The length of the output will be halved

[yizhibaiwuya]

how-to-optim-algorithm-in-cuda/reduce/reduce_v3_idle_threads_free.cu

Line 36 in 457fedd

int32_t block_num = (N + BLOCK_SIZE - 1) / BLOCK_SIZE / 2;

In reduce_v3_idle_threads_free.cu, each block is responsible for the reduction of 512 elements. However, this will result in the final output length being halved. But this is contrary to our desire to reduce every 256 elements.

[BBuf]