huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-17 11:08:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add cuda support for unary ops.

Browse Source
This commit is contained in:
laurent
2023-06-22 15:12:59 +01:00
parent b8f514d9c6
commit 5276755fb3
7 changed files with 101 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
src/cuda_backend.rs
View File

@ -145,21 +145,53 @@ impl CudaStorage {
match self {
Self::F32(arg) => {
let func = dev.get_or_load_func("affine_f32", kernels::AFFINE)?;
// SAFETY: if this function returns Ok(..), the kernel has been applied
// and has set the initially unset memory.
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.0.alloc::<f32>(elem_count) }?;
let params = (elem_count, arg, &out, mul as f32, add as f32);
// SAFETY: well, well, well...
// SAFETY: ffi.
unsafe { func.launch(cfg, params) }?;
Ok(Self::F32(out))
}
Self::F64(arg) => {
let func = dev.get_or_load_func("affine_f64", kernels::AFFINE)?;
// SAFETY: if this function returns Ok(..), the kernel has been applied
// and has set the initially unset memory.
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.0.alloc::<f64>(elem_count) }?;
let params = (elem_count, arg, &out, mul, add);
// SAFETY: well, well, well...
// SAFETY: ffi.
unsafe { func.launch(cfg, params) }?;
Ok(Self::F64(out))
}
}
}
pub(crate) fn unary_impl<U: crate::storage::UnaryOp>(
&self,
shape: &Shape,
stride: &[usize],
) -> Result<Self> {
if !shape.is_contiguous(stride) {
return Err(CudaError::RequiresContiguous { op: "affine" });
}
let elem_count = shape.elem_count();
let cfg = LaunchConfig::for_num_elems(elem_count as u32);
let dev = self.device();
match self {
Self::F32(arg) => {
let func = dev.get_or_load_func(U::KERNEL_F32, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.0.alloc::<f32>(elem_count) }?;
let params = (elem_count, arg, &out);
// SAFETY: ffi.
unsafe { func.launch(cfg, params) }?;
Ok(Self::F32(out))
}
Self::F64(arg) => {
let func = dev.get_or_load_func(U::KERNEL_F64, kernels::UNARY)?;
// SAFETY: Set later by running the kernel.
let out = unsafe { dev.0.alloc::<f64>(elem_count) }?;
let params = (elem_count, arg, &out);
// SAFETY: ffi.
unsafe { func.launch(cfg, params) }?;
Ok(Self::F64(out))
}