huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-17 02:58:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Switch from a macro to a trait to make things more generic.

Browse Source
This commit is contained in:
laurent
2023-06-28 21:06:56 +01:00
parent 0cfa21f26a
commit 2ae368e98e
2 changed files with 107 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

181
candle-core/src/cpu_backend.rs
View File

@ -1,10 +1,8 @@
use crate::op::{BinaryOp, UnaryOp};
use crate::{DType, Error, Layout, Result, Shape};
use crate::{DType, Error, Layout, Result, Shape, WithDType};
use gemm::{gemm, Parallelism};
use half::{bf16, f16};
// TODO: Think about whether we would be better off with a dtype and
// a buffer as an owned slice of bytes.
// TODO: Maybe we should not implement [Clone] here and instead have an explicit allocator +
// intercept the oom errors to avoid panicking and provide a proper error.
#[derive(Debug, Clone)]
@ -16,6 +14,24 @@ pub enum CpuStorage {
F64(Vec<f64>),
}
trait Map1 {
fn f<T: WithDType + Copy + num_traits::NumAssign>(
&self,
vs: &[T],
layout: &Layout,
) -> Result<Vec<T>>;
fn map(&self, vs: &CpuStorage, layout: &Layout) -> Result<CpuStorage> {
match vs {
CpuStorage::U32(vs) => Ok(CpuStorage::U32(self.f(vs, layout)?)),
CpuStorage::BF16(vs) => Ok(CpuStorage::BF16(self.f(vs, layout)?)),
CpuStorage::F16(vs) => Ok(CpuStorage::F16(self.f(vs, layout)?)),
CpuStorage::F32(vs) => Ok(CpuStorage::F32(self.f(vs, layout)?)),
CpuStorage::F64(vs) => Ok(CpuStorage::F64(self.f(vs, layout)?)),
}
}
}
fn wcond<T: Copy>(
pred: &[u32],
layout: &Layout,
@ -46,30 +62,30 @@ fn wcond<T: Copy>(
}
}
macro_rules! map1 {
($v: expr, $fn: ident, $( $args:expr ),*) => {{
let v = match $v {
CpuStorage::BF16(__s) => CpuStorage::BF16($fn::<bf16>(__s, $($args),*)?),
CpuStorage::F16(__s) => CpuStorage::F16($fn::<f16>(__s, $($args),*)?),
CpuStorage::F32(__s) => CpuStorage::F32($fn::<f32>(__s, $($args),*)?),
CpuStorage::F64(__s) => CpuStorage::F64($fn::<f64>(__s, $($args),*)?),
CpuStorage::U32(__s) => CpuStorage::U32($fn::<u32>(__s, $($args),*)?),
};
Ok(v)
}};
struct Sum<'a> {
dst_shape: &'a Shape,
sum_dims_and_stride: Vec<(usize, usize)>,
}
fn sum_impl1<T: Copy + num_traits::NumAssign>(
src: &[T],
dst_shape: &Shape,
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 src_layout.strided_index().enumerate() {
dst[to_dst_index(unstr_index)] += src[src_index];
impl<'a> Map1 for Sum<'a> {
fn f<T: WithDType + Copy + num_traits::NumAssign>(
&self,
src: &[T],
src_layout: &Layout,
) -> Result<Vec<T>> {
let mut dst = vec![T::zero(); self.dst_shape.elem_count()];
for (unstr_index, src_index) in src_layout.strided_index().enumerate() {
let mut dst_index = unstr_index;
// Set the sum_dims indexes to 0.
for &(dim, stride) in self.sum_dims_and_stride.iter() {
// The compiler is able to optimize the following in a single divmod op.
let (pre, post) = (dst_index / stride, dst_index % stride);
dst_index = (pre / dim) * stride + post;
}
dst[dst_index] += src[src_index];
}
Ok(dst)
}
Ok(dst)
}
fn unary_map<T: Copy, U: Copy, F: FnMut(T) -> U>(vs: &[T], layout: &Layout, mut f: F) -> Vec<U> {
@ -101,23 +117,48 @@ fn binary_map<T: Copy, F: FnMut(T, T) -> T>(
}
}
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 {
index,
vocab_size,
op: "take",
});
} else {
values.extend(&vs[hidden_size * index..hidden_size * (index + 1)]);
}
struct Affine(f64, f64);
impl Map1 for Affine {
fn f<T: WithDType + Copy + num_traits::NumAssign>(
&self,
vs: &[T],
layout: &Layout,
) -> Result<Vec<T>> {
let mul = T::from_f64(self.0);
let add = T::from_f64(self.1);
Ok(unary_map(vs, layout, |v| v * mul + add))
}
}
struct Embedding<'a> {
vocab_size: usize,
hidden_size: usize,
ids: &'a [u32],
ids_l: &'a Layout,
}
impl<'a> Map1 for Embedding<'a> {
fn f<T: WithDType>(&self, vs: &[T], layout: &Layout) -> Result<Vec<T>> {
// TODO: We assume that vs is contiguous here.
let vs = &vs[layout.start_offset()..];
let mut values = Vec::with_capacity(self.ids_l.shape().elem_count() * self.hidden_size);
// TODO: Optimize for the case where ids are contiguous.
for index in self.ids_l.strided_index() {
let index = self.ids[index].try_into()?;
if index >= self.vocab_size {
return Err(Error::InvalidIndex {
index,
vocab_size: self.vocab_size,
op: "take",
});
} else {
let hidden_size = self.hidden_size;
values.extend(&vs[hidden_size * index..hidden_size * (index + 1)]);
}
}
Ok(values)
}
Ok(values)
}
fn copy_strided_src_<T: Copy + std::fmt::Display>(
@ -348,19 +389,11 @@ impl CpuStorage {
.iter()
.map(|&d| (src_dims[d], src_dims[d + 1..].iter().product::<usize>()))
.collect();
let to_dst_index = |unstr_index: usize| {
// TODO: Optimize, the following does lots of slow division.
let mut dst_index = unstr_index;
// Set the sum_dims indexes to 0.
for &(dim, stride) in sum_dims_and_stride.iter() {
// The compiler is able to optimize the following in a single divmod op.
let (pre, post) = (dst_index / stride, dst_index % stride);
dst_index = (pre / dim) * stride + post;
}
dst_index
};
// TODO: Maybe provide an implementation with higher precision accumulators?
map1!(self, sum_impl1, &dst_shape, layout, to_dst_index)
Sum {
dst_shape: &dst_shape,
sum_dims_and_stride,
}
.map(self, layout)
}
pub(crate) fn divide_by_sum_over_dim(&mut self, shape: &Shape, dim: usize) -> Result<()> {
@ -447,36 +480,7 @@ impl CpuStorage {
}
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, 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, 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, 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, layout, |v| v * mul + add);
Ok(Self::F32(data))
}
Self::F64(storage) => {
let data = unary_map(storage, layout, |v| v * mul + add);
Ok(Self::F64(data))
}
}
Affine(mul, add).map(self, layout)
}
pub(crate) fn unary_impl<B: UnaryOp>(&self, layout: &Layout) -> Result<Self> {
@ -605,9 +609,16 @@ impl CpuStorage {
}
}
pub(crate) fn embedding(&self, layout: &Layout, rhs: &Self, rhs_l: &Layout) -> Result<Self> {
pub(crate) fn embedding(&self, ids_l: &Layout, rhs: &Self, rhs_l: &Layout) -> Result<Self> {
let ids = self.as_slice::<u32>()?;
map1!(rhs, take_impl1, ids, layout, rhs_l)
let (vocab_size, hidden_size) = rhs_l.shape().r2()?;
Embedding {
vocab_size,
hidden_size,
ids,
ids_l,
}
.map(rhs, rhs_l)
}
pub(crate) fn matmul(

17
candle-core/src/dtype.rs
View File

@ -34,6 +34,7 @@ impl DType {
pub trait WithDType: Sized + Copy {
const DTYPE: DType;
fn from_f64(v: f64) -> Self;
fn to_cpu_storage_owned(data: Vec<Self>) -> CpuStorage;
fn to_cpu_storage(data: &[Self]) -> CpuStorage {
@ -45,10 +46,14 @@ pub trait WithDType: Sized + Copy {
}
macro_rules! with_dtype {
($ty:ty, $dtype:ident) => {
($ty:ty, $dtype:ident, $from_f64:expr) => {
impl WithDType for $ty {
const DTYPE: DType = DType::$dtype;
fn from_f64(v: f64) -> Self {
$from_f64(v)
}
fn to_cpu_storage_owned(data: Vec<Self>) -> CpuStorage {
CpuStorage::$dtype(data)
}
@ -77,8 +82,8 @@ macro_rules! with_dtype {
}
};
}
with_dtype!(u32, U32);
with_dtype!(half::f16, F16);
with_dtype!(half::bf16, BF16);
with_dtype!(f32, F32);
with_dtype!(f64, F64);
with_dtype!(u32, U32, |v: f64| v as u32);
with_dtype!(half::f16, F16, half::f16::from_f64);
with_dtype!(half::bf16, BF16, half::bf16::from_f64);
with_dtype!(f32, F32, |v: f64| v as f32);
with_dtype!(f64, F64, |v: f64| v);