tests: lib: cmsis_dsp: fastmath: Adjust error threshold for MVE

The MVE vector version of the `vinverse` function is slightly less
accurate than the scalar version, so allow a higher relative error
threshold when the MVE is enabled.

Signed-off-by: Stephanos Ioannidis <root@stephanos.io>
This commit is contained in:
Stephanos Ioannidis 2022-05-03 21:12:39 +09:00
commit f560eebd86

View file

@ -20,6 +20,15 @@
#define ABS_ERROR_THRESH (1.0e-3)
#define ABS_LOG_ERROR_THRESH (3.0e-2)
#ifdef CONFIG_ARMV8_1_M_MVEF
/*
* NOTE: The MVE vector version of the `vinverse` function is slightly less
* accurate than the scalar version.
*/
#undef REL_ERROR_THRESH
#define REL_ERROR_THRESH (1.1e-3)
#endif
#if 0
/*
* NOTE: These tests must be enabled once the F16 sine and cosine function