Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Allow convtrans2d to use dynamic dimensions #639

Merged
merged 1 commit into from
Mar 30, 2023
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
76 changes: 37 additions & 39 deletions src/tensor_ops/convtrans2d/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -76,8 +76,10 @@ pub(super) trait ConvTrans2DKernel<E: Dtype>: DeviceStorage {
) -> Result<(), Self::Err>;
}

pub trait ConvTransAlgebra<const K: usize, const S: usize, const P: usize>: ConstDim {
type Convolved: ConstDim;
pub trait ConvTransAlgebra<const K: usize, const S: usize, const P: usize>: Dim {
type Convolved: Dim;

fn convolve_dim(&self) -> Self::Convolved;
}

impl<const D: usize, const K: usize, const S: usize, const P: usize> ConvTransAlgebra<K, S, P>
Expand All @@ -86,6 +88,18 @@ where
Const<{ D * S + K - S - 2 * P }>: Sized,
{
type Convolved = Const<{ D * S + K - S - 2 * P }>;

fn convolve_dim(&self) -> Self::Convolved {
Default::default()
}
}

impl<const K: usize, const S: usize, const P: usize> ConvTransAlgebra<K, S, P> for usize {
type Convolved = usize;

fn convolve_dim(&self) -> Self::Convolved {
(self * S + K).checked_sub(S + 2 * P).unwrap()
}
}

pub trait TryConvTrans2DTo<F, const S: usize, const P: usize>: HasErr {
Expand Down Expand Up @@ -118,40 +132,34 @@ impl<T, F> TryConvTrans2D<F> for T {}

impl<
const C: usize,
const H: usize,
const W: usize,
H: Dim + ConvTransAlgebra<K, S, P>,
W: Dim + ConvTransAlgebra<K, S, P>,
const O: usize,
const K: usize,
const S: usize,
const P: usize,
E: Dtype,
D: ConvTrans2DKernel<E> + ZerosTensor<E>,
T: 'static + Tape<E, D>,
> TryConvTrans2DTo<Tensor<Rank4<O, C, K, K>, E, D>, S, P> for Tensor<Rank3<C, H, W>, E, D, T>
where
Const<H>: ConvTransAlgebra<K, S, P>,
Const<W>: ConvTransAlgebra<K, S, P>,
> TryConvTrans2DTo<Tensor<Rank4<O, C, K, K>, E, D>, S, P>
for Tensor<(Const<C>, H, W), E, D, T>
{
type Output = Tensor<
(
Const<O>,
<Const<H> as ConvTransAlgebra<K, S, P>>::Convolved,
<Const<W> as ConvTransAlgebra<K, S, P>>::Convolved,
),
E,
D,
T,
>;
type Output = Tensor<(Const<O>, H::Convolved, W::Convolved), E, D, T>;

fn try_convtrans2d_to(
self,
filters: Tensor<Rank4<O, C, K, K>, E, D>,
) -> Result<Self::Output, Self::Err> {
let op = ConvTrans2DOp::new(S, P, K, [1, C, H, W], O);
let h = self.shape.1;
let w = self.shape.2;

let op = ConvTrans2DOp::new(S, P, K, [1, C, h.size(), w.size()], O);
let (lhs, ltape) = self.split_tape();
let (rhs, rtape) = filters.split_tape();
let mut tape = ltape.merge(rtape);
let mut out = lhs.device.try_zeros()?;
let mut out = lhs
.device
.try_zeros_like(&(Const, h.convolve_dim(), w.convolve_dim()))?;
lhs.device.forward(op, &lhs, &rhs, &mut out)?;
let phantom_out = out.clone();
tape.try_alloc_grad(&lhs)?;
Expand All @@ -169,8 +177,8 @@ where
impl<
B: Dim,
const C: usize,
const H: usize,
const W: usize,
H: Dim + ConvTransAlgebra<K, S, P>,
W: Dim + ConvTransAlgebra<K, S, P>,
const O: usize,
const K: usize,
const S: usize,
Expand All @@ -179,33 +187,23 @@ impl<
D: ConvTrans2DKernel<E> + ZerosTensor<E>,
T: 'static + Tape<E, D>,
> TryConvTrans2DTo<Tensor<Rank4<O, C, K, K>, E, D>, S, P>
for Tensor<(B, Const<C>, Const<H>, Const<W>), E, D, T>
where
Const<H>: ConvTransAlgebra<K, S, P>,
Const<W>: ConvTransAlgebra<K, S, P>,
for Tensor<(B, Const<C>, H, W), E, D, T>
{
type Output = Tensor<
(
B,
Const<O>,
<Const<H> as ConvTransAlgebra<K, S, P>>::Convolved,
<Const<W> as ConvTransAlgebra<K, S, P>>::Convolved,
),
E,
D,
T,
>;
type Output = Tensor<(B, Const<O>, H::Convolved, W::Convolved), E, D, T>;
fn try_convtrans2d_to(
self,
filters: Tensor<Rank4<O, C, K, K>, E, D>,
) -> Result<Self::Output, Self::Err> {
let h = self.shape.2;
let w = self.shape.3;

let batch = self.shape().0;
let op = ConvTrans2DOp::new(S, P, K, [batch.size(), C, H, W], O);
let op = ConvTrans2DOp::new(S, P, K, [batch.size(), C, h.size(), w.size()], O);
let (lhs, ltape) = self.split_tape();
let (rhs, rtape) = filters.split_tape();
let mut out =
lhs.device
.try_zeros_like(&(batch, Const, Default::default(), Default::default()))?;
.try_zeros_like(&(batch, Const, h.convolve_dim(), w.convolve_dim()))?;
let mut tape = ltape.merge(rtape);
lhs.device.forward(op, &lhs, &rhs, &mut out)?;
let phantom_out = out.clone();
Expand Down