 14, 14, 256]) ==========

placeholder = PLACEHOLDER [1, 28, 28, 128]

PaddedInput(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 29)) && (i2 >= 1)) && (i2 < 29)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

placeholder = PLACEHOLDER [3, 3, 128, 256]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

placeholder = PLACEHOLDER [1, 1, 1, 256]

T_add(ax0, ax1, ax2, ax3) = (Conv2dOutput[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 11 (workload key: ["c4500b4e2fd04e695c32d2f31bbdc14a", 1, 28, 28, 128, 4, 4, 128, 128, 1, 28, 28, 128, 1, 1, 1, 128, 1, 28, 28, 128]) ==========

placeholder = PLACEHOLDER [1, 28, 28, 128]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 29)) && (i2 >= 1)) && (i2 < 29)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*2) + eps), ((floormod(p, 14)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 128, 128]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*14)*14) + (floordiv(h, 2)*14)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 28, 28, 128]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

placeholder = PLACEHOLDER [1, 1, 1, 128]

T_add(ax0, ax1, ax2, ax3) = (T_add[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 12 (workload key: ["e4cdf917b876dbdd64488c3818d9c141", 1, 28, 28, 128, 4, 4, 128, 128, 1, 1, 1, 128, 1, 28, 28, 128]) ==========

placeholder = PLACEHOLDER [1, 28, 28, 128]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 29)) && (i2 >= 1)) && (i2 < 29)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*2) + eps), ((floormod(p, 14)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 128, 128]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*14)*14) + (floordiv(h, 2)*14)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 1, 1, 128]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 13 (workload key: ["dac19035dd5fe9424ee8617421b9c817", 1, 28, 28, 128, 4, 4, 128, 128, 1, 28, 28, 128, 1, 28, 28, 128]) ==========

placeholder = PLACEHOLDER [1, 28, 28, 128]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 29)) && (i2 >= 1)) && (i2 < 29)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*2) + eps), ((floormod(p, 14)*2) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [4, 4, 128, 128]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 2), floormod(w, 2), ((((n*14)*14) + (floordiv(h, 2)*14)) + floordiv(w, 2)), co]

placeholder = PLACEHOLDER [1, 28, 28, 128]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

========== Task 14 (workload key: ["12b88bedece6984af589a28b43e0f3c4", 1, 56, 56, 64, 3, 3, 64, 128, 1, 1, 1, 128, 1, 28, 28, 128]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

PaddedInput(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 57)) && (i2 >= 1)) && (i2 < 57)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

placeholder = PLACEHOLDER [3, 3, 64, 128]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

placeholder = PLACEHOLDER [1, 1, 1, 128]

T_add(ax0, ax1, ax2, ax3) = (Conv2dOutput[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 15 (workload key: ["1e3c4211ffd2f2db91078ae4d04b779d", 1, 56, 56, 64, 6, 6, 64, 64, 1, 56, 56, 64, 1, 1, 1, 64, 1, 56, 56, 64]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 57)) && (i2 >= 1)) && (i2 < 57)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*4) + eps), ((floormod(p, 14)*4) + nu), ci]

B(i, j) = select(((floormod(i, 6) == 5) && (floormod(j, 6) == 5)), 1f, select(((floormod(i, 6) == 5) && (floormod(j, 6) == 4)), ..(OMITTED).. (floormod(j, 6) == 1)), 0f, select(((floormod(i, 6) == 0) && (floormod(j, 6) == 0)), 1f, 0f))))))))))))))))))))))))))))))))))))

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [6, 6, 64, 64]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

A(i, j) = select(((floormod(i, 6) == 5) && (floormod(j, 4) == 3)), 1f, select(((floormod(i, 6) == 5) && (floormod(j, 4) == 2)), ..(OMITTED).. 6) == 0) && (floormod(j, 4) == 1)), 0f, select(((floormod(i, 6) == 0) && (floormod(j, 4) == 0)), 1f, 0f))))))))))))))))))))))))

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 4), floormod(w, 4), ((((n*14)*14) + (floordiv(h, 4)*14)) + floordiv(w, 4)), co]

placeholder = PLACEHOLDER [1, 56, 56, 64]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

placeholder = PLACEHOLDER [1, 1, 1, 64]

T_add(ax0, ax1, ax2, ax3) = (T_add[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 16 (workload key: ["b818b53148cd450f86569dfc3e04cb8a", 1, 56, 56, 64, 6, 6, 64, 64, 1, 1, 1, 64, 1, 56, 56, 64]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 57)) && (i2 >= 1)) && (i2 < 57)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*4) + eps), ((floormod(p, 14)*4) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [6, 6, 64, 64]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 4), floormod(w, 4), ((((n*14)*14) + (floordiv(h, 4)*14)) + floordiv(w, 4)), co]

placeholder = PLACEHOLDER [1, 1, 1, 64]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 17 (workload key: ["3ea73fb9b0364374730d09e068821f95", 1, 56, 56, 64, 6, 6, 64, 64, 1, 56, 56, 64, 1, 56, 56, 64]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

data_pad(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 1) && (i1 < 57)) && (i2 >= 1)) && (i2 < 57)), placeholder[i0, (i1 - 1), (i2 - 1), i3], 0f)

input_tile(eps, nu, p, ci) = data_pad[floordiv(p, 196), ((floormod(floordiv(p, 14), 14)*4) + eps), ((floormod(p, 14)*4) + nu), ci]

data_pack(eps, nu, p, ci) += ((input_tile[r_a, r_b, p, ci]*B[r_a, eps])*B[r_b, nu])

placeholder = PLACEHOLDER [6, 6, 64, 64]

bgemm(eps, nu, p, co) += (data_pack[eps, nu, p, ci]*placeholder[eps, nu, co, ci])

inverse(vh, vw, p, co) += ((bgemm[r_a, r_b, p, co]*A[r_a, vh])*A[r_b, vw])

conv2d_winograd(n, h, w, co) = inverse[floormod(h, 4), floormod(w, 4), ((((n*14)*14) + (floordiv(h, 4)*14)) + floordiv(w, 4)), co]

placeholder = PLACEHOLDER [1, 56, 56, 64]

T_add(ax0, ax1, ax2, ax3) = (conv2d_winograd[ax0, ax1, ax2, ax3] + placeholder[ax0, ax1, ax2, ax3])

========== Task 18 (workload key: ["a5612fdeb9db4d579a75ec225ea4c06a", 1, 112, 112, 64, 1, 1, 1, 64, 1, 56, 56, 64]) ==========

placeholder = PLACEHOLDER [1, 112, 112, 64]

pad_temp(ax0, ax1, ax2, ax3) = tir.if_then_else(((((ax1 >= 1) && (ax1 < 113)) && (ax2 >= 1)) && (ax2 < 113)), placeholder[ax0, (ax1 - 1), (ax2 - 1), ax3], -3.40282e+38f)

tensor(ax0, ax1, ax2, ax3) max= pad_temp[ax0, ((ax1*2) + dh), ((ax2*2) + dw), ax3]

placeholder = PLACEHOLDER [1, 1, 1, 64]

T_add(ax0, ax1, ax2, ax3) = (tensor[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 19 (workload key: ["12b88bedece6984af589a28b43e0f3c4", 1, 224, 224, 3, 7, 7, 3, 64, 1, 1, 1, 64, 1, 112, 112, 64]) ==========

placeholder = PLACEHOLDER [1, 224, 224, 3]

PaddedInput(i0, i1, i2, i3) = tir.if_then_else(((((i1 >= 3) && (i1 < 227)) && (i2 >= 3)) && (i2 < 227)), placeholder[i0, (i1 - 3), (i2 - 3), i3], 0f)

placeholder = PLACEHOLDER [7, 7, 3, 64]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

placeholder = PLACEHOLDER [1, 1, 1, 64]

T_add(ax0, ax1, ax2, ax3) = (Conv2dOutput[ax0, ax1, ax2, ax3] + placeholder[ax0, 0, 0, ax3])

T_relu(ax0, ax1, ax2, ax3) = max(T_add[ax0, ax1, ax2, ax3], 0f)

========== Task 20 (workload key: ["7006235cfc29b73be524cf390ed5a977", 1, 56, 56, 64, 1, 1, 64, 64, 1, 56, 56, 64]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

PaddedInput(i0, i1, i2, i3) = placeholder[i0, i1, i2, i3]

placeholder = PLACEHOLDER [1, 1, 64, 64]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, (yy + ry), (xx + rx), rc]*placeholder[ry, rx, rc, ff])

========== Task 21 (workload key: ["f4380bb1dc62422a69ad4a1a9771f927", 1, 56, 56, 64, 1, 1, 64, 128, 1, 28, 28, 128]) ==========

placeholder = PLACEHOLDER [1, 56, 56, 64]

PaddedInput(i0, i1, i2, i3) = placeholder[i0, i1, i2, i3]

placeholder = PLACEHOLDER [1, 1, 64, 128]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

========== Task 22 (workload key: ["f4380bb1dc62422a69ad4a1a9771f927", 1, 28, 28, 128, 1, 1, 128, 256, 1, 14, 14, 256]) ==========

placeholder = PLACEHOLDER [1, 28, 28, 128]

PaddedInput(i0, i1, i2, i3) = placeholder[i0, i1, i2, i3]

placeholder = PLACEHOLDER [1, 1, 128, 256]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

========== Task 23 (workload key: ["f4380bb1dc62422a69ad4a1a9771f927", 1, 14, 14, 256, 1, 1, 256, 512, 1, 7, 7, 512]) ==========

placeholder = PLACEHOLDER [1, 14, 14, 256]

PaddedInput(i0, i1, i2, i3) = placeholder[i0, i1, i2, i3]

placeholder = PLACEHOLDER [1, 1, 256, 512]

Conv2dOutput(nn, yy, xx, ff) += (PaddedInput[nn, ((yy*2) + ry), ((xx*2) + rx), rc]*placeholder[ry, rx, rc, ff])

开始调整

设置一些选项来优化和启动搜索任务

measure_ctx启动不同的测量过程以提供隔离。保护主进程免受测量期间GPU崩溃的影响，避免其它运行时冲突。
min_repeat_ms定义每次测量中一次“重复”的最小持续时间。这样可以预热GPU，对于获得准确的测量结果是必不可少的。通常，建议值> = 300毫秒。
num_measure_trials是在调整期间可以使用的测量试验的次数。可以将其设置为较小的数字（例如200），进行快速演示。实际上，建议将其设置为900 * len(tasks)，使搜索收敛。例如，resnet-18中有24个任务，将其设置为20000。根据时间预算调整此参数。
将测量记录转储到日志文件RecordToFile中，这些测量记录可用于最好地查询历史记录，恢复搜索，进行更多分析。
有关更多参数auto_scheduler.TuningOptions，请参见auto_scheduler.LocalRPCMeasureContext。

def run_tuning():

print("Begin tuning...")

measure_ctx = auto_scheduler.LocalRPCMeasureContext(repeat=1, min_repeat_ms=300, timeout=10)

tuner = auto_scheduler.TaskScheduler(tasks, task_weights)

tune_option = auto_scheduler.TuningOptions(

num_measure_trials=200, # change this to 20000 to achieve the best performance

runner=measure_ctx.runner,

measure_callbacks=[auto_scheduler.RecordToFile(log_file)],

)

tuner.tune(tune_option)

# We do not run the tuning in our webpage server since it takes too long.

# Uncomment the following line to run it by yourself.

# run_tuning()

笔记

调整期间说明打印的信息

在调整期间，控制台上会打印很多信息。用于调试目的。最重要的信息是任务调度程序的输出。下表是示例输出。

----------------------------------------------------------------------

------------------------------ [ Task Scheduler ]

----------------------------------------------------------------------

-------------------------------------------------

| 0 | 0.005 | 0.88 | 64 |

| 1 | 0.010 | 99.10 | 64 |

| 2 | 0.006 | 0.00 | 64 |

| 3 | 0.145 | 979.78 | 384 |

| 4 | 0.130 | 1097.02 | 384 |

| 5 | 0.143 | 992.69 | 384 |

| 6 | 0.076 | 1526.86 | 192 |

| 7 | 0.115 | 999.44 | 320 |

| 8 | 0.079 | 1449.39 | 320 |

| 9 | 0.122 | 938.73 | 384 |

| 10 | 0.063 | 1832.98 | 192 |

| 11 | 0.072 | 1763.62 | 256 |

| 12 | 0.062 | 2036.40 | 192 |

| 13 | 0.068 | 1874.44 | 192 |

| 14 | 0.049 | 2346.50 | 128 |

| 15 | 0.076 | 1694.31 | 256 |

| 16 | 0.067 | 1933.30 | 448 |

| 17 | 0.076 | 1680.90 | 256 |

| 18 | 0.022 | 98.43 | 64 |

| 19 | 0.076 | 3112.55 | 192 |

| 20 | 0.013 | 2026.44 | 64 |

| 21 | 0.011 | 1136.69 | 64 |

| 22 | 0.013 | 992.47 | 64 |

| 23 | 0.020 | 627.56 | 64 |

-------------------------------------------------

Estimated total latency: 1.587 ms Trials: 4992 Used time : 13296 s Next ID: 3

下表列出了所有任务的延迟和（估计）速度。列出了所有任务的测量试验分配。最后一行显示这些任务的总加权延迟，可以粗略估计网络的端到端执行时间。最后一行还显示测量试验的总数，自动调整所花费的总时间，要调整的下一个任务的ID。

自动调度程序将尝试某些无效的调度，出现一些“ dmlc :: Error”和CUDA错误。继续进行调整，放心地忽略，这些错误与主要过程是隔离的。

笔记

提前终止调整

可以通过强制终止此过程来提前终止调整。在日志文件中为每个任务至少获得一个有效的调度，能够进行编译（下面的部分）。

编译和评估

自动调整后，可以使用发现的最佳调度表来编译网络。在自动调整期间，所有测量记录都将转储到日志文件中，读取日志文件并加载最佳调度。

# Compile with the history best

print("Compile...")

with auto_scheduler.ApplyHistoryBest(log_file):

with tvm.transform.PassContext(opt_level=3, config={"relay.backend.use_auto_scheduler": True}):

lib = relay.build(mod, target=target, params=params)

# Create graph runtime

ctx = tvm.context(str(target), 0)

module = graph_runtime.GraphModule(lib["default"](ctx))

data_tvm = tvm.nd.array((np.random.uniform(size=input_shape)).astype(dtype))

module.set_input("data", data_tvm)

# Evaluate

print("Evaluate inference time cost...")

ftimer = module.module.time_evaluator("run", ctx, repeat=3, min_repeat_ms=500)

prof_res = np.array(ftimer().results) * 1e3 # convert to millisecond

print("Mean inference time (std dev): %.2f ms (%.2f ms)" % (np.mean(prof_res), np.std(prof_res)))

输出：

Compile...

Evaluate inference time cost...

Mean inference time (std dev): 3.22 ms (0.02 ms)

其它技巧

调整过程中，自动调度器需要编译许多程序并从中提取功能。该部分占用大量CPU，建议使用具有多个内核的高性能CPU，加快搜索速度。
提取大型日志文件，仅保存最有用的记录。python3 -m tvm.auto_scheduler.measure_record --mode distill -i log.json
从上一个日志文件继续搜索。load_log_file在function中创建任务调度程序时，只需添加一个新参数run_tuning。tuner = auto_scheduler.TaskScheduler(tasks, task_weights, load_log_file=log_file)
如果有多个目标GPU，全部用于测量，并行化测量。了解如何使用RPC跟踪器和RPC服务器。要在自动调度使用RPC跟踪，用auto_scheduler.RPCRunner，更换转轮TuningOptions 。

posted @ 2021-03-14 20:52 吴建明wujianming 阅读(542) 评论(0) 编辑收藏举报

刷新页面返回顶部

吴建明

NVIDIA GPU自动调度神经网络

公告