Currently the benchmarking harness (so no matter which implementation) does not support buffer / batch sizes, which are unaligned to the number of images in the input. E.g. if there are 48 images in the input, and the buffer size is 32, it will fail during the 2nd read (because only 16 images are left).
Related code is here:
|
for (int i = 0; i < bufferSize; i++) { |
Currently the benchmarking harness (so no matter which implementation) does not support buffer / batch sizes, which are unaligned to the number of images in the input. E.g. if there are 48 images in the input, and the buffer size is 32, it will fail during the 2nd read (because only 16 images are left).
Related code is here:
RabbitCT/src/main.c
Line 275 in 770569e