Auto-scheduling a Convolution Layer for GPU

Author: Lianmin Zheng, Chengfan Jia

This is a tutorial on how to use the auto-scheduler for GPUs.

Different from the template-based autotvm which relies on manual templates to define the search space, the auto-scheduler does not require any templates. Users only need to write the computation declaration without any schedule commands or templates. The auto-scheduler can automatically generate a large search space and find a good schedule in the space.

We use a convolution layer as an example in this tutorial.

Note that this tutorial will not run on Windows or recent versions of macOS. To get it to run, you will need to wrap the body of this tutorial in a if __name__ == "__main__": block.

import os

import numpy as np
import tvm
from tvm import te, auto_scheduler, topi
from tvm.topi.testing import conv2d_nchw_python

Define the computation

To begin with, let us define the computation of a convolution layer. The function should return the list of input/output tensors. From these tensors, the auto-scheduler can get the whole computational graph.

@auto_scheduler.register_workload
def conv2d_layer(N, H, W, CO, CI, KH, KW, stride, padding):
    data = te.placeholder((N, CI, H, W), name="data")
    kernel = te.placeholder((CO, CI, KH, KW), name="kernel")
    bias = te.placeholder((1, CO, 1, 1), name="bias")
    conv = topi.nn.conv2d_nchw(data, kernel, stride, padding, dilation=1, out_dtype="float32")
    out = topi.nn.relu(conv + bias)
    return [data, kernel, bias, out]

Create the search task

We then create a search task for the last convolution layer in the resnet.

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

# Use the last layer in ResNet-50
N, H, W, CO, CI, KH, KW, strides, padding = 1, 7, 7, 512, 512, 3, 3, (1, 1), (1, 1)
task = auto_scheduler.SearchTask(
    func=conv2d_layer, args=(N, H, W, CO, CI, KH, KW, strides, padding), target=target
)

# Inspect the computational graph
print("Computational DAG:")
print(task.compute_dag)

Computational DAG:
data = PLACEHOLDER [1, 512, 7, 7]
pad_temp(i0, i1, i2, i3) = tir.if_then_else(((((i2 >= 1) && (i2 < 8)) && (i3 >= 1)) && (i3 < 8)), data[i0, i1, (i2 - 1), (i3 - 1)], 0f)
kernel = PLACEHOLDER [512, 512, 3, 3]
conv2d_nchw(nn, ff, yy, xx) += (pad_temp[nn, rc, (yy + ry), (xx + rx)]*kernel[ff, rc, ry, rx])
bias = PLACEHOLDER [1, 512, 1, 1]
T_add(ax0, ax1, ax2, ax3) = (conv2d_nchw[ax0, ax1, ax2, ax3] + bias[ax0, ax1, 0, 0])
compute(i0, i1, i2, i3) = max(T_add[i0, i1, i2, i3], 0f)

Next, we set parameters for the auto-scheduler. These parameters mainly specify how we do the measurement during the search.

• measure_ctx launches a different process for measurement to provide isolation. It can protect the main process from GPU crashes during measurement and avoid other runtime conflicts.

• min_repeat_ms defines the minimum duration of one “repeat” in every measurement. This can warmup the GPU, which is necessary to get accurate measurement results. Typically, we recommend a value >= 300 ms.

• num_measure_trials is the number of measurement trials we can use during the search. We only make 10 trials in this tutorial for a fast demonstration. In practice, 1000 is a good value for the search to converge. You can do more trials according to your time budget.

• In addition, we use RecordToFile to dump measurement records into a file conv2d.json. The measurement records can be used to query the history best, resume the search, and do more analyses later.

• see auto_scheduler.TuningOptions, auto_scheduler.LocalRPCMeasureContext for more parameters.

log_file = "conv2d.json"
measure_ctx = auto_scheduler.LocalRPCMeasureContext(min_repeat_ms=300)
tune_option = auto_scheduler.TuningOptions(
    num_measure_trials=10,  # change this to 1000 to achieve the best performance
    runner=measure_ctx.runner,
    measure_callbacks=[auto_scheduler.RecordToFile(log_file)],
    verbose=2,
)

Get devices for measurement successfully!

Run the search

Now we get all inputs ready. Pretty simple, isn’t it? We can kick off the search and let the auto-scheduler do its magic. After some measurement trials, we can load the best schedule from the log file and apply it.

# Run auto-tuning (search)
# We do not run the tuning in our webpage server since it takes too long.
# Uncomment the following line to run it by yourself.
# task.tune(tune_option)
# Apply the best schedule
sch, args = task.apply_best(log_file)

# Kill the measurement process
del measure_ctx

We can lower the schedule to see the IR after auto-scheduling. The auto-scheduler correctly performs optimizations including multi-level tiling, cooperative fetching, unrolling and operator fusion.

print("Lowered TIR:")
print(tvm.lower(sch, args, simple_mode=True))

Lowered TIR:
# from tvm.script import ir as I
# from tvm.script import tir as T

@I.ir_module
class Module:
    @T.prim_func
    def main(data: T.Buffer((1, 512, 7, 7), "float32"), kernel: T.Buffer((512, 512, 3, 3), "float32"), bias: T.Buffer((1, 512, 1, 1), "float32"), compute: T.Buffer((1, 512, 7, 7), "float32")):
        T.func_attr({"from_legacy_te_schedule": T.bool(True), "tir.noalias": T.bool(True)})
        blockIdx_x = T.launch_thread("blockIdx.x", 28)
        conv2d_nchw = T.allocate([14], "float32", "local")
        pad_temp_shared = T.allocate([72], "float32", "shared")
        kernel_shared = T.allocate([3072], "float32", "shared")
        threadIdx_x = T.launch_thread("threadIdx.x", 64)
        conv2d_nchw_1 = T.Buffer((14,), data=conv2d_nchw, scope="local", align=32)
        conv2d_nchw_1[0] = T.float32(0)
        conv2d_nchw_1[1] = T.float32(0)
        conv2d_nchw_1[2] = T.float32(0)
        conv2d_nchw_1[3] = T.float32(0)
        conv2d_nchw_1[4] = T.float32(0)
        conv2d_nchw_1[5] = T.float32(0)
        conv2d_nchw_1[6] = T.float32(0)
        conv2d_nchw_1[7] = T.float32(0)
        conv2d_nchw_1[8] = T.float32(0)
        conv2d_nchw_1[9] = T.float32(0)
        conv2d_nchw_1[10] = T.float32(0)
        conv2d_nchw_1[11] = T.float32(0)
        conv2d_nchw_1[12] = T.float32(0)
        conv2d_nchw_1[13] = T.float32(0)
        for rc_outer_outer, ry_outer_outer in T.grid(64, 3):
            cse_var_2: T.int32 = rc_outer_outer * 72
            cse_var_1: T.int32 = ry_outer_outer * 3
            pad_temp_shared_1 = T.Buffer((72,), data=pad_temp_shared, scope="shared")
            with T.launch_thread("threadIdx.x", 64) as threadIdx_x_1:
                if T.likely(threadIdx_x_1 < 18):
                    cse_var_4: T.int32 = rc_outer_outer * 392
                    cse_var_3: T.int32 = ry_outer_outer * 7
                    data_1 = T.Buffer((25088,), data=data.data)
                    pad_temp_shared_1[threadIdx_x_1 * 4] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= threadIdx_x_1 * 4 % 9 and threadIdx_x_1 * 4 % 9 < 8, data_1[cse_var_4 + threadIdx_x_1 * 4 // 9 * 49 + cse_var_3 + blockIdx_x % 7 * 7 + threadIdx_x_1 * 4 % 9 - 8], T.float32(0))
                    pad_temp_shared_1[threadIdx_x_1 * 4 + 1] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 1) % 9 and (threadIdx_x_1 * 4 + 1) % 9 < 8, data_1[cse_var_4 + (threadIdx_x_1 * 4 + 1) // 9 * 49 + cse_var_3 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 1) % 9 - 8], T.float32(0))
                    pad_temp_shared_1[threadIdx_x_1 * 4 + 2] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 2) % 9 and (threadIdx_x_1 * 4 + 2) % 9 < 8, data_1[cse_var_4 + (threadIdx_x_1 * 4 + 2) // 9 * 49 + cse_var_3 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 2) % 9 - 8], T.float32(0))
                    pad_temp_shared_1[threadIdx_x_1 * 4 + 3] = T.if_then_else(1 <= ry_outer_outer + blockIdx_x % 7 and ry_outer_outer + blockIdx_x % 7 < 8 and 1 <= (threadIdx_x_1 * 4 + 3) % 9 and (threadIdx_x_1 * 4 + 3) % 9 < 8, data_1[cse_var_4 + (threadIdx_x_1 * 4 + 3) // 9 * 49 + cse_var_3 + blockIdx_x % 7 * 7 + (threadIdx_x_1 * 4 + 3) % 9 - 8], T.float32(0))
            threadIdx_x_1 = T.env_thread("threadIdx.x")
            kernel_shared_1 = T.Buffer((3072,), data=kernel_shared, scope="shared")
            kernel_1 = T.Buffer((2359296,), data=kernel.data)
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 64) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 64) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 128) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 128) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 192] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 36864]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 256) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 256) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 320) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 320) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 384] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 73728]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 448) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 448) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 512) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 512) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 576] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 110592]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 640) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 640) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 704) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 704) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 768] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 147456]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 832) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 832) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 896) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 896) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 960] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 184320]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1024) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1024) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1088) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1088) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 1152] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 221184]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1216) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1216) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1280) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1280) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 1344] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 258048]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1408) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1408) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1472) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1472) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 1536] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 294912]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1600) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1600) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1664) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1664) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 1728] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 331776]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1792) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1792) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1856) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1856) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 1920] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 368640]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 1984) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 1984) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2048) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2048) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 2112] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 405504]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2176) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2176) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2240) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2240) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 2304] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 442368]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2368) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2368) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2432) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2432) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 2496] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 479232]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2560) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2560) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2624) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2624) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 2688] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 516096]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2752) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2752) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2816) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2816) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[threadIdx_x_1 + 2880] = kernel_1[blockIdx_x // 7 * 589824 + threadIdx_x_1 // 24 * 4608 + cse_var_2 + threadIdx_x_1 % 24 // 3 * 9 + cse_var_1 + threadIdx_x_1 % 3 + 552960]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 2944) // 24 * 24 + (threadIdx_x_1 + 16) % 24 // 3 * 3 + (threadIdx_x_1 + 1) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 2944) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 16) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 1) % 3]
            with T.launch_thread(threadIdx_x_1, 64):
                kernel_shared_1[(threadIdx_x_1 + 3008) // 24 * 24 + (threadIdx_x_1 + 8) % 24 // 3 * 3 + (threadIdx_x_1 + 2) % 3] = kernel_1[blockIdx_x // 7 * 589824 + (threadIdx_x_1 + 3008) // 24 * 4608 + cse_var_2 + (threadIdx_x_1 + 8) % 24 // 3 * 9 + cse_var_1 + (threadIdx_x_1 + 2) % 3]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 3]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[0] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[9] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 24]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 27]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 1]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 4]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[1] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[10] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 25]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 28]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 2]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 5]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[2] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[11] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[3] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[12] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[4] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[13] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[5] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[14] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[6] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[15] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[7] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[16] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[8] * kernel_shared_1[threadIdx_x * 48 + 26]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[17] * kernel_shared_1[threadIdx_x * 48 + 29]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 6]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 9]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[18] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[27] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 30]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 33]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 7]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 10]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[19] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[28] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 31]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 34]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 8]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 11]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[20] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[29] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[21] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[30] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[22] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[31] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[23] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[32] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[24] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[33] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[25] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[34] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[26] * kernel_shared_1[threadIdx_x * 48 + 32]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[35] * kernel_shared_1[threadIdx_x * 48 + 35]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 12]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 15]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[36] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[45] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 36]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 39]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 13]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 16]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[37] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[46] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 37]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 40]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 14]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 17]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[38] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[47] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[39] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[48] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[40] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[49] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[41] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[50] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[42] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[51] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[43] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[52] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[44] * kernel_shared_1[threadIdx_x * 48 + 38]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[53] * kernel_shared_1[threadIdx_x * 48 + 41]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 18]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 21]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[54] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[63] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 42]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 45]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 19]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 22]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[55] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[64] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 43]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 46]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[0] = conv2d_nchw_1[0] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[1] = conv2d_nchw_1[1] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[2] = conv2d_nchw_1[2] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[3] = conv2d_nchw_1[3] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[4] = conv2d_nchw_1[4] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[5] = conv2d_nchw_1[5] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 20]
            conv2d_nchw_1[6] = conv2d_nchw_1[6] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 23]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[56] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[7] = conv2d_nchw_1[7] + pad_temp_shared_1[65] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[57] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[8] = conv2d_nchw_1[8] + pad_temp_shared_1[66] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[58] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[9] = conv2d_nchw_1[9] + pad_temp_shared_1[67] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[59] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[10] = conv2d_nchw_1[10] + pad_temp_shared_1[68] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[60] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[11] = conv2d_nchw_1[11] + pad_temp_shared_1[69] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[61] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[12] = conv2d_nchw_1[12] + pad_temp_shared_1[70] * kernel_shared_1[threadIdx_x * 48 + 47]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[62] * kernel_shared_1[threadIdx_x * 48 + 44]
            conv2d_nchw_1[13] = conv2d_nchw_1[13] + pad_temp_shared_1[71] * kernel_shared_1[threadIdx_x * 48 + 47]
        for i1_inner, i3_inner in T.grid(2, 7):
            compute_1 = T.Buffer((25088,), data=compute.data)
            bias_1 = T.Buffer((512,), data=bias.data)
            compute_1[blockIdx_x // 7 * 6272 + threadIdx_x * 98 + i1_inner * 49 + blockIdx_x % 7 * 7 + i3_inner] = T.max(conv2d_nchw_1[i1_inner * 7 + i3_inner] + bias_1[blockIdx_x // 7 * 128 + threadIdx_x * 2 + i1_inner], T.float32(0))

Check correctness and evaluate performance

We build the binary and check its correctness and performance.

func = tvm.build(sch, args, target)

# Check correctness
data_np = np.random.uniform(size=(N, CI, H, W)).astype(np.float32)
weight_np = np.random.uniform(size=(CO, CI, KH, KW)).astype(np.float32)
bias_np = np.random.uniform(size=(1, CO, 1, 1)).astype(np.float32)
conv_np = conv2d_nchw_python(data_np, weight_np, strides, padding)
out_np = np.maximum(conv_np + bias_np, 0.0)

dev = tvm.cuda()
data_tvm = tvm.nd.array(data_np, device=dev)
weight_tvm = tvm.nd.array(weight_np, device=dev)
bias_tvm = tvm.nd.array(bias_np, device=dev)
out_tvm = tvm.nd.empty(out_np.shape, device=dev)
func(data_tvm, weight_tvm, bias_tvm, out_tvm)

# Check results
np.testing.assert_allclose(out_np, out_tvm.numpy(), rtol=1e-3)

# Evaluate execution time
evaluator = func.time_evaluator(func.entry_name, dev, min_repeat_ms=500)
print(
    "Execution time of this operator: %.3f ms"
    % (np.median(evaluator(data_tvm, weight_tvm, bias_tvm, out_tvm).results) * 1000)
)

Execution time of this operator: 0.344 ms

Using the record file

During the search, all measurement records are dumped into the record file “conv2d.json”. The measurement records can be used to re-apply search results, resume the search, and perform other analyses.

Here is an example where we load the best schedule from a file, print the equivalent python schedule API and CUDA source code. They can be used for debugging and learning the behavior of the auto-scheduler.

print("Equivalent python schedule:")
print(task.print_best(log_file, print_mode="schedule"))

print("CUDA source code:")
print(task.print_best(log_file, print_mode="cuda"))

Equivalent python schedule:
pad_temp_i0, pad_temp_i1, pad_temp_i2, pad_temp_i3 = tuple(pad_temp.op.axis) + tuple(pad_temp.op.reduce_axis)
conv2d_nchw_nn, conv2d_nchw_ff, conv2d_nchw_yy, conv2d_nchw_xx, conv2d_nchw_rc, conv2d_nchw_ry, conv2d_nchw_rx = tuple(conv2d_nchw.op.axis) + tuple(conv2d_nchw.op.reduce_axis)
T_add_ax0, T_add_ax1, T_add_ax2, T_add_ax3 = tuple(T_add.op.axis) + tuple(T_add.op.reduce_axis)
compute_i0, compute_i1, compute_i2, compute_i3 = tuple(compute.op.axis) + tuple(compute.op.reduce_axis)
s[T_add].compute_inline()
conv2d_nchw_nn_o_i, conv2d_nchw_nn_i = s[conv2d_nchw].split(conv2d_nchw_nn, factor=1)
conv2d_nchw_nn_o_o_i, conv2d_nchw_nn_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_i, factor=1)
conv2d_nchw_nn_o_o_o_i, conv2d_nchw_nn_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_i, factor=1)
conv2d_nchw_nn_o_o_o_o, conv2d_nchw_nn_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_nn_o_o_o_i, factor=1)
conv2d_nchw_ff_o_i, conv2d_nchw_ff_i = s[conv2d_nchw].split(conv2d_nchw_ff, factor=1)
conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=2)
conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=64)
conv2d_nchw_ff_o_o_o_o, conv2d_nchw_ff_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_o_i, factor=1)
conv2d_nchw_yy_o_i, conv2d_nchw_yy_i = s[conv2d_nchw].split(conv2d_nchw_yy, factor=1)
conv2d_nchw_yy_o_o_i, conv2d_nchw_yy_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_i, factor=1)
conv2d_nchw_yy_o_o_o_i, conv2d_nchw_yy_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_i, factor=1)
conv2d_nchw_yy_o_o_o_o, conv2d_nchw_yy_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_yy_o_o_o_i, factor=1)
conv2d_nchw_xx_o_i, conv2d_nchw_xx_i = s[conv2d_nchw].split(conv2d_nchw_xx, factor=1)
conv2d_nchw_xx_o_o_i, conv2d_nchw_xx_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_i, factor=7)
conv2d_nchw_xx_o_o_o_i, conv2d_nchw_xx_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_i, factor=1)
conv2d_nchw_xx_o_o_o_o, conv2d_nchw_xx_o_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_o_i, factor=1)
conv2d_nchw_rc_o_i, conv2d_nchw_rc_i = s[conv2d_nchw].split(conv2d_nchw_rc, factor=2)
conv2d_nchw_rc_o_o, conv2d_nchw_rc_o_i = s[conv2d_nchw].split(conv2d_nchw_rc_o_i, factor=4)
conv2d_nchw_ry_o_i, conv2d_nchw_ry_i = s[conv2d_nchw].split(conv2d_nchw_ry, factor=1)
conv2d_nchw_ry_o_o, conv2d_nchw_ry_o_i = s[conv2d_nchw].split(conv2d_nchw_ry_o_i, factor=1)
conv2d_nchw_rx_o_i, conv2d_nchw_rx_i = s[conv2d_nchw].split(conv2d_nchw_rx, factor=1)
conv2d_nchw_rx_o_o, conv2d_nchw_rx_o_i = s[conv2d_nchw].split(conv2d_nchw_rx_o_i, factor=3)
s[conv2d_nchw].reorder(conv2d_nchw_nn_o_o_o_o, conv2d_nchw_ff_o_o_o_o, conv2d_nchw_yy_o_o_o_o, conv2d_nchw_xx_o_o_o_o, conv2d_nchw_nn_o_o_o_i, conv2d_nchw_ff_o_o_o_i, conv2d_nchw_yy_o_o_o_i, conv2d_nchw_xx_o_o_o_i, conv2d_nchw_nn_o_o_i, conv2d_nchw_ff_o_o_i, conv2d_nchw_yy_o_o_i, conv2d_nchw_xx_o_o_i, conv2d_nchw_rc_o_o, conv2d_nchw_ry_o_o, conv2d_nchw_rx_o_o, conv2d_nchw_rc_o_i, conv2d_nchw_ry_o_i, conv2d_nchw_rx_o_i, conv2d_nchw_nn_o_i, conv2d_nchw_ff_o_i, conv2d_nchw_yy_o_i, conv2d_nchw_xx_o_i, conv2d_nchw_rc_i, conv2d_nchw_ry_i, conv2d_nchw_rx_i, conv2d_nchw_nn_i, conv2d_nchw_ff_i, conv2d_nchw_yy_i, conv2d_nchw_xx_i)
compute_i0_o_i, compute_i0_i = s[compute].split(compute_i0, factor=1)
compute_i0_o_o_i, compute_i0_o_i = s[compute].split(compute_i0_o_i, factor=1)
compute_i0_o_o_o, compute_i0_o_o_i = s[compute].split(compute_i0_o_o_i, factor=1)
compute_i1_o_i, compute_i1_i = s[compute].split(compute_i1, factor=2)
compute_i1_o_o_i, compute_i1_o_i = s[compute].split(compute_i1_o_i, factor=64)
compute_i1_o_o_o, compute_i1_o_o_i = s[compute].split(compute_i1_o_o_i, factor=1)
compute_i2_o_i, compute_i2_i = s[compute].split(compute_i2, factor=1)
compute_i2_o_o_i, compute_i2_o_i = s[compute].split(compute_i2_o_i, factor=1)
compute_i2_o_o_o, compute_i2_o_o_i = s[compute].split(compute_i2_o_o_i, factor=1)
compute_i3_o_i, compute_i3_i = s[compute].split(compute_i3, factor=7)
compute_i3_o_o_i, compute_i3_o_i = s[compute].split(compute_i3_o_i, factor=1)
compute_i3_o_o_o, compute_i3_o_o_i = s[compute].split(compute_i3_o_o_i, factor=1)
s[compute].reorder(compute_i0_o_o_o, compute_i1_o_o_o, compute_i2_o_o_o, compute_i3_o_o_o, compute_i0_o_o_i, compute_i1_o_o_i, compute_i2_o_o_i, compute_i3_o_o_i, compute_i0_o_i, compute_i1_o_i, compute_i2_o_i, compute_i3_o_i, compute_i0_i, compute_i1_i, compute_i2_i, compute_i3_i)
s[conv2d_nchw].compute_at(s[compute], compute_i3_o_i)
kernel_shared = s.cache_read(kernel, "shared", [conv2d_nchw])
kernel_shared_ax0, kernel_shared_ax1, kernel_shared_ax2, kernel_shared_ax3 = tuple(kernel_shared.op.axis)
s[kernel_shared].compute_at(s[conv2d_nchw], conv2d_nchw_rx_o_o)
pad_temp_shared = s.cache_read(pad_temp, "shared", [conv2d_nchw])
pad_temp_shared_ax0, pad_temp_shared_ax1, pad_temp_shared_ax2, pad_temp_shared_ax3 = tuple(pad_temp_shared.op.axis)
s[pad_temp_shared].compute_at(s[conv2d_nchw], conv2d_nchw_rx_o_o)
s[pad_temp].compute_inline()
compute_i0_o_o_o_i1_o_o_o_fused_i2_o_o_o_fused_i3_o_o_o_fused = s[compute].fuse(compute_i0_o_o_o, compute_i1_o_o_o, compute_i2_o_o_o, compute_i3_o_o_o)
s[compute].bind(compute_i0_o_o_o_i1_o_o_o_fused_i2_o_o_o_fused_i3_o_o_o_fused, te.thread_axis("blockIdx.x"))
compute_i0_o_o_i_i1_o_o_i_fused_i2_o_o_i_fused_i3_o_o_i_fused = s[compute].fuse(compute_i0_o_o_i, compute_i1_o_o_i, compute_i2_o_o_i, compute_i3_o_o_i)
s[compute].bind(compute_i0_o_o_i_i1_o_o_i_fused_i2_o_o_i_fused_i3_o_o_i_fused, te.thread_axis("vthread"))
compute_i0_o_i_i1_o_i_fused_i2_o_i_fused_i3_o_i_fused = s[compute].fuse(compute_i0_o_i, compute_i1_o_i, compute_i2_o_i, compute_i3_o_i)
s[compute].bind(compute_i0_o_i_i1_o_i_fused_i2_o_i_fused_i3_o_i_fused, te.thread_axis("threadIdx.x"))
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[kernel_shared].fuse(kernel_shared_ax0, kernel_shared_ax1, kernel_shared_ax2, kernel_shared_ax3)
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=1)
s[kernel_shared].vectorize(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[kernel_shared].split(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
s[kernel_shared].bind(kernel_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused = s[pad_temp_shared].fuse(pad_temp_shared_ax0, pad_temp_shared_ax1, pad_temp_shared_ax2, pad_temp_shared_ax3)
pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused, factor=4)
s[pad_temp_shared].vectorize(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_i)
pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_o, pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i = s[pad_temp_shared].split(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o, factor=64)
s[pad_temp_shared].bind(pad_temp_shared_ax0_ax1_fused_ax2_fused_ax3_fused_o_i, te.thread_axis("threadIdx.x"))
s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "auto_unroll_max_step", 512)
s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "unroll_explicit", True)

CUDA source code:

#if (((__CUDACC_VER_MAJOR__ == 11) && (__CUDACC_VER_MINOR__ >= 4)) || \
     (__CUDACC_VER_MAJOR__ > 11))
#define TVM_ENABLE_L2_PREFETCH 1
#else
#define TVM_ENABLE_L2_PREFETCH 0
#endif

#ifdef _WIN32
  using uint = unsigned int;
  using uchar = unsigned char;
  using ushort = unsigned short;
  using int64_t = long long;
  using uint64_t = unsigned long long;
#else
  #define uint unsigned int
  #define uchar unsigned char
  #define ushort unsigned short
  #define int64_t long long
  #define uint64_t unsigned long long
#endif
extern "C" __global__ void __launch_bounds__(64) default_function_kernel(float* __restrict__ bias, float* __restrict__ compute, float* __restrict__ data, float* __restrict__ kernel);
extern "C" __global__ void __launch_bounds__(64) default_function_kernel(float* __restrict__ bias, float* __restrict__ compute, float* __restrict__ data, float* __restrict__ kernel) {
  float conv2d_nchw[14];
  __shared__ float pad_temp_shared[72];
  __shared__ float kernel_shared[3072];
  conv2d_nchw[0] = 0.000000e+00f;
  conv2d_nchw[1] = 0.000000e+00f;
  conv2d_nchw[2] = 0.000000e+00f;
  conv2d_nchw[3] = 0.000000e+00f;
  conv2d_nchw[4] = 0.000000e+00f;
  conv2d_nchw[5] = 0.000000e+00f;
  conv2d_nchw[6] = 0.000000e+00f;
  conv2d_nchw[7] = 0.000000e+00f;
  conv2d_nchw[8] = 0.000000e+00f;
  conv2d_nchw[9] = 0.000000e+00f;
  conv2d_nchw[10] = 0.000000e+00f;
  conv2d_nchw[11] = 0.000000e+00f;
  conv2d_nchw[12] = 0.000000e+00f;
  conv2d_nchw[13] = 0.000000e+00f;
  for (int rc_outer_outer = 0; rc_outer_outer < 64; ++rc_outer_outer) {
    for (int ry_outer_outer = 0; ry_outer_outer < 3; ++ry_outer_outer) {
      __syncthreads();
      if (((int)threadIdx.x) < 18) {
        pad_temp_shared[(((int)threadIdx.x) * 4)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= ((((int)threadIdx.x) * 4) % 9))) && (((((int)threadIdx.x) * 4) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + (((((int)threadIdx.x) * 4) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + ((((int)threadIdx.x) * 4) % 9)) - 8)] : 0.000000e+00f);
        pad_temp_shared[((((int)threadIdx.x) * 4) + 1)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 1) % 9))) && ((((((int)threadIdx.x) * 4) + 1) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 1) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 1) % 9)) - 8)] : 0.000000e+00f);
        pad_temp_shared[((((int)threadIdx.x) * 4) + 2)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 2) % 9))) && ((((((int)threadIdx.x) * 4) + 2) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 2) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 2) % 9)) - 8)] : 0.000000e+00f);
        pad_temp_shared[((((int)threadIdx.x) * 4) + 3)] = (((((1 <= (ry_outer_outer + (((int)blockIdx.x) % 7))) && ((ry_outer_outer + (((int)blockIdx.x) % 7)) < 8)) && (1 <= (((((int)threadIdx.x) * 4) + 3) % 9))) && ((((((int)threadIdx.x) * 4) + 3) % 9) < 8)) ? data[((((((rc_outer_outer * 392) + ((((((int)threadIdx.x) * 4) + 3) / 9) * 49)) + (ry_outer_outer * 7)) + ((((int)blockIdx.x) % 7) * 7)) + (((((int)threadIdx.x) * 4) + 3) % 9)) - 8)] : 0.000000e+00f);
      }
      kernel_shared[((int)threadIdx.x)] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 64) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 64) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 128) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 128) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 192)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 36864)];
      kernel_shared[(((((((int)threadIdx.x) + 256) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 256) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 320) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 320) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 384)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 73728)];
      kernel_shared[(((((((int)threadIdx.x) + 448) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 448) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 512) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 512) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 576)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 110592)];
      kernel_shared[(((((((int)threadIdx.x) + 640) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 640) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 704) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 704) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 768)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 147456)];
      kernel_shared[(((((((int)threadIdx.x) + 832) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 832) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 896) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 896) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 960)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 184320)];
      kernel_shared[(((((((int)threadIdx.x) + 1024) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1024) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 1088) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1088) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 1152)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 221184)];
      kernel_shared[(((((((int)threadIdx.x) + 1216) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1216) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 1280) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1280) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 1344)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 258048)];
      kernel_shared[(((((((int)threadIdx.x) + 1408) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1408) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 1472) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1472) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 1536)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 294912)];
      kernel_shared[(((((((int)threadIdx.x) + 1600) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1600) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 1664) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1664) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 1728)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 331776)];
      kernel_shared[(((((((int)threadIdx.x) + 1792) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1792) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 1856) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1856) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 1920)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 368640)];
      kernel_shared[(((((((int)threadIdx.x) + 1984) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 1984) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 2048) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2048) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 2112)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 405504)];
      kernel_shared[(((((((int)threadIdx.x) + 2176) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2176) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 2240) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2240) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 2304)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 442368)];
      kernel_shared[(((((((int)threadIdx.x) + 2368) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2368) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 2432) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2432) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 2496)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 479232)];
      kernel_shared[(((((((int)threadIdx.x) + 2560) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2560) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 2624) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2624) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 2688)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 516096)];
      kernel_shared[(((((((int)threadIdx.x) + 2752) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2752) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 2816) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2816) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      kernel_shared[(((int)threadIdx.x) + 2880)] = kernel[((((((((((int)blockIdx.x) / 7) * 589824) + ((((int)threadIdx.x) / 24) * 4608)) + (rc_outer_outer * 72)) + (((((int)threadIdx.x) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + (((int)threadIdx.x) % 3)) + 552960)];
      kernel_shared[(((((((int)threadIdx.x) + 2944) / 24) * 24) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 1) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 2944) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 16) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 1) % 3))];
      kernel_shared[(((((((int)threadIdx.x) + 3008) / 24) * 24) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 3)) + ((((int)threadIdx.x) + 2) % 3))] = kernel[(((((((((int)blockIdx.x) / 7) * 589824) + (((((int)threadIdx.x) + 3008) / 24) * 4608)) + (rc_outer_outer * 72)) + ((((((int)threadIdx.x) + 8) % 24) / 3) * 9)) + (ry_outer_outer * 3)) + ((((int)threadIdx.x) + 2) % 3))];
      __syncthreads();
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[0] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[1] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[2] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[3] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[4] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[5] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[6] * kernel_shared[(((int)threadIdx.x) * 48)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 3)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[0] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[9] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 24)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 27)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 1)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 4)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[1] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[10] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 25)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 28)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 2)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 5)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[2] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[11] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[3] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[12] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[4] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[13] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[5] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[14] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[6] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[15] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[7] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[16] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[8] * kernel_shared[((((int)threadIdx.x) * 48) + 26)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[17] * kernel_shared[((((int)threadIdx.x) * 48) + 29)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 6)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 9)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[18] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[27] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 30)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 33)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 7)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 10)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[19] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[28] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 31)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 34)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 8)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 11)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[20] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[29] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[21] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[30] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[22] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[31] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[23] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[32] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[24] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[33] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[25] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[34] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[26] * kernel_shared[((((int)threadIdx.x) * 48) + 32)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[35] * kernel_shared[((((int)threadIdx.x) * 48) + 35)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 12)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 15)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[36] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[45] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 36)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 39)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 13)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 16)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[37] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[46] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 37)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 40)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 14)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 17)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[38] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[47] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[39] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[48] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[40] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[49] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[41] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[50] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[42] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[51] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[43] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[52] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[44] * kernel_shared[((((int)threadIdx.x) * 48) + 38)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[53] * kernel_shared[((((int)threadIdx.x) * 48) + 41)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 18)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 21)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[54] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[63] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 42)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 45)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 19)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 22)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[55] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[64] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 43)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 46)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 20)]));
      conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 23)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[56] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[65] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[57] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[66] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[58] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[67] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[59] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[68] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[60] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[69] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[61] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[70] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[62] * kernel_shared[((((int)threadIdx.x) * 48) + 44)]));
      conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[71] * kernel_shared[((((int)threadIdx.x) * 48) + 47)]));
    }
  }
  for (int i1_inner = 0; i1_inner < 2; ++i1_inner) {
    for (int i3_inner = 0; i3_inner < 7; ++i3_inner) {
      compute[((((((((int)blockIdx.x) / 7) * 6272) + (((int)threadIdx.x) * 98)) + (i1_inner * 49)) + ((((int)blockIdx.x) % 7) * 7)) + i3_inner)] = max((conv2d_nchw[((i1_inner * 7) + i3_inner)] + bias[((((((int)blockIdx.x) / 7) * 128) + (((int)threadIdx.x) * 2)) + i1_inner)]), 0.000000e+00f);
    }
  }
}

A more complicated example is to resume the search. In this case, we need to create the search policy and cost model by ourselves and resume the status of search policy and cost model with the log file. In the example below we resume the status and do more 5 trials.

def resume_search(task, log_file):
    print("Resume search:")
    cost_model = auto_scheduler.XGBModel()
    cost_model.update_from_file(log_file)
    search_policy = auto_scheduler.SketchPolicy(
        task, cost_model, init_search_callbacks=[auto_scheduler.PreloadMeasuredStates(log_file)]
    )
    measure_ctx = auto_scheduler.LocalRPCMeasureContext(min_repeat_ms=300)
    tune_option = auto_scheduler.TuningOptions(
        num_measure_trials=5,
        runner=measure_ctx.runner,
        measure_callbacks=[auto_scheduler.RecordToFile(log_file)],
    )
    task.tune(tune_option, search_policy=search_policy)

    # Kill the measurement process
    del measure_ctx


# We do not run the tuning in our webpage server since it takes too long.
# Uncomment the following line to run it by yourself.
# resume_search(task, log_file)