huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 10:38:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #27 from LaurentMazare/layout-refactor

Browse Source 
Refactor the stride/shape handling
This commit is contained in:
Laurent Mazare
2023-06-28 15:59:53 +01:00
committed by GitHub
parent 8b4b2d1830 6c9e6b5a99
commit 0cfa21f26a
10 changed files with 518 additions and 487 deletions
Download Patch File Download Diff File Expand all files Collapse all files

315
candle-core/src/cpu_backend.rs
View File

@ -1,5 +1,5 @@
use crate::op::{BinaryOp, UnaryOp};
use crate::{DType, Error, Result, Shape, StridedIndex};
use crate::{DType, Error, Layout, Result, Shape};
use gemm::{gemm, Parallelism};
use half::{bf16, f16};
@ -18,31 +18,31 @@ pub enum CpuStorage {
fn wcond<T: Copy>(
pred: &[u32],
shape: &Shape,
stride: &[usize],
layout: &Layout,
t: &[T],
stride_t: &[usize],
layout_t: &Layout,
f: &[T],
stride_f: &[usize],
layout_f: &Layout,
) -> Vec<T> {
if shape.is_contiguous(stride) && shape.is_contiguous(stride_t) && shape.is_contiguous(stride_f)
{
let elem_count = shape.elem_count();
let pred = &pred[..elem_count];
let t = &t[..elem_count];
let f = &f[..elem_count];
pred.iter()
.zip(t.iter().zip(f.iter()))
.map(|(&p, (&t, &f))| if p > 0 { t } else { f })
.collect::<Vec<_>>()
} else {
let dims = shape.dims();
let it_p = StridedIndex::new(dims, stride);
let it_t = StridedIndex::new(dims, stride_t);
let it_f = StridedIndex::new(dims, stride_f);
it_p.zip(it_t.zip(it_f))
match (
layout.contiguous_offsets(),
layout_t.contiguous_offsets(),
layout_f.contiguous_offsets(),
) {
(Some((o1, o2)), Some((o_t1, o_t2)), Some((o_f1, o_f2))) => {
let pred = &pred[o1..o2];
let t = &t[o_t1..o_t2];
let f = &f[o_f1..o_f2];
pred.iter()
.zip(t.iter().zip(f.iter()))
.map(|(&p, (&t, &f))| if p > 0 { t } else { f })
.collect::<Vec<_>>()
}
_ => layout
.strided_index()
.zip(layout_t.strided_index().zip(layout_f.strided_index()))
.map(|(i_p, (i_t, i_f))| if pred[i_p] > 0 { t[i_t] } else { f[i_f] })
.collect::<Vec<_>>()
.collect::<Vec<_>>(),
}
}
@ -62,64 +62,50 @@ macro_rules! map1 {
fn sum_impl1<T: Copy + num_traits::NumAssign>(
src: &[T],
dst_shape: &Shape,
src_dims: &[usize],
stride: &[usize],
src_layout: &Layout,
to_dst_index: impl Fn(usize) -> usize,
) -> Result<Vec<T>> {
let mut dst = vec![T::zero(); dst_shape.elem_count()];
for (unstr_index, src_index) in StridedIndex::new(src_dims, stride).enumerate() {
for (unstr_index, src_index) in src_layout.strided_index().enumerate() {
dst[to_dst_index(unstr_index)] += src[src_index];
}
Ok(dst)
}
fn unary_map<T: Copy, U: Copy, F: FnMut(T) -> U>(
vs: &[T],
shape: &Shape,
stride: &[usize],
mut f: F,
) -> Vec<U> {
if shape.is_contiguous(stride) {
vs[..shape.elem_count()].iter().map(|&v| f(v)).collect()
} else {
StridedIndex::new(shape.dims(), stride)
.map(|i| f(vs[i]))
.collect()
fn unary_map<T: Copy, U: Copy, F: FnMut(T) -> U>(vs: &[T], layout: &Layout, mut f: F) -> Vec<U> {
match layout.contiguous_offsets() {
Some((o1, o2)) => vs[o1..o2].iter().map(|&v| f(v)).collect(),
None => layout.strided_index().map(|i| f(vs[i])).collect(),
}
}
// This function maps over two strided index sequences.
fn binary_map<T: Copy, F: FnMut(T, T) -> T>(
shape: &Shape,
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
lhs: &[T],
rhs: &[T],
mut f: F,
) -> Vec<T> {
let dims = shape.dims();
if shape.is_contiguous(lhs_stride) && shape.is_contiguous(rhs_stride) {
(0..shape.elem_count()).map(|i| f(lhs[i], rhs[i])).collect()
} else {
let lhs_index = StridedIndex::new(dims, lhs_stride);
let rhs_index = StridedIndex::new(dims, rhs_stride);
lhs_index
.zip(rhs_index)
match (lhs_l.contiguous_offsets(), rhs_l.contiguous_offsets()) {
(Some((o_l1, o_l2)), Some((o_r1, o_r2))) => lhs[o_l1..o_l2]
.iter()
.zip(rhs[o_r1..o_r2].iter())
.map(|(&l, &r)| f(l, r))
.collect(),
_ => lhs_l
.strided_index()
.zip(rhs_l.strided_index())
.map(|(lhs_i, rhs_i)| f(lhs[lhs_i], rhs[rhs_i]))
.collect()
.collect(),
}
}
fn take_impl1<T: Copy>(
vs: &[T],
ids: &[u32],
shape: &Shape,
stride: &[usize],
vocab_size: usize,
hidden_size: usize,
) -> Result<Vec<T>> {
let mut values = Vec::with_capacity(shape.elem_count() * hidden_size);
for index in StridedIndex::new(shape.dims(), stride) {
fn take_impl1<T: Copy>(vs: &[T], ids: &[u32], layout: &Layout, rhs_l: &Layout) -> Result<Vec<T>> {
// TODO: Optimize for the case where ids are contiguous.
let (vocab_size, hidden_size) = rhs_l.shape().r2()?;
let mut values = Vec::with_capacity(layout.shape().elem_count() * hidden_size);
for index in layout.strided_index() {
let index = ids[index].try_into()?;
if index >= vocab_size {
return Err(Error::InvalidIndex {
@ -138,37 +124,40 @@ fn copy_strided_src_<T: Copy + std::fmt::Display>(
src: &[T],
dst: &mut [T],
dst_offset: usize,
src_shape: &Shape,
src_stride: &[usize],
src_offset: usize,
src_l: &Layout,
) {
let src = &src[src_offset..];
if src_shape.is_contiguous(src_stride) {
let elem_to_copy = (dst.len() - dst_offset).min(src.len());
dst[dst_offset..dst_offset + elem_to_copy].copy_from_slice(&src[..elem_to_copy])
} else {
let src_indexes = StridedIndex::new(src_shape.dims(), src_stride);
for (dst_index, src_index) in src_indexes.enumerate() {
let dst_index = dst_index + dst_offset;
if dst_index >= dst.len() {
break;
match src_l.contiguous_offsets() {
Some((o_dst1, o_dst2)) => {
let elem_to_copy = (dst.len() - dst_offset).min(o_dst2 - o_dst1);
dst[dst_offset..dst_offset + elem_to_copy].copy_from_slice(&src[o_dst1..o_dst2])
}
None => {
for (dst_index, src_index) in src_l.strided_index().enumerate() {
let dst_index = dst_index + dst_offset;
if dst_index >= dst.len() {
break;
}
dst[dst_index] = src[src_index]
}
dst[dst_index] = src[src_index]
}
}
}
fn matmul_impl<T: 'static + num_traits::Num + Copy>(
fn matmul<T: 'static + num_traits::Num + Copy>(
lhs: &[T],
rhs: &[T],
(b, m, n, k): (usize, usize, usize, usize),
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<Vec<T>> {
let lhs = &lhs[lhs_l.start_offset()..];
let rhs = &rhs[rhs_l.start_offset()..];
let a_skip: usize = m * k;
let b_skip: usize = n * k;
let c_skip: usize = m * n;
let lhs_stride = lhs_l.stride();
let rhs_stride = rhs_l.stride();
let rank = lhs_stride.len();
let lhs_cs = lhs_stride[rank - 1];
let lhs_rs = lhs_stride[rank - 2];
@ -238,118 +227,114 @@ impl CpuStorage {
D::cpu_storage_as_slice(self)
}
pub fn as_mut_slice<D: crate::WithDType>(&mut self) -> Result<&mut [D]> {
D::cpu_storage_as_mut_slice(self)
}
pub(crate) fn to_dtype(&self, shape: &Shape, stride: &[usize], dtype: DType) -> Result<Self> {
pub(crate) fn to_dtype(&self, layout: &Layout, dtype: DType) -> Result<Self> {
// TODO: find a way around the quadratic number of cases below.
match (self, dtype) {
(Self::U32(storage), DType::BF16) => {
let data = unary_map(storage, shape, stride, |v| bf16::from_f32(v as f32));
let data = unary_map(storage, layout, |v| bf16::from_f32(v as f32));
Ok(Self::BF16(data))
}
(Self::BF16(storage), DType::BF16) => {
let data = unary_map(storage, shape, stride, |v| v);
let data = unary_map(storage, layout, |v| v);
Ok(Self::BF16(data))
}
(Self::F16(storage), DType::BF16) => {
let data = unary_map(storage, shape, stride, |v| bf16::from_f32(v.to_f32()));
let data = unary_map(storage, layout, |v| bf16::from_f32(v.to_f32()));
Ok(Self::BF16(data))
}
(Self::F32(storage), DType::BF16) => {
let data = unary_map(storage, shape, stride, bf16::from_f32);
let data = unary_map(storage, layout, bf16::from_f32);
Ok(Self::BF16(data))
}
(Self::F64(storage), DType::BF16) => {
let data = unary_map(storage, shape, stride, bf16::from_f64);
let data = unary_map(storage, layout, bf16::from_f64);
Ok(Self::BF16(data))
}
(Self::U32(storage), DType::F16) => {
let data = unary_map(storage, shape, stride, |v| f16::from_f32(v as f32));
let data = unary_map(storage, layout, |v| f16::from_f32(v as f32));
Ok(Self::F16(data))
}
(Self::BF16(storage), DType::F16) => {
let data = unary_map(storage, shape, stride, |v| f16::from_f32(v.to_f32()));
let data = unary_map(storage, layout, |v| f16::from_f32(v.to_f32()));
Ok(Self::F16(data))
}
(Self::F16(storage), DType::F16) => {
let data = unary_map(storage, shape, stride, |v| v);
let data = unary_map(storage, layout, |v| v);
Ok(Self::F16(data))
}
(Self::F32(storage), DType::F16) => {
let data = unary_map(storage, shape, stride, f16::from_f32);
let data = unary_map(storage, layout, f16::from_f32);
Ok(Self::F16(data))
}
(Self::F64(storage), DType::F16) => {
let data = unary_map(storage, shape, stride, f16::from_f64);
let data = unary_map(storage, layout, f16::from_f64);
Ok(Self::F16(data))
}
(Self::U32(storage), DType::F32) => {
let data = unary_map(storage, shape, stride, |v| v as f32);
let data = unary_map(storage, layout, |v| v as f32);
Ok(Self::F32(data))
}
(Self::BF16(storage), DType::F32) => {
let data = unary_map(storage, shape, stride, |v| v.to_f32());
let data = unary_map(storage, layout, |v| v.to_f32());
Ok(Self::F32(data))
}
(Self::F16(storage), DType::F32) => {
let data = unary_map(storage, shape, stride, |v| v.to_f32());
let data = unary_map(storage, layout, |v| v.to_f32());
Ok(Self::F32(data))
}
(Self::F32(storage), DType::F32) => {
let data = unary_map(storage, shape, stride, |v| v);
let data = unary_map(storage, layout, |v| v);
Ok(Self::F32(data))
}
(Self::F64(storage), DType::F32) => {
let data = unary_map(storage, shape, stride, |v| v as f32);
let data = unary_map(storage, layout, |v| v as f32);
Ok(Self::F32(data))
}
(Self::U32(storage), DType::U32) => {
let data = unary_map(storage, shape, stride, |v| v);
let data = unary_map(storage, layout, |v| v);
Ok(Self::U32(data))
}
(Self::BF16(storage), DType::U32) => {
let data = unary_map(storage, shape, stride, |v| v.to_f32() as u32);
let data = unary_map(storage, layout, |v| v.to_f32() as u32);
Ok(Self::U32(data))
}
(Self::F16(storage), DType::U32) => {
let data = unary_map(storage, shape, stride, |v| v.to_f32() as u32);
let data = unary_map(storage, layout, |v| v.to_f32() as u32);
Ok(Self::U32(data))
}
(Self::F32(storage), DType::U32) => {
let data = unary_map(storage, shape, stride, |v| v as u32);
let data = unary_map(storage, layout, |v| v as u32);
Ok(Self::U32(data))
}
(Self::F64(storage), DType::U32) => {
let data = unary_map(storage, shape, stride, |v| v as u32);
let data = unary_map(storage, layout, |v| v as u32);
Ok(Self::U32(data))
}
(Self::U32(storage), DType::F64) => {
let data = unary_map(storage, shape, stride, |v| v as f64);
let data = unary_map(storage, layout, |v| v as f64);
Ok(Self::F64(data))
}
(Self::BF16(storage), DType::F64) => {
let data = unary_map(storage, shape, stride, |v| v.to_f64());
let data = unary_map(storage, layout, |v| v.to_f64());
Ok(Self::F64(data))
}
(Self::F16(storage), DType::F64) => {
let data = unary_map(storage, shape, stride, |v| v.to_f64());
let data = unary_map(storage, layout, |v| v.to_f64());
Ok(Self::F64(data))
}
(Self::F32(storage), DType::F64) => {
let data = unary_map(storage, shape, stride, |v| v as f64);
let data = unary_map(storage, layout, |v| v as f64);
Ok(Self::F64(data))
}
(Self::F64(storage), DType::F64) => {
let data = unary_map(storage, shape, stride, |v| v);
let data = unary_map(storage, layout, |v| v);
Ok(Self::F64(data))
}
}
}
pub(crate) fn sum(&self, shape: &Shape, stride: &[usize], sum_dims: &[usize]) -> Result<Self> {
let src_dims = shape.dims();
pub(crate) fn sum(&self, layout: &Layout, sum_dims: &[usize]) -> Result<Self> {
let src_dims = layout.dims();
let mut dst_dims = src_dims.to_vec();
for &sum_dim in sum_dims.iter() {
dst_dims[sum_dim] = 1;
@ -375,7 +360,7 @@ impl CpuStorage {
dst_index
};
// TODO: Maybe provide an implementation with higher precision accumulators?
map1!(self, sum_impl1, &dst_shape, src_dims, stride, to_dst_index)
map1!(self, sum_impl1, &dst_shape, layout, to_dst_index)
}
pub(crate) fn divide_by_sum_over_dim(&mut self, shape: &Shape, dim: usize) -> Result<()> {
@ -461,65 +446,59 @@ impl CpuStorage {
Ok(())
}
pub(crate) fn affine_impl(
&self,
shape: &Shape,
stride: &[usize],
mul: f64,
add: f64,
) -> Result<Self> {
pub(crate) fn affine(&self, layout: &Layout, mul: f64, add: f64) -> Result<Self> {
match self {
Self::U32(storage) => {
let mul = mul as u32;
let add = add as u32;
let data = unary_map(storage, shape, stride, |v| v * mul + add);
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::U32(data))
}
Self::BF16(storage) => {
let mul = bf16::from_f64(mul);
let add = bf16::from_f64(add);
let data = unary_map(storage, shape, stride, |v| v * mul + add);
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::BF16(data))
}
Self::F16(storage) => {
let mul = f16::from_f64(mul);
let add = f16::from_f64(add);
let data = unary_map(storage, shape, stride, |v| v * mul + add);
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::F16(data))
}
Self::F32(storage) => {
let mul = mul as f32;
let add = add as f32;
let data = unary_map(storage, shape, stride, |v| v * mul + add);
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::F32(data))
}
Self::F64(storage) => {
let data = unary_map(storage, shape, stride, |v| v * mul + add);
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::F64(data))
}
}
}
pub(crate) fn unary_impl<B: UnaryOp>(&self, shape: &Shape, stride: &[usize]) -> Result<Self> {
pub(crate) fn unary_impl<B: UnaryOp>(&self, layout: &Layout) -> Result<Self> {
match self {
Self::BF16(storage) => {
let data = unary_map(storage, shape, stride, B::bf16);
let data = unary_map(storage, layout, B::bf16);
Ok(Self::BF16(data))
}
Self::F16(storage) => {
let data = unary_map(storage, shape, stride, B::f16);
let data = unary_map(storage, layout, B::f16);
Ok(Self::F16(data))
}
Self::F32(storage) => {
let data = unary_map(storage, shape, stride, B::f32);
let data = unary_map(storage, layout, B::f32);
Ok(Self::F32(data))
}
Self::F64(storage) => {
let data = unary_map(storage, shape, stride, B::f64);
let data = unary_map(storage, layout, B::f64);
Ok(Self::F64(data))
}
Self::U32(storage) => {
let data = unary_map(storage, shape, stride, B::u32);
let data = unary_map(storage, layout, B::u32);
Ok(Self::U32(data))
}
}
@ -528,29 +507,28 @@ impl CpuStorage {
pub(crate) fn binary_impl<B: BinaryOp>(
&self,
rhs: &Self,
shape: &Shape,
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<Self> {
match (self, rhs) {
(Self::BF16(lhs), Self::BF16(rhs)) => {
let data = binary_map(shape, lhs_stride, rhs_stride, lhs, rhs, B::bf16);
let data = binary_map(lhs_l, rhs_l, lhs, rhs, B::bf16);
Ok(Self::BF16(data))
}
(Self::F16(lhs), Self::F16(rhs)) => {
let data = binary_map(shape, lhs_stride, rhs_stride, lhs, rhs, B::f16);
let data = binary_map(lhs_l, rhs_l, lhs, rhs, B::f16);
Ok(Self::F16(data))
}
(Self::F32(lhs), Self::F32(rhs)) => {
let data = binary_map(shape, lhs_stride, rhs_stride, lhs, rhs, B::f32);
let data = binary_map(lhs_l, rhs_l, lhs, rhs, B::f32);
Ok(Self::F32(data))
}
(Self::F64(lhs), Self::F64(rhs)) => {
let data = binary_map(shape, lhs_stride, rhs_stride, lhs, rhs, B::f64);
let data = binary_map(lhs_l, rhs_l, lhs, rhs, B::f64);
Ok(Self::F64(data))
}
(Self::U32(lhs), Self::U32(rhs)) => {
let data = binary_map(shape, lhs_stride, rhs_stride, lhs, rhs, B::u32);
let data = binary_map(lhs_l, rhs_l, lhs, rhs, B::u32);
Ok(Self::U32(data))
}
_ => {
@ -568,29 +546,14 @@ impl CpuStorage {
&self,
dst: &mut Self,
dst_offset: usize,
src_shape: &Shape,
src_stride: &[usize],
src_offset: usize,
src_l: &Layout,
) -> Result<()> {
if src_shape.rank() != src_stride.len() {
panic!("incoherent shape and strides {src_shape:?} {src_stride:?}")
}
match (self, dst) {
(Self::U32(src), Self::U32(dst)) => {
copy_strided_src_(src, dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::BF16(src), Self::BF16(dst)) => {
copy_strided_src_(src, dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::F16(src), Self::F16(dst)) => {
copy_strided_src_(src, dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::F32(src), Self::F32(dst)) => {
copy_strided_src_(src, dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::F64(src), Self::F64(dst)) => {
copy_strided_src_(src, dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::U32(src), Self::U32(dst)) => copy_strided_src_(src, dst, dst_offset, src_l),
(Self::BF16(src), Self::BF16(dst)) => copy_strided_src_(src, dst, dst_offset, src_l),
(Self::F16(src), Self::F16(dst)) => copy_strided_src_(src, dst, dst_offset, src_l),
(Self::F32(src), Self::F32(dst)) => copy_strided_src_(src, dst, dst_offset, src_l),
(Self::F64(src), Self::F64(dst)) => copy_strided_src_(src, dst, dst_offset, src_l),
(_, dst) => {
// This should be covered by the dtype check above.
return Err(Error::DTypeMismatchBinaryOp {
@ -605,34 +568,33 @@ impl CpuStorage {
pub(crate) fn where_cond(
&self,
shape: &Shape,
stride: &[usize],
layout: &Layout,
t: &Self,
stride_t: &[usize],
layout_t: &Layout,
f: &Self,
stride_f: &[usize],
layout_f: &Layout,
) -> Result<Self> {
// TODO: Support types that could be casted to a boolean.
let pred = self.as_slice::<u32>()?;
match (t, f) {
(Self::BF16(t), Self::BF16(f)) => {
let data = wcond(pred, shape, stride, t, stride_t, f, stride_f);
let data = wcond(pred, layout, t, layout_t, f, layout_f);
Ok(Self::BF16(data))
}
(Self::F16(t), Self::F16(f)) => {
let data = wcond(pred, shape, stride, t, stride_t, f, stride_f);
let data = wcond(pred, layout, t, layout_t, f, layout_f);
Ok(Self::F16(data))
}
(Self::F32(t), Self::F32(f)) => {
let data = wcond(pred, shape, stride, t, stride_t, f, stride_f);
let data = wcond(pred, layout, t, layout_t, f, layout_f);
Ok(Self::F32(data))
}
(Self::F64(t), Self::F64(f)) => {
let data = wcond(pred, shape, stride, t, stride_t, f, stride_f);
let data = wcond(pred, layout, t, layout_t, f, layout_f);
Ok(Self::F64(data))
}
(Self::U32(t), Self::U32(f)) => {
let data = wcond(pred, shape, stride, t, stride_t, f, stride_f);
let data = wcond(pred, layout, t, layout_t, f, layout_f);
Ok(Self::U32(data))
}
_ => Err(Error::DTypeMismatchBinaryOp {
@ -643,36 +605,29 @@ impl CpuStorage {
}
}
pub(crate) fn embedding_impl(
&self,
shape: &Shape,
stride: &[usize],
vs: &Self,
hidden_size: usize,
vocab_size: usize,
) -> Result<Self> {
pub(crate) fn embedding(&self, layout: &Layout, rhs: &Self, rhs_l: &Layout) -> Result<Self> {
let ids = self.as_slice::<u32>()?;
map1!(vs, take_impl1, ids, shape, stride, vocab_size, hidden_size)
map1!(rhs, take_impl1, ids, layout, rhs_l)
}
pub(crate) fn matmul_impl(
pub(crate) fn matmul(
&self,
rhs: &Self,
bmnk: (usize, usize, usize, usize),
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<Self> {
match (self, rhs) {
(CpuStorage::F16(lhs), CpuStorage::F16(rhs)) => {
let dst = matmul_impl(lhs, rhs, bmnk, lhs_stride, rhs_stride)?;
let dst = matmul(lhs, rhs, bmnk, lhs_l, rhs_l)?;
Ok(Self::F16(dst))
}
(CpuStorage::F32(lhs), CpuStorage::F32(rhs)) => {
let dst = matmul_impl(lhs, rhs, bmnk, lhs_stride, rhs_stride)?;
let dst = matmul(lhs, rhs, bmnk, lhs_l, rhs_l)?;
Ok(Self::F32(dst))
}
(CpuStorage::F64(lhs), CpuStorage::F64(rhs)) => {
let dst = matmul_impl(lhs, rhs, bmnk, lhs_stride, rhs_stride)?;
let dst = matmul(lhs, rhs, bmnk, lhs_l, rhs_l)?;
Ok(Self::F64(dst))
}
_ => Err(Error::DTypeMismatchBinaryOp {

190
candle-core/src/cuda_backend.rs
View File

@ -1,4 +1,4 @@
use crate::{CpuStorage, DType, Shape};
use crate::{CpuStorage, DType, Layout, Shape};
use candle_kernels as kernels;
use cudarc::cublas::{Gemm, GemmConfig, StridedBatchedConfig};
use cudarc::driver::{CudaFunction, CudaSlice, DeviceSlice, LaunchAsync, LaunchConfig};
@ -26,6 +26,9 @@ pub enum CudaError {
#[error("internal error '{0}'")]
InternalError(&'static str),
#[error("internal error '{0}'")]
WrappedError(Box<dyn std::error::Error + 'static + std::marker::Send + std::marker::Sync>),
#[error("matmul is only supported for contiguous tensors lstride: {lhs_stride:?} rstride: {rhs_stride:?} mnk: {mnk:?}")]
MatMulNonContiguous {
lhs_stride: Vec<usize>,
@ -242,13 +245,14 @@ enum CudaStorageSlice {
fn slice_src_and_dst<'a, T>(
src: &'a CudaSlice<T>,
src_offset: usize,
src_l: &Layout,
dst: &'a mut CudaSlice<T>,
dst_offset: usize,
) -> (
cudarc::driver::CudaView<'a, T>,
cudarc::driver::CudaViewMut<'a, T>,
) {
let src_offset = src_l.start_offset();
let to_copy = dst
.len()
.saturating_sub(dst_offset)
@ -268,12 +272,14 @@ fn gemm_config<T>(
alpha: T,
beta: T,
(b, m, n, k): (usize, usize, usize, usize),
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<StridedBatchedConfig<T>> {
// https://docs.nvidia.com/cuda/cublas/index.html#cublas-t-gemm
use cudarc::cublas::sys::cublasOperation_t;
let lhs_stride = lhs_l.stride();
let rhs_stride = rhs_l.stride();
let rhs_m1 = rhs_stride[rhs_stride.len() - 1];
let rhs_m2 = rhs_stride[rhs_stride.len() - 2];
let lhs_m1 = lhs_stride[lhs_stride.len() - 1];
@ -352,20 +358,27 @@ impl CudaStorage {
&self.device
}
pub(crate) fn to_dtype(&self, shape: &Shape, stride: &[usize], dtype: DType) -> Result<Self> {
pub(crate) fn to_dtype(&self, layout: &Layout, dtype: DType) -> Result<Self> {
use cudarc::driver::DevicePtr;
let shape = layout.shape();
let dims = shape.dims();
let el = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el as u32);
let dev = self.device();
let ds = dev.htod_copy([dims, stride].concat())?;
let ds = dev.htod_copy([dims, layout.stride()].concat())?;
let start_o = layout.start_offset();
// This returns an i64 rather than a &i64, this is useful to get around some temporary
// lifetime issue and is safe as long as self.slice does not go out of scope before inp
// is used.
let inp = match &self.slice {
CudaStorageSlice::U32(inp) => inp.device_ptr(),
CudaStorageSlice::BF16(inp) => inp.device_ptr(),
CudaStorageSlice::F16(inp) => inp.device_ptr(),
CudaStorageSlice::F32(inp) => inp.device_ptr(),
CudaStorageSlice::F64(inp) => inp.device_ptr(),
CudaStorageSlice::U32(inp) => *inp.slice(start_o..).device_ptr(),
CudaStorageSlice::BF16(inp) => *inp.slice(start_o..).device_ptr(),
CudaStorageSlice::F16(inp) => *inp.slice(start_o..).device_ptr(),
CudaStorageSlice::F32(inp) => *inp.slice(start_o..).device_ptr(),
CudaStorageSlice::F64(inp) => *inp.slice(start_o..).device_ptr(),
};
let inp = &inp;
let kernel_name = format!("cast_{}_{}", self.dtype().as_str(), dtype.as_str());
let func = dev.get_or_load_func(&kernel_name, kernels::CAST)?;
let slice = match dtype {
@ -406,20 +419,16 @@ impl CudaStorage {
})
}
pub(crate) fn affine_impl(
&self,
shape: &Shape,
stride: &[usize],
mul: f64,
add: f64,
) -> Result<Self> {
pub(crate) fn affine(&self, layout: &Layout, mul: f64, add: f64) -> Result<Self> {
let shape = layout.shape();
let dims = shape.dims();
let el_count = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el_count as u32);
let dev = self.device();
let ds = dev.htod_copy([dims, stride].concat())?;
let ds = dev.htod_copy([dims, layout.stride()].concat())?;
let slice = match &self.slice {
CudaStorageSlice::U32(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("affine_u32", kernels::AFFINE)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<u32>(el_count) }?;
@ -429,6 +438,7 @@ impl CudaStorage {
CudaStorageSlice::U32(out)
}
CudaStorageSlice::BF16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("affine_bf16", kernels::AFFINE)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<bf16>(el_count) }?;
@ -446,6 +456,7 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
CudaStorageSlice::F16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("affine_f16", kernels::AFFINE)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f16>(el_count) }?;
@ -463,6 +474,7 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
CudaStorageSlice::F32(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("affine_f32", kernels::AFFINE)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f32>(el_count) }?;
@ -472,6 +484,7 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
CudaStorageSlice::F64(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("affine_f64", kernels::AFFINE)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f64>(el_count) }?;
@ -485,7 +498,8 @@ impl CudaStorage {
Ok(Self { slice, device })
}
pub(crate) fn sum(&self, shape: &Shape, stride: &[usize], sum_dims: &[usize]) -> Result<Self> {
pub(crate) fn sum(&self, layout: &Layout, sum_dims: &[usize]) -> Result<Self> {
let shape = layout.shape();
let src_dims = shape.dims();
let el = shape.elem_count();
let mut dst_el = el;
@ -503,9 +517,10 @@ impl CudaStorage {
.collect();
let cfg = LaunchConfig::for_num_elems(el as u32);
let dev = self.device();
let ds = dev.htod_copy([src_dims, stride, &sum_dims_l, &sum_dims_s].concat())?;
let ds = dev.htod_copy([src_dims, layout.stride(), &sum_dims_l, &sum_dims_s].concat())?;
let slice = match &self.slice {
CudaStorageSlice::U32(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("sum_u32", kernels::REDUCE)?;
let out = dev.alloc_zeros::<u32>(dst_el)?;
let params = (el, src_dims.len(), sum_dims.len(), &ds, arg, &out);
@ -514,6 +529,7 @@ impl CudaStorage {
CudaStorageSlice::U32(out)
}
CudaStorageSlice::BF16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("sum_bf16", kernels::REDUCE)?;
let out = dev.alloc_zeros::<bf16>(dst_el)?;
let params = (el, src_dims.len(), sum_dims.len(), &ds, arg, &out);
@ -522,6 +538,7 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
CudaStorageSlice::F16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("sum_f16", kernels::REDUCE)?;
let out = dev.alloc_zeros::<f16>(dst_el)?;
let params = (el, src_dims.len(), sum_dims.len(), &ds, arg, &out);
@ -530,6 +547,7 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
CudaStorageSlice::F32(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("sum_f32", kernels::REDUCE)?;
let out = dev.alloc_zeros::<f32>(dst_el)?;
let params = (el, src_dims.len(), sum_dims.len(), &ds, arg, &out);
@ -538,6 +556,7 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
CudaStorageSlice::F64(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func("sum_f64", kernels::REDUCE)?;
let out = dev.alloc_zeros::<f64>(dst_el)?;
let params = (el, src_dims.len(), sum_dims.len(), &ds, arg, &out);
@ -556,21 +575,19 @@ impl CudaStorage {
))
}
pub(crate) fn unary_impl<U: crate::op::UnaryOp>(
&self,
shape: &Shape,
stride: &[usize],
) -> Result<Self> {
pub(crate) fn unary_impl<U: crate::op::UnaryOp>(&self, layout: &Layout) -> Result<Self> {
let shape = layout.shape();
let dims = shape.dims();
let el_count = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el_count as u32);
let dev = &self.device;
let ds = dev.htod_copy([dims, stride].concat())?;
let ds = dev.htod_copy([dims, layout.stride()].concat())?;
let slice = match &self.slice {
CudaStorageSlice::U32(_arg) => {
todo!("No unary kernels for u32");
}
CudaStorageSlice::BF16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func(U::KERNEL_BF16, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<bf16>(el_count) }?;
@ -580,6 +597,7 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
CudaStorageSlice::F16(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func(U::KERNEL_F16, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f16>(el_count) }?;
@ -589,6 +607,7 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
CudaStorageSlice::F32(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func(U::KERNEL_F32, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f32>(el_count) }?;
@ -598,6 +617,7 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
CudaStorageSlice::F64(arg) => {
let arg = &arg.slice(layout.start_offset()..);
let func = dev.get_or_load_func(U::KERNEL_F64, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f64>(el_count) }?;
@ -614,17 +634,19 @@ impl CudaStorage {
pub(crate) fn binary_impl<B: crate::op::BinaryOp>(
&self,
rhs: &Self,
shape: &Shape,
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<Self> {
let shape = lhs_l.shape();
let dims = shape.dims();
let elem_count = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(elem_count as u32);
let dev = self.device();
let dims_and_strides = dev.htod_copy([dims, lhs_stride, rhs_stride].concat())?;
let dims_and_strides = dev.htod_copy([dims, lhs_l.stride(), rhs_l.stride()].concat())?;
let slice = match (&self.slice, &rhs.slice) {
(CudaStorageSlice::BF16(lhs), CudaStorageSlice::BF16(rhs)) => {
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func(B::KERNEL_BF16, kernels::BINARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<bf16>(elem_count) }?;
@ -634,6 +656,8 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
(CudaStorageSlice::F16(lhs), CudaStorageSlice::F16(rhs)) => {
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func(B::KERNEL_F16, kernels::BINARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f16>(elem_count) }?;
@ -643,6 +667,8 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
(CudaStorageSlice::F32(lhs), CudaStorageSlice::F32(rhs)) => {
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func(B::KERNEL_F32, kernels::BINARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f32>(elem_count) }?;
@ -652,6 +678,8 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
(CudaStorageSlice::F64(lhs), CudaStorageSlice::F64(rhs)) => {
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
// SAFETY: Set later by running the kernel.
let func = dev.get_or_load_func(B::KERNEL_F64, kernels::BINARY)?;
let out = unsafe { dev.alloc::<f64>(elem_count) }?;
@ -661,6 +689,8 @@ impl CudaStorage {
CudaStorageSlice::F64(out)
}
(CudaStorageSlice::U32(lhs), CudaStorageSlice::U32(rhs)) => {
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
// SAFETY: Set later by running the kernel.
let func = dev.get_or_load_func(B::KERNEL_U32, kernels::BINARY)?;
let out = unsafe { dev.alloc::<u32>(elem_count) }?;
@ -708,28 +738,32 @@ impl CudaStorage {
pub(crate) fn where_cond(
&self,
shape: &Shape,
stride: &[usize],
layout: &Layout,
t: &Self,
stride_t: &[usize],
layout_t: &Layout,
f: &Self,
stride_f: &[usize],
layout_f: &Layout,
) -> Result<Self> {
let ids = match &self.slice {
CudaStorageSlice::U32(slice) => slice,
CudaStorageSlice::U32(slice) => slice.slice(layout.start_offset()..),
_ => Err(CudaError::UnexpectedDType {
msg: "where conditions should be u32",
expected: DType::U32,
got: self.dtype(),
})?,
};
let ids = &ids;
let shape = layout.shape();
let dims = shape.dims();
let el = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el as u32);
let dev = self.device();
let ds = dev.htod_copy([dims, stride, stride_t, stride_f].concat())?;
let ds =
dev.htod_copy([dims, layout.stride(), layout_t.stride(), layout_f.stride()].concat())?;
let slice = match (&t.slice, &f.slice) {
(CudaStorageSlice::BF16(t), CudaStorageSlice::BF16(f)) => {
let t = &t.slice(layout_t.start_offset()..);
let f = &f.slice(layout_f.start_offset()..);
let func = dev.get_or_load_func("where_bf16", kernels::TERNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<bf16>(el) }?;
@ -739,6 +773,8 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
(CudaStorageSlice::F16(t), CudaStorageSlice::F16(f)) => {
let t = &t.slice(layout_t.start_offset()..);
let f = &f.slice(layout_f.start_offset()..);
let func = dev.get_or_load_func("where_f16", kernels::TERNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f16>(el) }?;
@ -748,6 +784,8 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
(CudaStorageSlice::F32(t), CudaStorageSlice::F32(f)) => {
let t = &t.slice(layout_t.start_offset()..);
let f = &f.slice(layout_f.start_offset()..);
let func = dev.get_or_load_func("where_f32", kernels::TERNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f32>(el) }?;
@ -757,6 +795,8 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
(CudaStorageSlice::F64(t), CudaStorageSlice::F64(f)) => {
let t = &t.slice(layout_t.start_offset()..);
let f = &f.slice(layout_f.start_offset()..);
// SAFETY: Set later by running the kernel.
let func = dev.get_or_load_func("where_f64", kernels::TERNARY)?;
let out = unsafe { dev.alloc::<f64>(el) }?;
@ -766,6 +806,8 @@ impl CudaStorage {
CudaStorageSlice::F64(out)
}
(CudaStorageSlice::U32(t), CudaStorageSlice::U32(f)) => {
let t = &t.slice(layout_t.start_offset()..);
let f = &f.slice(layout_f.start_offset()..);
// SAFETY: Set later by running the kernel.
let func = dev.get_or_load_func("where_u32", kernels::TERNARY)?;
let out = unsafe { dev.alloc::<u32>(el) }?;
@ -775,36 +817,36 @@ impl CudaStorage {
CudaStorageSlice::U32(out)
}
// The dtypes should have been checked at this point so this is an internal error.
_ => return Err(CudaError::InternalError("dtype mismatch in binary op")),
_ => Err(CudaError::InternalError("dtype mismatch in binary op"))?,
};
let device = dev.clone();
Ok(Self { slice, device })
}
pub(crate) fn embedding_impl(
&self,
shape: &Shape,
stride: &[usize],
rhs: &Self,
h_size: usize, // hidden size
v_size: usize, // vocab size
) -> Result<Self> {
pub(crate) fn embedding(&self, layout: &Layout, rhs: &Self, rhs_l: &Layout) -> Result<Self> {
let ids = match &self.slice {
CudaStorageSlice::U32(slice) => slice,
CudaStorageSlice::U32(slice) => slice.slice(layout.start_offset()..),
_ => Err(CudaError::UnexpectedDType {
msg: "embedding ids should be u32",
expected: DType::U32,
got: self.dtype(),
})?,
};
let ids = &ids;
let shape = layout.shape();
let (v_size, h_size) = rhs_l
.shape()
.r2()
.map_err(|e| CudaError::WrappedError(Box::new(e)))?;
let dims = shape.dims();
let el = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el as u32);
let dev = self.device();
let ds = dev.htod_copy([dims, stride].concat())?;
let ds = dev.htod_copy([dims, layout.stride()].concat())?;
let slice = match &rhs.slice {
// The kernels below assume that rhs is contiguous.
CudaStorageSlice::U32(arg) => {
let arg = &arg.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func("emb_u32", kernels::EMBEDDINGS)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<u32>(el * h_size) }?;
@ -814,6 +856,7 @@ impl CudaStorage {
CudaStorageSlice::U32(out)
}
CudaStorageSlice::BF16(arg) => {
let arg = &arg.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func("emb_bf16", kernels::EMBEDDINGS)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<bf16>(el * h_size) }?;
@ -823,6 +866,7 @@ impl CudaStorage {
CudaStorageSlice::BF16(out)
}
CudaStorageSlice::F16(arg) => {
let arg = &arg.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func("emb_f16", kernels::EMBEDDINGS)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f16>(el * h_size) }?;
@ -832,6 +876,7 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
CudaStorageSlice::F32(arg) => {
let arg = &arg.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func("emb_f32", kernels::EMBEDDINGS)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f32>(el * h_size) }?;
@ -841,6 +886,7 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
CudaStorageSlice::F64(arg) => {
let arg = &arg.slice(rhs_l.start_offset()..);
let func = dev.get_or_load_func("emb_f64", kernels::EMBEDDINGS)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.alloc::<f64>(el * h_size) }?;
@ -854,12 +900,12 @@ impl CudaStorage {
Ok(Self { slice, device })
}
pub(crate) fn matmul_impl(
pub(crate) fn matmul(
&self,
rhs: &Self,
(b, m, n, k): (usize, usize, usize, usize),
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_l: &Layout,
rhs_l: &Layout,
) -> Result<Self> {
let elem_count = b * m * n;
let dev = &self.device;
@ -868,7 +914,9 @@ impl CudaStorage {
todo!("bf16")
}
(CudaStorageSlice::F16(lhs), CudaStorageSlice::F16(rhs)) => {
let cfg = gemm_config(f16::ONE, f16::ZERO, (b, m, n, k), lhs_stride, rhs_stride)?;
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let cfg = gemm_config(f16::ONE, f16::ZERO, (b, m, n, k), lhs_l, rhs_l)?;
let mut out = unsafe { dev.alloc::<f16>(elem_count) }?;
unsafe {
self.device
@ -878,7 +926,9 @@ impl CudaStorage {
CudaStorageSlice::F16(out)
}
(CudaStorageSlice::F32(lhs), CudaStorageSlice::F32(rhs)) => {
let cfg = gemm_config(1., 0., (b, m, n, k), lhs_stride, rhs_stride)?;
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let cfg = gemm_config(1., 0., (b, m, n, k), lhs_l, rhs_l)?;
let mut out = unsafe { dev.alloc::<f32>(elem_count) }?;
unsafe {
self.device
@ -888,7 +938,9 @@ impl CudaStorage {
CudaStorageSlice::F32(out)
}
(CudaStorageSlice::F64(lhs), CudaStorageSlice::F64(rhs)) => {
let cfg = gemm_config(1., 0., (b, m, n, k), lhs_stride, rhs_stride)?;
let lhs = &lhs.slice(lhs_l.start_offset()..);
let rhs = &rhs.slice(rhs_l.start_offset()..);
let cfg = gemm_config(1., 0., (b, m, n, k), lhs_l, rhs_l)?;
let mut out = unsafe { dev.alloc::<f64>(elem_count) }?;
unsafe {
self.device
@ -907,22 +959,18 @@ impl CudaStorage {
&self,
dst: &mut Self,
dst_offset: usize,
src_shape: &Shape,
src_stride: &[usize],
src_offset: usize,
src_l: &Layout,
) -> Result<()> {
if src_shape.rank() != src_stride.len() {
panic!("incoherent shape and strides {src_shape:?} {src_stride:?}")
}
let src_shape = src_l.shape();
let dims = src_shape.dims();
let el_count = src_shape.elem_count();
let cfg = LaunchConfig::for_num_elems(el_count as u32);
let dev = &self.device;
let ds = dev.htod_copy([dims, src_stride].concat())?;
let ds = dev.htod_copy([dims, src_l.stride()].concat())?;
match (&self.slice, &mut dst.slice) {
(CudaStorageSlice::BF16(src), CudaStorageSlice::BF16(dst)) => {
let (src, mut dst) = slice_src_and_dst(src, src_offset, dst, dst_offset);
if src_shape.is_contiguous(src_stride) {
let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);
if src_l.is_contiguous() {
dev.dtod_copy(&src, &mut dst)?
} else {
let func = dev.get_or_load_func("ucopy_bf16", kernels::UNARY)?;
@ -933,8 +981,8 @@ impl CudaStorage {
}
}
(CudaStorageSlice::F16(src), CudaStorageSlice::F16(dst)) => {
let (src, mut dst) = slice_src_and_dst(src, src_offset, dst, dst_offset);
if src_shape.is_contiguous(src_stride) {
let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);
if src_l.is_contiguous() {
dev.dtod_copy(&src, &mut dst)?
} else {
let func = dev.get_or_load_func("ucopy_f16", kernels::UNARY)?;
@ -945,8 +993,8 @@ impl CudaStorage {
}
}
(CudaStorageSlice::F32(src), CudaStorageSlice::F32(dst)) => {
let (src, mut dst) = slice_src_and_dst(src, src_offset, dst, dst_offset);
if src_shape.is_contiguous(src_stride) {
let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);
if src_l.is_contiguous() {
dev.dtod_copy(&src, &mut dst)?
} else {
let func = dev.get_or_load_func("ucopy_f32", kernels::UNARY)?;
@ -957,8 +1005,8 @@ impl CudaStorage {
}
}
(CudaStorageSlice::U32(src), CudaStorageSlice::U32(dst)) => {
let (src, mut dst) = slice_src_and_dst(src, src_offset, dst, dst_offset);
if src_shape.is_contiguous(src_stride) {
let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);
if src_l.is_contiguous() {
dev.dtod_copy(&src, &mut dst)?
} else {
let func = dev.get_or_load_func("ucopy_u32", kernels::UNARY)?;
@ -969,8 +1017,8 @@ impl CudaStorage {
}
}
(CudaStorageSlice::F64(src), CudaStorageSlice::F64(dst)) => {
let (src, mut dst) = slice_src_and_dst(src, src_offset, dst, dst_offset);
if src_shape.is_contiguous(src_stride) {
let (src, mut dst) = slice_src_and_dst(src, src_l, dst, dst_offset);
if src_l.is_contiguous() {
dev.dtod_copy(&src, &mut dst)?
} else {
let func = dev.get_or_load_func("ucopy_64", kernels::UNARY)?;

12
candle-core/src/dtype.rs
View File

@ -41,7 +41,6 @@ pub trait WithDType: Sized + Copy {
}
fn cpu_storage_as_slice(s: &CpuStorage) -> Result<&[Self]>;
fn cpu_storage_as_mut_slice(s: &mut CpuStorage) -> Result<&mut [Self]>;
fn cpu_storage_data(s: CpuStorage) -> Result<Vec<Self>>;
}
@ -75,17 +74,6 @@ macro_rules! with_dtype {
}),
}
}
fn cpu_storage_as_mut_slice(s: &mut CpuStorage) -> Result<&mut [Self]> {
match s {
CpuStorage::$dtype(data) => Ok(data),
_ => Err(Error::UnexpectedDType {
expected: DType::$dtype,
got: s.dtype(),
msg: "unexpected dtype",
}),
}
}
}
};
}

46
candle-core/src/dummy_cuda_backend.rs
View File

@ -1,5 +1,5 @@
#![allow(dead_code)]
use crate::{CpuStorage, DType, Error, Result, Shape};
use crate::{CpuStorage, DType, Error, Layout, Result, Shape};
#[derive(thiserror::Error, Debug)]
pub enum DummyError {}
@ -60,11 +60,11 @@ impl CudaStorage {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn affine_impl(&self, _: &Shape, _: &[usize], _: f64, _: f64) -> Result<Self> {
pub(crate) fn affine(&self, _: &Layout, _: f64, _: f64) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn sum(&self, _: &Shape, _: &[usize], _: &[usize]) -> Result<Self> {
pub(crate) fn sum(&self, _: &Layout, _: &[usize]) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
@ -72,65 +72,49 @@ impl CudaStorage {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn to_dtype(&self, _: &Shape, _: &[usize], _: DType) -> Result<Self> {
pub(crate) fn to_dtype(&self, _: &Layout, _: DType) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn unary_impl<B: crate::op::UnaryOp>(&self, _: &Shape, _: &[usize]) -> Result<Self> {
pub(crate) fn unary_impl<B: crate::op::UnaryOp>(&self, _: &Layout) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn binary_impl<B: crate::op::BinaryOp>(
&self,
_: &Self,
_: &Shape,
_: &[usize],
_: &[usize],
_: &Layout,
_: &Layout,
) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn where_cond(
&self,
_: &Shape,
_: &[usize],
_: &Layout,
_: &Self,
_: &[usize],
_: &Layout,
_: &Self,
_: &[usize],
_: &Layout,
) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn embedding_impl(
&self,
_: &Shape,
_: &[usize],
_: &Self,
_: usize,
_: usize,
) -> Result<Self> {
pub(crate) fn embedding(&self, _: &Layout, _: &Self, _: &Layout) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn matmul_impl(
pub(crate) fn matmul(
&self,
_: &Self,
_: (usize, usize, usize, usize),
_: &[usize],
_: &[usize],
_: &Layout,
_: &Layout,
) -> Result<Self> {
Err(Error::NotCompiledWithCudaSupport)
}
pub(crate) fn copy_strided_src(
&self,
_: &mut Self,
_: usize,
_: &Shape,
_: &[usize],
_: usize,
) -> Result<()> {
pub(crate) fn copy_strided_src(&self, _: &mut Self, _: usize, _: &Layout) -> Result<()> {
Err(Error::NotCompiledWithCudaSupport)
}
}

140
candle-core/src/layout.rs Normal file
View File

@ -0,0 +1,140 @@
use crate::{Error, Result, Shape};
#[derive(Debug, PartialEq, Eq, Clone)]
pub struct Layout {
shape: Shape,
// The strides are given in number of elements and not in bytes.
stride: Vec<usize>,
start_offset: usize,
}
impl Layout {
pub fn contiguous_with_offset<S: Into<Shape>>(shape: S, start_offset: usize) -> Self {
let shape = shape.into();
let stride = shape.stride_contiguous();
Self {
shape,
stride,
start_offset,
}
}
pub fn contiguous<S: Into<Shape>>(shape: S) -> Self {
Self::contiguous_with_offset(shape, 0)
}
pub fn dims(&self) -> &[usize] {
self.shape.dims()
}
pub fn shape(&self) -> &Shape {
&self.shape
}
pub fn stride(&self) -> &[usize] {
&self.stride
}
pub fn start_offset(&self) -> usize {
self.start_offset
}
/// Returns the appropriate start and stop offset if the data is stored in a C
/// contiguous (aka row major) way.
pub fn contiguous_offsets(&self) -> Option<(usize, usize)> {
if self.is_contiguous() {
let start_o = self.start_offset;
Some((start_o, start_o + self.shape.elem_count()))
} else {
None
}
}
/// Returns true if the data is stored in a C contiguous (aka row major) way.
pub fn is_contiguous(&self) -> bool {
self.shape.is_contiguous(&self.stride)
}
/// Returns true if the data is stored in a Fortran contiguous (aka column major) way.
pub fn is_fortran_contiguous(&self) -> bool {
self.shape.is_fortran_contiguous(&self.stride)
}
pub(crate) fn narrow(&self, dim: usize, start: usize, length: usize) -> Result<Self> {
let dims = self.shape().dims();
if dim >= dims.len() {
Err(Error::UnexpectedNumberOfDims {
expected: dim + 1,
got: dims.len(),
shape: self.shape().clone(),
})?
}
if start + length > dims[dim] {
todo!("add a proper error: out of bounds for narrow {dim} {start} {length} {dims:?}")
}
let mut dims = dims.to_vec();
dims[dim] = length;
Ok(Self {
shape: Shape::from(dims),
stride: self.stride.clone(),
start_offset: self.start_offset + self.stride[dim] * start,
})
}
pub(crate) fn transpose(&self, dim1: usize, dim2: usize) -> Result<Self> {
let rank = self.shape.rank();
if rank <= dim1 || rank <= dim2 {
return Err(Error::UnexpectedNumberOfDims {
expected: usize::max(dim1, dim2),
got: rank,
shape: self.shape().clone(),
});
}
let mut stride = self.stride().to_vec();
let mut dims = self.shape().dims().to_vec();
dims.swap(dim1, dim2);
stride.swap(dim1, dim2);
Ok(Self {
shape: Shape::from(dims),
stride,
start_offset: self.start_offset,
})
}
pub fn broadcast_as<S: Into<Shape>>(&self, shape: S) -> Result<Self> {
let shape = shape.into();
if shape.rank() < self.shape().rank() {
Err(Error::BroadcastIncompatibleShapes {
src_shape: self.shape().clone(),
dst_shape: shape.clone(),
})?
}
let added_dims = shape.rank() - self.shape().rank();
let mut stride = vec![0; added_dims];
for (&dst_dim, (&src_dim, &src_stride)) in shape.dims()[added_dims..]
.iter()
.zip(self.dims().iter().zip(self.stride()))
{
let s = if dst_dim == src_dim {
src_stride
} else if src_dim != 1 {
return Err(Error::BroadcastIncompatibleShapes {
src_shape: self.shape().clone(),
dst_shape: shape,
});
} else {
0
};
stride.push(s)
}
Ok(Self {
shape,
stride,
start_offset: self.start_offset,
})
}
pub(crate) fn strided_index(&self) -> crate::StridedIndex {
crate::StridedIndex::new(self)
}
}

2
candle-core/src/lib.rs
View File

@ -7,6 +7,7 @@ pub mod display;
mod dtype;
mod dummy_cuda_backend;
mod error;
mod layout;
mod npy;
mod op;
mod shape;
@ -19,6 +20,7 @@ pub use cpu_backend::CpuStorage;
pub use device::{Device, DeviceLocation};
pub use dtype::{DType, WithDType};
pub use error::{Error, Result};
pub use layout::Layout;
pub use shape::Shape;
pub use storage::Storage;
use strided_index::StridedIndex;

94
candle-core/src/storage.rs
View File

@ -1,4 +1,4 @@
use crate::{op, CpuStorage, CudaStorage, DType, Device, Error, Result, Shape};
use crate::{op, CpuStorage, CudaStorage, DType, Device, Error, Layout, Result, Shape};
// We do not want to implement Clone on Storage as cloning may fail because of
// out of memory. Instead try_clone should be used.
@ -53,38 +53,33 @@ impl Storage {
}
}
pub(crate) fn affine_impl(
&self,
shape: &Shape,
stride: &[usize],
mul: f64,
add: f64,
) -> Result<Self> {
pub(crate) fn affine(&self, layout: &Layout, mul: f64, add: f64) -> Result<Self> {
match self {
Storage::Cpu(storage) => {
let storage = storage.affine_impl(shape, stride, mul, add)?;
let storage = storage.affine(layout, mul, add)?;
Ok(Self::Cpu(storage))
}
Self::Cuda(storage) => {
let storage = storage.affine_impl(shape, stride, mul, add)?;
let storage = storage.affine(layout, mul, add)?;
Ok(Self::Cuda(storage))
}
}
}
pub(crate) fn sum(&self, shape: &Shape, stride: &[usize], s: &[usize]) -> Result<Self> {
pub(crate) fn sum(&self, layout: &Layout, s: &[usize]) -> Result<Self> {
match self {
Storage::Cpu(storage) => {
let storage = storage.sum(shape, stride, s)?;
let storage = storage.sum(layout, s)?;
Ok(Self::Cpu(storage))
}
Self::Cuda(storage) => {
let storage = storage.sum(shape, stride, s)?;
let storage = storage.sum(layout, s)?;
Ok(Self::Cuda(storage))
}
}
}
// This assumes a contiguous layout and no offset.
pub(crate) fn divide_by_sum_over_dim(&mut self, shape: &Shape, dim: usize) -> Result<()> {
match self {
Storage::Cpu(storage) => storage.divide_by_sum_over_dim(shape, dim)?,
@ -93,32 +88,28 @@ impl Storage {
Ok(())
}
pub(crate) fn to_dtype(&self, shape: &Shape, stride: &[usize], dtype: DType) -> Result<Self> {
pub(crate) fn to_dtype(&self, layout: &Layout, dtype: DType) -> Result<Self> {
match self {
Storage::Cpu(storage) => {
let storage = storage.to_dtype(shape, stride, dtype)?;
let storage = storage.to_dtype(layout, dtype)?;
Ok(Self::Cpu(storage))
}
Self::Cuda(storage) => {
let storage = storage.to_dtype(shape, stride, dtype)?;
let storage = storage.to_dtype(layout, dtype)?;
Ok(Self::Cuda(storage))
}
}
}
pub(crate) fn unary_impl<B: op::UnaryOp>(
&self,
shape: &Shape,
stride: &[usize],
) -> Result<Self> {
pub(crate) fn unary_impl<B: op::UnaryOp>(&self, layout: &Layout) -> Result<Self> {
// TODO: Different code path for the contiguous case?
match self {
Storage::Cpu(storage) => {
let storage = storage.unary_impl::<B>(shape, stride)?;
let storage = storage.unary_impl::<B>(layout)?;
Ok(Self::Cpu(storage))
}
Self::Cuda(storage) => {
let storage = storage.unary_impl::<B>(shape, stride)?;
let storage = storage.unary_impl::<B>(layout)?;
Ok(Self::Cuda(storage))
}
}
@ -127,19 +118,18 @@ impl Storage {
pub(crate) fn binary_impl<B: op::BinaryOp>(
&self,
rhs: &Self,
shape: &Shape,
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_layout: &Layout,
rhs_layout: &Layout,
) -> Result<Self> {
self.same_device(rhs, B::NAME)?;
self.same_dtype(rhs, B::NAME)?;
match (self, rhs) {
(Storage::Cpu(lhs), Storage::Cpu(rhs)) => {
let storage = lhs.binary_impl::<B>(rhs, shape, lhs_stride, rhs_stride)?;
let storage = lhs.binary_impl::<B>(rhs, lhs_layout, rhs_layout)?;
Ok(Self::Cpu(storage))
}
(Self::Cuda(lhs), Self::Cuda(rhs)) => {
let storage = lhs.binary_impl::<B>(rhs, shape, lhs_stride, rhs_stride)?;
let storage = lhs.binary_impl::<B>(rhs, lhs_layout, rhs_layout)?;
Ok(Self::Cuda(storage))
}
(lhs, rhs) => {
@ -156,49 +146,41 @@ impl Storage {
pub(crate) fn where_cond(
&self,
shape: &Shape,
stride: &[usize],
layout: &Layout,
t: &Self,
stride_t: &[usize],
layout_t: &Layout,
f: &Self,
stride_f: &[usize],
layout_f: &Layout,
) -> Result<Self> {
self.same_device(t, "where")?;
self.same_device(f, "where")?;
t.same_dtype(f, "where")?;
match (self, t, f) {
(Storage::Cpu(cond), Storage::Cpu(t), Storage::Cpu(f)) => {
let storage = cond.where_cond(shape, stride, t, stride_t, f, stride_f)?;
let storage = cond.where_cond(layout, t, layout_t, f, layout_f)?;
Ok(Self::Cpu(storage))
}
(Self::Cuda(cond), Self::Cuda(t), Self::Cuda(f)) => {
let storage = cond.where_cond(shape, stride, t, stride_t, f, stride_f)?;
let storage = cond.where_cond(layout, t, layout_t, f, layout_f)?;
Ok(Self::Cuda(storage))
}
(_, lhs, rhs) => Err(Error::DeviceMismatchBinaryOp {
lhs: lhs.device().location(),
rhs: rhs.device().location(),
op: "embedding",
op: "where",
}),
}
}
pub(crate) fn embedding_impl(
&self,
shape: &Shape,
stride: &[usize],
rhs: &Self,
hidden_size: usize,
vocab_size: usize,
) -> Result<Self> {
pub(crate) fn embedding(&self, layout: &Layout, rhs: &Self, rhs_l: &Layout) -> Result<Self> {
self.same_device(rhs, "embedding")?;
match (self, rhs) {
(Storage::Cpu(lhs), Storage::Cpu(rhs)) => {
let storage = lhs.embedding_impl(shape, stride, rhs, hidden_size, vocab_size)?;
let storage = lhs.embedding(layout, rhs, rhs_l)?;
Ok(Self::Cpu(storage))
}
(Self::Cuda(lhs), Self::Cuda(rhs)) => {
let storage = lhs.embedding_impl(shape, stride, rhs, hidden_size, vocab_size)?;
let storage = lhs.embedding(layout, rhs, rhs_l)?;
Ok(Self::Cuda(storage))
}
(lhs, rhs) => Err(Error::DeviceMismatchBinaryOp {
@ -209,22 +191,22 @@ impl Storage {
}
}
pub(crate) fn matmul_impl(
pub(crate) fn matmul(
&self,
rhs: &Self,
bmnk: (usize, usize, usize, usize),
lhs_stride: &[usize],
rhs_stride: &[usize],
lhs_layout: &Layout,
rhs_layout: &Layout,
) -> Result<Self> {
self.same_device(rhs, "matmul")?;
self.same_dtype(rhs, "matmul")?;
match (self, rhs) {
(Self::Cpu(lhs), Self::Cpu(rhs)) => {
let storage = lhs.matmul_impl(rhs, bmnk, lhs_stride, rhs_stride)?;
let storage = lhs.matmul(rhs, bmnk, lhs_layout, rhs_layout)?;
Ok(Self::Cpu(storage))
}
(Self::Cuda(lhs), Self::Cuda(rhs)) => {
let storage = lhs.matmul_impl(rhs, bmnk, lhs_stride, rhs_stride)?;
let storage = lhs.matmul(rhs, bmnk, lhs_layout, rhs_layout)?;
Ok(Self::Cuda(storage))
}
(lhs, rhs) => Err(Error::DeviceMismatchBinaryOp {
@ -240,17 +222,11 @@ impl Storage {
&self,
dst: &mut Self,
dst_offset: usize,
src_shape: &Shape,
src_stride: &[usize],
src_offset: usize,
src_l: &Layout,
) -> Result<()> {
match (self, dst) {
(Self::Cpu(src), Self::Cpu(dst)) => {
src.copy_strided_src(dst, dst_offset, src_shape, src_stride, src_offset)
}
(Self::Cuda(src), Self::Cuda(dst)) => {
Ok(src.copy_strided_src(dst, dst_offset, src_shape, src_stride, src_offset)?)
}
(Self::Cpu(src), Self::Cpu(dst)) => src.copy_strided_src(dst, dst_offset, src_l),
(Self::Cuda(src), Self::Cuda(dst)) => Ok(src.copy_strided_src(dst, dst_offset, src_l)?),
(lhs, rhs) => Err(Error::DeviceMismatchBinaryOp {
lhs: lhs.device().location(),
rhs: rhs.device().location(),

25
candle-core/src/strided_index.rs
View File

@ -1,27 +1,28 @@
use crate::Layout;
/// An iterator over offset position for items of an N-dimensional arrays stored in a
/// flat buffer using some potential strides.
#[derive(Debug)]
pub(crate) struct StridedIndex<'a> {
next_storage_index: Option<usize>,
multi_index: Vec<usize>,
dims: &'a [usize],
stride: &'a [usize],
layout: &'a Layout,
}
impl<'a> StridedIndex<'a> {
pub(crate) fn new(dims: &'a [usize], stride: &'a [usize]) -> Self {
pub(crate) fn new(layout: &'a Layout) -> Self {
let dims = layout.dims();
let elem_count: usize = dims.iter().product();
let next_storage_index = if elem_count == 0 {
None
} else {
// This applies to the scalar case.
Some(0)
Some(layout.start_offset())
};
StridedIndex {
next_storage_index,
multi_index: vec![0; dims.len()],
dims,
stride,
layout,
}
}
}
@ -35,7 +36,12 @@ impl<'a> Iterator for StridedIndex<'a> {
Some(storage_index) => storage_index,
};
let mut updated = false;
for (multi_i, max_i) in self.multi_index.iter_mut().zip(self.dims.iter()).rev() {
for (multi_i, max_i) in self
.multi_index
.iter_mut()
.zip(self.layout.dims().iter())
.rev()
{
let next_i = *multi_i + 1;
if next_i < *max_i {
*multi_i = next_i;
@ -49,9 +55,10 @@ impl<'a> Iterator for StridedIndex<'a> {
let next_storage_index = self
.multi_index
.iter()
.zip(self.stride.iter())
.zip(self.layout.stride().iter())
.map(|(&x, &y)| x * y)
.sum();
.sum::<usize>()
+ self.layout.start_offset();
Some(next_storage_index)
} else {
None

177
candle-core/src/tensor.rs
View File

@ -1,4 +1,4 @@
use crate::{op::Op, storage::Storage, DType, Device, Error, Result, Shape};
use crate::{op::Op, storage::Storage, DType, Device, Error, Layout, Result, Shape};
use std::sync::Arc;
/// Unique identifier for tensors.
@ -17,9 +17,7 @@ impl TensorId {
pub struct Tensor_ {
id: TensorId,
storage: Arc<Storage>,
shape: Shape,
// The strides are given in number of elements and not in bytes.
stride: Vec<usize>,
layout: Layout,
op: Option<Op>,
is_variable: bool,
}
@ -50,7 +48,7 @@ macro_rules! unary_op {
let shape = self.shape();
let storage = self
.storage
.unary_impl::<crate::op::$op_name>(self.shape(), self.stride())?;
.unary_impl::<crate::op::$op_name>(self.layout())?;
let op = if self.track_op() {
Some(Op::$op_name(self.clone()))
} else {
@ -67,9 +65,8 @@ macro_rules! binary_op {
let shape = self.same_shape_binary_op(rhs, stringify!($fn_name))?;
let storage = self.storage.binary_impl::<crate::op::$op_name>(
&rhs.storage,
shape,
self.stride(),
rhs.stride(),
self.layout(),
rhs.layout(),
)?;
let op = if self.track_op() || rhs.track_op() {
Some(Op::$op_name(self.clone(), rhs.clone()))
@ -107,13 +104,10 @@ fn from_storage<S: Into<Shape>>(
op: Option<Op>,
is_variable: bool,
) -> Tensor {
let shape = shape.into();
let stride = shape.stride_contiguous();
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: Arc::new(storage),
shape,
stride,
layout: Layout::contiguous(shape),
op,
is_variable,
};
@ -323,6 +317,7 @@ impl Tensor {
unary_op!(sqrt, Sqrt);
unary_op!(gelu, Gelu);
unary_op!(relu, Relu);
pub fn to_scalar<S: crate::WithDType>(&self) -> Result<S> {
if self.rank() != 0 {
return Err(Error::UnexpectedNumberOfDims {
@ -342,8 +337,7 @@ impl Tensor {
}
pub fn affine(&self, mul: f64, add: f64) -> Result<Self> {
let shape = self.shape();
let storage = self.storage.affine_impl(shape, self.stride(), mul, add)?;
let storage = self.storage.affine(self.layout(), mul, add)?;
let op = if self.track_op() {
Some(Op::Affine {
arg: self.clone(),
@ -353,42 +347,25 @@ impl Tensor {
} else {
None
};
Ok(from_storage(storage, shape.clone(), op, false))
Ok(from_storage(storage, self.shape(), op, false))
}
/// Returns a new tensor that is a narrowed version of the input, the dimension `dim`
/// ranges from `start` to `start + length`.
// TODO: Once we've refactored the shape and strides, make this return a view of the same data
// rather than copying.
pub fn narrow(&self, dim: usize, start: usize, length: usize) -> Result<Self> {
let dims = self.shape().dims();
if dim >= dims.len() {
return Err(Error::UnexpectedNumberOfDims {
expected: dim + 1,
got: dims.len(),
shape: self.shape().clone(),
});
}
if start + length > dims[dim] {
todo!("add a proper error: out of bounds for narrow {dim} {start} {length} {dims:?}")
}
let mut dims = dims.to_vec();
dims[dim] = length;
let adjusted_shape = Shape::from(dims);
let mut storage = self.device().zeros(&adjusted_shape, self.dtype())?;
self.storage.copy_strided_src(
&mut storage,
/* dst_offset= */ 0,
&adjusted_shape,
&self.stride,
/* src_offest= */ self.stride[dim] * start,
)?;
let op = if self.track_op() {
Some(Op::Narrow(self.clone(), dim, start, length))
} else {
None
};
Ok(from_storage(storage, adjusted_shape, op, false))
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: self.storage.clone(),
layout: self.layout().narrow(dim, start, length)?,
op,
is_variable: false,
};
Ok(Tensor(Arc::new(tensor_)))
}
pub fn softmax(&self, dim: usize) -> Result<Self> {
@ -401,9 +378,7 @@ impl Tensor {
exp.broadcast_div(&sum_exp)
} else {
let shape = self.shape();
let mut storage = self
.storage
.unary_impl::<crate::op::Exp>(shape, self.stride())?;
let mut storage = self.storage.unary_impl::<crate::op::Exp>(self.layout())?;
// The resulting storage is contiguous.
storage.divide_by_sum_over_dim(shape, dim)?;
let op = if self.track_op() {
@ -416,7 +391,7 @@ impl Tensor {
}
pub fn sum(&self, sum_dims: &[usize]) -> Result<Self> {
let storage = self.storage.sum(self.shape(), &self.stride, sum_dims)?;
let storage = self.storage.sum(self.layout(), sum_dims)?;
let op = if self.track_op() {
Some(Op::Sum(self.clone(), sum_dims.to_vec()))
} else {
@ -458,11 +433,11 @@ impl Tensor {
let c_shape = Shape::from(&a_dims[..dim - 2]).extend(&[m, n]);
let batching: usize = a_dims[..dim - 2].iter().product();
let storage = self.storage.matmul_impl(
let storage = self.storage.matmul(
&rhs.storage,
(batching, m, n, k),
self.stride(),
rhs.stride(),
self.layout(),
rhs.layout(),
)?;
let op = if self.track_op() || rhs.track_op() {
Some(Op::Matmul(self.clone(), rhs.clone()))
@ -476,12 +451,11 @@ impl Tensor {
let _shap = self.same_shape_binary_op(on_true, "where_cond")?;
let shape = self.same_shape_binary_op(on_false, "where_cond")?;
let storage = self.storage.where_cond(
shape,
self.stride(),
self.layout(),
&on_true.storage,
on_true.stride(),
on_true.layout(),
&on_false.storage,
on_false.stride(),
on_false.layout(),
)?;
let op = if self.track_op() || on_true.track_op() || on_false.track_op() {
Some(Op::WhereCond(
@ -498,23 +472,19 @@ impl Tensor {
pub fn embedding(ids: &Self, rhs: &Self) -> Result<Self> {
if !rhs.is_contiguous() {
return Err(Error::RequiresContiguous { op: "embedding" });
} else if rhs.shape().rank() != 2 || ids.shape().rank() != 1 {
} else if rhs.rank() != 2 || ids.rank() != 1 {
return Err(Error::ShapeMismatchBinaryOp {
lhs: ids.shape.clone(),
rhs: rhs.shape.clone(),
lhs: ids.shape().clone(),
rhs: rhs.shape().clone(),
op: "embedding",
});
}
let ids_shape = ids.shape();
let seq_len = ids_shape.r1()?;
let (vocab_size, hidden_size) = rhs.shape().r2()?;
let storage = ids.storage.embedding_impl(
ids_shape,
&ids.stride,
&rhs.storage,
hidden_size,
vocab_size,
)?;
let (_, hidden_size) = rhs.shape().r2()?;
let storage = ids
.storage
.embedding(ids.layout(), &rhs.storage, rhs.layout())?;
let shape: Shape = (seq_len, hidden_size).into();
let op = if ids.track_op() || rhs.track_op() {
Some(Op::Embedding(ids.clone(), rhs.clone()))
@ -525,7 +495,7 @@ impl Tensor {
}
pub(crate) fn strided_index(&self) -> crate::StridedIndex {
crate::StridedIndex::new(self.dims(), self.stride())
self.layout.strided_index()
}
/// Returns data from the underlying storage, this does not take the strides
@ -618,15 +588,20 @@ impl Tensor {
}
pub fn shape(&self) -> &Shape {
&self.shape
self.layout().shape()
}
pub fn dims(&self) -> &[usize] {
self.shape().dims()
}
pub fn stride(&self) -> &[usize] {
&self.stride
pub fn layout(&self) -> &Layout {
&self.layout
}
// TODO: Rename to `stride` once the PR that introduced the layout has been merged.
pub fn stride_tmp(&self) -> &[usize] {
self.layout.stride()
}
pub fn rank(&self) -> usize {
@ -704,18 +679,6 @@ impl Tensor {
/// Returns a tensor that is a transposed version of the input, the given dimensions are
/// swapped.
pub fn transpose(&self, dim1: usize, dim2: usize) -> Result<Tensor> {
let rank = self.rank();
if rank <= dim1 || rank <= dim2 {
return Err(Error::UnexpectedNumberOfDims {
expected: usize::max(dim1, dim2),
got: rank,
shape: self.shape().clone(),
});
}
let mut stride = self.stride().to_vec();
let mut dims = self.shape().dims().to_vec();
dims.swap(dim1, dim2);
stride.swap(dim1, dim2);
let op = if self.track_op() {
Some(Op::Transpose(self.clone(), dim1, dim2))
} else {
@ -724,8 +687,7 @@ impl Tensor {
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: self.storage.clone(),
shape: Shape::from(dims),
stride,
layout: self.layout.transpose(dim1, dim2)?,
op,
is_variable: false,
};
@ -734,12 +696,12 @@ impl Tensor {
/// Returns true if the data is stored in a C contiguous (aka row major) way.
pub fn is_contiguous(&self) -> bool {
self.shape.is_contiguous(&self.stride)
self.layout.is_contiguous()
}
/// Returns true if the data is stored in a Fortran contiguous (aka column major) way.
pub fn is_fortran_contiguous(&self) -> bool {
self.shape.is_fortran_contiguous(&self.stride)
self.layout.is_fortran_contiguous()
}
/// Compared to clone, this copies the actual storage but may fail because of running out of
@ -748,8 +710,7 @@ impl Tensor {
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: Arc::new(self.storage.try_clone()?),
shape: self.shape.clone(),
stride: self.stride.clone(),
layout: self.layout.clone(),
op: None, // TODO
is_variable: false,
};
@ -762,8 +723,7 @@ impl Tensor {
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: self.storage.clone(),
shape: self.shape.clone(),
stride: self.stride.clone(),
layout: self.layout.clone(),
op: None,
is_variable: false,
};
@ -796,8 +756,7 @@ impl Tensor {
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: Arc::new(storage),
shape: self.shape.clone(),
stride: self.stride.clone(),
layout: self.layout.clone(),
op,
is_variable: false,
};
@ -810,7 +769,7 @@ impl Tensor {
pub fn broadcast_left<S: Into<Shape>>(&self, left_shape: S) -> Result<Self> {
let left_shape = left_shape.into();
let mut dims = left_shape.into_dims();
dims.extend(self.shape.dims());
dims.extend(self.dims());
self.broadcast_as(dims)
}
@ -820,36 +779,10 @@ impl Tensor {
} else {
None
};
let shape = shape.into();
if shape.rank() < self.rank() {
return Err(Error::BroadcastIncompatibleShapes {
src_shape: self.shape().clone(),
dst_shape: shape,
});
}
let added_dims = shape.rank() - self.rank();
let mut stride = vec![0; added_dims];
for (&dst_dim, (&src_dim, &src_stride)) in shape.dims()[added_dims..]
.iter()
.zip(self.dims().iter().zip(self.stride()))
{
let s = if dst_dim == src_dim {
src_stride
} else if src_dim != 1 {
return Err(Error::BroadcastIncompatibleShapes {
src_shape: self.shape().clone(),
dst_shape: shape,
});
} else {
0
};
stride.push(s)
}
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: self.storage.clone(),
shape,
stride,
layout: self.layout.broadcast_as(shape)?,
op,
is_variable: false,
};
@ -866,7 +799,7 @@ impl Tensor {
Ok(self.clone())
} else {
let shape = self.shape();
let storage = self.storage.to_dtype(shape, self.stride(), dtype)?;
let storage = self.storage.to_dtype(self.layout(), dtype)?;
let op = if self.track_op() {
Some(Op::ToDType(self.clone()))
} else {
@ -883,7 +816,7 @@ impl Tensor {
let shape = self.shape();
let mut storage = self.device().zeros(shape, self.dtype())?;
self.storage
.copy_strided_src(&mut storage, 0, &self.shape, &self.stride, 0)?;
.copy_strided_src(&mut storage, 0, self.layout())?;
Ok(from_storage(
storage,
shape.clone(),
@ -913,12 +846,10 @@ impl Tensor {
None
};
if self.is_contiguous() {
let stride = shape.stride_contiguous();
let tensor_ = Tensor_ {
id: TensorId::new(),
storage: self.storage.clone(),
shape,
stride,
layout: Layout::contiguous_with_offset(shape, self.layout.start_offset()),
op,
is_variable: false,
};
@ -926,7 +857,7 @@ impl Tensor {
} else {
let mut storage = self.device().zeros(&shape, self.dtype())?;
self.storage
.copy_strided_src(&mut storage, 0, &self.shape, &self.stride, 0)?;
.copy_strided_src(&mut storage, 0, self.layout())?;
Ok(from_storage(storage, shape, op, false))
}
}
@ -1063,7 +994,7 @@ impl Tensor {
for (arg, &offset) in args.iter().zip(offsets.iter()) {
let arg = arg.as_ref();
arg.storage
.copy_strided_src(&mut storage, offset, &arg.shape, &arg.stride, 0)?;
.copy_strided_src(&mut storage, offset, arg.layout())?;
}
Ok(from_storage(storage, shape, op, false))
}

4
candle-core/tests/tensor_tests.rs
View File

@ -263,12 +263,12 @@ fn matmul(device: &Device) -> Result<()> {
let a_tt = a.t()?.contiguous()?.t()?;
assert!(!a_tt.is_contiguous());
assert_eq!(a.dims(), a_tt.dims());
assert_eq!(a_tt.stride(), &[6, 1, 2]);
assert_eq!(a_tt.stride_tmp(), &[6, 1, 2]);
let b_tt = b.t()?.contiguous()?.t()?;
assert!(!b_tt.is_contiguous());
assert_eq!(b.dims(), b_tt.dims());
assert_eq!(b_tt.stride(), &[6, 1, 3]);
assert_eq!(b_tt.stride_tmp(), &[6, 1, 3]);
assert_eq!(a_tt.matmul(&b)?.to_vec3::<f32>()?, &expected);
assert_eq!(a.matmul(&b_tt)?.to_vec3::<f32>()?, &expected);