summaryrefslogtreecommitdiff
path: root/candle-flash-attn/kernels/block_info.h
diff options
context:
space:
mode:
Diffstat (limited to 'candle-flash-attn/kernels/block_info.h')
-rw-r--r--candle-flash-attn/kernels/block_info.h4
1 files changed, 2 insertions, 2 deletions
diff --git a/candle-flash-attn/kernels/block_info.h b/candle-flash-attn/kernels/block_info.h
index 65435e51..3a23a1e1 100644
--- a/candle-flash-attn/kernels/block_info.h
+++ b/candle-flash-attn/kernels/block_info.h
@@ -24,12 +24,12 @@ struct BlockInfo {
}
template <typename index_t>
- inline __device__ index_t q_offset(const index_t batch_stride, const index_t row_stride, const int bidb) const {
+ __forceinline__ __device__ index_t q_offset(const index_t batch_stride, const index_t row_stride, const int bidb) const {
return sum_s_q == -1 ? bidb * batch_stride : uint32_t(sum_s_q) * row_stride;
}
template <typename index_t>
- inline __device__ index_t k_offset(const index_t batch_stride, const index_t row_stride, const int bidb) const {
+ __forceinline__ __device__ index_t k_offset(const index_t batch_stride, const index_t row_stride, const int bidb) const {
return sum_s_k == -1 ? bidb * batch_stride : uint32_t(sum_s_k) * row_stride;
}