huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-21 04:10:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

Avoid the race condition on cuda sums.

Browse Source
This commit is contained in:
laurent
2023-06-26 16:19:06 +01:00
parent 687c5beb6a
commit f2ac5547fc
3 changed files with 15 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
kernels/src/reduce.cu
View File

@ -20,7 +20,7 @@ extern "C" __global__ void FN_NAME( \
for (unsigned int nd = 0; nd < num_sum_dims; ++nd) { \
size_t stride = sum_dims_s[nd]; \
size_t pre = dst_index / stride; \
size_t post = dst_index / stride; \
size_t post = dst_index % stride; \
dst_index = (pre / sum_dims_l[nd]) * stride + post; \
} \
out[dst_index] += inp[i]; \
@ -33,7 +33,7 @@ extern "C" __global__ void FN_NAME( \
for (unsigned int nd = 0; nd < num_sum_dims; ++nd) { \
size_t stride = sum_dims_s[nd]; \
size_t pre = dst_index / stride; \
size_t post = dst_index / stride; \
size_t post = dst_index % stride; \
dst_index = (pre / sum_dims_l[nd]) * stride + post; \
} \
out[dst_index] += inp[strided_i]; \