Merge pull request #2 from LaurentMazare/matmul

Adding matmul.

.pre-commit-config.yaml

@@ -0,0 +1,15 @@
+repos:
+  - repo: https://github.com/Narsil/pre-commit-rust
+    rev: 2eed6366172ef2a5186e8785ec0e67243d7d73d0
+    hooks:
+      - id: fmt
+        name: "Rust (fmt)"
+      - id: clippy
+        name: "Rust (clippy)"
+        args:
+          [
+            "--tests",
+            "--examples",
+            "--",
+            "-Dwarnings",
+          ]

Cargo.toml

@@ -20,12 +20,13 @@ safetensors = "0.3.1"
 thiserror = "1"
 cudarc = { version = "0.9.9", optional = true }
 candle-kernels = { path = "kernels", optional = true }
+gemm = "0.15.4"
 
 [dev-dependencies]
 anyhow = "1"
 clap = { version = "4.2.4", features = ["derive"] }
 rand = "0.8.5"
-tokenizers = "0.13.3"
+tokenizers = { version = "0.13.3", default-features=false, features=["onig"] }
 
 [features]
 default = []

src/cpu_backend.rs

@@ -1,5 +1,6 @@
 use crate::storage::{BinaryOp, UnaryOp};
 use crate::{DType, Error, Result, Shape, StridedIndex};
+use gemm::{gemm, Parallelism};
 
 // TODO: Think about whether we would be better off with a dtype and
 // a buffer as an owned slice of bytes.
@@ -17,6 +18,14 @@ impl CpuStorage {
         }
     }
 
+    pub fn as_slice<D: crate::WithDType>(&self) -> Result<&[D]> {
+        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 affine_impl(
         &self,
         shape: &Shape,
@@ -97,6 +106,93 @@ impl CpuStorage {
         }
     }
 
+    pub(crate) fn matmul_impl(
+        &self,
+        rhs: &Self,
+        (b, m, n, k): (usize, usize, usize, usize),
+        lhs_stride: &[usize],
+        rhs_stride: &[usize],
+    ) -> Result<Self> {
+        let a_skip: usize = m * k;
+        let b_skip: usize = n * k;
+        let c_skip: usize = m * n;
+
+        let rank = lhs_stride.len();
+        let lhs_cs = lhs_stride[rank - 1];
+        let lhs_rs = lhs_stride[rank - 2];
+
+        let rhs_cs = rhs_stride[rank - 1];
+        let rhs_rs = rhs_stride[rank - 2];
+
+        if lhs_stride.len() > 2 {
+            let lhs_batch_stride = &lhs_stride[..rank - 2];
+            let rhs_batch_stride = &rhs_stride[..rank - 2];
+
+            if lhs_batch_stride != [a_skip] || rhs_batch_stride != [b_skip] {
+                // Temporary error before we support abitrary striding.
+                return Err(Error::UnexpectedStriding);
+            }
+        }
+
+        let mut dst = vec![0.0; b * m * n];
+
+        let dst_shape: Shape = (m, n).into();
+        let dst_strides = dst_shape.stride_contiguous();
+        let dst_rs = dst_strides[0];
+        let dst_cs = dst_strides[1];
+
+        for step in 0..b {
+            let lhs_p = &self.as_slice::<f32>()?[step * a_skip..];
+            let rhs_p = &rhs.as_slice::<f32>()?[step * b_skip..];
+            let dst_p = &mut dst[step * c_skip..];
+            unsafe {
+                gemm(
+                    // m: usize,
+                    m,
+                    // n: usize,
+                    n,
+                    // k: usize,
+                    k,
+                    // dst: *mut T,
+                    dst_p.as_mut_ptr(),
+                    // dst_cs: isize,
+                    dst_cs as isize,
+                    // dst_rs: isize,
+                    dst_rs as isize,
+                    // read_dst: bool,
+                    false,
+                    // lhs: *const T,
+                    lhs_p.as_ptr(),
+                    // lhs_cs: isize,
+                    lhs_cs as isize,
+                    // lhs_rs: isize,
+                    lhs_rs as isize,
+                    // rhs: *const T,
+                    rhs_p.as_ptr(),
+                    // rhs_cs: isize,
+                    rhs_cs as isize,
+                    // rhs_rs: isize,
+                    rhs_rs as isize,
+                    // alpha: T,
+                    1.0,
+                    // beta: T,
+                    1.0,
+                    // conj_dst: bool,
+                    false,
+                    // conj_lhs: bool,
+                    false,
+                    // conj_rhs: bool,
+                    true,
+                    // parallelism: Parallelism
+                    Parallelism::None,
+                )
+            }
+        }
+
+        let c = Self::F32(dst);
+        Ok(c)
+    }
+
     pub(crate) fn ones_impl(shape: &Shape, dtype: DType) -> Self {
         let elem_count = shape.elem_count();
         match dtype {
@@ -125,3 +221,45 @@ impl CpuStorage {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::{Device, Tensor};
+
+    #[test]
+    fn simple_matmul() -> Result<()> {
+        let data = vec![1.0f32, 2.0, 3.0, 4.0];
+        let a = Tensor::from_slice(&data, (2, 2), &Device::Cpu)?;
+        let data = vec![1.0f32, 2.0, 3.0, 4.0];
+        let b = Tensor::from_slice(&data, (2, 2), &Device::Cpu)?;
+
+        let c = a.matmul(&b)?;
+        assert_eq!(c.to_vec2::<f32>()?, &[&[7.0f32, 10.0], &[15.0, 22.0]]);
+
+        let data = vec![1.0f32, 2.0];
+        let a = Tensor::from_slice(&data, (2, 1), &Device::Cpu)?;
+        let data = vec![3.0f32, 4.0];
+        let b = Tensor::from_slice(&data, (1, 2), &Device::Cpu)?;
+        let c = a.matmul(&b)?;
+        assert_eq!(c.to_vec2::<f32>()?, &[&[3.0, 4.0], &[6.0, 8.0]]);
+
+        let data: Vec<_> = (0..6).map(|i| i as f32).collect();
+        let a = Tensor::from_slice(&data, (2, 3), &Device::Cpu)?;
+        let data: Vec<_> = (0..6).map(|i| (i + 2) as f32).collect();
+        let b = Tensor::from_slice(&data, (3, 2), &Device::Cpu)?;
+        let c = a.matmul(&b)?;
+        assert_eq!(c.to_vec2::<f32>()?, &[&[16., 19.], &[52., 64.]]);
+
+        let data: Vec<_> = (0..12).map(|i| i as f32).collect();
+        let a = Tensor::from_slice(&data, (2, 2, 3), &Device::Cpu)?;
+        let data: Vec<_> = (0..12).map(|i| (i + 2) as f32).collect();
+        let b = Tensor::from_slice(&data, (2, 3, 2), &Device::Cpu)?;
+        let c = a.matmul(&b)?;
+        assert_eq!(
+            c.to_vec3::<f32>()?,
+            &[&[&[16., 19.], &[52., 64.]], &[&[214., 235.], &[304., 334.]]]
+        );
+        Ok(())
+    }
+}

src/device.rs

@@ -101,7 +101,7 @@ impl Device {
         }
     }
 
-    pub(crate) fn tensor<A: NdArray>(&self, array: A) -> Result<Storage> {
+    pub(crate) fn storage<A: NdArray>(&self, array: A) -> Result<Storage> {
         match self {
             Device::Cpu => Ok(Storage::Cpu(array.to_cpu_storage())),
             Device::Cuda(device) => {

src/dtype.rs

@@ -25,6 +25,7 @@ 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]>;
 }
 
 macro_rules! with_dtype {
@@ -45,6 +46,16 @@ 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(),
+                    }),
+                }
+            }
         }
     };
 }

src/error.rs

@@ -12,6 +12,11 @@ pub enum Error {
     #[error("the candle crate has not been built with cuda support")]
     NotCompiledWithCudaSupport,
 
+    #[error(
+        "Shape mismatch, got buffer of size {buffer_size} which is compatible with shape {shape:?}"
+    )]
+    ShapeMismatch { buffer_size: usize, shape: Shape },
+
     #[error("shape mismatch in {op}, lhs: {lhs:?}, rhs: {rhs:?}")]
     ShapeMismatchBinaryOp {
         lhs: Shape,
@@ -40,6 +45,10 @@ pub enum Error {
         shape: Shape,
     },
 
+    // TODO this is temporary when we support arbitrary matmul
+    #[error("temporary error where matmul doesn't support arbitrary striding")]
+    UnexpectedStriding,
+
     #[error(transparent)]
     Cuda(#[from] crate::CudaError),
 }

src/op.rs

@@ -5,6 +5,7 @@ pub(crate) enum Op {
     Mul(Tensor, Tensor),
     Sub(Tensor, Tensor),
     Div(Tensor, Tensor),
+    Matmul(Tensor, Tensor),
 
     #[allow(dead_code)] // add is currently unused.
     Affine {

src/storage.rs

@@ -241,4 +241,22 @@ impl Storage {
     pub(crate) fn sqrt_impl(&self, shape: &Shape, stride: &[usize]) -> Result<Self> {
         self.unary_impl::<Sqrt>(shape, stride)
     }
+
+    pub(crate) fn matmul_impl(
+        &self,
+        rhs: &Self,
+        bmnk: (usize, usize, usize, usize),
+        lhs_stride: &[usize],
+        rhs_stride: &[usize],
+    ) -> Result<Self> {
+        self.same_device(rhs, "matmul")?;
+        self.same_dtype(rhs, "matmul")?;
+        match (self, rhs) {
+            (Storage::Cpu(storage), Storage::Cpu(rhs_storage)) => {
+                let storage = storage.matmul_impl(rhs_storage, bmnk, lhs_stride, rhs_stride)?;
+                Ok(Self::Cpu(storage))
+            }
+            _ => todo!(),
+        }
+    }
 }

src/tensor.rs

@@ -147,11 +147,16 @@ impl Tensor {
 
     pub fn new_impl<A: crate::device::NdArray>(
         array: A,
+        shape: Shape,
         device: &Device,
         is_variable: bool,
     ) -> Result<Self> {
-        let shape = array.shape()?;
-        let storage = device.tensor(array)?;
+        let n: usize = shape.elem_count();
+        let buffer_size: usize = array.shape()?.elem_count();
+        if buffer_size != n {
+            return Err(Error::ShapeMismatch { buffer_size, shape });
+        }
+        let storage = device.storage(array)?;
         let stride = shape.stride_contiguous();
         let tensor_ = Tensor_ {
             id: TensorId::new(),
@@ -165,11 +170,29 @@ impl Tensor {
     }
 
     pub fn new<A: crate::device::NdArray>(array: A, device: &Device) -> Result<Self> {
-        Self::new_impl(array, device, false)
+        let shape = array.shape()?;
+        Self::new_impl(array, shape, device, false)
     }
 
     pub fn var<A: crate::device::NdArray>(array: A, device: &Device) -> Result<Self> {
-        Self::new_impl(array, device, true)
+        let shape = array.shape()?;
+        Self::new_impl(array, shape, device, true)
+    }
+
+    pub fn from_slice<S: Into<Shape>, D: crate::WithDType>(
+        array: &[D],
+        shape: S,
+        device: &Device,
+    ) -> Result<Self> {
+        Self::new_impl(array, shape.into(), device, false)
+    }
+
+    pub fn var_from_slice<S: Into<Shape>, D: crate::WithDType>(
+        array: &[D],
+        shape: S,
+        device: &Device,
+    ) -> Result<Self> {
+        Self::new_impl(array, shape.into(), device, true)
     }
 
     pub(crate) fn same_shape_binary_op(&self, rhs: &Self, op: &'static str) -> Result<&Shape> {
@@ -234,10 +257,65 @@ impl Tensor {
         Ok(Self(Arc::new(tensor_)))
     }
 
+    pub fn matmul(&self, rhs: &Self) -> Result<Self> {
+        let a_dims = self.shape().dims();
+        let b_dims = rhs.shape().dims();
+
+        let dim = a_dims.len();
+
+        if dim < 2 || b_dims.len() != dim {
+            return Err(Error::ShapeMismatchBinaryOp {
+                lhs: self.shape().clone(),
+                rhs: rhs.shape().clone(),
+                op: "matmul",
+            });
+        }
+
+        let m = a_dims[dim - 2];
+        let k = a_dims[dim - 1];
+        let k2 = b_dims[dim - 2];
+        let n = b_dims[dim - 1];
+        if k != k2 {
+            return Err(Error::ShapeMismatchBinaryOp {
+                lhs: self.shape().clone(),
+                rhs: rhs.shape().clone(),
+                op: "matmul",
+            });
+        }
+
+        let mut c_shape: Vec<_> = a_dims[..dim - 2].into();
+        c_shape.extend(&[m, n]);
+        let c_shape = Shape(c_shape);
+        let batching: usize = a_dims[..dim - 2].iter().product();
+
+        let storage = self.storage.matmul_impl(
+            &rhs.storage,
+            (batching, m, n, k),
+            self.stride(),
+            rhs.stride(),
+        )?;
+        let tensor_ = Tensor_ {
+            id: TensorId::new(),
+            storage,
+            shape: c_shape.clone(),
+            stride: c_shape.stride_contiguous(),
+            op: Some(Op::Matmul(self.clone(), rhs.clone())),
+            is_variable: false,
+        };
+        Ok(Self(Arc::new(tensor_)))
+    }
+
     pub(crate) fn strided_index(&self) -> crate::StridedIndex {
         crate::StridedIndex::new(self.dims(), self.stride())
     }
 
+    pub fn as_slice<S: crate::WithDType>(&self) -> Result<&[S]> {
+        match &self.storage {
+            Storage::Cpu(cpu_storage) => S::cpu_storage_as_slice(cpu_storage),
+            Storage::Cuda { .. } => todo!(),
+        }
+    }
+
     pub fn to_vec1<S: crate::WithDType>(&self) -> Result<Vec<S>> {
         if self.rank() != 1 {
             return Err(Error::UnexpectedNumberOfDims {
@@ -279,6 +357,28 @@ impl Tensor {
         }
     }
 
+    pub fn to_vec3<S: crate::WithDType>(&self) -> Result<Vec<Vec<Vec<S>>>> {
+        let (dim1, dim2, dim3) = self.shape().r3()?;
+        match &self.storage {
+            Storage::Cpu(cpu_storage) => {
+                let data = S::cpu_storage_as_slice(cpu_storage)?;
+                let mut top_rows = vec![];
+                let mut src_index = self.strided_index();
+                for _idx in 0..dim1 {
+                    let mut rows = vec![];
+                    for _jdx in 0..dim2 {
+                        let row = (0..dim3).map(|_| data[src_index.next().unwrap()]).collect();
+                        rows.push(row)
+                    }
+                    top_rows.push(rows);
+                }
+                assert!(src_index.next().is_none());
+                Ok(top_rows)
+            }
+            Storage::Cuda { .. } => todo!(),
+        }
+    }
+
     pub fn dtype(&self) -> DType {
         self.storage.dtype()
     }
@@ -311,6 +411,31 @@ impl Tensor {
         self.id
     }
 
+    pub fn t(&self) -> Result<Tensor> {
+        let mut stride = self.stride().to_vec();
+        let mut shape = self.shape().clone();
+        let n = stride.len();
+        if n < 2 {
+            return Err(Error::UnexpectedNumberOfDims {
+                expected: 2,
+                got: n,
+                shape: self.shape().clone(),
+            });
+        }
+        (shape.0[n - 2], shape.0[n - 1]) = (shape.0[n - 1], shape.0[n - 2]);
+        (stride[n - 2], stride[n - 1]) = (stride[n - 1], stride[n - 2]);
+        let tensor_ = Tensor_ {
+            id: TensorId::new(),
+            storage: self.storage.clone(),
+            shape,
+            stride,
+            // TODO The op should have a backward
+            op: None,
+            is_variable: false,
+        };
+        Ok(Tensor(Arc::new(tensor_)))
+    }
+
     pub fn is_contiguous(&self) -> bool {
         self.shape.is_contiguous(&self.stride)
     }
@@ -340,7 +465,8 @@ impl Tensor {
                     Op::Add(lhs, rhs)
                     | Op::Mul(lhs, rhs)
                     | Op::Sub(lhs, rhs)
-                    | Op::Div(lhs, rhs) => {
+                    | Op::Div(lhs, rhs)
+                    | Op::Matmul(lhs, rhs) => {
                         let (tg, nodes) = walk(lhs, nodes, already_seen);
                         track_grad |= tg;
                         let (tg, nodes) = walk(rhs, nodes, already_seen);
@@ -420,6 +546,18 @@ impl Tensor {
                         let rhs_sum_grad = grads.or_insert(rhs)?;
                         *rhs_sum_grad = rhs_sum_grad.add(&rhs_grad)?;
                     }
+                    Op::Matmul(lhs, rhs) => {
+                        // Skipping checks, the op went ok, we can skip
+                        // the matmul size checks for now.
+
+                        let lhs_grad = grad.matmul(&rhs.t()?)?;
+                        let lhs_sum_grad = grads.or_insert(lhs)?;
+                        *lhs_sum_grad = lhs_sum_grad.add(&lhs_grad)?;
+
+                        let rhs_grad = lhs.t()?.matmul(&grad)?;
+                        let rhs_sum_grad = grads.or_insert(rhs)?;
+                        *rhs_sum_grad = rhs_sum_grad.add(&rhs_grad)?;
+                    }
                     Op::Affine { arg, mul, .. } => {
                         let arg_grad = grad.affine(*mul, 0.)?;
                         let sum_grad = grads.or_insert(arg)?;

tests/grad_tests.rs

@@ -1,5 +1,5 @@
 use anyhow::{Context, Result};
-use candle::{Device, Tensor};
+use candle::{Device, Shape, Tensor};
 
 #[test]
 fn simple_grad() -> Result<()> {
@@ -14,3 +14,27 @@ fn simple_grad() -> Result<()> {
     assert_eq!(grad_x.to_vec1::<f32>()?, [11., 7., 13.]);
     Ok(())
 }
+
+#[test]
+fn matmul_grad() -> Result<()> {
+    let data: Vec<_> = (0..12).map(|i| i as f32).collect();
+    let x = Tensor::var_from_slice(&data, (2, 2, 3), &Device::Cpu)?;
+    let data: Vec<_> = (0..12).map(|i| i as f32).collect();
+    let y = Tensor::var_from_slice(&data, (2, 3, 2), &Device::Cpu)?;
+
+    let c = x.matmul(&y)?;
+    let grads = c.backward()?;
+    let grad_x = grads.get(&x).context("no grad for x")?;
+    let grad_y = grads.get(&y).context("no grad for y")?;
+    assert_eq!(grad_x.shape(), &Shape::from((2, 2, 3)));
+    assert_eq!(grad_y.shape(), &Shape::from((2, 3, 2)));
+    assert_eq!(
+        grad_x.as_slice::<f32>()?,
+        &[1., 5., 9., 1., 5., 9., 13., 17., 21., 13., 17., 21.]
+    );
+    assert_eq!(
+        grad_y.as_slice::<f32>()?,
+        &[3., 3., 5., 5., 7., 7., 15., 15., 17., 17., 19., 19.]
+    );
+    Ok(())
+}