diff options
| author | txh18 <cloudygooseg@gmail.com> | 2015-11-17 21:58:05 +0800 |
|---|---|---|
| committer | txh18 <cloudygooseg@gmail.com> | 2015-11-17 21:58:05 +0800 |
| commit | fc4c5a71053b837ed6143659a6e7b45792ed9e51 (patch) | |
| tree | a5d3fbf7c59dfb08f67b2156d70def64f98925ad /nerv/lib/matrix/generic/cukernel.cu | |
| parent | b8b6bb0a6b9fb9b8d72de42d27f598bfddd1cd0e (diff) | |
added atomicAdd for select_linear update, however, the result still seems unreproducable, I changed select_linear layer update back to line-by-line
Diffstat (limited to 'nerv/lib/matrix/generic/cukernel.cu')
| -rw-r--r-- | nerv/lib/matrix/generic/cukernel.cu | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/nerv/lib/matrix/generic/cukernel.cu b/nerv/lib/matrix/generic/cukernel.cu index 8885b41..f996fdd 100644 --- a/nerv/lib/matrix/generic/cukernel.cu +++ b/nerv/lib/matrix/generic/cukernel.cu @@ -232,7 +232,7 @@ __global__ void cudak_(update_select_rows)(MATRIX_ELEM *c, const MATRIX_ELEM *a, if (i >= nrow_a || j >= ncol_a) return; int i_c = lrintf(idx[i]); //c[j + i_c * stride_c] = c[j + i_c * stride_c] * (1 - beta * alpha) + a[j + i * stride_a] * alpha; - atomicAdd(c + j + i_c * stride_c, c[j + i_c * stride_c] * (- beta * alpha) + a[j + i * stride_a] * alpha); + atomicAdd_nvidia(&c[j + i_c * stride_c], c[j + i_c * stride_c] * (- beta * alpha) + a[j + i * stride_a] * alpha); } __global__ void cudak_(expand_frm)(const MATRIX_ELEM *a, MATRIX_ELEM *b, |