fixed the bug described in Issue#30 by rewrite backward kernels

[kawabata-tomoko]

modified: csrc/flashfftconv/conv1d/conv1d_bwd_cuda_bhl.cu
modified: csrc/flashfftconv/conv1d/conv1d_bwd_cuda_blh.cu

And could passed this sample script in Issue#30:

import torch
import torch.nn as nn
import torch.optim as optim
from flashfftconv import FlashDepthWiseConv1d
B=4
L=26000
d=512
k=3
padding=k-1
dtype=torch.bfloat16
device="cuda:4"
# set up PyTorch equivalent to get the weights
# in_channels = out_channels, and kernel size must be odd
x=torch.randn((B,L,d),device=device,dtype=dtype)
trans = nn.Linear(d,d).to(device=device,dtype=dtype)
conv1d_torch = nn.Conv1d(
    in_channels = d,
    out_channels = d,
    kernel_size = k,
    groups = d,
    padding = padding,
    dtype = dtype,
    device=device
)

flash_conv1d = FlashDepthWiseConv1d(
    channels = d,
    kernel_size=k,
    padding=padding,
    weights=conv1d_torch.weight,
    bias=conv1d_torch.bias,
    is_bhl=True,#or False
    dtype = dtype # this should be the dtype of the weights
).to(device=device)

x_input=trans(x).transpose(-1,-2).contiguous()
x_input_flash = x_input.detach().clone().requires_grad_(True)
x_input_torch = x_input.detach().clone().requires_grad_(True)

out_torch = conv1d_torch(x_input_torch)
out_flash = flash_conv1d(x_input_flash)
# out_flash = flash_conv1d(x_input_flash.transpose(-1, -2).contiguous())

out_flash.sum().backward()
out_torch.sum().backward()

print((x_input_flash.grad[0]-x_input_torch.grad[0]).sum())