sgemm.cpp:block_sgemm_external_pack_threading_8x8, block_sgemm_external_pack_threading

Tencent / FeatherCNN

FeatherCNN is a high performance inference engine for convolutional neural networks.

1.21k stars 284 forks source link

Closed ejianzheng closed 6 years ago

ejianzheng commented 6 years ago

unsigned int tN = N / num_threads / factor;
tN = (tN + 7) & 0xFFFFFFF8;

for example: if N = 26 && num_threads = 3 tN = 8; thread task:8 8 8 but left 26-24= 2

turbo0628 commented 6 years ago

All SGEMM related functions are under reconstruction, please wait for a few days and these things will be fixed.

turbo0628 commented 6 years ago

The SGEMM is already reconstructed. I think this problem no longer exists. Can you help me check with it?

turbo0628 commented 6 years ago