Note

This tutorial can be used interactively with Google Colab! You can also click here to run the Jupyter notebook locally.

https://raw.githubusercontent.com/tlc-pack/web-data/main/images/utilities/colab_button.svg

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!

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: 0.264 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=2)
conv2d_nchw_ff_o_o_i, conv2d_nchw_ff_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_i, factor=1)
conv2d_nchw_ff_o_o_o_i, conv2d_nchw_ff_o_o_i = s[conv2d_nchw].split(conv2d_nchw_ff_o_o_i, factor=2)
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=7)
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=1)
conv2d_nchw_xx_o_o_o_i, conv2d_nchw_xx_o_o_i = s[conv2d_nchw].split(conv2d_nchw_xx_o_o_i, factor=7)
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=1)
conv2d_nchw_ry_o_i, conv2d_nchw_ry_i = s[conv2d_nchw].split(conv2d_nchw_ry, factor=3)
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=2)
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=7)
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=1)
compute_i3_o_o_i, compute_i3_o_i = s[compute].split(compute_i3_o_i, factor=7)
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=6)
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=14)
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=1)
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=14)
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", 1024)
s[conv2d_nchw].pragma(conv2d_nchw_nn_o_o_o_o, "unroll_explicit", True)

CUDA source code:

#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__(14) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
  float conv2d_nchw[14];
  __shared__ float pad_temp_shared[162];
  __shared__ float kernel_shared[72];
  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 < 256; ++rc_outer_outer) {
    __syncthreads();
    pad_temp_shared[((int)threadIdx.x)] = ((((9 <= ((int)threadIdx.x)) && (1 <= (((int)threadIdx.x) % 9))) && ((((int)threadIdx.x) % 9) < 8)) ? data[((((rc_outer_outer * 98) + ((((int)threadIdx.x) / 9) * 7)) + (((int)threadIdx.x) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 14)] = (((1 <= ((((int)threadIdx.x) + 5) % 9)) && (((((int)threadIdx.x) + 5) % 9) < 8)) ? data[((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 14) / 9) * 7)) + ((((int)threadIdx.x) + 5) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 28)] = (((1 <= ((((int)threadIdx.x) + 1) % 9)) && (((((int)threadIdx.x) + 1) % 9) < 8)) ? data[((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 28) / 9) * 7)) + ((((int)threadIdx.x) + 1) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 42)] = (((1 <= ((((int)threadIdx.x) + 6) % 9)) && (((((int)threadIdx.x) + 6) % 9) < 8)) ? data[((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 42) / 9) * 7)) + ((((int)threadIdx.x) + 6) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 56)] = (((1 <= ((((int)threadIdx.x) + 2) % 9)) && (((((int)threadIdx.x) + 2) % 9) < 8)) ? data[((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 56) / 9) * 7)) + ((((int)threadIdx.x) + 2) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 70)] = (((((9 <= ((((int)threadIdx.x) + 70) % 81)) && (((((int)threadIdx.x) + 70) % 81) < 72)) && (1 <= ((((int)threadIdx.x) + 7) % 9))) && (((((int)threadIdx.x) + 7) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 70) / 81) * 49)) + ((((((int)threadIdx.x) + 70) % 81) / 9) * 7)) + ((((int)threadIdx.x) + 7) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 84)] = ((((6 <= ((int)threadIdx.x)) && (1 <= ((((int)threadIdx.x) + 3) % 9))) && (((((int)threadIdx.x) + 3) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 84) / 81) * 49)) + (((((int)threadIdx.x) + 3) / 9) * 7)) + ((((int)threadIdx.x) + 3) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 98)] = (((1 <= ((((int)threadIdx.x) + 8) % 9)) && (((((int)threadIdx.x) + 8) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 98) / 81) * 49)) + (((((int)threadIdx.x) + 17) / 9) * 7)) + ((((int)threadIdx.x) + 8) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 112)] = (((1 <= ((((int)threadIdx.x) + 4) % 9)) && (((((int)threadIdx.x) + 4) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 112) / 81) * 49)) + (((((int)threadIdx.x) + 31) / 9) * 7)) + ((((int)threadIdx.x) + 4) % 9)) - 8)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 126)] = (((1 <= (((int)threadIdx.x) % 9)) && ((((int)threadIdx.x) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 126) / 81) * 49)) + ((((int)threadIdx.x) / 9) * 7)) + (((int)threadIdx.x) % 9)) + 27)] : 0.000000e+00f);
    pad_temp_shared[(((int)threadIdx.x) + 140)] = ((((((int)threadIdx.x) < 13) && (1 <= ((((int)threadIdx.x) + 5) % 9))) && (((((int)threadIdx.x) + 5) % 9) < 8)) ? data[(((((rc_outer_outer * 98) + (((((int)threadIdx.x) + 140) / 81) * 49)) + (((((int)threadIdx.x) + 59) / 9) * 7)) + ((((int)threadIdx.x) + 5) % 9)) - 8)] : 0.000000e+00f);
    if (((int)threadIdx.x) < 8) {
      pad_temp_shared[(((int)threadIdx.x) + 154)] = 0.000000e+00f;
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[(((int)threadIdx.x) * 6)] = kernel[((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6))];
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[((((int)threadIdx.x) * 6) + 1)] = kernel[(((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6)) + 1)];
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[((((int)threadIdx.x) * 6) + 2)] = kernel[(((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6)) + 2)];
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[((((int)threadIdx.x) * 6) + 3)] = kernel[(((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6)) + 3)];
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[((((int)threadIdx.x) * 6) + 4)] = kernel[(((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6)) + 4)];
    }
    if (((int)threadIdx.x) < 12) {
      kernel_shared[((((int)threadIdx.x) * 6) + 5)] = kernel[(((((((int)blockIdx.x) * 18432) + ((((int)threadIdx.x) / 3) * 4608)) + (rc_outer_outer * 18)) + ((((int)threadIdx.x) % 3) * 6)) + 5)];
    }
    __syncthreads();
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[(((int)threadIdx.x) % 7)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 9)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[((((int)threadIdx.x) / 7) * 36)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[(((int)threadIdx.x) % 7)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 9)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 18)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 9)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 63)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 3)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 9)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 63)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 21)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 63)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 72)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 6)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 18)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 27)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 36)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 45)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 54)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 63)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 72)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 24)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 81)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 9)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 81)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 27)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 144)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 12)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 90)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 144)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 30)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 144)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 153)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 15)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 99)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 108)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 117)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 126)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 135)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 144)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 153)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 33)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 1)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 1)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 1)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 19)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 64)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 4)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 10)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 64)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 22)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 64)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 73)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 7)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 19)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 28)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 37)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 46)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 55)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 64)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 73)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 25)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 10)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 82)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 28)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 145)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 13)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 91)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 145)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 31)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 145)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 154)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 16)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 100)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 109)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 118)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 127)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 136)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 145)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 154)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 34)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 2)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 2)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 20)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 65)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 5)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 11)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 65)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 23)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 65)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 74)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 8)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 20)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 29)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 38)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 47)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 56)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 65)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 74)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 26)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 11)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 83)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 29)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 146)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 14)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 92)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 146)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 32)]));
    conv2d_nchw[0] = (conv2d_nchw[0] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[1] = (conv2d_nchw[1] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[2] = (conv2d_nchw[2] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[3] = (conv2d_nchw[3] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[4] = (conv2d_nchw[4] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[5] = (conv2d_nchw[5] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 146)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[6] = (conv2d_nchw[6] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 155)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 17)]));
    conv2d_nchw[7] = (conv2d_nchw[7] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 101)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[8] = (conv2d_nchw[8] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 110)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[9] = (conv2d_nchw[9] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 119)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[10] = (conv2d_nchw[10] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 128)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[11] = (conv2d_nchw[11] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 137)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[12] = (conv2d_nchw[12] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 146)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
    conv2d_nchw[13] = (conv2d_nchw[13] + (pad_temp_shared[((((int)threadIdx.x) % 7) + 155)] * kernel_shared[(((((int)threadIdx.x) / 7) * 36) + 35)]));
  }
  for (int i1_inner = 0; i1_inner < 2; ++i1_inner) {
    for (int i2_inner = 0; i2_inner < 7; ++i2_inner) {
      compute[(((((((int)blockIdx.x) * 196) + ((((int)threadIdx.x) / 7) * 98)) + (i1_inner * 49)) + (i2_inner * 7)) + (((int)threadIdx.x) % 7))] = max((conv2d_nchw[((i1_inner * 7) + i2_inner)] + bias[(((((int)blockIdx.x) * 4) + ((((int)threadIdx.x) / 7) * 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


resume_search(task, log_file)

Resume search:
/venv/apache-tvm-py3.7/lib/python3.7/site-packages/xgboost/training.py:17: UserWarning: Old style callback is deprecated.  See: https://xgboost.readthedocs.io/en/latest/python/callbacks.html
  warnings.warn(f'Old style callback is deprecated.  See: {link}', UserWarning)
Get devices for measurement successfully!

Total running time of the script: ( 6 minutes 8.038 seconds)

Gallery generated by Sphinx-Gallery