submission 515187

#!POPCORN leaderboard conv2d_v2
#!POPCORN gpu L4

from task import input_t, output_t
import torch
import torch.nn.functional as F


def custom_kernel(data: input_t) -> output_t:
    """
    Implementation of 2D convolution using PyTorch with no padding and no striding.
    Args:
        data: Tuple of (input tensor, kernel tensor)
        spec: Convolution specifications
    Returns:
        Output tensor after convolution
    """
    input_tensor, kernel, output = data

    # Avoid TensorFloat-32 accumulation drift vs. reference implementation.
    prev_matmul_tf32 = torch.backends.cuda.matmul.allow_tf32
    prev_cudnn_tf32 = torch.backends.cudnn.allow_tf32
    torch.backends.cuda.matmul.allow_tf32 = False
    torch.backends.cudnn.allow_tf32 = False
    try:
        output[...] = F.conv2d(input_tensor, kernel, stride=1, padding=0)
    finally:
        torch.backends.cuda.matmul.allow_tf32 = prev_matmul_tf32
        torch.backends.cudnn.allow_tf32 = prev_cudnn_tf32
    return output