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

Handle the case where the kernel is not contiguous in the cuda backend. (#809)

Browse Source
This commit is contained in:
Laurent Mazare
2023-09-11 09:48:31 +01:00
committed by GitHub
parent 6fb665004c
commit df712ecf64
1 changed files with 14 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
candle-core/src/cuda_backend.rs
View File

@ -1746,10 +1746,20 @@ impl BackendStorage for CudaStorage {
let k = params.k_h * params.k_w * params.c_in;
let m = h_out * w_out;
let col_l = Layout::contiguous((b, m, k));
let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
.transpose(1, 2)?
.broadcast_as((b, k, n))?;
let res = col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?;
let res = if kernel_l.is_contiguous() {
let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
.transpose(1, 2)?
.broadcast_as((b, k, n))?;
col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
} else {
// Make the kernel contiguous if not already the case.
let mut kernel_c = self.device().zeros_impl(kernel_l.shape(), kernel.dtype())?;
kernel.copy_strided_src(&mut kernel_c, 0, kernel_l)?;
let kernel_l = Layout::contiguous_with_offset((1, n, k), kernel_l.start_offset())
.transpose(1, 2)?
.broadcast_as((b, k, n))?;
col.matmul(kernel, (b, m, n, k), &col_l, &kernel_l)?
};
let res_l = Layout::contiguous((b, h_out, w_out, n))
.transpose(1, 2)?
.transpose(1, 3)?;