Note

Click here to download the full example code

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)

Out:

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]
compute(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) = (compute[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 master 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,
)

Out:

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)
)

Out:

Execution time of this operator: 0.362 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"))

Out:

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)
compute_nn, compute_ff, compute_yy, compute_xx, compute_rc, compute_ry, compute_rx = tuple(compute.op.axis) + tuple(compute.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()
compute_nn_o_i, compute_nn_i = s[compute].split(compute_nn, factor=1)
compute_nn_o_o_i, compute_nn_o_i = s[compute].split(compute_nn_o_i, factor=1)
compute_nn_o_o_o_i, compute_nn_o_o_i = s[compute].split(compute_nn_o_o_i, factor=1)
compute_nn_o_o_o_o, compute_nn_o_o_o_i = s[compute].split(compute_nn_o_o_o_i, factor=1)
compute_ff_o_i, compute_ff_i = s[compute].split(compute_ff, factor=1)
compute_ff_o_o_i, compute_ff_o_i = s[compute].split(compute_ff_o_i, factor=2)
compute_ff_o_o_o_i, compute_ff_o_o_i = s[compute].split(compute_ff_o_o_i, factor=64)
compute_ff_o_o_o_o, compute_ff_o_o_o_i = s[compute].split(compute_ff_o_o_o_i, factor=1)
compute_yy_o_i, compute_yy_i = s[compute].split(compute_yy, factor=1)
compute_yy_o_o_i, compute_yy_o_i = s[compute].split(compute_yy_o_i, factor=1)
compute_yy_o_o_o_i, compute_yy_o_o_i = s[compute].split(compute_yy_o_o_i, factor=1)
compute_yy_o_o_o_o, compute_yy_o_o_o_i = s[compute].split(compute_yy_o_o_o_i, factor=1)
compute_xx_o_i, compute_xx_i = s[compute].split(compute_xx, factor=1)
compute_xx_o_o_i, compute_xx_o_i = s[compute].split(compute_xx_o_i, factor=7)
compute_xx_o_o_o_i, compute_xx_o_o_i = s[compute].split(compute_xx_o_o_i, factor=1)
compute_xx_o_o_o_o, compute_xx_o_o_o_i = s[compute].split(compute_xx_o_o_o_i, factor=1)
compute_rc_o_i, compute_rc_i = s[compute].split(compute_rc, factor=2)
compute_rc_o_o, compute_rc_o_i = s[compute].split(compute_rc_o_i, factor=4)
compute_ry_o_i, compute_ry_i = s[compute].split(compute_ry, factor=1)
compute_ry_o_o, compute_ry_o_i = s[compute].split(compute_ry_o_i, factor=1)
compute_rx_o_i, compute_rx_i = s[compute].split(compute_rx, factor=1)
compute_rx_o_o, compute_rx_o_i = s[compute].split(compute_rx_o_i, factor=3)
s[compute].reorder(compute_nn_o_o_o_o, compute_ff_o_o_o_o, compute_yy_o_o_o_o, compute_xx_o_o_o_o, compute_nn_o_o_o_i, compute_ff_o_o_o_i, compute_yy_o_o_o_i, compute_xx_o_o_o_i, compute_nn_o_o_i, compute_ff_o_o_i, compute_yy_o_o_i, compute_xx_o_o_i, compute_rc_o_o, compute_ry_o_o, compute_rx_o_o, compute_rc_o_i, compute_ry_o_i, compute_rx_o_i, compute_nn_o_i, compute_ff_o_i, compute_yy_o_i, compute_xx_o_i, compute_rc_i, compute_ry_i, compute_rx_i, compute_nn_i, compute_ff_i, compute_yy_i, compute_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[compute].compute_at(s[compute], compute_i3_o_i)
kernel_shared = s.cache_read(kernel, "shared", [compute])
kernel_shared_ax0, kernel_shared_ax1, kernel_shared_ax2, kernel_shared_ax3 = tuple(kernel_shared.op.axis)
s[kernel_shared].compute_at(s[compute], compute_rx_o_o)
pad_temp_shared = s.cache_read(pad_temp, "shared", [compute])
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[compute], compute_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[compute].pragma(compute_nn_o_o_o_o, "auto_unroll_max_step", 512)
s[compute].pragma(compute_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__(64) default_function_kernel0(float* __restrict__ data, float* __restrict__ kernel, float* __restrict__ compute, float* __restrict__ bias) {
  float compute1[14];
  __shared__ float pad_temp_shared[72];
  __shared__ float kernel_shared[3072];
  compute1[(0)] = 0.000000e+00f;
  compute1[(1)] = 0.000000e+00f;
  compute1[(2)] = 0.000000e+00f;
  compute1[(3)] = 0.000000e+00f;
  compute1[(4)] = 0.000000e+00f;
  compute1[(5)] = 0.000000e+00f;
  compute1[(6)] = 0.000000e+00f;
  compute1[(7)] = 0.000000e+00f;
  compute1[(8)] = 0.000000e+00f;
  compute1[(9)] = 0.000000e+00f;
  compute1[(10)] = 0.000000e+00f;
  compute1[(11)] = 0.000000e+00f;
  compute1[(12)] = 0.000000e+00f;
  compute1[(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);
      }
      if (((int)threadIdx.x) < 18) {
        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);
      }
      if (((int)threadIdx.x) < 18) {
        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);
      }
      if (((int)threadIdx.x) < 18) {
        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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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))] = 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();
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(0)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(9)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(1)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(10)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(2)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(3)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(4)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(5)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(6)] * kernel_shared[((((int)threadIdx.x) * 48))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 3))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(0)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(9)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(1)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(10)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(2)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(3)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(4)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(5)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(6)] * kernel_shared[(((((int)threadIdx.x) * 48) + 24))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 27))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(1)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(10)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(2)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(3)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(4)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(5)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(6)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(7)] * kernel_shared[(((((int)threadIdx.x) * 48) + 1))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(16)] * kernel_shared[(((((int)threadIdx.x) * 48) + 4))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(1)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(10)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(2)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(3)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(4)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(5)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(6)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(7)] * kernel_shared[(((((int)threadIdx.x) * 48) + 25))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(16)] * kernel_shared[(((((int)threadIdx.x) * 48) + 28))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(2)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(3)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(4)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(5)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(6)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(7)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(16)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(8)] * kernel_shared[(((((int)threadIdx.x) * 48) + 2))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(17)] * kernel_shared[(((((int)threadIdx.x) * 48) + 5))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(2)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(11)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(3)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(12)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(4)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(13)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(5)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(14)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(6)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(15)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(7)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(16)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(8)] * kernel_shared[(((((int)threadIdx.x) * 48) + 26))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(17)] * kernel_shared[(((((int)threadIdx.x) * 48) + 29))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(18)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(27)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(19)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(28)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 6))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 9))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(18)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(27)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(19)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(28)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 30))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 33))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(19)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(28)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(25)] * kernel_shared[(((((int)threadIdx.x) * 48) + 7))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(34)] * kernel_shared[(((((int)threadIdx.x) * 48) + 10))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(19)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(28)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(25)] * kernel_shared[(((((int)threadIdx.x) * 48) + 31))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(34)] * kernel_shared[(((((int)threadIdx.x) * 48) + 34))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(25)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(34)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(26)] * kernel_shared[(((((int)threadIdx.x) * 48) + 8))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(35)] * kernel_shared[(((((int)threadIdx.x) * 48) + 11))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(20)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(29)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(21)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(30)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(22)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(31)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(23)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(32)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(24)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(33)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(25)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(34)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(26)] * kernel_shared[(((((int)threadIdx.x) * 48) + 32))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(35)] * kernel_shared[(((((int)threadIdx.x) * 48) + 35))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(36)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(45)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(37)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(46)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 12))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 15))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(36)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(45)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(37)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(46)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 36))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 39))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(37)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(46)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(43)] * kernel_shared[(((((int)threadIdx.x) * 48) + 13))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(52)] * kernel_shared[(((((int)threadIdx.x) * 48) + 16))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(37)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(46)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(43)] * kernel_shared[(((((int)threadIdx.x) * 48) + 37))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(52)] * kernel_shared[(((((int)threadIdx.x) * 48) + 40))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(43)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(52)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(44)] * kernel_shared[(((((int)threadIdx.x) * 48) + 14))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(53)] * kernel_shared[(((((int)threadIdx.x) * 48) + 17))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(38)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(47)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(39)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(48)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(40)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(49)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(41)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(50)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(42)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(51)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(43)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(52)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(44)] * kernel_shared[(((((int)threadIdx.x) * 48) + 38))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(53)] * kernel_shared[(((((int)threadIdx.x) * 48) + 41))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(54)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(63)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(55)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(64)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 18))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 21))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(54)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(63)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(55)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(64)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 42))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 45))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(55)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(64)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(61)] * kernel_shared[(((((int)threadIdx.x) * 48) + 19))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(70)] * kernel_shared[(((((int)threadIdx.x) * 48) + 22))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(55)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(64)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(61)] * kernel_shared[(((((int)threadIdx.x) * 48) + 43))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(70)] * kernel_shared[(((((int)threadIdx.x) * 48) + 46))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(0)] = (compute1[(0)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(1)] = (compute1[(1)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(2)] = (compute1[(2)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(3)] = (compute1[(3)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(4)] = (compute1[(4)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(61)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(5)] = (compute1[(5)] + (pad_temp_shared[(70)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(62)] * kernel_shared[(((((int)threadIdx.x) * 48) + 20))]));
      compute1[(6)] = (compute1[(6)] + (pad_temp_shared[(71)] * kernel_shared[(((((int)threadIdx.x) * 48) + 23))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(56)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(7)] = (compute1[(7)] + (pad_temp_shared[(65)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(57)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(8)] = (compute1[(8)] + (pad_temp_shared[(66)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(58)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(9)] = (compute1[(9)] + (pad_temp_shared[(67)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(59)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(10)] = (compute1[(10)] + (pad_temp_shared[(68)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(60)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(11)] = (compute1[(11)] + (pad_temp_shared[(69)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(61)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(12)] = (compute1[(12)] + (pad_temp_shared[(70)] * kernel_shared[(((((int)threadIdx.x) * 48) + 47))]));
      compute1[(13)] = (compute1[(13)] + (pad_temp_shared[(62)] * kernel_shared[(((((int)threadIdx.x) * 48) + 44))]));
      compute1[(13)] = (compute1[(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((compute1[(((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


resume_search(task, log_file)

Out:

Resume search:
/usr/local/lib/python3.6/dist-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: ( 2 minutes 13.406 seconds)

Gallery generated by Sphinx-Gallery