Fast correlation slower than regular?

[RMcCa]

Arm_correlate_fast_q31 takes almost twice as much time as arm_correlate_q31.

Am i missing something? I measured the calculating time for arm_correlate_q31 for various block sizes and then simply inserted the "fast_" in the function call and it takes almost twice as long.​

[waclawek.jan]

The DSP library is written in C, so particular compiler's behaviour may have dramatic impact.

And, especially on CM7, caching, buffering, the particularities of dual-issue, speed of memories involved etc. may have significant impact, too. And perhaps also other system conflicts (e.g. with DMA).

Try to make the comparison running both versions out of TCM, with data in TCM too. Of course on a minimal example, to exclude impact of other software.

JW

PS. Next time please start with stating what's your mcu and other relevant context information.