use ruda_kernel::dsl as kernel_dsl;
use ruda_kernel::dsl::Runtime;
use crate::pooling::pool2d::Position;
use crate::pooling::pool2d::view4d;
use ruda_kernel::tensor::layout::address_type;
use ruda_kernel::tensor::layout::decompose_linear;
use ruda_kernel::tensor::layout::shape_divmod;
use ruda_kernel::tensor::layout::max_vector_size;
use ruda_kernel::tensor::allocation::empty_device_dtype;
use ruda_kernel::tensor::permutation::permute_nchw_to_nhwc;
use ruda_kernel::tensor::permutation::permute_nhwc_to_nchw;
use ruda_kernel::tensor::RudaTensor;
use ruda_core::tensor::Shape;
use ruda_kernel::dsl::calculate_ruda_count_elemwise;
use ruda_kernel::dsl::num_traits::Zero;
use ruda_kernel::dsl::prelude::*;
use ruda_kernel::library::FastDivmod;
use ruda_kernel::library::tensor::View;
#[derive(RudaLaunch, RudaType)]
pub struct PoolBackwardArgs {
pub stride_0: i32,
pub stride_1: i32,
pub dilation_0: i32,
pub dilation_1: i32,
pub padding_0: i32,
pub padding_1: i32,
}
#[ruda(launch_unchecked, address_type = "dynamic")]
fn avg_pool2d_backward_kernel<E: Numeric, N: Size>(
grad: &Tensor<Vector<E, N>>,
output: &mut View<Vector<E, N>, Position, ReadWrite>,
out_shape: Sequence<FastDivmod<usize>>,
working_units: usize,
args: &PoolBackwardArgs,
#[comptime] kernel_size_0: i32,
#[comptime] kernel_size_1: i32,
#[comptime] count_include_pad: bool,
#[comptime] divisor_override: i64,
#[define(E)] _dtype: StorageType,
) {
if ABSOLUTE_POS >= working_units {
terminate!();
}
let vector_size = grad.vector_size();
let (_, pos) = decompose_linear(ABSOLUTE_POS * output.vector_size(), &out_shape);
let [batch, ih, iw, channel] = *pos else {
unreachable!()
};
let mut grad_acc = Vector::zero();
let (oh_start, oh_end, ow_start, ow_end) = loop_ranges(
ih as i32,
iw as i32,
grad.shape(1) as u32,
grad.shape(2) as u32,
args,
kernel_size_0,
kernel_size_1,
);
let padding_0 = args.padding_0 as u32;
let padding_1 = args.padding_1 as u32;
let stride_0 = args.stride_0 as u32;
let stride_1 = args.stride_1 as u32;
let kernel_size_0 = comptime![kernel_size_0 as u32];
let kernel_size_1 = comptime![kernel_size_1 as u32];
let index_base = batch * grad.stride(0) + channel * grad.stride(3);
let border_bottom = output.shape().1 as u32 + padding_0;
let border_right = output.shape().2 as u32 + padding_1;
let begin_h = ih as u32 + padding_0;
let begin_w = iw as u32 + padding_1;
for oh in oh_start..oh_end {
let ih_start = oh * stride_0;
let ih_end = clamp_max(ih_start + kernel_size_0, border_bottom);
let ih_start = clamp_min(ih_start, padding_0);
if begin_h >= ih_start && begin_h < ih_end {
for ow in ow_start..ow_end {
let index =
index_base + oh as usize * grad.stride(1) + ow as usize * grad.stride(2);
let iw_start = ow * stride_1;
let iw_end = clamp_max(iw_start + kernel_size_1, border_right);
let iw_start = clamp_min(iw_start, padding_1);
if begin_w >= iw_start && begin_w < iw_end {
if comptime![divisor_override != 0] {
grad_acc += grad[index / vector_size] / Vector::cast_from(divisor_override);
} else if count_include_pad {
let padded_h_end = clamp_max(oh * stride_0 + kernel_size_0, border_bottom + padding_0);
let padded_w_end = clamp_max(ow * stride_1 + kernel_size_1, border_right + padding_1);
let count = (padded_h_end - oh * stride_0) * (padded_w_end - ow * stride_1);
grad_acc += grad[index / vector_size]
/ Vector::cast_from(count);
} else {
let ih_diff = ih_end - ih_start;
let iw_diff = iw_end - iw_start;
let count = Vector::cast_from(ih_diff * iw_diff);
grad_acc += grad[index / vector_size] / count;
}
}
}
}
}
output[(batch, ih, iw, channel)] = grad_acc;
}
#[ruda]
fn loop_ranges(
ih: i32,
iw: i32,
grad_h: u32,
grad_w: u32,
args: &PoolBackwardArgs,
#[comptime] kernel_size_0: i32,
#[comptime] kernel_size_1: i32,
) -> (u32, u32, u32, u32) {
let h = (ih + args.padding_0) as u32;
let w = (iw + args.padding_1) as u32;
let kh = comptime![kernel_size_0 as u32];
let kw = comptime![kernel_size_1 as u32];
let sh = args.stride_0 as u32;
let sw = args.stride_1 as u32;
let oh_start = if h >= kh { (h - kh) / sh + 1 } else { 0u32.runtime() };
let ow_start = if w >= kw { (w - kw) / sw + 1 } else { 0u32.runtime() };
let oh_end = clamp_max(h / sh + 1, grad_h);
let ow_end = clamp_max(w / sw + 1, grad_w);
(oh_start, oh_end, ow_start, ow_end)
}
pub fn avg_pool2d_backward<R: Runtime>(
x: RudaTensor<R>,
grad: RudaTensor<R>,
kernel_size: [usize; 2],
stride: [usize; 2],
padding: [usize; 2],
count_include_pad: bool,
ceil_mode: bool,
) -> RudaTensor<R> {
avg_pool2d_backward_with_divisor(x, grad, kernel_size, stride, padding, count_include_pad, ceil_mode, None)
}
pub fn avg_pool2d_backward_with_divisor<R: Runtime>(
x: RudaTensor<R>,
grad: RudaTensor<R>,
kernel_size: [usize; 2],
stride: [usize; 2],
padding: [usize; 2],
count_include_pad: bool,
_ceil_mode: bool,
divisor_override: Option<i64>,
) -> RudaTensor<R> {
assert_ne!(divisor_override, Some(0), "average pooling divisor must be nonzero");
let [batches, channels, height, width] = x.meta.shape().dims();
let grad = permute_nchw_to_nhwc(grad);
let vector_size = if x.meta.strides()[3] == grad.meta.strides()[3] {
max_vector_size(&x)
} else {
1
};
let dilation = 1;
let out_shape = Shape::new([batches, height, width, channels]);
let output = empty_device_dtype(x.client.clone(), x.device.clone(), out_shape, x.dtype);
let working_units = output.meta.num_elements() / vector_size as usize;
let ruda_dim = RudaDim::new(x.client.properties(), working_units);
let ruda_count = calculate_ruda_count_elemwise(&x.client, working_units, ruda_dim);
unsafe {
avg_pool2d_backward_kernel::launch_unchecked(
&output.client,
ruda_count,
ruda_dim,
address_type!(grad, output),
vector_size,
grad.into_tensor_arg(),
view4d(output.clone(), vector_size),
shape_divmod(&output),
working_units,
PoolBackwardArgsLaunch::new(
stride[0] as i32,
stride[1] as i32,
dilation,
dilation,
padding[0] as i32,
padding[1] as i32,
),
kernel_size[0] as i32,
kernel_size[1] as i32,
count_include_pad,
divisor_override.unwrap_or(0),
output.dtype.into(),
)
};
permute_nhwc_to_nchw(output)
}